text
stringlengths 254
1.16M
|
|---|
---
title: Association Between Magnitude of Differential Blood Pressure Reduction and
Secondary Stroke Prevention
authors:
- Chia-Yu Hsu
- Jeffrey L. Saver
- Bruce Ovbiagele
- Yi-Ling Wu
- Chun-Yu Cheng
- Meng Lee
journal: JAMA Neurology
year: 2023
pmcid: PMC10028545
doi: 10.1001/jamaneurol.2023.0218
license: CC BY 4.0
---
# Association Between Magnitude of Differential Blood Pressure Reduction and Secondary Stroke Prevention
## Abstract
This meta-analysis and meta-regression of randomized clinical trials evaluates the association of magnitude of differential blood pressure reduction and recurrent stroke in patients with stroke or transient ischemic attack.
## Key Points
### Question
How much better is more blood pressure reduction vs less blood pressure reduction for secondary stroke prevention?
### Findings
In this meta-analysis and meta-regression that included 10 randomized clinical trials comprising 40 710 patients with stroke or transient ischemic attack, the risk of recurrent stroke was $8.4\%$ with more intensive blood pressure lowering vs $10.1\%$ with less intensive or no blood pressure lowering, a statistically significant difference. The greater the amount of differential blood pressure reduction, the greater the amount of reduction risk of recurrent stroke.
### Meaning
This study suggests that more intensive differential blood pressure–lowering therapy may be beneficial for secondary stroke prevention.
### Importance
The degree to which more intensive blood pressure reduction is better than less intensive for secondary stroke prevention has not been delineated.
### Objective
To perform a standard meta-analysis and a meta-regression of randomized clinical trials to evaluate the association of magnitude of differential blood pressure reduction and recurrent stroke in patients with stroke or transient ischemic attack (TIA).
### Data Sources
PubMed, Embase, the Cochrane Central Register of Controlled Trials, and ClinicalTrials.gov were searched from January 1, 1980, to June 30, 2022.
### Study Selection
Randomized clinical trials that compared more intensive vs less intensive blood pressure lowering and recorded the outcome of recurrent stroke in patients with stroke or TIA.
### Data Extraction and Synthesis
The Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline was used for abstracting data and assessing data quality and validity. Risk ratio (RR) with $95\%$ CI was used as a measure of the association of more intensive vs less intensive blood pressure lowering with primary and secondary outcomes. The univariate meta-regression analyses were conducted to evaluate a possible moderating effect of magnitude of differential systolic blood pressure (SBP) and diastolic blood pressure (DBP) reduction on the recurrent stroke and major cardiovascular events.
### Main Outcomes and Measures
The primary outcome was recurrent stroke and the lead secondary outcome was major cardiovascular events.
### Results
Ten randomized clinical trials comprising 40 710 patients (13 752 women [$34\%$]; mean age, 65 years) with stroke or TIA were included for analysis. The mean duration of follow-up was 2.8 years (range, 1-4 years). Pooled results showed that more intensive treatment compared with less intensive was associated with a reduced risk of recurrent stroke in patients with stroke or TIA (absolute risk, $8.4\%$ vs $10.1\%$; RR, 0.83; $95\%$ CI, 0.78-0.88). Meta-regression showed that the magnitude of differential SBP and DBP reduction was associated with a lower risk of recurrent stroke in patients with stroke or TIA in a log-linear fashion (SBP: regression slope, −0.06; $95\%$ CI, −0.08 to −0.03; DBP: regression slope, −0.17; $95\%$ CI, −0.26 to −0.08). Similar results were found in the association between differential blood pressure lowering and major cardiovascular events.
### Conclusions and Relevance
More intensive blood pressure–lowering therapy might be associated with a reduced risk of recurrent stroke and major cardiovascular events. These results might support the use of more intensive blood pressure reduction for secondary prevention in patients with stroke or TIA.
## Introduction
Hypertension is a major risk factor for recurrent stroke in patients with ischemic and hemorrhagic stroke or transient ischemic attack (TIA), and blood pressure lowering is therefore a guideline-recommended strategy to prevent recurrent stroke. Indeed, recurrent stroke has declined substantially over decades, with improved blood pressure control as a leading cause.1 However, most individual randomized clinical trials have not significantly showed that blood pressure–lowering therapy reduces recurrent stroke in patients with stroke or TIA. Relatively small magnitude of blood pressure reduction and relatively few participants and inadequate statistical power in single randomized clinical trials may have attenuated signals of efficacy.2,3 Furthermore, some trials2,4 compared antihypertensive therapy with placebo, while other trials3,5 compared a lower systolic blood pressure (SBP) target with a higher SBP-lowering target. Pooling of data from these 2 types of trials is needed to clarify the association between degree of increased blood pressure lowering attained with therapy and secondary stroke prevention.
Therefore, we performed a systematic review and meta-analysis of all relevant randomized clinical trials to evaluate the association between magnitude of differential blood pressure reduction and risk of recurrent stroke and performed a meta-regression to clarify whether larger differential blood pressure reduction magnitude was monotonically associated with a lower recurrent stroke risk in patients with stroke or TIA.
## Methods
The Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline was used for abstracting data and validity of this meta-analysis.6 The protocol was registered with PROSPERO (CRD42022309056).
## Search Methods and Resources
We searched PubMed, Embase, the Cochrane Central Register of Controlled Trials, and the clinical trial registry maintained at ClinicalTrials.gov from January 1, 1980, to June 30, 2022, using the following terms: stroke or cerebrovascular accident or brain vascular accident or cerebral infarct or cerebrovascular disorder or intracranial vascular disease or cerebrovascular disease or cerebrovascular occlusion or transient ischemic attack AND antihypertensive or blood pressure lowering or blood pressure reduction AND recurrent or secondary prevention or previous or prior or history. We restricted the search to studies in humans and randomized clinical trials and did not apply language restrictions. We also reviewed prior meta-analyses7,8,9 to identify additional trials. Two investigators (C.-Y.H. and Y.-L.W.) independently screened and identified potential trials, and discrepancies were resolved by discussion with a third investigator (M.L.).
## Study Selection and Data Extraction
Criteria for inclusion of a study were as follows: [1] the study design was a randomized clinical trial; [2] all or an identifiable subset of participants had a history of stroke or TIA; [3] the study evaluated more intensive vs less intensive blood pressure–lowering therapy, including the following possible comparisons: antihypertensive drug(s) vs placebo and a lower blood pressure target vs a higher blood pressure target; [4] recurrent stroke was reported as an end point; [5] the magnitude of SBP reduction between more intensive and less intensive blood pressure lowering was reported; and [6] treatment duration was of at least 1 year.
Criteria for exclusion of a study were as follows: [1] the study was published before 1980 because strategies for secondary stroke prevention, including antihypertensive drug use, were substantially different then1; [2] the study adopted only nonpharmaceutical approach, such as salt restriction or exercise, in the active treatment arm; [3] the study compared one antihypertensive drug vs another antihypertensive drug; [4] achieved SBP was higher in more intensive blood pressure–lowering group than less intensive blood pressure–lowering group after treatment; [5] more than $10\%$ of patients were enrolled within 3 days after stroke; or [6] more than $10\%$ of patients had end-stage kidney disease because of differential hemodynamic vulnerability of this disease. We extracted characteristics of each trial, which included patient age, sex, baseline blood pressure, number of patients in more intensive and less intensive blood pressure–lowering groups, duration of follow-up, magnitude of differential blood pressure reduction between more intensive and less intensive blood pressure–lowering groups, and number of recurrent stroke events and other outcomes in more intensive and less intensive blood pressure–lowering groups. Two investigators (C.-Y.H. and M.L.) independently abstracted the data and any discrepant judgments were resolved by joint discussion and by referencing the original report.
## Study Quality Assessment
Because all of the included studies were randomized clinical trials, the risk of bias (eg, selection bias, performance bias, detection bias, attrition bias, reporting bias, and other issues) of the included trials was assessed by the Cochrane risk-of-bias algorithm.10,11
## Outcomes
The primary outcome was recurrent stroke. The lead secondary outcome was major cardiovascular events. Major cardiovascular events were defined as the composition of nonfatal stroke, nonfatal myocardial infarction, and death from cardiovascular causes. Additional secondary outcomes were recurrent ischemic stroke, hemorrhagic stroke, fatal or disabling stroke, myocardial infarction, death from cardiovascular causes, death from any cause, and heart failure.
## Statistical Analysis
The analysis plan was performed on an intention-to-treat basis. We computed the fixed-effects estimate based on the Mantel-Haenszel method. Risk ratio (RR) with $95\%$ CI was used as a measure of the association of more intensive vs less intensive blood pressure lowering with the primary and secondary outcomes. To explore the association between magnitude of differential blood pressure reduction and risk of recurrent stroke, analyses were performed for thresholds of SBP reduction of 4 mm Hg or lower, 5 mm Hg or lower, more than 5 mm Hg, more than 7 mm Hg, and more than 11 mm Hg, as well as magnitude of differential diastolic blood pressure (DBP) reduction of 2 mm Hg or lower, 3 mm Hg or lower, more than 3 mm Hg, and more than 4 mm Hg between more intensive vs less intensive blood pressure–lowering groups. The univariate meta-regression analyses were conducted with the fixed-effects model to evaluate a possible moderating effect of magnitude of differential SBP and DBP reduction on the recurrent stroke and major cardiovascular events. All P values were from 2-sided tests, and results were deemed statistically significant at $P \leq .05.$ Heterogeneity was assessed by a P value determined by the use of χ2 statistics and I2 statistics, and I2 values of $0\%$ to $29\%$, $30\%$ to $49\%$, $50\%$ to $74\%$, and $75\%$ to $100\%$ represent not important, moderate, substantial, and considerable inconsistency, respectively.12 The trim-and-fill method to identify and correct for funnel plot asymmetry arising from publication bias was used.13 A sensitivity test was conducted to identify any trial that might have exerted a disproportionate influence on the summary treatment effect on the primary outcome by removing each individual trial from the meta-analysis one at a time. Another sensitivity test was conducted by restricting analysis within trials with recurrent stroke being the primary outcome in the original trial design. Grading of Recommendations, Assessment, Development and Evaluations was used to evaluate summaries of evidence for the primary and secondary outcomes.14,15 Subgroup analyses of included trials were conducted according to different study characteristics: mean baseline SBP levels (≥150 mm Hg vs 140-149 mm Hg), mean achieved SBP levels in the more intensive and less intensive blood pressure–lowering groups (≥140 mm Hg vs 130 to <140 mm Hg vs <130 mm Hg), study duration (<3 years vs ≥3 years), sample size (<3000 vs ≥3000 patients), time interval from index stroke to randomization (within 6 months from stroke vs within 3-5 years from stroke), entry event (ischemic stroke vs hemorrhagic stroke), study design (antihypertensive drugs vs placebo and a lower blood pressure target vs a higher blood pressure target), definition of differential blood pressure reduction (mean difference throughout the studies vs other definitions), and antihypertensive drugs used in the more intensive treated arm (angiotensin-converting enzyme [ACE] inhibitors vs angiotensin receptor blockers vs β-blockers vs diuretics vs ACE inhibitors plus diuretics). Since the risk of recurrent ischemic and hemorrhagic stroke is greater in Asian populations with high blood pressure compared with several other groups around the world,16 we also conducted a subgroup analysis for Asians vs non-Asians. The Cochrane Collaboration’s Review Manager Software Package version 5.4 (RevMan) and Stata/SE version 15.1 (StataCorp LLC) were used for this meta-analysis and mete-regression.
## Results
We identified 21 full articles for detailed assessment, of which 11 did not meet the inclusion criteria; therefore, the final analysis included 10 randomized clinical trials (eFigure 1 in Supplement 1).2,3,4,5,17,18,19,20,21,22 The characteristics of the included trials are shown in Table 1.2,3,4,5,17,18,19,20,21,22,23 Overall, 40 710 patients (13 752 women [$34\%$]; mean age, 65 years) with stroke or TIA were enrolled. The mean duration of follow-up was 2.8 years (range, 1-4 years). Among the 10 included trials, 6 compared antihypertensive drug(s) vs placebo or no antihypertensive therapy,2,4,17,18,19,20 and 4 compared a lower blood pressure target vs a higher blood pressure target.3,5,21,22 In the Perindopril Protection Against Recurrent Stroke Study (PROGRESS), 2 different active groups (perindopril alone and a combination of peridopril plus indapamide) were compared with placebo, and we analyzed separately perindopril alone vs placebo and perindopril plus indapamide vs placebo.4 Across all trials, the mean baseline SBP was 146 mm Hg and DBP was 85 mm Hg.
**Table 1.**
| Source | Population | Time interval from stroke to randomization | Intensity of BP lowering | Intensity of BP lowering.1 | Sample size, No. | Women, % | Mean, y | Mean, y.1 | Primary outcome of original trial | BP at baseline, mm Hg | Definition of differential BP reduction | Magnitude of differential BP reduction, mm Hg |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Source | Population | Time interval from stroke to randomization | More | Less | Sample size, No. | Women, % | Age | Follow-up duration | Primary outcome of original trial | BP at baseline, mm Hg | Definition of differential BP reduction | Magnitude of differential BP reduction, mm Hg |
| PAST-BP,21 2016, UK | Stroke or TIA | | Lowering SBP to <130 mm Hg | Lowering SBP to <140 mm Hg | 529 | 41 | 72 | 1 | Change in SBP | 143/80 | Mean difference between the 2 groups throughout the study | 2.9/1.6 |
| PRoFESS,2 2008, 35 countries | Ischemic stroke | Within 90 d | Telmisartan, 80 mg, daily | Placebo | 20 332 | 36 | 66 | 2.5 | Recurrent stroke | 144/84 | Mean difference between the 2 groups throughout the study | 3.8/2 |
| TEST,18 1995, Sweden | Stroke or TIA | Within 3 wk | Atenolol, 50 mg, daily | Placebo | 720 | 40 | 70 | 2.6 | Death from any cause, nonfatal MI, and nonfatal stroke | 161/89 | Difference between the 2 groups after 1 mo of treatment | 4/3 |
| PROGRESS (single),4,23 2001, Asia, Australasia, and Europe | Stroke or TIA | Within 5 y | Perindopril, 4 mg, daily | Placebo | 2561 | 32 | 65 | 3.9 | Recurrent stroke | 144/84 | Mean difference between the 2 groups throughout the study | 5/3 |
| DUTCH TIA,17 1993, the Netherlands | Nondisabling ischemic stroke or TIA | Within 3 mo | Atenolol, 50 mg, daily | Placebo | 1473 | 36 | | 2.6 | Death from vascular causes, nonfatal MI, and nonfatal stroke | 158/91 | Difference between the 2 groups at first follow-up after randomization (median at 4 mo) | 5.8/2.9 |
| RESPECT,3 2019, Japan | Stroke | Within 3 y | Lowering BP to <120/80 mm Hg | Lowering BP to <140/90 mm Hg | 1263 | 31 | 67 | 3.9 | Recurrent stroke | 145/85 | Mean difference between the 2 groups throughout the study | 6.5/3.3 |
| PATS,20 2009, China | Stroke or TIA | ≥4 wk | Indapamide, 2.5 mg, daily | Placebo | 5665 | 28 | 60 | 2 | Recurrent stroke | 154/93 | Mean difference between 2 groups after 2 y | 6.8/3.3 |
| PODCAST,22 2017, UK | Stroke | Within previous 3-7 mo | Lowering SBP to <125 mm Hg | Lowering SBP to <140 mm Hg | 83 | 23 | 74 | 2 | ACE-R | 147/82 | Difference between the 2 groups in first 6 mo of treatment | 10.6/5.5 |
| SPS 3,5 2013, North America, Latin America, and Spain | Lacunar infarction | Within 180 d | Lowering SBP to <130 mm Hg | Lowering SBP to 130-149 mm Hg | 3020 | 37 | 63 | 3.7 | Recurrent stroke | 143/79 | Difference at the last study visit | 11/NA |
| PROGRESS (combined),4,23 2001, Asia, Australasia, and Europe | Stroke or TIA | Within 5 y | Perindopril, 4 mg, daily plus indapamide, 2.5 mg, daily | Placebo | 3544 | 29 | 63 | 3.9 | Recurrent stroke | 149/87 | Mean difference between the 2 groups throughout the study | 12/5 |
| Liu et al,19 2005, China | Stroke or TIA | Within 5 y | Perindopril, 4 mg, daily plus indapamide, 2.5 mg, daily | Placebo | 1520 | 29 | 64 | 4 | Recurrent stroke | 145/87 | Mean difference between the 2 groups throughout the study | 14/6 |
The definition of differential blood pressure reduction varied among included trials. Among included trials, 5 adopted mean blood pressure difference between the 2 groups throughout the study2,3,4,19,21 and 1 adopted blood pressure difference after 1 month of treatment,18 1 adopted difference at first follow-up after randomization (median at 4 months),17 1 adopted mean difference after 2 years,20 1 adopted difference at first 6 months of treatment,22 and 1 adopted difference at the last study visit.5 The mean SBP reduction was 6.7 mm Hg and the mean DBP reduction was 2.8 mm Hg between active treatment and comparator groups. The Cochrane risk-of-bias assessment for the included trials is summarized in eFigure 2 in Supplement 1.
## Recurrent Stroke
Pooled results from the fixed-effects model showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of recurrent stroke in patients with stroke or TIA (10 trials; absolute risk, $8.4\%$ vs $10.1\%$; RR, 0.83; $95\%$ CI, 0.78-0.88; $P \leq .001$; number needed to treat [NNT] in 3 years, 58). There was considerable inconsistency among included trials (P for heterogeneity <.001; I2 = $79\%$) (Figure 1).2,3,4,5,17,18,19,20,21,22 Pooled results with the random-effects model obtained similar results.
**Figure 1.:** *Risk of Recurrent StrokeRisk ratio with 95% CI of recurrent stroke with more intensive compared with less intensive blood pressure (BP)–lowering therapy in patients with stroke or transient ischemic attack (TIA). PROGRESS indicates Perindopril Protection Against Recurrent Stroke Study; SBP, systolic blood pressure.*
Meta-regression showed that the magnitude of differential SBP reduction was associated with a lower risk of recurrent stroke in patients with stroke or TIA in a log-linear fashion (regression slope, −0.06; $95\%$ CI, −0.08 to −0.03; $$P \leq .001$$) (Figure 2A).2,3,4,5,17,18,19,20,21,22 A 5-mm Hg greater SBP reduction was associated with an RR of 0.90 and a 10-mm Hg greater SBP reduction with an RR of 0.67. Also, differential magnitude of DBP reduction was associated with a lower risk of recurrent stroke in patients with stroke or TIA in a log-linear fashion (regression slope, −0.17; $95\%$ CI, −0.26 to −0.08; $$P \leq .003$$) (Figure 2B).2,3,4,17,18,19,20,21,22 A 3-mm Hg greater DBP reduction was associated with an RR of 0.84 and a 5-mm Hg greater DBP reduction with an RR of 0.60.
**Figure 2.:** *Meta-Regression of Recurrent StrokeMeta-regression of included trials to explore the association between magnitude of differential systolic blood pressure (SBP) reduction (A) and differential diastolic blood pressure (DBP) reduction (B) vs recurrent stroke rate. RR indicates risk ratio.*
With regard to subgroups of trials with different magnitudes of differential SBP reduction, pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a lower effect magnitude estimand when the amount of differential SBP reduction was 4 mm Hg or lower (3 trials; RR, 0.95; $95\%$ CI, 0.87-1.03)2,18,21 or 5 mm Hg or lower (4 trials; RR, 0.95; $95\%$ CI, 0.88-1.02)2,4,18,21 and associated with a greater effect magnitude when the amount of differential SBP reduction was more than 5 mm Hg (7 trials; RR, 0.67; $95\%$ CI, 0.60-0.74; NNT = 28),3,4,5,17,19,20,22 more than 7 mm Hg (4 trials; RR, 0.62; $95\%$ CI, 0.54-0.70; NNT = 19),4,5,19,22 and more than 11 mm Hg (2 trials; RR, 0.54; $95\%$ CI, 0.46-0.63; NNT = 14) (Table 2).4,19
**Table 2.**
| Differential BP reduction magnitude | No. of events/No. of population (%) | No. of events/No. of population (%).1 | RR (95% CI) | NNT in 3 y |
| --- | --- | --- | --- | --- |
| Differential BP reduction magnitude | More intensive | Less intensive | RR (95% CI) | NNT in 3 y |
| Differential SBP reduction magnitude | Differential SBP reduction magnitude | Differential SBP reduction magnitude | Differential SBP reduction magnitude | Differential SBP reduction magnitude |
| ≤4 mm Hg2,18,21 | 954/10 784 (8.8) | 1006/10 797 (9.3) | 0.95 (0.87-1.03) | |
| ≤5 mm Hg2,4,18,21 | 1111/12 065 (9.2) | 1171/12 077 (9.7) | 0.95 (0.88-1.02) | |
| >5 mm Hg3,4,5,17,19,20,22 | 593/8279 (7.2) | 890/8289 (10.7) | 0.67 (0.60-0.74) | 28 (23-36) |
| >7 mm Hg4,5,19,22 | 343/4074 (8.4) | 557/4093 (13.6) | 0.62 (0.54-0.70) | 19 (16-25) |
| >11 mm Hg4,19 | 217/2532 (8.5) | 402/2532 (15.9) | 0.54 (0.46-0.63) | 14 (12-17) |
| Differential DBP reduction magnitude | Differential DBP reduction magnitude | Differential DBP reduction magnitude | Differential DBP reduction magnitude | Differential DBP reduction magnitude |
| ≤2 mm Hg2,21 | 880/10 412 (8.5) | 937/10 499 (8.9) | 0.94 (0.86-1.03) | |
| ≤3 mm Hg2,4,17,18,21 | 1163/12 797 (9.1) | 1233/12 818 (9.6) | 0.94 (0.87-1.02) | |
| >3 mm Hg3,4,19,20,22 | 416/6046 (6.9) | 676/6029 (11.2) | 0.61 (0.55-0.69) | 23 (20-29) |
| >4 mm Hg4,19,22 | 218/2573 (8.5) | 405/2574 (15.7) | 0.54 (0.46-0.63) | 14 (12-17) |
With regard to subgroups of trials with different magnitudes of differential DBP reduction, pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a lower effect magnitude estimand when the amount of differential DBP reduction was 2 mm Hg or lower (2 trials; RR, 0.94; $95\%$ CI, 0.86-1.03)2,21 or 3 mm Hg or lower (5 trials; RR, 0.94; $95\%$ CI, 0.87-1.02)2,4,17,18,21 and associated with a greater effect magnitude when the amount of differential DBP reduction was more than 3 mm Hg (5 trials; RR, 0.61; $95\%$ CI, 0.55-0.69; NNT = 23)3,4,19,20,22 and more than 4 mm Hg (3 trials; RR, 0.54; $95\%$ CI, 0.46-0.63; NNT = 14) (Table 2).4,19,22
## Major Cardiovascular Events
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of major cardiovascular events in patients with stroke or TIA (9 trials; absolute risk, $12.0\%$ vs $13.7\%$; RR, 0.88; $95\%$ CI, 0.83-0.92; $P \leq .001$; NNT in 3 years, 61). There was substantial heterogeneity among included trials (P for heterogeneity <.001; I2 = $71\%$) (eFigure 3 in Supplement 1).2,3,4,5,17,18,19,20,21,22 Meta-regression showed that the magnitude of differential SBP reduction was associated with a lower risk of major cardiovascular events in patients with stroke or TIA in a log-linear fashion (regression slope, −0.04; $95\%$ CI, −0.07 to −0.01; $$P \leq .01$$) (Figure 3A).2,3,4,5,17,18,19,21,22 A 5-mm Hg greater SBP reduction was associated with an RR of 0.92, and a 10-mm Hg greater SBP reduction with an RR of 0.75. Also, magnitude of differential DBP reduction was associated with a lower risk of major cardiovascular events in patients with stroke or TIA in a log-linear fashion (regression slope, −0.12; $95\%$ CI, −0.25 to 0; $$P \leq .048$$) (Figure 3B).2,3,4,17,18,20,21,22 A 3-mm Hg greater DBP reduction was associated with an RR of 0.88 and a 5-mm Hg greater DBP reduction with an RR of 0.69.
**Figure 3.:** *Meta-Regression of Major Cardiovascular EventsMeta-regression of included trials to explore the association between magnitude of (A) differential systolic blood pressure (SBP) reduction and (B) differential diastolic blood pressure (DBP) reduction vs major cardiovascular events. RR indicates risk ratio.*
## Ischemic Stroke, Hemorrhagic Stroke, and Fatal or Disabling Stroke
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of recurrent ischemic stroke in patients with stroke or TIA (6 trials; absolute risk, $7.5\%$ vs $8.7\%$; RR, 0.87; $95\%$ CI, 0.81-0.94; $P \leq .001$; NNT = 88). There was considerable inconsistency among included trials (P for heterogeneity =.001; I2 = $75\%$) (eFigure 4 in Supplement 1).2,3,4,5,19,22 Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of hemorrhagic stroke in patients with stroke or TIA (6 trials; absolute risk, $0.7\%$ vs $1.3\%$; RR, 0.54; $95\%$ CI, 0.43-0.68; $P \leq .001$; NNT = 167). There was substantial heterogeneity among included trials (P for heterogeneity =.007; I2 = $69\%$) (eFigure 5 in Supplement 1).2,3,4,5,19,22 Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of fatal or disabling stroke in patients with stroke or TIA (6 trials; absolute risk, $3.0\%$ vs $3.9\%$; RR, 0.76; $95\%$ CI, 0.64-0.89; $P \leq .001$; NNT = 107). Heterogeneity was not important among included trials (P for heterogeneity =.52; I2 = $0\%$) (eFigure 6 in Supplement 1).4,5,17,18,20,22
## Myocardial Infarction
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was not associated with a significantly reduced risk of myocardial infarction in patients with stroke or TIA (10 trials; absolute risk, $1.8\%$ vs $2.0\%$; RR, 0.89; $95\%$ CI, 0.78-1.03; $$P \leq .11$$). Heterogeneity was not important among included trials (P for heterogeneity =.31; I2 = $15\%$) (eFigure 7 in Supplement 1).2,3,4,5,17,18,19,20,21,22
## Death From Cardiovascular Causes
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a reduced risk of death from cardiovascular causes in patients with stroke or TIA (9 trials; absolute risk, $3.2\%$ vs $3.7\%$; RR, 0.86; $95\%$ CI, 0.78-0.96; $$P \leq .006$$; NNT = 193). Heterogeneity was not important among included trials (P for heterogeneity =.46; I2 = $0\%$) (eFigure 8 in Supplement 1).2,4,5,17,18,19,20,21,22
## Death From Any Cause
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was not associated with a significantly reduced risk of death from any cause in patients with stroke or TIA (10 trials; absolute risk, $7.4\%$ vs $7.6\%$; RR, 0.97; $95\%$ CI, 0.91-1.04; $$P \leq .42$$). Heterogeneity was not important among included trials (P for heterogeneity =.31; I2 = $14\%$) (eFigure 9 in Supplement 1).2,3,4,5,17,18,19,20,21,22
## Heart Failure
Pooled results showed that more intensive compared with less intensive blood pressure–lowering therapy was not associated with a reduced risk of heart failure in patients with stroke or TIA (2 trials; absolute risk, $1.2\%$ vs $1.1\%$; RR, 1.05; $95\%$ CI, 0.82-1.35; $$P \leq .68$$). Heterogeneity was not important among included trials (P for heterogeneity =.53; I2 = $0\%$) (eFigure 10 in Supplement 1).2,3 Summaries of evidence for the primary and secondary outcomes are presented in the eTable in Supplement 1.
## Sensitivity Tests
Sensitivity testing removing each individual trial from the meta-analysis one at a time yielded pooled results similar to the overall pooled estimates of the primary outcome. Sensitivity testing restricting analysis to trials with recurrent stroke being the primary outcome in the original trial design showed that more intensive treatment compared with less intensive or no treatment was associated with a reduced risk of recurrent stroke in patients with stroke or TIA (6 trials; RR, 0.82; $95\%$ CI, 0.77-0.88; $P \leq .001$) (eFigure 11 in Supplement 1).2,3,4,5,19,20,22
## Subgroup Analysis
Subgroup analyses of included trials showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a greater risk reduction of recurrent stroke in [1] trials with patients enrolled within 3 to 5 years from stroke (RR, 0.67; $95\%$ CI, 0.59-0.75) than patients enrolled within 6 months from stroke (RR, 0.93; $95\%$ CI, 0.86-1.00; P for interaction <.001; I2 = $95\%$) (eFigure 12 in Supplement 1); [2] trials with study follow-up of 3 years or more (RR, 0.75; $95\%$ CI, 0.68-0.83) than in trials with study duration less than 3 years (RR, 0.88; $95\%$ CI 0.81-0.95; P for interaction =.01; I2 = $85\%$) (eFigure 13 in Supplement 1); [3] trials enrolling an Asian population (RR, 0.63; $95\%$ CI, 0.54-0.73) than in trials enrolling a mostly non-Asian population (RR, 0.88; $95\%$ CI 0.82-0.94; P for interaction <.001; I2 = $94\%$) (eFigure 14 in Supplement 1); and [4] trials using an ACE inhibitor plus diuretics (RR, 0.54; $95\%$ CI, 0.46-0.63) than in trials using either ACE inhibitors, angiotensin receptor blockers, β-blockers, diuretics alone (P for interaction <.001; I2 = $91\%$) (eFigure 15 in Supplement 1). In contrast, no significant heterogeneity was found between trials with [1] mean baseline SBP of 140 to 149 mm Hg vs 150 mm Hg or higher; [2] achieved SBP in intensive treated group of less than 130 mm Hg vs 130 to 139 mm Hg vs 140 mm Hg or higher; [3] achieved SBP in less intensive or no therapy of less than 130 mm Hg vs 130 to 139 mm Hg vs 140 mm Hg or higher; [4] entry event ischemic vs hemorrhagic stroke; [5] trial sample size less than 3000 vs 3000 or more; [6] study design (antihypertensive drugs vs placebo and a lower blood pressure target vs a higher blood pressure target); and [7] definition of differential blood pressure reduction (mean difference throughout the studies vs other definitions) (eFigures 16-22 in Supplement 1).
## Publication Bias
There was no obvious publication bias assessed by the trim-and-fill method for the primary outcome (eFigure 23 in Supplement 1).
## Discussion
The present meta-analysis, comprising 10 randomized clinical trials for blood pressure–lowering therapy enrolling over 40 000 individuals with a history of stroke or TIA, revealed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a risk reduction of recurrent stroke. Further, meta-regression suggested the magnitudes of both differential SBP and DBP reduction were monotonically associated with a lower risk of recurrent stroke. More intensive compared with less intensive blood pressure–lowering therapy was also associated with a risk reduction of major cardiovascular events, ischemic stroke, hemorrhagic stroke, fatal or disabling stroke, and death from cardiovascular causes in patients with stroke or TIA.
High statistical heterogeneity as well as clinical heterogeneity was found in the forest plots in the primary outcome (ie, recurrent stroke) and the lead secondary outcome (ie, major cardiovascular events). In the forest plot of the recurrent stroke, the larger of the differential SBP reduction, the larger of the reduction of recurrent stroke was found visually. In the meta-regression analyses, magnitude of differential SBP and DBP reduction were significantly associated with a lower risk of recurrent stroke in patients with stroke or TIA. Therefore, heterogeneity might be primarily driven by the various magnitude of differential blood pressure lowering among included trials. Similar findings were noted in the major cardiovascular events.
The current study was distinct from the study by Katsanos et al8 in several aspects. First, we excluded trials comparing one antihypertensive drug vs another antihypertensive drug (eg, eprosartan vs nitrendipine)24 or achieved SBP higher in more intensive blood pressure–lowering group than less intensive blood pressure–lowering group after treatment.25 This approach was taken to be in line with the objective of our study. Second, the study by Katsanos et al8 only pooled data from trials with antihypertensive therapy vs placebo, whereas we pooled data from trials with antihypertensive therapy vs placebo and a lower blood pressure target vs a higher blood pressure target. The latter approach includes all relevant trials of blood pressure–lowering therapy for secondary stroke prevention. Third, the study by Katsanos et al8 adopted achieved blood pressure levels for meta-regression analyses, whereas this study adopted magnitude of differential blood pressure reduction for meta-regression analyses and showed that the larger the magnitude of differential blood pressure reduction, the larger the reduction of recurrent stroke and major cardiovascular events.
The current study is consonant with and extends prior investigations. In the current study, greater achieved differential blood pressure lowering was beneficial even among patients at the lower end of the blood pressure spectrum. The degree of separation between more intensive and less intensive SBP lowering was associated with a uniform relative risk reduction regardless of whether patients with less intensive treatment SBP levels of 140 mm Hg or higher, 130-139 mm Hg, or less than 130 mm Hg. This finding accords with analyses of intensive blood pressure lowering in stroke qualifying event trials, including PROGRESS and the Warfarin-Aspirin Symptomatic Intracranial Disease (WASID) trial.26,27 It accords as well with studies in broader cardiovascular event qualifying event trials, including the Action to Control Cardiovascular Risk in Diabetes Blood Pressure (ACCORD BP) trial,28 the Systolic Blood Pressure Intervention Trial (SPRINT),29,30 and a trial of older Asian individuals with hypertension.31 Furthermore, physiologically, intensive (SBP target <120 mm Hg) compared with standard (SBP target <140 mm Hg) blood pressure–lowering therapy was associated with increased, not decreased, cerebral perfusion, in participants with a history of cardiovascular disease.32 Taken together, this study does not refute the current practice guidelines recommending blood pressure lowering to an SBP target of less than 130 mm Hg as a major target for patients with stroke or TIA.
Among patients with high-grade vertebrobasilar or other intracranial stenoses, an inverse association of SBP with recurrent stroke risk was found in some studies.33,34 However, post hoc analyses of several trials showed that treating SBP to a target of less than 140 mm Hg in patients with severe intracranial stenosis might be beneficial.27,35,36,37 Since the risk-benefit profile of more aggressive blood pressure lowering is not consistent among studies, secondary prevention guidelines recommend a less aggressive SBP target (<140 mm Hg) in patients with $50\%$ to $99\%$ stenosis of a major intracranial artery.38 *Although this* meta-analysis excluded trials with blood pressure–lowering therapy used in acute stage of stroke or TIA, the time interval from stroke to randomization still varied among included trials. A subgroup analysis suggested that more intensive compared with less intensive blood pressure–lowering therapy was associated with substantial risk reduction of recurrent stroke in trials with patients enrolled within 3 to 5 years from stroke or TIA but only associated with modest risk reduction of recurrent stroke in trials with patients enrolled within 6 months from stroke or TIA. This finding accords with observational studies that have suggested that aggressive blood pressure lowering might not be beneficial within the first 6 months following the index ischemic stroke event.39,40 These findings support a strategy of a moderate blood pressure–lowering target during the first 6 months following stroke or TIA followed by a more aggressive target chronically.
A subgroup analysis also suggested that more intensive compared with less intensive blood pressure–lowering therapy was associated with a larger risk reduction of recurrent stroke in trials with study duration of 3 years or more vs less than 3 years. It is conceivable that a longer duration of high blood pressure levels may cause more profound damage of vessels thereby leading to a higher likelihood of recurrent stroke in patients with stroke or TIA. Also, we found that more intensive compared with less intensive blood pressure–lowering therapy was associated with a larger risk reduction of recurrent stroke in Asian populations than non-Asian populations. *In* general, the ratio of hemorrhagic stroke to ischemic stroke is higher in Asian than in non-Asian populations,41 and this meta-analysis showed that more intensive compared with less intensive blood pressure–lowering therapy was associated with a substantial relative risk reduction in hemorrhagic stroke (RR, 0.54; $95\%$ CI, 0.43-0.68). Also, the risk of recurrent ischemic and hemorrhagic stroke was greater in Asian population with high blood pressure compared with several other groups around the world.16 Therefore, it is conceivable that more intensive blood pressure–lowering treatment might be especially beneficial for Asian populations.
## Limitations
There are several limitations to this study. First, although it was conceivable that some patients in the included trials had baseline SBP less than 140 mm Hg, none of included trials had mean baseline SBP less than 140 mm Hg. Granular analysis related to baseline BP levels could not be conducted because this study was a trial-level meta-analysis, rather than an individual, patient-level pooled analysis. Second, the purpose of some included trials was not to evaluate association between blood pressure lowering and risk of recurrent stroke,17,18,21,22 and in such settings, there might be a higher chance that stroke events were not well recorded. Still, a sensitivity test restricting analysis within trials with recurrent stroke reported as the primary outcome obtained similar results. Third, although we excluded trials before 1980, the included trials reflect a long period of clinical practice during which concomitant treatments for stroke prevention further evolved.1 However, the consistency in benefit of blood pressure lowering suggests substantial benefit on top of a wide range of background therapies. Fourth, the individual studies mostly have wide CIs and many of the pooled results of secondary outcomes are driven by a single study with a large weight. Fifth, more than half of the participants are contributed by the Prevention Regimen for Effectively Avoiding Second Strokes (PRoFESS)2 and PROGRESS4 trials. Still, sensitivity testing removing either trial yielded pooled results similar to the overall pooled estimates of the primary outcome.
## Conclusions
This meta-analysis and meta-regression of randomized clinical trials for blood pressure–lowering therapy found that more intensive compared with less intensive blood pressure–lowering therapy may be associated with a reduction of recurrent stroke; further, the larger the magnitude of differential blood pressure reduction, the larger the reduction of cerebrovascular events. More intensive blood pressure–lowering therapy may be also associated with a risk reduction of major cardiovascular events, ischemic stroke, hemorrhagic stroke, fatal or disabling stroke, and death from cardiovascular causes in patients with stroke or TIA. These results might support the use of more intensive blood pressure reduction for secondary stroke prevention chronically.
## References
1. Hong KS, Yegiaian S, Lee M, Lee J, Saver JL. **Declining stroke and vascular event recurrence rates in secondary prevention trials over the past 50 years and consequences for current trial design**. *Circulation* (2011.0) **123** 2111-2119. DOI: 10.1161/CIRCULATIONAHA.109.934786
2. Yusuf S, Diener HC, Sacco RL. **Telmisartan to prevent recurrent stroke and cardiovascular events**. *N Engl J Med* (2008.0) **359** 1225-1237. DOI: 10.1056/NEJMoa0804593
3. Kitagawa K, Yamamoto Y, Arima H. **Effect of standard vs intensive blood pressure control on the risk of recurrent stroke: a randomized clinical trial and meta-analysis**. *JAMA Neurol* (2019.0) **76** 1309-1318. DOI: 10.1001/jamaneurol.2019.2167
4. **Randomised trial of a perindopril-based blood-pressure-lowering regimen among 6,105 individuals with previous stroke or transient ischaemic attack**. *Lancet* (2001.0) **358** 1033-1041. DOI: 10.1016/S0140-6736(01)06178-5
5. Benavente OR, Coffey CS, Conwit R. **Blood-pressure targets in patients with recent lacunar stroke: the SPS3 randomised trial**. *Lancet* (2013.0) **382** 507-515. DOI: 10.1016/S0140-6736(13)60852-1
6. 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
7. Zonneveld TP, Richard E, Vergouwen MD. **Blood pressure-lowering treatment for preventing recurrent stroke, major vascular events, and dementia in patients with a history of stroke or transient ischaemic attack**. *Cochrane Database Syst Rev* (2018.0) **7**. DOI: 10.1002/14651858.CD007858.pub2
8. Katsanos AH, Filippatou A, Manios E. **Blood pressure reduction and secondary stroke prevention: a systematic review and metaregression analysis of randomized clinical trials**. *Hypertension* (2017.0) **69** 171-179. DOI: 10.1161/HYPERTENSIONAHA.116.08485
9. Boncoraglio GB, Del Giovane C, Tramacere I. **Antihypertensive drugs for secondary prevention after ischemic stroke or transient ischemic attack: a systematic review and meta-analysis**. *Stroke* (2021.0) **52** 1974-1982. DOI: 10.1161/STROKEAHA.120.031945
10. 10Cochrane Handbook for Systematic Reviews of Interventions. Cochrane Training. Accessed October 18, 2022. https://training.cochrane.org/handbook
11. 11Higgins JPT, Green S, eds. Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0. The Cochrane Collaboration; 2011.. *Cochrane Handbook for Systematic Reviews of Interventions, version 5.1.0* (2011.0)
12. Higgins JP, Thompson SG, Deeks JJ, Altman DG. **Measuring inconsistency in meta-analyses**. *BMJ* (2003.0) **327** 557-560. DOI: 10.1136/bmj.327.7414.557
13. Duval S, Tweedie R. **Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis**. *Biometrics* (2000.0) **56** 455-463. DOI: 10.1111/j.0006-341X.2000.00455.x
14. Guyatt GH, Oxman AD, Vist GE. **GRADE: an emerging consensus on rating quality of evidence and strength of recommendations**. *BMJ* (2008.0) **336** 924-926. DOI: 10.1136/bmj.39489.470347.AD
15. Meader N, King K, Llewellyn A. **A checklist designed to aid consistency and reproducibility of GRADE assessments: development and pilot validation**. *Syst Rev* (2014.0) **3** 82. DOI: 10.1186/2046-4053-3-82
16. Estol CJ, Bath PM, Gorelick PB, Cotton D, Martin RH. **Differences in ischemic and hemorrhagic recurrence rates among race-ethnic groups in the PRoFESS secondary stroke prevention trial**. *Int J Stroke* (2014.0) **9** 43-47. DOI: 10.1111/ijs.12269
17. **Trial of secondary prevention with atenolol after transient ischemic attack or nondisabling ischemic stroke**. *Stroke* (1993.0) **24** 543-548. DOI: 10.1161/01.STR.24.4.543
18. Eriksson S, Olofsson B, Wesley P. **Atenolol in secondary prevention after stroke**. *Cerebrovasc Dis* (1995.0) **5** 21-25. DOI: 10.1159/000107813
19. Liu LS, Gong LS, Wang W. **[Effects of blood pressure lowering treatment on stroke recurrence in patients with cerebrovascular diseases-a large-scale, randomized, placebo controlled trial]**. *Zhonghua Xin Xue Guan Bing Za Zhi* (2005.0) **33** 613-617. PMID: 16080808
20. Liu L, Wang Z, Gong L. **Blood pressure reduction for the secondary prevention of stroke: a Chinese trial and a systematic review of the literature**. *Hypertens Res* (2009.0) **32** 1032-1040. DOI: 10.1038/hr.2009.139
21. Mant J, McManus RJ, Roalfe A. **Different systolic blood pressure targets for people with history of stroke or transient ischaemic attack: PAST-BP (Prevention After Stroke–Blood Pressure) randomised controlled trial**. *BMJ* (2016.0) **352** i708. DOI: 10.1136/bmj.i708
22. Bath PM, Scutt P, Blackburn DJ. **Intensive versus guideline blood pressure and lipid lowering in patients with previous stroke: main results from the pilot ‘Prevention of Decline in Cognition after Stroke Trial’ (PODCAST) randomised controlled trial**. *PLoS One* (2017.0) **12**. DOI: 10.1371/journal.pone.0164608
23. Chapman N, Huxley R, Anderson C. **Effects of a perindopril-based blood pressure-lowering regimen on the risk of recurrent stroke according to stroke subtype and medical history: the PROGRESS Trial**. *Stroke* (2004.0) **35** 116-121. DOI: 10.1161/01.STR.0000106480.76217.6F
24. Schrader J, Lüders S, Kulschewski A. **Morbidity and mortality after stroke, eprosartan compared with nitrendipine for secondary prevention: principal results of a prospective randomized controlled study (MOSES)**. *Stroke* (2005.0) **36** 1218-1226. DOI: 10.1161/01.STR.0000166048.35740.a9
25. Trenkwalder P, Elmfeldt D, Hofman A. **The Study on COgnition and Prognosis in the Elderly (SCOPE): major CV events and stroke in subgroups of patients**. *Blood Press* (2005.0) **14** 31-37. DOI: 10.1080/08037050510008823
26. Arima H, Chalmers J, Woodward M. **Lower target blood pressures are safe and effective for the prevention of recurrent stroke: the PROGRESS trial**. *J Hypertens* (2006.0) **24** 1201-1208. DOI: 10.1097/01.hjh.0000226212.34055.86
27. Turan TN, Cotsonis G, Lynn MJ, Chaturvedi S, Chimowitz M. **Relationship between blood pressure and stroke recurrence in patients with intracranial arterial stenosis**. *Circulation* (2007.0) **115** 2969-2975. DOI: 10.1161/CIRCULATIONAHA.106.622464
28. Cushman WC, Evans GW, Byington RP. **Effects of intensive blood-pressure control in type 2 diabetes mellitus**. *N Engl J Med* (2010.0) **362** 1575-1585. DOI: 10.1056/NEJMoa1001286
29. Wright JT, Williamson JD, Whelton PK. **A randomized trial of intensive versus standard blood-pressure control**. *N Engl J Med* (2015.0) **373** 2103-2116. DOI: 10.1056/NEJMoa1511939
30. Perkovic V, Rodgers A. **Redefining blood-pressure targets–SPRINT starts the marathon**. *N Engl J Med* (2015.0) **373** 2175-2178. DOI: 10.1056/NEJMe1513301
31. Zhang W, Zhang S, Deng Y. **Trial of Intensive blood-pressure control in older patients with hypertension**. *N Engl J Med* (2021.0) **385** 1268-1279. DOI: 10.1056/NEJMoa2111437
32. Dolui S, Detre JA, Gaussoin SA. **Association of intensive vs standard blood pressure control with cerebral blood flow: secondary analysis of the SPRINT MIND randomized clinical trial**. *JAMA Neurol* (2022.0) **79** 380-389. DOI: 10.1001/jamaneurol.2022.0074
33. Amin-Hanjani S, Turan TN, Du X. **Higher stroke risk with lower blood pressure in hemodynamic vertebrobasilar disease: analysis from the VERiTAS Study**. *J Stroke Cerebrovasc Dis* (2017.0) **26** 403-410. DOI: 10.1016/j.jstrokecerebrovasdis.2016.09.044
34. Park JM, Kim BJ, Kwon SU. **Intensive blood pressure control may not be safe in subacute ischemic stroke by intracranial atherosclerosis: a result of randomized trial**. *J Hypertens* (2018.0) **36** 1936-1941. DOI: 10.1097/HJH.0000000000001784
35. Turan TN, Nizam A, Lynn MJ. **Relationship between risk factor control and vascular events in the SAMMPRIS trial**. *Neurology* (2017.0) **88** 379-385. DOI: 10.1212/WNL.0000000000003534
36. Yu DD, Pu YH, Pan YS. **High blood pressure increases the risk of poor outcome at discharge and 12-month follow-up in patients with symptomatic intracranial large artery stenosis and occlusions: subgroup analysis of the CICAS study**. *CNS Neurosci Ther* (2015.0) **21** 530-535. DOI: 10.1111/cns.12400
37. Chaturvedi S, Turan TN, Lynn MJ. **Risk factor status and vascular events in patients with symptomatic intracranial stenosis**. *Neurology* (2007.0) **69** 2063-2068. DOI: 10.1212/01.wnl.0000279338.18776.26
38. Kleindorfer DO, Towfighi A, Chaturvedi S. **2021 Guideline for the prevention of stroke in patients with stroke and transient ischemic attack: a guideline from the American Heart Association/American Stroke Association**. *Stroke* (2021.0) **52** e364-e467. DOI: 10.1161/STR.0000000000000375
39. Ovbiagele B, Diener HC, Yusuf S. **Level of systolic blood pressure within the normal range and risk of recurrent stroke**. *JAMA* (2011.0) **306** 2137-2144. DOI: 10.1001/jama.2011.1650
40. Ovbiagele B. **Low-normal systolic blood pressure and secondary stroke risk**. *J Stroke Cerebrovasc Dis* (2013.0) **22** 633-638. DOI: 10.1016/j.jstrokecerebrovasdis.2011.12.003
41. Klatsky AL, Friedman GD, Sidney S, Kipp H, Kubo A, Armstrong MA. **Risk of hemorrhagic stroke in Asian American ethnic groups**. *Neuroepidemiology* (2005.0) **25** 26-31. DOI: 10.1159/000085310
|
---
title: Anti-obesity effect of collagen peptides obtained from Diplulmaris antarctica,
a jellyfish of the Antarctic region
authors:
- Nataliia Raksha
- Tetiana Halenova
- Tetiana Vovk
- Olexandra Kostyuk
- Tatyana Synelnyk
- Tatyana Andriichuk
- Tetiana Maievska
- Olexiy Savchuk
- Ludmila Ostapchenko
journal: Croatian Medical Journal
year: 2023
pmcid: PMC10028559
doi: 10.3325/cmj.2023.64.21
license: CC BY 2.5
---
# Anti-obesity effect of collagen peptides obtained from Diplulmaris antarctica, a jellyfish of the Antarctic region
## Body
Conclusion Collagen peptides obtained from *Diplulmaris antarctica* can be used to prevent and treat obesity caused by a high-calorie diet and pathologies associated with increased oxidative stress. Given the obtained results and the abundance of *Diplulmaris antarctica* in the Antarctic region, this species can be considered a sustainable source of collagen and its derivatives.
Obesity increases the risk of many health disorders, including cardiovascular disease, kidney disease, diabetes mellitus, hypertension, and cancer [1-3]. Although the existing medications for obesity treatment [4,5] are effective, their safety and side effects remain a concern. Therefore, there is a need to find natural substances to manage obesity with minimal side effects.
Peptides are currently being actively studied as possible alternatives to drugs for the prevention and treatment of various obesity complications. Endogenous and bioactive endogenous peptides possess a wide range of anti-oxidant, anti-inflammatory, anti-microbial, anti-ulcer, lipid-lowering, wound healing, and anti-skin-aging effects [6-8]. In addition, peptides are a part of a complex system of energy balance regulation, which, when impaired, is one of the causes of obesity. Endogenous orexigenic and anorexigenic peptides responsible for regulating the onset of hunger, satiety, adipocyte metabolism, and inflammation have been identified [9,10].
A promising source for obtaining peptides is collagen. Collagen and its fragments are included in pharmaceutical preparations due to their excellent biocompatibility, biodegradability, and low antigenicity. Collagen peptides affect glucose tolerance and insulin sensitivity in overweight individuals, modulate the immune status, and have a hypocholesterolemic effect [11-13]. Although sources of collagen are available and often quite cheap, frequent outbreaks of infectious diseases among land animals necessitate the search for alternative sources of proteins. Additionally, the use of molecules from warm-blooded animals may be unacceptable to individual patients due to their religious beliefs or lifestyle. In this context, hydrobionts, which make up about half of the world's biodiversity, can be an inexhaustible source of collagen. The rapid growth of the jellyfish population around the world seriously affects ecosystems and human activities in marine areas. On the other hand, for hundreds of years jellyfish has been an important food source in many countries [14]. There is growing interest in biologically active compounds isolated from jellyfish, which were found to exhibit antioxidant [15], anti-microbial [16], anti-cancer [17], as well as immune-modulatory and wound healing properties [18]. Jellyfish is rich in collagenous protein, which makes this species a promising source of collagen and collagen-related products. Despite the conservative structure, the collagen from hydrobionts has a lower content of alanine, glycine, and proline and a higher content of arginine, aspartic acid, threonine, tyrosine, cysteine, and methionine than the collagen isolated from mammals [19,20]. Proteins isolated from organisms living in atypical habitats (low or high temperatures, high pressure, low illumination) may have certain structural and functional features (conformational flexibility, high catalytic efficiency, thermolability, or thermostability) [21]. This allows us to suggest that the hydrobionts of the Antarctic region may contain compounds with new or more pronounced properties. Diplulmaris antarctica is abundant in the Antarctic region and may represent a sustainable source of collagen. Although the effects of collagen or total protein hydrolysates from jellyfish have already been investigated, the research using collagen peptides (less than 5 kDa) remains limited. In the current study, we evaluated the efficacy of collagen peptides obtained from *Diplulmaris antarctica* in preventing diet-induced obesity.
## Abstract
### Aim
To investigate the ability of collagen peptides derived from a jellyfish of the Antarctic region (Diplulmaris antarctica) to prevent the development of obesity in rats fed a high-calorie diet.
### Methods
Collagen peptides were produced by pepsin hydrolysis of jellyfish-derived collagen. The purity of collagen and collagen peptides was confirmed by SDS-polyacrylamide gel electrophoresis. Rats were fed a high-calorie diet for ten weeks and were simultaneously orally administered collagen peptides (1 g per 1 kg of body weight every other day) starting from the fourth week. Body mass index (BMI), body weight gain, selected nutritional parameters, the key parameters associated with insulin resistance, and the level of oxidative stress markers were assessed.
### Results
Compared with untreated obese rats, rats treated with hydrolyzed jellyfish collagen peptides had a decreased body weight gain and body mass index. They also had a decreased level of fasting blood glucose, glycated hemoglobin, insulin, lipid peroxidation products (conjugated dienes, Schiff bases), and oxidatively modified proteins, as well as a restored activity of superoxide dismutase.
## Preparation of collagen peptides
The jellyfish was caught near the island of Galindez (65°15′ S, 64°15′ W) in the Argentine Islands archipelago by the Ukrainian Antarctic expeditions. The whole jellyfish specimens were individually frozen in liquid nitrogen and stored at -80 °C to prevent enzyme deterioration. The samples were transported to the laboratory frozen. The jellyfish was authenticated by the Department of Zoology and Ecology of Educational and Scientific Center “Institute of Biology and Medicine” of Taras Shevchenko National University of Kyiv, Ukraine.
The jellyfish samples were washed at least three times with tap water and once with ultra-purified water. The clean jellyfish mass was homogenized with a blender. Collagen was extracted with a subsequent addition of NaCl and acetic acid. Briefly, dry NaCl was added to the homogenate to a final concentration of 1 M, and the mixture was constantly stirred for 48 hours. The mixture was centrifuged (10 000 g, 30 min), the pellet containing collagen was solubilized in 0.5 M acetic acid, and the supernatant was salted-out by adding dry NaCl to a final concentration of 1 M. The mixture was stirred for 24 h. The resulting precipitate was collected by centrifugation (10 000 g, 30 min) and then dissolved in 0.5 M acetic acid. The samples of dissolved collagen were pooled. After dialysis against pure water, the samples were lyophilized in a freeze dryer and used to obtain collagen peptides. For this purpose, lyophilized collagen (1 g) was suspended in 20 mL of 0.2 M acetic acid and mixed with pepsin (~ 2500 units mg/protein, Sigma-Aldrich, St. Louis, MO, USA) at an enzyme:substrate ratio of 1: 100 (w/w). The mixture was stirred for 8 h at 37 °C. Hydrolysis was stopped by heating the sample at 95 °C for 10 min in a temperature-controlled water bath shaker. The samples were then centrifuged at 10 000 g for 15 min. Collagen peptides with a molecular weight of less than 10 kDa were isolated by ultrafiltration with a Pierce Protein Concentrator PES, 10K MWCO (Thermo Fisher Scientific, Waltham, MA, USA). The molecular weight of the obtained collagen peptides was assessed with SDS-polyacrylamide gel electrophoresis according to the Laemmli method [22]. The obtained fraction of collagen peptides was further lyophilized. The quality control of peptide preparation is described in detail in the Supplemental material.
## Animals and experimental design
Thirty Wistar male rats (5 weeks old; an initial weight 90 ± 5 g) were used. All animal experiments complied with the principles of the Council of Europe Convention for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes. The study was approved by the Ethics Committee of Taras Shevchenko National University of Kyiv (protocol No. 1). The experiment was started one week after the acclimatization of the animals in the vivarium of Taras Shevchenko National University of Kyiv at constant temperature (22 ± 3 °C), humidity (60 ± $5\%$), and illumination (12 h light/12 h dark cycle). Standard food for rodents and water were provided ad libitum. On the eighth day, the animals were divided into the control (10 rats per group) and experimental (20 rats per group) group by simple randomization. The animals were kept in polypropylene cages (595 × 380 × 200 mm, floor area 1820 cm2); 5 rats in a cage. The control group received a basal rodent diet for the next 10 weeks. Rats in the experimental group received a high-calorie diet consisting of a standard meal ($60\%$), lard ($10\%$), eggs ($10\%$), sugar ($9\%$), peanuts ($5\%$), dry milk ($5\%$), and sunflower oil ($1\%$) [23]. All animals received food and water ad libitum. After four weeks, the experimental animals were randomly divided into two groups (10 rats per group). The rats in the first group continued to receive a high-calorie diet. The rats in the second group also received a high-calorie diet, but they were also intragastrically administered collagen peptides (1 g per kg of body weight) in $0.9\%$ NaCl every other day for the next six weeks. The control group and the first experimental group were administered an equal volume of $0.9\%$ NaCl. Food and water intake were measured daily at a fixed time. Body mass index (BMI) was calculated at the end of the experiment. At the end of the 10th week, the animals were not fed overnight and were then sacrificed. The serum was obtained by centrifugation (1000 g, 30 min) of blood samples preincubated at 37 °C for 30 min.
## Biochemical analysis
Glucose concentration was assessed with a Glucophot-II glucometer (Norma, Kyiv, Ukraine). Glycated hemoglobin was measured with an assay kit (Pliva-Lachema Diagnostika, Brno, Czech Republic). The insulin content in serum was measured by enzyme-linked immunosorbent assay according to a previously published protocol [24]. The level of lipid peroxidation products was determined in reaction with a thiobarbituric acid reagent and expressed as thiobarbituric acid reactive substances (TBARS) [25]. The level of Schiff bases and conjugated dienes was determined according to the method by Fletcher et al [26]. The level of oxidatively modified proteins was assessed spectrophotometrically in the reaction with 2,4-dinitrophenylhydrazine [25]. Superoxide dismutase activity (SOD) was determined spectrophotometrically based on the ability of the enzyme to inhibit the autoxidation of adrenaline [27]. The protein concentration was measured according to the method by Bradford [28].
## Statistical analysis
Data are expressed as mean ± standard deviation (SD). The normality of distribution was assessed with a Kolmogorov-Smirnov test. The significance of differences between the groups was assessed with a one-way analysis of variance (ANOVA) with a Tukey post hoc test. The level of significance was $P \leq 0.05.$ *Statistical analysis* was performed with STATISTICA, version 8.0 (StatSoft, Tulsa, OK, USA).
## RESULTS
Effect of collagen peptides on body mass index, body weight gain, and some nutritional parameters To determine whether obesity was present in rats fed a high-calorie diet, BMI and body weight gain were assessed in all experimental groups. The BMI of the control rats was 0.62 ± 0.05 g/cm2 (Table 1), which was within the reference range for rats of this age [22]. In rats fed high-calorie diet, BMI was 0.78 ± 0.05 g/cm2, which was 1.25 times higher than in control rats. The group treated with collagen peptides also experienced an increase in BMI, but it was significantly lower than that in obese rats.
The weight gain in the control group was $107.5\%$, among rats with HCD obesity it was $165.5\%$, and among rats administered collagen peptides it was significantly lower compared with the obese group ($132\%$; $P \leq 0.05$) (Table 1).
To elucidate the possible mechanisms underlying the action of collagen peptides, we analyzed whether treatment with collagen fragments affected food consumption and water intake. The control group consumed an average of 23.5 ± 2.4 g of standard food per day. Obese rats consumed approximately 30.5 ± 2.4 g of high-calorie food per day. Rats administered collagen peptides consumed 20.5 ± 2.2 g of food per day, which was significantly lower ($P \leq 0.05$) compared with obese animals. No significant changes in water consumption were found between all experimental groups, with only a tendency toward a decreased water intake by obese animals.
## The effect of collagen peptides on parameters associated with insulin resistance
In order to evaluate the effect of collagen peptides on insulin resistance development, we assessed the levels of glucose, insulin, and glycated hemoglobin (HbA1c). The concentration of blood glucose decreased significantly ($P \leq 0.05$) in rats treated with collagen peptides (Table 2) compared with the value in obese rats. Insulin level increased 1.6 times in obese rats, but returned to the reference range in rats that received collagen peptides. HbA1c content substantially increased among obese rats – this value was 0.85 ± 0.06 μmol fructose/g hemoglobin compared with 0.38 ± 0.06 μmol fructose/g hemoglobin in control animals. The administration of collagen peptides significantly decreased HbA1c level ($P \leq 0.05$) compared with the level in obese rats.
## Effect of collagen peptides on markers of oxidative stress
To elucidate the mechanism of action of collagen peptides, their effect on oxidative status was determined. In obese rats, the level of lipid peroxidation products increased – the levels of conjugated dienes, TBARS, and Schiff bases exceeded the corresponding control values (Table 3). In addition, the level of oxidatively modified proteins significantly increased (4.5 times) compared with the control group. Compared with obese rats, rats administered collagen peptides had significantly decreased concentrations of conjugated dienes and Schiff bases. In contrast, TBARS level in rats administered collagen peptides was slightly higher than in obese animals (0.81 ± 0.004 nmol/mg protein vs 0.71 ± 0.003 nmol/mg protein). The activity of superoxide dismutase was significantly reduced ($P \leq 0.05$) in obese animals and significantly increased ($P \leq 0.05$) in rats treated with collagen peptides compared with the control group and obese rats.
## DISCUSSION
This study showed that an intragastric administration of jellyfish-derived collagen peptides for 6 weeks slowed down weight gain, reduced several parameters associated with insulin resistance, and reduced oxidative stress compared with the value in obese rats.
In recent years, there has been growing interest in the use of bioactive peptides as alternative agents for the treatment of metabolic disorders [29-31]. Our previous study revealed the weight-lowering effect of collagen fragments prepared from fish scales [32].
In this study, animals that received a high-calorie diet and collagen peptides from *Dipulmaris antarctica* had a BMI similar to the control value. A similar effect of jellyfish collagen hydrolysate on body weight in mice fed a high-fat diet was previously demonstrated [33]. However, collagen peptides may have a number of advantages over collagen hydrolysates from both a pharmacological and biotechnological point of view. Peptides are more stable than proteins and can be stored for a long time without a loss of biological activity. In addition, they exhibit a more pronounced activity; peptide preparations are well absorbed by various routes of administration and do not provoke an immune response.
Given that most peptide-containing products are consumed as functional foods or supplements, in the current study, collagen peptides were administered intragastrically to mimic the administration route in humans. Regardless of the etiology, obesity development is accompanied by impaired control of appetite, which leads to excessive food intake. The decrease in both BMI and body weight gain of rats treated with collagen peptides compared with obese rats may be explained by their reduced appetite. This may indicate the ability of collagen peptides to affect satiety. The exact mechanism of action of collagen peptides was not established. However, given that we used a mixture of peptides, it can be assumed that the anti-obesity effect of jellyfish collagen peptides is complex, that it is realized at different levels, and that it involves various mechanisms. Enzymatic hydrolysis of jellyfish collagen results in the formation of many peptides, some of which are bioactive. As collagen peptides are possibly structurally similar to several gut hormones and neuropeptides involved in the regulation of energy homeostasis, they may influence food intake and satiety in rats by mimicking the action of gut peptide hormones. Another study also found that peptides isolated from shrimp influenced the release of cholecystokinin by STC-1 cells, leading to appetite suppression [34]. Similarly, milk protein hydrolysates were shown to bind to serotonin receptors, creating an appetite-suppression effect similar to that of physiological ligands [35].
Many studies show obesity to be an important trigger of type-2 diabetes mellitus, the progression of which is characterized by impaired glucose homeostasis [36]. In our experiments, an increased level of HbA1c in rats fed a high-calorie diet indicates an increase in glucose concentration over a long period. The simultaneous increase in glucose and insulin concentrations in obese animals may predict the development of insulin resistance. Additionally, hyperinsulinemia is a compensatory reaction of pancreatic beta cells to a decrease in tissue sensitivity to the action of insulin. The ability of collagen fragments derived from various sources to modulate glucose and lipid metabolism was previously confirmed [37,38]. Considering this fact, we examined whether collagen peptides have the same effect on the parameters associated with insulin resistance. The obtained data indicate a decrease in the level of glucose, insulin, and HbA1c in obese rats treated with collagen peptides. Since an increased content of abdominal fat (confirmed by a high BMI) and, accordingly, an increased level of non-esterified fatty acids and inflammatory markers lead to decreased insulin sensitivity, the recovery of parameters associated with insulin resistance may be explained by the effect of collagen peptides on body weight and body fat content.
Another trigger for the development of obesity-related disorders is systemic oxidative stress [39]. Uncontrolled formation of reactive oxygen species (ROS) can be deleterious by itself since it causes oxidative damage to proteins, lipids, and nucleic acids and leads to the formation of aggressive secondary by-products. ROS action can cause oxidative modification of enzymes and changes in their function, damage pancreatic cells [40], and impair the insulin responsiveness of muscle and liver cells. In addition, both ROS and lipid peroxidation products can disturb redox-dependent cellular homeostasis and signal-transduction pathways, leading to apoptosis, increased inflammation, adipokine imbalance, and even changes in neurotransmitter activity. All of these factors, individually or in combination, are involved in the induction of insulin resistance in obesity. Moreover, our study found an accumulation of oxidatively modified proteins in obese rats. This may further indicate the intensity and duration of oxidative stress as oxidatively modified proteins are an early criterion of tissue damage by free radicals. If the structure of cellular proteins is modified by free radicals, their function is decreased or completely abolished. This may cause an accumulation of protein aggregates – factors provoking the development of complications associated with obesity. In addition, proteins that undergo oxidative modification may stimulate the production of new antibodies, thus provoking an immune or autoimmune response. Oxidative stress often occurs as a result of a decreased activity of antioxidant defense system due to the depletion of non-enzymatic antioxidants with low-molecular-weight. It may also occur due to a decreased level and activity of antioxidant enzymes. In our experiment, the presence of oxidative stress in obese rats was confirmed by a decreased activity of superoxide dismutase. A previous study also found a decreased activity of antioxidant enzymes and a depletion of antioxidants in overweight and obese animals [41].
In our experiments, collagen peptides prevented the development of oxidative stress in rats fed a high-calorie diet. They restored the capacity of the antioxidant system, as shown by the normalized level of oxidative-modified proteins and an increased activity of superoxide dismutase. This may be the result of a direct action of peptides as free-radical scavengers. Previous in vitro and in vivo studies confirmed the ability of peptides to reduce the levels of superoxide radicals, hydroxyl radicals, and to chelate prooxidative transition metals [42,43]. In addition, peptides may be involved in the regulation of several antioxidant enzyme genes. Given that oxidative stress is involved in the development of diabetic complications, achieving oxidative homeostasis in rats treated with collagen peptides may additionally reduce the damage to pancreatic cells.
The limitation of this study is the lack of a control group of animals without obesity administered with collagen peptides. Our study also did not assess the levels of pro-inflammatory cytokines, as well as the levels of key gut and adipose tissue hormones, which is a step needed to comprehensively assess the anti-obesity effect of collagen peptides.
To our knowledge, this is the first report to evaluate the effect of collagen peptides from the jellyfish Diplulmaris antarctica. Due to its abundance and high proliferative potential, this jellyfish can be considered a sustainable source of collagen and its derivatives. Given the conservatism of the collagen structure, jellyfish from other regions can also be used to obtain collagen peptides, which may contribute to curbing their uncontrolled spread in the oceans. Our results indicate that jellyfish collagen peptides can be used for the prevention and treatment of obesity caused by a high-calorie diet, as well as of pathologies associated with increased oxidative stress. Further studies are needed to elucidate the mechanisms of the anti-obesity effect of jellyfish collagen peptides.
## References
1. Kopelman PG. **Obesity as a medical problem.**. *Nature* (2000.0) **404** 635-43. DOI: 10.1038/35007508
2. Lavie CJ, Milani RV, Ventura HO. **Obesity and cardiovascular disease: risk factor, paradox and impact of weight loss.**. *J Am Coll Cardiol* (2009.0) **53** 1925-32. DOI: 10.1016/j.jacc.2008.12.068
3. Bendor CD, Bardugo A, Pinhas-Hamiel O, Afek A, Twig G. **Cardiovascular morbidity, diabetes and cancer risk among children and adolescents with severe obesity.**. *Cardiovasc Diabetol* (2020.0) 79. DOI: 10.1186/s12933-020-01052-1
4. Kahan S, Manson JE. **Obesity treatment, beyond the guidelines: practical suggestions for clinical practice.**. *JAMA* (2019.0) **321** 1349-50. DOI: 10.1001/jama.2019.2352
5. Jackson VM, Breen DM, Fortin JP, Liou A, Kuzmiski JB, Loomis AK. **Latest approaches for the treatment of obesity.**. *Expert Opin Drug Discov* (2015.0) **10** 825-39. DOI: 10.1517/17460441.2015.1044966
6. Jang M, Kim J, Choi Y, Bang J, Kim Y. **Antiseptic effect of Ps-K18: mechanism of its antibacterial and anti-inflammatory activities.**. *Int J Mol Sci* (2019.0) **20** 4895. DOI: 10.3390/ijms20194895
7. Mahlapuu M, Hakansson J, Ringstad L, Bjorn C. **Antimicrobial peptides: an emerging category of therapeutic agents.**. *Front Cell Infect Microbiol* (2016.0) **6** 194. DOI: 10.3389/fcimb.2016.00194
8. Oludiran A, Courson D, Stuart MD, Radwan AR, Poutsma JC, Cotten ML. **How oxygen availability affects the antimicrobial efficacy of host defense peptides: lessons learned from studying the copper-binding peptides piscidins 1 and 3.**. *Int J Mol Sci* (2019.0) **20** 5289. DOI: 10.3390/ijms20215289
9. Szekely M, Petervari E, Szelenyi Z. **Orexigenic vs. anorexigenic peptides and feeding status in the modulation of fever and hypothermia.**. *Front Biosci* (2004.0) **9** 2746-63. DOI: 10.2741/1433
10. Bergen WG. **Peptide-based regulation of metabolism as related to obesity.**. *Endocrinology* (2017.0) **158** 3081-2. DOI: 10.1210/en.2017-00696
11. Zhu C-F, Li G-Z, Peng H-B, Zhang F, Chen Y, Li Y. **Treatment with marine collagen peptides modulates glucose and lipid metabolism in Chinese patients with type 2 diabetes mellitus.**. *Appl Physiol Nutr Metab* (2010.0) **35** 797-804. DOI: 10.1139/H10-075
12. Saito M, Kiyose C, Higuchi T, Uchida N, Suzuki H. **Effect of collagen hydrolysates from salmon and trout skins on the lipid profile in rats.**. *J Agric Food Chem* (2009.0) **57** 10477-82. DOI: 10.1021/jf902355m
13. Wang JB, Xie Y, Pei XR, Yang RY, Zhang ZF, Li Y. **The lipid-lowering and antioxidative effects of marine collagen peptides.**. *Zhonghua Yu Fang Yi Xue Za Zhi.* (2008.0) **42** 226-30. PMID: 18763630
14. Raposo A, Coimbra A, Amaral L, Gonçalves A, Morais Z. **Eating jellyfish: Safety, chemical and sensory properties.**. *J Sci Food Agric* (2018.0) **98** 3973-81. DOI: 10.1002/jsfa.8921
15. Yu H, Liu X, Xin R, Liu S, Li C, Li P. **Radical scavenging activity of protein from tentacles of jellyfish Rhopilema esculenta.**. *Bioorg Med Chem Lett* (2005.0) **15** 2659-64. DOI: 10.1016/j.bmcl.2005.03.044
16. Ovchinnikova TV, Balandin SV, Aleshina GM, Tagaev AA, Leonova YF, Krasnodembsky ED. **Aurelin, a novel antimicrobial peptide from jellyfish Aurelia aurita with structural features of defensins and channel-blocking toxins.**. *Biochem Biophys Res Commun* (2006.0) **348** 514-23. DOI: 10.1016/j.bbrc.2006.07.078
17. Ha SH, Jin F, Kwak CH, Abekura F, Park JY, Park NG. **Jellyfish extract induces apoptotic cell death through the p38 pathway and cell cycle arrest in chronic myelogenous leukemia K562 cells.**. *PeerJ* (2017.0) **5**. DOI: 10.7717/peerj.2895
18. Kawabata T, Lindsay DJ, Kitamura M, Konishi S, Nishikawa J, Nishida S. **Evaluation of the bioactivities of water-soluble extracts from twelve deep-sea jellyfish species.**. *Fish Sci* (2013.0) **79** 487-94. DOI: 10.1007/s12562-013-0612-y
19. Zeng S, Yin J, Yang S, Zhang C, Yang P, Wu W. **Structure and characteristics of acid and pepsin-solubilized collagens from the skin of cobia (Rachycentron canadum).**. *Food Chem* (2012.0) **135** 1975-84. DOI: 10.1016/j.foodchem.2012.06.086
20. Singh P, Benjakul S, Maqsood S, Kishimura H. **Isolation and characterisation of collagen extracted from the skin of striped catfish (Pangasianodon hypophthalmus).**. *Food Chem* (2011.0) **124** 97-105. DOI: 10.1016/j.foodchem.2010.05.111
21. 21Feller G. Psychrophilic enzymes: from folding to function and biotechnology. Hindawi Publishing Corporation Scientifca.2013;1-28.
22. Raksha N, Halenova T, Vovk T, Savchuk O, Samborska I, Zaichko N. **Protein-peptide composition in the lungs of rats with hyperhomocysteinemia.**. *Journal of Biological Research (Italy)* (2021.0) **94** 67-71. DOI: 10.4081/jbr.2021.9858
23. Shen X-H, Tang Q-Y, Huang J, Cai W. **Vitamin E regulates adipocytokine expression in a rat model of dietary-induced obesity.**. *Exp Biol Med (Maywood)* (2010.0) **235** 47-51. DOI: 10.1258/ebm.2009.009122
24. 24Crowther J. The ELISA guidebook. Methods Mol Biol. 2000;149:III-IV:1-413.
25. 25Hyshiktuev BS, Hyshiktueva NA, Ivanov VN. Metody opredelenija produktov perekisnogo okislenija lipidov v kondensate vydyhaemogo vozduha i ih klinicheskoe znachenie. Klinicheskaja laboratornaja diagnostika. 1996;3:13-15.
26. Fletcher BL, Dillard CL, Tappel AL. **Fluorescent products of lipid peroxidation of mitohondria and microsomes.**. *Anal Biochem* (1973.0) **52** 1-9. DOI: 10.1016/0003-2697(73)90327-8
27. Sirota TV. **A novel approach to study the reaction of adrenaline autooxidation: A possibility for polarographic determination of superoxide dismutase activity and antioxidant properties of various preparations.**. *Biochemistry* (2011.0) **3** 253-9
28. Bradford MM. **A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding.**. *Anal Biochem* (1976.0) **2** 248-54. DOI: 10.1016/0003-2697(76)90527-3
29. Aoyama T, Fukui K, Takamatsu K, Hashimoto Y, Yamamoto T. **Soy protein isolate and its hydrolysate reduce body fat of dietary obese rats and genetically obese mice (yellow KK).**. *Nutrition* (2000.0) **16** 349-54. DOI: 10.1016/S0899-9007(00)00230-6
30. de Campos Zani S, Wu J. **B. Chan C. Egg and soy-derived peptides and hydrolysates: a review of their physiological actions against diabetes and obesity.**. *Nutrients* (2018.0) **10** 549. DOI: 10.3390/nu10050549
31. Kwak JH, Ahn C, Park S, Jung S, Min B, Kim OY. *Food Funct* (2012.0) **3** 1019-24. DOI: 10.1039/c2fo10244g
32. Raksha NG, Potalitsyn PY, Yurchenko AV, Halenova TI, Savchuk OM, Ostapchenko LI. **Effect of collagen fragments on development of diet-induced obesity in rats.**. *Arch Biol Sci* (2018.0) **70** 77-86. DOI: 10.2298/ABS170401027R
33. 33Lv Z, Zhang C, Song W, Chen Q, Wang Y. Jellyfish collagen hydrolysate alleviates inflammation and oxidative stress and improves gut microbe composition in high-fat diet-fed mice. Hindawi Mediators of Inflammation. 2022;Article ID 5628702.
34. Cudennec B, Ravallec-Plé R, Courois E, Fouchereau-Peron M. **Peptides from fish and crustacean by-products hydrolysates stimulate cholecystokinin release in STC-1 cells.**. *Food Chem* (2008.0) **111** 970-5. DOI: 10.1016/j.foodchem.2008.05.016
35. Nongonierma AB, Schellekens H, Dinan TG, Cryan JF, FitzGerald RJ. **Milk protein hydrolysates activate 5-HT(2C) serotonin receptors: influence of the starting substrate and isolation of bioactive fractions.**. *Food Funct* (2013.0) **4** 728-37. DOI: 10.1039/c3fo30309h
36. Bell JA, Kivimaki M, Hamer M. **Metabolically healthy obesity and risk of incident type 2 diabetes: a meta-analysis of prospective cohort studies.**. *Obes Rev* (2014.0) **15** 504-15. DOI: 10.1111/obr.12157
37. Saito M, Kiyose C, Higuchi T, Uchida N, Suzuki H. **Effect of collagen hydrolysates from salmon and trout skins on the lipid profile in rats.**. *J Agric Food Chem* (2009.0) **57** 10477-82. DOI: 10.1021/jf902355m
38. Zhu C-F, Li G-Z, Peng H-B, Zhang F, Chen Y, Li Y. **Treatment with marine collagen peptides modulates glucose and lipid metabolism in Chinese patients with type 2 diabetes mellitus.**. *Appl Physiol Nutr Metab* (2010.0) **35** 797-804. DOI: 10.1139/H10-075
39. Hurrle S, Hsu WH. **The etiology of oxidative stress in insulin resistance.**. *Biomed J* (2017.0) **40** 257-62. DOI: 10.1016/j.bj.2017.06.007
40. Kaneto H, Kawamori D, Matsuoka TA, Kajimoto Y, Yamasaki Y. **Oxidative stress and pancreatic beta-cell dysfunction.**. *Am J Ther* (2005.0) **12** 529-33. DOI: 10.1097/01.mjt.0000178773.31525.c2
41. Hosseini B, Saedisomeolia A, Allman-Farinelli M. **Association between antioxidant intake/status and obesity: a systematic review of observational studies.**. *Biol Trace Elem Res* (2017.0) **175** 287-97. DOI: 10.1007/s12011-016-0785-1
42. Pozzolini M, Millo E, Oliveri C, Mirata S, Salis A, Damonte G. **Elicited ROS scavenging activity, photoprotective, and wound-healing properties of collagen-derived peptides from the marine sponge**. *Mar Drugs* (2018.0) **16** 465. DOI: 10.3390/md16120465
43. Sampath Kumar NS, Nazeer RA, Jaiganesh R. **Purification and identification of antioxidant peptides from the skin protein hydrolysate of two marine fishes, horse mackerel (**. *Amino Acids* (2012.0) **42** 1641-9. DOI: 10.1007/s00726-011-0858-6
|
---
title: 'Immunohistochemical expression of apolipoprotein B and 4-hydroxynonenal proteins
in colorectal carcinoma patients: a retrospective study'
authors:
- Phei Ying Ng
- Siti Norasikin Mohd Nafi
- Nur Asyilla Che Jalil
- Yee Cheng Kueh
- Yeong Yeh Lee
- Anani Aila Mat Zin
journal: Croatian Medical Journal
year: 2023
pmcid: PMC10028567
doi: 10.3325/cmj.2023.64.29
license: CC BY 2.5
---
# Immunohistochemical expression of apolipoprotein B and 4-hydroxynonenal proteins in colorectal carcinoma patients: a retrospective study
## Body
Colorectal carcinoma (CRC) is increasing in prevalence, being currently the second most common cancer in Malaysia and the third most common cancer in the world [1,2]. The cancer has considerable morbidity and mortality. The most affected age group are patients older than 50 years, and the disease is frequently diagnosed in advanced stages [3,4]. An increasingly recognized CRC risk factor is obesity [5] as it can cause aberrant lipid metabolism by speeding up cancer-cell cycles and replication rates [5,6].
Apolipoprotein B (apoB) and 4-hydroxynonenal (4HNE) are markers related to the oxidation of low-density lipoprotein (LDL). ApoB, predominantly synthesized in the intestines and liver [7,8], is the primary protein component of chylomicrons and LDL, particles that transport cholesterol to the outside of the synthesized organs [6,7,9]. It is associated with a poor prognosis in metastatic CRC patients [9] and in stage-III and high-risk stage-II CRC patients undergoing curative surgery [10].
4HNE is a lipid peroxidation marker formed when free radicals react with the lipid membrane of cells under oxidative stress [11]. It modifies DNA and binds to proteins to form 4HNE protein adducts, which act as growth-regulating signaling factors and cause inflammation and apoptosis [11,12]. 4HNE expression was decreased following apocynin treatment and hyperbaric oxygenation in acute kidney injury [13]. However, it was increased in non-alcoholic fatty liver disease [11] and in Barrett's esophagus with specific intestinal metaplasia compared with esophageal cancer [14]. It was also higher in well- and moderately-differentiated oropharyngeal cancers, but lower in poorly differentiated and advanced cancers [12].
During LDL oxidation, 4HNE forms protein adducts with the lysine of apoB, thus altering the apoB that binds to LDL [15]. A previous study assessing apoB and 4HNE found that modified LDL may promote pre-eclampsia in mothers with diabetes [16]. Although the link between apoB and 4HNE has been previously shown, the significance of apoB and 4HNE in cancer development remains to be elucidated. Hence, the current study aimed to determine the differential expression of apoB and 4HNE in human CRC tissues and the association of these markers with demographic factors, body mass index (BMI), and clinicopathological data.
## Abstract
### Aim
To assess the association of the expression of apolipoprotein B (apoB) and 4-hydroxynonenal (4HNE) with the clinicopathological data of patients with colorectal cancer (CRC).
### Methods
We obtained 80 CRC histopathological specimens sent to the Pathology Laboratory of Hospital Universiti Sains Malaysia from 2015 to 2019. Data on demographic factors, body mass index (BMI), and clinicopathological characteristics were also collected. Formalin-fixed paraffin-embedded tissues were stained by using an optimized immunohistochemical protocol.
### Results
Patients were mostly older than 50 years, male, Malay, and overweight or obese. A high apoB expression was observed in $87.5\%$ CRC samples ($\frac{70}{80}$), while a high 4HNE expression was observed in only $17.5\%$ ($\frac{14}{80}$) of CRCs. The expression of apoB was significantly associated with the sigmoid and rectosigmoid tumor sites ($$p \leq 0.001$$) and tumor size 3-5 cm ($$p \leq 0.005$$). 4HNE expression was significantly associated with tumor size 3-5 cm ($$p \leq 0.045$$). Other variables were not significantly associated with the expression of either marker.
### Conclusion
ApoB and 4HNE proteins may play a role in promoting CRC carcinogenesis.
## Study design and data extraction
This retrospective study was conducted at Hospital Universiti Sains Malaysia (HUSM), a tertiary referral hospital located in the northeastern region of Peninsular Malaysia. The specimens were obtained from the Laboratory Information System (LIS), a database of histopathological specimens of the Pathology Department of HUSM. The study was approved by the Human Ethics and Research Committee of Universiti Sains Malaysia (USM/JEPeM 19060354).
Of the 98 CRC patients registered in the LIS from 2015 to 2019, 18 were not included as the samples were inadequate for IHC staining. This left 80 specimens in the final sample. Demographic factors, BMI, and clinicopathological characteristics were extracted from the medical records.
## Sample collection and preparation
Formalin-fixed, paraffin-embedded tissue blocks of CRC patients were collected from the Pathology Laboratory. Each of the tissue blocks was sectioned into two-micrometer-thick sections on a poly-L-lysine slide by using a microtome (Leica, Harbourfront Centre, Singapore). The tissue slides were then prepared for IHC staining.
## Immunohistochemistry staining
The IHC staining protocol for apoB and 4HNE was optimized previously (unpublished data). The tissue slides were de-waxed in a hot air oven (Bionics Scientific Technologies, Delhi, India) for 20 minutes and deparaffinized with xylene. Then they were rehydrated with decreasing ethanol concentrations: absolute ethanol (HmbG®, Hamburg, Germany), $95\%$ ethanol, $80\%$ ethanol, $70\%$ ethanol, and $50\%$ ethanol, each for 5 minutes. ApoB and 4HNE antigens were retrieved with citrate and Tris-EDTA buffers (both from Dako, Glostrup, Denmark), respectively. The antigen retrieval was carried out by using a heat-induced process that included incubating slides in a decloaking chamber (Biocare Medical, Pacheco, CA, USA) at 121 psi for 30 seconds. Anti-apoB antibody and anti-4HNE antibody (both from Abcam, Cambridge, UK) [17,18] were incubated at 1:100 concentrations for one hour at room temperature. A negative control was created by omitting the antibody. REAL EnVision Detection System (Dako), which employs horseradish peroxidase-conjugated polymer method, was used to detect and visualize the bound antibodies using 3,3′-diaminobenzidine chromogen. The tissue slides were then counterstained with hematoxylin (Merck, Darmstadt, Germany), dehydrated with an increasing ethanol concentration ($50\%$ ethanol to absolute ethanol), cleared with xylene, and mounted by using DPX glue.
## Immunohistochemical scoring
A semi-quantitative IHC scoring protocol was modified from previous studies [6,12]. The IHC score, which was calculated only for the tumor area and not for the surrounding normal tissue, was based on two parameters: the proportion of positive tumor cells (0: $0\%$ positive cells; 1: 5-$25\%$ positive cells; 2: 26-$50\%$ positive cells; and 3: ˃$50\%$ positive cells) and the intensity grade (0: no staining; 1: mild staining; 2: moderate staining; and 3: strong staining). The immunoreactive score (IRS) was calculated by multiplying the positive proportion score with the intensity grade. The cut-off point for categorization of the scores into two expression groups was the maximum IRS of 9 divided in half. The expression lower than 4.5 was considered low, whereas the expression higher than 4.5 was considered high. To assess the reproducibility, the IHC scoring was evaluated by two independent pathologists, who were blinded to the clinical data. Disagreements were settled by consensus.
## Statistical analysis
The normality distribution of numerical variables was assessed with a histogram, box-plot, and Shapiro-Wilk test. Numerical variables are presented as mean and standard deviation (SD). Categorical variables are presented as frequency (percentage). The association of markers’ expression with demographic factors, BMI, and clinicopathological characteristics of CRC patients was assessed with a two-sided Fisher’s exact test or a Pearson’s chi-square test. A P ˂0.05 was considered statistically significant. The analyses were performed with SPSS, version 26.0 (IBM Corp., Armonk, NY, USA).
## RESULTS
A high expression of apoB was observed in 70 out of 80 CRC samples ($87.5\%$) (Table 1). A high 4HNE expression was observed in only 14 of 80 CRC samples ($17.5\%$) (Table 1). The cytoplasmic staining intensity of apoB was mild, moderate, and strong, while that of 4HNE was mild and moderate (Figure 1).
The majority of CRC patients ($76.3\%$) were older than 50 years at diagnosis, with a mean age of 59 years. The majority were men ($53.8\%$) and members of the Malay subpopulation ($90.0\%$) (Table 2). According to the Asian BMI categorization [19], $56.3\%$ of the CRC patients were overweight or obese (Table 2). However, the expression of apoB or 4HNE was not significantly associated with demographic factors or BMI (Table 2).
**Table 2**
| Variables | CRC patients (n=80), n (%) | ApoB expression, n (%) | ApoB expression, n (%).1 | Unnamed: 4 | 4HNE expression, n (%) | 4HNE expression, n (%).1 | Unnamed: 7 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | low | high | P | low | high | P |
| Age (mean age ±standard deviation 59±15.03) | | | | 0.999* | | | 0.500* |
| ≤50 | 19 (23.8) | 2 (10.5) | 17 (89.5) | | 17 (89.5) | 2 (10.5) | |
| >50 | 61 (76.3) | 8 (13.1) | 53 (86.9) | | 49 (80.3) | 12 (19.7) | |
| Sex | | | | 0.999* | | | 0.779† |
| male | 43 (53.8) | 5 (11.6) | 38 (88.4) | | 35 (81.4) | 8 (18.6) | |
| female | 37 (46.3) | 5 (13.5) | 32 (86.5) | | 31 (83.8) | 6 (16.2) | |
| Race | | | | 0.065* | | | 0.367* |
| Malay | 72 (90.0) | 7 (9.7) | 65 (90.3) | | 60 (83.3) | 12 (16.7) | |
| Chinese | 6 (7.5) | 2 (33.3) | 4 (66.7) | | 5 (83.3) | 1 (16.7) | |
| others | 2 (2.5) | 1 (50.0) | 1 (50.0) | | 1 (50.0) | 1 (50.0) | |
| Asian BMI classification | | | | 0.167* | | | 0.443* |
| underweight (˂18.5) | 6 (7.5) | 2 (33.3) | 4 (66.7) | | 6 (100.0) | 0 (0.0) | |
| normal
(18.5-22.9) | 28 (35.0) | 5 (17.9) | 23 (82.1) | | 24 (85.7) | 4 (14.3) | |
| overweight (23-24.9) | 10 (12.5) | 0 (0.0) | 10 (100.0) | | 9 (90.0) | 1 (10.0) | |
| obese
(≥25) | 35 (43.8) | 3 (8.6) | 32 (91.4) | | 26 (74.3) | 9 (25.7) | |
The most common sites of CRC were the sigmoid ($28.7\%$) and rectosigmoid ($31.3\%$). High apoB was expressed in all sigmoid tissues and in $84.0\%$ of rectosigmoid tissues. A high apoB expression was observed in most of the left-sided colon, including the sigmoid colon, descending colon, and transverse colon (distal) (Figure 2). The most common CRC tumor sizes were 3-5 cm ($47.5\%$) and >5 cm ($48.8\%$). Overall, $97.4\%$ of CRCs with tumor sizes of 3-5 cm and $82.1\%$ of those with tumor sizes >5 cm demonstrated high apoB expression. On the other hand, $73.7\%$ of CRCs with tumor sizes of 3-5 cm and $92.3\%$ of those with tumor sizes >5 cm expressed low 4HNE expression. ApoB expression was significantly associated with the sigmoid and rectosigmoid tumor site ($$p \leq 0.001$$) and tumor size of 3-5 cm ($$p \leq 0.005$$) (Table 3). However, 4HNE expression was significantly associated only with tumor size of 3-5 cm ($$p \leq 0.045$$) (Table 3).
**Figure 2:** *Percentages of high apolipoprotein B (apoB) (pink) or high hydroxynonenal (4HNE) expression (purple) in left-sided and right-sided colorectal cancer (CRC).* TABLE_PLACEHOLDER:Table 3 The majority of CRC patients had intestinal wall invasion ($93.8\%$, $\frac{75}{80}$) but no lymph node involvement, lymphovascular invasion, or perineural invasion (Table 3). The most common CRC subtypes were adenocarcinomas ($88.8\%$, $\frac{71}{80}$); $87.5\%$ ($\frac{70}{80}$) had moderate differentiation grade and $56.3\%$ ($\frac{45}{80}$) belonged to modified Dukes’ B class. No other clinicopathological variable was significantly associated with either apoB or 4HNE expression (Table 3).
## DISCUSSION
Our study demonstrated a high apoB IHC expression in the majority of CRC tissues. ApoB expression levels were not significantly associated with age, sex, race, or BMI status. Fang et al [19] also demonstrated no association between serum apoB and demographic factors in CRC patients. However, another study reported significantly higher apoB levels in CRC patients younger than 65 years [20].
A high apoB expression was significantly associated with the sigmoid and rectosigmoid locations, and tumor sizes of 3-5 cm. A high apoB expression was observed in most of the left-sided colon, including the sigmoid colon, descending colon, and transverse colon (distal). In other studies, the right-sided CRC had a poorer prognosis than the left-sided CRC due to a higher rate of metastasis [21,22]. Whether high apoB expression in the left-sided CRCs indicates a better prognosis needs to be further studied. However, we also found that tumors with a size of 3-5 cm were more likely to have a high apoB expression than tumors larger than 5 cm, which indicates that the tumor requires apoB at later stages of progression. A lower apoB expression in larger tumors was attributed to lower LDL levels in the late-stage CRC due to poor nutrition [23,24]. Later stages of cancer cell growth are presumed to be driven by a de novo synthesis of endogenous cholesterol via the 3-hydroxy-3-methylglutaryl coenzyme A reductase pathway [23].
4HNE also contributes to the mutagenic and carcinogenic effects of lipid peroxidation [25,26]. However, the role of 4HNE in mediating CRC growth is unclear. Increased 4HNE levels were shown to be associated with advanced stages of CRC [27]. However, 4HNE was also demonstrated to have an anticancerogenic effect, as shown by its ability to inhibit telomerase activity in intestinal cancer cell lines [26]. In KRAS-mutated CRC, 4HNE linked with MAP kinase and transforming growth factor to inhibit tumor growth [28]. The involvement of 4HNE in cancer development is also affected by its concentration. At low concentrations, 4HNE has a protective effect and inhibits cancer cell damage, but at high concentrations it can cause apoptosis or necrosis of cancer cells [25]. In our study, 4HNE expression was significantly increased in CRC with a size of 2 cm or less. In earlier research, high cytoplasmic 4HNE expression was typically detected in dysplastic and early-stage malignancies, when small tumor sizes are typical [12,29]. Due to the toxicity of high 4HNE accumulation, it was hypothesized that 4HNE levels decreased in later stages of cancer progression. The lack of dietary intake of linoleic acid, heme iron, and antioxidants, which may interfere with lipoproteins production, reduced 4HNE production at later stages of CRC [30].
Despite the fact that 4HNE forms protein adducts with apoB to change LDL recognition [15], the role of apoB-4HNE in cancer progression is still unknown. In our study, the intensity of 4HNE protein expression in CRC was lower than that of apoB. Different apoB and 4HNE levels in CRC may indicate that both markers are differently regulated throughout tumor growth. However, the exact mechanism by which 4HNE association with apoB affects cancer growth at the early vs later stages has to be elucidated. In this study, in contrast to apoB, 4HNE expression was not significantly associated with tumor site. This might be explained by previous findings that 4HNE at pathologically relevant concentrations interacts with the cells in the gut. This leads to an elevated inflammatory response and tumorigenesis and results in the progression of inflammatory bowel disease-related CRC [26,30]. However, it is not yet understood how changes to the inflammatory response associated with 4HNE in the gut affect CRC.
This study suffers from several limitations. The retrospective design prevented us from collecting sufficient data on certain clinical variables. The small sample size might explain the lack of association of apoB or 4HNE expression with demographic factors, BMI, or clinicopathological features, which was confirmed in other studies. Another limitation was a lack of mechanistic studies to determine the significance of high apoB expression and low 4HNE expression in certain tumor sites and sizes. This research, therefore, could be considered only hypothesis-generating. Although apoB has been linked to CRC development, little is known about its involvement in modulating CRC oncogenic or tumor-suppressor signaling. Attention should be paid to the APOB gene, which is able to impair DNA repair and regulate oncogenic and metastatic regulators (mTOR and PI3K pathways), as well as inhibit tumor suppressors [8,31]. As this study only assessed the cytoplasmic expression of 4HNE, future studies should address the co-localization of mitochondrial and cytoplasmic 4HNE by using an immunofluorescence approach. Previous CRC studies found that glycated apoB IHC expression increased from the normal tissue around the cancer site ($18\%$) to the cancer tissue ($45\%$). The same was true for the 4HNE level, as assessed with high-performance liquid chromatography [6,27]. We were unable to compare the expression of the antibody markers between normal tissues and colorectal cancer tissues as we focused only on cancer tissues. This issue remains to be investigated in future studies using IHC staining.
In conclusion, apoB expression in the cytoplasm of CRCs was high, but 4HNE protein expression was low, with weaker staining intensity compared with that of apoB. The differences in apoB and 4HNE expressions in different tumor sites and sizes may indicate a role of these proteins in CRC development. Additional research is needed to elucidate the roles of these proteins during colorectal carcinogenesis. Furthermore, apoB and 4HNE expression should be correlated with survival and disease-free times to better comprehend their roles in CRC progression. In addition, since apoB has been associated with the production of lipoproteins in blood, it would be interesting to compare the serum level of apoB and IHC expression of apoB in CRC tissues.
## References
1. Hancock KJ, Hsu W, Klimberg VS. **The clinical versatility of next-generation sequencing in colorectal cancer.**. *American Journal of Biomedical Science & Research.* (2020.0) **7** 548-50. DOI: 10.34297/AJBSR.2020.07.001220
2. 2(WHO) WHO. Globocan 2020: Cancer in Malaysia 2020. Available from: https://gco.iarc.fr/today/data/factsheets/populations/458-malaysia-fact-sheets.pdf. Accessed: January 25, 2021
3. Granados-Romero JJ, Valderrama-Treviño AI, Contreras-Flores EH, Barrera-Mera B, Enríquez MH, Uriarte-Ruíz K. **Colorectal cancer: a review.**. *International Journal of Research in Medical Sciences.* (2017.0) **5** 4667-76. DOI: 10.18203/2320-6012.ijrms20174914
4. Veettil SK, Lim KG, Chaiyakunapruk N, Ching SM, Hassan MRA. **Colorectal cancer in Malaysia: Its burden and implications for a multiethnic country.**. *Asian J Surg* (2017.0) **40** 481-9. DOI: 10.1016/j.asjsur.2016.07.005
5. Ye PF, Xi Y, Huang ZY, Xu PF. **Linking obesity with colorectal cancer: epidemiology and mechanistic insights.**. *Cancers (Basel)* (2020.0) **12** 1408-27. DOI: 10.3390/cancers12061408
6. Reddavide R, Misciagna G, Caruso MG, Notarnicola M, Armentano R, Caruso ML. **Tissue expression of glycated apolipoproteinb in colorectal adenoma and cancer.**. *Anticancer Res* (2011.0) **31** 555-60. PMID: 21378338
7. Chandler PD, Song YQ, Lin J, Zhang SM, Sesso HD, Mora S. **Lipid biomarkers and long-term risk of cancer in the Women’s Health Study.**. *Am J Clin Nutr* (2016.0) **103** 1397-407. DOI: 10.3945/ajcn.115.124321
8. Ren LW, Yi J, Li W, Zheng XJ, Liu JY, Wang JH. **Apolipoproteins and cancer.**. *Cancer Med* (2019.0) **8** 7032-43. DOI: 10.1002/cam4.2587
9. Yang D-D, Chen Z-H, Wang D-S, Yu H-E, Lu J-H, Xu R-H. **Prognostic value of the serum apolipoprotein B to apolipoprotein A-I ratio in metastatic colorectal cancer patients.**. *J Cancer* (2020.0) **11** 1063-74. DOI: 10.7150/jca.35659
10. Chen X-Q, Wu P-W, Liu D-H, Yan S-J, Shen X-M, Yang L-Y. **Prognostic significance of high triglyceride and apolipoproteinB levels in patients with stage III and high-risk stage II colorectal cancer undergoing curative surgery.**. *Oncol Lett* (2020.0) **20** 705-14. DOI: 10.3892/ol.2020.11617
11. Podszun MC, Chung JY, Ylaya K, Kleiner DE, Hewitt SM, Rotman Y. **4-HNE immunohistochemistry and image analysis for detection of lipid peroxidation in human liver samples using vitamin E treatment in NAFLD as a proof of concept.**. *J Histochem Cytochem* (2020.0) **68** 635-43. DOI: 10.1369/0022155420946402
12. Jakovčević A, Žarković K, Jakovčević D, Rakušić Z, Prgomet D, Waeg G. **The appearance of 4-Hydroxy-2-Nonenal (HNE) in squamous cell carcinoma of the oropharynx.**. *Molecules* (2020.0) **25** 868-80. DOI: 10.3390/molecules25040868
13. Kovacevic S, Ivanov M, Zivotic M, Brkic P, Miloradovic Z, Jeremic R. **Immunohistochemical analysis of 4-HNE, NGAL, and HO-1 tissue expression after apocynin treatment and HBO preconditioning in postischemic acute kidney injury induced in spontaneously hypertensive rats.**. *Antioxidants* (2021.0) **10** 1163-79. DOI: 10.3390/antiox10081163
14. O’Farrell NJ, Phelan JJ, Feighery R, Doyle B, Picardo SL, Ravi N. **Differential expression profiles of oxidative stress levels, 8-oxo-dG and 4-HNE, in Barrett’s esophagus compared to esophageal adenocarcinoma.**. *Int J Mol Sci* (2019.0) **20** 4449-63. DOI: 10.3390/ijms20184449
15. 15Jairam V, Uchida K, Narayanaswami V. Chapter 16: Pathophysiology of lipoprotein oxidation: Intech Open Science; 2012.
16. McLeese RH, Zhao JW, Fu DX, Yu JY, Brazil DP, Lyons TJ. **Effects of modified lipoproteins on human trophoblast cells: a role in preeclampsia in pregnancies complicated by diabetes.**. *BMJ Open Daib Res Care.* (2021.0) **9** 1696-705. DOI: 10.1136/bmjdrc-2020-001696
17. 17Abcam. Anti-apolipoprotein B antibody ab20737 2020. Available from: www.abcam.com/apolipoprotein-b-antibody-ab20737.pdf. Accessed: August 7, 2021.
18. 18Abcam. Anti-4 hydroxynonenal antibody ab46545 2020. Available from: www.abcam.com/4-hydroxynonenal-antibody-ab46545.pdf. Accessed: August 7, 2021.
19. Fang Z, He MM, Song MY. **Serum lipid profiles and risk of colorectal cancer: a prospective cohort study in the UK Biobank.**. *Br J Cancer* (2020.0) **124** 663-70. DOI: 10.1038/s41416-020-01143-6
20. Sirniö P, Väyrynen JP, Klintrup K, Mäkelä J, Mäkinen MJ, Karttunen TJ. **Decreased serum apolipoprotein A1 levels are associated with poor survival and systemic inflammatory response in colorectal cancer.**. *Sci Rep* (2017.0) **7** 5374-81. DOI: 10.1038/s41598-017-05415-9
21. Lai HW, Wei JCC, Hung HC, Lin CC. **Tumor sidedness influences prognostic impact of lymph node metastasis in colon cancer patients undergoing curative surgery.**. *Natural Research: Scientific Reports.* (2019.0) **9** 19892-901. DOI: 10.1038/s41598-019-56512-w
22. Xie M-Z, Li J-L, Cai Z-M, Li K-Z, Hu B-L. **Impact of primary colorectal Cancer location on the KRAS status and its prognostic value.**. *BMC Gastroenterol* (2019.0) **19**. DOI: 10.1186/s12876-019-0965-5
23. Notarnicola M, Caruso MG, Tutino V, De Nunzio V, Gigante I, De Leonardis G. **Nutrition and lipidomic profile in colorectal cancers.**. *Acta Biomed* (2018.0) **89** 87-96. PMID: 30561400
24. Wang CH, Li PW, Xuan JM, Zhua CP, Liu JJ, Shan LZ. **Cholesterol enhances colorectal cancer progression via ROS elevation and MAPK signaling pathway activation.**. *Cell Physiol Biochem* (2017.0) **42** 729-42. DOI: 10.1159/000477890
25. Zhong HQ, Yin HY. **Role of lipid peroxidation derived 4-hydroxynonenal (4-HNE) in cancer: Focusing on mitochondria.**. *Redox Biol* (2015.0) **4** 193-9. DOI: 10.1016/j.redox.2014.12.011
26. Pakiet A, Kobiela J, Stepnowski P, Sledzinski T, Mika A. **Changes in lipids composition and metabolism in colorectal cancer: a review.**. *Lipids Health Dis* (2019.0) **18**. DOI: 10.1186/s12944-019-0977-8
27. Skrzydlewska E, Sulkowski S, Koda M, Zalewski B, Kanczuga-Koda L, Sulkowska M. **Lipid peroxidation and antioxidant status in colorectal cancer.**. *World J Gastroenterol* (2005.0) **11** 403-6. DOI: 10.3748/wjg.v11.i3.403
28. Bartolacci C, Andreani C, El-Gammal Y, Scaglioni PP. **Lipid metabolism regulates oxidative stress and ferroptosis in RAS-driven cancers: A perspective on cancer progression and therapy.**. *Front Mol Biosci* (2021.0) **8** 1-19. DOI: 10.3389/fmolb.2021.706650
29. Blendea A, Şerban IL, Brănişteanu DC, Brănişteanu DE. **Evaluation of immunostaining for 4-Hydroxy-2-Nonenal receptors in cutaneous malignant melanoma immunohistochemical study of 55 cases.**. *J Mol Biomark Diagn* (2017.0) **8** 364-70
30. Lei L, Zhang JN, Decker EA, Zhang GD. **Roles of lipid peroxidation-derived electrophiles in pathogenesis of colonic inflammation and colon cancer.**. *Front Cell Dev Biol* (2021.0) 9. DOI: 10.3389/fcell.2021.665591
31. Kopylov AT, Stepanov AA, Malsagova KA, Soni D, Kushlinsky NE, Enikeev DV. **Revelation of proteomic indicators for colorectal cancer in initial stages of development.**. *Molecules* (2020.0) **25** 619-38. DOI: 10.3390/molecules25030619
|
---
title: Weight gain, poor mental health and increased sedentary hours among Malaysian
adults during the COVID-19 pandemic
authors:
- Guo Fu Ng
- Shi-Hui Cheng
journal: Nutrition and Health
year: 2023
pmcid: PMC10028682
doi: 10.1177/02601060231164434
license: CC BY 4.0
---
# Weight gain, poor mental health and increased sedentary hours among Malaysian adults during the COVID-19 pandemic
## Abstract
### Background
The Movement Control Orders (MCO) in Malaysia due to the COVID-19 pandemic had a profound impact on the lifestyle behaviours, weight changes, and mental health of the population.
### Aim
To determine the changes in physical activity, sedentary behaviour, body weight status and mental health status among Malaysian adults before and during the pandemic.
### Methods
A total of 338 Malaysian adults participated in this cross-sectional online study. Sociodemographic and anthropometric data were self-reported. Physical activity and sedentary behaviour were assessed using International Physical Activity Questionnaire-Short Form (IPAQ-SF) while the Perceived Stress Scale (PSS-10), Patient Health Questionnaire (PHQ-9) and Generalised Anxiety Disorder Assessment (GAD-7) were used to examine stress, depression and anxiety, respectively. All statistical analysis was performed using SPSS version 28.0.
### Results
The results showed an average weight gain of 0.6 kg among the participants with $45.5\%$ of them experiencing weight gain. In addition, sedentary behaviour ($p \leq 0.001$), PSS-10 score ($p \leq 0.001$), PHQ-9 score ($$p \leq 0.002$$) and GAD-7 score ($$p \leq 0.001$$) were significantly increased during the COVID-19 pandemic whereas the level of physical activity was significantly decreased ($$p \leq 0.003$$) during the pandemic. Weight changes during the pandemic were found to be associated with age, sedentary hours, and PHQ-9 score. Through binary logistic regression, sedentary hours (AOR = 1.068, $95\%$ CI = 1.002–1.139, $$p \leq 0.043$$) were identified to be a risk factor for weight gain during the pandemic.
### Conclusion
The findings suggested that public health interventions to prevent weight gain should focus on strategies to increase physical activity for sedentary lifestyles.
## Introduction
Since the outbreak of coronavirus disease 2019 (COVID-19) caused by SARS-CoV-2 in Wuhan, China, it has spread rapidly across the globe. As part of the measures to halt the spread of the disease, nationwide lockdowns, known as Movement Control Orders (MCO) have been enforced in Malaysia along with stringent Standard Operating Procedures (SOPs) such as social distancing, home confinement and temporary closure of unnecessary businesses. The SOPs in place have had an impact on everyone's lifestyle since outdoor activities are restricted (di Renzo et al., 2020). Physical inactivity is expected to deteriorate compared to pre-COVID-19, as remote learning and smart working have emerged as new norms. Moreover, the suspension of social activities and extended home confinement would allow more time for TV and most likely induce boredom, which is associated with an increased intake of energy-dense snacks that most people would treat as a way to escape boredom (di Renzo et al., 2020). Ultimately, physical inactivity and a sedentary lifestyle would have a detrimental synergistic effect on the weight status among adults (di Renzo et al., 2020).
Aside from altering people's lifestyle, adjournment of social activities and prolonged isolation due to COVID-19 would eventually lead to stress, anxiety and depression in the public thereby further deteriorating the mental health of the population (Al-Musharaf et al., 2021). The association between stress and overeating has been thoroughly researched, whereby those under stress are more likely to crave ‘comfort foods’ which contain a high amount of sugar (Husain and Ashkanani, 2020). Furthermore, the unpleasant experience of prolonged isolation also contributes to the physiological response of seeking reward and satisfaction, resulting in excess food intake and a positive energy balance (di Renzo et al., 2020).
Altogether, physical inactivity, sedentary lifestyle and deteriorated mental health have effects on weight changes during the pandemic. Although the SOPs has been gradually loosened, the ongoing mutation of COVID-19 and its continually high rate of transmission suggest that the pandemic will not end with the development of vaccines. Therefore, this study aims to determine the changes in lifestyle, mental health, and body weight status among Malaysian adults before and during COVID-19.
## Study design and ethics approval
A cross-sectional study was conducted among Malaysian adults before and during COVID-19. Data before COVID-19 were collected from September 2019 through means of recalling, whereas data during COVID-19 were collected from November 2021 to February 2022. A research ethics committee of the university approved this study (Ethics approval ID: NGF061121).
## Subjects and recruitment procedures
The sample size was computed with the formula n=z2pqd2, where n = desired sample size, z = confidence level, p = prevalence, $q = 1$–p and d = margin of error (Sharma et al., 2019). Based on the study by Zheng et al. [ 2020], physical inactivity increased by $70\%$ during the COVID-19 pandemic. Hence, the sample size required in this study was 323 at a $95\%$ confidence interval.
Snowball sampling was used to recruit participants. Malaysian adults aged 18 and above were invited to participate in this study via multiple communication platforms including email, WhatsApp, Facebook, and Instagram. Adults who had COVID-19 infection at the time of completing the survey, having any mental health issues, physical disabilities or chronic diseases and pregnant women were excluded from this study.
## Data collection
The questionnaire was made up of three sections: [1] sociodemographic and anthropometric information; [2] physical activity and sedentary behaviour; and [3] mental health status. All responses were self-reported by the participants.
## Sociodemographic and anthropometric information
Sociodemographic information including gender, age, highest education level, ethnicity, monthly household income, employment status, smoking status and lockdown status were collected. For anthropometric measurement, the height and weight of the participants before the pandemic were self-reported by the participants. On the other hand, the body weight of the participants during the pandemic was self-measured with a bathroom scale with minimal clothes, and shoes off. Body weight changes were reflected by the difference before and during the pandemic and were further categorised into ‘decreased weight’, ‘no change’ or ‘increased weight’. Additionally, body mass index (BMI) was derived by dividing body weight (kg) with the square of height (m) thereafter classified based on the Asia-Pacific cutoff points whereby below 18.5, 18.5–22.9, 23.0–27.4, and above 27.5 kg/m2 represents underweight, normal weight, overweight, and obesity, respectively, (Lim et al., 2017).
## Physical activity and sedentary behaviour
The physical activity level was examined by using the International Physical Activity Questionnaire-Short Form (IPAQ-SF). The questionnaire consisted of seven questions that aimed to gather data on the duration of sedentary behaviour, walking, and moderate, and vigorous physical activities by the participants for the past seven days (International Physical Activity Questionnaire, 2004). The weekly metabolic equivalent (MET) minute was computed by multiplying the MET factor allocated for each activity (walking = 3.3 MET, moderate physical activity = 4.0 MET, vigorous physical activity = 8.0 MET) with the duration spent on the activity and the number of days the activity was carried out in a week. To classify the participants according to how active they were, MET minutes per week were computed by summing up the weekly MET minutes from walking, moderate, and vigorous physical activity. Based on the WHO recommendation, participants who achieved 600 MET-minutes/week and above were classified as physically active whereas those who obtained less than 600 MET-minutes/week were classified as physically inactive. Duration spent on sedentary behaviour including sitting and lying down was addressed by asking the participants to list down the duration they spent sitting on a weekday for the past seven days in the final question. Subsequently, participants were divided into two groups of sedentary and non-sedentary, with an 8-hour cutoff time for sedentary behaviour (Tan et al., 2021).
## Perceived stress questionnaire
The Perceived Stress Scale (PSS-10) was used to evaluate the stress level among the participants for the past one month during the pandemic. Each aspect was based on a scale of zero to four whereby a higher score indicated less severe stress in questions 4, 5, 7 and 8 but more severe stress for the rest of the questions (Cohen et al., 1983). The final score ranging from 0 to 40 was obtained by summing up the score from all 10 questions. Participants were then classified into three separate groups of slightly (0–13), moderately (14–26) and severely stressed (27–40) based on their scores.
## Depression scale
Patient Health Questionnaire (PHQ-9) was used to assess the level of depression before and during the pandemic (Kroenke et al., 2001). The questionnaire consisted of nine items that required participants to answer on how frequently they were bothered by the problems for the past two weeks. The responses to the questions were corresponded to a score of zero to three. The points scored in each item were added up to obtain the final score of 0–4, 5–9, 10–14, 15–19 and 20–27 indicating minimal, mild, moderate, moderately severe and severe depression, respectively.
## Anxiety scale
The Generalized Anxiety Disorder (GAD-7) (Spitzer et al., 2006) was used to assess the level of anxiety before and during the pandemic. It consisted of seven questions which required the participants to report how frequently they felt the feelings for the past two weeks. The score was allocated and summed up to obtain the final score. Different ranges of final scores at 0–4, 5–9, 10–14 and 15–21 reflected minimal, mild, moderate and severe anxiety, respectively.
## Statistical analysis
Data collected was analysed by using IBM SPSS Statistics version 28.0 (IBM Corp., Armonk, NY, USA). Continuous variables were expressed as mean and standard deviation (SD) while categorical variables were presented as frequencies in percentage (%). The differences between the means of continuous variables before and during the pandemic were determined using paired t-tests. The association between lifestyle behaviours and weight change as well as mental health and weight change were examined using the chi-square test for categorical variables and ANOVA for continuous variables. Fisher's exact test was performed on a variable with a frequency of less than 5. All variables with $p \leq 0.25$ in the chi-square test and ANOVA test were included in the binary logistic regression to determine the factors associated with body weight changes. The reference group used in the binary logistic regression was weight loss or no changes in weight. The p-value was considered statistically significant at $p \leq 0.05.$
## Changes in weight change-related parameters before and during COVID-19
Table 1 presents the changes in body weight-related parameters before and during the COVID-19 pandemic. Overall, there was a significant weight gain in the participants from 63.7 ± 16.4 to 64.3 ± 16.9 kg during the COVID-19 pandemic. Besides, a significant increase in the BMI of the participants from 24.1 ± 5.5 to 24.3 ± 5.6 kg/m2 was found during the COVID-19 pandemic. The level of physical activity has dropped significantly during the COVID-19 pandemic. Conversely, the mean sedentary hours of the participants were significantly increase from 6.4 ± 3.6 to 7.1 ± 3.8 h during the pandemic. In terms of mental health, all the parameters have increased significantly as reflected by the score obtained in PSS-10, PHQ-9 and GAD-7, respectively.
**Table 1.**
| Parameters | Mean ± SD | Mean ± SD.1 | p-value |
| --- | --- | --- | --- |
| Parameters | Before COVID-19 (n = 338) | During COVID-19 (n = 338) | p-value |
| Anthropometry parameters | Anthropometry parameters | Anthropometry parameters | Anthropometry parameters |
| Body weight (kg) | 63.7 ± 16.4 | 64.3 ± 16.9 | 0.011 |
| BMI (kg/m2) | 24.1 ± 5.5 | 24.3 ± 5.6 | 0.017 |
| Lifestyle parameters | Lifestyle parameters | Lifestyle parameters | Lifestyle parameters |
| Physical activity (MET-minutes/week) | 2986.3 ± 3433.0 | 2600.0 ± 3324.5 | 0.003 |
| Sedentary behaviour (hour/day) | 6.4 ± 3.6 | 7.1 ± 3.8 | <0.001 |
| Mental health parameters | Mental health parameters | Mental health parameters | Mental health parameters |
| PSS-10 score | 18.4 ± 5.0 | 19.5 ± 5.9 | <0.001 |
| PHQ-9 score | 5.7 ± 5.2 | 6.5 ± 6.2 | 0.002 |
| GAD-7 score | 4.9 ± 4.8 | 5.6 ± 5.4 | 0.001 |
## Changes in body weight and BMI status during the COVID-19
Table 2 shows the overall changes in body weight and BMI status during COVID-19. Generally, the change in body weight during the COVID-19 pandemic was an average weight gain of 0.6 ± 4.2 kg with those in the underweight category experiencing the largest weight gain at 0.9 ± 2.6 kg. Before the COVID-19 outbreak, $38.8\%$ of the participants had normal BMI, $29.3\%$ were overweight, $20.7\%$ were obese and $11.2\%$ were classified as underweight. During the COVID-19 pandemic, about half of them experienced weight gain ($45.5\%$) while $32.0\%$ of them lost weight and $22.5\%$ with unchanged body weight.
**Table 2.**
| Parameters | Total (n = 338) | BMI status before COVID-19, n (%) | BMI status before COVID-19, n (%).1 | BMI status before COVID-19, n (%).2 | BMI status before COVID-19, n (%).3 | p-value |
| --- | --- | --- | --- | --- | --- | --- |
| Parameters | Total (n = 338) | Underweight (n = 38) | Normal (n = 131) | Overweight (n = 99) | Obesity (n = 70) | p-value |
| Body weight changes (kg) | 0.6 ± 4.2 | 0.9 ± 2.6 | 0.4 ± 3.0 | 0.8 ± 4.3 | 0.4 ± 6.2 | 0.860 |
| Decreased | 108 (32.0) | 10 (26.3) | 38 (29.0) | 32 (32.3) | 28 (40.0) | 0.617 |
| No change | 76 (22.5) | 7 (18.4) | 32 (24.4) | 24 (24.3) | 13 (18.6) | 0.617 |
| Increased | 154 (45.5) | 21 (55.3) | 61 (46.6) | 43 (43.4) | 29 (41.4) | 0.617 |
| BMI status during COVID-19 | | | | | | < 0.001 |
| Underweight | 30 (8.9) | 25 (65.8) | 5 (3.8) | 0 | 0 | |
| Normal | 137 (40.5) | 13 (34.2) | 113 (86.3) | 11 (11.1) | 0 | |
| Overweight | 95 (28.1) | 0 | 13 (9.9) | 77 (77.8) | 5 (7.1) | |
| Obesity | 76 (22.5) | 0 | 0 | 11 (11.1) | 65 (92.9) | |
Many of the participants experienced significant changes in BMI status whereby $34.2\%$ of underweight participants and $11.1\%$ of overweight participants managed to achieve normal BMI during the pandemic. However, $3.8\%$ and $9.9\%$ of those with normal BMI before the pandemic had developed to become underweight and overweight, respectively, during the pandemic. Besides, $11.1\%$ of overweight participants gained weight to become obese, while $7.1\%$ of the obese participants lose weight and returned to the overweight group during the pandemic.
## Associations between sociodemographic characteristics, lifestyle behaviours and mental health with body weight changes during the COVID-19
Table 3 shows the sociodemographic characteristics of the participants, with a mean age of 33.7 ± 14.0 years, and their associations with body weight changes during COVID-19. A majority of the participants were female ($68.6\%$) and had completed an undergraduate ($45.9\%$) as their highest education level. For ethnicity, the majority of the participants were of Chinese origin ($79.6\%$), while $18.0\%$ were Bumiputera and Malay, $1.2\%$ were Indian and $1.2\%$ were other from other minor ethnicities. More than half ($53.2\%$) were from households with a monthly income of RM5000–RM9999 (∼ USD1185‒2371). Most of the participants were non-smokers ($96.7\%$) and were under National Recovery Plan (NRP) Phase 3 ($73.4\%$). During the COVID-19 pandemic, the majority of them remained physically active ($71.9\%$) while $54.4\%$ were non-sedentary with 7.1 ± 3.8 h, on average, of sedentary behaviour in a day. About three-quarters of the participants had a moderate stress level ($73.4\%$) with a mean PSS-10 score of 19.5 ± 5.9. On the other hand, about half of the participants had minimal depression ($50.5\%$) and minimal anxiety ($50.6\%$) with a mean PHQ-9 score and mean GAD-7 score of 6.5 ± 6.2 and 5.6 ± 5.4, respectively.
**Table 3.**
| Parameters | Total (n = 338) | Weight changes | Weight changes.1 | Weight changes.2 | p-value |
| --- | --- | --- | --- | --- | --- |
| Parameters | Total (n = 338) | Decreased (n = 108) | No change (n = 76) | Increased (n = 154) | p-value |
| Parameters | n (%)/Mean ± SD | n (%)/Mean ± SD | n (%)/Mean ± SD | n (%)/Mean ± SD | p-value |
| Age | 33.7 ± 14.0 | 34.4 ± 14.6 | 37.3 ± 14.0 | 31.5 ± 13.2 | 0.010 |
| Gender | | | | | 0.430 |
| Male | 106 (31.4) | 29 (26.9) | 24 (31.6) | 53 (34.4) | |
| Female | 232 (68.6) | 79 (73.1) | 52 (68.4) | 101 (65.6) | |
| Highest education level | | | | | 0.910 |
| Secondary or lower | 24 (7.1) | 6 (5.6) | 4 (5.3) | 14 (9.1) | |
| A-levels/diploma | 68 (20.1) | 21 (19.4) | 16 (21.0) | 31 (20.1) | |
| Undergraduate | 155 (45.9) | 53 (49.1) | 35 (46.1) | 67 (43.5) | |
| Postgraduate | 91 (26.9) | 28 (25.9) | 21 (27.6) | 42 (27.3) | |
| Ethnicity | | | | | 0.411 |
| Chinese | 269 (79.6) | 93 (86.1) | 56 (73.7) | 120 (77.9) | |
| Indian | 4 (1.2) | 1 (0.9) | 1 (1.3) | 2 (1.3) | |
| Bumiputera/Malay | 61 (18.0) | 13 (12.1) | 17 (22.4) | 31 (20.1) | |
| Others | 4 (1.2) | 1 (0.9) | 2 (2.6) | 1 (0.7) | |
| Monthly household income | | | | | 0.164 |
| Below RM2500 | 79 (23.4) | 27 (25.0) | 17 (22.4) | 35 (22.7) | |
| RM2500 – RM4999 | 79 (23.4) | 19 (17.6) | 25 (32.9) | 35 (22.7) | |
| RM5000 – RM9999 | 129 (38.1) | 48 (44.4) | 26 (34.2) | 55 (35.7) | |
| RM10,000 and above | 51 (15.1) | 14 (13.0) | 8 (10.5) | 29 (18.9) | |
| Employment status | | | | | 0.163 |
| Employed – Full-time | 163 (48.2) | 53 (49.1) | 41 (53.9) | 69 (44.8) | |
| Employed – Part-time | 8 (2.4) | 1 (0.9) | 2 (2.6) | 5 (3.2) | |
| Self-employed | 15 (4.4) | 3 (2.8) | 6 (7.9) | 6 (3.9) | |
| Retired | 10 (3.0) | 4 (3.7) | 4 (5.3) | 2 (1.3) | |
| Student/not employed | 142 (42.0) | 47 (43.5) | 23 (30.3) | 72 (46.8) | |
| Smoking status | | | | | 0.787 |
| Smoker | 11 (3.3) | 4 (3.7) | 3 (3.9) | 4 (2.6) | |
| Non-smoker | 327 (96.7) | 104 (96.3) | 73 (96.1) | 150 (97.4) | |
| Lockdown status | | | | | 0.718 |
| National Recovery Plan (NRP) Phase 3 | 248 (73.4) | 80 (74.1) | 58 (76.3) | 110 (71.4) | |
| National Recovery Plan (NRP) Phase 4 | 90 (26.6) | 28 (25.9) | 18 (23.7) | 44 (28.6) | |
| MET-minutes/week | 2600.0 ± 3324.5 | 2408.2 ± 2962.2 | 3161.1 ± 4562.9 | 2457.7 ± 2792.7 | 0.246 |
| Physical activity status | | | | | 0.390 |
| Physically active | 243 (71.9) | 76 (70.4) | 51 (67.1) | 116 (75.3) | |
| Physically inactive | 95 (28.1) | 32 (29.6) | 25 (32.9) | 38 (24.7) | |
| Sedentary hours/day | 7.1 ± 3.8 | 6.8 ± 3.9 | 6.0 ± 3.6 | 7.8 ± 3.8 | 0.003 |
| Sedentary behaviour | | | | | 0.087 |
| Sedentary | 154 (45.6) | 45 (41.7) | 29 (38.2) | 80 (51.9) | |
| Not sedentary | 184 (54.4) | 63 (58.3) | 47 (61.8) | 74 (48.1) | |
| PSS-10 score | 19.5 ± 5.9 | 19.6 ± 6.1 | 19.0 ± 5.2 | 19.6 ± 6.2 | 0.729 |
| Stress level | | | | | 0.468 |
| Low | 46 (13.6) | 15 (13.9) | 7 (9.2) | 24 (15.6) | |
| Moderate | 248 (73.4) | 78 (72.2) | 62 (81.6) | 108 (70.1) | |
| High | 44 (13.0) | 15 (13.9) | 7 (9.2) | 22 (14.3) | |
| PHQ-9 score | 6.5 ± 6.2 | 6.6 ± 5.9 | 4.9 ± 6.0 | 7.3 ± 6.3 | 0.019 |
| Depression level | | | | | 0.035 |
| Minimal | 171 (50.5) | 54 (50.0) | 52 (68.4) | 65 (42.2) | |
| Mild | 72 (21.3) | 22 (20.4) | 10 (13.2) | 40 (26.0) | |
| Moderate | 52 (15.4) | 19 (17.6) | 7 (9.2) | 26 (16.9) | |
| Moderately severe | 35 (10.4) | 12 (11.1) | 5 (6.6) | 18 (11.7) | |
| Severe | 8 (2.4) | 1 (0.9) | 2 (2.6) | 5 (3.2) | |
| GAD-7 score | 5.6 ± 5.4 | 5.6 ± 5.2 | 4.7 ± 5.9 | 6.1 ± 5.3 | 0.166 |
| Anxiety level | | | | | 0.068 |
| Minimal | 171 (50.6) | 57 (52.8) | 48 (63.1) | 66 (42.9) | |
| Mild | 96 (28.4) | 31 (28.7) | 12 (15.8) | 53 (34.4) | |
| Moderate | 44 (13.0) | 11 (10.2) | 11 (14.5) | 22 (14.3) | |
| Severe | 27 (8.0) | 9 (8.3) | 5 (6.6) | 13 (8.4) | |
Among these, the parameters that were found to have significant associations with weight changes during the COVID-19 pandemic were age, sedentary hours and PHQ-9 score. Participants who had gained weight during the pandemic were significantly younger (31.5 ± 13.2 years) than those who had decreased (34.4 ± 14.6 years) or remained the same weight (37.3 ± 14.0 years). Additionally, participants who gained weight during the COVID-19 pandemic had significantly longer sedentary hours (7.8 ± 3.8 h) than those who lost weight (6.8 ± 3.9 h) or unchanged body weight (6.0 ± 3.6 h). Similarly, PHQ-9 scores were found to be significantly higher among participants with weight gain (7.3 ± 6.3) than participants with decreased (6.6 ± 5.9) or maintained body weight (4.9 ± 6.0).
## Risk factors associated with body weight changes during COVID-19
Table 4 demonstrates the risk factors associated with body weight changes during COVID-19. Among the factors assessed, the sedentary hour was found to be significantly associated with weight gain during COVID-19. Participants were found to be 1.068 times more likely to gain weight during the COVID-19 pandemic for every 1 h increase in sedentary behaviour (AOR = 1.068, $95\%$ CI = 1.002–1.139, $$p \leq 0.043$$). No significant associations were found between age, monthly household income, employment status, MET-minutes/week, PHQ-9 score and GAD-7 score with weight gain during the COVID-19.
**Table 4.**
| Parameters | Weight gain during COVID-19 | Weight gain during COVID-19.1 |
| --- | --- | --- |
| Parameters | Adjusted odds ratio (95% CI) | p-value |
| Age | 0.981 (0.961–1.001) | 0.058 |
| Monthly household income | | |
| < RM2500 | 1 | |
| ≥ RM2500 | 1.187 (0.942–1.496) | 0.145 |
| Employment status | | |
| Full-time employment | 1 | |
| Non-full-time employment | 0.937 (0.797–1.101) | 0.429 |
| MET-minutes/week | 1.000 (1.000–1.000) | 0.843 |
| Sedentary hours | 1.068 (1.002–1.139) | 0.043 |
| PHQ-9 score | 1.034 (0.969–1.103) | 0.310 |
| GAD-7 score | 0.984 (0.915–1.057) | 0.652 |
## Discussion
This study revealed $45.5\%$ of the participants had an average weight gain of 0.6 ± 4.2 kg during the pandemic. This finding indicated a relatively higher proportion of the population experiencing weight gain compared to the previous studies, which may be explained by the higher percentage of participants who reported a moderate level of stress. For instance, weight gain was reported among $12.8\%$ of the Spanish population (Rodríguez-Pérez et al., 2020), $33.6\%$ (Drywień et al., 2020) and $38\%$ (Reyes-Olavarría et al., 2020) of the Polish and Chilean population. These variations are probably due to the implementation of different restrictive measures as well as the diverse cultures and dietary habits among different countries (Al-Musharaf et al., 2021). Nonetheless, the average weight gain found in this study is consistent with the weight gain of 0.62 kg reported in the USA (Bhutani et al., 2021) and 0.5 kg reported in China (Zhu et al., 2021). Interestingly, a much higher weight gain of 1.5 kg was reported in Italy (Pellegrini et al., 2020) while 3 kg of weight gain was reported among the Polish population (Sidor and Rzymski, 2020). On the other hand, weight loss was reported among $32\%$ of the participants in this study which was comparable to the percentage previously reported in Malaysia (Chin et al., 2022) and China (Xia et al., 2020).
Weight gain was observed across all BMI categories in this study, with those in the underweight category experiencing the greatest weight gain. The changes in mean MET-minutes per week before and during the COVID-19 by BMI category are shown in Appendix 1. In addition, sedentary hours increased significantly in underweight group before and during COVID-19 ($$p \leq 0.026$$) (Appendix 1). Our findings are in line with previous studies, which have suggested that individuals with lower BMI were more likely to engage in physical activities and live a healthier lifestyle before the pandemic (He et al., 2020). However, because access to physical activity resources was very limited under the lockdown circumstances, a reduction in physical activity was more obvious among those with a lower BMI (Alshahrani et al., 2022). Additionally, individuals with lower BMI were less bothered by excess body weight, therefore, making them less aware of weight control and more prone to weight gain (He et al., 2020).
Among the obese, a surprisingly high percentage ($40.0\%$) of them lost weight during the pandemic. This finding can be explained by the greater awareness among those with higher BMI, who are more likely to control their food intake and increase physical activity. Before the pandemic, they were unable to effectively reduce their body weight due to frequent socializing and drinking as part of their work regimes (He et al., 2020). However, the circumstances under lockdown were ideal for them to lose weight as they now have more time to exercise at home compared to before the pandemic where they only exercise occasionally (di Renzo et al., 2020).
The change in BMI status during the pandemic was significant ($p \leq 0.001$), with $9.9\%$ of those with normal BMI becoming overweight and $11.1\%$ of those overweight developing obesity. An increase in overweight and obese individuals by $4.8\%$ and $3.3\%$, and $4.8\%$ and $5.1\%$ were reported in China (Liu et al., 2021) and Saudi Arabia (Alshahrani et al., 2022), respectively. The discrepancy between the findings can be explained by the timing of data collection since the previous studies were conducted soon after the lockdown was implemented while data in the present study was collected after the lockdown was implemented for an extended period. The overall increasing trend in body weight during the pandemic, resulting in an increase in overweight and obese individuals, is alarming and should be addressed when it comes to policy making.
Participants who gained weight during the pandemic were significantly younger ($$p \leq 0.010$$). Further statistical analysis revealed that the mean MET-minutes per week reduced significantly in the 18–30 younger age group before and during COVID-19 ($$p \leq 0.007$$) (Appendix 1). In addition, sedentary hours increased significantly in the 18–30 age group before and during COVID-19 ($p \leq 0.001$). This can be explained by previous studies which stated that the decline in physical activity was greater among younger adults (Curtis et al., 2021). Besides, over-eating to cope with stress during the pandemic was noted to be prevalent among young adults (Cheng and Kamil, 2020; Cheng and Wong, 2021). This finding is of clinical importance as weight gain among youths can substantially increase over the years and lead to comorbidities.
The current study revealed a significant decrease in physical activity ($$p \leq 0.003$$) from before the pandemic (2986.3 MET-minutes/week) to during the pandemic (2600.0 MET-minutes/week) due to the lockdown. The average physical activity during the pandemic found in this study is comparable with the median score of 2826.0 MET-minutes/week previously reported among Malaysian students (Tan et al., 2021). Despite the fact that overall physical activity decreased during the pandemic, a relatively large proportion of the population ($71.9\%$) remained physically active during the lockdown, which is consistent with previous studies that found $76.0\%$ (Chin et al., 2022) and $79.6\%$ (Tan et al., 2021) of Malaysians to be physically active during the pandemic. Additionally, the proportion of participants classified to be physically inactive ($28.1\%$) in this study is consistent with WHO statistics, which revealed that $23\%$ of adults worldwide were physically inactive (World Health Organization, 2022). The reduction in physical activity during the pandemic in the present study was in line with studies in Kuwait (AlMughamis et al., 2021), Lithuania (Kriaucioniene et al., 2020) as well as Chile (Reyes-Olavarría et al., 2020). Conversely, increased physical activity was observed in a British cohort study (Bann et al., 2021) and an Italian survey (di Renzo et al., 2020). The discrepancy can be explained by the different periods of data collection and different restrictive measures being implemented.
On the contrary, this study found a significant increase in sedentary behaviour ($p \leq 0.001$), with an average of 7.1 h spent on screens and sitting down during the pandemic. This finding contradicts the findings reported by Tan et al., who found that Malaysian university students spent 9.2 h on sedentary behaviour. The longer sedentary hour reported by Tan et al. may be explained by the previous study focused on university students alone, who were expected to commit to longer screen time due to online classes. Besides, almost half of the participants ($45.6\%$) were classified as sedentary in this study. A similar trend was observed in Kuwait (Husain and Ashkanani, 2020) and Saudi Arabia (Alshahrani et al., 2022), where $43.6\%$ and $36.2\%$ of the population, respectively, were found to be sedentary during the pandemic. Although smart working could be a factor, other platforms including TikTok and Netflix were reported to be the main reason for increased screen time.
More importantly, the present study found that participants are 1.068 times more likely to gain weight during the COVID-19 pandemic for every 1 h spent on sedentary behaviour. This is comparable to the previous research, which concluded that sedentary behaviour including watching TV and using the computer was a risk factor for weight gain (Ashdown-Franks et al., 2019). Similarly, among Arabian women, there was a significant positive association between weight change and a sitting time (Al-Musharaf et al., 2021). In another study, sedentary time was associated with an average weight gain of 0.53 kg in males and 0.46 kg in females (Liu et al., 2021). This association can be explained by increased snacking frequency, which leads to over-consumption of the next meal, eventually resulting in excess weight (Rodríguez-Pérez et al., 2020). Moreover, studies have shown that sedentary behaviour increased the risk of weight gain not only among obese individuals (Pellegrini et al., 2020), but also among those with normal BMI (Reyes-Olavarría et al., 2020).
The present study found that the stress level among Malaysian adults increased significantly ($p \leq 0.001$) during the pandemic, with $73.4\%$ of the population experiencing moderate stress. This is consistent with previous studies indicating that the pandemic has led to an increase in stress levels across different populations. The symptoms of mental distress were reported to be significantly higher in the USA (Ettman et al., 2020) and the UK (Pierce et al., 2020) during the pandemic than in pre-COVID. Increased stress is anticipated under the pandemic circumstances, particularly during the early phase of the lockdown, due to the presence of a novel and unknown virus (Al-Musharaf et al., 2021). Besides, constantly reading or hearing about the pandemic under restrictions due to the lockdown as well as the fear of disease and death can further deteriorate the stress level within a population (di Renzo et al., 2020).
Depression was observed to deteriorate significantly during the pandemic ($$p \leq 0.002$$) in the current study, with almost half of the population suffering from depression of varying degrees. This is in line with an Australian study that reported at least a doubling of depressive symptoms among 15,000 respondents (Curtis et al., 2021). Besides, our study reported a positive association between PHQ-9 score and weight change, which is corroborated by a previous study that indicated a significant association between depression and increased weight (Pellegrini et al., 2020). Additionally, depression score was reported to be independently associated with weight change whereby a 0.14 kg weight gain was observed in participants with a greater increase in depression score (Liu et al., 2021). To mitigate the negative experience of prolonged isolation and confinement, people became more susceptible to ‘food craving’ which often overrides the signals of satiety and hunger. Despite the advice to consume food containing high fat and sugar in moderation, chocolate and crisps were reported to be the most consumed snacks during the pandemic (Husain and Ashkanani, 2020). Ultimately, the over-consumption of ‘comfort food’ due to depression during the pandemic may contribute to an increased risk of weight gain and obesity.
Anxiety showed a similar trend, with the GAD-7 score increased significantly ($$p \leq 0.001$$) during the pandemic, and $49.4\%$ of the participants suffered from anxiety. This finding was in line with early studies that observed a higher prevalence of anxiety under lockdown circumstances in New Zealand (Sibley et al., 2020) and Spain (Rodríguez-Rey, Garrido-Hernansaiz and Collado, 2020). The deterioration of mental health during the pandemic can induce behavioural changes (di Renzo et al., 2020) as well as affect food choices (Cheng and Wong 2021). Besides, emotional eating associated with higher levels of stress and anxiety could lead to excessive intake of comfort food containing high amount of sugar and fat (Husain and Ashkanani, 2020). Furthermore, Malaysian adults reported an increased consumption of sugar-sweetened beverages (Cheng and Lau, 2022) and decreased intake of fruits and vegetables (Lo et al., 2022) during COVID-19. These findings suggest that stress, depression, and anxiety can synergistically cause adverse effects on body weight during the pandemic due to poor dietary choices. Hence, the importance of mental health during the pandemic should be emphasized.
The present study has several limitations that must be acknowledged. First, all data collected were self-reported by the participants. These data may be subjected to recall bias and over- and underestimation may be present to a certain degree. Secondly, an online questionnaire was used in this study and was disseminated through online platforms in which participants were recruited through the snowball sampling method. Therefore, oversampling of a particular group including a higher proportion of female gender and Chinese ethnicity in this study should be highlighted. Besides, dietary intake which could be a confounding factor in weight changes during the pandemic was not assessed in this study. Hence, a more inclusive longitudinal study assessing all possible risk factors and a larger sample size should be carried out to accurately determine the association between lifestyle behaviours, mental health, and weight change during the COVID-19 pandemic. Nonetheless, this study presented an insight into the changes in lifestyle behaviours and mental health along with their impacts on weight changes among Malaysian adults during the pandemic.
## Conclusion
In conclusion, weight gain was observed across all BMI categories among Malaysian adults during the COVID-19 pandemic. During the pandemic, physical inactivity and sedentary behaviour increased significantly, while the prevalence of stress, depression and anxiety has significantly heightened among Malaysian adults. In particular, increased sedentary behaviour was found to be associated with an increased risk of weight gain during the pandemic. While the lockdown has caused enormous changes in lifestyle and mental health leading to an increased risk of weight gain, these findings provide a preliminary insight for the development of nutrition and lifestyle guidelines to promote healthy body weight and well-being during the pandemic. Therefore, strategies to increase physical activity such as planning an exercise regimen should be implemented to reduce sedentary hours and prevent weight gain among Malaysian adults.
## References
1. Al-Musharaf S, Aljuraiban G, Bogis R. **Lifestyle changes associated with COVID-19 quarantine among young Saudi women: A prospective study**. *PLoS ONE* (2021) **16**
2. AlMughamis N, AlAsfour S, Mehmood S. **Poor eating habits and predictors of weight gain during the COVID-19 quarantine measures in Kuwait: A cross-sectional study**. *F1000Research* (2021) **9** 1-11
3. Alshahrani SM, Alghannam AF, Taha N. **The impact of COVID-19 pandemic on weight and body mass index in Saudi Arabia: A longitudinal study**. *Frontiers in Public Health* (2022) **9** 775022. PMID: 35111714
4. Ashdown-Franks G, Vancampfort D, Firth J. **Leisure-time sedentary behavior and obesity among 116,762 adolescents aged 12–15 years from 41 low- and middle-income countries**. *Obesity* (2019) **27** 830-836. PMID: 30859752
5. Bann D, Villadsen A, Maddock J. **Changes in the behavioural determinants of health during the COVID-19 pandemic: Gender, socioeconomic and ethnic inequalities in five British cohort studies**. *Journal of Epidemiology and Community Health* (2021) **75** 1136-1142. PMID: 34039660
6. Bhutani S, vanDellen MR, Cooper JA. **Longitudinal weight gain and related risk behaviors during the COVID-19 pandemic in adults in the US**. *Nutrients* (2021) **13** 671. PMID: 33669622
7. Cheng SH, Kamil M. **Stress and food intake among university students—Is there a relationship?**. *Sains Malaysiana* (2020) **49** 121-128
8. Cheng SH, Lau MY. **Increased consumption of sugar-sweetened beverages among Malaysian University students during the COVID-19**. *Malaysian Journal of Social Sciences and Humanities(MJSSH)* (2022) **7**
9. Cheng SH, Wong SE. **Stress, emotional eating and food choices among university students during the COVID-19**. *Malaysian Journal of Social Sciences and Humanities (MJSSH)* (2021) **6** 335-346
10. Chin YS, Woon FC, Chan YM. **The impact of movement control order during the COVID-19 pandemic on lifestyle behaviours and body weight changes: Findings from the MyNutriLifeCOVID-19 online survey**. *PLoS ONE* (2022) **17**
11. Cohen S, Kamarck T, Mermelstein R. **A global measure of perceived stress**. *Journal of Health and Social Behavior* (1983) **24** 385-396. PMID: 6668417
12. Curtis RG, Olds T, Ferguson T. **Changes in diet, activity, weight, and wellbeing of parents during COVID-19 lockdown**. *PLoS ONE* (2021) **16**
13. Di Renzo L, Gualtieri P, Pivari F. **Eating habits and lifestyle changes during COVID-19 lockdown: An Italian survey**. *Journal of Translational Medicine* (2020) **18** 229. PMID: 32513197
14. Drywień ME, Hamulka J, Zielinska-Pukos MA. **The COVID-19 pandemic lockdowns and changes in body weight among polish women. A cross-sectional online survey PLifeCOVID-19 study**. *Sustainability* (2020) **12** 7768
15. Ettman CK, Abdalla SM, Cohen GH. **Prevalence of depression symptoms in US adults before and during the COVID-19 pandemic**. *JAMA Network Open* (2020) **3**
16. He M, Xian Y, Lv X. (2020)
17. Husain W, Ashkanani F. **Does COVID-19 change dietary habits and lifestyle behaviours in Kuwait: A community-based cross-sectional study**. *Environmental Health and Preventive Medicine* (2020) **25** 61. PMID: 33045996
18. International Physical Activity Questionnaire (2004) Guidelines for Data Processing and Analysis of the International Physical Activity Questionnaire (IPAQ)-Short Form. Available at: www.ipaq.ki.se. (assessed 3 September 2022).. (2004)
19. Kriaucioniene V, Bagdonaviciene L, Rodríguez-Pérez C. **Associations between changes in health behaviours and body weight during the COVID-19 quarantine in Lithuania: The Lithuanian COVIDiet study**. *Nutrients* (2020) **12** 3119. PMID: 33065991
20. Kroenke K, Spitzer RL, Williams JB. **The PHQ-9: Validity of a brief depression severity measure**. *Journal of General Internal Medicine* (2001) **16** 606-613. PMID: 11556941
21. Lim JU, Lee JH, Kim JS. **Comparison of World Health Organization and Asia-Pacific body mass index classifications in COPD patients**. *International Journal of Chronic Obstructive Pulmonary Disease* (2017) **12** 2465-2475. PMID: 28860741
22. Liu S, Dun Y, Jeffrey R. (2021). DOI: 10.21203/rs.3.rs-69025/v2
23. Lo YL, Lee SS, Cheng SH. **Inadequate fruits and vegetables consumption among Malaysian adults during the COVID-19 pandemic**. *Nutrition and Health* (2022) **28** 741-750. PMID: 35522261
24. Pellegrini M, Ponzo V, Rosato R. **Changes in weight and nutritional habits in adults with obesity during the ‘lockdown’ period caused by the COVID-19 virus emergency**. *Nutrients* (2020) **12** 2016. PMID: 32645970
25. Pierce M, Hope H, Ford T. **Mental health before and during the COVID-19 pandemic: A longitudinal probability sample survey of the UK population**. *The Lancet Psychiatry* (2020) **7** 883-892. PMID: 32707037
26. Reyes-Olavarría D, Latorre-Román PÁ, Guzmán-Guzmán IP. **Positive and negative changes in food habits, physical activity patterns, and weight status during COVID-19 confinement: Associated factors in the Chilean population**. *International Journal of Environmental Research and Public Health* (2020) **17** 5431. PMID: 32731509
27. Rodríguez-Pérez C, Molina-Montes E, Verardo V. **Changes in dietary behaviours during the COVID-19 outbreak confinement in the Spanish COVIDiet study**. *Nutrients* (2020) **12** 1730. PMID: 32531892
28. 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**. *Frontiers in Psychology* (2020) **11** 1540. PMID: 32655463
29. Sharma SK, Mudgal SK, Rakhi KT. **How to calculate sample size for observational and experiential nursing research studies?**. *National Journal of Physiology, Pharmacy and Pharmacology* (2019) **10** 1-8
30. Sibley CG, Greaves LM, Satherley N. **Effects of the COVID-19 pandemic and nationwide lockdown on trust, attitudes toward government, and well-being**. *The American Psychologist* (2020) **75** 618-630. PMID: 32496074
31. Sidor A, Rzymski P. **Dietary choices and habits during COVID-19 lockdown: Experience from Poland**. *Nutrients* (2020) **12** 1657. PMID: 32503173
32. Spitzer RL, Kroenke K, Williams JBW. **A brief measure for assessing generalized anxiety disorder: The GAD-7**. *Archives of Internal Medicine* (2006) **166** 1092-1097. PMID: 16717171
33. Tan ST, Tan CX, Tan SS. **Physical activity, sedentary behavior, and weight status of university students during the COVID-19 lockdown: A cross-national comparative study**. *International Journal of Environmental Research and Public Health* (2021) **18** 7125. PMID: 34281062
34. World Health Organization (2022) Physical Activity. Available at: https://www.who.int/health-topics/physical-activity#tab=tab_1 (accessed April 24, 2022).. (2022)
35. Xia QS, Wu F, Gong M. (2020). DOI: 10.21203/rs.3.rs-55697/v1
36. Zheng C, Huang WY, Sheridan S. **COVID-19 pandemic brings a sedentary lifestyle in young adults: A cross-sectional and longitudinal study**. *International Journal of Environmental Research and Public Health* (2020) **17** 6035. PMID: 32825092
37. Zhu Q, Li M, Ji Y. **“Stay-at-Home” lifestyle effect on weight gain during the COVID-19 outbreak confinement in China**. *International Journal of Environmental Research and Public Health* (2021) **18** 1813. PMID: 33673375
|
---
title: Reversible skin microvascular hyporeactivity in patients with immune-mediated
thrombocytopenic thrombotic purpura
authors:
- Jérémie Joffre
- Lisa Raia
- Tomas Urbina
- Vincent Bonny
- Paul Gabarre
- Louai Missri
- Jean-Luc Baudel
- Paul Coppo
- Bertrand Guidet
- Eric Maury
- Hafid Ait-Oufella
journal: Critical Care
year: 2023
pmcid: PMC10028781
doi: 10.1186/s13054-023-04405-w
license: CC BY 4.0
---
# Reversible skin microvascular hyporeactivity in patients with immune-mediated thrombocytopenic thrombotic purpura
## Abstract
### Background
Immune-mediated thrombotic thrombocytopenic purpura (iTTP) is a rare disease characterized by arteriolar and capillary microthrombosis precipitating organ failure. However, the contribution of endothelial dysfunction on impaired microvascular blood flow in iTTP patients has been poorly explored. This pilot observational study aimed to explore endothelial-mediated vasoreactivity in iTTP patients at admission and its changes after plasma exchange therapy (PE).
### Methods
We conducted a prospective observational study in patients (> 18-year old) admitted in ICU for iTTP. Using laser Doppler flowmetry and acetylcholine (Ach) iontophoresis in the forearm, we recorded the skin microvascular blood flow and the endothelium-mediated vasoreactivity at admission and after PE. Demographics, biological, clinical courses, and outcomes were also collected. As a control group, we used a previously published cohort of young diabetic patients after correction of ketoacidosis.
### Results
Eighteen confirmed iTTP patients and 34 controls were included in the study, mainly female ($72\%$) aged 43 ± 16-year-old. At admission, $55\%$ had neurological abnormalities, $50\%$ cardiac issues and $27.8\%$ an acute kidney injury. Median platelet count was 19 G/mL [10–37]. Baseline microvascular blood flow was decreased in iTTP patients when compared to controls (5.97 ± 4.5 vs. 10.1 ± 6.3 PU, $$P \leq 0.03$$), associated with markedly impaired endothelial-mediated skin microvascular reactivity (AUC: 9627 ± 8122 vs. 16,475 ± 11,738, $$P \leq 0.03$$). Microvascular reactivity improved after the first PE session (AUC: 9627 ± 8122 vs 16,558 ± 10,699, $$P \leq 0.007$$, respectively, baseline and post-PE1) and much more after the second session (26,431 ± 23,181, $$P \leq 0.04$$ post-PE1 vs post-PE2). Hemolysis biomarkers (LDH and bilirubin) negatively correlated with skin microvascular flow and vasoreactivity.
### Conclusion
We highlighted a marked yet reversible skin endothelium-mediated microvascular hyporeactivity in iTTP patients that could participate in organ injury pathophysiology.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13054-023-04405-w.
## Introduction
Acquired or immune-mediated thrombotic thrombocytopenic purpura (TTP) is a rare thrombotic microangiopathy characterized by thrombocytopenia and hemolytic anemia [1–3]. Immune TTP is due to the presence of neutralizing anti-ADAMTS-13 autoantibodies responsible for impaired cleavage of the von Willebrand factor (VWF) mega multimers [4, 5]. Ultimately, VWF–platelet aggregates provoke microvascular thrombosis leading to inadequate microvascular blood flow, tissue ischemia and multiorgan failure. Therefore, the iTTP pattern combines hemolytic anemia, thrombocytopenia, often with neurologic, cardiac, or renal abnormalities, still associated with a 10–$20\%$ death rate [6, 7]. Current treatment consists of plasma exchange (PE) [8–10] combined with immunosuppressive therapy (e.g., glucocorticoids and rituximab) and caplacizumab [11], an anti–VWF factor monoclonal humanized antibody inhibiting interaction between VWF multimers and platelets [12, 13].
Arteriolar and capillary microthrombosis due to the accumulation of VWF–platelet aggregates lead to life-threatening organ hypoperfusion, affecting the heart and the brain. However, the consequences of VWF–platelet aggregates on the endothelium, a key regulator of blood flow, remain unknown. This observational study aimed to explore endothelial-dependent microvascular reactivity in iTTP patients in the intensive care unit (ICU) at admission and after treatment.
## Methods
We conducted a prospective observational study in our tertiary university hospital. We included iTTP patients (> 18-year old) experiencing their first acute event, admitted to our ICU between January 2016 and September 2022. Clinical and biological parameters were recorded. Skin microcirculatory reactivity in the right forearm area was recorded at ICU admission (baseline) and after PE. As a control group for microvascular reactivity, we used data from a previously published cohort of young diabetic patients recorded after the correction of metabolic acidosis [14]. The local ethical committee approved the protocol (Comité de Protection des Personnes, Hôpital Saint-Louis, France, No $\frac{2015}{64}$NI), and the database was registered according to the French legislation (No 2,228,742), and all patients consented to anonymous data use for academic research and publication. It was a noninvasive observational study without any specific intervention. All patients were managed following international guidelines for TTP [15] and in collaboration with the physician of the thrombotic microangiopathy national reference center. All patients received urgent therapeutic PE (1.5 plasmatic mass, $100\%$ fresh frozen plasma (FFP)), corticosteroids, and $\frac{17}{18}$ patients received caplacizumab on the first day of ICU admission.
## Skin microcirculatory endothelial function assessment
We recorded microvascular parameters at baseline and after PE, using laser Doppler flowmetry and acetylcholine iontophoresis in the forearm area (Additional file 1: Figure S1). Methods have been previously described and validated by our group in different clinical settings (15–18). Briefly, a calibrated laser Doppler flow meter probe (Periflux 5000; Perimed, Craponne, France) embedded within a drug delivery chamber loaded with 80 μg of acetylcholine (Miochol; Novartis, Cedex, France) was used in combination with a current delivering generator. After 1 min of baseline microvascular blood flow recording, three successive current pulses (0.12 mA, 12 ms) were delivered, leading to acetylcholine diffusion within small skin vessels. Microcirculatory skin blood flow was recorded for 10 min following the first impulse. Baseline blood flow (expressed as flow index), maximal blood flow (peak value), and area under the curve (AUC) after acetylcholine iontophoresis were determined for each patient at each time point, and curves were blindly analyzed offline (a representative record is shown as Additional file 1: Fig. S1).
## Statistics
Continuous variables were presented as mean ± SD or median and 25th–75th interquartile ranges (IQR). Discrete variables were presented as percentages. Comparisons between groups were made with Fisher test for discrete variables and Mann–Whitney U test for continuous variables. Comparisons between admission and post-PE values were made using a paired Wilcoxon signed-rank test. Statistical analysis and graphical representations were performed using GraphPad Prism 10.2 software (Graph Pad Software Inc., La Jolla, CA). A two-sided P-value of less than 0.05 was considered statistically significant.
## Results
Eighteen consecutive iTTP patients were included, $72\%$ were female, aged 43 ± 16-year-old. ADAMTS13 activity at baseline was below $10\%$ in all included patients [16] (below $5\%$ in 15 patients ($83.3\%$)) and all included patients had circulating anti-ADAMTS13 autoantibodies, which confirmed the final diagnosis of iTTP. At ICU admission, $55\%$ had neurological abnormalities, $50\%$ cardiac (troponin elevation, EKG, or echocardiography) issues and $27.8\%$ had stage ⩾1 acute kidney injury according to KDIGO classification [17]. At baseline, all included patients had severe thrombocytopenia with median platelet count at 19 G/mL [10–37] G/L, mild regenerative anemia (hemoglobin: 9.6 g/dl [7.6–10], reticulocytes 178 [128–285]G/L) associated with hemolysis markers (median bilirubin: 36 µmol/L [23–59], haptoglobin: 0 g/dL [0–0.035]). Only one patient was under mechanical ventilation and received vasopressors. None received renal replacement therapy. All patients were treated by PE (median number of PE: 4 [2–5]), $100\%$ corticosteroids (methylprednisolone 1 mg/Kg/Day i.v), and $88.9\%$ caplacizumab (one injection i.v. before the first PE, then 10 mg/Day s.c). Patients' baseline characteristics are reported in Table 1 and initial treatment and ICU stay characteristics are summarized in Additional file 1: Table S1. Overall, the in-ICU length of stay was 6.4 [4.8–8] days, and one patient died in ICU ($5.5\%$). Table 1Baseline patients’ characteristicsBaseline patient’s characteristicsiTTP ($$n = 18$$)Controls ($$n = 34$$)P value Age (years. Mean ± SD)43 ± 1644 ± 150.86 Male (n%)5 [28]23 [68]0.009Medical history (n%) Diabetes mellitus3 [17]34 [100] < 0.0001 Hypertension5 [28]7 [21]0.73 Cardiopathy1 (5.6)3 (8.8) > 0.99 Peripheral vascular disease1 (5.6)1 (2.9) > 0.99 CKD0 [0]7 [21]0.08 Neurological disease1 (5.6)0 [0]0.35 Cirrhosis0 [0]1 (2.9) > 0.99Vitals (Mean ± SD) Core temperature (°C)37 ± 0.8537 ± 0.340.41 SBP (mmHg)135 ± 20135 ± 250.87 HR (bpm)89 ± 12106 ± 140.0001 CRT (sec)1.3 ± 1.21.7 ± 0.90.15 Mottling score0.3 ± 0.80.2 ± 0.50.74Biologicals (Median [IQR]) Hemoglobin (g/dL)9.6 [7.6–10]13 [12–15] < 0.0001 WBC (103/mL)10.59 [7.8–13.6]10.51[6.5–14.4]0.87 Platelets (106/mL)19.5 [9.75–37]229 [145–272] < 0.0001 Urea (mmol/L)6.5 [5.1–9.2]4.7 [3.1–9.5]0.05 Creatinine (µmol/L)73 [54–115]76 [48–102]0.99 Haptoglobin (g/L)0 [0–0.035]–– Bilirubin (µmol/L)36 [23–59]–– LDH (UI/L)1494 [942–2423]–– Schizocytes (%)2.5 [1.5–4.4]–– *Troponin us* (ng/L)13 [0.52–236]–– BNP (ng/L)68 [20–155]–– Lactate (mmol/L)1.4 [0.9–2.1]1.1 [0.8–1.9]0.55 ADAMTS 13 < $5\%$15 (83.3)–– < $10\%$3 (16.7)–– Ab Anti-ADAMTS13 positivity18 [100]––SAPS II17.2 ± 12.730.1 ± 18 < 0.0001iTTP immune-mediated thrombocytopenic thrombotic purpura, SD standard deviation, BMI body mass index, CKD chronic kidney disease, SBP systolic blood pressure, HR heart rate, IQR interquartile range, WBC white blood cells, LDH lactate dehydrogenase, BNP brain natriuretic peptide, Ab antibodies, SOFA Sequential Organ Failure Assessment, CRT capillary refill time First, we compared the microvascular reactivity of iTTP patients with a cohort of diabetic patients admitted to our ICU after correction of keto-acidosis. Such a control cohort was relevant because patients were young with rare co-morbidities (Table 1) and no severe organ failure. At admission, when compared to the control group (Fig. 1 and Additional file 1: Table S2), we observed that iTTP patients had twofold lower skin microvascular blood flow (5.97 ± 4.5 vs. 10.1 ± 6.3 PU, $$P \leq 0.03$$) (Fig. 1A). In addition, we found marked impaired endothelial-mediated microvascular reactivity in iTTP patients characterized by a lower peak after *Ach iontophoresis* (31.9 ± 19.1 vs. 67.7 ± 39.9, $$P \leq 0.001$$) (Fig. 1B) and ultimately a lower AUC (9627 ± 8122 vs. 16,475 ± 11,738, $$P \leq 0.03$$) (Fig. 1C) (Fig. 2). Fig. 1Skin microvascular endothelium-mediated reactivity assessed by laser Doppler flowmetry at baseline in iTTP and controls. Comparison of skin microvascular laser Doppler flowmetry value between controls and iTTP patients at admission regarding the baseline flow index (A) and the response to *Ach iontophoresis* (Peak value (B) and AUC (C)). * $P \leq 0.05$, **$P \leq 0.01$, CTR versus ITTP, two-tailed Mann–Whitney U test. B Abbreviations: PU, perfusion index; CTR, controls, iTTP, immune-mediated thrombocytopenic thrombotic purpura; AUC, area under curveFig. 2iTTP patients’ skin microvascular endothelium-mediated reactivity assessed by laser Doppler flowmetry at baseline and after PE. Evolution of skin microvascular laser Doppler flowmetry value regarding the baseline flow index (A) and the response to *Ach iontophoresis* (Peak value (B) and AUC (C)), in iTTP patients at admission and after the two first PE. * $P \leq 0.05$, **$P \leq 0.01$, paired Wilcoxon signed-rank test at each time point versus admission value for. Abbreviations: PU, perfusion index; CTR, controls, iTTP, immune-mediated thrombocytopenic thrombotic purpura; PE, plasma exchange; AUC, area under curve; Ach, Acetylcholine Next, on iTTP patients, we analyzed the impact of combined treatment on endothelial-dependent microvascular hyporeactivity during ICU stay. Acetylcholine iontophoresis was repeated after the first and the second PE session. After the first PE session, platelet count significantly increased (26 ± 28 vs 42 ± 38 G/mL, $$P \leq 0.0003$$) while hemolysis parameters improved (LDH 1870 ± 1440 vs. 650 ± 203 UI/mL, $P \leq 0.0001$, Haptoglobin 0.09 ± 0.2 vs. 0.56 ± 0.16 g/L, $P \leq 0.0001$) and biological recovery was more pronounced after the second session (Additional file 1: Table S1, Fig. 3 and Additional file 1: Fig. S2A). We observed that global microvascular blood flow significantly increased after the first PE session (Baseline perfusion index: 5.97 ± 4.5 PU at admission vs 11.38 ± 8.6 post-PE1, $$P \leq 0.027$$,) and even more after the second session (Baseline perfusion index 12.89 ± 6.9 PU, $$P \leq 0.008$$ vs admission). Global microvascular reactivity improved after the first PE session (AUC: 9627 ± 8122 vs 16,558 ± 10,699, $$P \leq 0.007$$, respectively, baseline and post-PE1) and much more after the second session (26,431 ± 23,181, $$P \leq 0.04$$ post-PE1 vs post-PE2) (Fig. 2A–C and Additional file 1: Table S2). Changes in microvascular reactivity after PE were heterogeneous, some patients improved after the first PE while others improved their microvascular reactivity after the second. Finally, few iTTP patients had no variation of skin microvascular response to Ach (Additional file 1: Fig. S3). Figure 4 shows an archetypical example of endothelium-mediated microvascular hyporeactivity in a single patient which improved after the plasma exchange session and much more after the second session. Interestingly, microvascular blood flow across time positively correlated with platelet but not the vasoreactivity (Additional file 1: Fig. S2A). Hemolysis biomarkers (LDH and bilirubin) negatively correlated with microvascular flow and reactivity (Fig. 3). Conversely, we observed no correlation between microvascular flow/ reactivity and hemoglobin, haptoglobin, schizocytes or reticulocytes (Additional file 1: Fig. S2B).Fig. 3Biological variations during ICU stay and correlation with microvascular perfusion parameters. Courses of biomarkers in iTTP patients during the first days in ICU and correlation with flowmetry values. Hemolysis biomarkers (LDH and bilirubin) negatively correlate with microvascular flow and reactivity ***$P \leq 0.0001$, versus admission value, paired Wilcoxon signed-rank test at each time point. On correlations graph, the full line represents the linear regression and the dotted line show the $95\%$IC. Abbreviations: iTTP, immune-mediated thrombocytopenic thrombotic purpura; PE, plasma exchange; LDH, lactate dehydrogenase; PU, perfusion unitFig. 4Archetypical record of the gradual improvement of skin microvascular reactivity following PE. Example of the gradual improvement of the skin microvascular reactivity in a single iTTP patient. Arrows indicate the successive *Ach iontophoresis* application. Abbreviations: PU, perfusion index; iTTP, immune-mediated thrombocytopenic thrombotic purpura; PE, plasma exchange; Ach, Acetylcholine
## Discussion
In this prospective study, we showed a markedly impaired skin microvascular endothelial-mediated reactivity in iTTP patients which recovered quickly after plasma exchange therapy.
This profoundly impaired microvascular vasoreactivity is similar to what our group previously observed in other critical conditions characterized by patent endothelial dysfunction such as critically ill COVID-19 [18] septic shock [19] or severe keto-acidosis [14]. Endothelial cell (EC) involvement in the pathophysiology of thrombotic microangiopathy-associated organ failure has been suggested in animal models but remains poorly demonstrated in humans. Indeed, experimental models indicated that Adamts-13 deficiency, by itself, is not sufficient to trigger thrombotic microangiopathy. Endothelial activation is another necessary step to induce microvascular disease, probably by releasing of a large amount of UL-VWF [20–22]. In the same line, in primates, the injection of human anti-ADAMTS-13 neutralizing autoantibody provokes a transient biological thrombotic microangiopathy but not a severe disease responsible for organ failure [23]. In iTTP patients, circulating EC number and plasma biomarkers reflecting endothelial activation are increased and correlated with the outcome supporting a role of the endothelium in the pathophysiology of iTTP [24]. Recently, Tellier et al., reported that several plasmatic components, including anti-ADAMTS13 IgG, free heme and possibly others converge to induce EC activation ex vivo [25]. Interestingly the intensity of the ex vivo EC activation induced by iTTP patients' plasmas correlated with the disease severity [25]. Among soluble factors released in iTTP, hemolysis-derived products (including cell-free heme, free-hemoglobin and bilirubin at high concentration [26, 27]) are well known to be highly toxic for the endothelium [28, 29]. Moreover, in a murine model, Frei et al. reported that hemolysis causes direct vascular injury and functionally impaired vasodilation via increased scavenging of nitric oxide by plasma free hemoglobin [30]. Interestingly, in our study, we observed a negative correlation between endothelium-mediated skin microvascular vasodilation and hemolysis parameters.
We showed that the microvascular reactivity of iTTP patients is impaired and, therefore, could participate in organ injury besides the thrombotic process. Moreover, the endothelial vasoreactivity is restored after plasma exchange therapy at the same time than platelet count recovered. However, the improvement of endothelial reactivity could not be directly linked to exchange plasma therapy because at the same time, almost all the patient received additional treatment including steroids and Caplacizumab. Currently, there is no experimental data about the effect of Caplacizumab on endothelial microvascular reactivity [31].
We acknowledge some limitations to this observational and translational study. First, this is a single-center study with a limited number of patients. Second, given the multiple treatments received simultaneously and the synchronous correction of thrombocytopenia and hemolysis, one cannot speculate on which biological mechanism is responsible for the microcirculatory improvement. Next, as a control group, we used a previously published cohort of young diabetic patients after correction of ketoacidosis where we showed that microvascular hyporeactivity recovered after acidosis correction [14] Given that patients with cardiovascular risk factors are susceptible to a lower microvascular reactivity; difference between healthy subjects and iTTP patients could be even more important than differences between diabetic and iTTP patients [32–35]. The device used in this study only allows exploration of the skin microvasculature and we did not investigate the endothelium of key organs affected by iTTP such as the brain, heart and kidney microcirculation [36, 37]. Finally, we did not explore the endothelium-independent vasodilation, which requires either heating or nitroprusside challenge. Thus, we cannot rule out that iTTP patients have impaired endothelium-independent vasodilation reserve [38] or decreased NO bioavailability [39], on top of the observed impaired endothelium-mediated vasoreactivity.
## Conclusion
This prospective observational study highlights a marked endothelium-mediated microvascular hyporeactivity in acute iTTP patients that could participate in organ injury pathophysiology. Moreover, endothelium-mediated vasoreactivity dysfunction quickly recovered after PE therapy.
## Supplementary Information
Additional file 1. Supplementary Figures and Tables.
## References
1. Moake JL. **Thrombotic microangiopathies**. *N Engl J Med* (2002.0) **347** 589-600. DOI: 10.1056/NEJMra020528
2. George JN. **Clinical practice: thrombotic thrombocytopenic purpura**. *N Engl J Med* (2006.0) **354** 1927-1935. DOI: 10.1056/NEJMcp053024
3. George JN, Nester CM. **Syndromes of thrombotic microangiopathy**. *N Engl J Med* (2014.0) **371** 654-666. DOI: 10.1056/NEJMra1312353
4. Rieger M, Mannucci PM, Kremer Hovinga JA, Herzog A, Gerstenbauer G, Konetschny C, Zimmermann K, Scharrer I, Peyvandi F, Galbusera M. **ADAMTS13 autoantibodies in patients with thrombotic microangiopathies and other immunomediated diseases**. *Blood* (2005.0) **106** 1262-1267. DOI: 10.1182/blood-2004-11-4490
5. Thomas MR, de Groot R, Scully MA, Crawley JT. **Pathogenicity of anti-ADAMTS13 autoantibodies in acquired thrombotic thrombocytopenic purpura**. *EBioMedicine* (2015.0) **2** 942-952. DOI: 10.1016/j.ebiom.2015.06.007
6. Benhamou Y, Assie C, Boelle PY, Buffet M, Grillberger R, Malot S, Wynckel A, Presne C, Choukroun G, Poullin P. **Development and validation of a predictive model for death in acquired severe ADAMTS13 deficiency-associated idiopathic thrombotic thrombocytopenic purpura: the French TMA Reference Center experience**. *Haematologica* (2012.0) **97** 1181-1186. DOI: 10.3324/haematol.2011.049676
7. Peigne V, Perez P, Resche Rigon M, Mariotte E, Canet E, Mira JP, Coppo P, Veyradier A, Azoulay E. **Causes and risk factors of death in patients with thrombotic microangiopathies**. *Intensive Care Med* (2012.0) **38** 1810-1817. DOI: 10.1007/s00134-012-2638-5
8. Adamski J. **Thrombotic microangiopathy and indications for therapeutic plasma exchange**. *Hematology Am Soc Hematol Educ Progr* (2014.0) **2014** 444-449. DOI: 10.1182/asheducation-2014.1.444
9. Nguyen TC, Han YY. **Plasma exchange therapy for thrombotic microangiopathies**. *Organogenesis* (2011.0) **7** 28-31. DOI: 10.4161/org.7.1.14027
10. Winters JL. **Plasma exchange in thrombotic microangiopathies (TMAs) other than thrombotic thrombocytopenic purpura (TTP)**. *Hematol Am Soc Hematol Educ Progr* (2017.0) **2017** 632-638. DOI: 10.1182/asheducation-2017.1.632
11. Scully M, Cataland SR, Peyvandi F, Coppo P, Knobl P, Kremer Hovinga JA, Metjian A, de la Rubia J, Pavenski K, Callewaert F. **Caplacizumab treatment for acquired thrombotic thrombocytopenic purpura**. *N Engl J Med* (2019.0) **380** 335-346. DOI: 10.1056/NEJMoa1806311
12. George JN. **Corticosteroids and rituximab as adjunctive treatments for thrombotic thrombocytopenic purpura**. *Am J Hematol* (2012.0) **87** S88-91. DOI: 10.1002/ajh.23126
13. Sukumar S, Lammle B, Cataland SR. **Thrombotic Thrombocytopenic Purpura: Pathophysiology, Diagnosis, and Management**. *J Clin Med* (2021.0) **10** 536. DOI: 10.3390/jcm10030536
14. Joffre J, Bourcier S, Hariri G, Miailhe AF, Bige N, Dumas G, Dubee V, Boelle PY, Abdallah I, Baudel JL. **Reversible microvascular hyporeactivity to acetylcholine during diabetic ketoacidosis**. *Crit Care Med* (2018.0) **46** e772-e778. DOI: 10.1097/CCM.0000000000003224
15. Zheng XL, Vesely SK, Cataland SR, Coppo P, Geldziler B, Iorio A, Matsumoto M, Mustafa RA, Pai M, Rock G. **ISTH guidelines for treatment of thrombotic thrombocytopenic purpura**. *J Thromb Haemost* (2020.0) **18** 2496-2502. DOI: 10.1111/jth.15010
16. Lotta LA, Wu HM, Musallam KM, Peyvandi F. **The emerging concept of residual ADAMTS13 activity in ADAMTS13-deficient thrombotic thrombocytopenic purpura**. *Blood Rev* (2013.0) **27** 71-76. DOI: 10.1016/j.blre.2013.01.001
17. 17.Summary of recommendation statements. Kidney Int Suppl (2011) 2012; 2(1): 8–12.
18. Raia L, Urbina T, Gabarre P, Bonny V, Hariri G, Ehrminger S, Bige N, Baudel JL, Guidet B, Maury E. **Impaired skin microvascular endothelial reactivity in critically ill COVID-19 patients**. *Ann Intensive Care* (2022.0) **12** 51. DOI: 10.1186/s13613-022-01027-3
19. Bourcier S, Joffre J, Dubee V, Preda G, Baudel JL, Bige N, Leblanc G, Levy BI, Guidet B, Maury E. **Marked regional endothelial dysfunction in mottled skin area in patients with severe infections**. *Crit Care* (2017.0) **21** 155. DOI: 10.1186/s13054-017-1742-x
20. Joffre J, Hellman J, Ince C, Ait-Oufella H. **Endothelial responses in sepsis**. *Am J Respir Crit Care Med* (2020.0) **202** 361-370. DOI: 10.1164/rccm.201910-1911TR
21. Raia L, Zafrani L. **Endothelial activation and microcirculatory disorders in sepsis**. *Front Med* (2022.0) **9** 907992. DOI: 10.3389/fmed.2022.907992
22. Motto DG, Chauhan AK, Zhu G, Homeister J, Lamb CB, Desch KC, Zhang W, Tsai HM, Wagner DD, Ginsburg D. **Shigatoxin triggers thrombotic thrombocytopenic purpura in genetically susceptible ADAMTS13-deficient mice**. *J Clin Invest* (2005.0) **115** 2752-2761. DOI: 10.1172/JCI26007
23. Feys HB, Roodt J, Vandeputte N, Pareyn I, Lamprecht S, van Rensburg WJ, Anderson PJ, Budde U, Louw VJ, Badenhorst PN. **Thrombotic thrombocytopenic purpura directly linked with ADAMTS13 inhibition in the baboon (**. *Blood* (2010.0) **116** 2005-2010. DOI: 10.1182/blood-2010-04-280479
24. Widemann A, Pasero C, Arnaud L, Poullin P, Loundou AD, Choukroun G, Sanderson F, Lacroix R, Sabatier F, Coppo P. **Circulating endothelial cells and progenitors as prognostic factors during autoimmune thrombotic thrombocytopenic purpura: results of a prospective multicenter French study**. *J Thromb Haemost* (2014.0) **12** 1601-1609. DOI: 10.1111/jth.12681
25. Tellier E, Widemann A, Cauchois R, Faccini J, Lagarde M, Brun M, Robert P, Robert S, Bachelier R, Poullin P. **Immune thrombotic thrombocytopenic purpura plasmas induce calcium- and IgG-dependent endothelial activation: correlations with disease severity**. *Haematologica* (2022.0). DOI: 10.3324/haematol.2022.280651
26. Maruhashi T, Kihara Y, Higashi Y. **Bilirubin and endothelial function**. *J Atheroscler Thromb* (2019.0) **26** 688-696. DOI: 10.5551/jat.RV17035
27. Brito MA, Palmela I, Cardoso FL, Sa-Pereira I, Brites D. **Blood-brain barrier and bilirubin: clinical aspects and experimental data**. *Arch Med Res* (2014.0) **45** 660-676. DOI: 10.1016/j.arcmed.2014.11.015
28. Frimat M, Boudhabhay I, Roumenina LT. **Hemolysis derived products toxicity and endothelium: model of the second hit**. *Toxins* (2019.0) **11** 660. DOI: 10.3390/toxins11110660
29. Aslan M, Ryan TM, Adler B, Townes TM, Parks DA, Thompson JA, Tousson A, Gladwin MT, Patel RP, Tarpey MM. **Oxygen radical inhibition of nitric oxide-dependent vascular function in sickle cell disease**. *Proc Natl Acad Sci USA* (2001.0) **98** 15215-15220. DOI: 10.1073/pnas.221292098
30. Frei AC, Guo Y, Jones DW, Pritchard KA, Fagan KA, Hogg N, Wandersee NJ. **Vascular dysfunction in a murine model of severe hemolysis**. *Blood* (2008.0) **112** 398-405. DOI: 10.1182/blood-2007-12-126714
31. Goshua G, Sinha P, Hendrickson JE, Tormey C, Bendapudi PK, Lee AI. **Cost effectiveness of caplacizumab in acquired thrombotic thrombocytopenic purpura**. *Blood* (2021.0) **137** 969-976. DOI: 10.1182/blood.2020006052
32. Debbabi H, Bonnin P, Ducluzeau PH, Leftheriotis G, Levy BI. **Noninvasive assessment of endothelial function in the skin microcirculation**. *Am J Hypertens* (2010.0) **23** 541-546. DOI: 10.1038/ajh.2010.10
33. Kihara M, Low PA. **Impaired vasoreactivity to nitric oxide in experimental diabetic neuropathy**. *Exp Neurol* (1995.0) **132** 180-185. DOI: 10.1016/0014-4886(95)90023-3
34. Debbabi H, Bonnin P, Levy BI. **Effects of blood pressure control with perindopril/indapamide on the microcirculation in hypertensive patients**. *Am J Hypertens* (2010.0) **23** 1136-1143. DOI: 10.1038/ajh.2010.115
35. Vuletic V, Cengic L, Basic S, Sporis D, Rahelic D, Demarin V. **Impaired cerebral vasoreactivity in type 2 diabetes mellitus**. *Coll Antropol* (2011.0) **35** 27-31. PMID: 21661351
36. Gomez-Segui I, Pascual Izquierdo C, Mingot Castellano ME, de la Rubia CJ. **An update on the pathogenesis and diagnosis of thrombotic thrombocytopenic purpura**. *Expert Rev Hematol* (2023.0) **16** 17. DOI: 10.1080/17474086.2023.2159803
37. Fodil S, Zafrani L. **Severe thrombotic thrombocytopenic purpura (TTP) with organ failure in critically Ill patients**. *J Clin Med* (2022.0) **11** 1103. DOI: 10.3390/jcm11041103
38. Alba BK, Greaney JL, Ferguson SB, Alexander LM. **Endothelial function is impaired in the cutaneous microcirculation of adults with psoriasis through reductions in nitric oxide-dependent vasodilation**. *Am J Physiol Heart Circ Physiol* (2018.0) **314** H343-H349. DOI: 10.1152/ajpheart.00446.2017
39. Greaney JL, Saunders EFH, Santhanam L, Alexander LM. **Oxidative stress contributes to microvascular endothelial dysfunction in men and women with major depressive disorder**. *Circ Res* (2019.0) **124** 564-574. DOI: 10.1161/CIRCRESAHA.118.313764
|
---
title: How does parents’ social support impact children’s health practice? Examining
a mediating role of health knowledge
authors:
- Paulin Tay Straughan
- Chengwei Xu
journal: Global Health Research and Policy
year: 2023
pmcid: PMC10028785
doi: 10.1186/s41256-023-00291-5
license: CC BY 4.0
---
# How does parents’ social support impact children’s health practice? Examining a mediating role of health knowledge
## Abstract
### Background
Family environmental factors play a vital role in shaping children’s health practices (e.g., obesity prevention). It is still unclear how parents’ social support affects children’s obesity-related health practices. The present study argues that whether parents’ social support positively associates with children’s obesity-related health practice depends on if it could promote parents’ obesity-related health knowledge. Thus, we hypothesize that health knowledge mediates the relationship between parents’ social support and children’s health practice regarding weight management.
### Methods
To test the hypothesis, we conducted a questionnaire survey and collected a nationally representative sample of 1488 household responses in Singapore. The survey included questions about parents’ social support, health knowledge, children’s health practices, and socio-demographic variables. All participants have at least one child 14 years old or younger. In the sample, $66.1\%$ of the respondents are female, and $93.7\%$ are below 50 years old. Structural equation modeling (SEM) via Stata was used to examine the associations between parents’ social support, health knowledge, and children’s health practice.
### Results
The results of our analysis support our hypothesis. Specifically, [1] parents’ social support shows a positive relationship with health knowledge (Coef. = 0.17, $p \leq 0.001$ for BMI knowledge and Coef. = 0.18, $p \leq 0.001$ for nutrition knowledge); [2] parents’ social support (total effect of social support = 0.081, $$p \leq 0.071$$) and health knowledge positively associate with children’s obesity-related health practice (coefficient of BMI knowledge = 0.10, $p \leq 0.01$; coefficient of nutrition knowledge = 0.31, $p \leq 0.001$); and [3] the effects of parents’ social support on children’s health practice is fully mediated by parents’ health knowledge (mediating effect = $100\%$, $$p \leq 0.007$$).
### Conclusion
The present study provides fresh evidence from a multicultural context to understand the relationships between parents’ social support, health knowledge, and children’s obesity-related health practice. Our findings support the argument that social support from parents’ social networks does not necessarily promote health outcomes. The only social support that carries proper health knowledge can facilitate good health practice.
## Background
The increasing prevalence of childhood obesity has been one of the most challenging issues faced by both developed and developing countries [1]. Between 1980 and 2013, the rate of childhood overweight and obesity jumped from 16.6 to $23.2\%$ in developed countries, and from 8.3 to $13.2\%$ in developing countries [2]. In response to the heightened concern, parental influence over child weight merits attention. Many studies support the view that parents are highly responsible for childhood obesity and obesity prevention practices [3, 4]. Golan and colleagues even suggested that health promotion programs focusing on parents only are more effective than that involving both parents and their children with obesity [5]. How parental factors impacts children’s health practice merits attention. Notably, physical activity and food intake control have been identified as two critical means of intervention for parents to manage their children’s body weight. Thus, the present study focuses on children’s participation in physical activities and diet control as the primary obesity-related health practice.
Parents’ social support is critical for parenting practice, and whether social support has a positive impact depends very much on their social environment, in particular who parents can draw knowledge and advisories from [6, 7]. Social support for parenting comprises both formal and informal support. Formal support is conceptualized as caregiving help provided by professionals and formal organizations, where assistance is governed by contractual rather than affiliative norms [8, 9]. The operationalization of formal support typically determines if the care recipient and/or caregiver uses specific services. Typical services include home health, daycare, support groups, transportation, and referral services. In contrast, informal social support tends to be provided within an individual’s network, comprising mainly family and friends [10]. Yet, while the relationship between an individual’s social support and their health and weight management has been explored [6], it has not been explored as thoroughly with regards to parent’s child-rearing practices. That is, there is still a dearth of literature focusing on the effects of parents’ social support on their children’s obesity-related health practice (e.g., physical exercise and eating less junk food).
Further, whether social support always produces positive impacts is questionable. Specifically, there is a lack of empirical evidence to demonstrate what is likely to enable social support to positively impact health behaviors, and conversely what causes social support to fail to make a difference or even have a negative effect. Although many studies indicate a positive relationship between social support and health practice [11, 12], some scholars presented inconsistent findings [13]. Several studies indicate that social support can, in some cases, bring positive effects, and in other cases, negative effects on health behaviors, depending on whether social support carries salutary health-related knowledge or inadequate knowledge may have adverse health effects [14, 15]. We thus can speculate that health knowledge may mediate the relationship between social support and health practice. Put differently, the present study argues that the only social support that carries proper health knowledge can facilitate good health practice.
The primary hypothesis of this paper posits that the relationship between parents’ social support and children’s obesity-related health practice is determined by parents’ health knowledge of weight management (e.g., knowledge about body mass index and what constitutes a healthy diet). In particular, the mediating effect of health knowledge on the relationship between social support and health practice is to be examined. Health knowledge refers to parents’ general understanding and awareness of what constitutes obesity, how obesity is measured, what constitutes a healthy diet, and parents’ ability to comprehend nutritional labels on food products. We expect that parents’ social support will enhance children's health practice only when parents are knowledgeable about how to manage body weight. Hence, parents’ obesity-related knowledge may mediate the relationship between social support and health practice.
To examine the hypotheses, empirical evidence is drawn from a nationally representative survey of 1,488 parents with children who are 14 years old and younger in Singapore. We use structural equation modeling (SEM) to verify the hypotheses of the associations between parents’ social support, health knowledge, and children’s health practice. The following section reviews the existing literature and presents the study’s hypotheses. Next, we introduce data, analysis, and results. Lastly, we discuss the findings, implications, future directions, and conclusions.
## Social support and health practices
The effect of social support on health outcomes has been an important research topic for the past four decades [14]. Social support refers to support and resources that an individual can receive from his or her social networks (e.g., family members, friends, relatives, colleagues, and neighborhood residents) [16]. It can be given in the form of problem-solving information or advice, positive interactions, emotional or affective support, or even tangible aid [17, 18]. The concept is important for health outcomes as scholarship has shown that when people face health problems, they are very likely to seek support from people within their close networks [19].
Current research has indicated that there is a causal relationship between social support and positive health practices and outcomes [11, 12]. It has been argued that information and assistance from friends and relatives can promote patients’ awareness of seeking medical care [20]. For instance, one study found that women with obesity who received frequent support from family and friends had a higher possibility of losing weight [6]. Evidence from participants from 16 countries also indicated that low social support is associated with less physical activities [21].
Scholars have also linked parental social support to parenting practices and their children’s health outcomes. For example, social support from relatives and friends has significantly influenced parents’ parenting capacities and practices [17, 22]. Other studies have illustrated that parents with greater social support from their extended kin are more likely to have healthier children [23]. One of the possible mechanisms identified that enables this could be that support from family and friends helps parents start and maintain a healthier lifestyle with their children (e.g., regularly participating in physical exercises or eating less junk food) [24–26]. In light of this, we hypothesize that (see Fig. 1): Fig. 1Hypotheses
## H1
Parents’ social support is positively associated with their children’s health practices.
## Social support and health knowledge
An important dynamic in the way social support influences health practice is the transmission of health knowledge. Despite the lack of consensus on the definition, health-related knowledge is a concept that is commonly agreed upon to be key within the health literacy framework [27]. Baker argues that health-related knowledge could facilitate the development of health literacy as it is with prior knowledge that the individual can comprehend health information and is, in that sense, literate [28]. This runs counter to frameworks developed by Nutbeam and colleagues, who deem health knowledge as an aspect of health literacy [27, 29]. They posit that health knowledge is necessary as it is what the individual acts upon to be considered health literate.
Social support serves as a critical means for individuals to gain health knowledge. Coleman argued that social support is important in gaining new information and serves as information channels, whereby the use of social relations with others provides the means through which one can acquire more knowledge. He further argued that while individuals may maintain social relations for other purposes, knowledge is also passed through in the process [30]. In House’s conceptualization of social support [31], informational support, or the provision of useful information, is one of the four forms of providing social support. Studies have also shown that individuals tend to seek health information from interpersonal sources as they may provide information tailored to their needs [32, 33]. A person’s social support may significantly impact his or her health knowledge. Therefore, we hypothesize that:
## H2
Parents’ social support is positively associated with parents’ obesity-related health knowledge.
## Health knowledge and health practices
The impact of health knowledge on an individual’s health practice has been well explored. It has been argued that knowledge plays a major role in behavioral change [34]. Particular to health behavior, individuals with adequate health knowledge tend to adopt more preventive care [35–37]. A study based in Italy found that people with adequate nutrition knowledge are more likely to have healthier dietary patterns and a lower prevalence of obesity [38]. On the flipside, a lack of health knowledge has also been shown to lead to more health risk behaviors and poor health status [39, 40]. Studies found that the lack of health knowledge or literacy is associated with chronic diseases, higher rates of hospital admissions, longer hospital stays, and even unnecessary use of health care resources [41, 42].
However, the health knowledge of individuals is important not just for individuals’ own health outcomes, but in the case of parents, the health outcomes and practices of their children [43]. This is because parents play a dominant role in children’s lifestyles, particularly in the case of younger children in their formative years. It has been shown, for instance, that parents’ possession of health-related knowledge has positive effects on children’s health practice [44, 45]. Evidence also shows that children are more likely to have a better health status when their parents understand disease prevention [46] [. Concurrently, children whose caregivers had limited health literacy and less health knowledge tend to have worse health outcomes [47]. Yin and colleagues found that caregivers with inadequate health knowledge knew little about weight-based medication dosing and used non-standardized dosing instruments when administering medications [48]. Another study on children aged six and below showed that children whose caregivers with limited oral health knowledge tended to practice more harmful oral health behaviors, such as no daily cleaning or no brushing [49, 50]. Etelson and colleagues found that parents of children with excess weight are generally unable to recognize that their children have a weight problem [4]. And they argue that the success of any obesity prevention practices targeting young children depends on parents’ capability to recognize the overweight/obesity problem and to provide a healthy diet. Based on what we discussed above, we could argue that parents’ knowledge and capability to recognize obesity (e.g., identifying overweight/obesity) and to provide health interventions (e.g., providing a healthy diet) may significantly impact children’s health behaviors (e.g., participation in physical exercises). We thus hypothesize that:
## H3
Parents’ obesity-related health knowledge is positively associated with their children’s health practice.
Despite this, several studies have also argued that social support can bring about negative health outcomes. Specifically, adequate information from individuals’ social networks may facilitate healthful knowledge and practice, whereas inadequate support or negative information may have an adverse health influence, especially for those with low health literacy [39, 51]. Thus, while there is strong evidence that positive social support has protective effects against all-cause mortality [14, 52], and that adequate resources help individuals to cope with health issues [39, 53], misleading information or advice, on the other hand, may hinder patients from seeking appropriate medical care or even reinforce unhealthy practice [15]. For example, for individuals with risky health practices (e.g., smoking and heavy drinking), social support from people with similar habits may normalize and maintain those unhealthy practices [39, 54]. Therefore, we can speculate that social support promotes health practice only when it can provide adequate health knowledge. It is not how much social support parents have in their child-raising endeavors but rather, what this support contributes to their health knowledge that matters. The relationships between social support, health knowledge, and health practice are further illustrated in Fig. 1.
Building on all the discussions above, we hypothesize that:
## H4
The effect of parents’ social support on children’s health practices is mediated by parents’ obesity-related health knowledge.
## Participants
Data is drawn from a nationally representative survey of parents with young children (age 14 years and younger) conducted in Singapore between June and November 2018. The sampling was based on a representative sample of household addresses provided by the Singapore Department of Statistics (SDOS). As requested by the research team, SDOS randomly drew 2116 household addresses from the total population excluding those without children or children aged 15 and above. Once we received the list of addresses, our research team proceeded to visit the 2116 households and conduct face-to-face surveys. The questionnaire includes measurement scales of social support, health knowledge, health practices, and socio-democratic variables. Each survey took about ten to 15 min to be completed. We prepared questionnaires in Chinese, English, Malay, and Tamil though all participants responded in English. Of the 2116 households being visited, 1488 valid responses were returned and the response rate was $70.6\%$. We conceptualized that parents’ influence is strongest when children are 14 years old or younger as parents remain socially significant in these children’s everyday lives. At this age, children tend to be homebound and are less likely to be influenced as strongly by peers and social media (compared to older teenagers, for example). Thus, the unit of analysis for our study was parents with a child age 14 years or younger. In the sample, $66.1\%$ of the respondents are female, $30.1\%$ are younger than 36 years old, $30.7\%$ are between 36 and 40 years old, $33.0\%$ are between 41 and 49 years old, and $6.3\%$ are 50 years old and above; of the respondents, $40.4\%$ have a bachelor’s degree, $33.3\%$ have a post-secondary diploma, and $26.3\%$ have secondary education and below. Among the respondents, $61.2\%$ are Chinese, $20.2\%$ are Malay, and $18.6\%$ are Indian. This ethnic ratio is generally consistent with the country’s ethnic composition. More information about the sample profile is available in Table 1 which presents the frequencies of gender, age group, work status, housing type (as a proxy for social class), and education. Table 1Frequency table of gender, age group, work status, education, and residence typeGenderPercentAgePercentFemale66.1130 and below8.6Male33.8931 to 3521.47Total10036 to 4030.741 to 4932.9750 and above6.26Total100Work statusPercentEducation levelPercentNot working now22.07Below secondary8.81Working part-time7.81Secondary17.48Working full-time70.12Post-secondary (A levels and poly diploma)33.33Total100Bachelor and postgraduate40.38Total100Residence typePercentHDB 1-room to 3-room21.28HDB 4- to 5-room72.03Private apartment, condo or landed property6.68Total100N = 1488. HDB refers to the homes built by the Housing & Development Board (HDB) of Singapore
## Measures
Social support. To measure social support, Sarason and colleagues used a six-item index that operationalizes social support by counting the number of support sources [10]. Participants are asked to list the people whom they counted on to help them. A higher score indicated greater perceived availability of social support. Procidano and Heller employed a list of dichotomous items to count the number of support sources (e.g., My friends are good at helping me solve problems; 1 = yes, and 0 = No) [55]. Zimet and colleagues proposed a multi-dimensional scale that includes support from family, friends, and significant others (e.g., “My family really tries to help me”, “I can talk about my problems with my friends”, “*There is* a special person who is around when I am in need”, etc.) [ 56]. Building on the measurements developed in the above-mentioned studies, we employed five dichotomous items to measure the number of available support sources (see Table 2). Besides the two items on support from family and friends (e.g., “Do you have family members / close friends whom you trust to discuss childcare matters with?”), we also adapted Zimet and colleagues’ scale items of the support for significant other and created three new items: “Are you able to seek help from a doctor, when you need to?” and “Are you able to seek help from other health care providers like a nurse or dietitian, when you need to?”. Further, we include another item to capture parents’ general ability to look for help: “Do you know where to look for information on child nutrition and well-being?”. All indicators have a dichotomous outcome (1 = Yes and 0 = No). The number of ‘Yes’ answers is accumulated to create an index of social support. Table 2Measurement scale of social support, health knowledge, and health practicesVariablesMinMaxMean/percentage of positive responseSocial support [55, 56, 59]Do you know where to look for information on child nutrition and well-being?$0180.4\%$Do you have family members whom you trust to discuss childcare matters?$0188.8\%$Do you have close friends whom you trust to discuss childcare matters?$0178.9\%$Are you able to seek advice from a doctor, when you need to?$0190\%$Are you able to seek advice from other health care providers like a nurse or dietitian, when you need to?$0179.4\%$Health knowledge [57, 58]BMI knowledge Do you know how obesity is measured?$0166.8\%$ Do you know what is BMI?$0187.3\%$ Will you be able to tell if your child is obese by checking your child’s height and weight?$0158.7\%$Nutrition knowledge (of weight management) I have a good knowledge of what constitutes a healthy nutritious diet for children143.02 I know what my children should consume142.96 Children should eat home-cooked meals instead of food bought from outside143.31 I do not understand the details printed on nutrition labels (reversed)142.89 I read the nutrition labels on the food products142.93 I know how to help my child stay within the acceptable weight range143.05Health practiceHow often do your children exercise?152.64How often do your children eat out at a fast food restaurant? ( reversed)153.05 Obesity-related health knowledge. The present study assesses two aspects of health knowledge: knowledge on what constitutes obesity (e.g., knowledge of BMI) and knowledge on nutrition (e.g., what constitutes a healthy diet). Although many scales have been developed to measure disease-related knowledge, few studies address the measurement of knowledge about bodyweight management (e.g., knowledge of BMI and healthy diet). One study assessed nutrition knowledge by four items (e.g., Knowledge of recommended fruit servings a person should eat each day) [57]. Carter et al. measured patients’ cancer knowledge using a seven-item scale (e.g., “Do you know what breast cancer is?” “ Do you know what a mammogram is?” [ 58]. Building on Carter and colleagues’ work, we assess BMI knowledge with three items (e.g., “Do you know how obesity is measured?”, “ Do you know what is BMI?”, and “Will you be able to tell if your child is obese by checking your child’s height and weight?”). For knowledge of nutrition, we constructed six Likert scale instruments that captured respondents’ understanding of what constitutes a healthy diet and their confidence that they could provide for the nutritional needs of their children (see Table 2).
Obesity-related health practice. We use two items to capture weight management practices: “How often do your children exercise?” and “How often do your children eat out at a fast food restaurant?” Both items were rated by five-point-Likert scale items (from 1 = ‘Rarely or never’ to 5 = ‘Daily’). These items capture the frequency of health practices, which is the main weight control practice for children, and manifest both parents’ and children’s proactive role in weight management.
Control variables. To improve the robustness of the structural equation model, we include several control variables in the analysis. Parents’ social-economic status has been shown to play a significant influence on parenting practice and children’s participation in physical activities and eating [60, 61]. Therefore, we controlled the following social class and background factors in our model: gender, age, education (1 = below secondary; 4 = degree or above), employment status (1 = not working; 2 = working part-time; 3 = working full-time), and residence type (1 = one-room to three-room HDB; 2 = four-room or five-room HDB; 3 = private apartment, condominium or landed house). Recent studies also reported that fathers and mothers may have different perceptions about parenting, and fathers’ involvement in child-rearing is important for children’s health practice [62, 63]. These factors were also controlled for. Father/mother involvement was measured by asking whether mostly father/mother does the five types of household tasks (e.g., planning meals, feeding children, watching over child nutrition, cooking, and ensuring sufficient physical exercise; 1 = Yes and 0 = No).
## Analysis
The primary aim of the present study is to examine the mediating effects of health knowledge on the relationship between social support and health practice. We use structural equation modeling (SEM) via Stata 15.0 to conduct mediation analysis.
Descriptive analysis. Before SEM, we conducted descriptive analysis and correlation analysis to preliminarily describe our variables and their relationships. We also used a series of tests to examine the reliability and validity of our measurements. We first examine the reliability of each measurement using ordinal alpha as all the scale items are ordinal and non-normally distributed [64]. An alpha coefficient of 0.70 or higher is usually considered as a cutoff point for good internal consistency but a value between 0.50 and 0.60 is still acceptable for preliminary studies in social sciences, especially for scales with a limited number of items [65–69]. We then exmined composite reliability (CR) and average variance extracted (AVE). CR values ranges from 0.70 to 0.80. and AVE ranges from 0.40 to 0.60. According to Forrell and Lacker [70], AVE values below 0.50 are acceptable if CR is above 0.7. Regarding the diagnosis of convergent and discriminant validity, we followed the methods by Courvoisier et al. and Wingenfeld et al. [ 71, 72]. It is deemed adequate if within-scale item-to-total correlations are greater than between-scale item-to-total correlations.
Structural equation modeling (SEM). To test the mediating effects of BMI and nutrition knowledge in the relationship between parental social support and children’s health practice, structural equation modeling (SEM) via Stata 15.0 was employed. Comparing to traditional mediation analysis through step-by-step regression, SEM has many advantages, especially when models include latent variables and more than one mediator [73]. The SEM package with Stata can directly estimate the indirect effects (mediating effects) of the main predictor which makes the mediation test much easier. SEM also can produce model fit information about the consistency between the data and the hypothesized model.
We used multiple goodness-of-fit indices to assess the model fit [74]: root means squared error of approximation (RMSEA), standardized root means squared residual (SRMR), comparative fit index (CFI), and chi-square to the degree of freedom ratio (χ2/df). Values smaller than 0.1 for RMSEA indicate acceptable fit, and values between 0.05 and 0.08 indicate a good fit [75]. Values less than 0.08 for SRMR show a good fit [76]. Values of CFI greater than 0.90 indicate adequate fit [77]. Scholars also suggested that a value of SBχ2/df lower than 3 indicates a good fit [78].
We build two SEM models for comparison purposes. Model 1 (see Fig. 2) contains the predictor (social support), dependent variable (health practice), and two mediators (BMI and Nutrition knowledge). In Model 2 (see Fig. 3), we add additional control variables (e.g., age, gender, education, work status, and mother/father involvement. Fig. 2SEM model 1 testing the mediating effects of health knowledge on the relationship between social support and children’s health practiceFig. 3SEM model 1 testing the mediating effects of health knowledge on the relationship between social support and children’s health practice. Parameters of control variables are not presented in this graph due to parsimony consideration
## Descriptive analysis
The results in Table 3 show that respondents have a moderate level of knowledge about BMI and nutrition – out of the 3 questions asked on awareness of what BMI measures, most were able to respond affirmative to 2 out of 3 indicators. On average, children exercise less than once a week, with the mean falling just below the average of once a week (mean = 2.94). Correlations between the dependent variable and key independent variables range from 0.09 to 0.22, thus assuring that there are no issues with collinearity between variables in the model. Nutrition knowledge is positively associated with children’s health practices ($r = 0.22$). Social support is found to have significant positive associations with social support and health knowledge. Gender, age, and education level were included as control variables. Table 3Descriptive statistics and correlationVariableMeanSDMinMaxOrdinal αCRAVE12341. Health practice2.941.13150.640.710.58–2. BMI knowledge2.220.89030.720.680.530.09–3. Nutrition knowledge3.050.371.6740.700.800.500.220.09–4. Social Support4.201.21050.860.790.410.110.170.16–5. Gender––01–− 0.04− 0.06− 0.11− 0.046. Age group––15–0.060.001− 0.03- 0.107. Education––14–0.110.210.070.158. Work status––13–− 0.030.05− 0.130.0019. Residence type––13–0.010.11− 0.010.0311. Mother’s involvement2.441.8105–0.040.0020.10− 0.0612. Father’s involvement0.401.0105–0.01− 0.07− 0.03− 0.07N = 1484. Bold coefficients: $p \leq 0.05.$ AVE Average variance extracted. CR Composite reliability Table 3 also shows that ordinal alpha ranges from 0.64 to 0.86 for the four scales, demonstrating acceptable internal reliability of each measurement. CR values ranges from 0.68 to 0.80. and AVE ranges from 0.41 to 0.58. Results of AVE also support our measurement tools’ convergent validity. The convergent and discriminant validity were supported by correlation analysis in Appendix Table 5 showing that within-scale item-to-total correlations are stronger than between-scale item-to-total correlations.
## Structural equation modeling (SEM)
Results of Model 1 and 2 are shown in Figs. 2 and 3 respectively. Results for Model 1 show a good model fit (see Table 4): χ2/df = $\frac{69.11}{30}$ < 3, RMSEA = 0.039; CFI = 0.958; SRMR = 0.033. All predictors explain $9.02\%$ of the variance in health practice. All the path coefficients through mediators are significant ($p \leq 0.1$). Results of Model 1 (see Fig. 2) also show that $76.85\%$ of the total effects of social support on health practice are mediated by BMI and nutrition knowledge: Total effect = 0.108 ($$p \leq 0.008$$), indirect effect = 0.083 ($$p \leq 0.000$$), and direct effect = 0.025 ($$p \leq 0.575$$). Table 4Descriptive measures of the model fitModel fit indicesCriteriaObtained valueModel 1Model 2χ2/df < 3.002.301.93Root means squared error of approximation (RMSEA) < 0.100.0390.038Comparative fit index (CFI) >.900.9580.904Standardized root means squared residual (SRMR) < 0.080.0330.037 Estimation of Model 2 (Fig. 3) shows a satisfactory model fit (see Table 4): χ2/df = $\frac{154.08}{80}$ < 3, RMSEA = 0.038, CFI = 0.904, and SRMR = 0.037. All predictors explain $20.20\%$ of the variance in health practice. All the path coefficients through mediators are significant ($p \leq 0.1$). Most control variables are not significant except age (β = – 0.081, $$p \leq 0.06$$), father involvement (β = 0.088, $$p \leq 0.07$$), and residence type (β = – 0.074, $$p \leq 0.09$$). The total effect of social support is 0.081 ($$p \leq 0.071$$), the indirect effect through BMI knowledge and nutrition knowledge is 0.081 ($$p \leq 0.007$$), and its direct effect is not significant ($p \leq 0.1$). This demonstrates that the two types of health knowledge have full mediation effects on the relationship between social support and children’s health practice.
Since Model 2 explains more variance in the dependent variables, we use the results from Model 2 for further reference. According to estimation results from Model 2, all the path coefficients are significant ($p \leq 0.001$) except the direct path between social support and health practice. Therefore, Hypothesis H2 and H3 are supported. We can also conclude that the mediating effects of BMI and nutrition knowledge on the relationship between parental social support and children’s health practice are supported (Hypothesis H4 is supported). Although the direct path between social support and health practice is not significant due to the full mediation effects ($100\%$), the total effects of social support on health practice in both Model 1 and 2 are significant ($$p \leq 0.008$$ and $$p \leq 0.071$$). This means that the relationship between social support and health practice is significant and positive when mediators are not included in the model. Thus, Hypothesis H1 is supported (Fig. 3).
## Summary of findings
Using structural equation modeling on a representative sample of Singaporean households with children aged 14 or younger, we found that parents’ social support and health knowledge significantly associate with children’s participation in weight management practices (e.g., physical exercises). More importantly, our results support that parents’ health knowledge serves a mediating role in the relationship between parents’ social support and children’s health practices in weight management. Specifically, there is a significant positive relationship between parents’ social support and health knowledge, thus suggesting that parents draw pro-health information from their social support network. Further, it is noted that the direct effect of social support of parents on children’s health practices is not significant after the mediation effects of health knowledge are considered, which suggests the full mediation effects of health knowledge. These findings contextualize the relationship between social support and health outcomes and advance our theoretical appreciation of the impact of social support as an essential resource. The empirical distillation of the mediation effects advised how pro-health information can be effectively disseminated and will have helpful in framing public health initiatives.
## Theoretical contributions
Three theoretical contributions are noteworthy. First, the present study complements existing knowledge on social determinants (e.g., parents’ social support and health knowledge) of childhood obesity, that there is a direct and positive link between social support and health-related behaviors or outcomes [11, 12]. Our model shows that social support from an individual’s networks does not always necessarily impact pro-health behaviors. As with all peer influence, the normative behaviors of peers vary, as do their credibility as resource persons for health information.
Second, the current study tests a mediation model that bridges social support theories, health knowledge literature, and childhood obesity research. Our findings provide empirical evidence for how children’s health practice is influenced by parents’ social support and health knowledge. The mediating role of health knowledge in the relationship between social support and health practice was supported, which responded to the doubt about why inconsistent findings on the relationship between social support and health practice exist [11–15, 39, 51]. The SEM model demonstrates that while both parents’ BMI and nutrition knowledge fully mediates the relationship between parents’ social support and children’s health practices, compared to BMI knowledge, parents’ nutrition knowledge plays a stronger role.
Finally, this is one of the few studies on the effects of social support on health behavior conducted on an affluent and multi-cultural Southeast Asian population. Although obesity is not traditionally considered a big problem in Asian countries, the growing prevalence of obesity rates attracts increasing attention from researchers and policymakers. Our findings thus contribute to existing knowledge by grounding it within an Asian context.
## Policy values
The findings clarify how pro-health information can be more effectively disseminated to the general public. Health promotion and obesity prevention programs should target participants’ social support networks. Public health messages that are too broad-based and targeted at a general audience dissipate without impacting their target audience. In addition to focusing on parents with young children, our research suggests that another important avenue for disseminating pro-health messages through social support networks, perhaps with simple tag lines like “share this information with a parent of young children”. Against the backdrop of the persistent COVID-19 pandemic, public health educators or governments can be better informed by this study how to guarantee a successful vaccination campaign.
Concurrently, an effort to evolve a network of public health champions in the community may be an effective way of disseminating pro-health information and advisories. These champions can be positioned as support resources to partner parents in their childrearing endeavors. In parenting talks and community education events, invited parents can bring a friend so that information disseminated can reach a larger audience. Such interventions will encourage the provision of social support from sources with higher levels of health knowledge.
One highlight from our study alerted us to the lack of understanding on how the BMI is derived and what it can be used for, and how to make sense of food nutrition labels to support their children’s well-being. This is a reminder that while we have made many advances in pushing out tests and procedures to push out nutrition and health information, for these to impact health practice, we have to invest in educating the lay public on how to render relevance to such information in their everyday practice.
## Limitations and future research directions
Limitations appear in the present study. Our study is based on an analysis of cross-sectional data, which may limit the validity of our results and interpretation. Researchers elsewhere suggest the use of longitudinal rather than cross-sectional data to establish the inference of causality and mediation models [79]. Due to limited resources, the present study was also only able to collect data from parents to test our hypotheses. Information from the child’s position is absent. Future studies may consider a longitudinal research design and collect data based on a parent–child dyad approach. Further, the findings presented in this paper are a small section taken from a more extensive study on sociocultural environmental effects on childhood obesity, and have only limited instruments to measure for health practice and health knowledge. Although we believe that our results based on the current measurement scales are still trustworthy, there is a need to improve the validity and reliability of the scales. Future studies should include more detailed instruments to capture these constructs holistically. For example, the health knowledge should be expanded to include awareness of risk factors of childhood obesity on adult chronic diseases morbidity. Health practice scale should include more information about children’ physical activities and diet management. While this paper demonstrated the effect of access to social support on health knowledge, future research should elaborate on the more complex effects of social support on other aspects of pro-health behaviors.
## Conclusion
The present study aimed to investigate the joint influence of parents’ social support and health knowledge on children’s health practice. Results from our analysis on a nationally representative sample from Singapore support the view that parents’ obesity-related health knowledge has a mediating effect on the relationship between parents’ social support and children’s obesity-related health practice. This indicates that the influence that social support has on health practice is heterogeneous – while parents’ social support has a positive effect on children’s body weight management practices when social support could enhance obesity-related health knowledge, this is not the case when there is a lack of health knowledge embedded in parents’ social support. This study highlighted the family environmental factors of children’s health from the perspective of social support theories, health knowledge literature, and childhood obesity research. Future studies should adopt a longitudinal research design and include more comprehensive instruments to measure the constructs of social support, obesity-related health knowledge, and obesity-related health practices.
## Appendix
See Table Table 5Correlations between scale itemsQ32aQ32bQ32cQ32dQ32eQ1hQ1iQ1jQ34dQ34eQ34fBMIK1BMIK2BMIK3Q22hQ22iSocial supportQ32a1.00Q32b0.441.00Q32c0.280.251.00Q32d0.210.210.501.00Q32e0.230.250.300.401.00Nutrition knowledgeQ1h0.050.050.040.080.091.00Q1i0.050.040.040.110.100.341.00Q1j0.030.050.080.090.140.230.251.00Q34d0.010.03– 0.040.000.030.160.140.141.00Q34e0.040.030.020.040.060.160.190.170.401.00Q34f0.100.090.090.130.180.160.250.190.330.451.00BMI knowledeBMIK10.110.100.180.120.210.070.080.140.050.000.121.00BMIK20.140.140.200.140.230.070.050.120.060.010.060.471.00BMIK30.020.05– 0.010.030.030.020.050.050.02– 0.03– 0.030.220.121.00Health practiceQ22h0.110.090.050.050.120.150.160.100.100.090.140.110.120.051.00Q22i0.060.040.020.050.110.090.080.070.130.070.150.080.09– 0.010.471.005
## References
1. Staniford LJ, Breckon JD, Copeland RJ. **Treatment of childhood obesity: a systematic review**. *J Child Fam Stud* (2012.0) **21** 545-564. DOI: 10.1007/s10826-011-9507-7
2. Ng M, Fleming T, Robinson M, Thomson B, Graetz N, Margono C. **Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the global burden of disease study 2013**. *The lancet* (2014.0) **384** 766-781. DOI: 10.1016/S0140-6736(14)60460-8
3. Andrews KR, Silk KS, Eneli IU. **Parents as health promoters: a theory of planned behavior perspective on the prevention of childhood obesity**. *J Health Commun* (2010.0) **15** 95-107. DOI: 10.1080/10810730903460567
4. Etelson D, Brand DA, Patrick PA, Shirali A. **Childhood obesity: do parents recognize this health risk?**. *Obes Res* (2003.0) **11** 1362-1368. DOI: 10.1038/oby.2003.184
5. Golan M, Kaufman V, Shahar DR. **Childhood obesity treatment: targeting parents exclusively v. Parents and children**. *British J Nutrition* (2006.0) **95** 1008-1015. DOI: 10.1079/BJN20061757
6. Kiernan M, Moore SD, Schoffman DE, Lee K, King AC, Taylor CB. **Social support for healthy behaviors: scale psychometrics and prediction of weight loss among women in a behavioral program**. *Obesity* (2012.0) **20** 756-764. DOI: 10.1038/oby.2011.293
7. **The look AHEAD study: a description of the lifestyle intervention and the evidence supporting it**. *Obesity* (2006.0) **14** 737-752. DOI: 10.1038/oby.2006.84
8. Eustis NN, Fischer LR. **Relationships between home care clients and their workers: implications for quality of care**. *Gerontologist* (1991.0) **31** 447-456. DOI: 10.1093/geront/31.4.447
9. Miller B, McFall S, Campbell RT. **Changes in sources of community long-term care among African American and white frail older persons**. *J Gerontol* (1994.0) **49** S14-S24. DOI: 10.1093/geronj/49.1.S14
10. Sarason IG, Sarason BR, Shearin EN, Pierce GR. **A brief measure of social support: practical and theoretical implications**. *J Soc Pers Relat* (1987.0) **4** 497-510. DOI: 10.1177/0265407587044007
11. House JS, Landis KR, Umberson D. **Social relationships and health**. *Science* (1988.0) **241** 540-545. DOI: 10.1126/science.3399889
12. Smith SG, Jackson SE, Kobayashi LC, Steptoe A. **Social isolation, health literacy, and mortality risk: findings from the english longitudinal study of ageing**. *Health Psychol* (2018.0) **37** 160. DOI: 10.1037/hea0000541
13. 13.Wadden TA, Phelan S. Behavioral assessment of the obese patient. Handbook of obesity treatment. 2002;186–226.
14. Eaker ED. *Social support and physical health: understanding the health consequences of relationships* (2005.0)
15. Jorm AF. **Mental health literacy: public knowledge and beliefs about mental disorders**. *Br J Psychiatry* (2000.0) **177** 396-401. DOI: 10.1192/bjp.177.5.396
16. Lepore SJ, Evans GW, Schneider ML. **Dynamic role of social support in the link between chronic stress and psychological distress**. *J Pers Soc Psychol* (1991.0) **61** 899. DOI: 10.1037/0022-3514.61.6.899
17. Armstrong MI, Birnie-Lefcovitch S, Ungar MT. **Pathways between social support, family well being, quality of parenting, and child resilience: What we know**. *J Child Fam Stud* (2005.0) **14** 269-281. DOI: 10.1007/s10826-005-5054-4
18. Sherbourne CD, Stewart AL. **The MOS social support survey**. *Soc Sci Med* (1991.0) **32** 705-714. DOI: 10.1016/0277-9536(91)90150-B
19. 19.Verbrugge LM, Ascione FJ. Exploring the iceberg: common symptoms and how people care for them. Medical Care 1987:539–69.
20. Lydakis C, Kerr H, Hutchings K, Lip G. **Women's awareness of, and attitudes towards, hormone replacement therapy: ethnic differences and effects of age and education**. *Int J Clin Pract* (1998.0) **52** 7. PMID: 9536560
21. Allgöwer A, Wardle J, Steptoe A. **Depressive symptoms, social support, and personal health behaviors in young men and women**. *Health Psychol* (2001.0) **20** 223. DOI: 10.1037/0278-6133.20.3.223
22. Snowdon AW, Cameron S, Dunham K. **Relationships between stress, coping resources, and satisfaction with family functioning in families of children with disabilities**. *Canadian J Nurs Res Archiv* (1994.0) **26** 63
23. Kana'Iaupuni SM, Donato KM, Thompson-Colon T, Stainback M. **Counting on kin: social networks, social support, and child health status**. *Soc Forces* (2005.0) **83** 1137-1164. DOI: 10.1353/sof.2005.0036
24. Kim Y-H, Cardinal BJ. **Psychosocial correlates of Korean adolescents' physical activity behavior**. *J Exerc Sci Fit* (2010.0) **8** 97-104. DOI: 10.1016/S1728-869X(10)60015-9
25. Sylvia-Bobiak S, Caldwell LL. **Factors related to physically active leisure among college students**. *Leis Sci* (2006.0) **28** 73-89. DOI: 10.1080/01490400500332728
26. Theodoropoulou E, Karteroliotis K, Stavrou N. **Validity and reliability of Greek versions of two scales assessing family and friend support for exercise behaviour**. *Percept Mot Skills* (2014.0) **118** 26-40. DOI: 10.2466/06.08.PMS.118k13w7
27. Sørensen K, Van den Broucke S, Fullam J, Doyle G, Pelikan J, Slonska Z. **Health literacy and public health: a systematic review and integration of definitions and models**. *BMC Public Health* (2012.0) **12** 80. DOI: 10.1186/1471-2458-12-80
28. Baker DW. **The meaning and the measure of health literacy**. *J Gen Intern Med* (2006.0) **21** 878-883. DOI: 10.1111/j.1525-1497.2006.00540.x
29. Nutbeam D. **Health literacy as a public health goal: a challenge for contemporary health education and communication strategies into the 21st century**. *Health Promot Int* (2000.0) **15** 259-267. DOI: 10.1093/heapro/15.3.259
30. Coleman JS. **Social capital in the creation of human capital**. *Am J Sociol* (1988.0) **94** S95-S120. DOI: 10.1086/228943
31. 31.House JS. Work stress and social support. Addison-wesley series on occupational stress. 1983.
32. Albrecht TL, Adelman MB. *Communicating social support* (1987.0)
33. Johnson JD, Meishcke H. **Differences in evaluations of communication channels for cancer-related information**. *J Behav Med* (1992.0) **15** 429-445. DOI: 10.1007/BF00844940
34. Noar SM, Zimmerman RS. **Health behavior theory and cumulative knowledge regarding health behaviors: are we moving in the right direction?**. *Health Educ Res* (2005.0) **20** 275-290. DOI: 10.1093/her/cyg113
35. Lindau ST, Tomori C, Lyons T, Langseth L, Bennett CL, Garcia P. **The association of health literacy with cervical cancer prevention knowledge and health behaviors in a multiethnic cohort of women**. *Am J Obstet Gynecol* (2002.0) **186** 938-943. DOI: 10.1067/mob.2002.122091
36. Malla C, Aylward P, Ward P. **Knowledge translation for public health in low-and middle-income countries: a critical interpretive synthesis**. *Global Health Res Policy* (2018.0) **3** 1-12. DOI: 10.1186/s41256-018-0084-9
37. Scott TL, Gazmararian JA, Williams MV, Baker DW. **Health literacy and preventive health care use among medicare enrollees in a managed care organization**. *Med Care* (2002.0) **40** 395-404. DOI: 10.1097/00005650-200205000-00005
38. Bonaccio M, Di Castelnuovo A, Costanzo S, De Lucia F, Olivieri M, Donati MB. **Nutrition knowledge is associated with higher adherence to Mediterranean diet and lower prevalence of obesity**. *Results Moli-Sani Stud Appetite* (2013.0) **68** 139-146
39. Lee SY, Arozullah AM, Cho YI. **Health literacy, social support, and health: a research agenda**. *Soc Sci Med* (2004.0) **58** 1309-1321. DOI: 10.1016/S0277-9536(03)00329-0
40. Sudore RL, Mehta KM, Simonsick EM, Harris TB, Newman AB, Satterfield S. **Limited literacy in older people and disparities in health and healthcare access**. *J Am Geriatr Soc* (2006.0) **54** 770-776. DOI: 10.1111/j.1532-5415.2006.00691.x
41. Baker DW, Gazmararian JA, Williams MV, Scott T, Parker RM, Green D. **Functional health literacy and the risk of hospital admission among medicare managed care enrollees**. *Am J Public Health* (2002.0) **92** 1278-1283. DOI: 10.2105/AJPH.92.8.1278
42. Zarcadoolas C. **The simplicity complex: exploring simplified health messages in a complex world**. *Health Promot Int* (2011.0) **26** 338-350. DOI: 10.1093/heapro/daq075
43. Sanders LM, Thompson VT, Wilkinson JD. **Caregiver health literacy and the use of child health services**. *Pediatrics* (2007.0) **119** e86-e92. DOI: 10.1542/peds.2005-1738
44. Adair PM, Pine CM, Burnside G, Nicoll AD, Gillett A, Anwar S. **Familial and cultural perceptions and beliefs of oral hygiene and dietary practices among ethnically and socio-economicall diverse groups**. *Community Dent Health* (2004.0) **21** 102-111. PMID: 15072479
45. Saied-Moallemi Z, Virtanen J, Ghofranipour F, Murtomaa H. **Influence of mothers’ oral health knowledge and attitudes on their children’s dental health**. *Eur Arch Paediatr Dent* (2008.0) **9** 79-83. DOI: 10.1007/BF03262614
46. Sanders LM, Shaw JS, Guez G, Baur C, Rudd R. **Health literacy and child health promotion: implications for research, clinical care, and public policy**. *Pediatrics* (2009.0) **124** S306-S314. DOI: 10.1542/peds.2009-1162G
47. DeWalt DA, Hink A. **Health literacy and child health outcomes: a systematic review of the literature**. *Pediatrics* (2009.0) **124** S265-S274. DOI: 10.1542/peds.2009-1162B
48. Yin HS, Dreyer BP, Foltin G, van Schaick L, Mendelsohn AL. **Association of low caregiver health literacy with reported use of nonstandardized dosing instruments and lack of knowledge of weight-based dosing**. *Ambul Pediatr* (2007.0) **7** 292-298. DOI: 10.1016/j.ambp.2007.04.004
49. Miller E, Lee JY, DeWalt DA, Vann WF. **Impact of caregiver literacy on children's oral health outcomes**. *Pediatrics* (2010.0) **126** 107-114. DOI: 10.1542/peds.2009-2887
50. Vann W, Lee JY, Baker D, Divaris K. **Oral health literacy among female caregivers: impact on oral health outcomes in early childhood**. *J Dent Res* (2010.0) **89** 1395-1400. DOI: 10.1177/0022034510379601
51. 51.Lee S-YD, Arozullah AM, Cho YI, Crittenden K, Vicencio D. Health literacy, social support, and health status among older adults. Edu Gerontol 2009;35(3):191–201.
52. Uchino BN. *Social support and physical health: understanding the health consequences of relationships* (2004.0)
53. Lyyra T-M, Heikkinen R-L. **Perceived social support and mortality in older people**. *J Gerontol B Psychol Sci Soc Sci* (2006.0) **61** S147-S152. DOI: 10.1093/geronb/61.3.S147
54. 54.Engels RC, Knibbe RA, Drop MJ, Haan YTd. Homogeneity of cigarette smoking within peer groups: Influence or selection?. Health Educ. Behav. 1997;24(6):801–11.
55. Procidano ME, Heller K. **Measures of perceived social support from friends and from family: three validation studies**. *Am J Community Psychol* (1983.0) **11** 1-24. DOI: 10.1007/BF00898416
56. Zimet GD, Dahlem NW, Zimet SG, Farley GK. **The multidimensional scale of perceived social support**. *J Pers Assess* (1988.0) **52** 30-41. DOI: 10.1207/s15327752jpa5201_2
57. Variyam JN. **Overweight children: is parental nutrition knowledge a factor?**. *Food Rev Nat Food Rev* (2001.0) **24** 18-22
58. Carter J, Park ER, Moadel A, Cleary SD, Morgan C. **Cancer knowledge, attitudes, beliefs, and practices (KABP) of disadvantaged women in the South Bronx**. *J Cancer Educ* (2002.0) **17** 142-149. PMID: 12243219
59. Sarason BR, Shearin EN, Pierce GR, Sarason IG. **Interrelations of social support measures: theoretical and practical implications**. *J Pers Soc Psychol* (1987.0) **52** 813. DOI: 10.1037/0022-3514.52.4.813
60. Gibson LY, Allen KL, Byrne SM, Clark K, Blair E, Davis E. **Childhood overweight and obesity: maternal and family factors**. *J Child Fam Stud* (2016.0) **25** 3236-3246. DOI: 10.1007/s10826-016-0485-7
61. 61.Noh H-M, Park J, Sung E-J, Ju YS, Lee H-J, Jeong Y-K, et al. Family factors and obesity in relation to mental health among korean children and adolescents. J Child Family Stud 2019:1–9.
62. Alehagen S, Hägg M, Kalén-Enterlöv M, Johansson A. **Experiences of community health nurses regarding father participation in child health care**. *J Child Health Care* (2011.0) **15** 153-162. DOI: 10.1177/1367493511403952
63. Algarvio S, Isabel L, Maroco J. **Parental concerns’ prevalence and socio-demographic variables in general parenting**. *J Child Health Care* (2013.0) **17** 204-214. DOI: 10.1177/1367493512456107
64. McNeish D. **Thanks coefficient alpha, we’ll take it from here**. *Psychol Methods* (2018.0) **23** 412. DOI: 10.1037/met0000144
65. 65.Chen C-A, Xu C. ‘No, I cannot just walk away’: government career entrenchment in China. Int Rev Administrative Sci 2020:0020852319884624.
66. 66.Gliem JA, Gliem RR, editors. Calculating, interpreting, and reporting Cronbach’s alpha reliability coefficient for Likert-type scales2003: Midwest research-to-practice conference in adult, continuing, and community ….
67. Peterson RA. **A meta-analysis of cronbach's coefficient alpha**. *J Consumer Res* (1994.0) **21** 381-391. DOI: 10.1086/209405
68. Streiner DL, Norman GR, Cairney J. *Health measurement scales: a practical guide to their development and use* (2015.0)
69. Xu C. **Work motivation in the public service: a scale development based on the self-determination theory**. *SAGE Open* (2022.0) **12** 1-13. DOI: 10.1177/21582440221091263
70. Fornell C, Larcker DF. **Evaluating structural equation models with unobservable variables and measurement error**. *J Mark Res* (1981.0) **18** 39-50. DOI: 10.1177/002224378101800104
71. Courvoisier DS, Nussbeck FW, Eid M, Geiser C, Cole DA. **Analyzing the convergent and discriminant validity of states and traits: development and applications of multimethod latent state-trait models**. *Psychol Assess* (2008.0) **20** 270. DOI: 10.1037/a0012812
72. Wingenfeld SA, Lachar D, Gruber CP, Kline RB. **Development of the teacher-informant student behavior survey**. *J Psychoeduc Assess* (1998.0) **16** 226-249. DOI: 10.1177/073428299801600304
73. Gunzler D, Chen T, Wu P, Zhang H. **Introduction to mediation analysis with structural equation modeling**. *Shanghai Arch Psychiatry* (2013.0) **25** 390. PMID: 24991183
74. Greiff S, Heene M. **Why psychological assessment needs to start worrying about model fit**. *Eur J Psychol Assess* (2017.0) **33** 313-317. DOI: 10.1027/1015-5759/a000450
75. 75.Kenny DA. Measuring model fit 2015 [Available from: f://davidakenny.net/cm/fit.htm.
76. 76.Hu Lt, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Model: Multidisciplinary J 1999;6(1):1–55.
77. Gagné M, Forest J, Gilbert M-H, Aubé C, Morin E, Malorni A. **The motivation at work scale: validation evidence in two languages**. *Educ Psychol Measur* (2010.0) **70** 628-646. DOI: 10.1177/0013164409355698
78. Kline RB. *Principles and practice of structural equation modeling* (2015.0)
79. O'Laughlin KD, Martin MJ, Ferrer E. **Cross-sectional analysis of longitudinal mediation processes**. *Multivar Behav Res* (2018.0) **53** 375-402. DOI: 10.1080/00273171.2018.1454822
|
---
title: 'GoM DE: interpreting structure in sequence count data with differential expression
analysis allowing for grades of membership'
authors:
- Peter Carbonetto
- Kaixuan Luo
- Abhishek Sarkar
- Anthony Hung
- Karl Tayeb
- Sebastian Pott
- Matthew Stephens
journal: bioRxiv
year: 2023
pmcid: PMC10028846
doi: 10.1101/2023.03.03.531029
license: CC BY 4.0
---
# GoM DE: interpreting structure in sequence count data with differential expression analysis allowing for grades of membership
## Abstract
Parts-based representations, such as non-negative matrix factorization and topic modeling, have been used to identify structure from single-cell sequencing data sets, in particular structure that is not as well captured by clustering or other dimensionality reduction methods. However, interpreting the individual parts remains a challenge. To address this challenge, we extend methods for differential expression analysis by allowing cells to have partial membership to multiple groups. We call this grade of membership differential expression (GoM DE). We illustrate the benefits of GoM DE for annotating topics identified in several single-cell RNA-seq and ATAC-seq data sets.
## Background
A key methodological aim in single-cell genomics is to learn structure from single-cell sequencing data in a systematic, data-driven way [1–3]. Clustering [4–7] and dimensionality reduction techniques such as such as PCA [8–10], t-SNE [11] or UMAP [12] are commonly used for this aim. Despite the fact that many of these techniques have been applied “out-of-the-box” (with some caveats [13–18]), they have been remarkably successful in revealing and visualizing biologically interesting substructures from single-cell data [7,19–29].
Another class of dimensionality reduction approaches that have been used to identify structure from single-cell data are what are sometimes called parts-based representations—these approaches include non-negative matrix factorization (NMF) [30–44] and topic modeling [45–56], which also have formal connections [48, 57, 58]. Parts-based representations share some of the features of both a clustering and a dimensionality reduction: on the one hand, they learn a lower dimensional representation of the cells; on the other hand, the individual dimensions (the “parts”) of the reduced representation can identify discrete clusters or discrete subpopulations [59, 60]. However, parts-based representations are more flexible than clustering—the dimensions can also capture other features such as continuously varying cell states.
In this paper, we investigate the question of how to interpret the individual dimensions of a parts-based representation learned by fitting a topic model. ( In the topic model, the dimensions are also called “topics”.) For topics that assign observations to discrete clusters, one could apply a standard method for differential expression analysis [61, 62] to compare expression between topics, then annotate these topics by the genes that are differentially expressed. The question, therefore, is what to do with topics that do not assign observations to discrete clusters. To tackle this question, we extend models that compare expression between groups by allowing observations to have partial membership in multiple groups. This more flexible differential expression analysis is implemented by taking an existing model and modifying it to allow for partial memberships to groups or topics. This modified model is a “grade of membership” model [63], so we call our new method grade of membership differential expression (GoM DE). The idea is that, by generalizing existing methods, we can continue to take advantage of existing elements of differential expression analysis, but now apply them to learn about different types of cell features beyond discrete cell populations.
We describe the GoM DE approach more formally in the next section. Then we evaluate the GoM DE approach in simulations, showing, in particular, that it recovers the same results as existing differential expression analysis methods when the cells can be grouped into discrete clusters. In case studies, we demonstrate how the GoM DE analysis analysis can be used to uncover and interpret a variety of cell features from single-cell RNA-seq and ATAC-seq data sets.
## Methods overview and illustration
We begin by giving a brief overview of the topic model, then we describe the new methods for annotating topics. To illustrate key concepts, we analyze a single-cell RNA-seq (scRNA-seq) data set obtained from peripheral blood mononuclear cells (PBMCs) [29] that has been used in several benchmarking studies [e.g., 4, 7, 8, 64, 65]. We refer to these data as the “PBMC data.”
## Learning expression topics from single-cell RNA-seq data
The original aim of the topic model was to discover patterns from collections of text documents, in which text documents were represented as word counts [45, 50, 66–68]. By substituting genes for words and cells for documents, topic models can also be used to learn a reduced representation of cells by their membership in multiple “topics” [47].
When applied to scRNA-seq data generated using UMIs, the topic model assumes a multinomial distribution of the RNA molecule counts in a cell, [1] xi1,…,xim~Multinomialsi;πi1,…,πim. where si=xi1+⋯+xim, and m is the number of genes. That is, the number of RNA molecules xij observed for gene j in cell i is a noisy observation of an underlying true expression level, πij [8, 69].
For n cells, the topic model is a reduced representation of the underlying expression, [2] Π=LFT, where Π,L,F are n×m,n×K,m×K matrices, respectively, with entries πij,lik,fjk. Each cell i is represented by its “grade of membership” in K topics, a vector of proportions li1,…,liK, such that lik≥0, ∑$k = 1$K lik=1, and each “expression topic” is represented by a vector of (relative) expression levels f1k,…,fmk,fjk≥0. ( These are also constrained to sum to 1, which ensures that the πij’s are multinomial probabilities.) To efficiently fit the topic model to large single-cell data sets, we exploit the fact that the topic model is closely related to the Poisson NMF model [48].
The matrix L in [2], which contains the membership proportions for all cells and topics, can be visualized using a “Structure plot”. Structure plots have been used to visualize the results of population genetics analyses [e.g., 70–72], and, more recently, to visualize the topics learned from bulk and single-cell RNA-seq data [47].
A Structure plot visualizing the topic model fit to the PBMC data, with $K = 6$ topics, is given in Figure 1. In this data set, the cells have been “sorted” into different cell types which provides a cell labeling to compare against. From the Structure plot, it is apparent that a subset of topics—topics 1,2 and 3—correspond closely to the sorted subpopulations (B cells, CD14+ monocytes, CD34+ cells). ( Indeed, distinctive genes and enriched gene sets identified by the methods described below suggest these same subpopulations; Figure 1C.) Topics 4 and 5, on the other hand, are not confined to a single sorted cell type, and instead appear to capture biological processes common to T cells and natural killer (NK) cells. CD8+ cytotoxic T cells have characteristics of both NK cells and T cells—these are T cells that sometimes become “NK-like” [73]—and this is captured in the topic model by assigning membership to both topics. Topic 6 also captures continuous structure, but, unlike topics 4 and 5, it is present in almost all cells, and therefore its biological interpretation is not at all clear from the cell labeling. *More* generally, the topics, whether they capture largely discrete structure (topics 1–3) or more continuous structure (topics 4–6), can be thought of as a “soft” clustering [47].
## Learning chromatin accessibility topics from single-cell ATAC-seq data
For single-cell ATAC-seq data, the observations xij denote the number of reads mapping to region j in cell i. However, it is common to “binarize” the read counts such that xij=1 when at least one fragment in cell i maps to region j and xij=0 otherwise.
Using the topic model to analyze (binarized) single-cell ATAC-seq data was first suggested by [49]. Therefore, they implicitly assumed a multinomial model [1] in which the xij’s are binarized accessibility values instead of UMI counts. A binomial model for binarized accessibility data was proposed in [74]. As we explain in Methods, we view both models as approximations, and under reasonable assumptions the models are similar.
## Differential expression analysis allowing for grades of membership
Having learned the topics, our aim now is to identify genes that are distinctive to each topic. In the simplest case, the topic is a distinct or nearly distinct cluster of cells, such as topic 1 or 2 in Figure 1.
In the following, we describe methods for analyzing differences in expression, but they can also be understood as methods for analyzing differences in chromatin accessibility. Therefore, “expression,” “expressed” and “gene” in the descriptions below may be substituted with “accessibility”, “accessible” and “peak” (or “region”).
Consider a single gene, j. Provided unmodeled sources of variation are negligible relative to measurement error, a simple Poisson model of expression should suffice: [3] xij~Poissonsiθij. In this model, θij for gene j in cell i is controlled by the cell’s membership in the cluster: when cell i belongs to the cluster, θij=pj1; otherwise, θij=pj2. Under this model, differential expression (DE) analysis proceeds by estimating the log-fold change (LFC) in expression for each gene j, [4] LFC(j)=log2pj1pj2. Although simple, this Poisson model forms the basis for many DE analysis methods [75–80].
We now modify the Poisson model [3] in a simple way to analyze differential expression among topics. In a clustering, each cell belongs to a single cluster, whereas in the topic model, cells have grades of membership to the clusters [63] in which lik is the membership proportion for cluster or topic k. Therefore, we extend the model to allow for partial membership in the K topics: [5] xij~Poissonsiθijθij=∑$k = 1$Klikpjk, in which the membership proportions lik are treated as known, and the unknowns pj1,…,pjK represent relative expression levels. ( A related model is used in C-SIDE [80] to model cell-type mixtures in DE analysis of spatial transcriptomics data.) Note that pjk will be similar to, but not the same as, fjk in the topic model because the DE analysis is a gene-by-gene analysis, whereas the topic model considers all genes at once. The standard Poisson model [3] is recovered as a special case of [5] when $K = 2$ and all membership proportions lik are 0 or 1.
Recall, our aim is to identify genes that are distinctive to each topic. To this end, we estimate the least extreme LFC (l.e. LFC), which we define as [6] LFCkl.e.(j)≔LFCk,l(j)suchthatl=argminl′≠kLFCk,l′(j), in which LFCk,l(j) is the pairwise LFC, [7] LFCk,l(j)≔log2pjkpjl In words, the l.e. LFC for topic k is the LFC comparing topics k and l, in which l is chosen to be topic that results in the smallest (“least extreme”) change. By this definition, a “distinctive gene” is one in which its expression is significantly different from its expression in all other topics. ( Note the l.e. LFC reduces to the standard LFC [4] when $K = 2.$) We then annotate topics by the distinctive genes. The estimation of l.e. LFCs and computation of related posterior statistics is described in Methods.
To illustrate what the least extreme LFC does and does not do, consider the following toy example with $K = 10$ topics (Figure 2). Gene 1 has high expression in topic 1 and low expression in the other topics. Therefore, all the pairwise LFCs for topic 1 are large, LFC1,k[1]=log2[100],$k = 2$,…,10, and this results in an l.e. LFC for topic 1 of log2[100]≈6.6. *So* gene 1 is a distinctive gene for topic 1. Next consider gene 2, which has high expression in topics 1 and 2 and low expression in the other topics. *For* gene 2, the pairwise LFCs for topic 1 are mostly large, LFC1,k[2]=log2[100],$k = 3$,…,10, except for LFC1,2[2]=0. So the l.e. LFC for topic 1 is zero and, as a result, gene 2, although potentially helpful for interpreting topic 1, is not a distinctive gene for topic 1.
## Illustration of GoM DE analysis in PBMC data set
To illustrate, we applied the GoM DE analysis to the topic model shown in Figure 1, and visualized the results in “volcano plots” (Figure 3). We then used the GoM DE results (Additional file 2: Table S1) to perform gene set enrichment analysis (Additional file 3: Tables S3, S4).
For the topics that closely correspond to cell types, the GoM DE analysis, as expected, identified genes and gene sets reflecting these cell types. For example, topic 1 corresponds to FACS B cells, and is characterized by overexpression of CD79A (posterior mean l.e. LFC = 13.05) and enrichment of B cell receptor signaling genes (enrichment coefficient = 0.72). Topic 2 corresponds to myeloid cells and is characterized by overexpression of S100A9 (l.e. LFC = 15.45) and enrichment of genes down-regulated in hematopoietic stem cells (enrichment coefficient = 0.90).
The close correspondence between topics 1 and 2 and FACS cell types (B cells, myeloid cells) provides an opportunity to contrast the GoM DE analysis with a standard DE analysis of the FACS cell types (Figure 4). This is not a perfect comparison because the topics and FACS cell populations are not exactly the same, but the LFC estimates correlate well (Figure 4A, B). This comparison illustrates to two key differences: Other topics capture more continuous structure, such as topics 4 and 5 (Figure 1). Although the GoM DE analysis of these topics is not comparable to a standard DE analysis, many of the the distinctive genes and gene sets suggest NK and T cells, which are precisely the FACS-labeled cells with greatest membership to these topics: for example, for topic 4, overexpression of NKG7 (posterior mean l.e. LFC = 14.09), enrichment of cytolysis genes (enrichment coefficient = 2.22); for topic 5, overexpression of CD3D (l.e. LFC = 12.01), enrichment co-stimulatory signaling during T-cell activation (enrichment coefficient = 1.58).
Topic 6 captures continuous structure and is present in almost all cells, so knowledge of the FACS cell types is not helpful for understanding this topic. Still, the GoM DE results for topic 6 show a striking enrichment of ribosome-associated genes (Figure 3, Additional file 3: Tables S3, S4). ( These ribosomal protein genes also account for a large fraction of the total expression in the cells [5].) This ability to annotate distinctly non-discrete structure is a distinguishing feature of the grade-of-membership approach, and below we will show more examples where this feature contributes to understanding of the cell populations.
## Evaluation of DE analysis methods using simulated data
Having illustrated the features of this approach, we now evaluate the methods more systematically in simulated expression data sets. We began our evaluation by first considering the case of two groups in which there is no partial membership to these groups; that is, when the cells can be separated into two cell types. The GoM DE analysis should accommodate this special case, and should compare well with existing DE analysis methods. We compared with DESeq2 [79] and MAST [84], both popular methods that have been shown to be competitive in benchmarking studies [61, 62, 85] (and are included in Seurat [25]).
To compare the ability of these methods to discover differentially expressed genes, we simulated RNA molecule count data for 10,000 genes and 200 cells in which $98\%$ of cells were attributed to a single topic, with roughly the same number of cells assigned to each of the two topics (with membership proportions of $99\%$ or greater). Note that although half the simulated genes had different expression levels in the two topics, most of these expression differences were small, and therefore the methods were not expected to identify most expression differences. This mimics the typical situation in gene expression studies whereby most expression differences are small. Molecule counts were simulated using a Poisson measurement model so that variation in expression across cells was due to either measurement error or true differences in expression levels between the two groups. For all DE analyses, we took group/topic assignments to be known so that incorrect assignment of cells to topics was not a source of error. Other aspects of the simulations were chosen to emulate molecule count data from scRNA-seq studies (see Methods). We repeated the simulations 20 times, and summarized the results of the DE analyses in Figure 5 (also Additional file 1: Figures S2, S3).
DESeq2 and the GoM DE analysis have several features in common: both are based on a Poisson model, and both use adaptive shrinkage [81, 83] to improve accuracy of the LFC estimates and test statistics. Therefore, we expected the GoM DE results to closely resemble DESeq2 in these simulations. Indeed, both methods produced nearly identical posterior mean LFC estimates, posterior z-scores (Figure 5A, B) and s-values (Additional file 1: Figure S3), and achieved very similar performance (Figure 5C). Although DESeq2 additionally estimates an overdispersion level for each gene, in these simulations DESeq2 correctly determined that the level of overdispersion was small for genes with large expression differences, which explains the strong similarity of the LFC estimates and posterior z-scores. MAST, owing to an approach that is very different from DESeq2 and the GoM DE analysis, yielded estimates that were less similar (Additional file 1, Figure S3), yet achieved comparable performance (Figure 5C).
Next we evaluated the GoM DE analysis methods in data sets in which the cells had varying degrees of membership to multiple topics. Since existing DE methods cannot handle the situation in which there are partial memberships to groups, we mainly sought to verify that the method behaves as expected in the ideal setting when data sets are simulated from the topic model [2]. To provide some baseline for comparison, we also applied the method of Dey et al [47], which is not strictly a DE analysis method, but does provide a ranking of genes by their “distinctiveness” in each topic. This ranking is based on a simple Kullback-Leibler (K-L) divergence measure; large K-L divergences should signal large differences in expression, as well as high overall levels of expression, so large K-L divergences should correspond to small DE p-values. Since the K-L divergence is not a signed measure, we omitted tests for negative expression differences from the evaluations, which was roughly half of the total number of possible tests for differential expression.
We performed 20 simulations with $K = 2$ topics and $$n = 200$$ cells, and another 20 simulations with $K = 6$ topics and $$n = 1$$,000 cells. To simplify evaluation, all genes either had the same rate of expression in all topics, or the rate was different in exactly one topic. As a result, the total number of expression differences in each data set was roughly the same regardless of the number of simulated topics. Other aspects of the simulations were kept the same as the first set of simulations (see Methods). Similar to before, we took the membership proportions to be known so that misestimation of the membership proportions would not be source of error in the GoM DE analysis and in calculation of the K-L divergence scores.
The largest K-L divergence scores in the simulated data sets reliably recovered true expression differences (Figure 6A, E). Therefore, the K-L divergence scores achieved good true positives rates (i.e., good power) at low false positive rates, FPR = FP/(TN + FP) (see Figure 5 for notation). However, for DE analysis a more relevant performance measure is the false discovery rate, FDR = FP/(TP + FP). Because the K-L divergence score does not fully account for uncertainty in the unknown gene expression differences, many genes with no expression differences among topics were also highly ranked, leading to poor FDR control (Figure 6D, H). By contrast, the GoM DE analysis better accounted for uncertainty in the unknown expression levels. The GoM DE analysis also more accurately recovered true expression differences at small p-values or s-values (Figure 6B, C, F, G), and therefore obtained much lower false discovery rates at corresponding levels of power (Figure 6D, H). Comparing the GoM DE analysis with and without adaptive shrinkage, the adaptive shrinkage did not necessarily lead to better performance (Figure 6D, H), but did provide more directly interpretable measures of significance (s-values or local false sign rates) by shrinking the LFC estimates, and adapting the rate of shrinkage to the data; for example, the expression differences were shrunk more strongly in the $K = 6$ data sets, correctly reflecting the much smaller proportion of true expression differences (compare Figure 6C and G).
## Case study: scRNA-seq epithelial airway data from Montoro et al, 2018
We reanalyzed scRNA-seq data for $$n = 7$$,193 single cells sampled from the tracheal epithelium in wild-type mice [86]. The original analysis [86] used a combination of methods, including t-SNE, community detection [87], diffusion maps [88], and partitioning around medoids (PAM) to identify 7 epithelial cell types: abundant basal and secretory (club) cells; rare, specialized epithelial cell types, including ciliated, neuroendocrine and tuft cells; a novel subpopulation of “ionocytes”; and a novel basal-to-club transitional cell type, “hillock” cells. Although not large in comparison to other modern single-cell data sets, this data set is challenging to analyze, with complex structure, and a mixture of abundant and rare cell types. In contrast to the PBMC data set, there are no existing cell annotations to interpret the topics, so we must rely on inferences made from the expression data alone to make sense of the results.
The topic model fit to the UMI counts with $K = 7$ topics is shown in Figure 7A, and the results of the GoM DE analysis and subsequent GSEA are summarized in Figure 7. Although we do not have cell labels to compare with, distinctive genes emerging from the GoM DE analysis help connect some of the topics to known cell types. For example, the most abundant topics correspond well with predominant epithelial cell types in the lung: topic 1 shows strong overexpression of basal cell marker gene Krt5 [89] (posterior mean l.e. LFC = 4.62); and distinctive genes in topics 2 and 3 include key secretory genes in club cells such as Bpifa1/Splunc1 [90] (l.e. LFC = 4.93) and Scgb1a1 [91] (l.e. LFC = 5.90).
The “hillock” transitional cells, which were originally identified via a diffusion maps analysis [86], emerge as a single topic (topic 4, cyan), with Krt13 (l.e. LFC = 8.04) and Krt4 (l.e. LFC = 5.46) being among the most distinctive genes. The transitional nature of these cells is evoked by their mixed membership; only 237 out of the 7,193 cells have > $90\%$ membership to this topic.
Other less abundant epithelial cell types emerge as separate topics once a topic model is fit separately to the subpopulation of these rare cell types (Figure 7B). These topics recover ciliated cells (topics 8, 9; Ccdc153, posterior l.e. LFC = 5.39), neuroendocrine cells (topic 10; Chga, l.e. LFC = 6.92) and tuft cells (topic 11; Trpm5, l.e. LFC = 6.94). Note that Foxi1+ ionocytes were previously identified as a novel cell type from a small cluster of 26 cells [86], but our analysis failed to distinguish this very rare cell type from the neuroendocrine cells (Additional file 1: Figures S4, S5).
The topics also capture biologically relevant continuous substructure in club cells (topics 2 and 3) and ciliated cells (topics 8 and 9) that was not discovered in the original analysis [86]. This continuous substructure may be reflective of finer scale cell differentiation or specialization of function. In particular, we interpret topic 3 as capturing “canonical” or “mature” (Scgb1a1+, l.e. LFC = 5.90) club cells [90], with negative regulation of inflammation, whereas cells with greater membership to topic 2 are “club-like” (Bpifa1/Splunc1+, l.e. LFC = 3.94) [89, 91]. Topic 9, similarly, appears to represent “canonical” ciliated cells, featuring upregulated genes such as such as Ccdc67/Deup1 (l.e. LFC = 4.82) and Ccdc34 (3.29) [89, 92, 93], and enrichment of Gene Ontology terms [94] such as cilium organization (GO:0044782) and axonemal dynein inner arm assembly (GO:0036159).
In summary, by taking a topic-model-based approach we identified and annotated well-characterized cell types such as basal cells, as well less distinct but potentially interesting substructures such as “Hillock” cells and club cell subtypes.
## Case study: Mouse sci-ATAC-seq Atlas data from Cusanovich et al, 2018
We reanalyzed data from the Mouse sci-ATAC-seq Atlas [97], comprising 81,173 single cells in 13 tissues. First, to provide an overview of the primary structure in the whole data set, we fit a topic model with $K = 13$ topics to these data. The topics correspond closely to the clusters identified in [97] (Additional file 1: Figure S8), and several different tissues are distinguished by different topics (Figure 8A). For the 4 tissues that have replicates, the replicates show a similar composition of the topics (Figure 8A).
Next we performed a more detailed analysis of just the kidney (6,431 cells), fitting a topic model with $K = 10$ to just these cells. We focussed on the kidney cells because, as noted previously [97, 98], both expression and chromatin accessibility vary in relation to the spatial organization of the renal tubular cells, and we predicted that this spatial structure could be better captured by topics rather than by traditional clustering methods. To interpret these topics obtained from chromatin accessibility data, we first used the GoM DE analysis to identify differentially accessible peaks for each topic, then we used “co-accessibility” as predicted by Cicero [95, 97] to connect genes to peaks representing distal regulated sites. Finally, we performed a simple enrichment analysis to identify the “distinctive genes” for each topic, which we defined as the genes with many distal regulatory sites that were differentially accessible.
The results of these analyses are shown in Figure 8. Many of the distinctive genes (Figure 8, Additional file 1: Figure S9, Additional file 6: Table S7) clearly relate topics to known kidney cell types. For example, topic 1 is enriched for genes Klf5 and Elf5 which relate to the collecting duct [98, 99]; topic 3 is enriched for genes Umod, Slc12a1 associated with the loop of Henle [98, 100]); and topics 2, 6, 7 are respectively enriched for genes related to the distal convoluted tubule (Wnk1), podocytes (Col1a2) and glomerular endothelial cells (Ptprb).
Most interestingly, spatial organization of the proximal tubule is captured by two topics; topic 4 is enriched for Slc5a2 (also known as Sglt2) and Slc2a2 (also known as Glut2), associated with the S1 segment of the proximal tube [96, 101, 102], and topic 5 is enriched for Slc5a8 (Smct1) and Atp11a, related to the S3 segment [96, 103]. This result illustrates the ability of the topic model to capture continuous variation in membership of two somewhat complementary processes, which traditional clustering methods are not designed for.
## Case study: chromatin accessibility profiles of the hematopoietic system from Buenrostro et al, 2018
Buenrostro et al [104] studied 2,034 single-cell ATAC-seq profiles of 10 cell populations isolated by FACS to characterize regulation of the human hematopoietic system. Both PCA and t-SNE showed, visually, the expected structure into the main developmental branches (Figure 2 in [104]). However, neither PCA nor t-SNE isolated these branches as individual dimensions of the embedding. Identifying these branches may allow for more precise characterization of the underlying regulatory patterns. Here, by fitting a topic model to the data, the main developmental branches are identified as individual topics (Figure 9A): topic 3, pDC; topic 4, erythroid (MEP); topic 5, lymphoid (CLP); and topic 6, myeloid (GMP and monocytes). Another topic captures the cells at the top of the developmental path (topic 1; HSC and MPP). Other cells at intermediate points in the developmental trajectory, such as CMP, GMP and LMPP cells, are more heterogeneous, and this is reflected by their high variation in topic membership.
To better interpret the regulatory patterns behind each topic, we identified transcription factor (TF) motifs that were enriched for differentially accessible regions in each topic (Figure 9B, Additional file 7: Table S8). Many of the top TF motifs (as ranked by HOMER p-values [105]) point toward regulation of the main developmental trajectories, such as EBF motifs in topic 5 (lymphoid), CEBP motifs in topics 6 and 7 (myeloid), and Hox motifs in topic 1 (HSC and MPP cells). A few topics (topics 8–10) are much less abundant and do not align well with the FACS cell types, and their motif enrichment results were correspondingly more difficult to interpret.
A complication that arose in analyzing these data, which was also noted in [104], is that the cells were obtained from different sources, and this shows up as systematic variation in the chromatin accessibility. This donor effect is captured by topics 1 and 2 in HSC and MPP cells, and, to a lesser extent, in CMP and LMPP cells (Additional file 1: Figure S11). Topic 1 is enriched for Jun and Fos TF motifs, similar to what was found in [104].
## Discussion
The GoM DE analysis is part of a topic-model-based pipeline for analysis of single-cell RNA-seq [47] or ATAC-seq data [49]. This pipeline includes the following steps: [1] fit a topic model to the data; [2] visualize the structure inferred by the topic model; [3] run the GoM DE analysis with the estimated topics; and, optionally, [4] perform other downstream analyses using the results of the GoM DE analysis, e.g., gene set enrichment analysis (for RNA-seq data) or motif enrichment analysis (for ATAC-seq data). Unlike most analysis pipelines for clustering and dimensionality reduction (e.g., [4, 19, 23, 26, 27]), the topic-model-based pipeline is directly applied to the “raw” count data, and therefore does not require an initial step to transform and normalize the data which can lead to downstream issues in the statistical analysis [8, 106–108]. We presented several case studies illustrating the use of the topic-model-based pipeline to analyze single-cell RNA-seq and ATAC-seq data sets. From these case studies, we have drawn a few lessons on the practical challenges that may arise in applying topic modeling approaches to single-cell data, and we share these lessons here. ( See also [47, 49] for related discussion.) One practical question is how to choose K, the number of topics. Many papers have suggested different criteria for determining K. Our view, following [47], is that there is no single “best” K, and we recognize the advantages of learning topics at multiple settings of K; in some data sets, different K’s can reveal structure at different levels of granularity (for example, increasing the number of topics in the Mouse sci-ATAC-seq *Atlas data* revealed more structure within tissues; see https://tinyurl.com/2p99swdk). We have found that it is often helpful to start with a smaller K to elucidate the less granular structure, which is often easier to interpret, then rerun the topic modeling with larger K to identify finer structure.
We proposed annotating topics by distinctive genes identified using the l.e. LFC. One drawback is that this does not reveal the commonalities that may exist among multiple topics, for example, topics corresponding to subpopulations within a common class of cells. A simple alternative to the l.e. LFC, which is also implemented in the fastTopics R package, is to compare against expression under the “null model” (see Methods). We view this as a complementary LFC metric that may reveal additional insights into the topics.
Donor, batch or other technical effects in the single-cell RNA-seq or ATAC-seq data can complicate the analysis and interpretation of the topics if these effects are not small. Since these effects are usually not known, usually we must assess their impact indirectly [109]. For example, the Mouse sci-ATAC-seq *Atlas data* included several replicates, but the replicate effects appeared to be small judging by the fact that the replicates showed a similar composition of topics. By contrast, the donor effects in the human hematopoietic system data were much larger, and in the topic model these donor effects were at least partially captured by individual topics. The broader question of how to deal with non-ignorable donor or batch effects—in particular, how to separate technical effects from biological effects of interest—remains a question of considerable debate and continued investigation [25, 39, 109–116]. In particular, it has been noted that attempting to “correct” for effects can sometimes remove differences that we would like to learn about such as differences in cell type proportions among the batches.
For modeling UMI counts, an open question is whether the Poisson or multinomial model [1] is sufficient, or whether more flexible models are needed. ( This question was investigated in [69] for single-gene models, but not for multi-gene models.) Alternative models such as the negative binomial [117] or Poisson log-normal [80, 118], which can capture additional random variation (“overdispersion”) in underlying expression or measurement error, may result in more robust estimation of the topics.
In single-cell ATAC-seq data, the GoM DE analysis identifies differentially accessible peaks or regions. Usually these peak-level results need to be translated into biological units that are more useful for annotating the topics (e.g., genes, gene sets, transcription factors). In the analysis of the hematopoietic system single-cell ATAC-seq data, we used HOMER [105] to identify TF motifs enriched for differentially accessible peaks. In the analysis of the Mouse sci-ATAC-seq Atlas data, we identified genes enriched for differentially accessible distal regularity sites. Clearly, the quality of the gene enrichment results will depend on our ability to accurately associate peaks with genes. For this, we used the scores computed in [97] using Cicero [95]. However, there are now several alternatives to Cicero that may be preferred [19, 27, 28, 119–122], and in principle any of these approaches could be combined with the peak-level GoM DE results to identify relevant genes.
Recently developed technologies profile both transcription and chromatin accessibility in single cells [123, 124]. For such data, one could fit two topic models, one to the RNA-seq data and another to the ATAC-seq data. With a careful initialization of the topic model fitting algorithm, the topics may be more consistent across the two modalities. But it would be preferrable to analyze the multimodal data jointly for improved accuracy [125–130]. Potentially, the strategy used in MOFA [131, 132] could be adapted for topic modeling—that is, the transcripts and accessibility profiles would share the same membership proportions, L, but each modality would have a different F. However, it remains to be seen how well this strategy works in practice.
## Conclusions
To summarize, we have described a new method that aids in annotating and interpretating the “parts” of cells learned by fitting a topic model to scRNA-seq data or single-cell ATAC-seq data. Our method, GoM DE (differential expression analysis allowing for grades of membership), can be viewed as an extension of existing differential expression methods that allows for mixed membership to multiple groups or topics.
## Models for single-cell ATAC-seq data
In single-cell ATAC-seq data, xij is the number of unique reads mapping to peak or region j in cell i. Although xij can take non-negative integer values, it is common to “binarize” the accessibility data [e.g., 19, 74, 133–135], meaning that xij=1 when at least one read in cell i maps to region j and xij=0 otherwise. For this reason, one might prefer to model the binarized accessibility values as binomial (Bernoulli) random variables. A multinomial model, on the other hand, should better capture the sampling process for reads mapping to regions, but does not account for the truncation of read counts above 1. Therefore, we view both the binomial and multinomial models as approximations. As we explain next, under reasonable assumptions the binomial and multinomial models are similar to each other so it may not matter which model one chooses.
The multinomial topic model for analyzing single-cell ATAC-seq data was suggested by [49]. They assumed the multinomial model [1] in which the xij’s are binarized accessibility values instead of UMI counts.
A binomial model was proposed in [74], [8] xij~Binom1,tirjθij where ti>0 is a cell-specific factor that depends on sequencing coverage and other properties (e.g., amplification, read post-processing [136]), rj>0 is a region-specific factor (say, proportional to the size of the region), and the θij’s capture additional variation in accessibility across cells and regions. Moving forward, we make the simplifying assumption that the regions are all approximately the same size; that is, rj=1 for all $j = 1$,…,m.
The binomial model [8] is closely related to a multinomial model. To make the connection, we first note that the binomial model with rj=1 for all j can be approximated by a Poisson model, [9] xij~Poistiθij. This will be a good approximation when the θij’s are small and the cell-specific factors ti are large, which is usually the case in single-cell ATAC-seq data. Next, we note that the Poisson model [9] and multinomial model [1] are closely related if we choose the size factors to be ti=si [69, 137]; this implies Θ≈Π, where Θ is the n×m matrix with entries θij. By these arguments, the binomial model [8] (also the model used in [74]) and the multinomial model [1] (also the model used in [49]) are similar, and connecting the two models clarifies the assumptions made by each of the models. In particular, the multinomial topic model (1–2) used here and in [49] assumes a low-rank structure in the θij’s across cells and regions; i.e., Θ≈LFT.
## Derivation of GoM DE model
In the “Methods overview,” we motivated the GoM DE model [3] as extending a basic Poisson model expression to allow for partial membership to K groups or topics. The GoM DE model can also be motivated from an approximation to the topic model. Recall, the topic model is a multinomial model [1] in which the multinomial probabilities πij are given by affine combinations of the expression levels fjk in the K topics, πij=∑$k = 1$K likfjk. The non-negativity constraints lik≥0,fjk≥0 and sum-to-one constraints ∑$k = 1$K lik=1,∑$j = 1$p fjk=1 ensure that the πij’s are multinomial probabilities. From a basic identity relating the multinomial and Poisson distributions [138, 139], the multinomial likelihood for the topic model can be replaced with a likelihood formed by a simple product of independent Poissons; that is, [10] Mulinomialxi;si,πi∝∏$j = 1$m Poisxij;siπij, where xi=xi1,…,xim and πi=πi1,…,πim. The approximation then comes from no longer requiring the πij’s to be multinomial probabilities by removing the constraint that f1k+⋯+fmk=1. This allows us to analyze the genes $j = 1$,…,m independently. This is a good approximation so long as si is large and the fjk’s are small. ( A similar approximation was used for GLM-PCA [8].) To be explicit about this approximation, we say πij≈θij (which are no longer guaranteed to be multinomial probabilities) and fjk≈pjk (which are no longer guaranteed to sum to one), resulting in the GoM DE model, which for convenience we restate here: [11] xij~Poissonsiθij,θij=∑$k = 1$klikpjk.
## “Null” model
The simplest Poisson model of the form [3] is one in which θij is the same across all cells i; that is, θij= pj0 for all $i = 1$,…,n. We treat this a “null” model, which can be used to make certain comparisons, e.g., to estimate changes in expression in relative to expression in all cells. The maximum-likelihood estimate (MLE) of pj0 under the null model is [12] pˆj0=∑$i = 1$nxij∑$i = 1$nsi.
## Estimation of log-fold change
In practice, we have found the l.e. LFC to work well, so in our results we use the l.e. LFC. But the l.e. LFC may not be appropriate in all circumstances, and for this reason we note that the GoM DE analysis framework is quite general and accommodates alternatives to the l.e. LFC. Two alternatives are implemented in the software. One alternative is to compare with the “null” model, [13] LFCknull(j)≔log2pjkpj0 Another treats one topic l as a “reference topic”, and compares all other topics k≠l to l using [4].
## Maximum-likelihood estimation
A convenience of the Poisson model allowing for grades of membership is that we can reuse Poisson NMF computations (described below and in more detail in [48]) to compute MLEs of the unknowns pjk: if we consider all genes $j = 1$,…,m simultaneously, we recover a Poisson NMF model, xij~Poissonλij,λij= ∑$k = 1$K hikwjk, by setting hik=silik,wjk=pjk. Therefore, we can reuse the Poisson NMF algorithms to compute MLEs of the unknowns pjk.
## Maximum a posteriori estimation
To improve numerical stability in the parameter estimation, we compute maximum a posteriori (MAP) estimates of pj1,…,pjK in which each pjk is assigned a gamma prior, pjk~Gamma(α,β), with α=1+ε,β=1, and ε>0. Typically ε will be some small, positive number, e.g., ε=0.1. Here we use the parameterization of the gamma distribution from [140] in which α is the shape parameter and β is the inverse scale parameter; under this parameterization, the mean is α/β and the variance is α/β2. The maximum-likelihood computations can be reused for MAP estimation with this gamma prior by adding “pseudocounts” to the data; specifically, MAP estimation of pj1,…,pjK given counts x1j,…,xnj and membership proportions L and is equivalent to maximum-likelihood estimation of pj1,…,pjK given counts x1j,…,xnj,ε,…,ε and membership proportions matrix LIK, where IK is the K×K identity matrix. Unless otherwise stated, we added ε=0.1 pseudocounts to the data.
## Quantifying uncertainty and stabilizing LFC estimates
We implemented a simple Markov chain Monte Carlo (MCMC) algorithm [141, 142] to quantify uncertainty in the LFC estimates. Although normal approximations to likelihoods are typically used by DE methods to quickly obtain analytical measures of uncertainty (e.g., standard errors, confidence intervals) for LFCs, we found that normal approximations to the likelihoods from [5] were sometimes poorly behaved, particularly for lowly expressed genes. Another consideration was that the analytical solutions provide confidence intervals for the unknowns pjk, but ultimately we are interested in quantifying uncertainty in the l.e. LFCs [6] which do not have a simple linear relationship to the pjk’s. Therefore, it is unclear whether the standard analytical solutions can be applied to the l.e. LFCs without making further approximations or simplifications.
MCMC is typically computationally intensive, but with careful implementation (e.g., use of sparse matrix operations and multithreaded computations) the MCMC algorithm is quite fast. Other benefits of using MCMC is that the algorithm can straightforwardly accommodate different choices of LFC statistics and no normality assumptions are needed.
The basic idea behind the MCMC algorithm is as follows: for a given gene j, simulate the posterior distribution of the LFC statistic by performing a “random walk” on gj=gj1,…,gjK, where gjk≔logpjk, $k = 1$,…,K. The random walk generates a sequence of states gj1,…,gjns, in which ns denotes the prespecified length of the simulated Markov chain. After choosing an initial state gj[0], each new state gj(s+1) is generated from the current state gj(s) by the following procedure: first, a topic k∈{1,…,K} is chosen uniformly at random; next, a proposed state gj⋆ is generated as gjk⋆=gjk+δ,δ~N0,σ2, with gjk′⋆=gjk′ for all k′≠k. Assuming an (improper) uniform prior for the unknowns, Prpjk∝1, the proposed state is accepted into the Markov chain with probability [14] 𝒜gj(s),gj⋆=1,Prxj∣pj⋆Prxj∣pj(s)×pjk⋆pjk(s), in which xj is the jth column of the counts matrix X, xj=x1j,…,xnj, and Prxj∣pj is the likelihood at pj,Prxj∣pj=∏$i = 1$n Poissonxi;siθi. ( Note that xj may include pseudocounts.) The standard deviation of the Gaussian proposal distribution, σ, is a tuning parameter. ( Unless otherwise stated, we used = 0.3) The additional pjk⋆/pjk(s) term in the acceptance probability is needed to account for the fact that we are simulating the log-transformed parameters gj, not pj [143, p. 11]. When the proposal is not accepted, the new state is simply copied from the previous state, gj(s+1)=gj(s).
Most of the effort in running the MCMC goes into computing the acceptance probabilities [14], so we have carefully optimized these computations. For example, we have taken advantage of the fact that the count vectors xj are typically very sparse. Additionally, these computations can be performed in parallel since the Markov chains are simulated independently for each gene j.
Once Monte Carlo samples g(s), for $s = 1$,…,ns, have been simulated by this random-walk MCMC, we compute posterior mean LFC estimates, and quantify uncertainty in the LFC estimates. For example, expressing the l.e. LFC for gene j and topic k as a function of the unknowns, LFCkl.e.pj, the posterior mean l.e. LFCs are calculated as ELFCkl.e.pj≈∑$s = 1$ns LFCkl.e.pj(s)/ns.
The final step in the GoM DE analysis is to perform adaptive shrinkage [81] to stabilize the posterior mean estimates. To implement this step, we used the ash function from the ashr R package [144]. We used the same settings as DESeq2 to replicate as closely as possible the performance of DESeq2 with adaptive shrinkage. DESeq2 calls ash with method = “shrink”, which sets the prior to be a mixture of uniforms without a point-mass at zero.
The adaptive shrinkage method takes as input a collection of effect estimates βˆ1,…,βˆm and associated standard errors sˆ1,…,sˆm. In this setting, it is not immediately obvious what are the standard errors, in part because the posterior distribution of the unknowns is not always symmetric about the mean or median. To provide a reasonable substitute summarizing uncertainty in the estimates, we computed Monte Carlo estimates of highest posterior density (HPD) intervals. A (1-α) HPD interval is the smallest interval that contains 100(1-α)% of the probability mass [145, 146]. Specifically, let ajk,bjk denote the (1-α) HPD interval for the LFC estimate of gene j in topic k, and let βˆjk denote the posterior mean. We defined the standard error as sˆjk=bjk-βˆjk when βˆjk<0; otherwise, sˆjk=βˆjk-ajk. Defining the standard errors in this way prevented overshrinking of estimates that were uncertain but had little overlap with zero. We set the size of the HPD intervals to 1-α=0.68 so that the sˆjk would recover conventional standard error calculations when the posterior distribubtion is well approximated by the normal distribution. The revised posterior means and standard errors returned by the adaptive shrinkage method were then used by ashr to calculate test statistics including posterior z-scores (defined as the posterior mean divided by the posterior standard error [147]), local false sign rates (lfsr) and s-values.
An important question is the choice of ns. One heuristic way to assess whether ns is large enough is to perform two independent MCMC runs initialized with different pseudorandom number generator states (“seeds”) and check consistency of the posterior estimates from the two runs. ( We checked consistency of the posterior estimates after stabilizing the estimates using adaptive shrinkage, as described above.) In simulated data sets (below), comparison of two independent MCMC runs suggested that ns=10,000 was sufficient to obtain reasonably accurate estimates of posterior means and posterior z-scores for all genes (Additional file 1: Figure S2). Therefore, we performed initial MCMC simulations for all single-cell data sets using ns=10,000. The runtimes for performing these MCMC simulations on the single-cell data sets (described below), with ns=10,000, are given in Table 1.
Although this consistency check suggested that running a simulation with ns=10,000 would be “good enough”, to provide additional assurance we performed another consistency assessment in the PBMC data set. We found that even better consistency was achieved with ns=100,000 (Additional file 1: Figure S12). Therefore, to provide more reliable results, the final GoM DE results were generated with ns=100,000.
The GoM DE analysis methods are implemented in the de_analysis function in the fastTopics package [184].
## Single-cell data sets
All data sets analyzed were stored as sparse n×m matrices X, where n was the number of cells and m was the number of genes or regions. The data sets are summarized in Table 1.
## Preparation of scRNA-seq data
Since the topic model is a multinomial model of count data, no log-normalization or other transformation of the scRNA-seq molecule counts was needed. Further, we kept all genes other than those with no variation in the data set. ( *This is* done in part to demonstrate that our methods are robust to including genes with little variation.) Also note that due to the use of sparse matrix techniques in our software implementations, including genes with low variation did not greatly increase computational effort.
## Preparation of single-cell ATAC-seq data
As previously suggested [19, 133–135]), we “binarized” the single-cell ATAC-seq data; that is, we assigned xij=1 (“accessible”) when least one fragment in cell i mapped to peak or region j, otherwise xij=0 (“inaccessible”). There are at least a couple reasons for doing this. For small peaks (say, < 5 kb), read counts do not provide a reliable quantitative measure of accessibility in single cells. This is because the (random) first insertion restricts the space for subsequent insertions. Additionally, insertions could occur within the same site on the same allele or on each of the two alleles, complicating interpretation of the read counts.
Like the RNA molecule count data (see above), we kept all regions except those that showed no variation.
## PBMC data from Zheng et al, 2017
We combined reference transcriptome profiles generated from 10 bead-enriched subpopulations of PBMCs (Donor A) processed using Cell Ranger 1.1.0 [29, 148]. We downloaded the “Gene/cell matrix (filtered)” tar.gz file from the 10x Genomics website for each of the following 10 FACS-purified data sets: CD14+ monocytes, CD19+ B cells, CD34+ cells, CD4+ helper T Cells, CD4+/CD25+ regulatory T Cells, CD4+/CD45RA+/CD25− naive T cells, CD4+/CD45RO+ memory T Cells, CD56+ natural killer cells, CD8+ cytotoxic T cells and CD8+/CD45RA+ naive cytotoxic T cells. After combining these 10 data sets, then filtering out unexpressed genes, the combined data set contained molecule counts for 94,655 cells and 21,952 genes; $97.1\%$ of the molecule counts were zero.
In Figure 1, the 54,132 cells from these data sets were labeled as “T cells”: CD4+ helper T Cells, CD4+/CD25+ regulatory T Cells, CD4+/CD45RA+/CD25− naive T cells, CD4+/CD45RO+ memory T Cells and CD8+/CD45RA+ naive cytotoxic T cells.
## Epithelial airway data from Montoro et al, 2018
We analyzed a mouse epithelial airway data set from [86, 149]. These were gene expression profiles of trachea epithelial cells in C57BL/6 mice obtained using droplet-based 3’ scRNA-seq, processed using the GemCode Single Cell Platform. We downloaded file GSE103354_Trachea_droplet_UMIcounts.txt.gz. This file also contained the cluster assignments that we compared with. ( In [86], the samples were subdivided into 7 clusters using a community detection algorithm.) After removing genes that were not expressed in any of the cells, the data set contained molecule counts for 7,193 cells and 18,388 genes ($90.7\%$ of counts were zero).
## Mouse Atlas data from Cusanovich et al, 2018
Cusanovich et al [97] profiled chromatin accessibility by single-cell combinatorial indexing ATAC-seq (sciATAC-seq) [150, 151] in nuclei from 13 distinct tissues of a 8-week-old male C57BL/6J mouse. Replicates for 4 of the 13 tissues were obtained by profiling chromatin accessibility in a second mouse. We downloaded the (sparse) binarized peak × cell matrix in RDS format, atac matrix.binary.qc_filtered.rds, from the Mouse sci-ATAC-seq Atlas website [152]. We also downloaded cell_metadata.txt which included cell types estimated by a clustering of the cells (see Table S1 in [97]). The full data set used in our analysis (13 tissues, including 4 replicated tissues) consisted of the binary accessibility values for 81,173 cells and 436,206 peaks ($1.2\%$ overall rate of accessibility). Note that all peaks had fragments mapping to at least 40 cells, so no extra step was taken to filter out peaks.
Separately, we analyzed the sci-ATAC-seq data from kidney only, in which peaks with fragments mapping to fewer than 20 kidney cells were removed, resulting in data set containing binary accessibility values for 6,431 cells and 270,864 peaks. Base-pair positions of the peaks were based on Mouse Genome Assembly mm9 (NCBI and Mouse Genome Sequencing Consortium, Build 37, July 2007).
From the Mouse sci-ATAC-seq website, we also downloaded the file master_cicero_conns.rds containing the Cicero co-accessibility predictions [95, 152], which we used to link chromatin accessibility peaks to genes. For the kidney data, we connected a peak given in the “Peak2” column of the Cicero co-accessibility data table to a gene given in the “peak1.tss.gene_id” column if the “cluster” column was 11, 18, 22 or 25. ( These four clusters were the main kidney-related clusters identified in [97].) This extracted, for each gene, the distal and proximal sites connected to the gene associated with Peak1 (specifically, a gene in which the transcription start site overlaps with Peak1). Among the 22,194 genes associated with at least one peak, the median number of peaks connected to a gene was 19, and the largest number of peaks was 179 (for Bahcc1 on chromosome 11). Among the 270,864 peaks included in the topic modeling analysis, 113,489 ($42\%$) were connected to at least one gene, $95\%$ of peaks were connected to 10 genes or fewer, and the largest number of connected genes was 60.
## Human hematopoietic system data from Buenrostro et al, 2018
Buenrostro et al [104] used FACS to isolate 10 hematopoietic cell populations from human bone marrow and blood, then the cells were assayed using single-cell ATAC-seq. The processed single-cell ATAC-seq data were downloaded from [153]; specifically, file GSE96769_scATACseq_counts.txt.gz containing the fragment counts, and file GSE96769_PeakFile_-20160207.bed.gz containing peaks obtained from bulk ATAC-seq data [104]. Although there may be benefits to calling peaks using aggregated single-cell data instead [154], we used the original accessibility data based on the bulk ATAC-seq peaks so that our analysis was more directly comparable to the analysis of [104].
Following [104, 154], we extracted the 2,034 samples passing quality control filters, then we “binarized” the counts. The list of 2,034 cells considered “high quality” was obtained from file metadata.tsv included in the online benchmarking repository [154]. After removing peaks with fragments mapping to fewer than 20 cells, the final data set used in our analysis consisted of binary accessibility values for 2,034 cells and 126,719 peaks ($4.6\%$ overall rate of accessibility). Base-pair positions of the peaks were based on human genome assembly 19 (Genome Reference Consortium Human Build 37, February 2009).
In [104], a large, patient-specific batch effect was identified in the accessibility profiles for the HSC cells, and therefore steps were taken in [104] to normalize the accessibility data before performing PCA. We instead fit the topic model to the unnormalized binary accessibility values, in part to find out how well the topic model can cope with the complication of a batch effect. In agreement with [104], this batch effect is at least partly captured by the topics, although in our analysis the batch effect also appeared in MPP cells and, to a lesser extent, in CMP cells (Additional file 1: Figure S11).
## Fitting the topic models
In brief, we took the following steps to fit a topic model. All these steps are implemented in the R package fastTopics.
First, we fit a Poisson NMF model [37, 155], [15] xij~PoissonλijΛ=HWT, where Λ∈Rn×m is a matrix of the same dimension as X with entries λij≥0 giving the Poisson rates for the counts xij. The parameters of the Poisson NMF model are stored as two matrices, H∈Rn×K,W∈Rm×K, with non-negative entries hik,wjk. fastTopics has efficient implementations of algorithms for computing maximum-likelihood estimates (MLEs) of W,H [48].
Second, we recovered MLEs of F,L from MLEs of W,H by a simple reparameterization [48].
In an empirical comparison of Poisson NMF algorithms with count data sets, including scRNA-seq data sets [48], we found that a simple co-ordinate descent (CD) algorithm [156, 157], when accelerated with the extrapolation method of Ang and Gillis [158], almost always produced the best Poisson NMF (and topic model) fits, and in the least amount of time. To confirm this, we compared topic model fits obtained by running the same four algorithms that were compared in [48]—EM and CD, with and without extrapolation—on the PBMC data set, and assessed the quality of the fits. We evaluated the model fits in two ways: using the likelihood, and using the residuals of the KarushKuhn-Tucker (KKT) first-order conditions. ( The residuals of the KKT system should vanish as the algorithm approaches maximum-likelihood estimates of W,H.) Following [48], to reduce the possibility that multiple optimizations converge to different local maxima of the likelihood, which could complicate these comparisons, we first ran 1,000 EM updates, then we examined the performance of the algorithms after this initialization phase (Additional file 1: Figures S13, S14). Consistent with [48], the extrapolated CD updates always produced the best fit, or at the very least a fit that was no worse than the other algorithms, and almost always converged on a solution more quickly than the other algorithms. Therefore, subsequently we used the extrapolated CD updates to fit the topic models.
In more detail, the pipeline for fitting topic models consisted of the following steps: [1] Initialize W using Topic-SCORE [159]; [2] perform 10 CD updates of H, with W fixed; [3] perform 1,000 EM updates (without extrapolation) to get close to a solution (“prefitting phase”); [4] run an additional 1,000 extrapolated CD updates to improve the fit (“refinement phase”); and [5] recover F,L from W,H by a simple transformation. The prefitting phase was implemented by calling fit_poisson_nmf from fastTopics with these settings: numiter = 1000, method = “em”, control = list (numiter = 4). The refinement phase was implemented with a second call to fit_poisson_nmf, with numiter = 1000, method = “scd”, control = list (numiter = 4, extrapolate = TRUE), in which the model fit was initialized using the fit from the prefitting phase. The topic model fit was recovered by calling poisson2multinom in fastTopics. Note that only the estimates of L were used in the GoM DE analysis.
For each data set, we fit topic models with different choices of K and compared the fits for each K by comparing their likelihoods (Additional file 1: Figure S15).
## Visualizing the membership proportions
The membership proportions matrix L can be viewed as an embedding of the cells $i = 1$,…,n in a continuous space with K-1 dimensions [50]. ( *It is* K-1 dimensions because of the constraint that the membership proportions for each cell must add up to 1.) A simple way to visualize this embedding in 2-d is to apply a nonlinear dimensionality reduction technique such as t-SNE [11, 160] or UMAP [12] to L ([49] used t-SNE). We have also found that plotting principal components (PCs) of the membership proportions can be an effective way to explore the structure inferred by the topic model (Additional file 1: Figures S1, S4). However, we view these visualization techniques as primarily for exploration, and a more powerful approach is to visualize all K-1 dimensions simultaneously using a Structure plot [70, 71]. Here we describe some improvements to the Structure plot for better visualization. These improvements are implemented in the structure_plot function in fastTopics.
When cells were labeled, we compared topics against labels by grouping the cells by these labels in the Structure plot. We then applied t-SNE to the L matrix, separately for each group, to arrange the cells on a line within each group. For this, we used the R package Rtsne [161]. ( In fastTopics, we also implemented options to arrange the cells in each group using UMAP or PCA, but in our experience we found that UMAP and PCA produced “noisier” visualizations.) Arranging the cells by 1-d t-SNE worked best for smaller groups of cells with less complex structure. For large groups of cells, or for unlabeled single-cell data sets, we randomly subsampled the cells to reduce t-SNE runtime. ( When cells number in the thousands, it is nearly impossible to distinguish individual cells in the Structure plot anyhow.) Even with this sub-sampling, the Structure plot sometimes did not show fine-scale substructures or rare cell types. Therefore, in more complex cases, we first subdivided the cells into smaller groups based on the membership proportions, then ran t-SNE on these smaller groups. These groups were either identified visually from PCs of L, or in a more automated way by running k-means on PCs of L (see [162]).
## Gene enrichment analysis based on differential accessibility of peaks connected to genes
Here we describe a simple approach to obtain gene-level statistics from the results of a differential accessibility analysis. This approach was applied in the topic modeling analysis of the Mouse Atlas kidney cells.
Cusanovich et al [97] used the Cicero co-accessibility predictions and the binarized single-cell ATAC-seq data to compute a “gene activity score” Rki for each gene k and cell i. Here we have a related but different goal: we would like to use the results of the differential accessibility analysis, which generates differential accessibility estimates and related statistics for each peak and each topic, to rank genes according to their importance to a given topic. A difficulty, however, with ranking the genes is that the Cicero co-accessibility predictions are uncertain, and they are only partially informative about which peaks are relevant to a gene. In aggregate, however, the expectation is that the “most interesting” genes will be genes that are (a) predicted by Cicero to be connected many peaks that are differentially accessible and (b) the differences in accessibility are mainly in the same direction. This suggests an enrichment analysis in which, for each gene, we test for enrichment of differential acccessibility among the peaks connected to that gene. Here we describe a simple enrichment analysis for (a) and (b).
For (a), we computed a Bayes factor [163] measuring the support for the hypothesis that at least one of the peaks is differentially accessible (the LFC is not zero) against the null hypothesis that none of the peaks are differentially accessible. For (b), we computed the average LFC among all differentially accessible peaks (that is, peaks with nonzero LFC according to some significance criterion).
We implemented this gene enrichment analysis by running adaptive shrinkage [81] separately for each gene and topic. This had the benefit of adapting the shrinkage separately to each gene in each topic. In particular, in comparison to the usual adaptive shrinkage step for a GoM DE analysis (see above), it avoided overshrinking differences for genes exhibiting strong patterns of differential accessibility. We took the following steps to implement this adaptive shrinkage analysis. First, we ran function ash from the ashr package [144] once on the posterior mean l.e. LFC estimates βˆjk and their standard errors sˆjk for all topics k and all peaks j, with settings mixcompdist = “normal”, method = “shrink”. This was done only to determine the variances in the mixture prior and to get a “default” model fit to be used in the subsequent adaptive shrinkage analyses.
Next, we ran ash separately for gene and each topic k using the l.e. LFC estimates βˆjk and standard errors sˆjk from the peaks j connected to that gene. We set the variances in the mixture prior to the variances determined from all the l.e. LFC estimates, and used ash settings mixcompdist = “normal” and pointmass = FALSE. One issue with running adaptive shrinkage using only the l.e. LFC estimates for the peaks connected to a gene is that some genes have few Cicero connections, leading to potentially unstable fits and unreliable posterior estimates. We addressed this issue by encouraging the fits toward the “default model” that was fitted to all genes and all topics; specifically, we set the Dirichlet prior on the mixture proportions to be Dirichlet α1,…,αK with prior sample sizes αk=1.01+n0πˆkdefault, where here K denotes the number of components of the prior mixture (not the number of topics), πˆkdefault denotes the k th mixture proportion in the adaptive shrinkage prior for the fitted “default” model, and n0=20. This stabilized the fits for genes with few Cicero connections while still allowing some ability to adapt to genes with many connections.
Finally, we used the logLR output from ash as a measure of support for enrichment (this is the Bayes factor on the log-scale), and we computed the mean l.e. LFC as the average of the posterior mean estimates of the l.e. LFCs taken over all peaks j connected to the gene and with posterior lfsr < 0.05.
## Motif enrichment analysis for differentially accessible regions
We used HOMER [105] to identify transcription factor (TF) motifs enriched for differentially accessible regions, separately for each topic estimated from the single-cell ATAC-seq data. For each topic $k = 1$,…,K, we applied the HOMER Motif Analysis tool findMotifsGenome.pl to estimate motif enrichment in differentially accessible regions; specifically, we took “differentially accessible regions” to be those with p-value less than 0.05 in the GoM DE analysis (Additional file 1: Figure S16). These differentially accessible regions were stored in a BED file positions.bed. The exact call from the command-line shell was findMotifsGenome.pl positions.bed hg19 homer −len 8, 10, 12 −size 200 −mis 2 −S 25 −p 4 −h.
Note that the adaptive shrinkage step was skipped in the GoM DE analysis, so these are the p-values for the unmoderated l.e. LFC estimates. The reason for skipping the adaptive shrinkage step is that the shrinkage is performed uniformly for the LFC estimates for all regions, and since the vast majority of regions have l.e. LFC estimates are that are indistinguishable from zero, the result is that very few differentially accessible regions remain shrinkage.
## Gene sets
Human and mouse gene sets for the gene set enrichment analyses (GSEA) were compiled from the following gene set databases: NCBI BioSystems [164]; Pathway Commons [165, 166]; and MSigDB [167–169], which includes Gene Ontology (GO) gene sets [94, 170]. Specifically, we downloaded bsid2info.gz and biosystems_gene.gz from the NCBI FTP site (https://ftp.ncbi.nih.gov/gene) on March 22, 2020; PathwayCommons12.All.hgnc.gmt.gz from the Pathway Commons website (https://www.pathwaycommons.org) on March 20, 2020; and msigdb_v7.2.xml.gz from the MSigDB website (https://www.gsea-msigdb.org) on October 15, 2020. For the gene set enrichment analyses we also downloaded human and mouse gene information (“gene info”) files Homo_sapiens.gene_info.gz and Mus_musculus.gene_info.gz from the NCBI FTP site on October 15, 2020. Put together, we obtained 37,856 human gene sets and 33,380 mouse gene sets. In practice, we filtered gene sets based on certain criteria before running the GSEA. To facilitate integration of these gene sets into our analyses, we have compiled these gene sets into an R package [171].
## Gene set enrichment analysis
We took a simple multiple linear regression approach to the gene set enrichment analysis (GSEA), in which we modeled the l.e. LFC estimate for gene i in a given topic, here denoted by yi, as yi=μi+∑$j = 1$n xijbj+ei, ei~N0,σ2, in which xij∈{0,1} indicates gene set membership; xij=1 if gene i belongs to gene set j, otherwise xij=0. ( We represented the geneset membership as a sparse matrix since most xij’s are zero.) Here, n denotes the number of candidate gene sets, and σ2 is the residual variance to be estimated. The idea behind this simple approach was that the most relevant gene sets are those that best explain the log-fold changes yi, and therefore in the multiple regression we sought to identify these gene sets by finding coefficients bj that were nonzero with high probability. See [172, 173] for similar ideas using logistic regression. Additionally, since many genes were typically differentially expressed in a given topic, modeling LFCs helped distinguish among DE genes that showed only a slight increase in expression versus those that were highly overexpressed [174, 175]. Of course, this simple multiple linear approach ignores uncertainty in the LFC estimates yi, which is accounted for in most gene set enrichment analyses. We addressed this issue by shrinking the l.e. LFC estimates prior to running the GSEA; that is, we took yi to be the the posterior mean LFC estimate after applying adaptive shrinkage, as described above (see “Quantifying uncertainty and stabilizing LFC estimates”). The result was that genes that we were more uncertain about had have an l.e. LFC estimate yi that was zero or near zero.
We implemented this multiple linear regression approach using SuSiE (susieR version 0.12.10) [176]. A benefit to using SuSiE is that it automatically organized similar or redundant gene sets into “credible sets” (CSs), making it easier to quickly recognize complementary gene sets; see [177–182] for related ideas.
In detail, the GSEA was performed as follows. We performed a separate GSEA for each topic, $k = 1$,…,K. Specifically, for each topic k, we ran the susieR function susie with the following options: $L = 10$, intercept = TRUE, standardize = FALSE, estimate_residual_variance = TRUE, refine = FALSE compute_univariate_zscore = FALSE and min_abs_corr =0. We set $L = 10$ so that SuSiE returned at most 10 credible sets. For a given topic k, we reported a gene set as being enriched if it was included in at least one CS. We organized the enriched gene sets by ($95\%$) credible sets. We also recorded the Bayes factor for each CS, which gives a measure of the level of support for that CS. For each gene set included in a CS, we reported the posterior inclusion probability (PIP), and the posterior mean estimate of the regression coefficient bj. In the results, we refer to bj as the “enrichment coefficient” for gene set j since it is an estimate of the expected increase in the l.e. LFC for genes that belong to gene set j relative to genes that do not belong to the gene set.
Often, a CS contained only one gene set, in which case the PIP for that gene set was close to 1. In several other cases, the CS contained multiple similar gene sets; in these cases, the smaller PIPs indicated that it was difficult to choose among the gene sets because they are similar to each other. ( Note that the sum of the PIPs in a $95\%$ CS should always be above 0.95 and less than 1.) Occasionally, SuSiE returned a CS with a small Bayes factor containing a very large number of gene sets. We excluded such CSs from the results.
When repeated these gene set enrichment analyses with two collections of gene sets: [1] all gene sets other than the MSigDB collections C1, C3, C4 and C6, and “archived” MSigDB gene sets; and [2] only gene sets from curated pathway databases, specifically Pathway Commons, NCBI BioSystems and “canonical pathways” (CP) in the MSigDB C2 collection, and Gene Ontology (GO) gene sets in the MSigDB C5 collection. In all cases, we removed gene sets with fewer than 10 genes and with more than 400 genes. Table 2 gives the exact number of gene sets included in each GSEA.
## Simulations
For evaluating the DE analysis methods, we generated matrices of UMI counts X∈Rn×m for $m = 10$,000 genes and $$n = 200$$ or $$n = 1$$,000 cells. We simulated the UMI counts xij from a Poisson NMF model [15] in which W and H were chosen to emulate UMI counts from scRNA-seq experiments.
The matrices W and H were generated as follows. First, for each cell i, we generated membership proportions li1,…,liK then set hik=silik, for $k = 1$,…,K, where si is the total UMI count. To simulate the wide range of total UMI counts often seen in scRNA-seq data sets, total UMI counts si were normally distributed on the log-scale, si=10ui,ui~N(0,$\frac{1}{5}$), where N(μ,σ) denotes the univariate normal distribution with mean μ and standard deviation σ.
Membership proportions lik for each cell i were generated so as to obtain a wide range of mixed memberships, according to the following procedure: the number of nonzero proportions was set to K′∈ {1,…,K} with probability 2-K′; the K′ selected topics t1,…,tK′⊆{1,…,K} were drawn uniformly at random (without replacement) from 1,…,K; then the membership proportions for the selected topics were set to 1 when K′=1, or, when K′>1, they were drawn from the Dirichlet distribution with shape parameters αt1,…,αtK′.
Expression rates wjk were generated so as to emulate the wide distribution of gene expression levels observed in single-cell data sets, and to allow for differences in expression rates among topics. The procedure for generating the expression rates for each gene j was as follows: with probability 0.5, the expression rates were the same across all topics, and were generated as fj1=⋯=fjK=2vj,vj~N[-4,2]. Otherwise, with probability 0.5, the expression rates were the same in all topics except for one topic. The differing topic k′ was chosen uniformly at random from 1,…,K, then the expression rate for topic k′ was set to fjk′=2vj+ej,ej~N[0,1]. As a result, the expression rates were roughly normally distributed on the log-scale, and the expression differences were also normally distributed on the log-scale. About half of genes had an expression difference among the topics.
Using this simulation procedure, we generated three collections of data sets. The simulation settings were altered slightly for each collection. In the first, data sets were simulated with $K = 2$,α=($\frac{1}{100}$,$\frac{1}{100}$), $$n = 200$$ so that most membership proportions were equal or very close to 0 or 1. In the second, we used $K = 2$,α=[1,1],$$n = 200$$ to allow for a range of mixed memberships. In the third, we generated data sets with $K = 6$,α=(1,…,1),$$n = 1$$,000.
For the data sets simulated with $K = 2$,α=($\frac{1}{100}$,$\frac{1}{100}$), the cells could essentially be subdivided into two groups. Therefore, we ran MAST [84, 183] and DESeq2 [78, 83] to test for genes that were differentially expressed between the two groups. MAST (R package version 1.20.0) was called via the FindMarkers interface in Seurat [25] (Seurat 4.0.3, SeuratObject 4.0.2) with the following settings: ident. 1=“2”, ident. 2 = NULL, test. use = “MAST”, logfc.threshold =0, min.pct =0. DESeq was called from the DESeq2 R package (version 1.34.0) using settings recommended in the package vignette: test = “LRT”, reduced = ~1, useT = TRUE, minmu =1e-6, minReplicatesForReplace = Inf. Size factors were calculated using the calculateSumFactors method from scran version 1.22.1 [23]. The LFC estimates returned by DESeq were subsequently revised using adaptive shrinkage [81] by calling IfcShrink in DESeq2 with type = “ashr”, svalue = TRUE. ( As in the GoM DE analysis, the DESeq2 posterior z-scores were defined as the posterior means divided by the posterior standard errors returned by the adaptive shrinkage.) To perform the GoM DE analysis in each of the simulations, we first fit a Poisson NMF model to the simulated counts X using fit_poisson_nmf from the fastTopics R package [48, 184] (version 0.697). The loadings matrix H was fixed to the matrix used to simulate the data, and W was estimated by running 40 co-ordinate ascent updates on W alone (update. loadings = NULL, method = “scd”, numiter = 40). The equivalent topic model fit was then recovered. Three GoM DE analyses were performed using the de_analysis function from the fastTopics R package, with the topic model fit provided as input: one analysis without adaptive shrinkage (shrink.method = “none”), and two analyses with adaptive shrinkage (shrink.method = “ash”, ashr version 2.2–51 [144]) in which the MCMC was initialized with different pseudorandom number generator states. In all three runs, posterior calculations were performed with ns=10,000,ε=0.01. Comparison of the two MCMC runs (with adaptive shrinkage) suggested that ns=10,000 was sufficient to obtain reasonably accurate posterior estimates in these simulations (Additional file 1: Figure S2).
## Computing environment
Most computations on real data sets were run in R 3.5.1 [185], linked to the OpenBLAS 0.2.19 optimized numerical libraries, on Linux machines (Scientific Linux 7.4) with Intel Xeon E5–2680v4 (“Broad-well”) processors. For performing the Poisson NMF optimization, which included some multithreaded computations, as many as 8 CPUs and 16 GB of memory were used. The DESeq2 analysis of the PBMC data was performed in R 4.1.0, using 4 CPUs and 264 GB of memory. The evaluation of the DE analysis methods in simulated data sets was performed in R 4.1.0, using as many and 8 CPUs as 24 GB of memory. More details about the computing environment, including the R packages used, are recorded in the workflowr pages in the companion code repositories [186, 187].
## Availability of data and materials
The fastTopics R package is available on GitHub (https://github.com/stephenslab/fastTopics) and CRAN (https://cran.r-project.org/package=fastTopics). A Seurat wrapper for fastTopics is available from https://github.com/stephenslab/seurat-wrappers. The data sets supporting the conclusions of this article are available in Zenodo repositories [186, 187]. These Zenodo repositories also include the source code implementing the analyses and workflowr websites [188] for browsing the code and results. Permission to use the source code in these repositories is granted under the MIT license. Numerical implementations of the contributed statistical methods, including tools for visualizing the results generated by these methods, are available from the fastTopics R package [48, 184] under the MIT license. *The* gene sets used in the GSEA were compiled into an R package [171], also distributed under the MIT license. All data sets used in the study were obtained from public sources [148, 149, 152, 153]. A description of how these data sets were used is provided in Methods.
## Funding
This work was supported by the NHGRI at the National Institutes of Health under award number 5R01HG002585.
## References
1. Kharchenko P.V.. **The triumphs and limitations of computational methods for scRNA-seq**. *Nature Methods* (2021) **18** 723-732. PMID: 34155396
2. Stegle O., Teichmann S.A., Marioni J.C.. **Computational and analytical challenges in single-cell transcriptomics**. *Nature Reviews Genetics* (2015) **16** 133-145
3. Wagner A., Regev A., Yosef N.. **Revealing the vectors of cellular identity with single-cell genomics**. *Nature Biotechnology* (2016) **34** 1145-1160
4. Duò A., Robinson M.D., Soneson C.. **A systematic performance evaluation of clustering methods for single-cell RNA-seq data**. *F1000Research* (2018) **7** 1141. PMID: 30271584
5. Freytag S., Tian L., Lönnstedt I., Ng M., Bahlo M.. **Comparison of clustering tools in R for medium-sized 10x Genomics single-cell RNA-sequencing data**. *F1000Research* (2018) **7** 1297. PMID: 30228881
6. Kiselev V.Y., Andrews T.S., Hemberg M.. **Challenges in unsupervised clustering of single-cell RNA-seq data**. *Nature Reviews Genetics* (2019) **20** 273-282
7. Sun S., Zhu J., Ma Y., Zhou X.. **Accuracy, robustness and scalability of dimensionality reduction methods for single-cell RNA-seq analysis**. *Genome Biology* (2019) **20** 269. PMID: 31823809
8. Townes F.W., Hicks S.C., Aryee M.J., Irizarry R.A.. **Feature selection and dimension reduction for single-cell RNA-Seq based on a multinomial model**. *Genome Biology* (2019) **20** 295. PMID: 31870412
9. Hotelling H.. **Analysis of a complex of statistical variables into principal components**. *Journal of Educational Psychology* (1933) **24** 417-441
10. Risso D., Perraudeau F., Gribkova S., Dudoit S., Vert J.-P.. **A general and flexible method for signal extraction from single-cell RNA-seq data**. *Nature Communications* (2018) **9** 284
11. van der Maaten L., Hinton G.. **Visualizing data using t-SNE**. *Journal of Machine Learning Research* (2008) **9** 2579-2605
12. Mclnnes L., Healy J., Saul N., Großberger L.. **UMAP: uniform manifold approximation and projection**. *Journal of Open Source Software* (2018) **3** 861
13. Becht E., Mclnnes L., Healy J., Dutertre C.-A., Kwok I.W.H., Ng L.G., Ginhoux F., Newell E.W.. **Dimensionality reduction for visualizing single-cell data using UMAP**. *Nature Biotechnology* (2019) **37** 38-44
14. Cooley S.M., Hamilton T., Aragones S.D., Ray J.C.J., Deeds E.J.. **A novel metric reveals previously unrecognized distortion in dimensionality reduction of scRNA-seq data**. *bioRxiv* (2022). DOI: 10.1101/689851
15. Chari T., Pachter L.. **The specious art of single-cell genomics**. *PLoS Computational Biology* (2023) **19** 1011288
16. Kobak D., Berens P.. **The art of using t-SNE for single-cell transcriptomics**. *Nature Communications* (2019) **10** 5416
17. Kobak D., Linderman G.C.. **Initialization is critical for preserving global data structure in both t-SNE and UMAP**. *Nature Biotechnology* (2021) **39** 156-157
18. Wattenberg M., Viégas F., Johnson I.. **How to use t-SNE effectively**. *Distill* (2016). DOI: 10.23915/distill.00002
19. Granja J.M., Corces M.R., Pierce S.E., Bagdatli S.T., Choudhry H., Chang H.Y., Greenleaf W.J.. **ArchR is a scalable software package for integrative single-cell chromatin accessibility analysis**. *Nature Genetics* (2021) **53** 403-411. PMID: 33633365
20. Heiser C.N., Lau K.S.. **A quantitative framework for evaluating single-cell data structure preservation by dimensionality reduction techniques**. *Cell Reports* (2020) **31** 107576. PMID: 32375029
21. Linderman G.C., Rachh M., Hoskins J.G., Steinerberger S., Kluger Y.. **Fast interpolation-based t-SNE for improved visualization of single-cell RNA-seq data**. *Nature Methods* (2019) **16** 243-245. PMID: 30742040
22. Linderman G.C., Steinerberger S.. **Clustering with t-SNE, provably**. *SIAM Journal on Mathematics of Data Science* (2019) **1** 313-332. PMID: 33073204
23. Lun A.T.L., McCarthy D.J., Marioni J.C.. **A step-by-step workflow for low-level analysis of single-cell RNA-seq data with Bioconductor**. *F1000Research* (2016) **5** 2122. PMID: 27909575
24. McCarthy D.J., Campbell K.R., Lun A.T.L., Wills Q.F.. **Scater: Pre-processing, quality control, normalization and visualization of single-cell RNA-seq data in R**. *Bioinformatics* (2017) **33** 1179-1186. PMID: 28088763
25. Stuart T., Butler A., Hoffman P., Hafemeister C., Papalexi E., Mauck W.M., Hao Y., Stoeckius M., Smibert P., Satija R.. **Comprehensive integration of single-cell data**. *Cell* (2019) **177** 1888-1902. PMID: 31178118
26. Butler A., Hoffman P., Smibert P., Papalexi E., Satija R.. **Integrating single-cell transcriptomic data across different conditions, technologies, and species**. *Nature Biotechnology* (2018) **36** 411-420
27. Stuart T., Srivastava A., Madad S., Lareau C.A., Satija R.. **Single-cell chromatin state analysis with Signac**. *Nature Methods* (2021) **18** 1333-1341. PMID: 34725479
28. Fang R., Preissl S., Li Y., Hou X., Lucero J., Wang X., Motamedi A., Shiau A.K., Zhou X., Xie F., Mukamel E.A., Zhang K., Zhang Y., Behrens M.M., Ecker J.R., Ren B.. **Comprehensive analysis of single cell ATAC-seq data with SnapATAC**. *Nature Communications* (2021) **12** 1337
29. Zheng G.X.Y., Terry J.M., Belgrader P., Ryvkin P., Bent Z.W., Wilson R., Ziraldo S.B., Wheeler T.D., McDermott G.P., Zhu J., Gregory M.T., Shuga J., Montesclaros L., Underwood J.G., Masquelier D.A., Nishimura S.Y., Schnall-Levin M., Wyatt P.W., Hindson C.M., Bharadwaj R., Wong A., Ness K.D., Beppu L.W., Deeg H.J., McFarland C., Loeb K.R., Valente W.J., Ericson N.G., Stevens E.A., Radich J.P., Mikkelsen T.S., Hindson B.J., Bielas J.H.. **Massively parallel digital transcriptional profiling of single cells**. *Nature Communications* (2017) **8** 14049
30. Brunet J.-P., Tamayo P., Golub T.R., Mesirov J.P.. **Metagenes and molecular pattern discovery using matrix factorization**. *Proceedings of the National Academy of Sciences* (2004) **101** 4164-4169
31. Donoho D., Stodden V.. *Proceedings of the 16th International Conference on Neural Information Processing Systems* (2003) 1141-1148
32. Durif G., Modolo L., Mold J.E., Lambert-Lacroix S., Picard F.. **Probabilistic count matrix factorization for single cell expression data analysis**. *Bioinformatics* (2019) **35** 4011-4019. PMID: 30865271
33. Gong W., Rasmussen T.L., Singh B.N., Koyano-Nakagawa N., Pan W., Garry D.J.. **Dpath software reveals hierarchical haemato-endothelial lineages of Etv2 progenitors based on single-cell transcriptome analysis**. *Nature Communications* (2017) **8** 14362
34. Elyanow R., Dumitrascu B., Engelhardt B.E., Raphael B.J.. **netNMF-sc: leveraging gene-gene interactions for imputation and dimensionality reduction in single-cell expression analysis**. *Genome Research* (2020) **30** 195-204. PMID: 31992614
35. Ho Y.-J., Anaparthy N., Molik D., Mathew G., Aicher T., Patel A., Hicks J., Hammell M.G.. **Single-cell RNA-seq analysis identifies markers of resistance to targeted BRAF inhibitors in melanoma cell populations**. *Genome Research* (2018) **28** 1353-1363. PMID: 30061114
36. Kotliar D., Veres A., Nagy M.A., Tabrizi S., Hodis E., Melton D.A., Sabeti P.C.. **Identifying gene expression programs of cell-type identity and cellular activity with single-cell RNA-seq**. *eLife* (2019) **8** 43803
37. Lee D.D., Seung H.S.. **Learning the parts of objects by non-negative matrix factorization**. *Nature* (1999) **401** 788-791. PMID: 10548103
38. Levitin H.M., Yuan J., Cheng Y.L., Ruiz F.J., Bush E.C., Bruce J.N., Canoll P., lavarone A., Lasorella A., Blei D.M., Sims P.A.. **De novo gene signature identification from single-cell RNA-seq with hierarchical poisson factorization**. *Molecular Systems Biology* (2019) **15** 8557
39. Welch J.D., Kozareva V., Ferreira A., Vanderburg C., Martin C., Macosko E.Z.. **Single-cell multi-omic integration compares and contrasts features of brain cell identity**. *Cell* (2019) **177** 1873-188717. PMID: 31178122
40. Shao C., Höfer T.. **Robust classification of single-cell transcriptome data by nonnegative matrix factorization**. *Bioinformatics* (2016) **33** 235-242. PMID: 27663498
41. Sun S., Chen Y., Liu Y., Shang X.. **A fast and efficient count-based matrix factorization method for detecting cell types from single-cell RNAseq data**. *BMC Systems Biology* (2019) **13** 28. PMID: 30953530
42. Venkatasubramanian M., Chetal K., Schnell D.J., Atluri G., Salomonis N.. **Resolving single-cell heterogeneity from hundreds of thousands of cells through sequential hybrid clustering and NMF**. *Bioinformatics* (2020) **36** 3773-3780. PMID: 32207533
43. Zhang S., Yang L., Yang J., Lin Z., Ng M.K.. **Dimensionality reduction for single cell RNA sequencing data using constrained robust non-negative matrix factorization**. *NAR Genomics and Bioinformatics* (2020) **2**
44. Pierson E., Yau C.. **ZIFA: Dimensionality reduction for zero-inflated single-cell gene expression analysis**. *Genome Biology* (2015) **16** 241. PMID: 26527291
45. Blei D.M., Ng A.Y., Jordan M.I.. **Latent Dirichlet allocation**. *Journal of Machine Learning Research* (2003) **3** 993-1022
46. DuVerle D.A., Yotsukura S., Nomura S., Aburatani H., Tsuda K. **CellTree: an R/bioconductor package to infer the hierarchical structure of cell populations from single-cell RNA-seq data**. *BMC Bioinformatics* (2016) **17** 363. PMID: 27620863
47. Dey K.K., Hsiao C.J., Stephens M.. **Visualizing the structure of RNA-seq expression data using grade of membership models**. *PLoS Genetics* (2017) **13** 1006599
48. Carbonetto P., Sarkar A., Wang Z., Stephens M.. **Non-negative matrix factorization algorithms greatly improve topic model fits**. *arXiv 2105.13440* (2021)
49. González-Blas C., Minnoye L., Papasokrati D., Aibar S., Hulselmans G., Christiaens V., Davie K., Wouters J., Aerts S.. **cisTopic: cis-regulatory topic modeling on single-cell ATAC-seq data**. *Nature Methods* (2019) **16** 397-400. PMID: 30962623
50. Hofmann T.. **Probabilistic latent semantic indexing**. *Proceedings of the 22nd Annual International ACM SIGIR Conference* (1999) 50-57
51. Bielecki P., Riesenfeld S.J., Hütter J.-C., Torlai Triglia E., Kowalczyk M.S., Ricardo-Gonzalez R.R., Lian M., Amezcua Vesely M.C., Kroehling L., Xu H., Slyper M., Muus C., Ludwig L.S., Christian E., Tao L., Kedaigle A.J., Steach H.R., York A.G., Skadow M.H., Yaghoubi P., Dionne D., Jarret A., McGee H.M., Porter C.B.M., Licona-Limón P., Bailis W., Jackson R., Gagliani N., Gasteiger G., Locksley R.M., Regev A., Flavell R.A.. **Skin-resident innate lymphoid cells converge on a pathogenic effector state**. *Nature* (2021) **592** 128-132. PMID: 33536623
52. Housman G., Briscoe E., Gilad Y.. **Evolutionary insights into primate skeletal gene regulation using a comparative cell culture model**. *PLoS Genetics* (2022) **18** 1010073
53. Hung A., Housman G., Briscoe E.A., Cuevas C., Gilad Y.. **Characterizing gene expression in an in vitro biomechanical strain model of joint health**. *F1000Research* (2022) **11** 296
54. Rhodes K., Barr K.A., Popp J.M., Strober B.J., Battle A., Gilad Y.. **Human embryoid bodies as a novel system for genomic studies of functionally diverse cell types**. *eLife* (2022) **11** 71361
55. Schenkel J.M., Herbst R.H., Canner D., Li A., Hillman M., Shanahan S.-L., Gibbons G., Smith O.C., Kim J.Y., Westcott P., Hwang W.L., Freed-Pastor W.A., Eng G., Cuoco M.S., Rogers P., Park J.K., Burger M.L., Rozenblatt-Rosen O., Cong L., Pauken K.E., Regev A., Jacks T.. **Conventional type I dendritic cells maintain a reservoir of proliferative tumor-antigen specific TCF-1+ CD8+ T cells in tumor-draining lymph nodes**. *Immunity* (2021) **54** 2338-23536. PMID: 34534439
56. Xu H., Ding J., Porter C.B.M., Wallrapp A., Tabaka M., Ma S., Fu S., Guo X., Riesenfeld S.J., Su C., Dionne D., Nguyen L.T., Lefkovith A., Ashenberg O., Burkett P.R., Shi H.N., Rozenblatt-Rosen O., Graham D.B., Kuchroo V.K., Regev A., Xavier R.J.. **Transcriptional atlas of intestinal immune cells reveals that neuropeptide**. *Immunity* (2019) **51** 696-708. PMID: 31618654
57. Ding C., Li T., Peng W.. **On the equivalence between non-negative matrix factorization and probabilistic latent semantic indexing**. *Computational Statistics and Data Analysis* (2008) **52** 3913-3927
58. Gaussier E., Goutte C.. **Relation between PLSA and NMF and implications**. *Proceedings of the 28th Annual International ACM SIGIR Conference* (2005) 601-602
59. Gillis N.. *Society for Industrial and Applied Mathematics* (2021)
60. Kim J., Park H.. *Technical report* (2008)
61. Soneson C., Robinson M.D.. **Bias, robustness and scalability in single-cell differential expression analysis**. *Nature Methods* (2018) **15** 255-261. PMID: 29481549
62. Wang T., Li B., Nelson C.E., Nabavi S.. **Comparative analysis of differential gene expression analysis tools for single-cell RNA sequencing data**. *BMC Bioinformatics* (2019) **20** 40. PMID: 30658573
63. Erosheva E., Fienberg S., Lafferty J.. **Mixed-membership models of scientific publications**. *Proceedings of the National Academy of Sciences* (2004) **101** 5220-5227
64. Abdelaal T., Michielsen L., Cats D., Hoogduin D., Mei H., Reinders M.J.T., Mahfouz A.. **A comparison of automatic cell identification methods for single-cell RNA sequencing data**. *Genome Biology* (2019) **20** 194. PMID: 31500660
65. Diaz-Mejia J.J., Meng E.C., Pico A.R., MacParland S.A., Ketela T., Pugh T.J., Bader G.D., Morris J.H.. **Evaluation of methods to assign cell type labels to cell clusters from single-cell RNA-sequencing data**. *F1000Research* (2019) **8** 296
66. Blei D.M., Lafferty J.D., Srivastava A.N., Sahami M.. *Text Mining: Classification, Clustering, and Applications* (2009) 71-94
67. Griffiths T.L., Steyvers M.. **Finding scientific topics**. *Proceedings of the National Academy of Sciences* (2004) **101** 5228-5235
68. Hofmann T.. **Unsupervised learning by probabilistic latent semantic analysis**. *Machine Learning* (2001) **42** 177-196
69. Sarkar A., Stephens M.. **Separating measurement and expression models clarifies confusion in single-cell RNA sequencing analysis**. *Nature Genetics* (2021) **53** 770-777. PMID: 34031584
70. Rosenberg N.A.. **Genetic structure of human populations**. *Science* (2002) **298** 2381-2385. PMID: 12493913
71. Rosenberg N.A.. **distruct: a program for the graphical display of population structure**. *Molecular Ecology Notes* (2004) **4** 137-138
72. Haak W., Lazaridis I., Patterson N., Rohland N., Mallick S., Llamas B., Brandt G., Nordenfelt S., Harney E., Stewardson K., Fu Q., Mittnik A., Bánffy E., Economou C., Francken M., Friederich S., Pena R.G., Hallgren F., Khartanovich V., Khokhlov A., Kunst M., Kuznetsov P., Meller H., Mochalov O., Moiseyev V., Nicklisch N., Pichler S.L., Risch R., Rojo Guerra M.A., Roth C., Szécsényi-Nagy A., Wahl J., Meyer M., Krause J., Brown D., Anthony D., Cooper A., Alt K.W., Reich D.. **Massive migration from the steppe was a source for Indo-European languages in europe**. *Nature* (2015) **522** 207-211. PMID: 25731166
73. Pereira B.I., De Maeyer R.P.H., Covre L.P., Nehar-Belaid D., Lanna A., Ward S., Marches R., Chambers E.S., Gomes D.C.O., Riddell N.E., Maini M.K., Teixeira V.H., Janes S.M., Gilroy D.W., Larbi A., Mabbott N.A., Ucar D., Kuchel G.A., Henson S.M., Strid J., Lee J.H., Banchereau J., Akbar A.N.. **Sestrins induce natural killer function in senescent-like CD8+ T cells**. *Nature Immunology* (2020) **21** 684-694. PMID: 32231301
74. Ashuach T., Reidenbach D.A., Gayoso A., Yosef N.. **PeakVI: a deep generative model for single-cell chromatin accessibility analysis**. *Cell Reports Methods* (2022) **2** 100182. PMID: 35475224
75. Marioni J.C., Mason C.E., Mane S.M., Stephens M., Gilad Y.. **RNA-seq: an assessment of technical reproducibility and comparison with gene expression arrays**. *Genome Research* (2008) **18** 1509-1517. PMID: 18550803
76. Robinson M.D., Smyth G.K.. **Moderated statistical tests for assessing differences in tag abundance**. *Bioinformatics* (2007) **23** 2881-2887. PMID: 17881408
77. Robinson M.D., McCarthy D.J., Smyth G.K.. **edgeR: a Bioconductor package for differential expression analysis of digital gene expression data**. *Bioinformatics* (2009) **26** 139-140. PMID: 19910308
78. Anders S., Huber W.. **Differential expression analysis for sequence count data**. *Genome Biology* (2010) **11** 106. PMID: 20236492
79. Love M.I., Huber W., Anders S.. **Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2**. *Genome Biology* (2014) **15** 550. PMID: 25516281
80. Cable D.M., Murray E., Shanmugam V., Zhang S., Zou L.S., Diao M., Chen H., Macosko E.Z., Irizarry R.A., Chen F.. **Cell type-specific inference of differential expression in spatial transcriptomics**. *Nature Methods* (2022) **19** 1076-1087. PMID: 36050488
81. Stephens M.. **False discovery rates: a new deal**. *Biostatistics* (2016) **18** 275-294
82. Becker-Herman S., Lantner F., Shachar I.. **Id 2 negatively regulates B cell differentiation in the spleen**. *Journal of Immunology* (2002) **168** 5507-5513
83. Zhu A., Ibrahim J.G., Love M.I.. **Heavy-tailed prior distributions for sequence count data: removing the noise and preserving large differences**. *Bioinformatics* (2019) **35** 2084-2092. PMID: 30395178
84. Finak G., McDavid A., Yajima M., Deng J., Gersuk V., Shalek A.K., Slichter C.K., Miller H.W., McElrath M.J., Prlic M., Linsley P.S., Gottardo R.. **MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data**. *Genome Biology* (2015) **16** 278. PMID: 26653891
85. Squair J.W., Gautier M., Kathe C., Anderson M.A., James N.D., Hutson T.H., Hudelle R., Qaiser T., Matson K.J.E., Barraud Q., Levine A.J., La Manno G., Skinnider M.A., Courtine G.. **Confronting false discoveries in single-cell differential expression**. *Nature Communications* (2021) **12** 5692
86. Montoro D.T., Haber A.L., Biton M., Vinarsky V., Lin B., Birket S.E., Yuan F., Chen S., Leung H.M., Villoria J., Rogel N., Burgin G., Tsankov A.M., Waghray A., Slyper M., Waldman J., Nguyen L., Dionne D., Rozenblatt-Rosen O., Tata P.R., Mou H., Shivaraju M., Bihler H., Mense M., Tearney G.J., Rowe S.M., Engelhardt J.F., Regev A., Rajagopal J.. **A revised airway epithelial hierarchy includes CFTR-expressing ionocytes**. *Nature* (2018) **560** 319-324. PMID: 30069044
87. Rosvall M., Bergstrom C.T.. **Maps of random walks on complex networks reveal community structure**. *Proceedings of the National Academy of Sciences* (2008) **105** 1118-1123
88. Coifman R.R., Lafon S., Lee A.B., Maggioni M., Nadler B., Warner F., Zucker S.W.. **Geometric diffusions as a tool for harmonic analysis and structure definition of data: diffusion maps**. *Proceedings of the National Academy of Sciences* (2005) **102** 7426-7431
89. Ruiz García S., Deprez M., Lebrigand K., Cavard A., Paquet A., Arguel M.-J., Magnone V., Truchi M., Caballero I., Leroy S., Marquette C.-H., Marcet B., Barbry P., Zaragosi L.-E.. **Novel dynamics of human mucociliary differentiation revealed by single-cell RNA sequencing of nasal epithelial cultures**. *Development* (2019) **146**
90. Barkauskas C.E., Chung M.-I., Fioret B., Gao X., Katsura H., Hogan B.L.M.. **Lung organoids: current uses and future promise**. *Development* (2017) **144** 986-997. PMID: 28292845
91. Rawlins E.L., Okubo T., Xue Y., Brass D.M., Auten R.L., Hasegawa H., Wang F., Hogan B.L.M.. **The role of Scgb1a1+ clara cells in the long-term maintenance and repair of lung airway, but not elveolar, epithelium**. *Cell Stem Cell* (2009) **4** 525-534. PMID: 19497281
92. Spassky N., Meunier A.. **The development and functions of multiciliated epithelia**. *Nature Reviews Molecular Cell Biology* (2017) **18** 423-436. PMID: 28400610
93. Zhao H., Zhu L., Zhu Y., Cao J., Li S., Huang Q., Xu T., Huang X., Yan X., Zhu X.. **The cep63 paralogue deup1 enables massive de novo centriole biogenesis for vertebrate multiciliogenesis**. *Nature Cell Biology* (2013) **15** 1434-1444. PMID: 24240477
94. **The Gene Ontology resource: enriching a GOld mine**. *Nucleic Acids Research* (2020) **49** 325-334
95. Pliner H.A., Packer J.S., McFaline-Figueroa J.L., Cusanovich D.A., Daza R.M., Aghamirzaie D., Srivatsan S., Qiu X., Jackson D., Minkina A., Adey A.C., Steemers F.J., Shendure J., Trapnell C.. **Cicero predicts cis-regulatory DNA interactions from single-cell chromatin accessibility data**. *Molecular Cell* (2018) **71** 858-8718. PMID: 30078726
96. Wu C., Tao Y., Li N., Fei J., Wang Y., Wu J., Gu H.F.. **Prediction of cellular targets in diabetic kidney diseases with single-cell transcriptomic analysis of db/db mouse kidneys**. *Journal of Cell Communication and Signaling* (2022)
97. Cusanovich D.A., Hill A.J., Aghamirzaie D., Daza R.M., Pliner H.A., Berletch J.B., Filippova G.N., Huang X., Christiansen L., DeWitt W.S., Lee C., Regalado S.G., Read D.F., Steemers F.J., Disteche C.M., Trapnell C., Shendure J.. **A single-cell atlas of in vivo mammalian chromatin accessibility**. *Cell* (2018) **174** 1309-1324. PMID: 30078704
98. Der E., Ranabothu S., Suryawanshi H., Akat K.M., Clancy R., Morozov P., Kustagi M., Czuppa M., Izmirly P., Belmont H.M., Wang T., Jordan N., Bornkamp N., Nwaukoni J., Martinez J., Goilav B., Buyon J.P., Tuschl T., Putterman C.. **Single cell RNA sequencing to dissect the molecular heterogeneity in lupus nephritis**. *JCI Insight* (2017) **2** 93009. PMID: 28469080
99. Grassmeyer J., Mukherjee M., DeRiso J., Hettinger C., Bailey M., Sinha S., Visvader J.E., Zhao H., Fogarty E., Surendran K.. **Elf5 is a principal cell lineage specific transcription factor in the kidney that contributes to Aqp 2 and Avpr 2 gene expression**. *Developmental Biology* (2017) **424** 77-89. PMID: 28215940
100. Park J., Shrestha R., Qiu C., Kondo A., Huang S., Werth M., Li M., Barasch J., Suszták K.. **Single-cell transcriptomics of the mouse kidney reveals potential cellular targets of kidney disease**. *Science* (2018) **360** 758-763. PMID: 29622724
101. Ghezzi C., Loo D.D.F., Wright E.M.. **Physiology of renal glucose handling via SGLT1, SGLT2 and GLUT2**. *Diabetologia* (2018) **61** 2087-2097. PMID: 30132032
102. Thiagarajan R.D., Georgas K.M., Rumballe B.A., Lesieur E., Chiu H.S., Taylor D., Tang D.T.P., Grimmond S.M., Little M.H.. **Identification of anchor genes during kidney development defines ontological relationships, molecular subcompartments and regulatory pathways**. *PLoS ONE* (2011) **6** 17286
103. Gopal E., Umapathy N.S., Martin P.M., Ananth S., Gnana-Prakasam J.P., Becker H., Wagner C.A., Ganapathy V., Prasad P.D.. **Cloning and functional characterization of human SMCT2 (SLC5A12) and expression pattern of the transporter in kidney**. *Biochimica et Biophysica Acta—Biomembranes* (2007) **1768** 2690-2697
104. Buenrostro J.D., Corces M.R., Lareau C.A., Wu B., Schep A.N., Aryee M.J., Majeti R., Chang H.Y., Greenleaf W.J.. **Integrated single-cell analysis maps the continuous regulatory landscape of human hematopoietic differentiation**. *Cell* (2018) **173** 1535-1548. PMID: 29706549
105. Heinz S., Benner C., Spann N., Bertolino E., Lin Y.C., Laslo P., Cheng J.X., Murre C., Singh H., Glass C.K.. **Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities**. *Molecular Cell* (2010) **38** 576-589. PMID: 20513432
106. Booeshaghi A.S., Pachter L.. **Normalization of single-cell RNA-seq counts by log(x + 1) or log(1 + x)**. *Bioinformatics* (2021) **37** 2223-2224. PMID: 33676365
107. Lun A.. **Overcoming systematic errors caused by log-transformation of normalized single-cell RNA sequencing data**. *bioRxiv* (2018). DOI: 10.1101/404962
108. Warton D.I.. **Why you cannot transform your way out of trouble for small counts**. *Biometrics* (2018) **74** 362-368. PMID: 28504821
109. Amezquita R.A., Lun A.T.L., Becht E., Carey V.J., Carpp L.N., Geistlinger L., Marini F., Rue-Albrecht K., Risso D., Soneson C., Waldron L., Pagès H., Smith M.L., Huber W., Morgan M., Gottardo R., Hicks S.C.. **Orchestrating single-cell analysis with Bioconductor**. *Nature Methods* (2020) **17** 137-145. PMID: 31792435
110. Barkas N., Petukhov V., Nikolaeva D., Lozinsky Y., Demharter S., Khodosevich K., Kharchenko P.V.. **Joint analysis of heterogeneous single-cell RNA-seq dataset collections**. *Nature Methods* (2019) **16** 695-698. PMID: 31308548
111. Korsunsky I., Millard N., Fan J., Slowikowski K., Zhang F., Wei K., Baglaenko Y., Brenner M., Loh P. r., Raychaudhuri S.. **Fast, sensitive and accurate integration of single-cell data with Harmony**. *Nature Methods* (2019) **16** 1289-1296. PMID: 31740819
112. Haghverdi L., Lun A.T., Morgan M.D., Marioni J.C.. **Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors**. *Nature Biotechnology* (2018) **36** 421-427
113. Parker H.S., Leek J.T., Favorov A.V., Considine M., Xia X., Chavan S., Chung C.H., Fertig E.J.. **Preserving biological heterogeneity with a permuted surrogate variable analysis for genomics batch correction**. *Bioinformatics* (2014) **30** 2757-2763. PMID: 24907368
114. Tran H.T.N., Ang K.S., Chevrier M., Zhang X., Lee N.Y.S., Goh M., Chen J.. **A benchmark of batch-effect correction methods for single-cell RNA sequencing data**. *Genome Biology* (2020) **21** 12. PMID: 31948481
115. Richards L.M., Riverin M., Mohanraj S., Ayyadhury S., Croucher D.C., Díaz-Mejía J.J., Coutinho F.J., Dirks P.B., Pugh T.J.. **A comparison of data integration methods for single-cell RNA sequencing of cancer samples**. *bioRxiv* (2021). DOI: 10.1101/2021.08.04.453579
116. Fan J., Slowikowski K., Zhang F.. **Single-cell transcriptomics in cancer: computational challenges and opportunities**. *Experimental and Molecular Medicine* (2020) **52** 1452-1465. PMID: 32929226
117. Gouvert O., Oberlin T., Févotte C.. **Negative binomial matrix factorization for recommender systems**. *arXiv 1801.01708* (2018)
118. Gu J., Wang X., Halakivi-Clarke L., Clarke R., Xuan J.. **BADGE: A novel Bayesian model for accurate abundance quantification and differential analysis of RNA-Seq data**. *BMC Bioinformatics* (2014) **15** 6. PMID: 24410833
119. Wang C., Sun D., Huang X., Wan C., Li Z., Han Y., Qin Q., Fan J., Qiu X., Xie Y., Meyer C.A., Brown M., Tang M., Long H., Liu T., Liu X.S.. **Integrative analyses of single-cell transcriptome and regulome using MAESTRO**. *Genome Biology* (2020) **21** 198. PMID: 32767996
120. Lareau C.A., Duarte F.M., Chew J.G., Kartha V.K., Burkett Z.D., Kohlway A.S., Pokholok D., Aryee M.J., Steemers F.J., Lebofsky R., Buenrostro J.D.. **Droplet-based combinatorial indexing for massive-scale single-cell chromatin accessibility**. *Nature Biotechnology* (2019) **37** 916-924
121. Kartha V.K., Duarte F.M., Hu Y., Ma S., Chew J.G., Lareau C.A., Earl A., Burkett Z.D., Kohlway A.S., Lebofsky R., Buenrostro J.D.. **Functional inference of gene regulation using single-cell multi-omics**. *Cell Genomics* (2022) **2** 100166. PMID: 36204155
122. Bravo González-Blas C., De Winter S., Hulselmans G., Hecker N., Matetovici I., Christiaens V., Poovathingal S., Wouters J., Aibar S., Aerts S.. **SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks**. *Nature Methods* (2023) **20** 1355-1367. PMID: 37443338
123. Ma S., Zhang B., LaFave L.M., Earl A.S., Chiang Z., Hu Y., Ding J., Brack A., Kartha V.K., Tay T., Law T., Lareau C., Hsu Y.-C., Regev A., Buenrostro J.D.. **Chromatin potential identified by shared single-cell profiling of RNA and chromatin**. *Cell* (2020) **183** 1103-111620. PMID: 33098772
124. Cao J., Cusanovich D.A., Ramani V., Aghamirzaie D., Pliner H.A., Hill A.J., Daza R.M., McFaline-Figueroa J.L., Packer J.S., Christiansen L., Steemers F.J., Adey A.C., Trapnell C., Shendure J.. **Joint profiling of chromatin accessibility and gene expression in thousands of single cells**. *Science* (2018) **361** 1380-1385. PMID: 30166440
125. Stuart T., Satija R.. **Integrative single-cell analysis**. *Nature Reviews Genetics* (2019) **20** 257-272
126. Shiga M., Seno S., Onizuka M., Matsuda H.. **Sc-jnmf: single-cell clustering integrating multiple quantification methods based on joint non-negative matrix factorization**. *PeerJ* (2021) **9** 12087
127. Zhang S., Liu C.-C., Li W., Shen H., Laird P.W., Zhou X.J.. **Discovery of multi-dimensional modules by integrative analysis of cancer genomic data**. *Nucleic Acids Research* (2012) **40** 9379-9391. PMID: 22879375
128. Yang Z., Michailidis G.. **A non-negative matrix factorization method for detecting modules in heterogeneous omics multi-modal data**. *Bioinformatics* (2015) **32** 1-8. PMID: 26377073
129. Jin S., Zhang L., Nie Q.. **scAl: an unsupervised approach for the integrative analysis of parallel single-cell transcriptomic and epigenomic profiles**. *Genome Biology* (2020) **21** 25. PMID: 32014031
130. Hao Y., Stuart T., Kowalski M.H., Choudhary S., Hoffman P., Hartman A., Srivastava A., Molla G., Madad S., Fernandez-Granda C., Satija R.. **Dictionary learning for integrative, multimodal and scalable single-cell analysis**. *Nature Biotechnology* (2023). DOI: 10.1038/s41587-023-01767-y
131. Argelaguet R., Velten B., Arnol D., Dietrich S., Zenz T., Marioni J.C., Buettner F., Huber W., Stegle O.. **Multi-omics factor analysis—a framework for unsupervised integration of multi-omics data sets**. *Molecular Systems Biology* (2018) **14** 8124
132. Argelaguet R., Arnol D., Bredikhin D., Deloro Y., Velten B., Marioni J.C., Stegle O.. **MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data**. *Genome Biology* (2020) **21** 111. PMID: 32393329
133. Baker S.M., Rogerson C., Hayes A., Sharrocks A.D., Rattray M.. **Classifying cells with Scasat, a single-cell ATAC-seq analysis tool**. *Nucleic Acids Research* (2019) **47** 10
134. 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) **35** 108-116
135. Pott S., Lieb J.D.. **Single-cell ATAC-seq: strength in numbers**. *Genome Biology* (2015) **16** 172. PMID: 26294014
136. Yan F., Powell D.R., Curtis D.J., Wong N.C.. **From reads to insight: a hitchhiker’s guide to ATAC-seq data analysis**. *Genome Biology* (2020) **21** 22. PMID: 32014034
137. Taddy M.. **Distributed multinomial regression**. *Annals of Applied Statistics* (2015) **9** 1394-1414
138. Fisher R.A.. **On the interpretation of**. *Journal of the Royal Statistical Society* (1922) **85** 87-94
139. Good I.J.. **Some statistical applications of Poisson’s work**. *Statistical Science* (1986) **1** 157-170
140. Gelman A., Carlin J.B., Stern H.S., Dunson D.B., Vehtari A., Rubin D.B.. *Bayesian Data Analysis* (2013)
141. Andrieu C., de Freitas N., Doucet A., Jordan M.I.. **An introduction to MCMC for machine learning**. *Machine Learning* (2003) **50** 5-43
142. Robert C.P.. *Monte Carlo Statistical Methods* (2004)
143. Devroye L.. *Non-uniform Random Variate Generation* (1986)
144. Stephens M., Carbonetto P., Gerard D., Lu M., Sun L., Willwerscheid J., Xiao N.. **ashr: methods for adaptive shrinkage, using empirical Bayes**. *R package version 2.2–51* (2020)
145. Chen M.-H., Shao Q.-M.. **Monte Carlo estimation of Bayesian credible and HPD intervals**. *Journal of Computational and Graphical Statistics* (1999) **8** 69-92
146. Box G.E.P., Tiao G.C.. *Bayesian Inference in Statistical Analysis* (1992)
147. Gelman A., Hill J., Yajima M.. **Why we (usually) don’t have to worry about multiple comparisons**. *Journal of Research on Educational Effectiveness* (2012) **5** 189-211
148. Zheng G.X.Y., Terry J.M., Belgrader P., Ryvkin P., Bent Z.W., Wilson R., Ziraldo S.B., Wheeler T.D., McDermott G.P., Zhu J., Gregory M.T., Shuga J., Montesclaros L., Underwood J.G., Masquelier D.A., Nishimura S.Y., Schnall-Levin M., Wyatt P.W., Hindson C.M., Bharadwaj R., Wong A., Ness K.D., Beppu L.W., Deeg H.J., McFarland C., Loeb K.R., Valente W.J., Ericson N.G., Stevens E.A., Radich J.P., Mikkelsen T.S., Hindson B.J., Bielas J.H.. **Massively parallel digital transcriptional profiling of single cells**. *Data sets. 10x Genomics* (2017)
149. Montoro D.T., Haber A.L., Biton M., Vinarsky V., Chen S., Villoria J., Rogel N., Tata P.R., Rowe S.M., Engelhardt J.F., Regev A., Rajagopal J., Lin B., Birket S., Yuan F., Leung H., Burgin G., Tsankov A., Waghray A., Slyper M., Waldman J., Nguyen D. L. Dionne, Rozenblatt-Rosen O., Mou H., Shivaraju M., Bihler H., Mense M., Tearney G.. **A revised airway epithelial hierarchy includes CFTR-expressing ionocytes. Data sets**. *Gene Expression Omnibus* (2018)
150. Cusanovich D.A., Daza R., Adey A., Pliner H.A., Christiansen L., Gunderson K.L., Steemers F.J., Trapnell C., Shendure J.. **Multiplex single-cell profiling of chromatin accessibility by combinatorial cellular indexing**. *Science* (2015) **348** 910-914. PMID: 25953818
151. Cusanovich D.A., Reddington J.P., Garfield D.A., Daza R.M., Aghamirzaie D., Marco-Ferreres R., Pliner H.A., Christiansen L., Qiu X., Steemers F.J., Trapnell C., Shendure J., Furlong E.E.M.. **The cis-regulatory dynamics of embryonic development at single-cell resolution**. *Nature* (2018) **555** 538-542. PMID: 29539636
152. Cusanovich D.A., Hill A.J., Aghamirzaie D., Daza R.M., Pliner H.A., Berletch J.B., Filippova G.N., Huang X., Christiansen L., DeWitt W.S., Lee C., Regalado S.G., Read D.F., Steemers F.J., Disteche C.M., Trapnell C., Shendure J.. **A single-cell atlas of in vivo mammalian chromatin accessibility**. *Data sets. Mouse sci-ATAC-seq Atlas* (2018)
153. Buenrostro J.D., Corces M.R., Lareau C.A., Wu B., Schep A.N., Aryee M.J., Majeti R., Chang H.Y., Greenleaf W.J.. **Single-cell epigenomics maps the continuous regulatory landscape of human hematopoietic differentiation. Data sets**. *Gene Expression Omnibus* (2018)
154. Chen H., Lareau C., Andreani T., Vinyard M.E., Garcia S.P., Clement K., Andrade-Navarro M.A., Buenrostro J.D., Pinello L.. **Assessment of computational methods for the analysis of single-cell ATAC-seq data**. *Genome Biology* (2019) **20** 241. PMID: 31739806
155. Hien L.T.K., Gillis N.. **Algorithms for nonnegative matrix factorization with the Kullback-Leibler divergence**. *Journal of Scientific Computing* (2021) **87** 93
156. Hsieh C.-J., Dhillon I.S.. **Fast coordinate descent methods with variable selection for non-negative matrix factorization**. *Proceedings of the 17th ACM SIGKDD International Conference* (2011) 1064-1072
157. Lin X., Boutros P.C.. **Optimization and expansion of non-negative matrix factorization**. *BMC Bioinformatics* (2020) **21** 7. PMID: 31906867
158. Ang A.M.S., Gillis N.. **Accelerating nonnegative matrix factorization algorithms using extrapolation**. *Neural Computation* (2019) **31** 417-439. PMID: 30576616
159. Ke Z.T., Wang M.. **A new SVD approach to optimal topic estimation**. *arXiv 1704.07016* (2019)
160. van der Maaten L.. **Accelerating t-SNE using tree-based algorithms**. *Journal of Machine Learning Research* (2014) **15** 3221-3245
161. Krijthe J.H.. **Rtsne: t-distributed stochastic neighbor embedding using Barnes-Hut implementation**. *R package version 0.15* (2015)
162. Ding C., He X.. **K-means clustering via principal component analysis**. *21st International Conference on Machine Learning* (2004)
163. Kass R.E., Raftery A.E.. **Bayes factors**. *Journal of the American Statistical Association* (1995) **90** 773-795
164. Geer L.Y., Marchler-Bauer A., Geer R.C., Han L., He J., He S., Liu C., Shi W., Bryant S.H.. **The NCBI biosystems database**. *Nucleic Acids Research* (2009) **38** 492-496
165. Cerami E.G., Gross B.E., Demir E., Rodchenkov I., Babur Ö., Anwar N., Schultz N., Bader G.D., Sander C.. **Pathway Commons, a web resource for biological pathway data**. *Nucleic Acids Research* (2010) **39** 685-690
166. Rodchenkov I., Babur Ö., Luna A., Aksoy B.A., Wong J.V., Fong D., Franz M., Siper M.C., Cheung M., Wrana M., Mistry H., Mosier L., Dlin J., Wen Q., O’Callaghan C., Li W., Elder G., Smith P.T., Dallago C., Cerami E., Gross B., Dogrusoz U., Demir E., Bader G.D., Sander C.. **Pathway Commons 2019 update: integration, analysis and exploration of pathway data**. *Nucleic Acids Research* (2019) **48** 489-497
167. Subramanian A., Tamayo P., Mootha V.K., Mukherjee S., Ebert B.L., Gillette M.A., Paulovich A., Pomeroy S.L., Golub T.R., Lander E.S., Mesirov J.P.. **Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles**. *Proceedings of the National Academy of Sciences* (2005) **102** 15545-15550
168. Liberzon A., Birger C., Thorvaldsdóttir H., Ghandi M., Mesirov J.P., Tamayo P.. **The molecular signatures database hallmark gene set collection**. *Cell Systems* (2015) **1** 417-425. PMID: 26771021
169. Liberzon A., Subramanian A., Pinchback R., Thorvaldsdottir H., Tamayo P., Mesirov J.P.. **Molecular signatures database (MSigDB) 3.0**. *Bioinformatics* (2011) **27** 1739-1740. PMID: 21546393
170. Ashburner M., Ball C.A., Blake J.A., Botstein D., Butler H., Cherry J.M., Davis A.P., Dolinski K., Dwight S.S., Eppig J.T., Harris M.A., Hill D.P., Issel-Tarver L., Kasarskis A., Lewis S., Matese J.C., Richardson J.E., Ringwald M., Rubin G.M., Sherlock G.. **Gene Ontology: tool for the unification of biology**. *Nature Genetics* (2000) **25** 25-29. PMID: 10802651
171. Carbonetto P., Stephens M.. **pathways: gene set enrichment analysis using human and mouse gene sets**. *R package version 0.1–20* (2021)
172. Fang T., Davydov I., Marbach D., Zhang J.D.. **Gene-set enrichment with regularized regression**. *bioRxiv* (2019). DOI: 10.1101/659920
173. Sartor M.A., Leikauf G.D., Medvedovic M.. **LRpath: A logistic regression approach for identifying enriched biological groups in gene expression data**. *Bioinformatics* (2009) **25** 211-217. PMID: 19038984
174. Harrison P.F., Pattison A.D., Powell D.R., Beilharz T.H.. **Topconfects: a package for confident effect sizes in differential expression analysis provides a more biologically useful ranked gene list**. *Genome Biology* (2019) **20** 67. PMID: 30922379
175. McCarthy D.J., Smyth G.K.. **Testing significance relative to a fold-change threshold is a TREAT**. *Bioinformatics* (2009) **25** 765-771. PMID: 19176553
176. Wang G., Sarkar A., Carbonetto P., Stephens M.. **A simple new approach to variable selection in regression, with application to genetic fine mapping**. *Journal of the Royal Statistical Society, Series B* (2020) **82** 1273-1300
177. Bauer S., Gagneur J., Robinson P.N.. **GOing Bayesian: model-based gene set analysis of genome-scale data**. *Nucleic Acids Research* (2020) **38** 3523-3532
178. Ebrahimpoor M., Spitali P., Hettne K., Tsonaka R., Goeman J.. **Simultaneous enrichment analysis of all possible gene-sets: unifying self-contained and competitive methods**. *Briefings in Bioinformatics* (2019) **21** 1302-1312
179. Fontanillo C., Nogales-Cadenas R., Pascual-Montano A., Rivas J.D.L.. **Functional analysis beyond enrichment: non-redundant reciprocal linkage of genes and biological terms**. *PLoS ONE* (2011) **6** 24289
180. Lu Y., Rosenfeld R., Simon I., Nau G.J., Bar-Joseph Z.. **A probabilistic generative model for GO enrichment analysis**. *Nucleic Acids Research* (2008) **36** 109
181. Simillion C., Liechti R., Lischer H.E.L., loannidis V., Bruggmann R.. **Avoiding the pitfalls of gene set enrichment analysis with SetRank**. *BMC Bioinformatics* (2017) **18** 151. PMID: 28259142
182. Vivar J.C., Pemu P., McPherson R., Ghosh S.. **Redundancy control in pathway databases (ReCiPa): an application for improving gene-set enrichment analysis in omics studies and “big data” biology**. *Omics* (2013) **17** 414-422. PMID: 23758478
183. McDavid A., Finak G., Yajima M.. **MAST: model-based analysis of single cell transcriptomics**. *R package version 1.20.0* (2021)
184. Carbonetto P., Luo K., Dey K., Hsiao J., Sarkar A., Hung A., Stephens M.. **fastTopics: fast algorithms for fitting topic models and non-negative matrix factorizations to count data**. *R package version 0.6–142* (2022)
185. 185.R Core Team: R: a language and environment for statistical computing, Vienna, Austria. R Foundation for Statistical Computing (2018). https://www.R-project.org. *R: a language and environment for statistical computing* (2018)
186. Carbonetto P., Lao K., Sarkar A., Hung A., Tayeb K., Pott S., Stephens M.. **Analysis of single-cell RNA-seq data sets for this manuscript**. *Zenodo* (2023). DOI: 10.5281/zenodo.7962782
187. Carbonetto P., Lao K., Sarkar A., Hung A., Tayeb K., Pott S., Stephens M.. **Analysis of single-cell ATAC-seq data sets for this manuscript**. *Zenodo* (2023). DOI: 10.5281/zenodo.7962831
188. Blischak J.D., Carbonetto P., Stephens M.. **Creating and sharing reproducible research code the workflowr way**. *F1000Research* (2019) **8** 1749. PMID: 31723427
|
---
title: Hadza Prevotella Require Diet-derived Microbiota Accessible Carbohydrates to
Persist in Mice
authors:
- Rebecca H Gellman
- Matthew R Olm
- Nicolas Terrapon
- Fatima Enam
- Steven K Higginbottom
- Justin L Sonnenburg
- Erica D Sonnenburg
journal: bioRxiv
year: 2023
pmcid: PMC10028851
doi: 10.1101/2023.03.08.531063
license: CC BY 4.0
---
# Hadza Prevotella Require Diet-derived Microbiota Accessible Carbohydrates to Persist in Mice
## Summary
Industrialization has transformed the gut microbiota, reducing the prevalence of Prevotella relative to Bacteroides. Here, we isolate Bacteroides and Prevotella strains from the microbiota of Hadza hunter-gatherers of Tanzania, a population with high levels of Prevotella. We demonstrate that plant-derived microbiota-accessible carbohydrates (MACs) are required for persistence of Prevotella copri but not Bacteroides thetaiotaomicron in vivo. Differences in carbohydrate metabolism gene content, expression, and in vitro growth reveal that Hadza Prevotella strains specialize in degrading plant carbohydrates, while Hadza Bacteroides isolates use both plant and host-derived carbohydrates, a difference mirrored in Bacteroides from non-Hadza populations. When competing directly, P. copri requires plant-derived MACs to maintain colonization in the presence of B. thetaiotaomicron, as a no MAC diet eliminates P. copri colonization. Prevotella’s reliance on plant-derived MACs and Bacteroides’ ability to use host mucus carbohydrates could explain the reduced prevalence of Prevotella in populations consuming a low-MAC, industrialized diet.
## Introduction
The industrialized lifestyle is defined by the consumption of highly-processed foods, high rates of antibiotic administration, cesarean section births, sanitation of the living environment, and reduced contact with animals and soil–all of which can impact the human gut microbiota (Sonnenburg and Sonnenburg, 2019). Certain taxa are influenced by industrialization, i.e., are prevalent and abundant in non-industrialized populations and diminished or absent in industrialized populations, or vice versa. ( De Filippo et al., 2010; Jha et al., 2018; Merrill et al., 2022; Olm et al., 2022; Smits et al., 2017; Vangay et al., 2018; Yatsunenko et al., 2012). The microbiome of 1000–2000 year-old North American paleofeces is more similar to the modern non-industrialized than industrialized gut (Wibowo et al., 2021). The industrialized microbiota appears to be a product of both microbial extinction, as once-dominant taxa disappear, and expansion of less dominant or new taxa (Sonnenburg and Sonnenburg, 2014).
The industrialized diet differs drastically from non-industrialized diets, including in reduced amount of microbiota-accessible carbohydrates (MACs), a major metabolic input for microbes in the distal gastrointestinal tract (Cordain et al., 2005; Flint et al., 2012; Sonnenburg and Sonnenburg, 2014). Some gut-resident microbes use host mucin, which is heavily glycosylated, as a carbon source, depending on the availability of dietary MACs (Bell and Juge, 2021; Desai et al., 2016; Pudlo et al., 2022; Salyers et al., 1977; Sonnenburg et al., 2005). Shifts in dietary MACs alter microbial relative abundances and may increase inflammation and susceptibility to intestinal pathogens (Desai et al., 2016; Earle et al., 2015; Martens et al., 2018). Taxa are lost due to a lack of dietary MACs over generations in a mouse model, (Sonnenburg et al., 2016) and in humans as they immigrate to the U.S. (Vangay et al., 2018).
As human populations adopt an industrialized lifestyle, the prevalence of Prevotella decreases and that of Bacteroides increases (De Filippo et al., 2010; Jha et al., 2018; Kaplan et al., 2019). While Bacteroides are well-studied, Prevotella species remain understudied with few tools available for mechanistic investigation (Abdill et al., 2022; Accetto and Avguštin, 2015; Li et al., 2021; Xu et al., 2003). *Both* genera harbor well-documented carbohydrate utilization capabilities, encoded in carbohydrate active enzymes (CAZymes), often organized into polysaccharide utilization loci (PULs) (Bjursell et al., 2006; Dodd et al., 2010; Fehlner-Peach et al., 2019). Characterization of intestinal Prevotella species have been limited by challenges with colonization, particularly mono-colonization of germ-free mice. Here we overcome these barriers to establish a causal link between diet and P. copri abundance in a gnotobiotic mouse model.
The decreased prevalence of Prevotella in industrial populations is likely linked to a decline in relative abundance within individual microbiomes (Sprockett et al., 2020). Decreased abundance of bacterial taxa in individuals reduces the likelihood of transmission from mother to infant (Olm et al., 2022; Sonnenburg and Sonnenburg, 2019). When compounded over generations, decreased abundance can result in a population-level decline in prevalence and eventually taxa loss or extinction (Vangay et al., 2018; Sonnenburg et al., 2016). The factors driving the decline in Prevotella and the increase in Bacteroides during industrialization remain to be defined. The abundance and prevalence of specific strains of P. copri, the dominant Prevotella species in the human gut, vary among populations based on host lifestyle, particularly diet (De Filippis et al., 2019; Tett et al., 2019). Here we use gnotobiotic mice to investigate the role of diet in sustaining Prevotella and Bacteroides colonization; we demonstrate that dietary MACs play a key role in controlling the abundances of Bacteroides and Prevotella.
## Bacteroides and Prevotella genomes from the Hadza microbiota vary in prevalence across populations
To compare Prevotella and Bacteroides from non-industrialized lifestyle populations, we isolated and sequenced 6 Bacteroides strains (from 4 species) and 7 Prevotella copri strains from stool samples collected from 13 Hadza individuals. Importantly, P. copri is a species known to encompass extensive genomic diversity and its division into multiple species has been discussed (Tett et al., 2019). Single isolate genomes were assembled using both MiSeq generated short reads (146bp) and nanopore generated long reads (10–100kb) (Table 1).
To determine relatedness of the Hadza genomes to previously sequenced genomes, we calculated the average nucleotide identity (ANI) distance to the closest genome present in the National Center for Biotechnology Information (NCBI) GenBank. The Hadza Prevotella genomes are statistically distinct from P. copri genomes classified as complete in NCBI ($$p \leq 0.0002$$; Wilcoxon rank-sums test). The Hadza Bacteroides genomes, however, are not statistically distinct from existing complete Bacteroides genomes in NCBI ($$p \leq 0.76$$; Wilcoxon rank-sums test). The difference in distinctness of the Hadza Prevotella and Bacteroides genomes is unlikely due to an underrepresentation of Prevotella genomes as each genus has a similar number of published genomes (3482 and 4199, Prevotella and Bacteroides, respectively). It is also worth noting that the vast majority of human gut microbiota genome sequences were collected from North American and European samples (Abdill et al., 2022).
A phylogenetic comparison of the sequenced Hadza strains to representative genomes of the same species reveals that the Hadza Prevotella and Bacteroides strains cluster with, but are distinct from, type strains and other respective species (Fig. 1A, Fig. S1). All 7 Hadza P. copri strains belong to Clade A of the 4 proposed P. copri subgroups possessing >$10\%$ inter-clade genetic divergence (Tett et al., 2019).
To understand the prevalence of these genomes across human populations, we compared Prevotella and Bacteroides prevalence among Hadza adults and infants, four populations from Nepal living on a lifestyle gradient including foraging (Chepang), recent agriculturalist (Raute, Raj), longer term agriculturalist (Tharu), and industrial lifestyle populations (California) (Fig. 1B). We chose these groups due to their varied lifestyles and the exceptional metagenomic sequencing depth achieved, averaging 23 Gbp per sample (Jha et al., 2018; Merrill et al., 2022). Of the populations analyzed, the prevalence of Hadza Prevotella and Bacteroides isolate genomes are most similar to another foraging group, the Chepang, and most distinct from industrial lifestyle individuals (California). Prevotella genomes are rare or absent from the industrialized populations, while more prevalent and abundant in the Hadza and agriculturist samples. Conversely, nearly all Bacteroides genomes, including those isolated from the Hadza, are more prevalent in industrialized populations. The clear lifestyle shift associated with Bacteroides and Prevotella prevalence leads to the question of what aspects of the industrial lifestyle have driven these changes.
## Dietary MACs are necessary for P. copri persistence
While many factors differentiate the industrial and non-industrial lifestyles, diet serves as the top candidate for driving microbiome alterations (Sonnenburg and Sonnenburg, 2014). The Hadza diet is rich in dietary MACs from foraged tubers, berries, and baobab (Marlowe and Berbesque, 2009). In contrast, the industrialized diet is typified by high caloric intake and foods rich in fat and low in MACs (Monteiro et al., 2013). We wondered whether diet alone could impact the ability of Hadza Bacteroides and Prevotella to colonize mice. Germ-free (GF) mice were colonized with either Hadza B. thetaiotaomicron (Bt) H-2622, or Hadza P. copri (Pc) H-2477. Mice were maintained on a high MAC diet for 7 days and then switched to either a diet devoid of MACs (no MAC), a high-fat/low-MAC diet (Western), or maintained on the high MAC diet for 7 days (Fig. 2A). Bt H-2622 colonization density (109 CFU/ml in feces) on the high MAC diet was maintained in all three diet conditions (Fig. 2B). Pc H-2477 colonized to a lower degree on the high MAC diet (107 CFU/ml) and declined drastically following the change to the Western or no MAC diet, with no fecal CFUs detectable 7 days post diet switch (Fig. 2C). The lack of detectable Pc H-2477 in the absence of dietary MACs was particularly striking given the absence of competition from other microbes in this mono-associated state. To our knowledge this is the first example of a strain’s apparent eradication in a mono-associated state due to a diet change. Two other P. copri strains (Hadza Pc H-2497 and a non-Hadza strain isolated from an individual of African origin Pc N-01) are also lost in vivo in the absence of dietary MACs (Fig. S2 A, B), indicating that survival of P. copri in vivo depends on the presence of dietary MACs.
To better understand the strategies used by Hadza Pc and Bt to persist in vivo, we analyzed transcriptional profiling data from cecal contents of mice monocolonized with either Pc H-2477 or Bt H-2622 fed a high MAC diet relative to in vitro growth in peptone yeast glucose broth (PYG). Bt H-2622 and Pc H-2477 upregulate many genes in vivo under high MAC diet conditions (Fig. S2C, D). Despite the fact that $18\%$ and $13\%$ of genes in Bt H-2622 and Pc H-2477, respectively, encode for predicted carbohydrate utilization proteins, $86\%$ (in Bt H-2622) and $65\%$ (in Pc H-2477) of those upregulated in vivo encode for carbohydrate utilization ($p \leq 3.7$e-12 for Bt, $p \leq 4.5$e-13 for Pc, Fisher’s Exact test), indicating that carbohydrate utilization is the major metabolic function of these organisms in vivo (Fig. 2D).
A comparison of glycosidic linkage-breaking CAZymes, glycoside hydrolases (GH) and polysaccharide lyases (PL), reveals that Bt H-2622 upregulates a higher proportion of GHs and PLs devoted to animal-derived carbohydrate utilization relative to Pc H-2477 (Fig. 2E). Specifically, in vivo under high MAC diet conditions Bt H-2622 upregulates 8 of 22 encoded mucus targeted GHs ($\frac{3}{10}$ GH18; $\frac{5}{12}$ GH20) whereas Pc H-2477 encodes no GH18s and only one GH20, which is not upregulated in the high MAC diet condition. In addition to targeting mucus carbohydrates, Bt H-2622 also upregulates 40 of its 97 plant-targeting GHs and PLs whereas Pc H-2477 upregulates all 38 of its plant-targeting GHs and PLs in the high MAC diet (Fig. 2E). On the no MAC diet, Bt H-2622 upregulates 2 additional GH20s (along with the other mucin CAZymes upregulated on the high MAC diet) as well as 27 plant-targeting GH and PLs relative to the in vitro condition (Fig. S2E). When comparing the high MAC and no MAC in vivo conditions, Bt H-2622 upregulates only 3 GHs, 2 of which degrade mucin (GH18) (Fig. 2F) (Sonnenburg et al., 2005). These data indicate that in the absence of diet derived MACs, Hadza Bt H-2622 relies on mucus carbohydrates and that limited mucin degrading capabilities render Pc H-2477 incapable of sustaining colonization in the absence of dietary MACs.
## Carbohydrate degradation capacity differs between Hadza Bacteroides and Prevotella mirrors industrialized strains
Hadza Pc and Bacteroides isolates have a similar number and predicted function of GHs and PLs to reference strains of the corresponding species (Table 2). Unsupervised clustering of GHs and PLs reveals that the Hadza strains cluster with their type strain counterparts (Fig. 3A). When comparing the total number of GHs and PLs encoded within the Hadza strains to non Hadza strains, we found similar total numbers of these genes and distribution of substrate specificity between strains of the same species (Fig. 3B).
While Hadza Bacteroides and Prevotella strains mirror the carbohydrate degrading capacity of their non Hadza counterparts, large differences exist between the Bacteroides and Prevotella strains. The Bacteroides encode more GHs and PLs than Prevotella strains even when corrected for genome size ($\frac{251}{21}$ average GH/PL in Bacteroides; $\frac{101}{5}$ in Prevotella; Welch Two Sample t-test $$p \leq 0.00561$$) (Table 2, Fig. 3B). The proportion of Bacteroides GHs and PLs that are predicted to target plant carbohydrates or animal carbohydrates are equivalent (average $34\%$ and $37\%$, respectively) whereas the Prevotella-encoded carbohydrate degradation is biased toward plant over animal carbohydrates (average $44\%$ and $19\%$, respectively) (Fig. 3C). The Bacteroides also encode a greater breadth of GH and PL families (averaging 68 CAZyme families per genome) while Pc isolates average 40 CAZy families per genome (Fig. S3A), consistent with previously reported distributions for industrial lifestyle derived Bacteroides and Prevotella strains (Fehlner-Peach et al., 2019). The two genera also differ in their predicted mucin-degradation capacity. CAZyme families GH18 and GH20 target carbohydrates found within the intestinal mucus lining (Luis et al., 2021). All Hadza Bacteroides isolates harbor 11–14 GH20 and 1–13 GH18 CAZymes, however the Hadza Prevotella isolates contain only 1 or 2 GH20s and only one isolate, Pc H-2497, contains a single GH18 (Fig. 3D, S3B).
The CAZyme content of Hadza Bacteroides and Prevotella isolates are similar to their non-Hadza counterparts. Hadza Bacteroides isolates contain both more GHs and PLs overall as well as broader substrate degrading capabilities that include both plant and animal derived carbohydrates relative to the Hadza Prevotella isolates. This difference between the Hadza Bacteroides and Prevotella strains is similar to that seen in non-Hazda strains suggesting that the Prevotella niche is more reliant upon plant carbohydrates compared to Bacteroides (Gálvez et al., 2020).
## Dietary MACs are sufficient to maintain Pc colonization in the presence of Bt
To test whether Hadza Bacteroides and Prevotella isolates differ in their ability to use plant and mucus derived carbohydrates, we cultured Hadza and type strain Bacteroides and Pc isolates in media containing the plant carbohydrate inulin, porcine gastric mucin glycans, porcine intestinal heparin, or fructose as the sole carbon source. There is a range of ability to utilize inulin across the strains, consistent with previous work (Fig. 4A) (Sonnenburg et al., 2010). Growth in the presence of mucin, however, is divided by genera; most Bacteroides isolates grow well on mucin but the P. copri isolates do not (Fig. 4A). These data are consistent with the lack of mucin degrading capacity within the Pc genomes and the loss of Pc colonization in vivo when the host is the sole carbohydrate source.
To determine whether the lack of diet-derived MACs is responsible for the loss of Pc H-2477 colonization in vivo, we fed mice mono-colonized with Pc H-2477 a high MAC diet and then switched to either a custom diet containing $34\%$ inulin by weight as the sole fermentable carbohydrate to match MAC content of the high MAC diet (custom diets use gelatin as a binding agent and are noted by a “-g”; Inulin-g) or a no MAC diet (no MAC-g) (Fig. 4B) (Dubos and Pierce, 1948). The no MAC-g diet did not sustain Pc H-2477 colonization, with the strain becoming undetectable within one week (Fig. 4B). However, Pc H-2477 maintained colonization in the presence of the Inulin-g diet to levels similar to those observed in the high MAC diet (Fig. 2C, 4B), consistent with the requirement of dietary MACs for Pc H-2477 colonization in vivo.
We were curious how dietary MACs impact the relative abundance of Pc and Bt in mice when colonized together. GF mice were co-colonized with Pc H-2477 and Bt H-2622 and fed a high MAC diet for 7 days and then either maintained on the high MAC diet, switched to the no MAC-g diet, or the Inulin-g diet for 2 weeks, followed by a one week period in which all mice consumed the high MAC diet (Fig. 4C). Prior to the diet switch (Day 0), mice harbored both Pc H-2477 and Bt H-2622. However, 7 days after the switch to either the no MAC-g, Pc H-2477 decreased dramatically in abundance relative to Bt H-2622; a decrease of Pc also occurred in the Inulin-g diet, however the drop was not as severe as the no MAC diet indicating that inulin provided support to this strain (Fig 4D). When mice were returned to the high MAC diet, those fed the Inulin-g diet regained relative abundance of Pc H-2477 equivalent to that of baseline and to mice fed the high MAC diet throughout the experiment (Fig. 4E). In mice switched to the high MAC diet from the no MAC-g diet, Pc H-2477 colonization was detectable, but remained low after 7 days on the high MAC diet. These data are consistent with the requirement of dietary MACs for Pc colonization in the presence of Bt and may show that the variety of carbohydrates in the high MAC diet (derived from wheat, corn, oats, and alfalfa) better supports Pc colonization than a single MAC source like inulin under competition from Bacteroides. Furthermore, prolonged absence of dietary MACs restricts the ability of Pc to regain abundance when MACs are reintroduced.
## Discussion
The tradeoff between a microbiome dominated by Bacteroides or Prevotella based on host lifestyle has been well described, but its basis is not well understood (Gorvitovskaia et al., 2016; Yatsunenko et al., 2012). Here we demonstrate that Hadza isolates of Bacteroides and Prevotella do not differ dramatically from their non-Hazda counterparts in terms of genome-wide average nucleotide identity and carbohydrate utilization, suggesting that differences in their relative abundance and prevalence across lifestyle is not due to an inherent property of the population-specific strains themselves, but to differences in their environments. Furthermore, we demonstrate that dietary MACs are crucial for Prevotella to maintain colonization: even as the sole microbe, *Prevotella is* eradicated when dietary MACs are removed. Bacteroides species, however, can maintain colonization in the absence of dietary MACs due to their ability to use both plant- and host-derived carbohydrates, enabling continued colonization in low MAC industrialized diets. Our data demonstrates that in the presence of dietary MACs in gnotobiotic models, Hadza Bacteroides and Prevotella can co-exist, as is seen in the Hadza microbiome. However removal of dietary MACs results in a precipitous decline in Prevotella, which is slow to recover when MACs are reintroduced. The presence of a single MAC in the diet, inulin, was sufficient to maintain an intermediate level of colonization that then rebounded when a more complete palate of MACs was available. These data are reminiscent of the seasonal pattern of Prevotella abundance in the Hadza, which cycles in abundance with the seasonality of their diet.
All together these data are consistent with the model that prior to industrialization, human microbiomes harbored both Bacteroides and Prevotella species. As diets shift from high MAC foraged foods to low MAC industrially produced foods, abundance and prevalence of Prevotella diminished to the point of extinction in some individuals (Merrill et al., 2022). How the loss of Prevotella and increased abundance of Bacteroides within the industrialized microbiome impacts human physiology remains an important question.
## LEAD CONTACT AND MATERIALS AVAILABILITY
All information and requests for further resources should be directed to and will be fulfilled by the Lead Contact, Erica Sonnenburg, erica.sonnenburg@stanford.edu
## Data and code availability
Datasets and code for analysis are available at https://github.com/SonnenburgLab/. Raw data files for WGS are in the process of being uploaded to public databases and will be freely available upon publication of this manuscript.
## Bacterial Culture
Bacteria not isolated in this study were purchased from DSMZ (P. copri DSM 18205), or ATCC (all other reference strains). Glycerol stocks were struck out on Brain Heart Infusion plates with $10\%$ defibrinated horse blood (BHIBA) and incubated anaerobically for 24–48 h at 37°C. All growth and culturing of Bacteroides and Prevotella strains were performed anaerobically in a Coy anaerobic chamber containing $87\%$ N2, $10\%$ CO2, and $3\%$ H2.
## Mouse Husbandry
All mouse experiments were performed in accordance with the Stanford Institutional Animal Care and Use Committee. Mice were maintained on a 12-h light/dark cycle at 20.5 °C at ambient humidity, fed ad libitum, and maintained in flexible film gnotobiotic isolators for the duration of all experiments (Class Biologically Clean). Swiss-Webster mice were used for gnotobiotic experiments and the sterility of germ-free mice was verified by 16S PCR amplification and anaerobic culture of feces. Sample sizes were chosen on the basis of litter numbers and controlled for sex and age within experiments. Researchers were unblinded during sample collection (Pruss and Sonnenburg, 2021).
## Strain Isolation from Fecal Samples
Samples for strain isolation were chosen from the samples reported previously based on the 16S abundance of either Bacteroides or *Prevotella* genera (Smits et al., 2017). All isolations were performed under anaerobic conditions on YCFA-Glucose and YCFA-Baobab agar. Visible colonies from the initial plates were identified via colony PCR and re-plated onto BBE and LKV plates (Anaerobe Systems).
## Whole Genome Sequencing
Genomic DNA was extracted from single-isolate cultures grown for 24h using a MasterPure Gram Positive DNA Purification Kit. Long-read sequencing was performed using a Nanopore MinION and short read sequencing was performed using an Illumina MiSeq. Short read sequence quality was assessed using Fastqc with the command “fastqc --nogroup -q”, and adapters were trimmed with BBTools using the command “bbduk.sh -Xmx2g -eoom ref=adapters, phix threads=8 ktrim=r $k = 23$ mink=11 edist=2 entropy=0.05 tpe tbo qtrim=rl minlength=100 trimq=30 pigz=t unpigz=t samplerate=0.25.” If there was more than 100x coverage of the genome, reads were normalized using the command “bbnorm.sh target=100 min=2”. Hybrid assembly of the short and long reads was performed using SPAdes with the command “spades.py —careful --cov-cutoff auto -k 21,33,55,77,99,127” (Prjibelski et al., 2020). RagOUT was used for chromosome-level scaffolding using either the matched reference genome of the same species for Bacteroides (Table 1), or Pc H-2477 for Prevotella (Kolmogorov et al., 2018). Assembly quality was assessed with Quast (Mikheenko et al., 2018). Gene annotation was performed using RASTtk (Brettin et al., 2015).
## Clustering Genomes into Subspecies
All public Bacteroides and Prevotella genomes were downloaded from NCBI GenBank on $\frac{1}{26}$/2021 using the program ncbi-genome-download (https://github.com/kblin/ncbi-genome-download). For Bacteroides, all genomes marked as “representative genome” in RefSeq ($$n = 53$$) and genomes marked as assembly level “Complete Genome” or “Chromosome” in Genbank ($$n = 71$$) were retained for further analysis ($$n = 113$$ genomes retained of 1,229 total genomes). For Prevotella, all available public genomes were retained ($$n = 368$$). Public genomes were clustered along with the isolate genomes recovered in the study using dRep v3.2.1(Olm et al., 2017) using the command “dRep dereplicate –S_algorithm fastANI - sa 0.98 –SkipMash -nc 0.65” to ensure that genomes with ≥ $98\%$ ANI and ≥ $65\%$ alignment coverage according to FastANI (Jain et al., 2018) are considered to be the same “subspecies”. These specific thresholds were chosen manually based on histograms of reported ANI and alignment coverage values. Representative genomes were chosen using dRep’s default scoring system with the following adjustments: public genomes marked as “representative genome” in Refseq were given an additional 50 points, and genomes recovered in this study were given an additional 200 points.
## Evaluating Subspecies Prevalence and Phylogenetic Analysis
Metagenomic reads were downloaded from Merrill et. al. ( Merrill et al., 2022) (all other populations). Metagenomic reads were mapped to Prevotella and Bacteroides subspecies representative genomes using Bowtie2 (Langmead and Salzberg, 2012), and the resulting.bam files were profiled using inStrain (Olm et al., 2021). Genomes detected with ≥ $65\%$ genome breadth were considered “present” in a metagenome. The prevalence of each genome in each population was calculated as the percentage of metagenomes in which the genome was detected. Phylogenetic trees were made all for Bacteroides and Prevotella subspecies representative genomes detected in at least one metagenome using GToTree v1.5.36 with the command “GToTree -H Bacteria”. One outgroup from a different genus was included in each tree. Tree leaves were labeled based on GTDB taxonomy release 202 (Chaumeil et al., 2020), which in some cases classified genomes as belonging to other genera than they were deposited in in GenBank. Trees were visualized using iTol (Letunic and Bork, 2021).
## CAZyme Annotation
CAZyme annotations were performed for each isolate. An additional 20 strains of Prevotella copri available at NCBI, with variable assembly levels, were annotated as well for comparative purpose, with the isolates and two model strains. All amino acid sequences were first compared to the full-length sequences stored in the CAZy database (Sept. 2021) (Drula et al., 2022) using BlastP (version 2.3.0+) (Camacho et al., 2009). Queries obtaining $100\%$ coverage, >$50\%$ sequence identity and E-value ≤10−6 were automatically annotated with the same domain composition as the closest reference homolog. All remaining sequences were subject to human curation to verify the presence of each putative modules. During this process, the curator could rely on (i) bioinformatics tools, including BLAST against libraries on either full-length protein, modules only or characterized modules only, and HMMER version 3.1 (Mistry et al., 2013) against in-house built models for each CAZy (sub)family; (ii) human expertise on the appropriate coverage, sequence identity and E-value thresholds which vary across (sub)families, and ultimately on the verification of the catalytic amino-acid conservation. Hierarchical clustering of isolates’ CAZyme repertoires was performed using ComplexHeatmap (Gu, 2022). Predicted substrate assignment was compiled from previously published works (Desai et al., 2016; Smits et al., 2017).
## In Vitro Polysaccharide Growth Assays
Glycerol stocks were struck out on Brain Heart Infusion plates with $10\%$ defibrinated horse blood and incubated anaerobically for 24 h at 37°C. Isolates were passaged overnight in BHI-S (Bacteroides), and YCFA-G (Prevotella). After 16h, cultures were diluted 1:50 for Bacteroides and 1:10 for Prevotella into 200uL of culture media in a clear, flat bottomed 96-well plate. Growth media was composed of a YCFA background, plus $0.5\%$ carbohydrate, with the exception of inulin, which was added at a $1.5\%$ concentration. OD600 was measured every 15 minutes for 48h using a BioTek Epoch2 plate reader, with 30 seconds of shaking prior to each reading. Normalized OD was calculated for each carbohydrate condition by subtracting the average blank OD600 from the raw OD600 for each isolate grown in the corresponding polysaccharide. Maximum OD was calculated as the highest normalized OD in the first 24h period.
## Colonization and Enumeration of Gnotobiotic Mice
For colonization with B. thetaiotaomicron H-2622, mice were gavaged with 300uL of a 3mL culture grown for 16h in BHI-S. For colonization with P. copri, mice were gavaged with 300uL of a 3mL culture grown for 16h in YCFAC, in which was suspended 10–15 lawns (~1 per mouse) of P. copri grown on BHIBA for 48 hours. For Prevotella colonization, food removed from mouse cages and bedding changed 12h before gavage. Before the gavage of Prevotella, mice were gavaged with 300uL of $10\%$ sodium bicarbonate in water. Food was returned 2h post-gavage.
For bicolonization experiments, mice were first colonized with Pc H-2477, then gavaged with Bt H-2622 7 days later. Bicolonization was allowed to stabilize for 5–7 days before the diet switch. Feces were collected from individual mice. Two biological replicates of 1 μl feces were resuspended in 200 μl sterile PBS, serially diluted 1:10, and 2μl of each dilution was plated on BHIBA. CFUs were counted after 36h anaerobic growth at 37 °C.
## In Vivo Competition Assays
Feces were collected from individual mice. Genomic DNA was extracted from 2 biological replicates of fecal pellets using DNeasy PowerLyzer PowerSoil kit (Qiagen). Concentration of Pc and Bt DNA was assessed using species-specific qPCR primers (Key Resources Table). qPCR was performed using the Brilliant III, Ultra Fast SYBR Green QPCR Master Mix and a Bio Rad CFX thermocycler. Genomic DNA from Bt H-2622 and Pc H-2477 were used to generate a standard curve for each primer pair. The standard curves were used to calculate the absolute quantity of Bt or Pc DNA in the sample. The efficiency value (E) for each primer pair was calculated as 10(1/−slope) of log10(DNA input) against Ct value. Competitive index was calculated using this equation: E−Ct Pc primer pair/E−Ct Bt primer pair. A competitive index of 1 denotes equal abundance.
## Mouse Diets
The Inulin-g and No MAC-g diets were created using $32\%$ AIN-93G Basal Mix (CHO, Cellulose Free) and $68\%$ carbohydrates, to match the carbohydrate content of the No MAC diet (TD.150689). The Basal Mix and carbohydrate components were suspended in a mixture of water (1100ml per 250g package of Basal Mix) and $5\%$ bovine gelatin as a binder. The carbohydrates ($100\%$ glucose, no MAC-g; $50\%$ glucose and $50\%$ inulin, Inulin-g) and gelatin were dissolved separately in MilliQ water and autoclaved. The gelatin mix and AIN-93G Basal Mix (CHO, Cellulose Free) (TD.200788) were added to the carbohydrate solution in a tissue culture hood, and the mix was allowed to solidify at 4°C. Diets are listed in the Key Resources Table. After one week post-colonization, standard chow was removed and replaced with the desired test diet, and the bedding was changed. Gelatin chow was replaced every 3 days as the chow dried out.
## RNAseq
RNA was extracted from mouse cecal contents and in vitro cultures using the RNeasy PowerMicrobiome Kit (Qiagen). Ribosomal RNA depletion was performed using the RiboMinus™ Transcriptome Isolation Kit (Invitrogen). A cDNA library was constructed using the TruSeq® Stranded Total RNA Library Prep Human/Mouse/Rat kit. Sequencing was performed on a NovaSeq SP flow cell.
Quality of raw reads was assessed with Multiqc using the command “multiqc”(Ewels et al., 2016).Adapters were trimmed using Trimmomatic and the command “trimmomatic PE ILLUMINACLIP - PE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36” (Bolger et al., 2014). Reads were aligned to the Pc H-2477 and Bt H-2622 genomes using HiSAT2 commands “hisat2-build” to generate indexes, and “hisat2 -p 8 --dta -x” to align reads to the indexes (Kim et al., 2019). SAMtools was used to generate.bam files with the commands “ samtools sort -@ 8 -o” and “samtools index” (Danecek et al., 2021). Transcripts were assembled using the Stringtie commands “stringtie”, “stringtie-merge”, and “stringtie -e -B -p 11 -G” (Pertea et al., 2016). Differential expression was analyzed using DESeq2 (Love et al., 2014).
## References
1. Abdill R.J., Adamowicz E.M., Blekhman R.. **Public human microbiome data are dominated by highly developed countries**. *PLOS Biol* (2022.0) **20** e3001536. DOI: 10.1371/journal.pbio.3001536
2. Accetto T., Avguštin G.. **Polysaccharide utilization locus and CAZYme genome repertoires reveal diverse ecological adaptation of Prevotella species**. *Syst. Appl. Microbiol.* (2015.0) **38** 453-461. DOI: 10.1016/J.SYAPM.2015.07.007
3. Bell A., Juge N.. **Mucosal glycan degradation of the host by the gut microbiota**. *Glycobiology* (2021.0) **31** 691-696. DOI: 10.1093/GLYCOB/CWAA097
4. Bjursell M.K., Martens E.C., Gordon J.I.. **Functional Genomic and Metabolic Studies of the Adaptations of a Prominent Adult Human Gut Symbiont, Bacteroides thetaiotaomicron, to the Suckling Period J**. *Biol. Chem.* (2006.0) **281** 36269. DOI: 10.1074/jbc.M606509200
5. Bolger A.M., Lohse M., Usadel B.. **Trimmomatic: A flexible trimmer for Illumina sequence data**. *Bioinformatics* (2014.0) **30** 2114-2120. DOI: 10.1093/bioinformatics/btu170
6. Brettin T., Davis J.J., Disz T., Edwards R.A., Gerdes S., Olsen G.J., Olson R., Overbeek R., Parrello B., Pusch G.D., Shukla M., Thomason J.A., Stevens R., Vonstein V., Wattam A.R., Xia F.. **RASTtk: A modular and extensible implementation of the RAST algorithm for building custom annotation pipelines and annotating batches of genomes**. *Sci. Rep.* (2015.0) **5** 8365. DOI: 10.1038/srep08365
7. Camacho C., Coulouris G., Avagyan V., Ma N., Papadopoulos J., Bealer K., Madden T.L.. **BLAST+: architecture and applications**. *BMC Bioinformatics* (2009.0) **10** 421. DOI: 10.1186/1471-2105-10-421
8. Chaumeil P.-A., Mussig A.J., Hugenholtz P., Parks D.H.. **GTDB-Tk: a toolkit to classify genomes with the Genome Taxonomy Database**. *Bioinformatics* (2020.0) **36** 1925-1927. DOI: 10.1093/bioinformatics/btz848
9. Cordain L., Eaton S.B., Sebastian A., Mann N., Lindeberg S., Watkins B.A., O’Keefe J.H., Brand-Miller J.. **Origins and evolution of the Western diet: health implications for the 21st century**. *Am. J. Clin. Nutr.* (2005.0) **81** 341-354. DOI: 10.1093/ajcn.81.2.341
10. Danecek P., Bonfield J.K., Liddle J., Marshall J., Ohan V., Pollard M.O., Whitwham A., Keane T., McCarthy S.A., Davies R.M., Li H.. **Twelve years of SAMtools and BCFtools**. *GigaScience* (2021.0) **10**. DOI: 10.1093/gigascience/giab008
11. De Filippis F., Pasolli E., Tett A., Tarallo S., Naccarati A., De Angelis M., Neviani E., Cocolin L., Gobbetti M., Segata N., Ercolini D.. **Distinct Genetic and Functional Traits of Human Intestinal Prevotella copri Strains Are Associated with Different Habitual Diets**. *Cell Host Microbe* (2019.0) **25** 444-453.e3. DOI: 10.1016/j.chom.2019.01.004
12. De Filippo C., Cavalieri D., Di Paola M., Ramazzotti M., Poullet J.B., Massart S., Collini S., Pieraccini G., Lionetti P.. **Impact of diet in shaping gut microbiota revealed by a comparative study in children from Europe and rural Africa**. *Proc. Natl. Acad. Sci. U. S. A.* (2010.0) **107** 14691-6. DOI: 10.1073/pnas.1005963107
13. Desai M.S., Seekatz A.M., Koropatkin N.M., Kamada N., Hickey C.A., Wolter M., Pudlo N.A., Kitamoto S., Terrapon N., Muller A., Young V.B., Henrissat B., Wilmes P., Stappenbeck T.S., Núñez G., Martens E.C.. **A Dietary Fiber-Deprived Gut Microbiota Degrades the Colonic Mucus Barrier and Enhances Pathogen Susceptibility**. *Cell* (2016.0) **167** 1339-1353.e21. DOI: 10.1016/J.CELL.2016.10.043
14. Dodd D., Moon Y.-H., Swaminathan K., Mackie R.I., Cann I.K.O.. **Transcriptomic analyses of xylan degradation by Prevotella bryantii and insights into energy acquisition by xylanolytic bacteroidetes**. *J. Biol. Chem.* (2010.0) **285** 30261-73. DOI: 10.1074/jbc.M110.141788
15. Drula E., Garron M.-L., Dogan S., Lombard V., Henrissat B., Terrapon N.. **The carbohydrate-active enzyme database: functions and literature**. *Nucleic Acids Res* (2022.0) **50** D571-D577. DOI: 10.1093/nar/gkab1045
16. Dubos R.J., Pierce C.. **The effect of diet on experimental tuberculosis of mice**. *Am. Rev. Tuberc. Pulm. Dis.* (1948.0) **57** 287-293
17. Earle K.A., Billings G., Sigal M., Lichtman J.S., Hansson G.C., Elias J.E., Amieva M.R., Huang K.C., Sonnenburg J.L.. **Quantitative Imaging of Gut Microbiota Spatial Organization**. *Cell Host Microbe* (2015.0) **18** 478-488. DOI: 10.1016/j.chom.2015.09.002
18. 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. DOI: 10.1093/bioinformatics/btw354
19. Fehlner-Peach H., Magnabosco C., Raghavan V., Scher J.U., Tett A., Cox L.M., Gottsegen C., Watters A., Wiltshire-Gordon J.D., Segata N., Bonneau R., Littman D.R.. **Distinct Polysaccharide Utilization Profiles of Human Intestinal Prevotella copri Isolates**. *Cell Host Microbe* (2019.0) **26** 680-690.e5. DOI: 10.1016/j.chom.2019.10.013
20. Flint H.J., Scott K.P., Duncan S.H., Louis P., Forano E.. *Microbial degradation of complex carbohydrates in the gut* (2012.0) **3** 289-306. DOI: 10.4161/gmic.19897
21. Fragiadakis G.K., Smits S.A., Sonnenburg E.D., Van Treuren W., Reid G., Knight R., Manjurano A., Changalucha J., Dominguez-Bello M.G., Leach J., Sonnenburg J.L.. **Links between environment, diet, and the hunter-gatherer microbiome**. *Gut Microbes* (2019.0) **10** 216-227. DOI: 10.1080/19490976.2018.1494103
22. Gálvez E.J.C., Iljazovic A., Amend L., Lesker T.R., Renault T., Thiemann S., Hao L., Roy U., Gronow A., Charpentier E., Strowig T.. **Distinct Polysaccharide Utilization Determines Interspecies Competition between Intestinal Prevotella spp**. *Cell Host Microbe* (2020.0) **28** 838-852.e6. DOI: 10.1016/j.chom.2020.09.012
23. Gorvitovskaia A., Holmes S.P., Huse S.M.. **Interpreting Prevotella and Bacteroides as biomarkers of diet and lifestyle**. *Microbiome* (2016.0) **4** 15. DOI: 10.1186/s40168-016-0160-7
24. Green E.D., Gunter C., Biesecker L.G., Di Francesco V., Easter C.L., Feingold E.A., Felsenfeld A.L., Kaufman D.J., Ostrander E.A., Pavan W.J., Phillippy A.M., Wise A.L., Dayal J.G., Kish B.J., Mandich A., Wellington C.R., Wetterstrand K.A., Bates S.A., Leja D., Vasquez S., Gahl W.A., Graham B.J., Kastner D.L., Liu P., Rodriguez L.L., Solomon B.D., Bonham V.L., Brody L.C., Hutter C.M., Manolio T.A.. **Strategic vision for improving human health at The Forefront of Genomics**. *Nature* (2020.0) **586** 683-692. DOI: 10.1038/s41586-020-2817-4
25. Gu Z.. **Complex heatmap visualization**. *iMeta* (2022.0) **1** e43. DOI: 10.1002/imt2.43
26. Gu Z., Gu L., Eils R., Schlesner M., Brors B.. **circlize implements and enhances circular visualization in R**. *Bioinformatics* (2014.0) **30** 2811-2812. DOI: 10.1093/bioinformatics/btu393
27. Hahsler M., Hornik K., Buchta C.. **Getting Things in Order: An Introduction to the R Package seriation**. *J. Stat. Softw.* (2008.0) **25** 1-34. DOI: 10.18637/jss.v025.i03
28. Jain C., Rodriguez-R L.M., Phillippy A.M., Konstantinidis K.T., Aluru S.. **High throughput ANI analysis of 90K prokaryotic genomes reveals clear species boundaries**. *Nat. Commun.* (2018.0) **9** 5114. DOI: 10.1038/s41467-018-07641-9
29. Jha A.R., Davenport E.R., Gautam Y., Bhandari D., Tandukar S., Ng K.M., Fragiadakis G.K., Holmes S., Gautam G.P., Leach J., Sherchand J.B., Bustamante C.D., Sonnenburg J.L.. **Gut microbiome transition across a lifestyle gradient in Himalaya**. *PLOS Biol* (2018.0) **16** e2005396. DOI: 10.1371/journal.pbio.2005396
30. Kaplan R.C., Wang Z., Usyk M., Sotres-Alvarez D., Daviglus M.L., Schneiderman N., Talavera G.A., Gellman M.D., Thyagarajan B., Moon J.-Y., Vázquez-Baeza Y., McDonald D., Williams-Nguyen J.S., Wu M.C., North K.E., Shaffer J., Sollecito C.C., Qi Q., Isasi C.R., Wang T., Knight R., Burk R.D.. **Gut microbiome composition in the Hispanic Community Health Study/Study of Latinos is shaped by geographic relocation, environmental factors, and obesity**. *Genome Biol* (2019.0) **20** 219. DOI: 10.1186/s13059-019-1831-z
31. Kim D., Paggi J.M., Park C., Bennett C., Salzberg S.L.. **Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype**. *Nat. Biotechnol.* (2019.0) **37** 907-915. DOI: 10.1038/s41587-019-0201-4
32. Kolmogorov M., Armstrong J., Raney B.J., Streeter I., Dunn M., Yang F., Odom D., Flicek P., Keane T.M., Thybert D., Paten B., Pham S.. **Chromosome assembly of large and complex genomes using multiple references**. *Genome Res* (2018.0) **28** 1720-1732. DOI: 10.1101/gr.236273.118
33. Langmead B., Salzberg S.L.. **Fast gapped-read alignment with Bowtie 2**. *Nat. Methods* (2012.0) **9** 357-359. DOI: 10.1038/nmeth.1923
34. Lee M.D.. **GToTree: a user-friendly workflow for phylogenomics**. *Bioinformatics* (2019.0) **35** 4162-4164. DOI: 10.1093/bioinformatics/btz188
35. Letunic I., Bork P.. **Interactive Tree Of Life (iTOL) v5: an online tool for phylogenetic tree display and annotation**. *Nucleic Acids Res* (2021.0) **49** W293-W296. DOI: 10.1093/nar/gkab301
36. Li J., Gálvez E.J.C., Amend L., Almási É., Iljazovic A., Lesker T.R., Bielecka A.A., Schorr E.-M., Strowig T.. **A versatile genetic toolbox for Prevotella copri enables studying polysaccharide utilization systems**. *EMBO J* (2021.0) **40** e108287. DOI: 10.15252/embj.2021108287
37. Love M.I., Huber W., Anders S.. **Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2**. *Genome Biol* (2014.0) **15** 550. DOI: 10.1186/s13059-014-0550-8
38. Luis A.S., Jin C., Pereira G.V., Glowacki R.W.P., Gugel S.R., Singh S., Byrne D.P., Pudlo N.A., London J.A., Baslé A., Reihill M., Oscarson S., Eyers P.A., Czjzek M., Michel G., Barbeyron T., Yates E.A., Hansson G.C., Karlsson N.G., Cartmell A., Martens E.C.. **A single sulfatase is required to access colonic mucin by a gut bacterium**. *Nat. 2021 5987880* (2021.0) **598** 332-337. DOI: 10.1038/s41586-021-03967-5
39. Marlowe F.W., Berbesque J.C.. **Tubers as fallback foods and their impact on Hadza hunter-gatherers**. *Am. J. Phys. Anthropol.* (2009.0) **140** 751-758. DOI: 10.1002/ajpa.21040
40. Martens E.C., Neumann M., Desai M.S.. **Interactions of commensal and pathogenic microorganisms with the intestinal mucosal barrier**. *Nat. Rev. Microbiol.* (2018.0) **16** 457-470. DOI: 10.1038/s41579-018-0036-x
41. Merrill B.D., Carter M.M., Olm M.R., Dahan D., Tripathi S., Spencer S.P., Yu B., Jain S., Neff N., Jha A.R., Sonnenburg E.D., Sonnenburg J.L.. *Ultra-deep Sequencing of Hadza Hunter-Gatherers Recovers Vanishing Gut Microbes* (2022.0). DOI: 10.1101/2022.03.30.486478
42. Mikheenko A., Prjibelski A., Saveliev V., Antipov D., Gurevich A.. **Versatile genome assembly evaluation with QUAST-LG**. *Bioinformatics* (2018.0) **34** i142-i150. DOI: 10.1093/bioinformatics/bty266
43. Mistry J., Finn R.D., Eddy S.R., Bateman A., Punta M.. **Challenges in homology search: HMMER3 and convergent evolution of coiled-coil regions**. *Nucleic Acids Res* (2013.0) **41** e121. DOI: 10.1093/nar/gkt263
44. Monteiro C.A., Moubarac J.-C., Cannon G., Ng S.W., Popkin B.. **Ultra-processed products are becoming dominant in the global food system**. *Obes. Rev.* (2013.0) **14** 21-28. DOI: 10.1111/obr.12107
45. Olm M.R., Brown C.T., Brooks B., Banfield J.F.. **dRep: a tool for fast and accurate genomic comparisons that enables improved genome recovery from metagenomes through de-replication**. *ISME J* (2017.0) **11** 2864-2868. DOI: 10.1038/ismej.2017.126
46. Olm M.R., Crits-Christoph A., Bouma-Gregson K., Firek B.A., Morowitz M.J., Banfield J.F.. **inStrain profiles population microdiversity from metagenomic data and sensitively detects shared microbial strains**. *Nat. Biotechnol.* (2021.0) **39** 727-736. DOI: 10.1038/s41587-020-00797-0
47. Olm M.R., Dahan D., Carter M.M., Merrill B.D., Yu F.B., Jain S., Meng X., Tripathi S., Wastyk H., Neff N., Holmes S., Sonnenburg E.D., Jha A.R., Sonnenburg J.L.. **Robust variation in infant gut microbiome assembly across a spectrum of lifestyles**. *Science* (2022.0) **376** 1220-1223. DOI: 10.1126/science.abj2972
48. Pertea M., Kim D., Pertea G.M., Leek J.T., Salzberg S.L.. **Transcript-level expression analysis of RNA-seq experiments with HISAT, StringTie and Ballgown**. *Nat. Protoc.* (2016.0) **11** 1650-1667. DOI: 10.1038/nprot.2016.095
49. Prjibelski A., Antipov D., Meleshko D., Lapidus A., Korobeynikov A.. **Using SPAdes De Novo Assembler**. *Curr. Protoc. Bioinforma.* (2020.0) **70** e102. DOI: 10.1002/cpbi.102
50. Pruss K.M., Sonnenburg J.L.. **C. difficile exploits a host metabolite produced during toxin-mediated disease**. *Nat. 2021 5937858* (2021.0) **593** 261-265. DOI: 10.1038/s41586-021-03502-6
51. Pudlo N.A., Urs K., Crawford R., Pirani A., Atherly T., Jimenez R., Terrapon N., Henrissat B., Peterson D., Ziemer C., Snitkin E., Martens E.C.. **Phenotypic and Genomic Diversification in Complex Carbohydrate-Degrading Human Gut Bacteria**. *mSystems* (2022.0) **7**. DOI: 10.1128/MSYSTEMS.00947-21
52. Sakai R., Winand R., Verbeiren T., Moere A.V., Aerts J.. *dendsort: modular leaf ordering methods for dendrogram representations in R* (2014.0). DOI: 10.12688/f1000research.4784.1
53. Salyers A.A., West S.E., Vercellotti J.R., Wilkins T.D.. **Fermentation of mucins and plant polysaccharides by anaerobic bacteria from the human colon**. *Appl. Environ. Microbiol.* (1977.0) **34** 529-533. DOI: 10.1128/aem.34.5.529-533.1977
54. Smits S.A., Leach J., Sonnenburg E.D., Gonzalez C.G., Lichtman J.S., Reid G., Knight R., Manjurano A., Changalucha J., Elias J.E., Dominguez-Bello M.G., Sonnenburg J.L.. *Seasonal Cycling in the Gut Microbiome of the Hadza Hunter- Gatherers of Tanzania Authors: Science* (2017.0) **806** 1-18. DOI: 10.1126/science.aan4834
55. Sonnenburg E.D., Smits S.A., Tikhonov M., Higginbottom S.K., Wingreen N.S., Sonnenburg J.L.. **Diet-induced extinctions in the gut microbiota compound over generations**. *Nature* (2016.0) **529** 212-215. DOI: 10.1038/nature16504
56. Sonnenburg E.D., Sonnenburg J.L.. **Starving our microbial self: The deleterious consequences of a diet deficient in microbiota-accessible carbohydrates**. *Cell Metab* (2014.0) **20** 779-786. DOI: 10.1016/j.cmet.2014.07.003
57. Sonnenburg E.D., Zheng H., Joglekar P., Higginbottom S.K., Firbank S.J., Bolam D.N., Sonnenburg J.L.. **Specificity of Polysaccharide Use in Intestinal Bacteroides Species Determines Diet-Induced Microbiota Alterations**. *Cell* (2010.0) **141** 1241-1252. DOI: 10.1016/j.cell.2010.05.005
58. Sonnenburg J.L., Sonnenburg E.D.. **Vulnerability of the industrialized microbiota**. *Science* (2019.0) **366**. DOI: 10.1126/science.aaw9255
59. Sonnenburg J.L., Xu J., Leip D.D., Chen C.H., Westover B.P., Weatherford J., Buhler J.D., Gordon J.I.. **Glycan foraging in vivo by an intestine-adapted bacterial symbiont**. *Science* (2005.0) **307**. DOI: 10.1126/SCIENCE.1109051
60. Sprockett D.D., Martin M., Costello E.K., Burns A.R., Holmes S.P., Gurven M.D., Relman D.A.. **Microbiota assembly, structure, and dynamics among Tsimane horticulturalists of the Bolivian Amazon**. *Nat. Commun.* (2020.0) **11** 3772. DOI: 10.1038/s41467-020-17541-6
61. Tett A., Huang K.D., Asnicar F., Fehlner-Peach H., Pasolli E., Karcher N., Armanini F., Manghi P., Bonham K., Zolfo M., De Filippis F., Magnabosco C., Bonneau R., Lusingu J., Amuasi J., Reinhard K., Rattei T., Boulund F., Engstrand L., Zink A., Collado M.C., Littman D.R., Eibach D., Ercolini D., Rota-Stabelli O., Huttenhower C., Maixner F., Segata N.. **The Prevotella copri Complex Comprises Four Distinct Clades Underrepresented in Westernized Populations**. *Cell Host Microbe* (2019.0) **0**. DOI: 10.1016/j.chom.2019.08.018
62. Vangay P., Johnson A.J., Ward T.L., Kashyap P.C., Culhane-Pera K.A.. *US Immigration Westernizes the Human Gut Microbiome* (2018.0). DOI: 10.1016/j.cell.2018.10.029
63. Wibowo M.C., Yang Z., Borry M., Hübner A., Huang K.D., Tierney B.T., Zimmerman S., Barajas-Olmos F., Contreras-Cubas C., García-Ortiz H., Martínez-Hernández A., Luber J.M., Kirstahler P., Blohm T., Smiley F.E., Arnold R., Ballal S.A., Pamp S.J., Russ J., Maixner F., Rota-Stabelli O., Segata N., Reinhard K., Orozco L., Warinner C., Snow M., LeBlanc S., Kostic A.D.. **Reconstruction of ancient microbial genomes from the human gut**. *Nature* (2021.0) **594** 234-239. DOI: 10.1038/s41586-021-03532-0
64. Wickham H., Wickham H.. *Ggplot2: Elegant Graphics for Data Analysis, Use R!* (2016.0) 241-253. DOI: 10.1007/978-3-319-24277-4_12
65. Wickham H.. *An SVG Graphics Device*
66. Wickham H., Averick M., Bryan J., Chang W., McGowan L.D., François R., Grolemund G., Hayes A., Henry L., Hester J., Kuhn M., Pedersen T.L., Miller E., Bache S.M., Müller K., Ooms J., Robinson D., Seidel D.P., Spinu V., Takahashi K., Vaughan D., Wilke C., Woo K., Yutani H.. **Welcome to the Tidyverse**. *J. Open Source Softw.* (2019.0) **4** 1686. DOI: 10.21105/joss.01686
67. Xu J., Bjursell M.K., Himrod J., Deng S., Carmichael L.K., Chiang H.C., Hooper L.V., Gordon J.I.. **A Genomic View of the Human-Bacteroides thetaiotaomicron Symbiosis**. *Science* (2003.0) **299** 2074-2076. DOI: 10.1126/science.1080029
68. Yatsunenko T., Rey F.E., Manary M.J., Trehan I., Dominguez-Bello M.G., Contreras M., Magris M., Hidalgo G., Baldassano R.N., Anokhin A.P., Heath A.C., Warner B., Reeder J., Kuczynski J., Caporaso J.G., Lozupone C.A., Lauber C., Clemente J.C., Knights D., Knight R., Gordon J.I.. **Human gut microbiome viewed across age and geography**. *Nature* (2012.0) **486** 222-227. DOI: 10.1038/nature11053
|
---
title: Understanding risk factors and microbial trends implicated in the development
of Whipple-related surgical-site infections
authors:
- Wendy Feng
- Ahmer Irfan
- Molly Fleece
- Vikas Dudeja
- Sushanth Reddy
- Salila Hashmi
- J. Bart Rose
- Rachael A. Lee
journal: 'Antimicrobial Stewardship & Healthcare Epidemiology : ASHE'
year: 2023
pmcid: PMC10028940
doi: 10.1017/ash.2022.377
license: CC BY 4.0
---
# Understanding risk factors and microbial trends implicated in the development of Whipple-related surgical-site infections
## Body
The Whipple procedure, or pancreatoduodenectomy, is the only curative treatment option for patients with proximal pancreatic, biliary, or ampullary malignancies. 1–3 Although outcomes for patients undergoing pancreatoduodenectomy have significantly improved with the advancement of surgical techniques, pancreatoduodenectomy remains a high-risk procedure that carries an overall $30\%$–$60\%$ morbidity risk. 4,5 These percentages are influenced by postoperative complications—most commonly, delayed gastric emptying, pancreatic leak, and surgical site infection (SSI)—which develop in almost half of patients. 1,3,4,6 Delayed gastric emptying and pancreatic leak development can be predicted, but risk factors are largely nonmodifiable and interrelated. However, the ease and widespread availability of microbial testing offers an opportunity to study SSIs and re-evaluate our ability to prevent and manage infections following pancreatoduodenectomies.
Incidence of SSI is estimated to be $1\%$–$3\%$ of all surgical interventions but tends to be higher in abdominal surgeries. 7,8 Gastroduodenal procedures confer a unique infection risk and require special consideration when choosing antibiotic prophylaxis due to potential contamination of bowel contents onto a sterile field. This risk is amplified by patient factors such as achlorhydria, bowel perforation, morbid obesity, bleeding, and cancer. 9 Concerns for postoperative SSIs in pancreatoduodenectomies arise from manipulation of the bile duct more specifically. 10 *It is* unsurprising that cultured microbials are overwhelmingly gram-negative bacteria, followed by gram-positive cocci such as Staphylococcus spp, Streptococcus spp, and Enterococcus spp. 7,11 First- and second-generation cephalosporins have been conventionally used as the perioperative antibiotic choice for pancreatoduodenectomy. Prior studies documented statistically significant differences between experimental and placebo groups, particularly in patients with biliary stents. 11–13 However, increasing awareness and antibiotic stewardship has prompted concerns over rising local resistance patterns. 14 *In this* retrospective study, we sought to understand SSI-related pathogens and their respective antibiotic susceptibilities within our patient population and geographical region. Outcomes of this analysis may serve as a framework for optimizing patients undergoing pancreatoduodenectomies and thus improve patient morbidity while reducing hospital burden.
## Abstract
### Objective:
The purpose of this study is to understand the role of risk factors and postoperative complications seen in patients undergoing Whipple procedures in the development of surgical site infections. Our secondary goal was to evaluate whether microbial patterns differed between preoperative antibiotic classes, offering insight into the effectiveness of current practices while promoting antibiotic stewardship.
### Design:
We performed a retrospective cohort study comparing patients with and without SSIs.
### Setting:
This study was conducted at a tertiary-care center in the southeastern United States.
### Participants:
Patients who underwent a Whipple procedure between 2012 and 2021 were acquired from the National Surgical Quality Improvement Program (NSQIP) database.
### Results:
Patients with a bleeding disorder reported higher SSI rates ($$P \leq .04$$), whereas patients with a biliary stent reported lower surgical site infection (SSI) rates ($$P \leq .02$$) Those with postoperative complications had higher SSI rates, including delayed gastric emptying ($P \leq .001$) and pancreatic fistula ($P \leq .001$). Patients with longer operative times were 1.002 times more likely to develop SSIs (adjusted odds ratio [aOR], 1.002; $95\%$ confidence interval [CI], 1.001–1.004; $$P \leq .006$$) whereas surgical indications for malignancy correlated with decreased SSIs risk (aOR, 0.578; $95\%$ CI, 0.386–866) when adjusting for body mass index, surgical indication, and duration of surgical procedure.
### Conclusions:
Optimizing preoperative management of modifiable risk factors for patients undergoing pancreatoduodenectomies and decreasing operative times may reduce SSI rates and patient and hospital burden. Further research is needed to understand whether stent placement reduces SSI risk in pancreatoduodenectomy.
## Data collection
Aggregate data on patients who underwent open pancreatoduodenectomy from a single-institution was obtained from the National Surgical Quality and Improvement Program (NSQIP) database (provided by the American College of Surgeons) over a 10-year period (January 1, 2012, to December 31, 2021). Furthermore, 10 attending surgeons performed all pancreatoduodenectomies.
SSIs were defined using the Centers for Disease Control and Prevention (CDC) guidelines. 13 *The data* set specified patients who developed SSIs within 30 days of surgery. SSIs were defined as any superficial or deep wound infection, organ-space infection, or positive microbiological data from drain cultures. Microbial speciation and respective sensitivities were acquired through chart review (Supplementary Fig. 1).
Fig. 1.Surgical-site infection (SSI) and multidrug resistance (MDR) rates by year. Superimposed depiction of SSI and MDR rates over 2012–2021. Values represent percentages.
## Statistical analysis
Surgical indications for pancreatoduodenectomy were grouped into categories: cancer or malignancy, benign mass, neoplasm of unknown behavior, pancreatitis, or other. Preoperative antibiotic class was categorized as: first-generation cephalosporins, second- or third-generation cephalosporins, broad-spectrum agents, and unknown. Records with missing preoperative antibiotic information were excluded from certain analyses ($$n = 322$$). Organisms were considered multidrug resistant (MDR) if they recorded nonsusceptibility to antimicrobials in 3 or more classes. 15 Candida species were excluded from MDR analysis. Antibiograms were constructed based on sensitivity and resistance patterns gathered from chart review.
Comparison of categorical variables was performed using χ2 analysis or the Fisher exact test where appropriate. Continuous variables were tested for normality and were compared using either the Mann-Whitney U test for nonparametric results or Student t test for parametric results. P values <.05 were considered statistically significant. Risk factors were determined for SSI, and multivariable logistic regression models were performed to calculate adjusted odds ratios (ORs) for development of SSI. All statistical analysis was performed using Stata version 16.0 software (StataCorp, College Station, TX).
## Demographics
In total, 645 patients underwent pancreatoduodenectomies between 2012 and 2021 (Table 1). The mean age of patients was 63.3 years (SD, 11.4) with a balanced distribution of women ($48.8\%$) and men ($51.2\%$). Patient race was self-identified as white ($$n = 504$$, $78.1\%$), black ($$n = 121$$, $18.8\%$), Asian ($$n = 8$$, $1.2\%$), or other or unknown ($$n = 12$$, $1.9\%$). Pancreatoduodenectomy was most frequently performed for cancer ($$n = 445$$, $69\%$) followed by benign mass ($$n = 69$$, $11\%$) and other ($$n = 69$$, $11\%$). Although SSI rates following operative intervention for benign processes were higher, this difference was not statistically significant ($$P \leq .08$$).
Table 1.Demographics, Risk Factors, Operative Characteristics, and Postoperative Complications Associated With Surgical-Site Infection (SSI)VariableTotal population($$n = 645$$),No. (%) a No SSI($$n = 514$$),No. (%) a SSI($$n = 131$$),No. (%) a P ValueAge, mean y (SD)63.3 (11.4)63.4 (11.5)63.0 (11.2).55Body mass index, mean (SD)27.9 (6.5)27.7 (6.7)28.7 (6.02).11Sex, male330 [51]256 [50]74 [56].17 Race.89White504 [78]400 [78]104 [79]Black121 [19]99 [19]22 [17]Asian8 [1]6 [1]2 [2]Other/unknown12 [2]9 [2]3 [2] Comorbidities Diabetes195 [30]152 [30]43 [33].47Hypertension381 [59]301 [59]80 [61].60Immunosuppressive therapy16 [2]11 [2]5 [4].34Malnourishment101 [16]86 [18]15 [12].15Bleeding disorder17 [3]10 [2]7 [5].04Smoking within 1 year137 [21]111 [22]26 [20].66Presence of biliary stent($$n = 553$$)349 [63]($$n = 436$$)286 [66]($$n = 117$$)63 [54].02Preoperative obstructive jaundice($$n = 581$$)357 [61]($$n = 463$$)290 [63]($$n = 118$$)67 [57].24 Reason for pancreatoduodenectomy.08Cancer445 [69]367 [71]78 [59]Benign mass69 [11]51 [10]18 [14]Pancreatitis36 [6]27 [5]9 [7]Neoplasm of unspecified behavior22 [3]14 [3]8 [6]Other77 [11]55 [11]18 [14] Preoperative antibiotics ($$n = 324$$)($$n = 233$$)($$n = 91$$).43First-generation cephalosporin127 [39]94 [40]33 [36]Second or third-generation cephalosporin23 [7]14 [6]9 [10]Broad spectrum174 [54]125 [54]49 [54] Preoperative laboratory results, mean (SD) WBC7.6 (3.3)7.6 (3.0)7.8 (4.2).36HCT37.5 (5.2)37.3 (5.2)38.3 (5.2).05Platelet260.7 (94.5)261.1 (94.2)259.1 (96.1).83INR1.05 (0.19)1.05 (0.2)1.05 (0.2).96Creatinine0.89 (0.40)0.90 (0.43)0.88 (0.25).77BUN14.1 (6.3)13.8 (6.1)15.2 (7.0).03Albumin3.7 (0.57)3.7 (0.6)3.9 (0.5)<.001T bili1.8 (2.8)1.81 (2.8)1.87 (2.8).85AST/SGOT47.8 (67.5)48.9 (68.8)43.6 (61.8).44Alk phos188.8 (187.7)190.9 (185.2)180.4 (198.3).58 Preoperative treatment Chemotherapy110 [17]91 [18]19 [14].38Radiation therapy15 [2]14 [3]1 (0.8).21Surgery duration, mean minutes (SD)290.0 (123.2)283.4 (5.3)315.8 (11.4).007 Wound classification ($$n = 614$$)($$n = 491$$)($$n = 123$$)Clean1 (0.2)1 (0.2)0 [0]Clean-contaminated548 [89]438 [89]110 [89].07Contaminated55 (9.0)47 (9.6)8 (6.5)Dirty/infected10 (1.6)5 [1]5 [4] Complications Delayed gastric emptying67 (10.4)32 (6.2)35 (26.7)<.001Fistula127 (19.7)64 (13.2)63 [50]<.001Sepsis35 (0.5)9 (1.8)26 (19.8)<.001UTI8 (1.2)7 (1.4)1 (0.8).59Pneumonia16 (2.5)8 (1.6)8 (6.1).003Ventilator requirement >48 hours19 (2.9)8 (1.6)11 (8.4)<.001Intubated29 (4.5)14 (2.7)15 (11.5).002MI6 (0.9)3 (0.6)3 (2.3).09CVA3 (0.5)2 (0.4)1 (0.8).58PE2 (0.3)1 (0.2)1 (0.8).33Renal insufficiency10 (1.6)5 [1]5 (3.8).03Dialysis requirement12 (1.9)4 (0.8)8 (6.1)<.001106 (16.4)81 (15.8)25 (19.1).36Elevated postoperative amylase, median (IQR)31.5 [264]26 [117]526 [5,663]<.001Length of hospital stay, median d (IQR)($$n = 639$$)8 [5]($$n = 513$$)7 [3]($$n = 126$$)12 [13]<.001Readmissions within 30 d105 [16]56 (10.9)49 (37.4)<.001Note. SD, standard deviation; WBC, white blood cell; HCT, hematocrit; INR, international normalized ratio; BUN, blood urea nitrogen; Tbili, total bilirubin; AST/SGOT, aspartate aminotransferase; Alk phos, alkaline phosphate; UTI, urinary tract infection; MI, myocardial infarction; CVA, cardiovascular accident; PE, pulmonary embolism; IQR, interquartile range.aNo. (%) unless otherwise indicated.
## Culture data
Among 645 patients, 131 ($20.3\%$) developed SSIs, and 75 ($57.3\%$) of those contained culture data documenting speciation and/or antibiotic sensitivities. Of 131 SSI cases, 33 ($25\%$) were considered superficial incisional, 10 ($8\%$) were deep incisional, and 88 ($67\%$) were organ-space infections. Yearly SSI and MDR rates are shown in Figure 1. Highest rates of SSI occurred in years 2016 and 2017 ($18.3\%$ and $19.3\%$; $P \leq .001$). MDR rates fluctuated drastically between 2012 ($20.0\%$), 2013 ($0\%$), 2014 ($29.0\%$), 2015 ($4.8\%$), and 2016 ($16.3\%$) before stabilizing in 2017 ($10.8\%$), 2018 ($13.8\%$), 2019 ($7.7\%$), 2020 ($15.0\%$), and 2021 ($12.2\%$). Bacteria were most commonly gram-negative isolates ($42\%$) and gram-positive isolates ($39\%$) and were less commonly yeast ($16\%$) or anaerobes ($3\%$) (Fig. 2a). Independent studies from 2012–2016 and 2017–2021 showed similar bacterial distributions (Fig. 2b and 2c). Polymicrobial infections comprised $57.8\%$ of all SSIs. Select patients had their bile ducts cultured intraoperatively at the time of pancreatoduodenectomy. Among those with SSIs, 10 patients had intraoperative bile-duct cultures, of which all returned positive. Only 5 patients were identified to have concordant bacteria causing SSI as bile-duct cultures at the time of pancreatoduodenectomy; all of these patients acquired new pathogens, and 4 of 10 patients acquired MDROs (Table 2).
Fig. 2.Surgical-site infection (SSI) microbial distributions. Aggregated from (a) 2012–2021 and divided into (b) 2012–2016 and (c) 2017–2021 intervals. Note. gram−, gram negative; gram+, gram positive.
Table 2.Preoperative Biliary Stenting in Patients with Surgical-Site Infection (SSI) ($$n = 10$$)YearPreop AbxClassPreop Stent CulturesPostop SSI CulturesPostop Blood CulturesPostop Culture SummaryNo. of PathogensCulturedNo. of PathogensSensitiveNo. of Pathogens CulturedNo. of Pathogens SensitiveNo. of Pathogens CulturedNo. of Pathogens SensitiveNo. of Pathogens C/W Preop CulturesNo. of New MDR PathogensMDRPathogen2012UTO52014UTO$\frac{261}{61}$VRE2015UTO$\frac{310}{12}$/220171st-gen$\frac{30}{320}$/$\frac{22}{21}$VRE2020BS$\frac{33}{3020201}$st-gen$\frac{20}{210}$/$\frac{100}{11}$MRSA2021BS$\frac{43}{443}$/$\frac{41}{420211}$st-gen$\frac{52}{530}$/$\frac{32}{31}$ E. coli (ESBL)2021BS$\frac{32}{302021}$BS$\frac{65}{610}$/$\frac{10}{1}$Note. MDR, multidrug resistant; UTO, unable to obtain; Abx, antibiotics; Preop, preoperative; Postop, postoperative; 1st-gen, first-generation cephalosporins; BS, broad-spectrum; VRE, vancomycin-resistant Enterococcus; MRSA, methicillin-resistant Staphylococcus aureus; ESBL, extended-spectrum β-lactamase;
## Preoperative antibiotics
There were no differences between preoperative antibiotic class and rate of SSIs: first-generation cephalosporins had an SSI rate of $26.0\%$ (uOR, 0.84; $95\%$ CI, 0.51–1.39), second- or third-generation cephalosporins had an SSI rate of $39.1\%$ (uOR, 1.72; $95\%$ CI, 0.72–4.12), and broad-spectrum had an SSI rate of $28.3\%$ (uOR, 1.01; $95\%$ CI, 0.62–1.64; $$P \leq .43$$) (Table 1, Fig. 3). Additionally, the rates of polymicrobial infection did not differ between classes ($$P \leq .46$$). Microbials cultured from SSIs after administration of the different preoperative antibiotics were variable. When comparing first-generation cephalosporins, second- or third-generation cephalosporins, and broad-spectrum antibiotics, the prevalence differed between Enterococcus spp ($23\%$ vs $43\%$ vs $18\%$), Klebsiella ($18\%$ vs $0\%$ vs $21\%$), and Staphylococcus ($3\%$ vs $14\%$ vs $11\%$). However, the prevalences of Candida ($12\%$ vs $15\%$ vs $13\%$) and Esherichia coli ($14\%$ vs $14\%$ vs $12\%$) were comparable. Antibiotic sensitivities varied between culture data collected over two 5-year periods (2012–2016 and 2017–2021) as depicted in antibiograms (Supplementary Fig. 2), and a decrease in “resistant” bacteria was demonstrated over time. No significant changes were seen in similar analyses for postoperative blood cultures or Candida infections.
Fig. 3.Effect of preoperative antibiotic class. ( a) Absolute number and (b) percentages of positive surgical-site infection (SSI) cultures according to preoperative antibiotic class ($$P \leq .43$$). Polymicrobial infections incidence is also observed by (c) absolute number and (d) percentages of positive SSIs ($$P \leq .46$$).
## Associations between risk factors and SSIs
Our analyses showed no differences regarding patient demographics or comorbidities among patients with SSI compared to those without SSI, apart from presence of a bleeding disorder ($$P \leq .04$$) (Table 1). Patients with preoperative biliary stents had a lower SSI rate ($63\%$ vs $57\%$; uOR, 0.61; $95\%$ CI; 0.40–0.93; $$P \leq .02$$;).
Certain preoperative laboratory values were decreased in patients with an SSI, including BUN ($$P \leq .03$$), albumin ($P \leq .001$), and hematocrit ($$P \leq .05$$) (Table 1). However, no differences between white blood cell count ($$P \leq .36$$), creatinine ($$P \leq .77$$), total bilirubin ($$P \leq .85$$), or alkaline phosphate ($$P \leq .58$$) levels were demonstrated. Preoperative interventions, such as chemotherapy (uOR, 0.79; $95\%$ CI, 0.46–1.3; $$P \leq .38$$) or radiation therapy (uOR, 0.27; $95\%$ CI, 0.035–2.08; $$P \leq .21$$), were not significant. In our study, the overwhelming majority ($89\%$) of operative fields were considered clean contaminated. No differences in SSI rates were seen between wound classifications: clean, clean-contaminated, contaminated, or dirty-infected ($$P \leq .07$$).
In a multivariable analysis adjusting for BMI, surgical indication, and duration of surgical procedure, increased BMI was not predictive of SSI risk (aOR, 1.017; $95\%$ CI, 0.989–1.047; $$P \leq .24$$,) (Table 3). A one-minute increase in operative time lead to a $0.2\%$ increase in the odds of developing an SSI (aOR, 1.002; $95\%$ CI, 1.001–1.004; $$P \leq .006$$), whereas patients with cancer as a surgical indication had a $42\%$ decrease in the odds of developing an SSI (aOR, 0.578; $95\%$ CI, 0.386–866; $$P \leq .008$$).
Table 3.Multivariable Logistic Analysis of Predictors for Post-Whipple Surgical-Site Infections (SSIs)PredictorsOdds RatioConfidence Interval P ValueDuration of surgical procedure1.0021.001–1.004.006Body mass index1.0170.989–1.047.243Surgical indication0.5780.386–0.866.008 Patients with certain reported postoperative complications also developed SSIs. These included postoperative sepsis (uOR, 13.9; $95\%$ CI, 6.3–30.5; $P \leq .001$,), pneumonia (uOR, 4.1; $95\%$ CI,1.5–11.2; $$P \leq .003$$), ventilator requirement for >48 hours (uOR, 5.8; $95\%$ CI, 2.3–14.7; $P \leq .001$), intubation requirement (uOR, 2.6; $95\%$ CI, 1.4–4.8; $$P \leq .002$$), renal insufficiency (uOR, 4.0; $95\%$ CI, 1.2–14.2; $$P \leq .03$$), and dialysis requirement (uOR, 8.3; $95\%$ CI, 2.5–28.0; $P \leq .001$) (Table 1). Patients undergoing pancreatoduodenectomies are at increased risk for developing complications, and those reporting delayed gastric emptying also had higher rates of SSI (uOR, 5.2; $95\%$ CI, 3.1–8.9; $P \leq .001$), as did those with fistula formation (uOR, 6.6; $95\%$ CI, 4.2–10.2; $P \leq .001$). Peak amylase levels measured on postoperative days 2 and 3 were on average statistically different between those with and without SSIs (8,508.5 U/L vs 1,529.5 U/L; $P \leq .001$). Average length of stay increased by 7.6 days in patients with reported SSIs compared to those without (8.9 vs 16.5 days; $P \leq .001$). Readmission rate—defined as readmission within 30 days of hospitalization—also increased in patients with SSIs ($P \leq .001$).
## Discussion
In this retrospective study, we assessed risk factors and microbial patterns of SSIs developing after pancreatoduodenectomies at a single institution over 10 years. From data extracted from the NSQIP database, the overall SSI rate was $20.3\%$, similar to previously reported rates of $20\%$–$40\%$. 16 Stabilized SSI rates in recent years may be related to increased availability and standardization of microbial testing. Moreover, the hospital system in this study transitioned to a new electronic medical record system in 2011, so variability in provider SSI reporting and familiarity with EMR documentation over time may have also contributed to perceived increase in SSI rate over time. 17 At the start of this study in 2012, our hospital pancreatoduodenectomy protocol called for broad-spectrum as the preoperative antibiotic of choice. In late 2016, a change to first-generation cephalosporins was implemented and later switched back to broad-spectrum in early 2018. During this interval, we observed a surge in the SSI rate from $29.0\%$ in 2016 to $37.1\%$ in 2017 and subsequently down to $23\%$ in 2018. Given missing data on preoperative antibiotics, we could not correlate antimicrobial use with SSI rates. Minor discrepancies between percentages can occur because patients may steer away from protocol-assigned preoperative antibiotics due to patient-specific contraindications (ie, allergies or chronic kidney disease stage 4 or 5 for ertapenem). 18 Despite changes in SSI rate, the incidence of multidrug-resistant organisms during this interval did not change. The antibiogram format allows visualization of changes in resistance patterns over time (2012–2016 vs 2017–2021). We were able to observe decreases in antimicrobial resistance patterns between year ranges. Interestingly, Klebsiella oxytoca, Pseudomonas, E. coli, and Citrobacter spp increased in sensitivities.
Unmodifiable risk factors, such as age, sex, and race, were not linked to post-Whipple SSIs in this study. The data set did not explore socioeconomic status, access to healthcare, or degree of health literacy. These social determinants of health may contribute to patient outcomes and offer important information because our healthcare system serves rural communities in the state of Alabama. 19 Regarding modifiable risk factors, smoking is strongly correlated with delayed wound healing in the literature, and obesity has been associated with infection, specifically of skin and soft tissue. 20,21 Interestingly, we did not identify associations between increased BMI or smoking within 1 year and the development of SSIs. Additional patient comorbidities have also been cited as SSI risk factors, such as diabetes mellitus, hypertension, immunosuppression, malnourishment, and bleeding disorders. 8 Only the presence of bleeding disorder demonstrated a higher SSI rate. This finding may be attributed to standardized preoperative management of the other comorbidities studied compared to bleeding disorders. Guidelines recommending target A1C, blood pressure range, or albumin levels have been well established. Previous reports describe similar increased SSI risks in thoracic and orthopedic cases among patients with bleeding disorders; however, no documentation in abdominal surgeries has been cited. 22,23 No differences in corresponding preoperative platelet, INR, or PTT values were found between SSI and no SSI cohorts. Nonetheless, low BUN, albumin, and hematocrit were associated with increased SSIs. Collectively, these values illustrate a state of malnutrition either from hypermetabolic states associated with malignancy or decreased oral intake. This discrepancy between laboratory and clinical malnutrition may be due to error in documentation or delayed symptom recognition. Furthermore, the documentation of a comorbidity does not indicate whether patient comorbidities were well managed versus poorly managed and symptomatic, which may be better understood through more objective measures such as preoperative laboratory values. For cancer patients who underwent chemotherapy or radiation therapy within 90 days prior to surgery, no increases in SSI rate were established, reassuring providers that patients may proceed with neoadjuvant therapy to improve cancer outcomes without compromising postoperative infection risk.
Overwhelmingly, the most common indication for undergoing pancreatoduodenectomy is malignancy. Our investigation demonstrates surgical indication of cancer as a negative prognostic factor. This finding has not been previously reported and may warrant further study. Patients with pancreatic head or biliary tract masses often present with preoperative biliary obstruction and may require biliary stenting (metal, plastic, or both). Patients with biliary obstruction did exhibit higher SSI rates; unexpectedly, patients with documented preoperative biliary stents had lower SSI rates in an unadjusted analysis. Although it has been recognized that endoscopic stenting harbors an increased risk of infection by exposing the biliary tree to duodenal bacteria, we hypothesize that confounding variables may be present and are worth investigating. 24,25 Controversy remains over the choice of preoperative antibiotic therapy as the effectiveness of antibiotic classes is debated, and emerging evidence suggests that institution-specific and/or targeted therapy is more effective in lowering SSI risk than any single antibiotic choice. 26–29 Intrabdominal surgery most commonly uses early-generation cephalosporins and broad-spectrum antibiotics to cover for gram-positives and/or enterococci and coliform bacteria. Moreover, coverage varies by regional resistance patterns, though antibiotic choices may be limited by patient-specific characteristics (ie, drug reactions, renal insufficiency, or liver disease). There is also concern that preoperative antibiotics may select for bacterial growth. Although we did not detect differences between antibiotic class and SSI rate, bacterial culture and speciation demonstrated a notable discrepancy. Specifically, the use of first-, second-, and third-generation cephalosporins later developed SSIs composed primarily of Enterococcus spp and E. coli. In contrast, the use of broad-spectrum antibiotics led to Klebsiella spp–dominated cultures, followed by Enterococcus spp. The rate of polymicrobial infections did not differ according to antibiotic class, suggesting low concern for developing resistance to preoperative antibiotics over time.
We evaluated duration of surgery and wound classification as potential risk factors for developing SSI. As expected, surgery duration was predictive of increased SSI risk; therefore, systemic and team-based approaches to reducing operative length may benefit both patient outcomes and hospital efficiency. 30 The degree of intraoperative wound contamination has also been suggestive of SSIs in prior studies. 31 Wounds were categorized as clean, clean-contaminated, contaminated, or dirty-infected. Pancreatic surgeries often involve dividing the bile duct and proximal small bowel, therefore considered clean-contaminated. 32 The vast majority ($89\%$) of our pancreatoduodenectomies reported clean-contaminated wounds with a $20.1\%$ SSI rate, comparable to our overall $20.3\%$ rate. Unsurprisingly, 5 of 5 dirty or infected wounds developed an SSI.
Almost half of patients undergoing Whipple procedures develop postoperative complications, with delayed gastric emptying and pancreatic leak being the most common. 1,3,4,6 Pancreatic leaks themselves have been linked to both delayed gastric emptying and SSIs and therefore represents an important prognostic factor to study. Biochemical leaks may be approximated by drain amylase levels monitored in the postoperative course and indicate insufficient pancreatojejunostomy anastomosis. As such, leaked bile, containing autodigestive enzymes, that may seep into surrounding tissues and promote necrosis and bacterial colonization. Our study identified an increase in SSI incidence in patients reporting pancreatic leak, consistent with literature documenting pancreatic leak as an independent risk factor. 33 Those with increased extra-abdominal postoperative complications related to anesthesia/induction (pneumonia, increased ventilator requirement) or kidney function (renal insufficiency, dialysis requirements) had increased SSI rates. Confounding variables contributing to renal insufficiency include microbial infection or adverse effects of antimicrobial treatment itself. No preoperative differences in creatinine were noted; postoperative creatinine values were not documented. The SSI patient cohort was admitted for 1 week longer on average, further increasing patient morbidity risk.
As predicted, SSI cultures primarily grew gram-negative isolates (E. coli) and gram-positive isolates (Enterococcus spp), commonly known as culprits of hepatobiliary infection. 34 Although >$99\%$ of gastrointestinal flora are anaerobic, we found that anaerobes only comprised $3\%$ of cultured bacteria. 35 Therefore, anaerobes are either sufficiently covered by preoperative antibiotics or are difficult to isolate. However, yeasts (specifically Candida) make up a considerable percentage of SSIs cultured from our institution, possibly related to antibiotic-induced yeast infections within bowels and superficial skin where they are colonized. More than half of cultures were polymicrobial; polymicrobial infections are common in abdominal surgeries yet do not have inferior outcomes compared to monomicrobial infections. 36–38 Distributions of bacterial grown from year ranges were comparable, suggesting no significant microbial surge or acquired antibiotic resistance (Fig. 2b and 2c).
This study had several limitations.
Given a patient population of 645, our study utilized patient specific data reported in the NSQIP database. Missing values in this database limited our statistical analyses. Nearly half of our patients (322 of 645) lacked data on preoperative antibiotic class, which restricted our ability to perform logistic regression analyses to evaluable whether choice of preoperative antibiotics was predictive of SSI. Moreover, completion of antibiograms was hindered by unstandardized laboratory testing because bacteria were often tested against different antibacterial agents. For instance, *Pseudomonas grew* from 4 SSI cultures between 2017 and 2021: 2 were resistant to meropenem, and the other 2 were suppressed per laboratory protocol. With data missing from the other 2 cultures, we were unable to calculate a resistance percentage for the antibiogram depiction.
In conclusion, we identified surgery duration as a prognostic indicator of SSIs in patients undergoing the Whipple procedure. Additionally, we identified a strong correlation between cancer as a surgical indication and decreased SSI rates. This finding has not been reported in previous epidemiologic studies, and future analyses may be warranted. We argue that choice of antimicrobials may be tailored to specific patient needs without compromising SSI risk. Maintaining current Whipple-protocol preoperative antibiotics at our institution while making systematic efforts to decrease operative times may help improve patient outcomes after pancreatoduodenectomy.
## Financial support
No financial support was provided relevant to this article.
## Conflicts of interest
All authors report no conflicts of interest relevant to this article.
## References
1. Saraee A, Vahedian-Ardakani J, Saraee E. **Whipple procedure: a review of a 7-year clinical experience in a referral center for hepatobiliary and pancreas diseases**. *World J Surg Onc* (2015) **13** 98
2. Leichtle SW, Kaoutzanis C, Mouawad NJ. **Classic Whipple versus pylorus-preserving pancreaticoduodenectomy in the ACS NSQIP**. *J Surg Res* (2013) **183** 170-176. PMID: 23410660
3. Changazi SH, Ahmed Q, Bhatti S. **Whipple procedure: a five-year clinical experience in tertiary-care center**. *Cureus* (2020) **12** e11466. PMID: 33329963
4. Clancy TE, Ashley SW.. **Pancreaticoduodenectomy (Whipple operation)**. *Surg Oncol Clin N Am* (2005) **14** 533-552. PMID: 15978428
5. Jakhmola CK, Kumar A.. **Whipple’s pancreaticoduodenectomy: outcomes at a tertiary-care hospital**. *Med J Armed Forces India* (2014) **70** 321-326. PMID: 25382904
6. Karim SAM, Abdulla KS, Abdulkarim QH, Rahim FH.. **The outcomes and complications of pancreaticoduodenectomy (Whipple procedure): cross-sectional study**. *Int J Surg* (2018) **52** 383-387. PMID: 29438817
7. Alkaaki A, Al-Radi OO, Khoja A. **Surgical site infection following abdominal surgery: a prospective cohort study**. *Can J Surg* (2019) **62** 111-117. PMID: 30907567
8. Azoury S, Farrow N, Hu Q. **Postoperative abdominal wound infection—epidemiology, risk factors, identification, and management**. *Chron Wound Care Manage Res* (2015) **2** 137-148
9. LoCiero J, Nichols RL.. **Sepsis after gastroduodenal operations; relationship to gastric acid, motility, and endogenous microflora**. *South Med J* (1980) **73** 878-880. PMID: 7384847
10. Cortes A, Sauvanet A, Bert F. **Effect of bile contamination on immediate outcomes after pancreaticoduodenectomy for tumor**. *J Am Coll Surg* (2006) **202** 93-99. PMID: 16377502
11. Nichols RL, Webb WR, Jones JW, Smith JW, LoCicero J. **Efficacy of antibiotic prophylaxis in high-risk gastroduodenal operations**. *Am J Surg* (1982) **143** 94-98. PMID: 7032332
12. Uchiyama K, Takifuji K, Tani M, Ueno M, Kawai M, Ozawa S, Yamaue H.. **Prevention of postoperative infections by administration of antimicrobial agents immediately before surgery for patients with gastrointestinal cancers**. *Hepatogastroenterology* (2007) **54** 1487-1493. PMID: 17708282
13. Naffouje SA, Allenson K, Hodul P. **Prophylactic perioperative antibiotics in open pancreaticoduodenectomy: when less is more and when it is not. a national surgical quality improvement program propensity-matched analysis**. *J Surg Res* (2022) **279** 722-732. PMID: 35933790
14. Decker BK, Nagrebetsky A, Lipsett PA, Wiener-Kronish JP, O’Grady NP.. **Controversies in perioperative antimicrobial prophylaxis**. *Anesthesiology* (2020) **132** 586-597. PMID: 31841446
15. Magiorakos AP, Srinivasan A, Carey RB. **Multidrug-resistant, extensively drug-resistant and pandrug-resistant bacteria: an international expert proposal for interim standard definitions for acquired resistance**. *Clin Microbiol Infect* (2012) **18** 268-281. PMID: 21793988
16. Ceppa EP, Pitt HA, House MG. **Reducing surgical site infections in hepatopancreatobiliary surgery**. *HPB (Oxford)* (2013) **15** 384-391. PMID: 23557410
17. 17.
Alabama hospital links medical devices to EMR. Healthcare Innovation website. https://www.hcinnovationgroup.com/home/news/13014840/alabama-hospital-links-medical-devices-to-emr. Published June 23, 2011. Accessed July 16, 2022.. (2011)
18. Kara E, Sahin OZ, Ayaz T, Yildirim S, Sahutoglu T, Tufekci A.. **Ertapenem-induced acute reversible peripheral neuropathy in chronic kidney disease: 3 case reports**. *Clin Nephrol* (2015) **84** 353-357. PMID: 26558370
19. **Surgical site infection after gastrointestinal surgery in high-income, middle-income, and low-income countries: a prospective, international, multicentre cohort study**. *Lancet Infect Dis* (2018) **18** 516-525. PMID: 29452941
20. Nolan MB, Martin DP, Thompson R, Schroeder DR, Hanson AC, Warner DO.. **Association between smoking status, preoperative exhaled carbon monoxide levels, and postoperative surgical site infection in patients undergoing elective surgery**. *JAMA Surg* (2017) **152** 508
21. Huttunen R., Syrjänen J.. **Obesity and the risk and outcome of infection**. *Int J Obes* (2013) **37** 333-340
22. Aeschbacher P, Nguyen TL, Dorn P, Kocher GJ, Lutz JA.. **Surgical site infections are associated with higher blood loss and open access in general thoracic practice**. *Front Surg* (2021) **8** 656249. PMID: 34250005
23. Lai Q, Song Q, Guo R. **Risk factors for acute surgical site infections after lumbar surgery: a retrospective study**. *J Orthop Surg Res* (2017) **12** 116. PMID: 28724387
24. Gavazzi F, Ridolfi C, Capretti G. **Role of preoperative biliary stents, bile contamination and antibiotic prophylaxis in surgical site infections after pancreaticoduodenectomy**. *BMC Gastroenterol* (2016) **16** 43. PMID: 27036376
25. Lamarca A, Rigby C, McNamara MG, Hubner RA, Valle JW.. **Impact of biliary stent-related events in patients diagnosed with advanced pancreatobiliary tumors receiving palliative chemotherapy**. *World J Gastroenterol* (2016) **22** 6065-6075. PMID: 27468198
26. Barreto SG, Singh A, Perwaiz A. **Perioperative antimicrobial therapy in preventing infectious complications following pancreatoduodenectomy**. *Indian J Med Res* (2017) **146** 514-519. PMID: 29434066
27. Donald GW, Sunjaya D, Lu X. **Perioperative antibiotics for surgical site infection in pancreaticoduodenectomy: does the SCIP-approved regimen provide adequate coverage?**. *Surgery* (2013) **154** 190-196. PMID: 23664266
28. Fong ZV, McMillan MT, Marchegiani G. **Discordance between perioperative antibiotic prophylaxis and wound infection cultures in patients undergoing pancreaticoduodenectomy**. *JAMA Surg* (2016) **151** 432-439. PMID: 26720272
29. Pham H, Chen A, Nahm CB, Lam V, Pang T, Richardson AJ.. **The role of targeted versus standard antibiotic prophylaxis in pancreatoduodenectomy in reducing postoperative infectious complications: a systematic review and meta-analysis**. *Ann Surg* (2022) **275** 315-323. PMID: 33630442
30. Cheng H, Chen BP, Soleas IM, Ferko NC, Cameron CG, Hinoul P.. **Prolonged operative duration increases risk of surgical site infections: a systematic review**. *Surg Infect (Larchmt)* (2017) **18** 722-735. PMID: 28832271
31. Onyekwelu I, Yakkanti R, Protzer L, Pinkston CM, Tucker C, Seligson D.. **Surgical wound classification and surgical site infections in the orthopaedic patient**. *J Am Acad Orthop Surg Glob Res Rev* (2017) **1** e022. PMID: 30211353
32. Speicher PJ, Nussbaum DP, Scarborough JE. **Wound classification reporting in HPB surgery: can a single word change public perception of institutional performance?**. *HPB (Oxford)* (2014) **16** 1068-1073. PMID: 24852206
33. Suragul W, Rungsakulkij N, Vassanasiri W. **Predictors of surgical site infection after pancreaticoduodenectomy**. *BMC Gastroenterol* (2020) **20** 201. PMID: 32586351
34. Melzer M, Toner R, Lacey S, Bettany E, Rait G.. **Biliary tract infection and bacteremia: presentation, structural abnormalities, causative organisms and clinical outcomes**. *Postgrad Med J* (2007) **83** 773-776. PMID: 18057178
35. Vedantam G, Hecht DW.. **Antibiotics and anaerobes of gut origin**. *Curr Opin Microbiol* (2003) **6** 457-461. PMID: 14572537
36. Frisan T.. **Co- and polymicrobial infections in the gut mucosa: the host-microbiota–pathogen perspective**. *Cell Microbiol* (2021) **23** e13279. PMID: 33040471
37. De Vos MGJ, Zagorski M, McNally A, Bollenbach T.. **Interaction networks, ecological stability, and collective antibiotic tolerance in polymicrobial infections**. *Proc Nat Acad Sci* (2017) **114** 10666-10671. PMID: 28923953
38. Shah PM, Edwards BL, Dietch ZC. **Do polymicrobial intra-abdominal infections have worse outcomes than monomicrobial intra-abdominal infections?**. *Surg Infect (Larchmt)* (2016) **17** 27-31. PMID: 26397376
|
---
title: Impact of a syndrome-specific antibiotic stewardship intervention on antipseudomonal
antibiotic use in inpatient diabetic foot infection management
authors:
- Randy J. McCreery
- Elizabeth Lyden
- Matthew Anderson
- Trevor C. Van Schooneveld
journal: 'Antimicrobial Stewardship & Healthcare Epidemiology : ASHE'
year: 2023
pmcid: PMC10028944
doi: 10.1017/ash.2023.123
license: CC BY 4.0
---
# Impact of a syndrome-specific antibiotic stewardship intervention on antipseudomonal antibiotic use in inpatient diabetic foot infection management
## Body
Antimicrobial stewardship programs (ASPs) are an integral part of the United States public health strategy to combat the emergence of antimicrobial resistant bacteria. 1 Syndromic stewardship interventions have been used by ASPs to reduce unnecessary antibiotic use and to improve patient care in common infections. 2,3 In 2010, 77,491 hospital discharges were attributed to diabetic foot infections (DFIs), making this a common hospital-managed infection. However, published strategies to improve antibiotic use in these cases are lacking. 2–4 National guidelines for the treatment of DFI were published in 2012 and include antibiotic recommendations based upon severity and prior culture data. 5 A few reports have suggested that adherence to antibiotic prescribing guidelines in DFI is suboptimal, but when guidelines are followed, improved outcomes have been observed. 6–9 Clinical decision support systems (CDSSs) have been utilized with the goal of improving guideline-adherent treatment for infectious diseases, although mixed success has been reported. 10 In DFI, CDSS associated with guideline adherence and reduction in unnecessary antibiotic use has not been reported.
A review of DFIs at our institution (July 2014–June 2015) revealed that *Pseudomonas aeruginosa* was an uncommon DFI pathogen, present in 1 ($2\%$) of 56 cases, but DFI was empirically treated in 41 ($73\%$) of 56 cases. 11 This finding suggested that DFI was an excellent opportunity for a syndromic stewardship intervention. Between November 2017 and March 2018, we conducted a 3-part ASP intervention that included provider education and the creation of an institutional management guideline and DFI order set in the electronic medical record (EMR). We evaluated the effect of this intervention (education, guideline, and order set) on DFI antibiotic use and clinical outcomes.
## Abstract
### Objective:
To demonstrate that a syndromic stewardship intervention can safely reduce antipseudomonal antibiotic use in the treatment of inpatient diabetic foot infections (DFIs).
### Intervention and method:
From November 2017 through March 2018, we performed an antimicrobial stewardship intervention that included creation of a DFI best-practice guideline, implementation of an electronic medical record order set, and targeted education of key providers. We conducted a retrospective before-and-after study evaluating guideline adherent antipseudomonal antibiotic use 1 year before and after the intervention using interrupted time-series analysis.
### Setting:
University of Nebraska Medical Center, a 718-bed academic medical center in Omaha, Nebraska.
### Patients:
The study included 193 adults aged ≥19 years (105 in the preintervention group and 88 in the postintervention group) admitted to non–intensive care units whose primary reason for antibiotic treatment was diabetic foot infection (DFI).
### Results:
Guideline-adherent use of antipseudomonal antibiotics increased from $39\%$ before the intervention to $68\%$ after the intervention (P ≤.0001). Antipseudomonal antibiotic use decreased from 538 days of therapy (DOT) per 1,000 DFI patient days (PD) before the intervention to 272 DOT per 1,000 DFI PD after the intervention ($P \leq .0001$), with a statistically significant decrease in both level of use and slope of change. We did not detect any changes in length of stay, readmission, amputation rate, subsequent positive Clostridioides difficile testing, or mortality.
### Conclusions:
Our 3-component intervention of guideline creation, implementation of an order set, and targeted education was associated with a significant decrease in antipseudomonal antibiotic use in the management of inpatient DFIs. DFIs are common and should be considered as opportunities for syndromic stewardship intervention.
## Study setting
The University of Nebraska Medical Center (UNMC) is a 718-bed, academic medical center located in Omaha, Nebraska. Institution-specific guidelines for the treatment of common infectious diseases with associated EMR order sets (ie, for urinary tract infections, pneumonia, etc) has been a core strategy of the program, coupled with audit and feedback.
## Study design and population
We retrospectively assessed empiric antibiotic use in hospitalized patients with DFI before and after a stewardship intervention. Guideline-adherent antibiotic use was measured before the intervention (November 1, 2016–October 31, 2017) and was compared to guideline-adherent antibiotic use after the intervention (April 1, 2018–March 31, 2019), excluding an implementation period (November 1, 2017,–March 31, 2018). We included adults aged ≥19 years (19 is the age of majority in Nebraska) who had been admitted to general medical wards through internal medicine (IM) or family medicine (FM) practices and received antibiotic treatment for a DFI. The IM and FM services include hospitalist teams and comprise >$90\%$ of DFI admissions. We identified eligible cases using International Classification of Diseases Tenth Revision (ICD-10) codes for DFI. Code E8-E13 identifies patients with diabetes mellitus and among these, they must have had at least 1 of the following codes: 0J, 0Q, 0S, 0Y for procedures involving lower extremity tissue, bones or joints or A48.0, I96, L02, L03, L97, M86 Z44, Z89 identifying gangrene, cellulitis, abscess, ulcer, osteomyelitis, or limb loss. Code definitions are included in the Supplementary Material. Cases were manually reviewed, and only cases in which infection was described as originating in the foot were included. Infections originating at the ankle and above were excluded, as were surgical wound infections. If an infection was present or suspected at any other site not contiguous with the foot, the case was excluded. Additional exclusion criteria included intensive care unit (ICU) admission, transfer from an outside hospital, admission to non-IM/FM or hospitalist services (surgical services rarely admit these patients at our institution), puncture wound or foreign body present, or admitted from hospice. Only the first admission for DFI was included. All patients were unique.
## Study definitions
The presence of a DFI and infection severity were defined based on national guidelines. 5 Infection was defined as having a wound present with at least 2 of the following: swelling, erythema, tenderness, warmth, or purulence. Severe infection was defined as an infection plus ≥2 of the following within 8 hours of admission: fever or hypothermia (temperature >38.0°C or <36.0°C), heart rate >90 beats per minute, respiratory rate >20 breaths per minute, or white blood cell count (WBC) >12,000/µL. Infections without evidence of systemic inflammatory response were consider nonsevere. Empiric coverage for P. aeruginosa was considered guideline adherent only if risk factors for P. aeruginosa were present (eg, water exposure or previous isolation as outlined in our institutional guideline). Empiric coverage was defined as the first systemic antibiotic regimen received. Topical antibiotics were excluded. Antibiotic use was measured in days of therapy (DOT), whereas any dose of an antibiotic in a single day constituted 1 day of therapy. Formulary antipseudomonal antibiotics included meropenem, piperacillin/tazobactam, cefepime, ceftazidime, aztreonam, levofloxacin, ceftolozane-tazobactam, ceftazidime-avibactam, and aminoglycosides. Only deep-tissue cultures were evaluated; they were categorized as having been obtained via surgical debridement (operative or bedside), amputation, or bone biopsy. All other wound cultures were excluded.
## Intervention
In July 2017, we produced an institutional guideline for the management of DFIs modeled after national guidelines and supplemented by local microbiologic data and formulary preferences (Appendix A). 5 This guideline was reviewed by our local ASP committee and was made available online. Recommendations included defining severity and risk factors for resistant pathogens, such as MRSA and P. aeruginosa, with empiric antibiotic recommendations based upon these factors. Additionally, a comprehensive DFI admission order set was created in the EMR with severity- and risk-factor–stratified empiric antibiotic orders available. The order set included options for additional consultation (eg, surgical, infectious diseases, endocrinology, nephrology, etc), suggestions for appropriate imaging and laboratory tests, as well as options for other medications including insulin.
Between November 2017 and March 2018, education was provided to clinicians who commonly prescribe antibiotics in DFIs. Emergency medicine residents received 1 session, hospitalists including advanced practice providers received 1 session, internal medicine residents received 1 session, and family medicine residents and attending physicians received 2 sessions. Content included education on local microbiology highlighting the rarity of P. aeruginosa as a DFI pathogen, current prescribing data noting opportunities for improvement, and an introduction to our institutional guideline and order set outlining recommended antibiotic regimens. Education was conducted at either department meetings or educational conferences typically lasting 15–20 minutes (5 sessions total). Our ASP conducts audit and feedback but did not specifically target DFI for review. The UNMC Institutional Review Board classified this work as a quality improvement project.
## Data collection
We collected demographic, clinical, laboratory, microbiologic, and treatment data. EMR order-set use was tabulated via request from our local EMR team. Online guideline document downloads were collected using local information technology resources. For vital signs, maximum or minimum values along with maximum WBCs within 8 hours of admission were recorded. To calculate the Elixhauser comorbidity score for each patient, ICD-10 codes from the source encounter file were parsed using SAS codes. If any of the predetermined ICD codes for a particular condition existed for a patient in the source file, that patient was said to have that condition and was assigned the standard Elixhauser score for that condition. All Elixhauser scores for each condition were summed to yield the final Elixhauser comorbidity index for each patient.
Conditions that make up the Elixhauser comorbidity index included peripheral vascular disorders, depression, lymphoma, metastatic cancer, solid tumor without metastasis, renal failure, and congestive heart failure.
## Study outcomes
The primary end point of interest in this study was adherence to empiric antipseudomonal antibiotic use guidelines. Secondary outcomes included antipseudomonal antibiotic use measured in DOT per 1,000 DFI patient days (PD), length of stay (LOS), 30-day readmission, amputation, subsequent positive Clostridioides difficile testing within 30 days of admission, 30-day mortality, and 1-year mortality.
## Statistical analysis
Descriptive statistics including means, standard deviations, medians, minimums, maximums, counts, and percentages were used to summarize the data. Categorical data were compared between the 2 periods using the Fisher exact test. The independent sample t test or Mann-Whitney test was used to compare continuous data.
A negative binomial regression model for interrupted time series was used to compare the change in adherence rates prior to and after the intervention. We hypothesized that the intervention was associated with both a level change and a slope change in the rate of antipseudomonal therapy prescribed between the 2 periods following the approach outlined by Bernal et al. 12
## Results
In total, 874 cases were identified and reviewed, and 681 were excluded (Fig. 1). We included 193 unique patients in the analysis: 105 in the preintervention period and 88 in the postintervention period.
Fig. 1.Study patient flow. Note. DFI, diabetic foot infection; IM, internal medicine; FM, family medicine; ICU, intensive care unit.
Demographics and clinical characteristics are shown in Table 1. The groups were similar with respect to demographics, infection severity, and comorbidities. Clinical outcomes, including mortality, readmission, Clostridioides difficile infection (CDI) rates, LOS, and amputation rate, were not significantly different between the 2 periods (Table 2).
Table 1.Baseline Patient Demographic and Clinical CharacteristicsCharacteristicPreintervention Period($$n = 105$$),No. (%) a Postintervention Period($$n = 88$$),No. (%) a P ValueAge, mean y (SD)57.6 (12.6)57.7 (11.7).95Sex61 (58.1)59 (67.0).23BMI, mean (SD)35.0 (8.7)34.1 (11.1).55 Ethnicity White74 (70.4)64 (72.7).66Black18 (17.1)11 (12.5)Hispanic or Latino10 (9.5)8 (9.0)Other3 (2.8)5 (5.6)HbA1C, mean (SD)9.1 (2.5)9.2 (2.4).87Heart rate, mean (SD)95.1 (15.2)95.1 (16.1).99Respiratory rate, mean (SD)19.6 (2.0)19.7 (2.0).77WBC, mean (SD)11.5 (4.7)12.1 (5.9).45Systolic BP, mean (SD)126.4 (21.4)127.4 (20.2).73Fever or hypothermia27 (25.7)27 (30.6).52Elixhauser comorbidity index, mean (SD)4.4 (5.5)4.0 (4.9).67Severe DFI50 (47.6)45 (51.1).66Note. SD, standard deviation; no., number; BP, blood pressure; HbA1C, hemoglobin A1C; WBC, white blood cell count; DFI, diabetic foot infection. a Units unless otherwise specified.
Table 2.Clinical OutcomesOutcomePreintervention Period($$n = 105$$),No. (%) a Postintervention Period($$n = 88$$),No. (%) a P ValueGuideline adherent empiric antipseudomonal antibiotic use41 [39]60 [68]<.0001Empiric antipseudomonal antibiotic use no. (%) 68 (64.7)35 (39.7).000830-day mortality b 1 (0.9)1 (1.2)1.01-year mortality b 10.4 ($\frac{10}{96}$)11.1 ($\frac{6}{54}$)1.030-day readmission18 (17.1)18 (20.4).58 C. difficile within 30 d (n/N)0 ($\frac{0}{105}$)2.2 ($\frac{2}{88}$).20LOS median d, (min–max)4.9 (0.91–61)5.1 (0.21–49).84Amputation during admission36 (34.2)25 (28.4).43Note. C. difficile, Clostridioides difficile; LOS, length of stay. a Units unless otherwise specified. b *Mortality data* were unavailable for 3 patients in the 30-day preintervention group and for 9 patients in the 1-year preintervention group. Mortality data were unavailable for 6 patients in the 30-day postintervention group and for 34 patients in the 1-year postintervention group.
Figure 2 presents deep-tissue culture results for 3 different periods: the preintervention period, the postintervention period, and the preliminary 2014–2015 DFI case-review period briefly described in the introduction. Of 170 cases with data, P. aeruginosa was isolated in $3.5\%$ of cases.
Fig. 2.Deep-tissue culture results from 3 periods.*Chart displays the percent of cases with deep-tissue culture where a specific organism was present. Often >1 isolate per case. Note. MSSA, methicillin-susceptible Staphylococcus aureus; strep, streptococci; MRSA, methicillin-resistant Staphylococcus aureus; VRE, vancomycin-resistant Enterococcus; VGS, viridans group streptococci; P. aeruginosa, Pseudomonas aeruginosa.
A decrease in the rate of empiric antipseudomonal antibiotic use was observed in the postintervention group: 68 of 105 ($64.7\%$) in the preintervention period versus 35 of 88 ($39.7\%$) in the postintervention period ($$P \leq .0008$$). The antipseudomonal antibiotic DOT per 1,000 DFI PD decreased as well: 538.3 in the preintervention period versus 272.9 in the postintervention period ($P \leq .0001$). Guideline adherence to antipseudomonal antibiotic use recommendations improved in the postintervention group: 41 of 105 ($39\%$) in the preintervention period versus 60 of 88 ($68\%$) in the postintervention period ($P \leq 0.0001$). Most of this decrease was related to the avoidance of antipseudomonal antibiotics when not recommended.
A negative binomial regression model comparing rates of antipseudomonal therapy between the pre- and postintervention periods showed a $54\%$ reduction in empiric antipseudomonal therapy (OR, 0.46; $95\%$ CI, 0.32–0.66). We conducted an interrupted time-series analysis to explore the changes in the level and slope between the 2 periods (Fig. 3). Statistically significant decreases occurred in the level and slope of monthly antipseudomonal therapy after the intervention.
Fig. 3.Antipseudomonal days of therapy per 1,000 patient days before and after the intervention. Note. DOT, days of therapy; IRR, incidence rate ratio; CI, confidence interval.
The guideline was downloaded 687 times (409 times in Nebraska and 36 times outside the United States), and the order set was used 41 times during the postintervention period. Order-set usage remained stable, with 49 uses in the 11 months after the postintervention period (April 1, 2019–February 29, 2020).
## Discussion
Our syndromic stewardship intervention was associated with a significant decrease in empiric and overall antipseudomonal antibiotic use in the postintervention period. We targeted antipseudomonal antibiotics because we, and others, had found that *Pseudomonas aeruginosa* was an uncommon DFI pathogen. 13,14 In comparing the preintervention and postintervention periods, we did not detect differences in important patient-centered outcomes such as mortality, readmission, length of stay, or rates of CDI and amputation. These findings indicate that avoiding empiric antipseudomonal antibiotics in DFI patients admitted to general medical floors is safe when local rates of *Pseudomonas causing* infection are low. Similar interventions targeting skin and soft-tissue infections, and DFI specifically, have yielded reductions in the use of broad-spectrum antibiotics without subsequent increases in adverse clinical outcomes. 9,15 Our intervention was relatively simple; we utilized 3 components: [1] an evidence-based, locally customized, DFI treatment guideline available on our ASP website; [2] targeted education to providers who frequently manage DFI cases; and [3] implementation of an order set in the EMR.
Our guideline document provided antibiotic recommendations based on infection severity, pathogen-specific risk factors, local microbiologic data, and formulary preferences. Measuring guideline downloads could not be assessed with location specificity beyond the level of the state of Nebraska, but our website has an established reputation at our institution as a source of guidance on treatment of common infections. Access and review of a guideline, however, does not equate to implementation or causation. We were not able to specifically link online utilization to treatment decisions, although no other interventions were performed during the study period to limit antipseudomonal antibiotic use in skin and soft-tissue infections, sepsis, or DFI.
When developing our education, we targeted those who most frequently prescribe antibiotics for DFI at our institution (EM residents, hospitalists, IM/FM residents), and we presented local microbiologic and treatment data in the context of national guidelines to demonstrate opportunities for improvement.
Coupling the guideline and education with a EMR order set also likely contributed to our success. Order sets should be efficiency tools that provide convenient access to best practices. 16 Our guideline excluded orders not supported by guidelines (ie, swab cultures) while making correct management easy. Along with orders needed for the care of complex diabetic patients (eg, insulin orders and nephrology consultation), we included all items needed for DFI management: support for determining infection severity, prepopulated antibiotic orders, medical and surgical consultations, preferred imaging, vascular studies, etc. The order set was a “one-stop shop” for DFI admissions. The order set was used 41 times during the postintervention period among 88 cases, suggesting that $100\%$ adoption is not necessary for a successful intervention.
We had concern about order-set use degradation over time because much of the education targeted residents and one-third of internal medicine and family medicine residents exit the programs yearly. During the 11 months following our postintervention period (April 2019–February 2020), the order set was utilized an additional 49 times, suggesting durability. Residency training is heavily dependent on practice passed down from senior to junior, and we believe that our practice-improving recommendations have been integrated into this process. Despite our efforts, a substantial portion of patients were not treated in a guideline-adherent manner. This may have occurred for a variety of reasons, including care by providers not included in education (ie, they may not have attended a departmental education event), incomplete penetrance of our education, and other clinical factors.
This study had several limitations. Case selection bias may have been introduced by our retrospective design. This was a single-center study, and our results may not be generalizable to other centers. One potential confounder was that additional education was provided around the beginning of the intervention period when clinicians were recommended to generally avoid vancomycin and piperacillin-tazobactam combination therapy to reduce the risk of kidney injury. 17 Severe DFI was specifically mentioned as a situation in which combination therapy with vancomycin and piperacillin-tazobactam may be appropriate. During this study, we noted significant decreases in piperacillin-tazobactam use, but we also noted subsequent increases in the use of cefepime. Specific to DFI, we detected an overall decrease in antipseudomonal antibiotic use, suggesting that clinicians changed their spectrum of the therapy rather than replacing one antipseudomonal agent with another. We did not specifically capture the presence of osteomyelitis, although we do not expect the diagnosis of osteomyelitis (which is often not established at the time empiric antibiotics are selected) to have had an effect on empiric antibiotic choices. Additionally, similar LOS and amputation rates between groups does not indicate that major differences in rates of osteomyelitis were present. Other unmeasured factors may also have influenced antibiotic use, although no major changes were made in the activities of the ASP nor were DFIs specifically targeted for review. Although no significant differences were detected in patient-centered outcomes (LOS, mortality, C. difficile infection, etc), this study was not powered to detect differences in these outcomes, and these results should be interpreted with caution.
Overall, this pattern of creating a DFI institutional guideline, best-practice order set coupled with targeted education was associated with a safe and significant reduction in antipseudomonal antibiotic use. Our experience suggests that DFI should be recognized as an additional opportunity for syndromic stewardship intervention for hospitals and healthcare systems.
## Financial support
No financial support was provided relevant to this article.
## Conflicts of interest
All authors report no conflicts of interest relevant to this article.
## References
1. 1.
The White House national action plan for combating antibiotic-resistant bacteria. Centers for Disease Control and Prevention website. https://www.cdc.gov/drugresistance/us-activities/national-action-plan.html. Published 2015. Accessed February 16, 2022.. (2015)
2. Barlam TF, Cosgrove S, Abbo L. **Executive summary. Implementing an antibiotic stewardship program: guidelines by the Infectious Diseases Society of America and the Society for Healthcare Epidemiology of America**. *Clin Infect Dis* (2016) **62** 1197-1202. PMID: 27118828
3. Pollack LA, Srinivasan A.. **Core elements of hospital antibiotic stewardship programs from the Centers for Disease Control and Prevention**. *Clin Infect Dis* (2014) **3** S97-S100
4. Duhon BM, Hand E, Howell C. **Retrospective cohort study evaluating the incidence of diabetic foot infections among hospitalized adults with diabetes in the United States from 1996–2010**. *Am J Infect Control* (2016) **44** 199-202. PMID: 26541067
5. Lipsky BA, Berendt A, Cornia P. **2012 Infectious Diseases Society of America clinical practice guideline for the diagnosis and treatment of diabetic foot infections**. *Clin Infect Dis* (2012) **54** e132-e173. PMID: 22619242
6. Pence LM, Mock CM, Kays MB. **Correlation of adherence to the 2012 Infectious Diseases Society of America practice guidelines with patient outcomes in the treatment of diabetic foot infections in an outpatient parenteral antimicrobial programme**. *Diabet Med* (2014) **31** 1114-1120. PMID: 24825001
7. Yun R, Shakowski C, Fish D.. **Adherence to IDSA Guidelines for the initial treatment of diabetic foot infections in the emergency department (ED)**. *Open Forum Infect Dis* (2015) **2** 1520
8. Cahn A, Elishuv O, Olshtain-Pops K.. **Establishing a multidisciplinary diabetic foot team in a large tertiary hospital: a workshop**. *Diabetes Metab Res Rev* (2014) **30** 350-353. PMID: 24446250
9. Sotto A, Richard JL, Combescure C. **Beneficial effects of implementing guidelines on microbiology and costs of infected diabetic foot ulcers**. *Diabetologia* (2010) **53** 2249-2255. PMID: 20571753
10. Rittmann B, Stevens MP.. **Clinical decision support systems and their role in antibiotic stewardship: a systematic review**. *Curr Infect Dis Rep* (2019) **21** 29. PMID: 31342180
11. McCreery R, Bergman S, VanSchooneveld T.. **Adherence to practice guidelines for treating diabetic foot infections: an opportunity for syndromic stewardship**. *Open Forum Infect Dis* (2018) **1** S536
12. Bernal JL, Cummins S, Gasparrini A.. **Interrupted time-series regression for the evaluation of public health interventions: a tutorial**. *Int J Epidemiol* (2017) **46** 348-355. PMID: 27283160
13. Uckay I, Holy D, Schoni M. **How good are clinicians in predicting the presence of**. *Endocrinol Diabetes Metab* (2021) **4** e00225. PMID: 33855224
14. Young H, Knepper B, Hernandez W. *J Am Podiatr Med Assoc* (2015) **105** 125-129. PMID: 25815651
15. Jenkins TC, Knepper B, Sabel A. **Decreased antibiotic utilization after implementation of a guideline for inpatient cellulitis and cutaneous abscess**. *Arch Intern Med* (2011) **171** 1072-1079. PMID: 21357799
16. McGreevey JD. **Order sets in electronic health records: principles of good practice**. *Chest* (2013) **143** 228-235. PMID: 23276846
17. Rutter WC, Cox J, Martin C. **Nephrotoxicity during vancomycin therapy in combination with piperacillin-tazobactam or cefepime**. *Antimicrob Agents Chemother* (2017) **61** e02089-16. PMID: 27895019
|
---
title: Mitochondrial electron transport chain, ceramide and Coenzyme Q are linked
in a pathway that drives insulin resistance in skeletal muscle
authors:
- Alexis Diaz-Vegas
- Soren Madsen
- Kristen C. Cooke
- Luke Carroll
- Jasmine X. Y. Khor
- Nigel Turner
- Xin Ying Lim
- Miro A. Astore
- Jonathan Morris
- Anthony Don
- Amanda Garfield
- Simona Zarini
- Karin A. Zemski Berry
- Andrew Ryan
- Bryan C. Bergman
- Joseph T. Brozinick
- David E. James
- James G. Burchfield
journal: bioRxiv
year: 2023
pmcid: PMC10028964
doi: 10.1101/2023.03.10.532020
license: CC BY 4.0
---
# Mitochondrial electron transport chain, ceramide and Coenzyme Q are linked in a pathway that drives insulin resistance in skeletal muscle
## Summary
Insulin resistance (IR) is a complex metabolic disorder that underlies several human diseases, including type 2 diabetes and cardiovascular disease. Despite extensive research, the precise mechanisms underlying IR development remain poorly understood. Here, we provide new insights into the mechanistic connections between cellular alterations associated with IR, including increased ceramides, deficiency of coenzyme Q (CoQ), mitochondrial dysfunction, and oxidative stress. We demonstrate that elevated levels of ceramide in the mitochondria of skeletal muscle cells results in CoQ depletion and loss of mitochondrial respiratory chain components, leading to mitochondrial dysfunction and IR. Further, decreasing mitochondrial ceramide levels in vitro and in animal models (under chow and high fat diet) increased CoQ levels and was protective against IR. CoQ supplementation also rescued ceramide-associated IR. Examination of the mitochondrial proteome from human muscle biopsies revealed a strong correlation between the respirasome system and mitochondrial ceramide as key determinants of insulin sensitivity. Our findings highlight the mitochondrial Ceramide-CoQ-respiratory chain nexus as a potential foundation of an IR pathway that may also play a critical role in other conditions associated with ceramide accumulation and mitochondrial dysfunction, such as heart failure, cancer, and aging. These insights may have important clinical implications for the development of novel therapeutic strategies for the treatment of IR and related metabolic disorders.
## Introduction:
Insulin is the primary hormone responsible for lowering blood glucose, in part, by stimulating glucose transport into muscle and adipose tissue. This is mediated by the phosphatidylinositol 3-kinase/Akt dependent delivery of insulin sensitive glucose transporters (GLUT4) to the plasma membrane (PM)1,2. This process is defective in insulin resistance, a significant risk factor for cardiometabolic diseases such as type 2 diabetes3, heart failure4, and some types of cancer5 and so defective GLUT4 translocation represents one of the hallmarks of insulin resistance.
The development of insulin resistance in skeletal muscle and adipocytes has been associated with multiple intracellular lesions, including mitochondrial Coenzyme Q (CoQ) deficiency6, accumulation of intracellular lipids such as ceramides7 and increased mitochondrial reactive oxygen species (ROS)8–10. However, delineating the relative contribution of these lesions to whole body insulin resistance and their interconnectivity remains a challenge.
Coenzyme Q (CoQ, CoQ9 in rodents and CoQ10 in humans) is a mitochondrial cofactor and antioxidant synthesised and localised in the inner mitochondrial membrane (IMM). This cofactor is essential for mitochondrial respiration11, fatty acid oxidation12 and nucleotide biosynthesis13. We reported that mitochondrial, but not global, CoQ9 depletion is both necessary and sufficient to induce insulin resistance in vitro and in vivo6, suggesting a causal role of CoQ$\frac{9}{10}$ depletion in insulin resistance. CoQ deficiency can result from primary mutation in the CoQ biosynthetic machinery (named Complex Q)14 or secondary from other cellular defects such as deletion of the oxidative phosphorylation system (OXPHOS)15,16. Low levels of CoQ10 are associated with human metabolic disease, including diabetes17,18, cardiovascular disease19 and aging20. Strikingly, many of these conditions are also associated with loss of OXPHOS and mitochondrial dysfunction21,22. However, it is unclear what causes these mitochondrial defects or if they are mechanistically linked.
Ceramides belong to the sphingolipid family, and high levels are strongly associated with insulin resistance23. Whilst it has been proposed that ceramides cause insulin resistance by inhibition of PI3K/Akt signalling24–26, there is now considerable evidence that does not support this3,6,27,28. Hence, ceramides may induce insulin resistance by a non-canonical mechanism. The rapid decline of mitochondrial oxidative phosphorylation in isolated mitochondria in the presence of N-acetylsphingosine (C2-ceramide), a synthetic ceramide analog that can penetrate cells, suggests that ceramides may be responsible for defective mitochondria29. This effect seems to be ceramide-specific as neither diacylglycerides (DAGs) nor triacylglycerides (TAGs) affect mitochondrial respiration30. Recent evidence suggests a link between mitochondrial ceramides and insulin sensitivity, with the observation that reducing mitochondrial, but not global, ceramide in the liver protects against the development of diet induced insulin resistance and obesity30,31. Consistent with this, mitochondrial ceramide levels are more strongly associated with insulin resistance than with whole tissue ceramide in human skeletal muscle30. Despite this association, no direct evidence exists linking mitochondrial ceramides with insulin sensitivity in skeletal muscle.
Here we describe the linkage between mitochondrial Ceramide, CoQ, OXPHOS and ROS in the aetiology of insulin resistance. We show that a strong inverse relationship between mitochondrial CoQ and ceramide levels is intimately linked to the control of cellular insulin sensitivity. For example, increasing mitochondrial ceramide using either chemical or genetic tools, decreased mitochondrial CoQ levels, and induced insulin resistance. Conversely, genetic or pharmacologic manipulations that lowered mitochondrial ceramide levels increased CoQ levels and protected against insulin resistance. Increased mitochondrial ceramides also led to a reduction in several OXPHOS components, hindering mitochondrial respiration and elevating mitochondrial ROS in vitro. This was further supported in human skeletal muscle, where we observed a strong association between insulin sensitivity, abundance of OXPHOS and mitochondrial ceramides. We propose that increased mitochondrial ceramides cause a depletion in various OXPHOS components, leading to mitochondrial malfunction and deficiency in CoQ, resulting in increased ROS and insulin resistance. This provides a significant advance in our understanding of how ceramide causes mitochondrial dysfunction and insulin resistance in mammals.
## Palmitate induces insulin resistance by increasing ceramides and lowering CoQ9 levels in L6 –myotubes
Lipotoxicity plays a major role in insulin resistance and Cardiometabolic disease32. Excess lipids accumulate in insulin target tissues, such as muscle, impairing insulin-stimulated GLUT4 translocation as well as other metabolic actions of insulin. For this reason, several in vitro models have been employed involving incubation of insulin sensitive cell types with lipids such as palmitate to mimic lipotoxicity in vivo9. In this study we have used cell surface GLUT4-HA abundance as the main readout of insulin response. As shown (Fig. 1A), incubation of L6 myotubes with palmitate (150 μM for 16 h) reduced the insulin-stimulated translocation of GLUT4 to the cell surface by ~30 %, consistent with impaired insulin action. Despite this marked defect in GLUT4 translocation, we did not observe any defect in proximal insulin signalling as measured by phosphorylation of Akt or TBC1D4 (Fig. 1C & D - 100nM insulin; 20mins). Previous studies linking ceramides to defective insulin signalling have utilised the short chain ceramide analogue (C2-ceramide)24–26. Intriguingly, we were able to replicate that C2-ceramide inhibited both GLUT4 translocation and Akt phosphorylation in L6 myocytes (Fig. 1B, C & D). One possibility is that palmitate induces insulin resistance in L6 myotubes via a ceramide-independent pathway. However, this is unlikely as palmitate-induced insulin resistance was prevented by the ceramide biosynthesis inhibitor myriocin (Fig. 1A) and we observed a specific increase in C16-ceramide levels in L6 cells following incubation with palmitate, which was also prevented by myriocin (Fig. 1E & F, Sup. Fig. 1). Based on these data we surmise that C2-ceramide does not faithfully recapitulate physiological insulin resistance, in contrast to that seen with incubation with palmitate.
We previously demonstrated that insulin resistance was associated with CoQ depletion in muscle from high-fat diet fed mice6. To test if CoQ supplementation reversed palmitate-induced insulin resistance, L6-myotubes were co-treated with palmitate plus CoQ9. Addition of CoQ9 had no effect on control cells but overcame insulin resistance in palmitate treated cells (Fig. 1A). Notably, the protective effect of CoQ9 appears to be downstream of ceramide accumulation, as it had no impact on palmitate-induced ceramide accumulation (Fig. 1E–F). Strikingly, both myriocin and CoQ9 reversed insulin resistance, suggesting that there might be an interaction between ceramides and CoQ in the induction of insulin resistance with palmitate in these cells. Moreover, we have previously shown that mitochondrial CoQ is a key determinant of insulin resistance6 suggesting that ceramides and CoQ may interact in mitochondria. To explore this link, we next examined the effect of palmitate on mitochondrial CoQ levels. As shown (Fig. 1G), palmitate lowered mitochondrial CoQ9 abundance by ~40 %, and this was prevented with myriocin. To test whether CoQ depletion is downstream of ceramide accumulation, we exposed GLUT4-HA-L6 myotubes to 4-nitrobenzoic acid (4NB) to competitively inhibit 4-hydroxybenzoate:prolyprenyl transferase (Coq2), a limiting step in CoQ9 synthesis33. 4NB (2.5 mM for 16 h) decreased mitochondrial CoQ9 to a similar extent as observed in palmitate-treated myocytes (Fig. 1 I) and generated insulin resistance in GLUT4-HA-L6 myotubes (Fig. 1 H). Notably, 4NB mediated insulin resistance was prevented by provision of CoQ9, as previously described6. Interestingly, total ceramide abundance was increased in 4NB treated cells albeit to a lesser extent than observed with palmitate, without affecting other lipid species (Fig. 1 J & K, Sup. Fig. 1). One possibility is that CoQ directly controls ceramide turnover34. An alternate possibility is that CoQ inside mitochondria is necessary for fatty acid oxidation12 and CoQ depletion triggers lipid overload in the cytoplasm promoting ceramide production35. In fact, increased fatty acid oxidation is protective against insulin resistance in several model organisms36–38. Future studies are required to determine how CoQ depletion promotes Cer accumulation. Regardless, these data indicate that ceramide and CoQ have a central role in regulating cellular insulin sensitivity.
Since palmitate treatment can have a number of effects beyond ceramides, we next attempted to increase intracellular ceramides by inhibiting the ceramide degradation pathway. We exposed L6 myotubes to different concentrations of Saclac, an inhibitor of acid ceramidase (Kao et al., 2019), for 24 h. Saclac increases ceramides in L6 cells in a dose-dependent fashion, with the largest effect on C16:0 ceramides (Sup. Fig. 1F). Interestingly, Saclac also promoted accumulation of DAGs, sphingosine-1 phosphate (S1P) and sphingosine (SPH), demonstrating the tight interaction between these lipid species (Sup. Fig. 1 B, C & E). Consistent with a role of ceramide in insulin sensitivity, Saclac (10 μM for 24 h) reduced insulin stimulated GLUT4 translocation by $40\%$ (Fig. 1L, vs Control; $p \leq 0.001$) and this was prevented by myriocin or CoQ9 supplementation (Fig. 1L). Notably, no detectable defects in Akt phosphorylation were observed (Sup. Fig. 1G & H).
To explore if ceramides promote CoQ depletion beyond skeletal muscle, human cervical cancer cells (HeLa) were exposed to Saclac, as previously described (2 μM for 24 h)39. Consistent with our observation in L6-myotubes, Saclac increased total ceramide levels (~6 fold over basal) (Sup. Fig. 2 A & B) and lowered CoQ levels inside mitochondria (Sup. Fig 2C). Of interest, myriocin prevented Saclac-induced-CoQ depletion demonstrating that there is a similar interaction between ceramide and CoQ levels in this human cell line as observed in L6 cells (Sup. Fig. 2D). Moreover, this was relatively specific to CoQ as we did not observe any change in mitochondrial mass with Saclac (Sup. Fig 2 E–G), cell viability (Sup. Fig. 2H) or DAGs abundance (Sup. Fig. 2 I) Regardless, these data indicate that there is a strong association between ceramide and CoQ and that this has a central role in regulating cellular insulin sensitivity.
## Mitochondrial ceramide promotes insulin resistance by lowering CoQ levels.
Although mitochondrial ceramides have been linked with insulin resistance in human skeletal muscle30, to date, there is no direct evidence linking mitochondrial ceramides with insulin sensitivity. We wanted to determine if ceramide accumulation specifically in mitochondria is associated with altered CoQ levels and insulin resistance. To achieve this, we employed doxycycline-Tet-On inducible40 overexpression of a mitochondrial-targeted Sphingomyelin Phosphodiesterase 5 (mtSMPD5) in GLUT4-HA-L6 cells (GLUT4-HA-L6-mtSMPD5) (Fig. 2A). SMPD5 is a murine mitochondria-associated enzyme41 that hydrolyses sphingomyelin to produce ceramides42. Thus, overexpressing mtSMPD5 should specifically increase ceramides within mitochondria and avoid potential non-specific effects associated with small molecule inhibitors. As expected, doxycycline induced mitochondrial expression of mtSMPD5, as demonstrated by enrichment of SMPD5 in mitochondria isolated from L6 cells (Fig. 2B) and this was associated with increased total mitochondrial ceramides to the same extent as observed with palmitate treatment (Fig. 2C), with the largest increase in C16-ceramide (Fig. 2D). Importantly, mtSMPD5 overexpression did not affect ceramide abundance in the whole cell lysate nor other lipid species inside mitochondria such as cardiolipin, cholesterol and DAGs (Sup. Fig. 3 A, D–J). Intriguingly, mtSMD5 did not affect sphingomyelin levels in mitochondria (Sup. Fig. 3G), consistent with exchange between mitochondrial and extra-mitochondrial sphingomyelin pools to compensate for the degradation induced by SMPD5 overexpression43. Consistent with our hypothesis, mtSMPD5 was sufficient to promote insulin resistance in response to submaximal and maximal insulin doses (Fig. 2E). Furthermore, mtSMPD5 overexpression promoted insulin resistance without affecting Akt phosphorylation (Fig. 2F – H), and no differences in total GLUT4 levels were observed across the treatments (Fig. 2F & Sup. Fig. 3B). We next explored if mitochondrial ceramide-induced insulin resistance was mediated by lowering CoQ within mitochondria. In line with our previous results, Mitochondrial CoQ levels were depleted in both palmitate-treated and mtSMPD5-overexpressing cells without any additive effects. This suggests that these strategies to increase ceramides share a common mechanism for inducing CoQ depletion in L6 myotubes (Fig. 2I). Importantly, CoQ9 supplementation prevented both palmitate- and mtSMPD5 induced-insulin resistance (Fig. 2J), suggesting that CoQ depletion is an essential mediator of insulin resistance.
## Mitochondrial ceramides are necessary for palmitate-induced CoQ depletion and insulin resistance.
Given increased mitochondrial ceramides are sufficient to induce CoQ depletion and insulin resistance, we next asked whether increased mitochondrial ceramides are necessary to drive these phenotypes. Using the doxycycline-Tet-On inducible system40 we overexpressed a mitochondrial-targeted Acid Ceramidase 1 (mtASAH1) in GLUT4-HA-L6 cells (GLUT4-HA-L6-mtASAH1) (Fig. 3A). ASAH1 degrades ceramides to fatty acid and sphingosine44. Hence, mitochondrial overexpression of ASAH1 was expected to selectively lower ceramides inside mitochondria. Doxycycline increased the abundance of mtASAH1 in the mitochondrial fraction, demonstrating the correct localisation of this construct (Fig. 3B). Furthermore, mtASAH1 induction prevented palmitate-induced mitochondrial ceramide accumulation (total levels and 18:1\16:0 ceramides) (Fig. 3C), indicating the enzyme was functioning as expected. Similar to our observations with mtSMD5 overexpression, mtASAH1 did not alter ceramide abundance in the whole cell lysate or mitochondrial sphingomyelin levels (Sup. Fig. 4 A&F). Notably, mtASAH1 overexpression protected cells from palmitate-induced insulin resistance without affecting basal insulin sensitivity (Fig. 3E). Similar results were observed using insulin-induced glycogen synthesis as an orthologous technique for GLUT4 translocation. These results provide additional evidence highlighting the role of dysfunctional mitochondria in muscle cell glucose metabolism (Sup. Fig. 5K). Importantly, mtASAH1 overexpression did not rescue insulin sensitivity in cells depleted of CoQ (2.5 mM 4NB for 24 h) supporting the notion that mitochondrial ceramides are upstream of CoQ (Fig. 3E). Neither palmitate nor mtASAH1 overexpression attenuated insulin-dependent Akt phosphorylation (Fig. 3 F–H) nor total GLUT4 abundance (Fig. 3F, Sup. Fig. 4B). Finally, mtASAH1 overexpression increased CoQ levels. In both control and mtASAH1 cells, palmitate induced a depletion of CoQ, however the levels in palmitate treated mtASAH1 cells remained similar to control untreated cells (Fig. 3I). This suggests that the absolute concentration of CoQ is crucial for insulin sensitivity, rather than the relative depletion compared to basal conditions, thus supporting the causal role of mitochondrial ceramide accumulation in reducing CoQ levels in insulin resistance.
In order to demonstrate the connection between ceramide and CoQ in vivo, we examined whether a reduction of ceramides in mouse skeletal muscle, using a Ceramide Synthase 1 (CerS1) inhibitor, would alter mitochondrial CoQ levels. Treatment of adult mice with the CerS1 inhibitor P053 for 6 wks selectively lowered muscle ceramides without affecting other lipid species (Fig 3J, Sup. Fig 5). Notably, CerS1 inhibition increased CoQ in mitochondrial fractions isolated from skeletal muscle (Fig. 3K), similar effect was observed in mice exposed to a high fat diet (HFD) for 5 wks (Supp. Fig. 4H–I). These animals exhibited an improvement in mitochondrial function and reduced muscle triglycerides and adiposity upon HFD (further phenotypic and metabolic characterization of these animals can be found in45) demonstrating the existence of the ceramide/CoQ relationship in muscle in vivo.
## Mitochondrial ceramides induce depletion of the electron transport chain.
We have established that both increased mitochondrial ceramides and a loss of mitochondrial CoQ are necessary for the induction of insulin resistance. As such these changes are likely to induce other mitochondrial defects. To gain insight into how increased mitochondrial ceramides drive changes in mitochondrial function we performed MS-based proteomics on L6 cells overexpressing mtSMPD5.
mtSMPD5-L6 myotubes were treated with doxycycline for various time points (2, 8, 24, 48 and 72 h) and positive induction was observed after 24 h of treatment (Fig. Sup. 6A). Subsequently, control, 24, and 72 h time points were selected for further studies. Mitochondria were purified via gradient separation and analysed using liquid chromatography-tandem mass spectrometry (LC-MS/MS) in data-independent acquisition (DIA) mode (Fig. 4A). Across control and mtSMPD5 cells we quantified 2501 proteins where 555 were annotated as mitochondrial proteins (MitoCarta 3.0 and uniprot localisation)46.
Analysis of the proteome revealed that 9 and 19 % of mitochondrially annotated proteins were significantly changed at 24 and 72 h respectively (adj. $p \leq 0.05$, Absolute log2 FC > 0.4) indicative of a temporal progression of changes following induction of mSMPD5 expression (Fig.4 B & C). 60 proteins were decreased at 72 h and of these $47\%$ were functionally annotated as components of oxidative phosphorylation (OXPHOS (rank 1, $\frac{28}{135}$ proteins). Within OXPHOS, we observed a significant depletion of the electron transport chain components (ETC). The ETC is composed of several complexes (complex I-V, CI-CV). In this dataset, CI ($\frac{14}{15}$ decreased), CIII ($\frac{5}{6}$ decreased) and CIV ($\frac{13}{13}$ decreased) but not CII ($\frac{0}{3}$ decreased) or CV ($\frac{3}{15}$ decreased) were depleted after mtSMPD5 overexpression (Fig. 4D). Despite the bulk downregulation of CI, III and IV, the assembly machinery associated with each complex was either upregulated or unchanged after mtSMPD5 overexpression (Fig. 4J) suggesting that mitochondrial ceramides somehow alter ETC stability. Intriguingly, neither CII nor CV were affected by mtSMPD5 suggesting that ceramides preferentially affect those ETC complexes that are part of structures known as supercomplexes (SCs)47. Importantly, as part of CoQ is found in SCs binding CI (Fig. 4F) we mapped the levels of individual subunits of CI onto the recently solved structure of bovine CI48. This revealed the loss of subunits around the N-module (Ndufs1, Ndufs4, Ndufs6, Ndufv2 & Ndufv3) and Q-module (Ndufa5 & Ndufs2) in CI (Fig. 4H). Importantly, CI downregulation was not associated with reduction in gene expression as shown in Sup. Fig. 6J. The N-module is essential for NADH oxidation, and the Q-module is where CoQ binds CI. Hence, loss of the Q-module might trigger a stoichiometric depletion of CoQ upon ceramide accumulation.
Of note, we observed a heterogeneous response of the mitochondrial proteome after mtSMPD5 overexpression. For instance, proteins associated with glucose oxidation and mitochondrial translation/transcription did not change after mtSMPD5 induction (Fig. Sup. 6D, F & G), proteins involved in fatty acid oxidation and OXPHOS were consistently downregulated after 24 h treatment (Fig. 4B, C, & D and Sup. Fig. 6E), proteins related with the mitochondrial import machinery were consistently upregulated (Fig. Sup. 6C) and proteins associated with CoQ production were transiently downregulated after 24 h induction (Fig. 4E).
## Mitochondrial ceramides impair mitochondrial function
Based on the ceramide-dependent depletion of ETC members, we hypothesised that mitochondrial ceramides would impair mitochondrial function. To test this, we evaluated several aspects of mitochondrial function upon mtSMPD5 overexpression. Mitochondrial respiration is broadly considered to be the best measure for describing mitochondrial activity22,49. Respiration was assessed in intact mtSMPD5-L6 myotubes treated with CoQ9 by Seahorse extracellular flux analysis. mtSMPD5 overexpression decreased basal and ATP-linked mitochondrial respiration (Fig. 5 A, B &C), as well as maximal, proton-leak and non-mitochondrial respiration (Fig. 5 A, D, E & F) suggesting that mitochondrial ceramides induce a generalised attenuation in mitochondrial function. Notably, we did not observe evidence of energy deficiency in our model (data not shown). Interestingly, CoQ9 supplementation partially recovered basal and ATP-linked mitochondrial respiration, suggesting that part of the mitochondrial defects are induced by CoQ9 depletion. The attenuation in mitochondrial respiration is consistent with a depletion of the ETC subunits observed in our proteomic dataset (Fig. 4). Since mitochondrial respiratory activity is limited by several factors including nutrient supply, bioenergetic demands, among others50 we tested whether the ETC generally or a specific respiratory complex was affected by mtSMPD5 overexpression. We measured the activity of the respiratory chain by providing substrates for each respiratory chain complex to permeabilized cells and analysed oxygen consumption. mtSMPD5 overexpressing cells exhibited attenuated mitochondrial respiration irrespective of the substrate provided (Fig. 5 G–L) supporting the notion that mitochondrial ceramides induce a generalised defect in mitochondrial respiration. In line with defective mitochondrial function, cells with mtSMPD5 overexpression also exhibited increased oxidative stress (Fig. 5M) measured by the redox sensitive dye MitoSOX. Interestingly, no difference in mitochondrial membrane potential was observed across conditions (Fig. 5N). Collectively, these data suggest that increased mitochondrial ceramides cause a loss of mitochondrial respiratory capacity and an increase in ROS production as a result of ETC depletion in L6-myotubes.
## Association of mitochondrial proteome with insulin sensitivity and mitochondrial ceramides in human muscle
To further characterise the effect of mitochondrial ceramides on ETC abundance in a more physiological context we performed a cross-sectional study assessing the mitochondrial lipid profile and protein abundance in muscle biopsies obtained from four groups of people (athletes, lean, obese and type 2 diabetics (T2D). The demographic information and detailed lipidomic analysis of these individuals was previously reported30, Fig. Sup. 7 A). In line with our in-vitro data, long tail ceramides (C18:0) in the mitochondria/ *Endoplasmic reticulum* (ER) enriched fraction, but not whole tissue, were inversely correlated with muscle insulin sensitivity30. To expand this observation, we employed proteomics analysis of the mitochondrial/ER fraction from the same subjects (Fig. 6A). A total of 2,058 unique protein groups were quantified in at least one sample, where 571 were annotated as mitochondrial associated proteins (Human MitoCarta 3.0)46. After filtering (proteins in >$50\%$ of samples within each group), 492 mitochondrial proteins were reliably quantified across 67 samples (Fig. 6B). We noted that the mitochondrial fraction from athletes were enriched for mitochondrial proteins, and this could be corrected by global median normalisation (Fig. Sup. 7D). Pairwise comparison of the mitochondrial proteome between all groups revealed differences between groups, although relatively small in effect size (Fig. 6C & D). For instance, $16\%$ of all mitochondrial proteins were significantly different between T2D and athletes (Fig. 6C), however $56\%$ ($\frac{45}{80}$) of these proteins were changed by less than 1.5-fold. This trend was even stronger when comparing the obese group to the athletes, where $18\%$ of mitochondrial proteins were changed, but ~$80\%$ were changed less than 1.5-fold. Gene set enrichment revealed a highly significant general trend following the difference in insulin sensitivity (measure by the rate of glucose disappearance -Rd- using a stable isotope - [6,6-2H2]glucose - during a hyperinsulinemic-euglycemic clamp30, where TCA cycle and respiratory electron transport and Complex I biogenesis were enriched as follows: Athletes > Lean > Obese > T2D (Sup. Table 2). Of note, the T2D group had an enrichment of mitochondrial translation compared to the obese group.
To further explore the relationship between mitochondria and insulin sensitivity, the mitochondrial proteome was correlated to the muscle insulin sensitivity measured using 2H2 glucose Rd. As a group, all detected proteins within CI of the ETC were highly correlated with muscle insulin sensitivity ($$p \leq 4$$e-15) (Fig. 6E), and to a lesser extent proteins within CIV ($$p \leq 0.08$$) and CV ($$p \leq 0.09$$; Fig. Sup. 7E). The abundance of CII and CIII, together with the small and large mitochondrial ribosome subunits, were not associated with insulin sensitivity across all the samples (Fig. Sup. 7F). Next, we determined the association between the mitochondrial proteome and the levels of C18:0 ceramide in the mitochondria/ER fraction. In line with our previous observations, as a group, CI proteins were inversely correlated with mitochondrial ceramides ($$p \leq 8$$e-14) and no association was observed between CII and C18:0 ceramides across samples (Fig. 6F). Furthermore, components of CIV were also negatively correlated with mitochondrial ceramides although to a lesser extent (Sup. Fig. 7F, $$p \leq 0.026$$) and CV was not associated with mitochondrial ceramides (Sup. Fig. 7F, $$p \leq 0.737$$). According to these results, ETC subunits exhibit differential sensitivity to mitochondrial ceramides, with CI subunits being the most sensitive in human muscle. To uncover structural changes in CI that could correlate with increased ceramide we mapped those proteins significatively associated with mitochondrial ceramides to the bovine CI structure48. Consistent with L6-mtSMPD5 myotubes, the N and Q modules were the regions with the most negative associated subunits with mitochondria ceramides in human muscle (Fig. 6G). To determine the conservation in the changes in the mitochondrial proteome induced by increased ceramides, we compared the proteomes of mtSMPD5-L6-myotubes (72 h after induction) and human muscle biopsies. We observed that across the two datasets, CI and CIV subunits were downregulated after mtSMPD5 overexpression and were negatively associated with C18 ceramides in human samples (Fig. 6H). In turn, CI and CIV were positively associated with muscle insulin sensitivity (Fig. 6I), suggesting that these ETC subunits exhibited a conserved sensitivity to ceramide accumulation with a potential role in insulin sensitivity.
## Discussion
Insulin resistance is characterised by attenuated insulin-dependent glucose uptake in relevant target tissues, such as muscle and fat, and it plays a central role in cardiometabolic diseases3. In skeletal muscle, mitochondrial ceramides have been linked to insulin resistance30, however, to date no direct link connecting mitochondrial ceramides with insulin resistance has been established. Furthermore, CoQ depletion and defective mitochondria have also been independently associated with insulin resistance6,7. In the current study we present evidence suggesting that these factors are mechanistically linked inside mitochondria. Our data demonstrate that increased mitochondrial ceramides are both necessary and sufficient to induce insulin resistance in skeletal muscle. This is likely a function of increased ROS production that results from the specific depletion of the OXPHOS subunits and the concomitant loss of CoQ. Analysis of the human muscle mitochondrial proteome strongly supports mitochondrial ceramide linked changes in the OXPHOS machinery as major drivers of insulin sensitivity. In this study, we mainly utilised L6-myotubes, which share many important characteristics with primary muscle fibres. Both types of cells exhibit high sensitivity to insulin and respond similarly to maximal doses of insulin, with GLUT4 translocation stimulated between 2 to 4 times over basal levels in response to 100 nM insulin (as shown in Fig. 1–4,51,52). Additionally, mitochondrial respiration in L6-myotubes have a similar sensitivity to mitochondrial poisons, as observed in primary muscle fibres (as shown in Fig. 5,53). Finally, inhibiting ceramide production increases CoQ levels in both L6-myotubes and adult muscle tissue (as shown in Fig. 2–3). Therefore, L6-myotubes possess the necessary metabolic features to investigate the role of mitochondria in insulin resistance, and this relationship is likely applicable to primary muscle fibres.
Many stressors, including chronic inflammation and anticancer drugs, stimulate endogenous ceramide generation54 and CoQ depletion in mitochondria55,56. Nevertheless, experimental evidence testing the link between these molecules has been lacking. We observed that increased mitochondrial ceramides drive a depletion of mitochondrial CoQ leading to insulin resistance (Fig. 1 & 2), and that reducing mitochondrial ceramide protects against the loss of CoQ and IR (Fig. 3). These findings align with our earlier observations demonstrating that mice exposed to HFHSD exhibit mitochondrial CoQ depletion in skeletal muscle6. Given that CoQ supplementation is sufficient to overcome ceramide induced-IR (Fig. 1 & 2), but a reduction of mitochondrial ceramide does not overcome a loss of CoQ (Fig. 3), our data support a pathway whereby an increase in mitochondrial ceramides precedes loss of CoQ. Interestingly, inhibition of CoQ synthesis also increased ceramides, suggesting a bidirectionality to the ceramide-CoQ nexus. That said, this effect was modest (Fig. 1) and we cannot exclude off target effects of the inhibitor. It is possible that CoQ directly controls ceramide turnover34 or alternatively that CoQ inside mitochondria is necessary for fatty acid oxidation12 and CoQ depletion triggers lipid overload in the cytoplasm promoting ceramide production35. Further studies will be needed to determine how CoQ depletion promotes ceramide accumulation.
Our proteomics analysis revealed that the loss of CoQ parallels a loss of mitochondrial ETC complexes CI, CIII and CIV. These are known to form supercomplexes or respirasomes where ~25 – 35 % of CoQ is localised in mammals57,16. This bulk downregulation of the respirasome induced by ceramides may lead to CoQ depletion. The observation that both palmitate and SMPD5 overexpression trigger CoQ depletion without additive effects support the notion that ceramides may trigger the depletion of a specific CoQ9 pool localised within the inner mitochondrial membrane. Despite the significant impact of ceramide on mitochondrial respiration, we did not observe any indications of cell damage in any of the treatments, suggesting that our models are not explained by toxicity and increased cell death (Sup. Fig. 2H & J). Whilst the physiological role of respirasomes is still a subject of discussion, it has been suggested that they may enhance energy generation by optimising electron flow while reducing production of ROS58,59 and therefore their loss can be predicted to increase ROS generation. The 2 major mechanisms that might account for the loss of the respirasomes are decreased synthesis or increased degradation. Proteomics data suggests no deficiency in the OXPHOS biosynthetic machinery or assembly proteins and an increase in the machinery for protein import (Fig. 4). In addition, the absence of mRNA downregulation in mtSMPD5 overexpressing cells strongly suggests that at least a portion of the observed protein depletion within CI is attributed to diminished protein stability. It therefore seems reasonable to speculate that the loss of these mitochondrial complexes is driven by increased degradation. Interestingly, pharmacological CIII inhibition leads to respirasome degradation via oxidative stress produced by reverse electron transfer (RET)60. Since ceramides can directly inhibit CIII61 it is possible that a similar mechanism mediates the effect of ceramides on the respirasome (Fig. 5). This suggests that defective respirasome activity (e.g. induced by ceramides) triggers ROS, which over time depletes respirasome subunits and a stoichiometric CoQ depletion, leading to further ROS production as a consequence. Another possibility is that, because of its highly hydrophobic nature, ceramides impact membrane fluidity promoting a gel/fluid phase transition62. These alterations in membrane fluidity could decrease respirasome stability. It is likely that bound lipids stabilise the interactions between the complexes in the respirasome and that this is impaired by ceramides. In fact, bound lipid molecules are observed in the structure of the porcine respirasome63 and the isolated bovine CI63 but none of the lipids identified thus far directly bridge different complexes. In order to understand the role of lipids in stabilising respirasomes and the role of ceramides in such stabilisation, higher-resolution structures will be required64.
The current studies pose a number of key unanswered questions. First, how does ceramide accumulate in mitochondria in insulin resistance? This could involve transfer from a different subcellular compartment65,66 or in situ mitochondrial ceramide synthesis. Consistent with the latter, previous studies have suggested that various enzymes involved in ceramide metabolism are specifically found in mitochondria41,67–71. Notably, CerS1-derived ceramide induces insulin-resistance in skeletal muscle72. Although this enzyme has not been reported as a mitochondrial protein, it can be transferred from the endoplasmic reticulum surface to the mitochondria under cellular stress in metabolically active tissues such as muscle and brain73. This provides a potential mechanism where cellular stress, like nutrient overload, may induce transfer of CerS1 to mitochondria, increasing mitochondrial ceramide to trigger insulin resistance.
A further question is how ceramide regulates insulin sensitivity. We present evidence that mitochondrial dysfunction precedes insulin resistance. However, previous studies have failed to observe changes in mitochondrial morphology, respiration or ETC components during early stages of insulin resistance22,74. However, in many cases such studies fail to document changes in insulin-dependent glucose metabolism in the same tissue as was used for assessment of mitochondrial function. This is crucial because we and others do not observe impaired insulin action in all muscles from high fat fed mice for example6,9. In addition, surrogate measures such as insulin-stimulated Akt phosphorylation may not accurately reflect tissue specific insulin action as demonstrated in figure 1C. Thus, further work is required to clarify some of these inconsistencies. We observed that mitochondrial ceramides were associated with the loss of CoQ, increased production of mitochondrial ROS and impaired mitochondrial respiration6,9,10. As we discussed above, this is likely a direct result of respirasome depletion. The molecular linkage between ROS production and IR remains unknown. Early studies suggested that ceramides and ROS impaired canonical insulin signalling24–26, however, our current data do not support this, with the caveat that these were static signalling measures. One possibility is the release of a signalling molecule from the mitochondria that impairs insulin action75. The mitochondrial permeability transition pore (mPTP) is an attractive candidate for this release since its activity is increased by mitochondrial ROS76 and ceramides77. It has been shown that mPTP inhibition protects against insulin resistance in either palmitate- or ceramide-induced L6-myotubes and mice on a high-fat diet78. Furthermore, mPTP deletion in the liver protects against liver steatosis and insulin resistance in mice79. Strikingly, CoQ is an antioxidant and also an inhibitor of mPTP suggesting that part of the protective mechanism of CoQ may involve the mPTP80. Because CoQ can accumulate in various intracellular compartments, it’s important to consider that its impact on insulin resistance might be due to its overall antioxidant properties rather than being limited to a mitochondrial effect. Excitingly, mtSMPD5 increased the abundance of mPTP associated proteins suggesting a role of this pore in ceramide induced insulin resistance (Sup. Fig. 6E). It is also possible that ceramides generated within mitochondria in SMPD5 cells leak out from the mitochondria into other membranes (e.g. PM and GLUT4vesicles) affecting other aspects of GLUT4 trafficking and insulin action. However, the observation that ASAH1 overexpression reversed IR without affecting whole cell ceramides argues against this possibility.
Ultimately, the significant challenge for the field is the discovery of the unknown factor(s) released from mitochondria that cause insulin resistance, their molecular target(s), and the transduction mechanism(s).
The observations described above led us to speculate on whether there is a teleological reason for why these mitochondrial perturbations occur and why they drive insulin resistance? Under conditions of stress, nutrient incorporation into the cell needs to be adjusted to keep the balance between energy supply and utilisation. In situations where the mitochondrial respirasome is depleted, the mitochondria’s ability to oxidise nutrients can be easily overwhelmed without a corresponding reduction in nutrient uptake. In this scenario, insulin resistance may be a protective mechanism to prevent mitochondrial nutrient oversupply9. Beyond nutrient uptake, the respirasome depletion could also affect the ability of the mitochondria to switch between different energy substrates depending on fuel availability, named “metabolic Inflexibility”81 this mechanism may potentially play a role in the ectopic lipid accumulation seen in individuals with obesity, a condition linked with cardio-metabolic disease.
In summary, our results provide evidence for the existence of a mechanism inside mitochondria connecting ceramides, mitochondrial respiratory complexes, CoQ and mitochondrial dysfunction as part of a core pathway leading to insulin resistance. We identified that CoQ depletion links ceramides with insulin resistance and define the respirasome as a critical connection between ceramides and mitochondrial dysfunction. While many pieces of the puzzle remain to be solved, identifying the temporal link between ceramide, mitochondrial dysfunction and CoQ in mitochondria is an important step forward in understanding insulin resistance and other human diseases affecting mitochondrial function.
## Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, David James (David.james@sydney.edu.au) or James Burchfield (James.burchfield@sydney.edu.au)
## Materials availability
This study generated two new molecular tools to overexpress mitochondria targeted SMPD5 and ASAH1. The plasmids are available upon request.
## Data availability
All Lipidomic analyses are available as supplementary information.
## Administration of P053 to mice
Mice of the C57BL6/J strain were obtained from the Animal Resources Centre of Perth (WA, Australia). Mice were housed in a controlled 12:12 h light-dark cycle, and they had ad libitum access to water and food. The oral gavage administration of P053 (5 mg/kg) was performed daily, while the control animals received vehicle ($2\%$ DMSO). The experiments were approved by the UNSW animal care and ethics committee (ACEC $\frac{15}{48}$B), and followed guidelines issued by the National Health and Medical Research Council of Australia.
## Cell lines
Mycoplasma-free L6 myotubes overexpressing HA-GLUT4 and HeLa cell lines were used for all in vitro experiments (detailed below each legend). HA-GLUT4 overexpression is essential for studying insulin sensitivity in vitro as we previously described9. L6-myoblast and HeLa cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco by Life Technologies) supplemented with $10\%$ foetal bovine serum (FBS) (v/v) (Gibco by Life Technologies) and 2 mM Glutamax (Gibco by Life Technologies) at 37 °C and $10\%$ CO2. L6 myoblasts were differentiated in DMEM/Glutamax/$2\%$ *Horse serum* as previously described9. The media was replaced every 48 h for 6 d. For induction of SMPD5 or ASAH1, L6 myotubes were incubated with doxycycline from day 3 until day 6 after the initiation of differentiation. L6-myotubes were used day 7 after the initiation of differentiation. At least 90 % of the cells were differentiated prior to experiments.
## Lentiviral transduction
Lentivirus was made by transfecting LentiX-293T (Takara Bio) cells with Lenti-X Packaging Single Shot (Takara Bio) with one of the following plasmids pLVX-Tet3G, pLVX-TRE3G-SMPD5-Myc-DDK or pLVX-TRE3G-ASAH-Myc-DDK according to the manufacturer’s specifications. Virus containing media was collected from the LentiX-293T cells and concentrated using Lenti-X Concentrator (Takara Bio). pLVX-Tet3G virus and polybrene was added to L6-myoblast cells and cells were positively selected using neomycin to create Tet3G expressing cells. The Tet3G expressing L6 cells then were subsequently infected with polybrene and either the pLVX-TRE3G-SMPD5-Myc-DDK or pLVX-TRE3G-ASAH-Myc-DDK virus. Cells were selected using puromycin to create a Tet-inducible SMPD5-Myc-Flag-DDK or ASAH-Myc-Flag-DDK L6 cell line.
## Lipid extraction
Two-phase extraction of lipids from frozen tissue samples (20 mg) or cells was carried out using the methyl-tert-butyl ether (MTBE)/methanol/water (10:3:2.5, v/v/v) method(Matyash et al., 2008). Frozen tissue samples (~20 mg) were homogenised in 0.2 mL methanol ($0.01\%$ butylated hydroxytoluene, BHT) using a Precellys 24 homogenizer and Cryolys cooling unit (Betin Technologies) with CK14 (1.4-mm ceramide) beads. HeLa cells and L6-myotubes were washed with PBS and scraped into 0.6 mL of ice-cold methanol(Turner et al., 2018). Mitochondrial pellets were washed twice to remove BSA from the fraction (see below), and 30 ug of mitochondrial protein was used for extraction. The homogenates were spiked with an internal standard mixture (2 nmole of 18:$\frac{1}{15}$:0 d5-diacylglycerol and 18:$\frac{1}{17}$:0 SM, 4 nmole 14:$\frac{0}{14}$:$\frac{0}{14}$:$\frac{0}{14}$:0 cardiolipin, 5 nmole d7-cholesterol, 500 pmole 18:$\frac{1}{17}$:0 ceramide, and 200 pmole d17:1 sphingosine and d17:1 S1P), then transferred to 10 mL screw cap glass tubes. MTBE (1.7 mL) was added and the samples were sonicated for 30 min in an ice-cold sonicating water bath (Thermoline Scientific, Australia). Phase separation was induced by adding 417 μL of mass spectrometry-grade water with vortexing (max speed for 30 sec), then centrifugation (1000 × g for 10 min). The upper organic phase was transferred into 5 mL glass tubes. The aqueous phase was re-extracted 3 times (MTBE/methanol/water 10:3:2.5), combining the organic phase in the 5 mL glass tube. The organic phase was dried under vacuum in a Savant SC210 SpeedVac (Thermo Scientific). Dried lipids were resuspended in 500 μL of $80\%$ MeOH/ 0.2 % formic acid / 2mM ammonium formate and stored at −20°C until analysis.
## Lipid quantification
Lipids were quantified by selected reaction monitoring on a TSQ Altis triple quadrupole mass spectrometer coupled to a Vanquish HPLC system (ThermoFisher Scientific). Lipids were separated on a 2.1 100 mm Waters Acquity UPLC C18 column (1.7 μM pore size) using a flow rate of 0.28 mL/min. Mobile phase A was $0.1\%$ formic acid, 10 mM ammonium formate in $60\%$ acetonitrile/$40\%$ water. Mobile phase B was $0.1\%$ formic acid and 10mM ammonium formate in $90\%$ isopropanol/$10\%$ acetonitrile. Total run time was 25 min, starting at $20\%$ B and holding for 3 min, increasing to $100\%$ B from 3–14 min, holding at $100\%$ from 14–20 min, returning to $20\%$ B at 20.5 min, and holding at $20\%$ B for a further 4.5 min. Ceramides, sphingomyelin, sphingosine, and sphingosine 1-phosphate were identified as the [M+H]+ precursor ion, with m/z 262.3 (sphinganine), 264.3 (sphingosine), or 266.3 (sphinganine) product ion, and m/z 184.1 product ion in the case of sphingomyelin. Diacylglycerols (DAGs) were identified as the [M+NH4]+ precursor ion and product ion corresponding to neutral loss of a fatty acid + NH3. Cardiolipins were identified as the [M+H]+ precursor ion and product ion corresponding to neutral loss of a DAG. Cholesterol was detected using precursor m/z 369.4 and product m/z 161.1. TraceFinder software (ThermoFisher) was used for peak alignment and integration. The amount of each lipid was determined relative to its class-specific internal standard. Lipidomic profiling of skeletal muscle tissue was performed exactly as described (Turner et al, 2018).
## Mass spectrometry sample preparation
Isolated mitochondria were defrosted and centrifuged at 4 °C at 4,000 × g for 15 min, and supernatant was removed. The mitochondrial pellet was resuspended in 100 uL 2 % SDC in Tris-HCl buffer (100 mM; pH 8.0) and the protein concentration determined by BCA assay. 10 ug of each sample was aliquoted and volume adjusted to 50 uL with milli-Q water, and samples were reduced and alkylated by addition of TCEP and CAA (10 and 40 mM respectively) at 60 °C for 20 minutes. Once cooled to room temperature, 0.4 ug MS grade trypsin and Lys-C were added to each sample, and proteins were digested overnight (16 h) at 37 °C. Peptides were prepared for MS analysis by SDB-RPS stage tips. 2 layers of SDB-RPS material was packed into 200 μL tips and washed by centrifugation of StageTips at 1,000 × g for 2 min in a 96-well adaptor with 50 μL acetonitrile followed by $0.2\%$ TFA in $30\%$ methanol and then $0.2\%$ TFA in water. 50 μL of samples were loaded to StageTips by centrifugation at 1,000 g for 3 min. Stage tips were washed with subsequent spins at 1,000 g for 3 min with 50 uL $1\%$ TFA in ethyl acetate, then $1\%$ TFA in isopropanol, and $0.2\%$ TFA in $5\%$ ACN. Samples were eluted by addition of 60 μL $60\%$ ACN with $5\%$ NH4OH4. Samples were dried by vacuum centrifugation and reconstituted in 30 μL $0.1\%$ TFA in $2\%$ ACN.
## Mass spectrometry acquisition and analysis
Samples were analysed using a Dionex UltiMate™ 3000 RSLCnano LC coupled to a Exploris Orbitrap mass spectrometer. 3 μL of peptide sample was injected onto an in-house packed 75 μm × 40 cm column (1.9 μm particle size, ReproSil Pur C18-AQ) and separated using a gradient elution, with Buffer A consisting of 0.1 % formic acid in water and Buffer B consisting of $0.1\%$ formic acid in $80\%$ ACN. Samples were loaded to the column at a flow rate 0.5 μL min-1 at $3\%$ B for 14 min, before dropping to 0.3 μL min-1 over 1 min and subsequent ramping to $30\%$ B over 110 min, then to $60\%$ B over 5 min and $98\%$ B over 3 min and held for 6 min, before dropping to $50\%$ and increasing flow rate to 0.5 μL min-1 over 1 min. Eluting peptides were ionised by electrospray with a spray voltage of 2.3 kV and a transfer capillary temperature of 300°C. Mass spectra were collected using a DIA method with varying isolation width windows (widths of m/z 22–589) between 350 – 1650 according to Supplementary Table 1. MS1 spectra were collected between m/z 350 – 1650 m/z at a resolution of 60,000. Ions were fragmented with an HCD collision energy at $30\%$ and MS2 spectra collected between m/z 300–2000 at resolution of 30,000, with an AGC target of 3e5 and the maximum injection time set to automatic. Raw data files were searched using DIA-NN using library generated from a 16-fraction high pH reverse phase library83. The protease was set to Trypsin/P with 1 missed cleavage, N-term M excision, carbamidomethylation and M oxidation options on. Peptide length was set to 7–30, precursor range 350–1650, and fragment range 300–2000, and FDR set to $1\%$.
## Statistical Analysis of L6 mitochondrial proteome
Mouse MitoCarta(REF) was mapped to *Rattus norvegicus* proteins using OrthoDB identifiers downloaded from Uniprot. The Rat MitoCarta was used to annotate the L6 proteome. Manual scanning of the annotation revealed a number of known mitochondrial proteins not captured using this approach. Proteins were therefore classified as mitochondria if they were annotated by our mouse:rat Mitocarta (380 proteins), contained “mitochondrial” in the protein name (78 additional proteins; 231 overlap with mitocarta) or if the first entry under Uniprot “Subcellular location” was mitochondria (97 additional proteins; 374 overlap with Mitocarta or protein name). LFQ intensities were Log 2 transformed and normalised to the median of the mitochondrially annotated proteins. Identification of differentially regulated proteins was performed using moderated t-tests84. Functional enrichment was performed using the STRING web-based platform85.
## Statistical Analysis of human proteome and mito-ER lipidome
Analysis of the human proteome and mito-ER lipidome were performed with R (version 4.2.1). Identification of differentially regulated proteins between each group were performed using the R package limma86 and p-values were corrected with p.adjust (method = “fdr”) within each comparison. Correlations were calculated with biweight midcorrelations from the R package WGCNA87. Gene set enrichment was performed with the R package clusterProfiler88 utilising pathways from Reactome for differentially regulated proteins89. Custom mitochondrial genes were constructed from HGNC Database90 and enrichment and enrichment figures were done with the R package fgsa (https://www.biorxiv.org/content/$\frac{10.1101}{060012}$v3).
## High pH reverse phase fractionation and library generation
A pooled sample was made by combining 1 uL of each sample and fractionated by high pH reverse phase liquid chromatography. 50 uL of pooled sample was injected onto a Waters XBridge Peptide BEH C18 column (4.6 × 250 mm, 130 Å, 3.5 um) using a ThermoScientific UltiMate 3000 BioRS System and peptides separated using gradient elution at 1 mL min-1, with the column oven set to 30 °C. Buffer A consisted of 10 mM ammonium formate, and Buffer B consisted of 10 mM ammonium formate in 80 % acetonitrile, which both adjusted to pH 9.0 with ammonium hydroxide. Initially Buffer B was set to 10 % and ramped up to 40 % over 11 minutes, before ramping up to 100 % B over 1 minute and held for 5 min before returning to 10 % for re-equilibration. Peptides were separated into 64 fractions collected between 3.45 min to 14.5 min, and samples were concatenated into 16 final fractions. Fractions were dried using a GeneVac 2.0 vacuum centrifuge using the HPLC program, with a max temperature of 60 °C. Fractions were resuspended in 10 uL $0.1\%$ TFA in $2\%$ ACN and 2 uL was injected and separated as described for DIA samples above, however, MS was acquired in a DDA manner. An MS1 was collected between m/z 350 −1650 with a resolution of 60,000. The top 15 most intense precursors were selected from fragmentation with an isolation window of 1.4 m/z, resolution of 15,000, HCD collision energy of 30 %, with an exclusion window of 30 s. Raw files were searched with MaxQuant against a FASTA file containing the reviewed UniProt human proteome (downloaded May 2020).
## Matrigel-coated plates
Matrigel diluted 1:100 v/v in ice-cold PBS was dispensed into 96-well plates (Eppendorf Cell Imaging plate, UNSPSC 41122107; and Perkin Elmer Cell Carrier Ultra, Cat# 6055300) and incubated for 2 h at room temperature. Before use, plates were washed twice in PBS at room temperature.
## HA-GLUT4 assay
HA-GLUT4 levels on the plasma membrane were determined as previously described9,91. L6-myotubes stably overexpressing HA-Glut4 were washed twice with warm PBS and serum-starved for two hours (in DMEM/$0.2\%$ BSA/GlutaMAX/with 220 mM bicarbonate (pH 7.4) at 37 °C, 10 % CO2). Cells were then stimulated with insulin for twenty minutes, after which the cells were placed on ice and washed three times with ice cold PBS. Cells were blocked with ice cold 10 % horse serum in PBS for 20 min, fixed with 4 % paraformaldehyde (PFA) for 5 min on ice and 20 min at room temperature. PFA was quenched with 50 mM glycine in PBS for 5 min at room temperature. We measured the accessibility of the HA epitope to an anti-HA antibody (Covance, 16B12) for 1 h at room temperature. Cells were then incubated with 20 mg/mL goat anti-mouse Alexa-488-conjugated secondary antibody (Thermo Fisher Scientific) for 45 min at room temperature. The determination of total HA-GLUT4 was performed in a separate set of cells following permeabilization with $0.01\%$ saponin (w/v) and anti-HA staining (as above). Each experimental treatment group had its own total HA-GLUT4. A FLuostar Galaxy microplate reader (BMG LABTECH) was used to measure fluorescence (excitation 485 nm/emission 520 nm). Surface HA-GLUT4 was expressed as a fold over control insulin condition.
## Induction of insulin resistance
To promote insulin resistance, cells were stimulated for 16 h with 150 μM palmitate-BSA or EtOH-BSA as control. The palmitate was complexed with BSA as previously described9. Briefly, fatty acid was dissolved in $50\%$ ethanol and then diluted 25 times in 10.5 % fatty acid free BSA solution. These stock solutions were further diluted in culture media to reach a final concentration of 150 μM (Final lipid:BSA ratio 4:1).
## Glycogen synthesis assay
L6 myotubes overexpressing ASAH were grown and differentiated in 12-well plates, as described in the Cell lines section, and stimulated for 16 h with palmitate-BSA or EtOH-BSA, as detailed in the Induction of insulin resistance section.
On day seven of differentiation, myotubes were serum starved in plain DMEM for 3 and a half hours. After incubation for 1 hour at 37 °C with 2 μCi/ml D-[U-14C]-glucose in the presence or absence of 100 nM insulin, glycogen synthesis assay was performed, as previously described92
## Coenzyme Q determination
CoQ9 and CoQ10 content in cell lysates and mitochondrial fractions were determined as described previously(Burger et al., 2020). Aliquots of 15 μg mitochondrial protein as prepared below were adjusted to a volume of 100 μL with water and subsequently mixed with 250 μL ice-cold methanol containing $0.1\%$ HCl, 20 μL internal standard (CoQ8, 200 pmol in hexane, Avanti Polar Lipids) and 300 μL of hexane. The mixture was vortexed for 30 sec, centrifuged (9,000 g × 5 min) and the supernatant was transferred into deepwell plate $\frac{96}{1000}$ uL (Cat. numb. 951032905). Samples were dried using a rotary evaporator (GeneVac, low BP at 45 °C for 40 min). The resulting dried lipids were re-dissolved in 100 uL of 100 % EtOH (HPLC grade), transferred into HPLC vials and stored at −20 °C until analysis by LC/MS.
LC-MS/MS was performed on a Vanquish LC (ThermoFisher) coupled to a TSQ Altis triple quadrupole mass spectrometer (Thermo Fisher Scientific). Samples were kept at in the autosampler at 4 °C and 15 μL was injected on onto column (50 × 2.1 mm, Kinetex 2.6 μm XB-X18 100 A) at 45 °C, and CoQ8, CoQ9 and CoQ10 were separated by gradient elution using mobile phase A (2.5 mM ammonium formate in 95:5 methanol:isopropanol) and mobile phase B (2.5 mM ammonium formate in $100\%$ isopropanol) at 0.8 mL/min. An initial concentration of 0 % B was held for 1 min before increasing to 45 % B over 1 min and held for 1 min, before decreasing back to 0 % B over 0.5 min and column re-equilibrated over 1.5 min. Under these conditions, CoQ8 eluted at 1.0 min, CoQ9 at 1.6 min and CoQ10 at 2.0 min. Eluent was then directed into the QqQ with a source voltage of 3.5 kV, sheath gas set to 2, auxiliary gas set to 2, and a transfer capillary temperature of 350 °C Ammonium adducts of each of the analytes were detected by SRM with Q1 and Q3 resolution set to 0.7 FWHM with the following parameters: [CoQ8+NH4]+, m/z 744.9 ® 197.1 with collision energy 32.76; [CoQ9+NH4]+, m/z 812.9®197.1 with collision energy 32.76; [CoQ9H2+NH4]+, m/z 814.9®197.1 with collision energy 36.4; and [CoQ10+NH4]+, m/z 880.9 ®197.1 with collision energy 32.76. CoQ9 and CoQ10 areas were normalised to the internal standard CoQ8 levels (20 ng/mL). CoQ9 and CoQ10 were quantified against external standard curves generated from authentic commercial standards obtained from Sigma Aldrich (USA).
## Mitochondrial isolation
Mitochondrial isolation from cultured L6-myotubes and was performed as described elsewhere93,94. Briefly, cells were homogenised in an ice-cold mitochondrial isolation buffer (5 mM HEPES, 0.5 mM EGTA, 200 mM mannitol and 0.1 % BSA, pH 7.4 containing protease inhibitors) using a Cell Homogenizer with 18-micron ball. Cells were passed through the Cell Homogenizer 10 times using 1 mL syringe. Cell Homogenizer was equilibrated with 1 mL of ice-cold isolation buffer prior to the experiment. Homogenates were centrifuged at 700 g for 10 min and the supernatant centrifuged at 10,300 g for 10 min to generate the crude mitochondrial pellet. The 10,300 g pellet was resuspended in 1 mL of isolation buffer and transferred into a polycarbonate tube containing 7.9 mL of $18\%$ Percoll in the homogenization buffer and centrifuged at 95,000 g at 4 °C for 30 min. The mitochondrial pellet was collected and diluted in a homogenization buffer (1 mL) and centrifuged at 10,000 g for 10 min at 4 °C. The supernatant was discarded, and the pellet was washed with a homogenization buffer without BSA followed by protein quantification with BCA protein assay.
Mitochondria from adult skeletal muscle (from mixed hindlimb muscle) were isolated by differential centrifugation as described previously95. Briefly, muscle was diced in CP-1 medium (100 mM KCl, 50 mM Tris/HCl, pH 7.4, and 2 mM EGTA), digested on ice for 3 min in CP-2 medium [CP-1, to which was added $0.5\%$ (w/v) BSA, 5 mM MgCl2, 1 mM ATP and 2.45 units ml–1 Protease Type VIII (Sigma P 5380)] and homogenised using an ultra-turrax homogenizer. The homogenate was spun for 5 min at 500 g and 4°C. The resulting supernatant was subjected to a high-speed spin (10,600 g, 10 min, 4°C) and the mitochondrial pellet was resuspended in CP-1. The 10,600 g spin cycle was repeated, the supernatant removed and the mitochondrial pellet snapped frozen.
## Western Blotting
After insulin stimulation or mitochondrial isolation, samples were tip sonicated in $2\%$ SDS-RIPA. Insoluble material was removed by centrifugation at 21,000 g × 10 min. Protein concentration was determined by bicinchoninic acid method (Thermo Scientific). 10 μg of protein was resolved by SDS-PAGE and transferred to PDVF membranes. Membranes were blocked in Tris-buffered saline (TBS) 4 % skim milk for 30 min at room temperature, followed by primary antibody incubation (detailed antibody is provided in “Key Resource Table”). Membranes were washed in TBS $0.1\%$ tween (TBS-T) and incubated with appropriate secondary antibodies (IRDye700- or IRDye800-conjugated) in TBS-T $2\%$ skim milk for 45 min at room temperature. Images were obtained by using 700- or 800-nm channels using Odyssey IR imager. Densitometry analysis of immunoblots was performed using Image Studio Lite (version 5.2). Uncropped Western blots are provided in Supplementary Material.
## Seahorse extracellular flux analyses
Mitochondrial respiration (JO2) of intact cells were measured using an XF HS mini Analyser Extracellular Flux Analyzer (Seahorse Bioscience, Copenhagen, Denmark). L6 myoblasts were seeded and differentiated in Seahorse XFp culture plates coated with matrigel and assayed after incubation at 37°C without CO2 for 1 hour. Prior to the assay, cells were washed 3 times with PBS, once with bicarbonate-free DMEM buffered with 30 mM Na-HEPES, pH 7.4 (DMEM/HEPES), and then incubated in DMEM/HEPES supplemented with $0.2\%$ (w/v) BSA, 25 mM glucose, 1 mM GlutaMAX and 1 mM glutamine (Media B), for 1.5 h in a non-CO2 incubator at 37 °C. During the assay, respiration was assayed with mix/wait/read cycles of $\frac{2}{0}$/2 min for L6 myotubes. Following assessment of basal respiration, the following compounds (final concentrations in parentheses) were injected sequentially: oligoymcin (10 μg/ml), BAM15 (10 mM), rotenone/antimycin A (5 μM / 10 μM). All of these reagents were obtained from Sigma-Aldrich. basal (baseline - Ant./Rot), ATP-linked respiration (determined by basal – oligomycin), maximal respiration (calculated by FCCP – AntA/Rot) and non-mitochondrial respiration (equal to AntA/Rot) was determined as previously described96. Protein concentration was determined immediately after the assy and data are presented as O2/min. Complex specific activity in permeabilized cells were performed according to97. The cells were seeded and the media was changed to a buffer consisting of 70 mM sucrose, 220 mM Mannitol, 10 mM KH2PO4, 5 mM MgCl2, 2 mM Hepes (pH 7.2), 1 mM EGTA, and $0.4\%$ BSA. Then, flux measurements began after taking three baseline measurements. The cells were permeabilized by adding digitonin (1 nM) and 1 mM ADP, followed by injecting respiratory complex substrates or ADP only (complex I, glutamate/malate (5 mM/2.5 mM); complex II, succinate/rotenone (10 mM/1 μM); complex III, and complex IV, N,N,N,N-tetramethyl-p-phenylenediamine/ascorbate (0.5 mM/2 mM)). Subsequently, oligomycin (1 μg/ml) and respective complex inhibitors were added (complex I, 1 μM rotenone; complexes II and III, 20 μM antimycin A; complex IV, 20 mM sodium azide). Wells where cells detached from the plate during the assay were excluded from the analysis.
## Mitochondrial membrane potential
Mitochondrial membrane potential was measured by loading cells with 20nM tetramethylrhodamine, ethyl ester (TMRM+, Life Technologies) for 30 min at 37°C plus mitotracker deep Red (MTDR). MTDR was used to normalize the fluorescence among the different mitochondrial populations as previously reported96. TMRM+ fluorescence was detected using the excitation-emission λ545–$\frac{580}{590}$ nm and MTDR was detected using an ex/em ~$\frac{644}{665}$ nm using confocal microscopy. The mitochondrial membrane potential was evaluated as raw fluorescence intensity of background-corrected images.
## Mitochondrial oxidative stress
MitoSOX Red was administered as described by the manufacturer (Molecular Probes); At the end of the induction period cells were washed twice with PBS and incubated with 0.5 μM MitoSOX Red for 30 min plus mitotracker deep Red (MTDR). MTDR was used to normalize the fluorescence among the different mitochondrial populations as previously reported. Cells were cultured in low-absorbance, black-walled 96-well plates. After MitoSOX treatment cells were quickly washed with PBS and fluorescence was detected on a confocal microscope. MitoSOX fluorescence was detected using the excitation-emission λ$\frac{396}{610}$ nm and MTDR was detected using an ex/em ~$\frac{644}{665}$ nm using confocal microscopy.
## RT-PCR
RNA was extracted by addition of TRIzol (Thermo-fisher) followed by addition of 0.1x volume of 1-bromo-3-chloropropane. Samples were centrifuged at 13,000xg for 15 minutes for phase separation. The clear phase was transferred to a fresh tube and an equal volume of isopropanol was added. Samples were centrifuged at 13,000xg for 10 minutes to precipitate RNA. RNA was washed three times in $70\%$ ethanol with centrifugation. RNA was resuspended in DEPC treated water and quantified on a NanoDrop 2000 (Thermo Scientific). RNA was reverse transcribed to cDNA using PrimeScript Reverse Transcriptase(Takara) as per the manufacturer’s instructions. qPCR was performed using FastStart SYBR Green MasterMix (2x) (Biorad) as per the manufacturer’s instructions. All primers sequence van be found in sup. Fig. 8
## Statistical analysis
Data are presented as mean ± S.E.M. Statistical tests were performed using GraphPad Prism version 9. HA-GLUT4 and Western blot assay was analysed by using Kruskal-Wallis with Dunn’s multiple comparisons test. CoQ and ceramide abundance were analysed with ordinary one-Way ANOVA and Dunnett’s multiple comparison test. Finally, for comparison of two groups (CoQ and Ceramides levels in mice) we use Student’s t-test. Significant effects were defined as $p \leq 0.05$ by these tests as reported in the Figures.
## References
1. Hill MM, Clark SF, Tucker DF, Birnbaum MJ, James DE, Macaulay SL. **A role for protein kinase Bbeta/Akt2 in insulin-stimulated GLUT4 translocation in adipocytes**. *Mol Cell Biol* (1999) **19** 7771-7781. PMID: 10523666
2. Cong LN, Chen H, Li Y, Zhou L, McGibbon MA, Taylor SI. **Physiological role of Akt in insulin-stimulated translocation of GLUT4 in transfected rat adipose cells**. *Mol Endocrinol* (1997) **11** 1881-1890. PMID: 9415393
3. James DE, Stöckli J, Birnbaum MJ. **The aetiology and molecular landscape of insulin resistance**. *Nat Rev Mol Cell Biol* (2021) **22** 751-771. PMID: 34285405
4. Riehle C. **Abel ED. Insulin Signaling and Heart Failure**. *Circ Res* (2016) **118** 1151-1169. PMID: 27034277
5. Leitner BP, Siebel S, Akingbesote ND, Zhang X, Perry RJ. **Insulin and cancer: a tangled web**. *Biochem J* (2022) **479** 583-607. PMID: 35244142
6. Fazakerley DJ, Chaudhuri R, Yang P, Maghzal GJ, Thomas KC, Krycer JR. **Mitochondrial CoQ deficiency is a common driver of mitochondrial oxidants and insulin resistance**. *Elife* (2018) **7**. DOI: 10.7554/eLife.32111
7. Holland WL, Summers SA. **Sphingolipids, insulin resistance, and metabolic disease: new insights from in vivo manipulation of sphingolipid metabolism**. *Endocr Rev* (2008) **29** 381-402. PMID: 18451260
8. Anderson EJ, Lustig ME, Boyle KE, Woodlief TL, Kane DA, Lin C-T. **Mitochondrial H2O2 emission and cellular redox state link excess fat intake to insulin resistance in both rodents and humans**. *J Clin Invest* (2009) **119** 573-581. PMID: 19188683
9. Hoehn KL, Salmon AB, Hohnen-Behrens C, Turner N, Hoy AJ, Maghzal GJ. **Insulin resistance is a cellular antioxidant defense mechanism**. *Proc Natl Acad Sci U S A* (2009) **106** 17787-17792. PMID: 19805130
10. Fazakerley DJ, Minard AY, Krycer JR, Thomas KC, Stöckli J, Harney DJ. **Mitochondrial oxidative stress causes insulin resistance without disrupting oxidative phosphorylation**. *J Biol Chem* (2018) **293** 7315-7328. PMID: 29599292
11. Hatefi Y, Haavik AG, Fowler LR, Griffiths DE. **Studies on the electron transfer system. XLII. Reconstitution of the electron transfer system**. *J Biol Chem* (1962) **237** 2661-2669. PMID: 13905326
12. Frerman FE. **Acyl-CoA dehydrogenases, electron transfer flavoprotein and electron transfer flavoprotein dehydrogenase**. *Biochem Soc Trans* (1988) **16** 416-418. PMID: 3053288
13. Jones ME. **Pyrimidine nucleotide biosynthesis in animals: genes, enzymes, and regulation of UMP biosynthesis**. *Annu Rev Biochem* (1980) **49** 253-279. PMID: 6105839
14. Montini G, Malaventura C, Salviati L. **Early coenzyme Q10 supplementation in primary coenzyme Q10 deficiency**. *N Engl J Med* (2008) **358** 2849-2850. PMID: 18579827
15. Kühl I, Miranda M, Atanassov I, Kuznetsova I, Hinze Y, Mourier A. **Transcriptomic and proteomic landscape of mitochondrial dysfunction reveals secondary coenzyme Q deficiency in mammals**. *Elife* (2017) **6**. DOI: 10.7554/eLife.30952
16. Calvo E, Cogliati S, Hernansanz-Agustín P, Loureiro-López M, Guarás A, Casuso RA. **Functional role of respiratory supercomplexes in mice: SCAF1 relevance and segmentation of the Qpool**. *Sci Adv* (2020) **6**
17. Ates O, Bilen H, Keles S, Alp HH, Keleş MS, Yıldırım K. **Plasma coenzyme Q10 levels in type 2 diabetic patients with retinopathy**. *Int J Ophthalmol* (2013) **6** 675-679. PMID: 24195048
18. El-ghoroury EA, Raslan HM, Badawy EA, El-Saaid GS, Agybi MH, Siam I. **Malondialdehyde and coenzyme Q10 in platelets and serum in type 2 diabetes mellitus: correlation with glycemic control**. *Blood Coagul Fibrinolysis* (2009) **20** 248-251. PMID: 19530339
19. Zozina VI, Covantev S, Goroshko OA, Krasnykh LM, Kukes VG. **Coenzyme Q10 in Cardiovascular and Metabolic Diseases: Current State of the Problem**. *Curr Cardiol Rev* (2018) **14** 164-174. PMID: 29663894
20. de Barcelos IP, Haas RH. **CoQ10 and Aging**. *Biology* (2019) **8**. DOI: 10.3390/biology8020028
21. Højlund K, Yi Z, Lefort N, Langlais P, Bowen B, Levin K. **Human ATP synthase beta is phosphorylated at multiple sites and shows abnormal phosphorylation at specific sites in insulin-resistant muscle**. *Diabetologia* (2010) **53** 541-551. PMID: 20012595
22. Diaz-Vegas A, Sanchez-Aguilera P, Krycer JR, Morales PE, Monsalves-Alvarez M, Cifuentes M. **Is Mitochondrial Dysfunction a Common Root of Noncommunicable Chronic Diseases?**. *Endocr Rev* (2020) **41**. DOI: 10.1210/endrev/bnaa005
23. Chaurasia B, Summers SA. **Ceramides in Metabolism: Key Lipotoxic Players**. *Annu Rev Physiol* (2021) **83** 303-330. PMID: 33158378
24. Summers SA, Garza LA, Zhou H, Birnbaum MJ. **Regulation of insulin-stimulated glucose transporter GLUT4 translocation and Akt kinase activity by ceramide**. *Mol Cell Biol* (1998) **18** 5457-5464. PMID: 9710629
25. Schubert KM, Scheid MP, Duronio V. **Ceramide inhibits protein kinase B/Akt by promoting dephosphorylation of serine 473**. *J Biol Chem* (2000) **275** 13330-13335. PMID: 10788440
26. Powell DJ, Hajduch E, Kular G, Hundal HS. **Ceramide disables 3-phosphoinositide binding to the pleckstrin homology domain of protein kinase B (PKB)/Akt by a PKCzeta-dependent mechanism**. *Mol Cell Biol* (2003) **23** 7794-7808. PMID: 14560023
27. Kono T, Barham FW. **The relationship between the insulin-binding capacity of fat cells and the cellular response to insulin. Studies with intact and trypsin-treated fat cells**. *J Biol Chem* (1971) **246** 6210-6216. PMID: 5127426
28. Hoehn KL, Hohnen-Behrens C, Cederberg A, Wu LE, Turner N, Yuasa T. **IRS1-independent defects define major nodes of insulin resistance**. *Cell Metab* (2008) **7** 421-433. PMID: 18460333
29. Di Paola M, Cocco T, Lorusso M. **Ceramide interaction with the respiratory chain of heart mitochondria**. *Biochemistry* (2000) **39** 6660-6668. PMID: 10828984
30. Perreault L, Newsom SA, Strauss A, Kerege A, Kahn DE, Harrison KA. **Intracellular localization of diacylglycerols and sphingolipids influences insulin sensitivity and mitochondrial function in human skeletal muscle**. *JCI Insight* (2018) **3**. DOI: 10.1172/jci.insight.96805
31. Hammerschmidt P, Ostkotte D, Nolte H, Gerl MJ, Jais A, Brunner HL. **CerS6-Derived Sphingolipids Interact with Mff and Promote Mitochondrial Fragmentation in Obesity**. *Cell* (2019) **177** 1536-1552.e23. PMID: 31150623
32. Ertunc ME, Hotamisligil GS. **Lipid signaling and lipotoxicity in metaflammation: indications for metabolic disease pathogenesis and treatment**. *J Lipid Res* (2016) **57** 2099-2114. PMID: 27330055
33. Forsman U, Sjöberg M, Turunen M, Sindelar PJ. **4-Nitrobenzoate inhibits coenzyme Q biosynthesis in mammalian cell cultures**. *Nat Chem Biol* (2010) **6** 515-517. PMID: 20526342
34. Fernández-Ayala DJ, Martín SF, Barroso MP, Gómez-Díaz C, Villalba JM, Rodríguez-Aguilera JC. **Coenzyme Q protects cells against serum withdrawal-induced apoptosis by inhibition of ceramide release and caspase-3 activation**. *Antioxid Redox Signal* (2000) **2** 263-275. PMID: 11229531
35. Koves TR, Ussher JR, Noland RC, Slentz D, Mosedale M, Ilkayeva O. **Mitochondrial overload and incomplete fatty acid oxidation contribute to skeletal muscle insulin resistance**. *Cell Metab* (2008) **7** 45-56. PMID: 18177724
36. Bruce CR, Hoy AJ, Turner N, Watt MJ, Allen TL, Carpenter K. **Overexpression of carnitine palmitoyltransferase-1 in skeletal muscle is sufficient to enhance fatty acid oxidation and improve high-fat diet-induced insulin resistance**. *Diabetes* (2009) **58** 550-558. PMID: 19073774
37. Sebastián D, Herrero L, Serra D, Asins G, Hegardt FG. **CPT I overexpression protects L6E9 muscle cells from fatty acid-induced insulin resistance**. *Am J Physiol Endocrinol Metab* (2007) **292** E677-86. PMID: 17062841
38. Perdomo G, Commerford SR, Richard A-MT, Adams SH, Corkey BE, O’Doherty RM. **Increased beta-oxidation in muscle cells enhances insulin-stimulated glucose metabolism and protects against fatty acid-induced insulin resistance despite intramyocellular lipid accumulation**. *J Biol Chem* (2004) **279** 27177-27186. PMID: 15105415
39. Fisher-Wellman KH, Hagen JT, Neufer PD, Kassai M, Cabot MC. **On the nature of ceramide-mitochondria interactions - Dissection using comprehensive mitochondrial phenotyping**. *Cell Signal* (2021) **78** 109838. PMID: 33212155
40. Das AT, Tenenbaum L, Berkhout B. **Tet-On Systems For Doxycycline-inducible Gene Expression**. *Curr Gene Ther* (2016) **16** 156-167. PMID: 27216914
41. Wu BX, Rajagopalan V, Roddy PL, Clarke CJ, Hannun YA. **Identification and characterization of murine mitochondria-associated neutral sphingomyelinase (MA-nSMase), the mammalian sphingomyelin phosphodiesterase 5**. *J Biol Chem* (2010) **285** 17993-18002. PMID: 20378533
42. Bienias K, Fiedorowicz A, Sadowska A, Prokopiuk S, Car H. **Regulation of sphingomyelin metabolism**. *Pharmacol Rep* (2016) **68** 570-581. PMID: 26940196
43. Feng S, Harayama T, Montessuit S, David FP, Winssinger N, Martinou J-C. **Mitochondria-specific photoactivation to monitor local sphingosine metabolism and function**. *Elife* (2018) **7**. DOI: 10.7554/eLife.34555
44. Li CM, Park JH, He X, Levy B, Chen F, Arai K. **The human acid ceramidase gene (ASAH): structure, chromosomal location, mutation analysis, and expression**. *Genomics* (1999) **62** 223-231. PMID: 10610716
45. Turner N, Lim XY, Toop HD, Osborne B, Brandon AE, Taylor EN. **A selective inhibitor of ceramide synthase 1 reveals a novel role in fat metabolism**. *Nat Commun* (2018) **9** 3165. PMID: 30131496
46. Rath S, Sharma R, Gupta R, Ast T, Chan C, Durham TJ. **MitoCarta3.0: an updated mitochondrial proteome now with sub-organelle localization and pathway annotations**. *Nucleic Acids Res* (2021) **49** D1541-D1547. PMID: 33174596
47. Letts JA, Sazanov LA. **Clarifying the supercomplex: the higher-order organization of the mitochondrial electron transport chain**. *Nat Struct Mol Biol* (2017) **24** 800-808. PMID: 28981073
48. Zhu J, Vinothkumar KR, Hirst J. **Structure of mammalian respiratory complex I**. *Nature* (2016) **536** 354-358. PMID: 27509854
49. Brand MD, Nicholls DG. **Assessing mitochondrial dysfunction in cells**. *Biochem J* (2011) **435** 297-312. PMID: 21726199
50. Krycer JR, Elkington SD, Diaz-Vegas A, Cooke KC, Burchfield JG, Fisher-Wellman KH. **Mitochondrial oxidants, but not respiration, are sensitive to glucose in adipocytes**. *J Biol Chem* (2020) **295** 99-110. PMID: 31744882
51. Sánchez-Aguilera P, Diaz-Vegas A, Campos C, Quinteros-Waltemath O, Cerda-Kohler H, Barrientos G. **Role of ABCA1 on membrane cholesterol content, insulin-dependent Akt phosphorylation and glucose uptake in adult skeletal muscle fibers from mice**. *Biochim Biophys Acta Mol Cell Biol Lipids* (2018) **1863** 1469-1477. PMID: 30254016
52. Rosales-Soto G, Diaz-Vegas A, Casas M, Contreras-Ferrat A, Jaimovich E. **Fibroblast growth factor-21 potentiates glucose transport in skeletal muscle fibers**. *J Mol Endocrinol* (2020). DOI: 10.1530/JME-19-0210
53. Valladares D, Utreras-Mendoza Y, Campos C, Morales C, Diaz-Vegas A, Contreras-Ferrat A. **IP3 receptor blockade restores autophagy and mitochondrial function in skeletal muscle fibers of dystrophic mice**. *Biochim Biophys Acta Mol Basis Dis* (2018) **1864** 3685-3695. PMID: 30251688
54. Chalfant CE, Rathman K, Pinkerman RL, Wood RE, Obeid LM, Ogretmen B. **De novo ceramide regulates the alternative splicing of caspase 9 and Bcl-x in A549 lung adenocarcinoma cells. Dependence on protein phosphatase-1**. *J Biol Chem* (2002) **277** 12587-12595. PMID: 11801602
55. Tesfay L, Paul BT, Konstorum A, Deng Z, Cox AO, Lee J. **Stearoyl-CoA Desaturase 1 Protects Ovarian Cancer Cells from Ferroptotic Cell Death**. *Cancer Res* (2019) **79** 5355-5366. PMID: 31270077
56. Morad SAF, Levin JC, Shanmugavelandy SS, Kester M, Fabrias G, Bedia C. **Ceramide--antiestrogen nanoliposomal combinations--novel impact of hormonal therapy in hormone-insensitive breast cancer**. *Mol Cancer Ther* (2012) **11** 2352-2361. PMID: 22962326
57. Schägger H, Pfeiffer K. **Supercomplexes in the respiratory chains of yeast and mammalian mitochondria**. *EMBO J* (2000) **19** 1777-1783. PMID: 10775262
58. Lapuente-Brun E, Moreno-Loshuertos R, Acín-Pérez R, Latorre-Pellicer A, Colás C, Balsa E. **Supercomplex assembly determines electron flux in the mitochondrial electron transport chain**. *Science* (2013) **340** 1567-1570. PMID: 23812712
59. Maranzana E, Barbero G, Falasca AI, Lenaz G, Genova ML. **Mitochondrial respiratory supercomplex association limits production of reactive oxygen species from complex I**. *Antioxid Redox Signal* (2013) **19** 1469-1480. PMID: 23581604
60. Guarás A, Perales-Clemente E, Calvo E, Acín-Pérez R, Loureiro-Lopez M, Pujol C. **The CoQH2/CoQ Ratio Serves as a Sensor of Respiratory Chain Efficiency**. *Cell Rep* (2016) **15** 197-209. PMID: 27052170
61. Gudz TI, Tserng KY, Hoppel CL. **Direct inhibition of mitochondrial respiratory chain complex III by cell-permeable ceramide**. *J Biol Chem* (1997) **272** 24154-24158. PMID: 9305864
62. Pinto SN, Silva LC, Futerman AH, Prieto M. **Effect of ceramide structure on membrane biophysical properties: the role of acyl chain length and unsaturation**. *Biochim Biophys Acta* (2011) **1808** 2753-2760. PMID: 21835161
63. Wu M, Gu J, Guo R, Huang Y, Yang M. **Structure of Mammalian Respiratory Supercomplex I1III2IV1**. *Cell* (2016) **167** 1598-1609.e10. PMID: 27912063
64. Mileykovskaya E, Dowhan W. **Cardiolipin-dependent formation of mitochondrial respiratory supercomplexes**. *Chem Phys Lipids* (2014) **179** 42-48. PMID: 24220496
65. Babiychuk EB, Atanassoff AP, Monastyrskaya K, Brandenberger C, Studer D, Allemann C. **The targeting of plasmalemmal ceramide to mitochondria during apoptosis**. *PLoS One* (2011) **6** e23706. PMID: 21886813
66. Babiychuk EB, Monastyrskaya K, Draeger A. **Fluorescent annexin A1 reveals dynamics of ceramide platforms in living cells**. *Traffic* (2008) **9** 1757-1775. PMID: 18694456
67. Bionda C, Portoukalian J, Schmitt D, Rodriguez-Lafrasse C, Ardail D. **Subcellular compartmentalization of ceramide metabolism: MAM (mitochondria-associated membrane) and/or mitochondria?**. *Biochem J* (2004) **382** 527-533. PMID: 15144238
68. Novgorodov SA, Wu BX, Gudz TI, Bielawski J, Ovchinnikova TV, Hannun YA. **Novel pathway of ceramide production in mitochondria: thioesterase and neutral ceramidase produce ceramide from sphingosine and acyl-CoA**. *J Biol Chem* (2011) **286** 25352-25362. PMID: 21613224
69. Birbes H, El Bawab S, Hannun YA, Obeid LM. **Selective hydrolysis of a mitochondrial pool of sphingomyelin induces apoptosis**. *FASEB J* (2001) **15** 2669-2679. PMID: 11726543
70. Yu J, Novgorodov SA, Chudakova D, Zhu H, Bielawska A, Bielawski J. **JNK3 signaling pathway activates ceramide synthase leading to mitochondrial dysfunction**. *J Biol Chem* (2007) **282** 25940-25949. PMID: 17609208
71. García-Ruiz C, Colell A, Marí M, Morales A, Fernández-Checa JC. **Direct effect of ceramide on the mitochondrial electron transport chain leads to generation of reactive oxygen species. Role of mitochondrial glutathione**. *J Biol Chem* (1997) **272** 11369-11377. PMID: 9111045
72. Turpin-Nolan SM, Hammerschmidt P, Chen W, Jais A, Timper K, Awazawa M. **CerS1-Derived C18:0 Ceramide in Skeletal Muscle Promotes Obesity-Induced Insulin Resistance**. *Cell Rep* (2019) **26** 1-10.e7. PMID: 30605666
73. Oleinik N, Kim J, Roth BM, Selvam SP, Gooz M, Johnson RH. **Mitochondrial protein import is regulated by p17/PERMIT to mediate lipid metabolism and cellular stress**. *Sci Adv* (2019) **5**
74. Bonnard C, Durand A, Peyrol S, Chanseaume E, Chauvin M-A, Morio B. **Mitochondrial dysfunction results from oxidative stress in the skeletal muscle of diet-induced insulin-resistant mice**. *J Clin Invest* (2008) **118** 789-800. PMID: 18188455
75. Picard M, Shirihai OS. **Mitochondrial signal transduction**. *Cell Metab* (2022) **34** 1620-1653. PMID: 36323233
76. Zorov DB, Juhaszova M, Sollott SJ. **Mitochondrial reactive oxygen species (ROS) and ROS-induced ROS release**. *Physiol Rev* (2014) **94** 909-950. PMID: 24987008
77. Siskind LJ, Kolesnick RN, Colombini M. **Ceramide channels increase the permeability of the mitochondrial outer membrane to small proteins**. *J Biol Chem* (2002) **277** 26796-26803. PMID: 12006562
78. Taddeo EP, Laker RC, Breen DS, Akhtar YN, Kenwood BM, Liao JA. **Opening of the mitochondrial permeability transition pore links mitochondrial dysfunction to insulin resistance in skeletal muscle**. *Mol Metab* (2014) **3** 124-134. PMID: 24634818
79. Cho J, Zhang Y, Park S-Y, Joseph A-M, Han C, Park H-J. **Mitochondrial ATP transporter depletion protects mice against liver steatosis and insulin resistance**. *Nat Commun* (2017) **8** 14477. PMID: 28205519
80. Walter L, Nogueira V, Leverve X, Heitz MP, Bernardi P, Fontaine E. **Three classes of ubiquinone analogs regulate the mitochondrial permeability transition pore through a common site**. *J Biol Chem* (2000) **275** 29521-29527. PMID: 10889201
81. Galgani JE, Fernández-Verdejo R. **Pathophysiological role of metabolic flexibility on metabolic health**. *Obes Rev* (2021) **22** e13131. PMID: 32815226
82. Hernansanz-Agustín P, Enríquez JA. **Functional segmentation of CoQ and cyt c pools by respiratory complex superassembly**. *Free Radic Biol Med* (2021) **167** 232-242. PMID: 33722627
83. Yang F, Shen Y, Camp DG, Smith RD. **High-pH reversed-phase chromatography with fraction concatenation for 2D proteomic analysis**. *Expert Rev Proteomics* (2012) **9** 129-134. PMID: 22462785
84. Smyth GK. **Linear models and empirical bayes methods for assessing differential expression in microarray experiments**. *Stat Appl Genet Mol Biol* (2004) **3**
85. Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J. **STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets**. *Nucleic Acids Res* (2019) **47** D607-D613. PMID: 30476243
86. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. **limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res* (2015) **43** e47. PMID: 25605792
87. Langfelder P, Horvath S. **WGCNA: an R package for weighted correlation network analysis**. *BMC Bioinformatics* (2008) **9** 559. PMID: 19114008
88. Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z. **clusterProfiler 4.0: A universal enrichment tool for interpreting omics data**. *Innovation (Camb)* (2021) **2** 100141. PMID: 34557778
89. Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P. **The Reactome Pathway Knowledgebase**. *Nucleic Acids Res* (2018) **46** D649-D655. PMID: 29145629
90. Seal RL, Braschi B, Gray K, Jones TEM, Tweedie S, Haim-Vilmovsky L. *Nucleic Acids Res* (2023) **51** D1003-D1009. PMID: 36243972
91. Govers R, Coster ACF, James DE. **Insulin increases cell surface GLUT4 levels by dose dependently discharging GLUT4 into a cell surface recycling pathway**. *Mol Cell Biol* (2004) **24** 6456-6466. PMID: 15226445
92. Zarini S, Brozinick JT, Zemski Berry KA, Garfield A, Perreault L, Kerege A. **Serum dihydroceramides correlate with insulin sensitivity in humans and decrease insulin sensitivity in vitro**. *J Lipid Res* (2022) **63** 100270. PMID: 36030929
93. Bui M, Gilady SY, Fitzsimmons REB, Benson MD, Lynes EM, Gesson K. **Rab32 modulates apoptosis onset and mitochondria-associated membrane (MAM) properties**. *J Biol Chem* (2010) **285** 31590-31602. PMID: 20670942
94. Frezza C, Cipolat S, Scorrano L. **Organelle isolation: functional mitochondria from mouse liver, muscle and cultured fibroblasts**. *Nat Protoc* (2007) **2** 287-295. PMID: 17406588
95. Montgomery MK, Osborne B, Brandon AE, O’Reilly L, Fiveash CE, Brown SHJ. **Regulation of mitochondrial metabolism in murine skeletal muscle by the medium-chain fatty acid receptor Gpr84**. *FASEB J* (2019) **33** 12264-12276. PMID: 31415180
96. Díaz-Vegas AR, Cordova A, Valladares D, Llanos P, Hidalgo C, Gherardi G. **Mitochondrial Calcium Increase Induced by RyR1 and IP3R Channel Activation After Membrane Depolarization Regulates Skeletal Muscle Metabolism**. *Front Physiol* (2018) **9** 791. PMID: 29988564
97. Kory N, Uit de Bos J, van der Rijt S, Jankovic N, Güra M, Arp N. **MCART1/SLC25A51 is required for mitochondrial NAD transport**. *Sci Adv* (2020) **6**. DOI: 10.1126/sciadv.abe5310
|
---
title: Leveraging Base Pair Mammalian Constraint to Understand Genetic Variation and
Human Disease
authors:
- Patrick F. Sullivan
- Jennifer R. S. Meadows
- Steven Gazal
- BaDoi N. Phan
- Xue Li
- Diane P. Genereux
- Michael X. Dong
- Matteo Bianchi
- Gregory Andrews
- Sharadha Sakthikumar
- Jessika Nordin
- Ananya Roy
- Matthew J. Christmas
- Voichita D. Marinescu
- Ola Wallerman
- James R. Xue
- Yun Li
- Shuyang Yao
- Quan Sun
- Jin Szatkiewicz
- Jia Wen
- Laura M. Huckins
- Alyssa J. Lawler
- Kathleen C. Keough
- Zhili Zheng
- Jian Zeng
- Naomi R. Wray
- Jessica Johnson
- Jiawen Chen
- Benedict Paten
- Steven K. Reilly
- Graham M. Hughes
- Zhiping Weng
- Katherine S. Pollard
- Andreas R. Pfenning
- Karin Forsberg-Nilsson
- Elinor K. Karlsson
- Kerstin Lindblad-Toh
journal: bioRxiv
year: 2023
pmcid: PMC10028973
doi: 10.1101/2023.03.10.531987
license: CC BY 4.0
---
# Leveraging Base Pair Mammalian Constraint to Understand Genetic Variation and Human Disease
## Abstract
Although thousands of genomic regions have been associated with heritable human diseases, attempts to elucidate biological mechanisms are impeded by a general inability to discern which genomic positions are functionally important. Evolutionary constraint is a powerful predictor of function that is agnostic to cell type or disease mechanism. Here, single base phyloP scores from the whole genome alignment of 240 placental mammals identified $3.5\%$ of the human genome as significantly constrained, and likely functional. We compared these scores to large-scale genome annotation, genome-wide association studies (GWAS), copy number variation, clinical genetics findings, and cancer data sets. Evolutionarily constrained positions are enriched for variants explaining common disease heritability (more than any other functional annotation). Our results improve variant annotation but also highlight that the regulatory landscape of the human genome still needs to be further explored and linked to disease.
## Introduction
In the past 15 years, increasingly larger genomic studies have delivered many novel associations for a wide array of human diseases, disorders, biomarkers, and other traits. Approximately 200K genetic associations have been identified that span the allelic spectrum, from ultra-rare variants in large sequencing datasets to variants common in all humans, in both coding and regulatory regions (see Supplementary Methods, Section 1). Although these associations meet rigorous standards for statistical significance and replicability, their functional importance is generally unknown. Inferring functional importance is crucial to translating the results of rare and common variant association studies into the biological, clinical, and therapeutic knowledge required to understand and treat human disease. Exceptional efforts have been made to annotate the human genome using functional genomics—e.g., ENCODE [1] and GTEx [2]—as well as inferring deleterious effects from allele frequencies and location in coding sequence—e.g., gnomAD [3] and TOPMed [4]. Although these seminal projects greatly expanded knowledge, this “central problem in biology” is unresolved and motivated the NHGRI Impact of Genomic Variation on Function initiative.
Evolutionary constraint is complementary to these efforts. Functional importance is inferred from the signatures of evolution in the human genome: “constraint” indicates genomic positions that have changed more slowly than expected under neutral drift due to purifying selection. A key advantage of constraint lies in its mechanistic agnosticism; a highly constrained base has an impact on some biological process, in some cell, at some life stage (discussed in Supplementary Methods, Section 2). Constraint has been used in efforts to understand the human genome for over 50 years beginning with cross-species protein sequence comparisons. More recently, at the extremes of the allelic spectrum, constraint is often used by clinical geneticists to prioritize potentially causal rare variants [5, 6], and common variants in regions under constraint are highly enriched in genome-wide association study (GWAS) results (7–9). Despite its reported importance, evolutionary constraint is not systematically leveraged in interpreting the function of GWAS loci (10–15).
Our companion paper describes the Zoonomia reference-free alignment of 240 placental mammals spanning ~100 million years of evolution (Companion paper #1, Christmas et al.). The analyses showed the unprecedented informativeness of this alignment at multiple scales: from exceptionally constrained 100 kb bins (e.g., all HOX clusters) to smaller ultra-conserved and human accelerated regions, non-coding regulatory regions, nuances of the genetic code, and specific base positions in binding motifs. These results strongly suggested the utility of constraint as a functional annotation that can be leveraged to deepen our understanding of heritable human diseases. In this paper, we demonstrate the importance of mammalian constraint for connecting genotype to phenotype for human disease.
## Defining constraint.
Placental mammalian constraint was estimated using phyloP scores [16] across 240 species for 2,852,623,265 bases in the human genome (chr1–22, X, Y; Supplementary Methods, Section 3). In our companion paper we estimated that ~$13\%$ of the human genome is under some degree of constraint due to purifying selection; for these disease-focused analyses, we used an empirical subset with the strongest constraint signatures. We defined a base as constrained in mammals if its phyloP score was ≥ 2.27 (FDR 0.05 threshold, 100,651,377 bases or $3.53\%$ of the genome). We defined constraint across 43 primates using a phastCons [17] threshold (≥ 0.961, 101,134,907 bases) selected to match the same fraction of the genome annotated as constrained in mammals. Mammalian and primate constraint overlapped significantly but not fully (Jaccard index 0.30). In Supplementary Methods, Section 4, we describe the properties of constrained genomic positions, from base level to higher order annotations. Briefly, we found that mammalian constrained bases had a marked tendency to cluster (median distance 2 bases) compared to random expectations (median 24 bases), and that specific genomic elements were highly enriched in constrained bases (particularly coding sequence, CDS, as expected) as well as multiple regulatory features (Figs. 1A and S1), and that constraint scores captured nuances of the genetic code (fig. S2).
## Constraint across the allelic spectrum.
Genetic variation is fundamental to heritable human diseases, disorders, and other traits. We thus evaluated the relationship between allele frequency and constraint (Fig. 1B). Using whole genome sequencing data from over 140K humans (TOPMed, v8) [4], we observed an inverse correlation between allele count and phyloP score (rho = −0.07) with stronger correlations in CDS regions and for non-synonymous variants (rho = −0.12 and −0.18, all $P \leq 2.2$×10−308). As expected due to negative selection, common genetic variants were depleted for constrained bases ($1.85\%$ vs. $3.53\%$ expected by chance, $P \leq 2.2$×10−308). However, this relatively high fraction of constrained bases highlights the ability of mammalian constraint to predict deleterious effects across the allele frequency spectrum. To evaluate these relations more formally, genome-wide models contrasting singletons (AC = 1) to common variants (AF ≥ 0.005) found that common variants had lower phyloP scores and a marked increase in CG context (fig. S3, Supplementary Methods, Section 4). Models for CDS SNPs found an inverse association of AC with constraint, and that common SNPs had greater odds of occurring at a C or G base, and tend not to occur in important CDS positions (e.g., codon position 1 or 2, or at bases that could mutate to stop).
## Common constrained SNPs are relevant for human diseases.
We conducted additional analyses of common SNPs (AF ≥ 0.005) as these variants are foundational for GWAS (Supplementary Methods, Section 4). Of these 15,777,878 SNPs in TOPMed, $1.85\%$ ($$n = 291$$,669) are constrained, far less than genome-wide constraint ($3.53\%$). Our modeling showed that constrained SNPs were 22x more likely to occur in CDS bases, 3x more likely to occur in promoters, and ~2x more likely to be a “fine-mapped” eQTL-SNP or to occur in open chromatin or an enhancer.
The strong tendency of these constrained SNPs to occur in CDS was unexpected given that (by definition) these positions are highly constrained in placental mammals and yet variable in humans. We hypothesized that this could occur if selection effects were variable across genes (some generate peptide variability whereas others are highly intolerant of CDS variation). We found that $37.8\%$ of protein-coding (PC) genes had no constrained CDS SNPs and other genes had appreciable fractions (up to $10\%$ of all CDS bases are common SNPs). The top $5\%$ ($$n = 980$$) of genes with the most constrained CDS SNPs have medical relevance (131 have an OMIM entry including multiple neurological disorders) and were strongly enriched for G-protein coupled receptors (GPCR), “druggable” genes (both GPCR and non-GPCR) [18], taste receptors, skin development, and multiple immune processes. These biological processes are at the interface of a mammal and its environment and allow adaptation to an environmental niche. We suggest that many of these genes could be prioritized for gene-environment interactions searches as constrained variants reaching high frequency in human populations are relevant for human diseases.
## Base pair resolution of deleterious effects.
We contrasted constraint scores to metrics used to aid the interpretation of functional variation for human health. First, pathogenic ClinVar [19] variants were significantly skewed to higher phyloP in comparison to benign variants (two-tailed Wilcoxon rank sum test, $P \leq 2.2$×10−16, Fig. 1C), and phyloP scores were strongly associated with the improvement in annotations of variants in ClinVar from 2016 to 2021 (e.g. uncertain to benign or to pathogenic; Supplementary Methods, Section 5). For a second metric, CADD [6], which incorporates evolutionary constraint, we found variant positions with a higher likelihood of deleteriousness were also enriched for constrained phyloP scores (two-tailed Wilcoxon rank sum test, $P \leq 2.2$×10−16, Fig. 1C). A focused analysis of human non-synonymous variants at constrained sites across the mammalian tree using TOGA (Tool to infer Orthologs from Genome Alignments, Companion paper #1, Christmas et al.; Companion paper #10, Kirilenko et al), identified 1,570 genes for which a non-synonymous change resulted in a ClinVar pathogenic or likely pathogenic phenotype in humans (Supplementary Methods, Section 5). For example, the CFTR gene underlying cystic fibrosis [20] showed a high burden of pathogenic compared to benign sites (123 vs. 1 out of 1,585 alignment sites). A further 12,889 genes had identifiable constrained sites, but lacked records of non-synonymous pathogenic alterations (Supplementary Methods, Section 5). Several of these constrained positions, currently lacking ClinVar pathogenic annotations, likely represent novel sources of deleterious variation resulting in a disease state. We tested this by leveraging functionally explored variation in two G-protein coupled receptors, GPR75 [21] and ADRB2 [22], and showed that functionally important SNP or amino acid sites respectively, were marked by higher constraint scores (Supplementary Methods, Section 5). Species alignments at this scale also allow for the identification of potential model systems, those for which a substitution may result in a human disease state, but is otherwise naturally occurring in non-human mammals. We found 697 such sites across 330 genes, including multiple positions in SOD1 (pathogenic sites for amyotrophic lateral sclerosis). These observations open the avenue for natural adaptive variants to inform the development of new therapies for treatment (Supplementary Methods, Section 5).
## Common variation and human diseases and complex traits
GWAS have found that the genetic architecture of human diseases and complex traits is highly polygenic and dominated by common variants with weak effects [10]. Here, we dissected the impact of common variants (defined in this section as AF ≥ 0.05) on this architecture via polygenic analyses of disease SNP-heritability (h2) using stratified LD score regression (S-LDSC) [7, 23, 24] using the results of 63 independent European ancestry GWAS [25] (mean $$n = 314$$K; table S1, Supplemental Methods, Section 6).
## Constraint scores are proportional to common variant SNP-heritability enrichments.
We first validated the relevance of our constraint scores to investigate the role of common variants in human diseases and complex traits. We found that common variants in the highest constraint score percentiles had greater enrichment for GWAS trait associated variants (measured by SNP-h2 enrichment, the proportion of h2 divided by the proportion of SNPs; Fig. 2A and table S2). We observed decreasing but significant enrichments ($P \leq 0.05$/15) for SNPs in the four first percentiles of mammalian constraint scores (phyloP) (in line with $3.53\%$ of the genome bases being considered as constrained using a $5\%$ FDR threshold), and in the first five percentiles of primate (phastCons) constraint scores. We justified the use of different scores to measure constraint in mammals and primates by the fact that phyloP scores were unable to detect single-base constraint in primates due to lack of power and were too noisy to lead to significant h2 enrichment (fig. S4). While both phyloP and phastCons scores performed similarly in heritability analyses, phyloP is superior for having single-base resolution (fig. S4 and additional justification in Supplemental Methods, Section 6).
## Mammal constraint scores are base pair specific.
We evaluated the resolution of constraint scores by estimating SNP-h2 with different distances to a constrained base. First, we confirmed the base pair resolution of mammalian constraint scores by observing that SNPs ~1 bp from a constrained variant were significantly less enriched than constrained SNPs (P ≤ 3.35×10−3) (Fig. 2B and table S3). We also observed log-linear decrease of h2 enrichment as a function of the distance to a constrained base, with significant h2 enrichment up to 100 kb from constrained bases, confirming the larger-scale clustering of constrained bases. Finally, demonstrating the power of a broad placental mammal-wide genome sampling, constraint scores obtained only from primate species have lower resolution (~10 bp, Fig. 2B) as these are based on fewer species [43], from a single mammalian order, and thus less branch length.
## Zoonomia constraint is uniquely informative.
Annotations derived from mammal and primate constrained positions were more informative for human diseases than key functional annotations, including previously published constrained annotations [17, 26, 27] (Fig. 2D and table S4). First, their degrees of enrichment (7.84 ± 0.37 fold for mammals and 11.10 ± 0.40 fold for primates) exceeded those of previously published constraint and key functional annotations, such as non-synonymous coding variants (7.20 ± 0.78 fold) or fine-mapped eQTL-SNPs (4.81 ± 0.31 fold) [28]. Second, in conditional analyses involving 106 annotations analyzed jointly (Supplemental Methods, Section 6), we observed that these constrained annotations were among the most significant ($$P \leq 1.17$$×10−10 for mammals, and $$P \leq 1.19$$×10−53 for primates, respectively), and more significant than previously published constrained annotations (Fig. 2D and table S4).
## Variants at constrained positions are less enriched in blood and immune traits heritability than in other complex traits.
We did not observe disease-specific patterns for our constrained annotations, without any trait exhibiting significantly higher h2 enrichment than the mean calculated for the mammal and primate constrained annotations (fig. S5 and table S5). However, we observed consistently lower h2 enrichments for constrained annotations in a meta-analysis of 11 blood and immune traits, as previously observed [7], but no differential enrichment in 9 brain disorders (Fig. 2C, table S1, and table S6).
## Variants at positions constrained in primates are informative for non-coding common variants.
Surprisingly, SNPs constrained in primates have greater SNP-h2 enrichment than SNPs constrained in mammals (Figs. 2A-C). To investigate, we intersected mammalian and primate constraint information, and observed significantly higher h2 enrichment in SNPs constrained in both mammals and primates (16.52 ± 0.73 fold), compared with constraint only in primates (8.66 ± 0.38 fold), or only in mammals (3.56 ± 0.40 fold) (Fig. 2E and table S7). We verified that these results are mostly driven by the intersection of mammal and primate constrained bases (and not due to the different scoring tests, fig. S6). By stratifying constrained mammalian bases by their primate constraint scores, we found that variants identified as constrained in mammals but not in primates are not significantly enriched in h2, whereas SNPs constrained in primates were significantly enriched regardless of their constraint scores in mammals (fig. S7). These results explain the lower SNP-h2 for constraint in mammals, and demonstrate increased informativeness when combining information from primates and mammals. Interestingly, we observed consistently higher h2 enrichment for SNPs that are constrained in both mammals and primates when stratifying by genomic function (i.e., coding regions, promoters, and enhancers), but that constraint is more informative in primates than in mammals only for non-coding variants (Fig. 2E). Strikingly, we observed that constrained SNPs defined as non-functional (see Supplemental Methods, Section 6) were still enriched in h2 (>2.67 fold with $P \leq 1.22$×10−4, except for SNPs constrained only in mammals or primates; Fig. 2E), emphasizing the informativeness of our constrained annotations to annotate non-coding variants with unknown function.
## Disease effect sizes of common variants at constrained positions differ across human populations.
While our heritability analyses focused on European ancestry GWAS, variant effect sizes differ across human populations, especially for variants with stronger gene-environment interactions [29]. To quantify how effect sizes of constrained common variants differ across populations, we applied S-LDXR [29] on 31 diseases and complex traits with GWAS data from East Asian (mean $$n = 90$$K) and European (mean $$n = 267$$K) populations. Variants at constrained sites in mammals and primates were among the most significantly depleted in squared trans-ancestry genetic correlation ($$P \leq 4.38$$×10−9 and $$P \leq 1.63$$×10−14, the third and most significant investigated annotation, respectively; Fig. 2F and table S8). These results highlight more population-specific causal effect sizes for variants at constrained positions, in line with stronger gene-environment interactions at these loci, and potentially explain how genetic variations at constrained bases could have become common in human populations.
## Strong effect sizes for coding low-frequency variants at constrained positions.
Annotations constrained by purifying selection tend to have low-frequency variants ($0.5\%$ ≤ AF < $5\%$) with larger effect sizes leading to higher enrichment in low-frequency variant h2 compared to common variant h2 [8]. We quantified low-frequency SNP-h2 enrichments of constrained annotations by analyzing 27 well-powered independent UK Biobank traits (same as in [8]; mean $$n = 355$$K; table S9). We observed that constrained annotations had consistently larger low-frequency h2 enrichment than common h2 enrichment, especially for variants at constrained sites in mammals (16.83 ± 0.92 vs. 8.70 ± 0.72 fold; $$P \leq 3.22$$×10−11 for difference) (fig. S8 and table S10) in line with greater effect sizes as allele frequency decreases (Fig. 2G and table S11). This enrichment difference was driven by coding variants at constrained sites in mammals (48.84 ± 3.10 vs. 19.42 ± 1.91 fold; $$P \leq 6.36$$×10−16 for difference); we note that the low-frequency h2 enrichment for these variants was similar to that of non-synonymous variants (40.38 ± 2.40 fold), suggesting that constraint information is as informative as protein change information at the coding level.
In conclusion, we observed that our mammalian constraint scores have unprecedented base pair resolution to investigate common variants in GWAS findings for human complex traits and diseases, are uniquely informative compared to known functional annotations and previously published constraint scores, are even more informative when combined with primate constraint scores, and could be utilized to investigate variants defined as non-functional.
## Zoonomia constraint scores improve functionally-informed fine-mapping analyses.
Based on our heritability results, we expect that our constraint scores will improve functionally-informed fine-mapping of constrained genetic variants associated with common traits. We compared PolyFun [30] fine-mapping results obtained with no annotations (non-functional model), with its default set of annotations (baseline-LF model), and with an augmented baseline-LF annotations containing multiple Zoonomia constrained annotations (baseline-LF+Zoonomia model) on 24 well-powered UK Biobank diseases and complex traits [30, 31] (mean $$n = 440$$K; table S12 and Supplemental Methods, Section 7). We observed significantly ($P \leq 1.00$×10−4) greater posterior inclusion probability (PIP) for variants at constrained sites in mammals and primates when using PolyFun with the baseline-LF+Zoonomia model compared to the non-functional and baseline-LF models (Figs. 3A and 3B). Notably, PolyFun with the baseline-LF+Zoonomia model detected 1,407 variants at constrained sites in mammals fine-mapped with high confidence (PIP > 0.75) across all the UK Biobank traits ($32.80\%$ of high confidence fine-mapped variants), against 732 and 1,216 when using the non-functional and baseline-LF and models, respectively ($24.50\%$ and $29.67\%$ of high confidence fine-mapped variants, respectively) (fig. S9).
## Fine-mapping examples.
We highlight the utility of evolutionary constraint scores in fine-mapping analyses. First, rs1421085 has a causal and experimentally validated association with BMI (the SNP is located in FTO but has regulatory effects on IRX5 and IRX3) [32, 33]; this variant is extremely constrained in mammals (phyloP = 6.31) and primates (phastCons = 1.00), leading to a higher PIP when using the baseline-LF+Zoonomia model (0.84) than when using the non-functional and baseline-LF models (0.13 and 0.58, respectively; Fig. 3C). Interestingly, the fraction of CDS and promoter bases that are constrained for IRX5 (0.79 and 0.58) and IRX3 (0.74 and 0.34) were higher than for FTO (0.61 and 0.23), suggesting that constrained variant in regulatory regions could be more likely to target genes with constrained CDS and/or promoters (see below). Second, rs6914622 is constrained in mammals and primates (phyloP = 2.37 and phastCons = 1.00) and may be causal in hypothyroidism via the baseline-LF+Zoonomia model (PIP = 0.76; Fig. 3D) but not via the non-functional and baseline-LF models (PIP ≤ 0.14). Conversely, the sentinel variant rs9497965 is not evolutionarily constrained but has a notable PIP in the baseline-LF model (PIP ≥ 0.85) but not in the baseline-LF+Zoonomia model (PIP = 0.24). Using epigenetic marks from four thyroid cell types [34] (functional information not in the fine-mapping models), rs6914622 was in an active enhancer in all thyroid cell-types and rs9497965 was inferred as being in an enhancer in only one thyroid cell type (weak transcription and quiescent for the others), suggesting a causal role for rs6914622 over rs9497965. While functional follow-up is necessary, these examples illustrated how Zoonomia constraint scores can significantly impact fine-mapping. One method of functional follow-up, Cell-TACIT, is explored in a companion paper (Companion paper #11, Phan et al.), in which the conservation of human neural cell type-specific open chromatin across mammals is used to improve the fine-mapping of GWAS for brain disorders. Some regulatory elements may not be conserved at the nucleotide level but lie in a cell type regulatory element predicted to be conserved across mammalians. Fine-mapping genetic variants with constraint and Cell-TACIT provide examples of how mammalian genomes can be leveraged to discover nucleotide and regulatory conservation to link variation to function. Finally, as discussed in another companion paper, Human Accelerated Regions can also improve fine-mapping interpretation (Companion paper #8, Keough et al,).
## Measures of constraint can reveal unannotated variants impacting human health.
Due to the challenge of generating functional datasets in all cell-types and cell-states, much of the genome’s regulatory space is still not fully annotated [35]. The high levels of constraint and low levels of variant diversity in UNannotated Intergenic COnstraint RegioNs (UNICORNs, Supplemental Methods, Section 8, Companion paper #1, Christmas et al.) suggest that they are likely of functional importance despite lacking functional annotations (consistent with our observation that non-functional constrained SNPs are enriched in h2, Fig. 2E). While fewer fine-mapped SNPs were located within UNICORNs (833 SNPs) compared to a matched set of random unannotated non-constrained intergenic regions (5,895 SNPs) and to SNPs located elsewhere in the genome (305,599 SNPs), those variants had higher mean PIP scores (0.15 UNICORNs vs 0.05 for the other two regions). This demonstrates that UNICORNs can reveal unannotated variants impacting human health and disease. UNICORNs contain fine-mapped SNPs with significantly higher PIP scores compared to the background sets across multiple traits (linear regression, $P \leq 0.01$ in all cases after correcting for multiple testing; table S13). For example, a 163 bp UNICORN contains rs72782676 with fine-mapping evidence for multiple traits (e.g., eosinophil count, asthma, eczema, respiratory and ENT diseases; AFTOPMed = 0.005; PIP > 0.99 in all GWAS) (Fig. 3E). The nearest gene, GATA3, sits 915 kb upstream, and is a master transcriptional regulator for T Helper 2 lineage commitment [36], and is known to play an important role in inflammatory disease [37, 38]. This UNICORN highlights a strong regulatory candidate for GATA3 in a disease-relevant region currently lacking annotation.
## Predicted variant effect validated at single base resolution.
Massively parallel reporter assays (MPRAs), have been used to rapidly test thousands of genomic variants for their potential regulatory effects on gene expression. While the functional output from these high-throughput methods are useful for localising putative causal alleles, overlaying constraint scores may help further elucidate functional variants (Supplemental Methods, Section 8). To investigate this, we integrated our Zoonomia-derived phyloP scores with > 35,000 assayed variants from existing 3’UTR [39] and eQTL [40] MPRAs. Using the 3’UTR MPRA data to highlight our results, we found that phyloP scores could differentiate between sequence backgrounds with and without regulatory activity, (e.g. across multiple tissues, Neutral vs Active: Polig = 2.32×10−5, Fig. 3F). PhyloP scores further highlighted variants with allele-specific regulatory effects (e.g. Neutral vs Skew: Pbase = 1.4×10−5; Fig. 3G). Additionally, we found that selection on constrained phyloP positions enriched the allele-specific regulatory effects by 1.3 fold (Supplemental Methods, Section 8). Similar trends were observed in promoter and enhancer saturation mutagenesis MPRAs [41]. For example, phyloP constraint was a strong predictor for variant effect within the LDLR promoter (Spearman rho = 0.51), with five of the most constrained sites providing the strongest regulatory effects and also tagging pathogenic ClinVar positions (Fig. 3H). Further, in our companion paper (Companion paper #? CONDEL, Xue et al), we use MPRAs to directly assess the regulatory impacts of bases under high constraint that have been deleted specifically in the human lineage. For many we can precisely identify how the deletions impact transcription factor binding which is well correlated with the observed regulatory changes, linking sequence change to mechanism. We found these human-specific deletions were enriched to overlie psychiatric disease GWAS signals (i.e. Schizophrenia, Bipolar), and discovered 717 deletions with significant species-specific regulatory effects, providing candidates targets that may have contributed to the prevalence of human neurological disorders.
## Evolutionary constraint, protein-coding genes, and human disease
Gene-based measures of evolutionary constraint have an important role in understanding the impact of genetic variation on human disease (e.g., LOEUF) [3]. As detailed in Supplementary Methods, Section 9, we defined 7 measures of gene constraint based on the Zoonomia alignment including fraction of CDS constrained, normalization against 32.13 million CDS bases, a model-based approach adjusting for 12 covariates (codon information, mutational consequences, and positional features), and cross-species amino acid constraint (normalized Shannon entropy). After evaluation, we selected the fraction of constrained CDS bases per gene (fracCdsCons) as a simple measure of gene constraint, given its continuous distribution, low missingness, high correlations with more complex measures of gene constraint, and external validation (Fig. 4A). *These* gene-based constraint metrics are provided in table S14.
Given the complexities of human PC genes, it would be surprising if any one gene metric applies to all genes (e.g., LOEUF and pLI are missing for $10.1\%$ of PC genes). We used an empirical approach to identify gene outliers, and identified 277 genes ($1.43\%$) inaccessible to fracCdsCons (clusters A-B, Fig. 4A; Supplementary Methods, Section 10).
We validated fracCdsCons in several ways (Supplementary Methods, Section 10). First, given its widespread use, we compared fracCdsCons to the inverse-scored LOEUF [3] and found rho = −0.55. This is notable given the markedly different basis of each measure—constraint over ~100 million years of mammalian evolution vs statistical modeling of pLoF counts in human WES catalogs (Supplemental Methods, Section 2): empirical confirmation is an important validator for both measures. We next compared fracCdsCons to external gene sets with established patterns of constraint (similar to the LOEUF validation strategy)[3] and obtained similar patterns between both scores (Figs. 4B and 4C).
Second, we used an empirical approach to cluster genes based on different constrained metrics (Fig. 4A; Supplementary Methods, Section 10; table S14). We identified 277 gene outliers ($1.43\%$) inaccessible to fracCdsCons (clusters A-B), and conducted gene set analyses for 19,109 PC genes (clusters C-E, tables S15 and S16). The $5\%$ most constrained genes ($$n = 955$$, fracCdsCons 0.811–0.975) were strongly enriched in gene sets: basic embryology (stem cell proliferation/differentiation, tube formation, anterior/posterior patterning, endoderm/mesoderm formation); organ morphogenesis (central/peripheral nervous system, connective tissue, ear, epithelium, eye, gastrointestinal tract, heart, kidney, lung, muscle, myeloid, pancreas, skeleton); cell cycle (phase transition, fate, WNT), cell signaling, positive and negative regulatory processes; and pre-/post-synaptic processes (synapse assembly, postsynaptic density, neurotransmitter regulation, synaptic vesicle cycle, modulation of transsynaptic signaling). The $5\%$ least constrained genes ($$n = 956$$, fracCdsCons 0–0.150) were strongly enriched in gene sets: microbial defense response (adaptive immunity, bacteria/virus, cell killing, cytokine/interferon); bitter taste and olfaction; and skin development (keratinization, keratinocyte differentiation, epidermal cell differentiation, and epidermis development). The most constrained genes captured processes fundamental to the making of a mammal and the least constrained genes are central to the adaptive evolution of a mammal to its environment—i.e., the specific microbiota, adaptations of smell and taste to detect mates, prey, predators, and poisons, and adaptations of skin for temperature regulation, camouflage, and defense.
Finally, we evaluated the relevance of mammalian gene constraint to human disease. Fig. S10A shows the relationship of fracCdsCons with multiple human disease annotations. For all comparisons, increasing constraint is correlated with increasing relevance for human disease. Fig. S10B depicts the relation with GTEx gene expression, and greater gene constraint is correlated with greater expression in all tissues. “ Housekeeping” genes that are uniformly expressed across tissues had greater constraint ($P \leq 3$×10−197) and comprised $3.0\%$ of the least constrained decile and $30.5\%$ of the most constrained decile. Finally, we evaluated the impact of common SNPs linked to PC genes in each fracCdsCons decile by estimating their gene h2 enrichment (defined as h2 enrichment for the decile annotation divided by the mean h2 enrichment over all deciles) using S-LDSC on 63 independent GWAS datasets (Supplemental Methods, Section 10). We observed significantly higher gene h2 enrichment for SNPs linked to genes in the most constrained deciles ($$P \leq 6.96$$×10−59; Fig. 4D and table S17). Interestingly, we observed stronger gene h2 enrichment patterns in a meta-analysis of nine brain disorders, and gene h2 enrichment patterns nearly independent of gene constraint in a meta-analysis of 11 blood and immune traits (Fig. 4D and table S17).
## Mammalian constraint is correlated between coding and regulatory elements.
We extended our approach to measure gene constraint on different regulatory features (including promoters, and ENCODE3 distal enhancers linked to their genes using EpiMap [34]), as human diseases and complex traits are predominantly impacted by common regulatory variants. We found substantial correlations of constraint between CDS and the regulatory parts of protein-coding genes, with a higher correlation between CDS and promoter gene constraint ($r = 0.55$) than between CDS and distal enhancer gene constraint ($r = 0.25$) (Figs. 4E-G; gene scores reported in table S18). These correlations are consistent with the idea that if the function of a gene in mammals requires high conservation of protein structure, then its regulatory sequences tend to also be constrained. Interestingly, we observed families of genes with shared constrained patterns (such as HOX genes that have constrained exons, promoters and enhancers), and with distinct constrained patterns (such as defensin beta (DEFB) genes, which only have constrained enhancers). Finally, we observed that common SNPs linked to genes with constrained promoters and distal enhancers are as enriched in h2 as genes with constrained CDS, suggesting that constraint in regulatory elements can be leveraged in the analyses of human diseases and complex traits (Fig. 4F and table S17).
## Mammalian constraint and copy number variation
Copy number variants (CNVs) are genomic segments that have fewer or more copies compared to a reference genome. CNVs are important drivers of evolution and risk factors for multiple human diseases (42–44). However, CNVs often occur in high repeat/low mappability regions meaning that detecting their presence and significance often carries uncertainty [45, 46]. We thus evaluated whether mammalian constraint could help prioritize potentially disease-related CNVs. First, as a qualitative check, we evaluated a pathogenic CNV—a small distal enhancer upstream of SOX9 with a ClinVar pathogenic annotation as a cause of Pierre Robin sequence—and found that it was highly constrained [47] (Supplemental Methods, Section 11). Second, we evaluated constraint in structural variants (SV) identified in TOPMed [4]. We found that singleton (AC=1) SV deletions, inversions, and duplications had similar fractions of constrained bases. However, common (AF ≥ 0.005) SV deletions had far less constraint than SV inversions or duplications. We speculate that singletons are recent mutations relatively unexposed to purifying selection whereas common SV deletions are directly exposed to selection pressures due to the impacts of haploinsufficiency.
Third, these analyses suggest that constrained bases could have utility in CNV prioritization and burden calculations. Given that CNVs are known risk factors for schizophrenia [48], we obtained the CNV call set from the largest published study (21,094 cases, 20,227 controls) [49]. After replicating the main analysis, we found that schizophrenia cases had greater CNV constraint burden (the total number of conserved bases impacted by a CNV) compared to controls. The case-control differences were 4–5 logs more significant than two commonly used measures of CNV burden (total number and total bases per person). The improvements were particularly notable for CNV deletions. We suggest that the number of constrained bases impacted by a CNV is a more direct assessment of functional impact—e.g., a large CNV with no constrained bases is less likely to be deleterious than a far smaller CNV that deletes constrained exons, promoters, and/or enhancer elements.
## Cancer driver genes identified with mammalian constraint
Moving from the germline to the somatic genomes, we demonstrated how mammalian constraint in non-coding regions of the genome could be applied to detect candidate cancer driver genes (Supplementary Methods, Section 12). Non-coding constraint mutations (NCCMs, phyloP ≥ 1.2 [50]) were identified using whole genome sequencing data (International Cancer Genome Consortium) [51] for two types of brain tumors primarily affecting children. Pilocytic astrocytoma is a low-grade tumor [52] and medulloblastomas are malignant brain tumors with intertumoral heterogeneity informed by subgroups determined by molecular profiling (i.e., Wingless/Integrated (WNT), Sonic Hedgehog Signaling (SHH), Group 3 and Group 4) [53]. We identified NCCMs within introns, 5ánd 3ÚTRs, and regions within 100kb of each gene [50].
We found drastically different NCCM rates between the two cancers. In pilocytic astrocytoma, known to have coding/translocation mutations primarily in BRAF, high NCCM rates were restricted to the BRAF locus, in line with the low somatic mutation burden of this tumor type. Strikingly, for medulloblastoma, 114 genes had ≥ 2 NCCMs/100 kb (Fig. 5A) and 525 genes had ≥ 5 NCCMs per gene. *These* genes were enriched for the GO biological processes “nervous system development” ($$P \leq 1.32$$×10−26) and “generation of neurons” ($$P \leq 1.68$$×10−22.). Among the top 114 genes, 15 gene loci were primarily seen in adult cases (≥18 years of age) and 7 loci in pediatric cases (<18 years of age). A subset of these loci is shown in Fig. 5B (Companion paper #12, Sakthikumar et al). An example is ZFHX4, previously reported to be differentially expressed in medulloblastoma [54], where NCCMs were predominantly identified in adult patients of the SHH subgroup, and found in high constraint ZFHX4 intronic regions (Fig. 5C). For the pediatric set of medulloblastoma, potential driver genes included BMP4 and the HOXB locus (containing multiple genes), mostly in patients diagnosed as Group 3 or Group 4. Multiple NCCMs in these two loci were shown to have differential DNA binding capacity in a medulloblastoma cell line (Companion paper #12, Sakthikumar et al). Further, we noted differential gene expression in medulloblastoma compared to cerebellum for multiple NCCM genes, e.g. HOXB2 [55], for which expression levels correlate with patient survival [56].
The addition of evolutionary constraint measures may help advance stratification of medulloblastoma, both with regard to age, and molecular subgroups. *More* generally, we demonstrate how NCCM analysis can be used as a tool for the identification of novel driver genes in cancer. We suggest that NCCM analysis should be evaluated in more cancer types for its potential to yield a better understanding of disease biology and improved diagnosis and prognosis.
## Discussion
The strength of evolutionary constraint can deepen our understanding of human diseases. The alignment of 240 placental mammals, representing ~100 million years of evolution, achieved single base resolution that allows detailed evaluation of individual mutations in contrast to previous methodologies of only gene-sized resolution. Evolutionary constraint compares favourably to huge amounts of functional genomics data as functionality in any tissue at any time point will be detected by constraint. We demonstrate that constraint can be used to detect candidate causal mutations in both rare and common disease as well as in cancer, and could be particularly leveraged for brain diseases that are more impacted by constrained genes and biological processes. Finally, we note that primate constraint has a stronger heritability enrichment than when measured across placental mammals in non-coding regions suggesting that sequencing more primates would complement the current efforts to validate function of the multitude of regulatory elements present in the human lineage.
## Funding
Swedish Research Council and Knut and Alice Wallenberg Foundation, Swedish Cancer Society, Swedish Childhood Cancer Fund, NIMH U01MH116438, Gladstone Institutes, NIDA DP1DA04658501, NIDA F30DA053020, UCD Ad Astra Fellowship, R00 HG010160 and NHGRI U41HG002371.
## References
1. Moore J. E., Purcaro M. J., Pratt H. E., Epstein C. B., Shoresh N., Adrian J., Kawli T., Davis C. A., Dobin A., Kaul R., Halow J., Van Nostrand E. L., Freese P., Gorkin D. U., Shen Y., He Y., Mackiewicz M., Pauli-Behn F., Williams B. A., Mortazavi A., Keller C. A., Zhang X.-O., Elhajjajy S. I., Huey J., Dickel D. E., Snetkova V., Wei X., Wang X., Rivera-Mulia J. C., Rozowsky J., Zhang J., Chhetri S. B., Zhang J., Victorsen A., White K. P., Visel A., Yeo G. W., Burge C. B., Lécuyer E., Gilbert D. M., Dekker J., Rinn J., Mendenhall E. M., Ecker J. R., Kellis M., Klein R. J., Noble W. S., Kundaje A., Guigó R., Farnham P. J., Cherry J. M., Myers R. M., Ren B., Graveley B. R., Gerstein M. B., Pennacchio L. A., Snyder M. P., Bernstein B. E., Wold B., Hardison R. C., Gingeras T. R., Stamatoyannopoulos J. A., Weng Z.. **Expanded encyclopaedias of DNA elements in the human and mouse genomes.**. *Nature.* (2020.0) **583** 699-710. PMID: 32728249
2. **The GTEx Consortium atlas of genetic regulatory effects across human tissues.**. *Science.* (2020.0) **369** 1318-1330. PMID: 32913098
3. 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., Gauthier L. D., Brand H., Solomonson M., Watts N. A., Rhodes D., Singer-Berk M., England E. M., Seaby E. G., Kosmicki J. A., Walters R. K., Tashman K., Farjoun Y., Banks E., Poterba T., Wang A., Seed C., Whiffin N., Chong J. X., Samocha K. E., Pierce-Hoffman E., Zappala Z., O’Donnell-Luria A. H., Minikel E. V., Weisburd B., Lek M., Ware J. S., Vittal C., Armean I. M., Bergelson L., Cibulskis K., Connolly K. M., Covarrubias M., Donnelly S., Ferriera S., Gabriel S., Gentry J., Gupta N., Jeandet T., Kaplan D., Llanwarne C., Munshi R., Novod S., Petrillo N., Roazen D., Ruano-Rubio V., Saltzman A., Schleicher M., Soto J., Tibbetts K., Tolonen C., Wade G., Talkowski M. E., Neale B. M., Daly M. J., MacArthur D. G.. **The mutational constraint spectrum quantified from variation in 141,456 humans.**. *Nature.* (2020.0) **581** 434-443. PMID: 32461654
4. Taliun D., Harris D. N., Kessler M. D., Carlson J., Szpiech Z. A., Torres R., Taliun S. A. G., Corvelo A., Gogarten S. M., Kang H. M., Pitsillides A. N., LeFaive J., Lee S.-B., Tian X., Browning B. L., Das S., Emde A.-K., Clarke W. E., Loesch D. P., Shetty A. C., Blackwell T. W., Smith A. V., Wong Q., Liu X., Conomos M. P., Bobo D. M., Aguet F., Albert C., Alonso A., Ardlie K. G., Arking D. E., Aslibekyan S., Auer P. L., Barnard J., Barr R. G., Barwick L., Becker L. C., Beer R. L., Benjamin E. J., Bielak L. F., Blangero J., Boehnke M., Bowden D. W., Brody J. A., Burchard E. G., Cade B. E., Casella J. F., Chalazan B., Chasman D. I., Chen Y.-D. I., Cho M. H., Choi S. H., Chung M. K., Clish C. B., Correa A., Curran J. E., Custer B., Darbar D., Daya M., de Andrade M., DeMeo D. L., Dutcher S. K., Ellinor P. T., Emery L. S., Eng C., Fatkin D., Fingerlin T., Forer L., Fornage M., Franceschini N., Fuchsberger C., Fullerton S. M., Germer S., Gladwin M. T., Gottlieb D. J., Guo X., Hall M. E., He J., Heard-Costa N. L., Heckbert S. R., Irvin M. R., Johnsen J. M., Johnson A. D., Kaplan R., Kardia S. L. R., Kelly T., Kelly S., Kenny E. E., Kiel D. P., Klemmer R., Konkle B. A., Kooperberg C., Köttgen A., Lange L. A., Lasky-Su J., Levy D., Lin X., Lin K.-H., Liu C., Loos R. J. F., Garman L., Gerszten R., Lubitz S. A., Lunetta K. L., Mak A. C. Y., Manichaikul A., Manning A. K., Mathias R. A., McManus D. D., McGarvey S. T., Meigs J. B., Meyers D. A., Mikulla J. L., Minear M. A., Mitchell B. D., Mohanty S., Montasser M. E., Montgomery C., Morrison A. C., Murabito J. M., Natale A., Natarajan P., Nelson S. C., North K. E., O’Connell J. R., Palmer N. D., Pankratz N., Peloso G. M., Peyser P. A., Pleiness J., Post W. S., Psaty B. M., Rao D. C., Redline S., Reiner A. P., Roden D., Rotter J. I., Ruczinski I., Sarnowski C., Schoenherr S., Schwartz D. A., Seo J.-S., Seshadri S., Sheehan V. A., Sheu W. H., Shoemaker M. B., Smith N. L., Smith J. A., Sotoodehnia N., Stilp A. M., Tang W., Taylor K. D., Telen M., Thornton T. A., Tracy R. P., Van Den Berg D. J., Vasan R. S., Viaud-Martinez K. A., Vrieze S., Weeks D. E., Weir B. S., Weiss S. T., Weng L.-C., Willer C. J., Zhang Y., Zhao X., Arnett D. K., Ashley-Koch A. E., Barnes K. C., Boerwinkle E., Gabriel S., Gibbs R., Rice K. M., Rich S. S., Silverman E. K., Qasba P., Gan W., Papanicolaou G. J., Nickerson D. A., Browning S. R., Zody M. C., Zöllner S., Wilson J. G., Cupples L. A., Laurie C. C., Jaquish C. E., Hernandez R. D., O’Connor T. D., Abecasis G. R.. **Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program.**. *Nature.* (2021.0) **590** 290-299. PMID: 33568819
5. Cooper G. M., Shendure J.. **Needles in stacks of needles: finding disease-causal variants in a wealth of genomic data.**. *Nature Reviews Genetics.* (2011.0) **12** 628-640
6. Kircher M., Witten D. M., Jain P., O’Roak B. J., Cooper G. M., Shendure J.. **A general framework for estimating the relative pathogenicity of human genetic variants.**. *Nat. Genet.* (2014.0) **46** 310-315. PMID: 24487276
7. Finucane H. K., Bulik-Sullivan B., Gusev A., Trynka G., Reshef Y., Loh P.-R., Anttila V., Xu H., Zang C., Farh K., Ripke S., Day F. R., Purcell S., Stahl E., Lindstrom S., Perry J. R. B., Okada Y., Raychaudhuri S., Daly M. J., Patterson N., Neale B. M., Price A. L.. **Partitioning heritability by functional annotation using genome-wide association summary statistics.**. *Nat. Genet.* (2015.0) **47** 1228-1235. PMID: 26414678
8. Gazal S., Loh P.-R., Finucane H. K., Ganna A., Schoech A., Sunyaev S., Price A. L.. **Functional architecture of low-frequency variants highlights strength of negative selection across coding and non-coding annotations.**. *Nat. Genet.* (2018.0) **50** 1600-1607. PMID: 30297966
9. Hujoel M. L. A., Gazal S., Hormozdiari F., van de Geijn B., Price A. L.. **Disease Heritability Enrichment of Regulatory Elements Is Concentrated in Elements with Ancient Sequence Age and Conserved Function across Species.**. *Am. J. Hum. Genet.* (2019.0) **104** 611-624. PMID: 30905396
10. Visscher P. M., Wray N. R., Zhang Q., Sklar P., McCarthy M. I., Brown M. A., Yang J.. **10 Years of GWAS Discovery: Biology, Function, and Translation.**. *Am. J. Hum. Genet.* (2017.0) **101** 5-22. PMID: 28686856
11. Gallagher M. D., Chen-Plotkin A. S.. **The Post-GWAS Era: From Association to Function.**. *Am. J. Hum. Genet.* (2018.0) **102** 717-730. PMID: 29727686
12. Tam V., Patel N., Turcotte M., Bossé Y., Paré G., Meyre D.. **Benefits and limitations of genome-wide association studies.**. *Nat. Rev. Genet.* (2019.0) **20** 467-484. PMID: 31068683
13. Claussnitzer M., Cho J. H., Collins R., Cox N. J., Dermitzakis E. T., Hurles M. E., Kathiresan S., Kenny E. E., Lindgren C. M., MacArthur D. G., North K. N., Plon S. E., Rehm H. L., Risch N., Rotimi C. N., Shendure J., Soranzo N., McCarthy M. I.. **A brief history of human disease genetics.**. *Nature.* (2020.0) **577** 179-189. PMID: 31915397
14. Uffelmann E., Huang Q. Q., Munung N. S., de Vries J., Okada Y., Martin A. R., Martin H. C., Lappalainen T., Posthuma D.. **Genome-wide association studies.**. *Nature Reviews Methods Primers.* (2021.0) **1** 1-21
15. Lappalainen T., MacArthur D. G.. **From variant to function in human disease genetics.**. *Science.* (2021.0) **373** 1464-1468. PMID: 34554789
16. Siepel A., Pollard K. S., Haussler D.. **New Methods for Detecting Lineage-Specific Selection.**. *Lecture Notes in Computer Science* (2006.0) 190-205
17. Siepel A., Bejerano G., Pedersen J. S., Hinrichs A. S., Hou M., Rosenbloom K., Clawson H., Spieth J., Hillier L. W., Richards S., Weinstock G. M., Wilson R. K., Gibbs R. A., Kent W. J., Miller W., Haussler D.. **Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes.**. *Genome Res.* (2005.0) **15** 1034-1050. PMID: 16024819
18. Finan C., Gaulton A., Kruger F. A., Lumbers R. T., Shah T., Engmann J., Galver L., Kelley R., Karlsson A., Santos R., Overington J. P., Hingorani A. D., Casas J. P.. **The druggable genome and support for target identification and validation in drug development.**. *Sci. Transl. Med.* (2017.0) **9**. DOI: 10.1126/scitranslmed.aag1166
19. Landrum M. J., Lee J. M., Benson M., Brown G. R., Chao C., Chitipiralla S., Gu B., Hart J., Hoffman D., Jang W., Karapetyan K., Katz K., Liu C., Maddipatla Z., Malheiro A., McDaniel K., Ovetsky M., Riley G., Zhou G., Holmes J. B., Kattman B. L., Maglott D. R.. **ClinVar: improving access to variant interpretations and supporting evidence.**. *Nucleic Acids Res.* (2017.0) **46** D1062-D1067
20. Lopes-Pacheco M.. **CFTR Modulators: Shedding Light on Precision Medicine for Cystic Fibrosis.**. *Front. Pharmacol.* (2016.0). DOI: 10.3389/fphar.2016.00275
21. Akbari P., Gilani A., Sosina O., Kosmicki J. A., Khrimian L., Fang Y.-Y., Persaud T., Garcia V., Sun D., Li A., Mbatchou J., Locke A. E., Benner C., Verweij N., Lin N., Hossain S., Agostinucci K., Pascale J. V., Dirice E., Dunn M., Kraus W. E., Shah S. H., Chen Y.-D. I., Rotter J. I., Rader D. J., Melander O., Still C. D., Mirshahi T., Carey D. J., Berumen-Campos J., Kuri-Morales P., Alegre-Díaz J., Torres J. M., Emberson J. R., Collins R., Balasubramanian S., Hawes A., Jones M., Zambrowicz B., Murphy A. J., Paulding C., Coppola G., Overton J. D., Reid J. G., Shuldiner A. R., Cantor M., Kang H. M., Abecasis G. R., Karalis K., Economides A. N., Marchini J., Yancopoulos G. D., Sleeman M. W., Altarejos J., Della Gatta G., Tapia-Conyer R., Schwartzman M. L., Baras A., Ferreira M. A. R., Lotta L. A.. **Sequencing of 640,000 exomes identifies variants associated with protection from obesity.**. *Science.* (2021.0) **373**. DOI: 10.1126/science.abf8683
22. Jones E. M., Lubock N. B., Venkatakrishnan A. J., Wang J., Tseng A. M., Paggi J. M., Latorraca N. R., Cancilla D., Satyadi M., Davis J. E., Babu M. M., Dror R. O., Kosuri S.. **Structural and functional characterization of G protein-coupled receptors with deep mutational scanning.**. *Elife* (2020.0) **9**. DOI: 10.7554/eLife.54895
23. Gazal S., Finucane H. K., Furlotte N. A., Loh P.-R., Palamara P. F., Liu X., Schoech A., Bulik-Sullivan B., Neale B. M., Gusev A., Price A. L.. **Linkage disequilibrium-dependent architecture of human complex traits shows action of negative selection.**. *Nat. Genet.* (2017.0) **49** 1421-1427. PMID: 28892061
24. Gazal S., Marquez-Luna C., Finucane H. K., Price A. L.. **Reconciling S-LDSC and LDAK functional enrichment estimates.**. *Nat. Genet.* (2019.0) **51** 1202-1204. PMID: 31285579
25. Gazal S., Weissbrod O., Hormozdiari F., Dey K., Nasser J., Jagadeesh K., Weiner D., Shi H., Fulco C., O’Connor L., Pasaniuc B., Engreitz J. M., Price A. L.. **Combining SNP-to-gene linking strategies to pinpoint disease genes and assess disease omnigenicity.**. *medRxiv* (2021.0)
26. Davydov E. V., Goode D. L., Sirota M., Cooper G. M., Sidow A., Batzoglou S.. **Identifying a high fraction of the human genome to be under selective constraint using GERP++.**. *PLoS Comput. Biol.* (2010.0) **6** e1001025. PMID: 21152010
27. Lindblad-Toh K., Garber M., Zuk O., Lin M. F., Parker B. J., Washietl S., Kheradpour P., Ernst J., Jordan G., Mauceli E., Ward L. D., Lowe C. B., Holloway A. K., Clamp M., Gnerre S., Alföldi J., Beal K., Chang J., Clawson H., Cuff J., Di Palma F., Fitzgerald S., Flicek P., Guttman M., Hubisz M. J., Jaffe D. B., Jungreis I., Kent W. J., Kostka D., Lara M., Martins A. L., Massingham T., Moltke I., Raney B. J., Rasmussen M. D., Robinson J., Stark A., Vilella A. J., Wen J., Xie X., Zody M. C., Baldwin J., Bloom T., Chin C. W., Heiman D., Nicol R., Nusbaum C., Young S., Wilkinson J., Worley K. C., Kovar C. L., Muzny D. M., Gibbs R. A., Cree A., Dihn H. H., Fowler G., Jhangiani S., Joshi V., Lee S., Lewis L. R., Nazareth L. V., Okwuonu G., Santibanez J., Warren W. C., Mardis E. R., Weinstock G. M., Wilson R. K., Delehaunty K., Dooling D., Fronik C., Fulton L., Fulton B., Graves T., Minx P., Sodergren E., Birney E., Margulies E. H., Herrero J., Green E. D., Haussler D., Siepel A., Goldman N., Pollard K. S., Pedersen J. S., Lander E. S., Kellis M.. **A high-resolution map of human evolutionary constraint using 29 mammals.**. *Nature.* (2011.0) **478** 476-482. PMID: 21993624
28. Hormozdiari F., Gazal S., van de Geijn B., Finucane H. K., Ju C. J.-T., Loh P.-R., Schoech A., Reshef Y., Liu X., O’Connor L., Gusev A., Eskin E., Price A. L.. **Leveraging molecular quantitative trait loci to understand the genetic architecture of diseases and complex traits.**. *Nat. Genet.* (2018.0) **50** 1041-1047. PMID: 29942083
29. Shi H., Gazal S., Kanai M., Koch E. M., Schoech A. P., Siewert K. M., Kim S. S., Luo Y., Amariuta T., Huang H., Okada Y., Raychaudhuri S., Sunyaev S. R., Price A. L.. **Population-specific causal disease effect sizes in functionally important regions impacted by selection.**. *Nat. Commun.* (2021.0) **12** 1098. PMID: 33597505
30. Weissbrod O., Hormozdiari F., Benner C., Cui R., Ulirsch J., Gazal S., Schoech A. P., van de Geijn B., Reshef Y., Márquez-Luna C., O’Connor L., Pirinen M., Finucane H. K., Price A. L.. **Functionally informed fine-mapping and polygenic localization of complex trait heritability.**. *Nat. Genet.* (2020.0) **52** 1355-1363. PMID: 33199916
31. Loh P.-R., Kichaev G., Gazal S., Schoech A. P., Price A. L.. **Mixed-model association for biobank-scale datasets.**. *Nat. Genet.* (2018.0) **50** 906-908. PMID: 29892013
32. Claussnitzer M., Dankel S. N., Kim K.-H., Quon G., Meuleman W., Haugen C., Glunk V., Sousa I. S., Beaudry J. L., Puviindran V., Abdennur N. A., Liu J., Svensson P.-A., Hsu Y.-H., Drucker D. J., Mellgren G., Hui C.-C., Hauner H., Kellis M.. **FTO Obesity Variant Circuitry and Adipocyte Browning in Humans.**. *N. Engl. J. Med.* (2015.0) **373** 895-907. PMID: 26287746
33. Claussnitzer M., Hui C.-C., Kellis M.. **FTO Obesity Variant and Adipocyte Browning in Humans.**. *N. Engl. J. Med.* (2016.0) **374** 192-193. PMID: 26760096
34. Boix C. A., James B. T., Park Y. P., Meuleman W., Kellis M.. **Regulatory genomic circuitry of human disease loci by integrative epigenomics.**. *Nature.* (2021.0) **590** 300-307. PMID: 33536621
35. Umans B. D., Battle A., Gilad Y.. **Where Are the Disease-Associated eQTLs?**. *Trends Genet.* (2021.0) **37** 109-124. PMID: 32912663
36. Zhu J., Yamane H., Cote-Sierra J., Guo L., Paul W. E.. **GATA-3 promotes Th2 responses through three different mechanisms: induction of Th2 cytokine production, selective growth of Th2 cells and inhibition of Th1 cell-specific factors.**. *Cell Res.* (2006.0) **16** 3-10. PMID: 16467870
37. Mjösberg J., Bernink J., Golebski K., Karrich J. J., Peters C. P., Blom B., te Velde A. A., Fokkens W. J., van Drunen C. M., Spits H.. **The transcription factor GATA3 is essential for the function of human type 2 innate lymphoid cells.**. *Immunity.* (2012.0) **37** 649-659. PMID: 23063330
38. Wohlfert E. A., Grainger J. R., Bouladoux N., Konkel J. E., Oldenhove G., Ribeiro C. H., Hall J. A., Yagi R., Naik S., Bhairavabhotla R., Paul W. E., Bosselut R., Wei G., Zhao K., Oukka M., Zhu J., Belkaid Y.. **GATA3 controls Foxp3**. *J. Clin. Invest.* (2011.0) **121** 4503-4515. PMID: 21965331
39. Griesemer D., Xue J. R., Reilly S. K., Ulirsch J. C., Kukreja K., Davis J., Kanai M., Yang D. K., Montgomery S. B., Novina C. D., Tewhey R., Sabeti P. C.. **Genome-Wide Functional Screen of 3’UTR Variants Uncovers Causal Variants for Human Disease and Evolution.**. *SSRN Electronic Journal*. DOI: 10.2139/ssrn.3762769
40. Tewhey R., Kotliar D., Park D. S., Liu B., Winnicki S., Reilly S. K., Andersen K. G., Mikkelsen T. S., Lander E. S., Schaffner S. F., Sabeti P. C.. **Direct Identification of Hundreds of Expression-Modulating Variants using a Multiplexed Reporter Assay.**. *Cell.* (2018.0) **172** 1132-1134. PMID: 29474912
41. Kircher M., Xiong C., Martin B., Schubach M., Inoue F., Bell R. J. A., Costello J. F., Shendure J., Ahituv N.. **Saturation mutagenesis of twenty disease-associated regulatory elements at single base-pair resolution.**. *Nat. Commun.* (2019.0) **10** 3583. PMID: 31395865
42. Innan H., Kondrashov F.. **The evolution of gene duplications: classifying and distinguishing between models.**. *Nat. Rev. Genet.* (2010.0) **11** 97-108. PMID: 20051986
43. Zarrei M., MacDonald J. R., Merico D., Scherer S. W.. **A copy number variation map of the human genome.**. *Nat. Rev. Genet.* (2015.0) **16** 172-183. PMID: 25645873
44. Mérot C., Oomen R. A., Tigano A., Wellenreuther M.. **A Roadmap for Understanding the Evolutionary Significance of Structural Genomic Variation.**. *Trends Ecol. Evol.* (2020.0) **35** 561-572. PMID: 32521241
45. Lappalainen T., Scott A. J., Brandt M., Hall I. M.. **Genomic Analysis in the Age of Human Genome Sequencing.**. *Cell.* (2019.0) **177** 70-84. PMID: 30901550
46. Mahmoud M., Gobet N., Cruz-Dávalos D. I., Mounier N., Dessimoz C., Sedlazeck F. J.. **Structural variant calling: the long and the short of it.**. *Genome Biol.* (2019.0) **20** 246. PMID: 31747936
47. Long H. K., Osterwalder M., Welsh I. C., Hansen K., Davies J. O. J., Liu Y. E., Koska M., Adams A. T., Aho R., Arora N., Ikeda K., Williams R. M., Sauka-Spengler T., Porteus M. H., Mohun T., Dickel D. E., Swigut T., Hughes J. R., Higgs D. R., Visel A., Selleri L., Wysocka J.. **Loss of Extreme Long-Range Enhancers in Human Neural Crest Drives a Craniofacial Disorder.**. *Cell Stem Cell.* (2020.0) **27** 765-783. PMID: 32991838
48. Sullivan P. F., Geschwind D. H.. **Defining the Genetic, Genomic, Cellular, and Diagnostic Architectures of Psychiatric Disorders.**. *Cell.* (2019.0) **177** 162-183. PMID: 30901538
49. Marshall C. R., Howrigan D. P., Merico D., Thiruvahindrapuram B., Wu W., Greer D. S., Antaki D., Shetty A., Holmans P. A., Pinto D., Gujral M., Brandler W. M., Malhotra D., Wang Z., Fajarado K. V. F., Maile M. S., Ripke S., Agartz I., Albus M., Alexander M., Amin F., Atkins J., Bacanu S. A., Belliveau R. A., Bergen S. E., Bertalan M., Bevilacqua E., Bigdeli T. B., Black D. W., Bruggeman R., Buccola N. G., Buckner R. L., Bulik-Sullivan B., Byerley W., Cahn W., Cai G., Cairns M. J., Campion D., Cantor R. M., Carr V. J., Carrera N., Catts S. V., Chambert K. D., Cheng W., Cloninger C. R., Cohen D., Cormican P., Craddock N., Crespo-Facorro B., Crowley J. J., Curtis D., Davidson M., Davis K. L., Degenhardt F., Del Favero J., DeLisi L. E., Dikeos D., Dinan T., Djurovic S., Donohoe G., Drapeau E., Duan J., Dudbridge F., Eichhammer P., Eriksson J., Escott-Price V., Essioux L., Fanous A. H., Farh K.-H., Farrell M. S., Frank J., Franke L., Freedman R., Freimer N. B., Friedman J. I., Forstner A. J., Fromer M., Genovese G., Georgieva L., Gershon E. S., Giegling I., Giusti-Rodríguez P., Godard S., Goldstein J. I., Gratten J., de Haan L., Hamshere M. L., Hansen M., Hansen T., Haroutunian V., Hartmann A. M., Henskens F. A., Herms S., Hirschhorn J. N., Hoffmann P., Hofman A., Huang H., Ikeda M., Joa I., Kähler A. K., Kahn R. S., Kalaydjieva L., Karjalainen J., Kavanagh D., Keller M. C., Kelly B. J., Kennedy J. L., Kim Y., Knowles J. A., Konte B., Laurent C., Lee P., Lee S. H., Legge S. E., Lerer B., Levy D. L., Liang K.-Y., Lieberman J., Lönnqvist J., Loughland C. M., Magnusson P. K. E., Maher B. S., Maier W., Mallet J., Mattheisen M., Mattingsdal M., McCarley R. W., McDonald C., McIntosh A. M., Meier S., Meijer C. J., Melle I., Mesholam-Gately R. I., Metspalu A., Michie P. T., Milani L., Milanova V., Mokrab Y., Morris D. W., Müller-Myhsok B., Murphy K. C., Murray R. M., Myin-Germeys I., Nenadic I., Nertney D. A., Nestadt G., Nicodemus K. K., Nisenbaum L., Nordin A., O’Callaghan E., O’Dushlaine C., Oh S.-Y., Olincy A., Olsen L., O’Neill F. A., Van Os J., Pantelis C., Papadimitriou G. N., Parkhomenko E., Pato M. T., Paunio T., Perkins D. O., Pers T. H., Pietiläinen O., Pimm J., Pocklington A. J., Powell J., Price A., Pulver A. E., Purcell S. M., Quested D., Rasmussen H. B., Reichenberg A., Reimers M. A., Richards A. L., Roffman J. L., Roussos P., Ruderfer D. M., Salomaa V., Sanders A. R., Savitz A., Schall U., Schulze T. G., Schwab S. G., Scolnick E. M., Scott R. J., Seidman L. J., Shi J., Silverman J. M., Smoller J. W., Söderman E., Spencer C. C. A., Stahl E. A., Strengman E., Strohmaier J., Stroup T. S., Suvisaari J., Svrakic D. M., Szatkiewicz J. P., Thirumalai S., Tooney P. A., Veijola J., Visscher P. M., Waddington J., Walsh D., Webb B. T., Weiser M., Wildenauer D. B., Williams N. M., Williams S., Witt S. H., Wolen A. R., Wormley B. K., Wray N. R., Wu J. Q., Zai C. C., Adolfsson R., Andreassen O. A., Blackwood D. H. R., Bramon E., Buxbaum J. D., Cichon S., Collier D. A., Corvin A., Daly M. J., Darvasi A., Domenici E., Esko T., Gejman P. V., Gill M., Gurling H., Hultman C. M., Iwata N., Jablensky A. V., Jönsson E. G., Kendler K. S., Kirov G., Knight J., Levinson D. F., Li Q. S., McCarroll S. A., McQuillin A., Moran J. L., Mowry B. J., Nöthen M. M., Ophoff R. A., Owen M. J., Palotie A., Pato C. N., Petryshen T. L., Posthuma D., Rietschel M., Riley B. P., Rujescu D., Sklar P., St Clair D., Walters J. T. R., Werge T., Sullivan P. F., O’Donovan M. C., Scherer S. W., Neale B. M., Sebat J.. **CNV and Schizophrenia Working Groups of the Psychiatric Genomics Consortium, Contribution of copy number variants to schizophrenia from a genome-wide study of 41,321 subjects.**. *Nat. Genet.* (2017.0) **49** 27-35. PMID: 27869829
50. Sakthikumar S., Roy A., Haseeb L., Pettersson M. E., Sundström E., Marinescu V. D., Lindblad-Toh K., Forsberg-Nilsson K.. **Whole-genome sequencing of glioblastoma reveals enrichment of non-coding constraint mutations in known and novel genes.**. *Genome Biol.* (2020.0) **21** 127. PMID: 32513296
51. Zhang J., Bajari R., Andric D., Gerthoffert F., Lepsa A., Nahal-Bose H., Stein L. D., Ferretti V.. **The International Cancer Genome Consortium Data Portal.**. *Nat. Biotechnol.* (2019.0) **37** 367-369. PMID: 30877282
52. Louis D. N., Perry A., Reifenberger G., von Deimling A., Figarella-Branger D., Cavenee W. K., Ohgaki H., Wiestler O. D., Kleihues P., Ellison D. W.. **The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary.**. *Acta Neuropathol.* (2016.0) **131** 803-820. PMID: 27157931
53. Northcott P. A., Buchhalter I., Morrissy A. S., Hovestadt V., Weischenfeldt J., Ehrenberger T., Gröbner S., Segura-Wang M., Zichner T., Rudneva V. A., Warnatz H.-J., Sidiropoulos N., Phillips A. H., Schumacher S., Kleinheinz K., Waszak S. M., Erkek S., Jones D. T. W., Worst B. C., Kool M., Zapatka M., Jäger N., Chavez L., Hutter B., Bieg M., Paramasivam N., Heinold M., Gu Z., Ishaque N., Jäger-Schmidt C., Imbusch C. D., Jugold A., Hübschmann D., Risch T., Amstislavskiy V., Gonzalez F. G. R., Weber U. D., Wolf S., Robinson G. W., Zhou X., Wu G., Finkelstein D., Liu Y., Cavalli F. M. G., Luu B., Ramaswamy V., Wu X., Koster J., Ryzhova M., Cho Y.-J., Pomeroy S. L., Herold-Mende C., Schuhmann M., Ebinger M., Liau L. M., Mora J., McLendon R. E., Jabado N., Kumabe T., Chuah E., Ma Y., Moore R. A., Mungall A. J., Mungall K. L., Thiessen N., Tse K., Wong T., Jones S. J. M., Witt O., Milde T., Von Deimling A., Capper D., Korshunov A., Yaspo M.-L., Kriwacki R., Gajjar A., Zhang J., Beroukhim R., Fraenkel E., Korbel J. O., Brors B., Schlesner M., Eils R., Marra M. A., Pfister S. M., Taylor M. D., Lichter P.. **The whole-genome landscape of medulloblastoma subtypes.**. *Nature.* (2017.0) **547** 311-317. PMID: 28726821
54. Smits M., van Rijn S., Hulleman E., Biesmans D., van Vuurden D. G., Kool M., Haberler C., Aronica E., Vandertop W. P., Noske D. P., Würdinger T.. **EZH2-regulated DAB2IP is a medulloblastoma tumor suppressor and a positive marker for survival.**. *Clin. Cancer Res.* (2012.0) **18** 4048-4058. PMID: 22696229
55. Weishaupt H., Johansson P., Sundström A., Lubovac-Pilav Z., Olsson B., Nelander S., Swartling F. J.. **Batch-normalization of cerebellar and medulloblastoma gene expression datasets utilizing empirically defined negative control genes.**. *Bioinformatics.* (2019.0) **35** 3357-3364. PMID: 30715209
56. Cavalli F. M. G., Remke M., Rampasek L., Peacock J., Shih D. J. H., Luu B., Garzia L., Torchia J., Nor C., Morrissy A. S., Agnihotri S., Thompson Y. Y., Kuzan-Fischer C. M., Farooq H., Isaev K., Daniels C., Cho B.-K., Kim S.-K., Wang K.-C., Lee J. Y., Grajkowska W. A., Perek-Polnik M., Vasiljevic A., Faure-Conter C., Jouvet A., Giannini C., Nageswara Rao A. A., Li K. K. W., Ng H.-K., Eberhart C. G., Pollack I. F., Hamilton R. L., Gillespie G. Y., Olson J. M., Leary S., Weiss W. A., Lach B., Chambless L. B., Thompson R. C., Cooper M. K., Vibhakar R., Hauser P., van Veelen M.-L. C., Kros J. M., French P. J., Ra Y. S., Kumabe T., López-Aguilar E., Zitterbart K., Sterba J., Finocchiaro G., Massimino M., Van Meir E. G., Osuka S., Shofuda T., Klekner A., Zollo M., Leonard J. R., Rubin J. B., Jabado N., Albrecht S., Mora J., Van Meter T. E., Jung S., Moore A. S., Hallahan A. R., Chan J. A., Tirapelli D. P. C., Carlotti C. G., Fouladi M., Pimentel J., Faria C. C., Saad A. G., Massimi L., Liau L. M., Wheeler H., Nakamura H., Elbabaa S. K., Perezpeña-Diazconti M., Chico Ponce de León F., Robinson S., Zapotocky M., Lassaletta A., Huang A., Hawkins C. E., Tabori U., Bouffet E., Bartels U., Dirks P. B., Rutka J. T., Bader G. D., Reimand J., Goldenberg A., Ramaswamy V., Taylor M. D.. **Intertumoral Heterogeneity within Medulloblastoma Subgroups.**. *Cancer Cell.* (2017.0) **31** 737-754. PMID: 28609654
57. Buniello A., MacArthur J. A. L., Cerezo M., Harris L. W., Hayhurst J., Malangone C., McMahon A., Morales J., Mountjoy E., Sollis E., Suveges D., Vrousgou O., Whetzel P. L., Amode R., Guillen J. A., Riat H. S., Trevanion S. J., Hall P., Junkins H., Flicek P., Burdett T., Hindorff L. A., Cunningham F., Parkinson H.. **The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019.**. *Nucleic Acids Res.* (2019.0) **47** D1005-D1012. PMID: 30445434
58. Liu B., Gloudemans M. J., Rao A. S., Ingelsson E., Montgomery S. B.. **Abundant associations with gene expression complicate GWAS follow-up.**. *Nat. Genet.* (2019.0) **51** 768-769. PMID: 31043754
59. Amberger J. S., Bocchini C. A., Scott A. F., Hamosh A.. **OMIM.org: leveraging knowledge across phenotype-gene relationships.**. *Nucleic Acids Res.* (2019.0) **47** D1038-D1043. PMID: 30445645
60. Landrum M. J., Lee J. M., Riley G. R., Jang W., Rubinstein W. S., Church D. M., Maglott D. R.. **ClinVar: public archive of relationships among sequence variation and human phenotype.**. *Nucleic Acids Res.* (2014.0) **42** D980-5. PMID: 24234437
61. Lek M., Karczewski K. J., Minikel E. V., Samocha K. E., Banks E., Fennell T., O’Donnell-Luria A. H., Ware J. S., Hill A. J., Cummings B. B., Tukiainen T., Birnbaum D. P., Kosmicki J. A., Duncan L. E., Estrada K., Zhao F., Zou J., Pierce-Hoffman E., Berghout J., Cooper D. N., Deflaux N., DePristo M., Do R., Flannick J., Fromer M., Gauthier L., Goldstein J., Gupta N., Howrigan D., Kiezun A., Kurki M. I., Moonshine A. L., Natarajan P., Orozco L., Peloso G. M., Poplin R., Rivas M. A., Ruano-Rubio V., Rose S. A., Ruderfer D. M., Shakir K., Stenson P. D., Stevens C., Thomas B. P., Tiao G., Tusie-Luna M. T., Weisburd B., Won H.-H., Yu D., Altshuler D. M., Ardissino D., Boehnke M., Danesh J., Donnelly S., Elosua R., Florez J. C., Gabriel S. B., Getz G., Glatt S. J., Hultman C. M., Kathiresan S., Laakso M., McCarroll S., McCarthy M. I., McGovern D., McPherson R., Neale B. M., Palotie A., Purcell S. M., Saleheen D., Scharf J. M., Sklar P., Sullivan P. F., Tuomilehto J., Tsuang M. T., Watkins H. C., Wilson J. G., Daly M. J., MacArthur D. G.. **Exome Aggregation Consortium, Analysis of protein-coding genetic variation in 60,706 humans.**. *Nature.* (2016.0) **536** 285-291. PMID: 27535533
62. Cummings B. B., Karczewski K. J., Kosmicki J. A., Seaby E. G., Watts N. A., Singer-Berk M., Mudge J. M., Karjalainen J., Satterstrom F. K., O’Donnell-Luria A. H., Poterba T., Seed C., Solomonson M., Alföldi J., Daly M. J., MacArthur D. G.. **Transcript expression-aware annotation improves rare variant interpretation.**. *Nature.* (2020.0) **581** 452-458. PMID: 32461655
63. Johansson P. A., Brattås P. L., Douse C. H., Hsieh P., Adami A., Pontis J., Grassi D., Garza R., Sozzi E., Cataldo R., Jönsson M. E., Atacho D. A. M., Pircs K., Eren F., Sharma Y., Johansson J., Fiorenzano A., Parmar M., Fex M., Trono D., Eichler E. E., Jakobsson J.. **A cis-acting structural variation at the ZNF558 locus controls a gene regulatory network in human brain development.**. *Cell Stem Cell* (2021.0). DOI: 10.1016/j.stem.2021.09.008
64. Zoonomia Consortium A. **comparative genomics multitool for scientific discovery and conservation.**. *Nature.* (2020.0) **587** 240-245. PMID: 33177664
65. Frankish A., Diekhans M., Jungreis I., Lagarde J., Loveland J. E., Mudge J. M., Sisu C., Wright J. C., Armstrong J., Barnes I., Berry A., Bignell A., Boix C., Carbonell Sala S., Cunningham F., Di Domenico T., Donaldson S., Fiddes I. T., García Girón C., Gonzalez J. M., Grego T., Hardy M., Hourlier T., Howe K. L., Hunt T., Izuogu O. G., Johnson R., Martin F. J., Martínez L., Mohanan S., Muir P., Navarro F. C. P., Parker A., Pei B., Pozo F., Riera F. C., Ruffier M., Schmitt B. M., Stapleton E., Suner M.-M., Sycheva I., Uszczynska-Ratajczak B., Wolf M. Y., Xu J., Yang Y. T., Yates A., Zerbino D., Zhang Y., Choudhary J. S., Gerstein M., Guigó R., Hubbard T. J. P., Kellis M., Paten B., Tress M. L., Flicek P.. **GENCODE 2021.**. *Nucleic Acids Res.* (2021.0) **49** D916-D923. PMID: 33270111
66. Hickey G., Paten B., Earl D., Zerbino D., Haussler D.. **HAL: a hierarchical format for storing and analyzing multiple genome alignments.**. *Bioinformatics.* (2013.0) **29** 1341-1342. PMID: 23505295
67. Earl D., Nguyen N., Hickey G., Harris R. S., Fitzgerald S., Beal K., Seledtsov I., Molodtsov V., Raney B. J., Clawson H., Kim J., Kemena C., Chang J.-M., Erb I., Poliakov A., Hou M., Herrero J., Kent W. J., Solovyev V., Darling A. E., Ma J., Notredame C., Brudno M., Dubchak I., Haussler D., Paten B.. **Alignathon: a competitive assessment of whole-genome alignment methods.**. *Genome Res.* (2014.0) **24** 2077-2089. PMID: 25273068
68. Felsenstein J., Churchill G. A.. **A Hidden Markov Model approach to variation among sites in rate of evolution.**. *Mol. Biol. Evol.* (1996.0) **13** 93-104. PMID: 8583911
69. Storey J. D.. **A direct approach to false discovery rates.**. *J. R. Stat. Soc. Series B Stat. Methodol.* (2002.0) **64** 479-498
70. Pollard K. S., Hubisz M. J., Rosenbloom K. R., Siepel A.. **Detection of nonneutral substitution rates on mammalian phylogenies.**. *Genome Res.* (2010.0) **20** 110-121. PMID: 19858363
71. Navarro Gonzalez J., Zweig A. S., Speir M. L., Schmelter D., Rosenbloom K. R., Raney B. J., Powell C. C., Nassar L. R., Maulding N. D., Lee C. M., Lee B. T., Hinrichs A. S., Fyfe A. C., Fernandes J. D., Diekhans M., Clawson H., Casper J., Benet-Pagès A., Barber G. P., Haussler D., Kuhn R. M., Haeussler M., Kent W. J.. **The UCSC Genome Browser database: 2021 update.**. *Nucleic Acids Res.* (2021.0) **49** D1046-D1057. PMID: 33221922
72. Zdobnov E. M., Kuznetsov D., Tegenfeldt F., Manni M., Berkeley M., Kriventseva E. V.. **OrthoDB in 2020: evolutionary and functional annotations of orthologs.**. *Nucleic Acids Research.* (2021.0) **49** D389-D393. PMID: 33196836
73. Partridge E. C., Chhetri S. B., Prokop J. W., Ramaker R. C., Jansen C. S., Goh S.-T., Mackiewicz M., Newberry K. M., Brandsmeier L. A., Meadows S. K., Messer C. L., Hardigan A. A., Coppola C. J., Dean E. C., Jiang S., Savic D., Mortazavi A., Wold B. J., Myers R. M., Mendenhall E. M.. **Occupancy maps of 208 chromatin-associated proteins in one human cell type.**. *Nature.* (2020.0) **583** 720-728. PMID: 32728244
74. Snyder M. P., Gingeras T. R., Moore J. E., Weng Z., Gerstein M. B., Ren B., Hardison R. C., Stamatoyannopoulos J. A., Graveley B. R., Feingold E. A., Pazin M. J., Pagan M., Gilchrist D. A., Hitz B. C., Cherry J. M., Bernstein B. E., Mendenhall E. M., Zerbino D. R., Frankish A., Flicek P., Myers R. M.. **Perspectives on ENCODE.**. *Nature.* (2020.0) **583** 693-698. PMID: 32728248
75. Grubert F., Srivas R., Spacek D. V., Kasowski M., Ruiz-Velasco M., Sinnott-Armstrong N., Greenside P., Narasimha A., Liu Q., Geller B., Sanghi A., Kulik M., Sa S., Rabinovitch M., Kundaje A., Dalton S., Zaugg J. B., Snyder M.. **Landscape of cohesin-mediated chromatin loops in the human genome.**. *Nature.* (2020.0) **583** 737-743. PMID: 32728247
76. Van Nostrand E. L., Freese P., Pratt G. A., Wang X., Wei X., Xiao R., Blue S. M., Chen J.-Y., Cody N. A. L., Dominguez D., Olson S., Sundararaman B., Zhan L., Bazile C., Bouvrette L. P. B., Bergalet J., Duff M. O., Garcia K. E., Gelboin-Burkhart C., Hochman M., Lambert N. J., Li H., McGurk M. P., Nguyen T. B., Palden T., Rabano I., Sathe S., Stanton R., Su A., Wang R., Yee B. A., Zhou B., Louie A. L., Aigner S., Fu X.-D., Lécuyer E., Burge C. B., Graveley B. R., Yeo G. W.. **A large-scale binding and functional map of human RNA-binding proteins.**. *Nature.* (2020.0) **583** 711-719. PMID: 32728246
77. Fullard J. F., Giambartolomei C., Hauberg M. E., Xu K., Voloudakis G., Shao Z., Bare C., Dudley J. T., Mattheisen M., Robakis N. K., Haroutunian V., Roussos P.. **Open chromatin profiling of human postmortem brain infers functional roles for non-coding schizophrenia loci.**. *Hum. Mol. Genet.* (2018.0). DOI: 10.1093/hmg/ddy229
78. de la Torre-Ubieta L., Stein J. L., Won H., Opland C. K., Liang D., Lu D., Geschwind D. H.. **The Dynamic Landscape of Open Chromatin during Human Cortical Neurogenesis.**. *Cell.* (2018.0) **172** 289-304. PMID: 29307494
79. Fullard J. F., Hauberg M. E., Bendl J., Egervari G., Cirnaru M.-D., Reach S. M., Motl J., Ehrlich M. E., Hurd Y. L., Roussos P.. **An atlas of chromatin accessibility in the adult human brain.**. *Genome Res.* (2018.0) **28** 1243-1252. PMID: 29945882
80. Bryois J., Garrett M. E., Song L., Safi A., Giusti-Rodriguez P., Johnson G. D., Shieh A. W., Buil A., Fullard J. F., Roussos P., Sklar P., Akbarian S., Haroutunian V., Stockmeier C. A., Wray G. A., White K. P., Liu C., Reddy T. E., Ashley-Koch A., Sullivan P. F., Crawford G. E.. **Evaluation of chromatin accessibility in prefrontal cortex of individuals with schizophrenia.**. *Nat. Commun.* (2018.0) **9** 3121. PMID: 30087329
81. Wang D., Liu S., Warrell J., Won H., Shi X., Navarro F. C. P., Clarke D., Gu M., Emani P., Yang Y. T., Xu M., Gandal M. J., Lou S., Zhang J., Park J. J., Yan C., Rhie S. K., Manakongtreecheep K., Zhou H., Nathan A., Peters M., Mattei E., Fitzgerald D., Brunetti T., Moore J., Jiang Y., Girdhar K., Hoffman G. E., Kalayci S., Gümüş Z. H., Crawford G. E., Roussos P., Akbarian S., Jaffe A. E., White K. P., Weng Z., Sestan N., Geschwind D. H., Knowles J. A., Gerstein M. B.. **Comprehensive functional genomic resource and integrative model for the human brain.**. *Science.* (2018.0) **362**. DOI: 10.1126/science.aat8464
82. Hopkins A. L., Groom C. R.. **The druggable genome.**. *Nat. Rev. Drug Discov.* (2002.0) **1** 727-730. PMID: 12209152
83. Rentzsch P., Witten D., Cooper G. M., Shendure J., Kircher M.. **CADD: predicting the deleteriousness of variants throughout the human genome.**. *Nucleic Acids Res.* (2019.0) **47** D886-D894. PMID: 30371827
84. Katoh K., Standley D. M.. **MAFFT multiple sequence alignment software version 7: improvements in performance and usability.**. *Mol. Biol. Evol.* (2013.0) **30** 772-780. PMID: 23329690
85. Jones C. T., Swingler R. J., Brock D. J.. **Identification of a novel SOD1 mutation in an apparently sporadic amyotrophic lateral sclerosis patient and the detection of Ile113Thr in three others.**. *Hum. Mol. Genet.* (1994.0) **3** 649-650. PMID: 8069312
86. Draper A. C. E., Wilson Z., Maile C., Faccenda D., Campanella M., Piercy R. J.. **Species-specific consequences of an E40K missense mutation in superoxide dismutase 1 (SOD1).**. *FASEB J.* (2020.0) **34** 458-473. PMID: 31914665
87. Auton A., Brooks L. D., Durbin R. M., Garrison E. P., Kang H. M., Korbel J. O., Marchini J. L., McCarthy S., McVean G. A., Abecasis G. R.. **A global reference for human genetic variation.**. *Nature.* (2015.0) **526** 68-74. PMID: 26432245
88. Bulik-Sullivan B. K., Loh P.-R., Finucane H. K., Ripke S., Yang J., Patterson N., Daly M. J., Price A. L., Neale B. M.. **LD Score regression distinguishes confounding from polygenicity in genome-wide association studies.**. *Nature Genetics.* (2015.0) **47** 291-295. PMID: 25642630
89. Bulik-Sullivan B., Finucane H. K., Anttila V., Gusev A., Day F. R., Loh P.-R., Duncan L., Perry J. R. B., Patterson N., Robinson E. B., Daly M. J., Price A. L., Neale B. M.. **Psychiatric Genomics Consortium, Genetic Consortium for Anorexia Nervosa of the Wellcome Trust Case Control Consortium, An atlas of genetic correlations across human diseases and traits.**. *Nature Genetics.* (2015.0) **47** 1236-1241. PMID: 26414676
90. Wang G., Sarkar A., Carbonetto P., Stephens M.. **A simple new approach to variable selection in regression, with application to genetic fine mapping.**. *Journal of the Royal Statistical Society: Series B (Statistical Methodology).* (2020.0) **82** 1273-1300
91. Petrovski S., Gussow A. B., Wang Q., Halvorsen M., Han Y., Weir W. H., Allen A. S., Goldstein D. B.. **The Intolerance of Regulatory Sequence to Genetic Variation Predicts Gene Dosage Sensitivity.**. *PLoS Genet.* (2015.0) **11** e1005492. PMID: 26332131
92. Grant B. J., Skjaerven L., Yao X.-Q.. **The Bio3D packages for structural bioinformatics.**. *Protein Sci.* (2021.0) **30** 20-30. PMID: 32734663
93. McInnes L., Healy J., Melville J.. *UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction* (2018.0)
94. Hahsler M., Piekenbrock M., Doran D.. **dbscan: Fast Density-Based Clustering with R.**. *J. Stat. Softw.* (2019.0) **91**. DOI: 10.18637/jss.v091.i01
95. Howe K. L., Achuthan P., Allen J., Allen J., Alvarez-Jarreta J., Amode M. R., Armean I. M., Azov A. G., Bennett R., Bhai J., Billis K., Boddu S., Charkhchi M., Cummins C., Da Rin Fioretto L., Davidson C., Dodiya K., El Houdaigui B., Fatima R., Gall A., Garcia Giron C., Grego T., Guijarro-Clarke C., Haggerty L., Hemrom A., Hourlier T., Izuogu O. G., Juettemann T., Kaikala V., Kay M., Lavidas I., Le T., Lemos D., Gonzalez Martinez J., Marugán J. C., Maurel T., McMahon A. C., Mohanan S., Moore B., Muffato M., Oheh D. N., Paraschas D., Parker A., Parton A., Prosovetskaia I., Sakthivel M. P., Salam A. I. A., Schmitt B. M., Schuilenburg H., Sheppard D., Steed E., Szpak M., Szuba M., Taylor K., Thormann A., Threadgold G., Walts B., Winterbottom A., Chakiachvili M., Chaubal A., De Silva N., Flint B., Frankish A., Hunt S. E., IIsley G. R., Langridge N., Loveland J. E., Martin F. J., Mudge J. M., Morales J., Perry E., Ruffier M., Tate J., Thybert D., Trevanion S. J., Cunningham F., Yates A. D., Zerbino D. R., Flicek P.. **Ensembl 2021.**. *Nucleic Acids Res.* (2021.0) **49** D884-D891. PMID: 33137190
96. Han X., Zhou Z., Fei L., Sun H., Wang R., Chen Y., Chen H., Wang J., Tang H., Ge W., Zhou Y., Ye F., Jiang M., Wu J., Xiao Y., Jia X., Zhang T., Ma X., Zhang Q., Bai X., Lai S., Yu C., Zhu L., Lin R., Gao Y., Wang M., Wu Y., Zhang J., Zhan R., Zhu S., Hu H., Wang C., Chen M., Huang H., Liang T., Chen J., Wang W., Zhang D., Guo G.. **Construction of a human cell landscape at single-cell level.**. *Nature.* (2020.0) **581** 303-309. PMID: 32214235
97. **The Gene Ontology resource: enriching a GOld mine.**. *Nucleic Acids Res.* (2021.0) **49** D325-D334. PMID: 33290552
98. Koopmans F., van Nierop P., Andres-Alonso M., Byrnes A., Cijsouw T., Coba M. P., Cornelisse L. N., Farrell R. J., Goldschmidt H. L., Howrigan D. P., Hussain N. K., Imig C., de Jong A. P. H., Jung H., Kohansalnodehi M., Kramarz B., Lipstein N., Lovering R. C., MacGillavry H., Mariano V., Mi H., Ninov M., Osumi-Sutherland D., Pielot R., Smalla K.-H., Tang H., Tashman K., Toonen R. F. G., Verpelli C., Reig-Viader R., Watanabe K., van Weering J., Achsel T., Ashrafi G., Asi N., Brown T. C., De Camilli P., Feuermann M., Foulger R. E., Gaudet P., Joglekar A., Kanellopoulos A., Malenka R., Nicoll R. A., Pulido C., de Juan-Sanz J., Sheng M., Südhof T. C., Tilgner H. U., Bagni C., Bayés À., Biederer T., Brose N., Chua J. J. E., Dieterich D. C., Gundelfinger E. D., Hoogenraad C., Huganir R. L., Jahn R., Kaeser P. S., Kim E., Kreutz M. R., McPherson P. S., Neale B. M., O’Connor V., Posthuma D., Ryan T. A., Sala C., Feng G., Hyman S. E., Thomas P. D., Smit A. B., Verhage M.. **SynGO: An Evidence-Based, Expert-Curated Knowledge Base for the Synapse.**. *Neuron.* (2019.0) **103** 217-234. PMID: 31171447
99. Sanyal A., Lajoie B. R., Jain G., Dekker J.. **The long-range interaction landscape of gene promoters.**. *Nature.* (2012.0) **489** 109-113. PMID: 22955621
100. Innan H., Kondrashov F.. **The evolution of gene duplications: classifying and distinguishing between models.**. *Nat. Rev. Genet.* (2010.0) **11** 97-108. PMID: 20051986
101. McKenna A., Hanna M., Banks E., Sivachenko A., Cibulskis K., Kernytsky A., Garimella K., Altshuler D., Gabriel S., Daly M., DePristo M. A.. **The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data.**. *Genome Res.* (2010.0) **20** 1297-1303. PMID: 20644199
102. Casper J., Zweig A. S., Villarreal C., Tyner C., Speir M. L., Rosenbloom K. R., Raney B. J., Lee C. M., Lee B. T., Karolchik D., Hinrichs A. S., Haeussler M., Guruvadoo L., Navarro Gonzalez J., Gibson D., Fiddes I. T., Eisenhart C., Diekhans M., Clawson H., Barber G. P., Armstrong J., Haussler D., Kuhn R. M., Kent W. J.. **The UCSC Genome Browser database: 2018 update.**. *Nucleic Acids Res.* (2018.0) **46** D762-D769. PMID: 29106570
103. Thomas-Chollier M., Hufton A., Heinig M., O’Keeffe S., Masri N. E., Roider H. G., Manke T., Vingron M.. **Transcription factor binding predictions using TRAP for the analysis of ChIP-seq data and regulatory SNPs.**. *Nat. Protoc.* (2011.0) **6** 1860-1869. PMID: 22051799
104. Portales-Casamar E., Thongjuea S., Kwon A. T., Arenillas D., Zhao X., Valen E., Yusuf D., Lenhard B., Wasserman W. W., Sandelin A.. **JASPAR 2010: the greatly expanded open-access database of transcription factor binding profiles.**. *Nucleic Acids Res.* (2010.0) **38** D105-10. PMID: 19906716
105. Bandopadhayay P., Bergthold G., Nguyen B., Schubert S., Gholamin S., Tang Y., Bolin S., Schumacher S. E., Zeid R., Masoud S., Yu F., Vue N., Gibson W. J., Paolella B. R., Mitra S. S., Cheshier S. H., Qi J., Liu K.-W., Wechsler-Reya R., Weiss W. A., Swartling F. J., Kieran M. W., Bradner J. E., Beroukhim R., Cho Y.-J.. **BET bromodomain inhibition of MYC-amplified medulloblastoma.**. *Clin. Cancer Res.* (2014.0) **20** 912-925. PMID: 24297863
|
---
title: Microglia play beneficial roles in multiple experimental seizure models
authors:
- Synphane Shelton-Gibbs
- Jordan Benderoth
- Ronald P. Gaykema
- Justyna Straub
- Kenneth A. Okojie
- Joseph O. Uweru
- Dennis H. Lentferink
- Binita Rajbanshi
- Maureen N. Cowan
- Brij Patel
- Anthony Brayan Campos-Salazar
- Edward Perez-Reyes
- Ukpong B. Eyo
journal: bioRxiv
year: 2023
pmcid: PMC10028974
doi: 10.1101/2023.03.04.531090
license: CC BY 4.0
---
# Microglia play beneficial roles in multiple experimental seizure models
## Abstract
Seizure disorders are common, affecting both the young and the old. Currently available antiseizure drugs are ineffective in a third of patients and have been developed with a focus on known neurocentric mechanisms, raising the need for investigations into alternative and complementary mechanisms that contribute to seizure generation or its containment. Neuroinflammation, broadly defined as the activation of immune cells and molecules in the central nervous system (CNS), has been proposed to facilitate seizure generation, although the specific cells involved in these processes remain inadequately understood. The role of microglia, the primary inflammation-competent cells of the brain, is debated since previous studies were conducted using approaches that were less specific to microglia or had inherent confounds. Using a selective approach to target microglia without such side effects, we show a broadly beneficial role for microglia in limiting chemoconvulsive, electrical, and hyperthermic seizures and argue for a further understanding of microglial contributions to contain seizures.
## INTRODUCTION
Seizure disorders affect 65–70 million individuals globally. Prolonged seizures, known as status epilepticus, can cause significant brain alterations and even damage. Notably, seizures are often self-limiting and understanding endogenous mechanisms by which seizures develop and lead to chronic epilepsy can offer insights into developing promising therapeutic approaches. Development of antiepileptic drugs (AEDs) have been based on targeting neuronal mechanisms of excitability. However, while neurons are central to the development of seizures, neurons do not function independent of other cells and other cellular factors. Indeed, a network of glia, traditionally considered the support cells of the CNS, are critical for proper neuronal development and function (Allen & Lyons, 2018; Thion, Ginhoux, & Garel, 2018). Interestingly, there is growing evidence that glia contribute to epileptic phenotypes (Eyo, Murugan, & Wu, 2017; Robel & Sontheimer, 2016).
Inflammation has been implicated in both sterile and infection-associated seizures (Devinsky, Vezzani, Najjar, De Lanerolle, & Rogawski, 2013; Vezzani, 2014; Vezzani, Aronica, Mazarati, & Pittman, 2013) including paraneoplastic encephalitis (Serafini et al., 2016) and *Rasmussen encephalitis* *Rasmussen encephalitis* (Cay-Martinez, Hickman, McKhann Ii, Provenzano, & Sands, 2020). For example, mice lacking interleukin-1 receptor (IL1R1), the receptor to IL-1α and IL-1β, are more resistant to seizures (C. Dube, Vezzani, Behrens, Bartfai, & Baram, 2005). Notably, IL-1β infusions enhanced seizures, while infusions of IL-1ra, an endogenous IL-1 antagonist mitigated seizures (Heida & Pittman, 2005). Similarly, TNF-α infusions in neonatal rats facilitated chemoconvulsive seizure development in adulthood, while treatment of mice with function blocking TNF-α antibodies reduced experimental seizures in adulthood (Galic et al., 2008). Moreover, interferon-alpha (IFN-α) has been implicated in febrile seizures (Masuyama et al., 2002). Thus, inflammatory cytokines like IL-1α, IL-1β, TNF-α and IL-1α facilitate the initiation of seizures and subsequent progression to epilepsy.
Despite the above observations, inhibiting inflammation during seizures has yielded conflicting results. For example, COX inhibitors failed to reduce epileptogenesis (Holtman, van Vliet, Edelbroek, Aronica, & Gorter, 2010; Holtman et al., 2009) and dexamethasone (DEX), a broad acting anti-inflammatory agent, gave conflicting results in seizure severity (Al-Shorbagy, El Sayeh, & Abdallah, 2012; Duffy, Chun, Ma, Lythgoe, & Scott, 2014; Marchi et al., 2011). Moreover, cytokines are sometimes significantly elevated (Haspolat et al., 2002; Virta, Hurme, & Helminen, 2002), unchanged (Lahat, Livne, Barr, & Katz, 1997), or partially elevated (Ichiyama, Nishikawa, Yoshitomi, Hayashi, & Furukawa, 1998) in children following febrile seizures, raising questions as to the power of the cytokine hypothesis in explaining seizure generation. Poignantly, inflammation can be induced by different cell types including astrocytes, microglia, and peripheral cells that may differ in their actions. For example, IL-1β is principally produced by astrocytes and not microglia during seizures (C. M. Dube et al., 2010). Therefore, broadly targeting inflammation may mask distinct roles employed by each of these cells during seizures raising a need for more selective cell-type specific targeting.
Microglia are the primary brain-resident inflammation-competent cells. Conflicting experimental results suggest that microglia either promote seizures (Abraham, Fox, Condello, Bartolini, & Koh, 2012; Di Nunzio et al., 2021; Eun, Abraham, Mlsna, Kim, & Koh, 2015; Kim et al., 2015) or dampen neuronal hyperactivity (Cserep et al., 2020; Kato et al., 2016; Y. Li, Du, Liu, Wen, & Du, 2012; Merlini et al., 2021) and perform seizure-limiting functions as evidenced when they are eliminated by pharmacogenetic approaches (Mirrione et al., 2010; Wu et al., 2020). However, concerns about these pharmacogenetic elimination approaches have arisen with evidence that they cause disrupted structural brain features, behavioral, and cellular abnormalities, as well as increased cytokine release and glial reactivity (Bedolla et al., 2022; Rubino et al., 2018). These results confound the interpretation of previous findings as to the roles microglia play in seizures. Moreover, the extent to which microglia regulate seizure phenotypes have been explored mainly in chemoconvulsive seizure paradigms while their roles in alternative seizure paradigms have remained largely unexplored. In the current study, we used a pharmacological approach that eliminates brain microglia without altering either cytokine levels, glial reactivity, or affecting behavior. We then applied this approach in multiple seizure models and provide compelling results to argue for beneficial roles for microglia in limiting seizures.
## Animals
All animal experiments were carried out pursuant to the relevant guidelines and regulations of the University of Virginia and approved by the Institutional Animal Care and Use Committee with protocol number 4237–08-21. The animals were housed under controlled temperature, humidity, and light (12:12 hr. light: dark cycle), with food and water readily available ad libitum. This study used both male and female mice on a C57BL/6J background between 6–12 weeks of age, and consisted of the following genotypes: CX3CR1GFP/+ expressing GFP under control of the fractalkine receptor (CX3CR1) promoter (Jung et al., 2000); and CX3CR1Cre/+:Rosa26iDTR/+ mice (Buch et al., 2005; Parkhurst et al., 2013); VGAT-Cre mice (Vong et al., 2011); TRAP2 mice capable of expressing tamoxifen-inducible tdTomato under the control of cFos promoter (DeNardo et al., 2019) and C57Bl/6J mice as wildtype mice. Mice were housed in groups for the experiment without special environmental enrichment.
## Pharmacogenetic elimination
Both CX3CR1Cre/+ and CX3CR1Cre/+:Rosa26iDTR/+ mice received two doses of diphtheria toxin (DT, 50 μg/kg) by intraperitoneal injection at 48h intervals and seizure experiments were conducted 24 h after the last DT injection.
## Pharmacological elimination
Microglia were depleted using a potent CSF1R inhibitor, PLX3397, that has been previously shown to lack the inflammatory consequences during the elimination process. Microglia from the adult brain were depleted by feeding adult mice with chow containing PLX3397 (660 mg/kg) for 7 days. Similarly, for febrile seizure studies, pre-weaned P10-P12 pups were administered 50μl of PLX3397 (40 mg/kg) or vehicle ($10\%$ DMSO; $50\%$ polyethylene glycol; $40\%$ saline) by daily intraperitoneal (i.p.) injections from P10–12. For studies using TRAP2 mice to label hyperactive neurons at the desired time points, tdTomato expression coupled to cFos expression was induced by administering mice with a single intraperitoneal (i.p.) injection of tamoxifen (40 mg/kg) dissolved in corn oil. Mice were allowed to return to their home cages for another week, after which they were perfused, and used for further imaging and analysis.
## Chemoconvulsive seizure induction
For chemoconvulsive seizure initiation model, 2–4-month-old mice were used, and i.p. injected with kainic acid (KA) at 24–27 mg/kg body weight. Control mice were injected with equal volume of saline that was used as the vehicle to dissolve the kainic acid. Seizures were scored using a modified Racine scale as follows: [1] freezing behavior; [2] rigid posture with raised tail; [3] continuous head bobbing and forepaws shaking; [4] rearing, falling, and jumping; [5] continuous occurrence of level 4; and [6] loss of posture and generalized convulsion activity (Avignone, Ulmann, Levavasseur, Rassendren, & Audinat, 2008; Eyo et al., 2014; Racine, 1972). Scores were recorded every 5 minutes for up to 4 hours. Scores were given based on the summary of the mouse behavior over the 5 minute interval. The experimenter was “blind” to the prior condition of the mice. The seizure scores are presented as the median of the scores and the area under the curve (AUC) was analyzed.
## Febrile seizure induction
For hyperthermia-induced febrile seizure induction, P13-P15 mice pups were placed in a fiberglass hyperthermia chamber heated by means of an overhead heat lamp and a basal heating plate. Temperatures of the two heat sources was precisely regulated by a temperature regulator. In experiments measuring the threshold temperature to seizure initiation, the body temperature of the pups was monitored using a rectal temperature probe connected to a digital thermometer. Baseline activity of mice was first recorded for 15–30 minutes at normothermic conditions 35°C, following which hyperthermia was induced by subjecting them to temperatures around 41–43°C, for 30 minutes, or until seizures were induced. Video monitoring and recording of the seizure induction was carried out. Latency to first signs of inactivity and convulsions, and temperature threshold for seizure initiation were all recorded and used to compare seizure outcomes between experimental and control groups.
## EEG implantation, recordings, and electrical stimulation protocols
Surgeries were performed on 8–12 week-old mice using isoflurane anesthesia. Ketoprofen was used as an anesthetic for the surgery. An EEG recording headset included bipolar depth electrodes for electrical stimulation composed of Teflon-coated stainless-steel wire (A-M Systems, diameter = 0.008”, #791400). Electrodes were soldered to a plastic pedestal (Plastics One) and secured to the skull with dental cement. A Kopf stereotaxic apparatus was used to place the depth electrodes (from bregma): 3 mm posterior, 3 mm lateral, and 3 mm depth.
For EEG recording, mice were connected to a video-EEG monitoring system (AURA LTM64 using TWin software, Grass) via a flexible cable and commutator (Plastics One). One day later, the after-discharge threshold (ADT) was measured. The hippocampal depth electrode was connected to a constant current stimulator (A-M Systems, Model 2100), which delivered a 2 s train of 1-ms pulses at 50 Hz. The current was initially set at 10 μA and increased in 10-μA increments until an electrographic discharge was observed (20–120 μA).
For kindling of VGAT-Cre mice, the current intensity was set to 1.5 × the magnitude of ADT for that mouse and was delivered six times per day separated by at least 1 hour. Animals were considered fully kindled when stimulations evoked five consecutive seizures that triggered 5 motor seizures with bilateral clonus and loss of posture control. Animals were monitored by video/EEG throughout the experiment (24 h/d, 7 d/wk). Spontaneous seizures were defined as high amplitude spike-wave discharges with >2-Hz frequency lasting at least 15 s. Electrographic seizures were verified by examining the corresponding video and a behavioral score was assigned. All seizures reported in this study had a motor component.
Similar methods were used for continuous hippocampal stimulation (CHS) including electroencephalography and behavior patterns during experimental status epilepticus as detailed previously (Lewczuk et al., 2018). These studies used C57BL/6J mice obtained from Jackson Labs ($$n = 32$$). Only males were used since female reproductive cycles affect seizure frequency (Joshi et al., 2018). The electrical stimulation protocol is similar to kindling, but the duration is increased to 10s and the total train duration is 15 s. The 5 s off interval is used to assess brain activity. Typically, brain activity progresses from: 0 spikes (post-ictal depression); to 1 spike (~13 min); to 2 spikes (~18 min); to 3 spikes (21 min); to 4 spikes (24 min); 5 spikes (27 min); and to 6 spikes (29 min). The slope of this seizure progression is a key metric for seizure susceptibility (see Fig. 3b). Intermittently during CHS, mice would have a discrete seizure lasting over 30 s; another metric for seizure susceptibility (see Fig. 3c). After 30 min the stimulation was stopped, because the mice entered status epilepticus. In this study, a large proportion of mice died after 30 min of status due to tonic phase apnea, as observed in other mouse models of epilepsy (Wenker et al., 2021).
To study the effect of microglia depletion on epilepsy we electrically kindled VGAT-Cre mice after IP injection of 10–15 mg/kg kainic acid (hybrid kindling). Mice were maintained on control chow (low PLX3397) until the mice developed epilepsy, which is defined as having 2 or more spontaneous seizures (King et al., 1998). Mice were then randomized to stay on control chow or switched to high dose PLX3397 chow ($$n = 14$$). Video/EEG recording were continued for at least 3 weeks to quantify spontaneous recurring seizures (SRS). As noted previously, epilepsy in VGAT-Cre mice spontaneously remits (Straub, Vitko, Gaykema, & Perez-Reyes, 2021).
## Open Field Behavioral Test
The open field test was carried out in a custom-built arena using a white plastic material with 35 (L) × 35 (B) × 21 (H) cm dimensions. Mouse cages were moved into the testing room and allowed to acclimate for 1 hour. Illumination in the room was maintained at 150 Lux intensity, temperature and relative humidity were also relatively constant at 70.2 ± 0.9°F and 40.9 ± $5\%$, respectively. All experiments were done during the light cycle. Four to five mice from the same cage were simultaneously placed into different arenas that had been cleaned with $70\%$ ethanol. Each mouse was placed adjacent to the wall of the arena and allowed to freely explore the space. Their open field activities (horizontal locomotion and mobility) in the arenas were video monitored for 10 min using EthoVision® XT (Noldus, Wagenigen, Netherlands), and then subsequently tracked offline for activity analysis using the same software.
## Chronic window implantation and two photon imaging
Mice were implanted with a chronic cranial window as previously described (Bisht, Sharma, & Eyo, 2020). Briefly, during surgery, mice were anesthetized with isoflurane ($5\%$ for induction; 1–$2\%$ for maintenance) and placed on a heating pad. Mice were treated subcutaneously at the site of surgery with 100μL of $0.25\%$ bupivacaine as a local anesthetic before cutting into the skin above the head for the craniotomy. Using a dental drill, a circular craniotomy of > 3 mm diameter was drilled at 2 mm posterior and 1.5 mm lateral to bregma, the craniotomy center was around the limb/trunk region of the somatosensory cortex. A $70\%$ ethanol-sterilized 3mm glass coverslip was placed inside the craniotomy. A light-curing dental cement (Tetric EvoFlow) was applied and cured with a Kerr Demi Ultra LED Curing Light (DentalHealth Products). iBond Total Etch glue (Heraeus) was applied to the rest of the skull, except for the region with the window. This was also cured with the LED light. The light-curing dental glue was used to attach a custom-made head bar onto the other side of the skull from which the craniotomy was performed. Mice were treated subcutaneously with buprenorphine slow release (SR) immediately after the surgery as an analgesic. Mice were allowed to recover from anesthesia for 10 minutes on a heating pad before returning to their home cage. Mice were allowed to recover from the cranial window surgery for 2–4 weeks before commencement of chronic imaging. Only surviving mice with a clear glass window were used for the imaging studies.
For chronic imaging, mice were anesthetized with isoflurane. The head of the anesthetized mice was stabilized and mounted by the head plate and the animal was placed on a heating plate at ~35°C under the two-photon microscope. A hundred μl of Rhodamine B dye (2 mg/mL) was injected intraperitoneally to label the vasculature. As previously described, for longitudinal imaging, the blood vessel architecture visible through the craniotomy window was carefully recorded as a precise map of the brain region being visualized and was used to trace back to the original imaging site for chronic imaging studies (Bisht et al., 2020). Imaging was conducted using a Leica SP8 Multiphoton microscope with a coherent laser. A wavelength of 880 nm was optimal for imaging both microglia and the blood vessel dye as well as tdTomato. The power output at the brain was maintained at 25 mW or below. Images were collected at a 1024 × 1024 pixel resolution using a 25X 0.9 NA objective at a 1.5X optical zoom. Several fields of view of z-stack images were collected every 1–2 μm through a volume of tissue and used for analysis. To observe microglial dynamics, z-stack images were acquired every minute at 2 μm steps in depth.
## Tissue preparation and immunostaining
For confocal microscopy studies, mice were anesthetized with $5\%$ isoflurane, and transcardially perfused with sodium phosphate buffer (PBS; 50 mM at pH 7.4) followed by $4\%$ paraformaldehyde (PFA). All perfusion solutions were chilled on ice prior to use. We used a perfusion pump (Masterflex ® Ismatec ®) at a perfusion flow rate of 7mL/min. Brains were then fixed in $4\%$ PFA overnight. Using a vibratome (Leica VT100S), 50μm thick sections of the brain were cut in chilled PBS. Slices were then stored in cryoprotectant ($40\%$ PBS, $30\%$ ethylene glycol, and $30\%$ glycerol) at −20°C while further processing took place. Brain sections containing the ventral hippocampus CA1 (Bregma −3.27 and −4.03), the frontal cortex (Bregma 2.93 and −2.57), and sensorimotor cortex (Bregma −2.5 and +2.0) were examined. For WFA (1:800, Wako, # 019–19741) immunohistochemical staining for fluorescence microscopy analysis, brain sections were washed in PBS, blocked with blocking buffer, and incubated overnight at 4°C with primary antibody solution against WFA. After washing, brain slices were incubated in appropriate secondary antibody solution for 1 hour at room temperature
## Fluorescence Microscopy
Fluorescently immunolabelled brain sections were imaged with a confocal microscope or an EVOS microscope. Image analysis to quantify the WFA area, cell number staining intensity as well as the number of cFos+ cells through z-stacks was done using ImageJ. WFA cell number was automatically counted using the ImageJ cell counter plugin. The WFA staining intensity was quantified as a measure of the florescence intensity for WFA and normalized to the area of the interest. The perineuronal net (PNN) area was quantified as a ration of the total area occupied by WFA signal over the total area of the tissue examined. cFos+ cells were identified in z-stack images as distinct from Rhodamine labelled blood vessels in two ways. First, morphological distinctions through the z-stack could be observed as blood vessels are tubular and can be determined through the z-stack and cFos+ cells often displayed a cell body with radiating florescent signals. Second, while blood vessels were always present in images from adjacent weeks, cFos+ cells sometimes were absent in a previous week and appeared in a subsequent week.
## Tissue processing and flow cytometry
Mice were euthanized with CO2 and sprayed with $70\%$ ethanol. The spleen was collected and placed in cold complete RPMI media (cRPMI) ($10\%$ FBS, $1\%$ Sodium pyruvate, $1\%$ non-essential amino acids, $1\%$ penicillin/streptomysin, $0.1\%$ 2-ME). The spleen was mashed through a 40-μm filter in 50mL conical using a syringe plunger and through with 15-mL cRPMI. The suspension was centrifuged at 1600 rpm for 5 min at 4°C using Eppendorf Centrifuge 5804R with an S-4–72 rotor. The spleen was resuspended in 2mL RBC lysis buffer for 2 min and the reaction was stopped by adding 13mL cRPMI. The suspension was once again centrifuged, and the resulting pellet were resuspended in 5mL cRPMI and kept on ice. The femur and tibia were carefully removed without splintering them to obtain the bone marrow. The bones were placed in a petri dish containing $70\%$ ethanol for 1–2 minutes to disinfect and transferred to a petri dish containing 4mL cRPMI on ice. The femora and tibiae were flushed with 10mL ice cold cRPMI using a 21G needle attached to a 10mL syringe. A single-cell suspension was generated by gently triturating the cells through the needle until large clumps were no longer present and the suspension was ran through a 40μm filter in 50mL conical tube. The suspension was centrifuge at 1600 rpm for 5 min at 4°C. The bone marrow was resuspended in 2mL RBC lysis buffer for 2 min and the reaction was stopped by adding 13mL cRPMI. The suspension was centrifuged at 1600 rpm for 5 min at 4°C, the supernatant was aspirated and the pellet was resuspended in 5mL cRPMI. Following the generation of a single-cell suspension, cells were counted using automated cell counter (C100, RWD, China), 150 μL (~1×10E6) of each sample were placed in a 96-well plate and incubated for 10 minutes in 50 μL Fc block (1:1000, CD$\frac{16}{32}$, Clone 93, eBioscience) at room temperature. Cells were then incubated in primary antibodies at a concentration of 1:200 and fixable viability dye eFluor 506 (eBioscience) at a concentration of 1:800 for 30 minutes at 4°C. Antibody clones used for experiments included: NK1.1 (PK136), CD3e (145–2C11), CD8a (53–6.7), CD4 (RM4–5), CD19 (eBio 1D3), Ly6C (HK1.4), F$\frac{4}{80}$ (BM8), CD11b (M$\frac{1}{70}$), Ly6G (1A8), CD45 (30-F11), MHC II (M$\frac{5}{114.15.2}$), and CD11c (N418) (Invitrogen). After staining, cells were washed twice and fixed overnight in $2\%$ PFA at 4°C. Cells were washed twice with cell staining buffer (BioLegend, Cat. # 420201) and transferred into a 5mL Polystyrene round-bottom tube with cell-strainer cap tubes (Falcon), then were analyzed on a Gallios flow cytometer (Beckman-Coulter) by gating on 250,000 live events. Flow cytometry data was analyzed using FlowJo (version 10.8.1).
## Statistical analysis
Data were initially measured for normality and homoscedasticity and upon comparing normal distributions and variances further analyzed with the respective tests. Student’s t-test was used to compare two-groups. Other comparisons were evaluated using one-way ANOVA (more than 2 groups) or two-way ANOVA (experiments with 2 variables), followed by suitable post hoc test for multiple comparisons within the tested groups. Other specific tests are stated in the figure legend for each of the experiments.
## Pharmacogenetic microglial elimination worsens kainic acid-induced seizures.
To begin, we attempted to confirm previous results showing that pharmacogenetic microglial elimination worsens chemoconvulsive seizures (Mirrione et al., 2010; Wu et al., 2020) by using a pharmacogenetic model of microglial elimination in a chemoconvulsive seizure model. We used IP injection of the chemoconvulsant, kainic acid (KA), a glutamate receptor analog at 24–27 mg/kg to investigate seizure severity (Supplemental Figure 1). CX3CR1Cre mice were crossed with Rosa26iDTR mice to generate CX3CR1Cre/+:Rosa26iDTR/+ that express the diphtheria toxin receptor (DTR) on CX3CR1+ cells (Supplemental Figure 1a). In adulthood, Rosa26iDTR/+ mice treated with saline or diphtheria toxin (DT, 50 μg/kg) and therefore not expected to have depleted microglia showed similar seizure scores when injected with KA a day after DT treatment (Supplemental Figure 1b–c) indicating that DT treatment in microglial-sufficient Rosa26iDTR/+ mice does not alter seizure severity. Interestingly, compared to saline-treated CX3CR1Cre/+:Rosa26iDTR/+ mice which are microglia-sufficient, DT-treated CX3CR1Cre/+:Rosa26iDTR/+ mice which become microglia depleted exhibited more severe seizures (Supplemental Figure 1d–e) and all the mice died shortly after an hour of KA treatment (Supplemental Figure 1d) indicating aggravated seizures with pharmacogenetic microglial elimination.
## Pharmacological microglial elimination and repopulation.
The above results are consistent with previous pharmacogenetic studies that eliminated microglia in chemoconvulsive seizures (Wu et al., 2020). However, recent findings suggest that pharmacogenetic approaches to eliminate microglia result in additional confounds ranging from altered brain ventricles as a structural abnormality, to increased neuronal cell death, to ataxic behavior, and to increased glial reactivity and cytokine levels (Bedolla et al., 2022; Rubino et al., 2018). Therefore, to clarify microglial contributions to seizures, we employed a pharmacological microglial elimination approach using treatment with PLX3397 (PLX), a CSF1R antagonist. CSF1R is required for microglial survival. Therefore, PLX treatment efficiently eliminates microglia without inflammatory reactions or altered brain structures (Bedolla et al., 2022; M. R. Elmore et al., 2014; Green, Crapser, & Hohsfield, 2020; Szalay et al., 2016). We confirmed that PLX (660mg/kg) delivered through the chow progressively eliminated microglia by 7 days (Figure 1a–b). As with PLX-treated mice that we refer to as “microglial depleted” mice, “control” mice received chow with either 0 mg/kg PLX or 75 mg/kg for 7 days. The latter concentration was previously shown to have minimal brain penetration of PLX, but significant plasma concentration that was higher than the concentration required in the brain to eliminate microglia (M. R. Elmore et al., 2014). We confirmed that the 75mg/kg PLX dose did not affect microglial density (Figure 1g) compared to the 660mg/kg dose (Figure 1b). These “control” mice we refer to as “microglia sufficient” mice. Interestingly, although PLX5622 was recently shown to reduce lymphoid and myeloid cell numbers with a 3-week treatment (Lei et al., 2020; Spiteri et al., 2022), we speculated that the prolonged treatment regimen could have yielded this result. Therefore, we used PLX at a high concentration (660mg/kg) to facilitate rapid microglial cell elimination. We found that a one-week PLX3397 treatment did not elicit significant changes to immune cell numbers in the spleen or bone marrow except for a slight but significant increase in monocyte cell numbers (Figure 1c–d). To assess whether microglia repopulate the brain rapidly following PLX3397 withdrawal, treated mice with either control or PLX chow for a week, returned all mice to control chow for another week and noted that microglia had repopulated the brain by 7 days (Figure 1e–f) Therefore, we used this approach of pharmacological microglial elimination and repopulation to interrogate microglial roles in seizure severity.
## Pharmacological microglial elimination worsens kainic acid-induced seizures.
Mice were treated with either control or PLX chow for 7 days before KA treatment (Figure 2a). PLX-treated (and therefore microglial deficient) mice showed worsened status epilepticus to KA (Figure 2b–c) when compared to control-treated (and therefore microglial sufficient) mice without altering mortality (Figure 2d). Seizure severity differences began to be apparent as early as 30 mins into KA treatment and persisted through 4 h (Figure 2b; Supplemental Figure 2a–e). Furthermore, increased seizure severity in microglial-deficient mice compared to microglial-sufficient mice was observed in both male and females (Supplemental Figure S2f–i).
Next, we examined the level of cell activity following KA treatment in the presence or absence of microglia. To this end, we used TRAP2 (targeting recombination in active populations) mice crossed with floxed tdTomato reporter mice (DeNardo et al., 2019; Madisen et al., 2010) where tamoxifen (TAM) treatment induces tdTomato expression under the control of the cFos promoter. cFos is an immediate early gene expressed by active cells. Littermate TRAP2 mice were treated with either PLX or control chow for 7 days to eliminate microglia. After this, mice were treated with KA and at 3 h, mice were injected with TAM (40 mg/kg) and then euthanized after 7 days to allow for tdTomato expression (Supplemental Figure S3a). With KA treatment, PLX-treated (and therefore microglial deficient) mice displayed increased cFos expression in both the cortex and hippocampus compared to control (and therefore microglial sufficient) mice (Supplemental Figure S3b).
To further visualize these changes in vivo, we crossed the TRAP2 mice with the microglial GFP reporter mice to generate double transgenic “TRAP-GFP” mice and conducted longitudinal in vivo two photon imaging (Supplemental Figure S3c). Consistent with the above findings in TRAP2 mice, longitudinal imaging of the same fields of view in TRAP-GFP mice revealed that KA treatment induced the expression of cFos in new cells (arrows in Supplemental Figure S3d, quantification in S3e). Interestingly, microglial elimination with PLX did not itself increase cFos expressing cell numbeers but coupled with subsequent KA treatment resulted in a noticeable labelling on several cortical cells (arrows in Supplementary Figure S3d, quantification in S3e).
Because inflammatory mediators are correlated with seizure severity (Choi, Min, & Shin, 2011; Terrone, Frigerio, Balosso, Ravizza, & Vezzani, 2019; Wang & Chen, 2018), we also evaluated inflammatory cytokine expression following KA-induced seizures in control (and therefore microglial sufficient) and PLX-treated (and therefore microglial deficient) littermate mice. We examined tissue expression of various cytokines at 3 h of KA treatment when cytokines were previously shown to be significantly elevated in this model of seizures (Avignone et al., 2008). We noted no differences between the microglial deficient and microglial sufficient groups (Supplemental Figure S4) suggesting that cytokine levels following KA treatment are maintained independent of microglia.
In addition, perineuronal nets (PNNs) are components of the extracellular matrix whose modulation can regulate seizure phenotypes (Chaunsali, Tewari, & Sontheimer, 2021). Since microglia regulate the ECM, it is therefore possible that they may influence seizure phenotypes by modulating PNNs. However, consistent with a previous report that documented a lack of gross alterations in PNN density following a one-month PLX treatment (Strackeljan et al., 2021), our PLX treatment did not alter gross perineuronal net (PNN) density or intensity in either adult (Supplemental Figure S5a–d) or neonatal mice (Supplemental Figure S5e–h) indicating that the increased KA-induced seizures in with microglial elimination occurs independent of tissue PNN density in the KA model of seizures. Together, these results indicate that a neither a cytokine-dependent or a PNN-dependent mechanism is sufficient to explain the exacerbated KA-induced seizures with microglial elimination.
## Endogenous and repopulated microglia are similarly beneficial in acute kainic acid-induced seizures.
Eliminating and repopulating microglia following PLX-induced elimination is neuro-beneficial with aging (M. R. P. Elmore et al., 2018) and following intracerebral hemorrhage (X. Li et al., 2022). Therefore, we speculated that repopulated microglia may have a stronger seizure-limiting potential than endogenous microglia. To this end we treated mice with either control or PLX chow for a week, returned all mice to control chow for another week by which time microglia had repopulated the brain (Figure 1e–f) and then exposed all mice to KA (Figure 2e). Although both cohorts of mice were microglial sufficient, we referred to the mice previously exposed to PLX as the “repopulated microglia” [Repop MG] and the mice treated with control chow as the “endogenous microglia” [Endo MG] group. Seizure severity was similar in mice with endogenous and repopulated microglia (Figure 2f–g). Because microglia are known to progressively mature during repopulation following PLX treatment (Zhan et al., 2019), we asked whether more mature repopulated microglia could show a more protective effect by conducting KA-induced seizures 4 weeks after PLX treatment when microglial maturation following PLX is complete (Zhan et al., 2019). As with one-week, mice with repopulated microglia at one-month showed similar seizure phenotypes to mice with endogenous microglia during KA seizures (Figure 2h–j). Together, these results indicate that both endogenous and repopulated microglia play seizure-limiting roles in chemoconvulsive seizures.
## Microglia play beneficial roles in the recovery from kainic acid-induced seizures.
Next, we attempted to determine microglial contributions to recovery from seizures. At 7 days of PLX or control treatment, we measured mice weight and open field behaviors. By itself, PLX did not affect mouse weight, distance travelled, or mobile activity in an open field (Figure 3a–d). Therefore, following a 7-day PLX or control treatment, both groups of mice were treated with KA and returned to control chow. We selected mice that exhibited similar seizure severity whether previously treated with control or PLX chow and monitored them for 7 days. Mouse weight and activity in an open field were monitored for 7 days (Figure 3e). For these studies, we selected only mice that showed similar seizure scores as they stayed or were in and out at least seizure stage 4 between the 1st and 3rd hours of seizures independent of PLX treatment (Figure 4f–g) to account for a similar seizure severity. Nevertheless, PLX-treated mice displayed increased weight reductions (Figure 4h), reduced distance travelled (Figure 4i), and reduced mobile activity (Figure 4j, see Supplemental Video S1–3) in an open field during the earlier phases of recovery when microglia were still reduced following PLX withdrawal (Figure 2e–g). These deficits were restored by 7 days (Figure 4f–h) when microglia had repopulated the brain (Figure 2e–g) suggesting a correlation between microglial absence/reduction and behavioral deficits on the one hand and microglial repopulation and behavioral recovery on the other. Together, these results indicate that microglia are important for containing seizures and facilitating recovery from seizures.
## Pharmacological microglial elimination worsens seizure phenotypes in electrically-induced seizure paradigms.
Next, we sought to extend our findings in the chemoconvulsive paradigm to other seizure paradigms. Therefore, we employed a continuous hippocampal stimulation (CHS) model (Lewczuk et al., 2018). Following implantation of EEG electrodes, we treated mice with PLX or control chow for 7 days and then performed CHS (Figure 5a). Here, microglial elimination accelerated the progression to status epilepticus (Figure 5b), accelerated the appearance of discrete seizures during CHS (Figure 5c), and accelerated animal mortality (Figure 5d). Furthermore, we used our recently developed VGAT-Cre model of temporal lobe epilepsy (TLE) where VGAT-Cre mice develop spontaneous seizures following mild electrical stimulation to induce kindling (Straub et al., 2020). As with CHS, following implantation of EEG electrodes, mice were treated with PLX or control chow for 7 days and a kindling regimen was implemented, which consisted of kindling six times a day with each kindling session separated by at least an hour (see Methods for details; Figure 5d). PLX-treated VCAT-Cre mice experienced a longer duration of evoked seizures (Figure 5f), and displayed an acceleration to fully the kindled state (5 evoked convulsive seizures, BSS≥5; Figure 5g). Moreover, we confirmed that PLX-treatment indeed eliminated microglia in this model (Figure 5h). Together, these results indicate that microglial elimination facilitates the development of electrically induced seizures.
## Pharmacological microglial elimination worsens seizure phenotypes in subsequent induced or spontaneous seizures.
Given the evidence above of microglia playing beneficial roles in primary or initial seizures, we sought to investigate possible roles for microglia in secondary or subsequent seizures to an initial seizure. We interrogated this possibility first using our chemoconvulsive model. Two groups of mice were treated with control chow for 7 days and exposed to KA. One group was then maintained in control chow and the second group in PLX chow to deplete their microglia. After another 7 days of the respective chow treatment, mice were treated with a second round of KA (Figure 6a) and their seizure scores monitored over time. Interestingly, seizures scores were similar between microglia-sufficient mice that experienced a single or two rounds of KA-induced seizures (Figure 6b) as well as between microglia-deficient mice that experienced a single or two rounds of seizures (Figure 6c) indicating that an initial KA-induced bout of seizures does not potentiate a second KA-induced bout of seizures 7 days later. However, in the second KA-induced bout of seizures experience, mice that lacked microglia between the first and second bouts also showed a greater seizure severity compared to mice that were microglia-sufficient (Figure 6d–e).
Furthermore, we extended these studies into a paradigm where electrically induced seizures can generate spontaneous recurrent seizures (SRS) in VGAT-Cre mice. To this end, electrode-implanted mice were kindled using a hybrid kindling paradigm (see Methods) and then allowed to recover and develop spontaneous seizures over a 2-week period while SRS were measured (Figure 6f). Consistent with our findings with KA, PLX-treated mice showed an increase in the number of SRS in the chronic phase (Figure 6g). These results in both a chemoconvulsive and electrical seizure model indicate that that microglia also play beneficial roles in subsequent/secondary seizures following an initial seizure.
## Pharmacological microglial elimination worsens seizure phenotypes in hyperthermia-induced seizures.
Finally, we sought to translate these findings to a more clinically relevant seizure paradigm. Febrile seizures (FS) are the most common neurological disorder in children globally. They occur in 2–$5\%$ of children in the United States and Western world (Leung, Hon, & Leung, 2018), $4\%$ of children in Tanzania (Winkler, Tluway, & Schmutzhard, 2013) and 7–$10\%$ of children in Asia (Byeon, Kim, & Eun, 2018; Hackett, Hackett, & Bhakta, 1997; Tsuboi, 1984). Furthermore, a subset (up to $18\%$) of these are prolonged and can have long-term effects on children including developmental delays (Hesdorffer et al., 2011; Martinos et al., 2019), increased association with neurodevelopmental disorders (Bertelsen, Larsen, Petersen, Christensen, & Dalsgaard, 2016; Gillberg, Lundstrom, Fernell, Nilsson, & Neville, 2017; Salehi, Yousefichaijan, Safi Arian, Ebrahimi, & Naziri, 2016), and the development of temporal lobe epilepsy (TLE), the most common form of focal epilepsy in adults (Cendes et al., 1993; Lewis et al., 2014; Shinnar, 2003; Yokoi et al., 2019).
Therefore, we developed a febrile seizure system where exposure of postnatal day (P)13–15 mice to increased temperatures in a hyperthermia (HT) chamber (Figure 7a) reliably elicited behavioral convulsions at temperatures ≥ 41°C (Supplemental Video S4). These convulsions correlated with electrographical seizure activity in both the cortex and hippocampus (Figure 7b-c) beginning to occur at 41°C (Figure 7d). Importantly, hyperthermia increased cFos expression which was further increased by PLX-induced microglial elimination (Figure 7e). Expectedly, compared to vehicle-treated mice, PLX-treated developing mice showed eliminated microglia (Figure 7f) and Consistent with findings in other seizure models, microglial-depleted developing mice displayed convulsions earlier (Figure 7g) and a greater percent of mice showed convulsions at 41°C (Figure 7h) and greater mortality (Figure 7i) from the hyperthermic exposure indicating exacerbated hyperthermia-induced seizures in the absence of microglia. Taken together, our results from a chemoconvulsive, two electrical, and a hyperthermia-induced seizures models with non-inflammatory microglial elimination comprehensively suggest beneficial roles for microglia in acute seizures.
## DISCUSSION
The precise role of microglia in seizure disorders has been elusive with suggested beneficial (Liu et al., 2020; Mirrione et al., 2010; Waltl et al., 2018; Wan et al., 2020; Wu et al., 2020; Zhao et al., 2020) and detrimental roles (Abraham et al., 2012; Di Nunzio et al., 2021; Eun et al., 2015; Kim et al., 2015). Using an approach that lacks inflammatory sequelae, glial reactivity, or brain structural aberrations in three different seizure paradigms, we provide comprehensive evidence for beneficial roles for microglia in mitigating seizures. Specifically, we show that: [1] endogenous or repopulated microglia are important in limiting chemoconvulsive, electrical, and hyperthermic seizures; [2] endogenous microglia facilitate recovery from chemoconvulsive seizures and [3] following an initial seizure, endogenous microglia still play seizure-limiting roles in both chemoconvulsive and electrically induced seizures.
Our studies suggest that microglia are important in regulating seizure severity and indicate that microglia are not required for seizure initiation as seizures occur in all models tested in the presence or absence of microglia. This is reasonable since neuronal excitability and/or excitation/inhibition imbalance underly seizure initiation. However, our results suggest that microglia play roles that likely facilitate seizure reduction. For example, in the chemoconvulsive model where we measured seizure severity temporally, microglial-sufficient and microglial-deficient mice show similar rises in seizure severity during the first hour. However, microglial-deficient mice show a slower decline from peak seizure severity when compared to microglial-sufficient mice. However, since our studies in the chemoconvulsive studies only measured the behavioral convulsions and not the electrical activity, this hypothesis remains to be fully validated.
Our findings are consistent with recent observations that microglia dampen seizure and network excitability (Cserep et al., 2020; Kato et al., 2016; Y. Li et al., 2012; Szalay et al., 2016) through Gi-dependent mechanisms such as microglial surveillance (Merlini et al., 2021) and well as previous results showing that microglia may facilitate the restoration of dysfunctional neuronal structures following seizures (Eyo et al., 2021). They are also consistent with other findings suggesting that microglial elimination enhances pilocarpine and pentylenetetrazol (PTZ) induced seizures and neurodegeneration (Liu et al., 2020) as well as neurodegeneration following kainic acid-induced seizures (Araki, Ikegaya, & Koyama, 2019). However, these previous studies assessed microglial roles usually in one paradigm i.e., chemoconvulsive models. Ours is therefore a first to comprehensively interrogate microglial roles in different seizure paradigms: chemoconvulsive, electrical, and hyperthermic.
Of interest, in our studies using both chemoconvulsive and electrical seizure paradigms, microglia were also protective in subsequent seizures either induced experimentally or spontaneously following kindling (Figure 6). This suggests that even in the early stages following an initial bout of seizures microglia provide beneficial roles. Since the mechanisms of seizure induction in chemoconvulsive, electrical, and hyperthermic paradigms differ but microglia are beneficial in all models, it is possible that microglia employ similar beneficial mechanism(s) that once understood could be enhanced to facilitate seizure containment and mitigation of its detrimental consequences. However, it is also possible that microglia employ different mechanisms for seizure mitigation in the different models of seizures. Therefore, while we could not find a cytokine-dependent mechanism in the chemoconvulsive model, we cannot rule out a cytokine-mediated mechanism in either the hyperthermia-induced or electrical models.
While our results provide supportive evidence for a beneficial role for microglia in several acute seizure models, the precise molecular mechanism(s) microglia employ in this context remain to be determined. Recently, microglial Gi-dependent signaling was shown to regulate seizure phenotypes through microglial surveillance (Merlini et al., 2021). Among the many Gi-coupled receptors expressed by microglia, the P2RY12 and CX3CR1 receptors have been shown to regulate injury-induced microglial surveillance (Haynes et al., 2006; Liang et al., 2009). In addition, the findings from the current study are consistent with molecular results that show that microglial P2RY12 play beneficial roles in adult chemoconvulsive (Eyo et al., 2014) and developing febrile (Wan et al., 2020) seizures. Similarly, CX3CR1-depedent mechanisms were shown to be protective beneficial in chemoconvulsive seizures (Eyo, Peng, et al., 2017). Therefore, one possible mechanism of microglial neuroprotection might be via Gi-dependent P2RY12 and/or CX3CR1 regulated seizures. The precise cellular mechanisms by which microglia dampen seizures remain undetermined. Possible mechanisms could include the clearance of excess neurotransmitters, the facilitation of neurotransmitter receptor recycling, the displacement of synapses, the secretion of anti-excitable or pro-inhibitory factors all to reduce neuronal excitability. Future studies will need to test each of these possibilities.
Importantly, it should be noted that our models are not models of epilepsy proper which takes weeks to months to develop and microglial roles in such paradigms could be different from that delineated herein. In addition, given recent evidence for microglial heterogeneity (Masuda, Sankowski, Staszewski, & Prinz, 2020), our approach to broadly eliminate microglia cannot distinguish possibly different roles for distinct subsets of microglia. Indeed, a recent study reports that selectively eliminating proliferating microglia reduced hippocampal neurodegeneration early following status epilepticus and spontaneous seizures during established epilepsy(Di Nunzio et al., 2021). These results therefore suggest that proliferating microglia are enhance neurodegeneration and reduce seizures in epilepsy. Since our approach eliminated all microglia, we cannot rule out this possibility and future more selective manipulations may need to be employed to assess microglial contributions in such contexts.
## DATA AVAILABILITY STATEMENT
Data will be made available upon reasonable request to corresponding author as well as detailed protocol provided for any specific experimental procedures.
## References
1. Abraham J., Fox P. D., Condello C., Bartolini A., Koh S.. **Minocycline attenuates microglia activation and blocks the long-term epileptogenic effects of early-life seizures**. *Neurobiol Dis* (2012) **46** 425-430. DOI: 10.1016/j.nbd.2012.02.006
2. Al-Shorbagy M. Y., El Sayeh B. M., Abdallah D. M.. **Diverse effects of variant doses of dexamethasone in lithium-pilocarpine induced seizures in rats**. *Can J Physiol Pharmacol* (2012) **90** 13-21. DOI: 10.1139/y11-096
3. Allen N. J., Lyons D. A.. **Glia as architects of central nervous system formation and function**. *Science* (2018) **362** 181-185. DOI: 10.1126/science.aat0473
4. Araki T., Ikegaya Y., Koyama R.. **Microglia attenuate the kainic acid-induced death of hippocampal neurons in slice cultures**. *Neuropsychopharmacol Rep* (2019). DOI: 10.1002/npr2.12086
5. Avignone E., Ulmann L., Levavasseur F., Rassendren F., Audinat E.. **Status epilepticus induces a particular microglial activation state characterized by enhanced purinergic signaling**. *J Neurosci* (2008) **28** 9133-9144. DOI: 10.1523/JNEUROSCI.1820-08.2008
6. Bedolla A., Taranov A., Luo F., Wang J., Turcato F., Fugate E. M., Luo Y.. **Diphtheria toxin induced but not CSF1R inhibitor mediated microglia ablation model leads to the loss of CSF/ventricular spaces in vivo that is independent of cytokine upregulation**. *J Neuroinflammation* (2022) **19** 3. DOI: 10.1186/s12974-021-02367-w
7. Bertelsen E. N., Larsen J. T., Petersen L., Christensen J., Dalsgaard S.. **Childhood Epilepsy, Febrile Seizures, and Subsequent Risk of ADHD**. *Pediatrics* (2016) **138**. DOI: 10.1542/peds.2015-4654
8. Bisht K., Sharma K., Eyo U. B.. **Precise Brain Mapping to Perform Repetitive In Vivo Imaging of Neuro-Immune Dynamics in Mice**. *J Vis Exp* (2020). DOI: 10.3791/61454
9. Buch T., Heppner F. L., Tertilt C., Heinen T. J., Kremer M., Wunderlich F. T., Waisman A.. **A Cre-inducible diphtheria toxin receptor mediates cell lineage ablation after toxin administration**. *Nat Methods* (2005) **2** 419-426. DOI: 10.1038/nmeth762
10. Byeon J. H., Kim G. H., Eun B. L.. **Prevalence, Incidence, and Recurrence of Febrile Seizures in Korean Children Based on National Registry Data**. *J Clin Neurol* (2018) **14** 43-47. DOI: 10.3988/jcn.2018.14.1.43
11. Cay-Martinez K. C., Hickman R. A., McKhann Ii G. M., Provenzano F. A., Sands T. T.. **Rasmussen Encephalitis: An Update**. *Semin Neurol* (2020) **40** 201-210. DOI: 10.1055/s-0040-1708504
12. Cendes F., Andermann F., Dubeau F., Gloor P., Evans A., Jones-Gotman M.. **Early childhood prolonged febrile convulsions, atrophy and sclerosis of mesial structures, and temporal lobe epilepsy: an MRI volumetric study**. *Neurology* (1993) **43** 1083-1087. DOI: 10.1212/wnl.43.6.1083
13. Chaunsali L., Tewari B. P., Sontheimer H.. **Perineuronal Net Dynamics in the Pathophysiology of Epilepsy**. *Epilepsy Curr* (2021) **21** 273-281. DOI: 10.1177/15357597211018688
14. Choi J., Min H. J., Shin J. S.. **Increased levels of HMGB1 and pro-inflammatory cytokines in children with febrile seizures**. *J Neuroinflammation* (2011) **8** 135. DOI: 10.1186/1742-2094-8-135
15. Cserep C., Posfai B., Lenart N., Fekete R., Laszlo Z. I., Lele Z., Denes A.. **Microglia monitor and protect neuronal function through specialized somatic purinergic junctions**. *Science* (2020) **367** 528-537. DOI: 10.1126/science.aax6752
16. DeNardo L. A., Liu C. D., Allen W. E., Adams E. L., Friedmann D., Fu L., Luo L.. **Temporal evolution of cortical ensembles promoting remote memory retrieval**. *Nat Neurosci* (2019) **22** 460-469. DOI: 10.1038/s41593-018-0318-7
17. Devinsky O., Vezzani A., Najjar S., De Lanerolle N. C., Rogawski M. A.. **Glia and epilepsy: excitability and inflammation**. *Trends Neurosci* (2013) **36** 174-184. DOI: 10.1016/j.tins.2012.11.008
18. Di Nunzio M., Di Sapia R., Sorrentino D., Kebede V., Cerovic M., Gullotta G. S., Vezzani A.. **Microglia proliferation plays distinct roles in acquired epilepsy depending on disease stages**. *Epilepsia* (2021) **62** 1931-1945. DOI: 10.1111/epi.16956
19. Dube C., Vezzani A., Behrens M., Bartfai T., Baram T. Z.. **Interleukin-1beta contributes to the generation of experimental febrile seizures**. *Ann Neurol* (2005) **57** 152-155. DOI: 10.1002/ana.20358
20. Dube C. M., Ravizza T., Hamamura M., Zha Q., Keebaugh A., Fok K., Baram T. Z.. **Epileptogenesis provoked by prolonged experimental febrile seizures: mechanisms and biomarkers**. *J Neurosci* (2010) **30** 7484-7494. DOI: 10.1523/JNEUROSCI.0551-10.2010
21. Duffy B. A., Chun K. P., Ma D., Lythgoe M. F., Scott R. C.. **Dexamethasone exacerbates cerebral edema and brain injury following lithium-pilocarpine induced status epilepticus**. *Neurobiol Dis* (2014) **63** 229-236. DOI: 10.1016/j.nbd.2013.12.001
22. Elmore M. R., Najafi A. R., Koike M. A., Dagher N. N., Spangenberg E. E., Rice R. A., Green K. N.. **Colony-stimulating factor 1 receptor signaling is necessary for microglia viability, unmasking a microglia progenitor cell in the adult brain**. *Neuron* (2014) **82** 380-397. DOI: 10.1016/j.neuron.2014.02.040
23. Elmore M. R. P., Hohsfield L. A., Kramar E. A., Soreq L., Lee R. J., Pham S. T., Green K. N. **Replacement of microglia in the aged brain reverses cognitive, synaptic, and neuronal deficits in mice**. *Aging Cell* (2018) **17** e12832. DOI: 10.1111/acel.12832
24. Eun B. L., Abraham J., Mlsna L., Kim M. J., Koh S.. **Lipopolysaccharide potentiates hyperthermia-induced seizures**. *Brain Behav* (2015) **5** e00348. DOI: 10.1002/brb3.348
25. Eyo U. B., Haruwaka K., Mo M., Campos-Salazar A. B., Wang L., Speros X. S. t., Wu L. J.. **Microglia provide structural resolution to injured dendrites after severe seizures**. *Cell Rep* (2021) **35** 109080. DOI: 10.1016/j.celrep.2021.109080
26. Eyo U. B., Murugan M., Wu L. J.. **Microglia-Neuron Communication in Epilepsy**. *Glia* (2017) **65** 5-18. DOI: 10.1002/glia.23006
27. Eyo U. B., Peng J., Murugan M., Mo M., Lalani A., Xie P., Wu L. J.. **Regulation of Physical Microglia-Neuron Interactions by Fractalkine Signaling after Status Epilepticus**. *eNeuro* (2017) **3**. DOI: 10.1523/ENEURO.0209-16.2016
28. Eyo U. B., Peng J., Swiatkowski P., Mukherjee A., Bispo A., Wu L. J.. **Neuronal hyperactivity recruits microglial processes via neuronal NMDA receptors and microglial P2Y12 receptors after status epilepticus**. *J Neurosci* (2014) **34** 10528-10540. DOI: 10.1523/JNEUROSCI.0416-14.2014
29. Galic M. A., Riazi K., Heida J. G., Mouihate A., Fournier N. M., Spencer S. J., Pittman Q. J.. **Postnatal inflammation increases seizure susceptibility in adult rats**. *J Neurosci* (2008) **28** 6904-6913. DOI: 10.1523/JNEUROSCI.1901-08.2008
30. Gillberg C., Lundstrom S., Fernell E., Nilsson G., Neville B.. **Febrile Seizures and Epilepsy: Association With Autism and Other Neurodevelopmental Disorders in the Child and Adolescent Twin Study in Sweden**. *Pediatr Neurol* (2017) **74** 80-86 e82. DOI: 10.1016/j.pediatrneurol.2017.05.027
31. Green K. N., Crapser J. D., Hohsfield L. A.. **To Kill a Microglia: A Case for CSF1R Inhibitors**. *Trends Immunol* (2020) **41** 771-784. DOI: 10.1016/j.it.2020.07.001
32. Hackett R., Hackett L., Bhakta P.. **Febrile seizures in a south Indian district: incidence and associations**. *Dev Med Child Neurol* (1997) **39** 380-384. DOI: 10.1111/j.1469-8749.1997.tb07450.x
33. Haspolat S., Mihci E., Coskun M., Gumuslu S., Ozben T., Yegin O.. **Interleukin-1beta, tumor necrosis factor-alpha, and nitrite levels in febrile seizures**. *J Child Neurol* (2002) **17** 749-751. DOI: 10.1177/08830738020170101501
34. Heida J. G., Pittman Q. J.. **Causal links between brain cytokines and experimental febrile convulsions in the rat**. *Epilepsia* (2005) **46** 1906-1913. DOI: 10.1111/j.1528-1167.2005.00294.x
35. Hesdorffer D. C., Benn E. K., Bagiella E., Nordli D., Pellock J., Hinton V., Team F. S.. **Distribution of febrile seizure duration and associations with development**. *Ann Neurol* (2011) **70** 93-100. DOI: 10.1002/ana.22368
36. Holtman L., van Vliet E. A., Edelbroek P. M., Aronica E., Gorter J. A.. **Cox-2 inhibition can lead to adverse effects in a rat model for temporal lobe epilepsy**. *Epilepsy Res* (2010) **91** 49-56. DOI: 10.1016/j.eplepsyres.2010.06.011
37. Holtman L., van Vliet E. A., van Schaik R., Queiroz C. M., Aronica E., Gorter J. A.. **Effects of SC58236, a selective COX-2 inhibitor, on epileptogenesis and spontaneous seizures in a rat model for temporal lobe epilepsy**. *Epilepsy Res* (2009) **84** 56-66. DOI: 10.1016/j.eplepsyres.2008.12.006
38. Ichiyama T., Nishikawa M., Yoshitomi T., Hayashi T., Furukawa S.. **Tumor necrosis factor-alpha, interleukin-1 beta, and interleukin-6 in cerebrospinal fluid from children with prolonged febrile seizures. Comparison with acute encephalitis/encephalopathy**. *Neurology* (1998) **50** 407-411. DOI: 10.1212/wnl.50.2.407
39. Joshi S., Sun H., Rajasekaran K., Williamson J., Perez-Reyes E., Kapur J.. **A novel therapeutic approach for treatment of catamenial epilepsy**. *Neurobiol Dis* (2018) **111** 127-137. DOI: 10.1016/j.nbd.2017.12.009
40. Jung S., Aliberti J., Graemmel P., Sunshine M. J., Kreutzberg G. W., Sher A., Littman D. R.. **Analysis of fractalkine receptor CX(3)CR1 function by targeted deletion and green fluorescent protein reporter gene insertion**. *Mol Cell Biol* (2000) **20** 4106-4114. PMID: 10805752
41. Kato G., Inada H., Wake H., Akiyoshi R., Miyamoto A., Eto K., Nabekura J.. **Microglial Contact Prevents Excess Depolarization and Rescues Neurons from Excitotoxicity**. *eNeuro* (2016) **3**. DOI: 10.1523/ENEURO.0004-16.2016
42. Kim I., Mlsna L. M., Yoon S., Le B., Yu S., Xu D., Koh S.. **A postnatal peak in microglial development in the mouse hippocampus is correlated with heightened sensitivity to seizure triggers**. *Brain Behav* (2015) **5** e00403. DOI: 10.1002/brb3.403
43. King M. A., Newton M. R., Jackson G. D., Fitt G. J., Mitchell L. A., Silvapulle M. J., Berkovic S. F.. **Epileptology of the first-seizure presentation: a clinical, electroencephalographic, and magnetic resonance imaging study of 300 consecutive patients**. *Lancet* (1998) **352** 1007-1011. DOI: 10.1016/S0140-6736(98)03543-0
44. Lahat E., Livne M., Barr J., Katz Y.. **Interleukin-1beta levels in serum and cerebrospinal fluid of children with febrile seizures**. *Pediatr Neurol* (1997) **17** 34-36. DOI: 10.1016/s0887-8994(97)00034-9
45. Lei F., Cui N., Zhou C., Chodosh J., Vavvas D. G., Paschalis E. I.. **CSF1R inhibition by a small-molecule inhibitor is not microglia specific; affecting hematopoiesis and the function of macrophages**. *Proc Natl Acad Sci U S A* (2020). DOI: 10.1073/pnas.1922788117
46. Leung A. K., Hon K. L., Leung T. N.. **Febrile seizures: an overview**. *Drugs Context* (2018) **7** 212536. DOI: 10.7573/dic.212536
47. Lewczuk E., Joshi S., Williamson J., Penmetsa M., Shan S., Kapur J.. **Electroencephalography and behavior patterns during experimental status epilepticus**. *Epilepsia* (2018) **59** 369-380. DOI: 10.1111/epi.13972
48. Lewis D. V., Shinnar S., Hesdorffer D. C., Bagiella E., Bello J. A., Chan S., Team F. S.. **Hippocampal sclerosis after febrile status epilepticus: the FEBSTAT study**. *Ann Neurol* (2014) **75** 178-185. DOI: 10.1002/ana.24081
49. Li X., Gao X., Zhang W., Liu M., Han Z., Li M., Liu Q.. **Microglial replacement in the aged brain restricts neuroinflammation following intracerebral hemorrhage**. *Cell Death Dis* (2022) **13** 33. DOI: 10.1038/s41419-021-04424x
50. Li Y., Du X.-F., Liu C.-S., Wen Z.-L., Du J.-L.. **Reciprocal Regulation between Resting Microglial Dynamics and Neuronal Activity In Vivo**. *Dev Cell* (2012) **23** 1189-1202. DOI: 10.1016/j.devcel.2012.10.027
51. Liu M., Jiang L., Wen M., Ke Y., Tong X., Huang W., Chen R.. **Microglia depletion exacerbates acute seizures and hippocampal neuronal degeneration in mouse models of epilepsy**. *Am J Physiol Cell Physiol* (2020) **319** C605-C610. DOI: 10.1152/ajpcell.00205.2020
52. Madisen L., Zwingman T. A., Sunkin S. M., Oh S. W., Zariwala H. A., Gu H., Zeng H.. **A robust and high-throughput Cre reporting and characterization system for the whole mouse brain**. *Nat Neurosci* (2010) **13** 133-140. DOI: 10.1038/nn.2467
53. Marchi N., Granata T., Freri E., Ciusani E., Ragona F., Puvenna V., Janigro D.. **Efficacy of anti-inflammatory therapy in a model of acute seizures and in a population of pediatric drug resistant epileptics**. *PLoS One* (2011) **6** e18200. DOI: 10.1371/journal.pone.0018200
54. Martinos M. M., Pujar S., O’Reilly H., de Haan M., Neville B. G. R., Scott R. C., Chin R. F. M.. **Intelligence and memory outcomes within 10years of childhood convulsive status epilepticus**. *Epilepsy Behav* (2019) **95** 18-25. DOI: 10.1016/j.yebeh.2019.03.039
55. Masuda T., Sankowski R., Staszewski O., Prinz M.. **Microglia Heterogeneity in the Single-Cell Era**. *Cell Rep* (2020) **30** 1271-1281. DOI: 10.1016/j.celrep.2020.01.010
56. Masuyama T., Matsuo M., Ichimaru T., Ishii K., Tsuchiya K., Hamasaki Y.. **Possible contribution of interferon-alpha to febrile seizures in influenza**. *Pediatr Neurol* (2002) **27** 289-292. DOI: 10.1016/s0887-8994(02)00452-6
57. Merlini M., Rafalski V. A., Ma K., Kim K. Y., Bushong E. A., Rios Coronado P. E., Akassoglou K.. **Microglial Gi-dependent dynamics regulate brain network hyperexcitability**. *Nat Neurosci* (2021) **24** 19-23. DOI: 10.1038/s41593-020-00756-7
58. Mirrione M. M., Konomos D. K., Gravanis I., Dewey S. L., Aguzzi A., Heppner F. L., Tsirka S. E.. **Microglial ablation and lipopolysaccharide preconditioning affects pilocarpine-induced seizures in mice**. *Neurobiol Dis* (2010) **39** 85-97. DOI: 10.1016/j.nbd.2010.04.001
59. Parkhurst C. N., Yang G., Ninan I., Savas J. N., Yates J. R., Lafaille J. J., Gan W.-B.. **Microglia promote learning-dependent synapse formation through brain-derived neurotrophic factor**. *Cell* (2013) **155** 1596-1609. DOI: 10.1016/j.cell.2013.11.030
60. Racine R. J.. **Modification of seizure activity by electrical stimulation. II. Motor seizure**. *Electroencephalogr Clin Neurophysiol* (1972) **32** 281-294. DOI: 10.1016/0013-4694(72)90177-0
61. Robel S., Sontheimer H.. **Glia as drivers of abnormal neuronal activity**. *Nat Neurosci* (2016) **19** 28-33. DOI: 10.1038/nn.4184
62. Rubino S. J., Mayo L., Wimmer I., Siedler V., Brunner F., Hametner S., Weiner H. L.. **Acute microglia ablation induces neurodegeneration in the somatosensory system**. *Nat Commun* (2018) **9** 4578. DOI: 10.1038/s41467-018-05929-4
63. Salehi B., Yousefichaijan P., Safi Arian S., Ebrahimi S., Naziri M.. **Comparison of Relation between Attention Deficit Hyperactivity Disorder in Children with and without Simple Febrile Seizure Admitted in Arak Central Iran**. *Iran J Child Neurol* (2016) **10** 56-61. PMID: 27843467
64. Serafini A., Lukas R. V., VanHaerents S., Warnke P., Tao J. X., Rose S., Wu S.. **Paraneoplastic epilepsy**. *Epilepsy Behav* (2016) **61** 51-58. DOI: 10.1016/j.yebeh.2016.04.046
65. Shinnar S.. **Febrile Seizures and Mesial Temporal Sclerosis**. *Epilepsy Curr* (2003) **3** 115-118. DOI: 10.1046/j.1535-7597.2003.03401.x
66. Spiteri A. G., Ni D., Ling Z. L., Macia L., Campbell I. L., Hofer M. J., King N. J. C.. **PLX5622 Reduces Disease Severity in Lethal CNS Infection by Off-Target Inhibition of Peripheral Inflammatory Monocyte Production**. *Front Immunol* (2022) **13** 851556. DOI: 10.3389/fimmu.2022.851556
67. Strackeljan L., Baczynska E., Cangalaya C., Baidoe-Ansah D., Wlodarczyk J., Kaushik R., Dityatev A.. **Microglia Depletion-Induced Remodeling of Extracellular Matrix and Excitatory Synapses in the Hippocampus of Adult Mice**. *Cells* (2021) **10**. DOI: 10.3390/cells10081862
68. Straub J., Gawda A., Ravichandran P., McGrew B., Nylund E., Kang J., E. Perez-Reyes. **Characterization of kindled VGAT-Cre mice as a new animal model of temporal lobe epilepsy**. *Epilepsia* (2020). DOI: 10.1111/epi.16651
69. Straub J., Vitko I., Gaykema R. P., Perez-Reyes E.. **Preparation and Implantation of Electrodes for Electrically Kindling VGAT-Cre Mice to Generate a Model for Temporal Lobe Epilepsy**. *J Vis Exp* (2021). DOI: 10.3791/62929
70. Szalay G., Martinecz B., Lenart N., Kornyei Z., Orsolits B., Judak L., Denes A.. **Microglia protect against brain injury and their selective elimination dysregulates neuronal network activity after stroke**. *Nat Commun* (2016) **7** 11499. DOI: 10.1038/ncomms11499
71. Terrone G., Frigerio F., Balosso S., Ravizza T., Vezzani A.. **Inflammation and reactive oxygen species in status epilepticus: Biomarkers and implications for therapy**. *Epilepsy Behav* (2019) 106275. DOI: 10.1016/j.yebeh.2019.04.028
72. Thion M. S., Ginhoux F., Garel S.. **Microglia and early brain development: An intimate journey**. *Science* (2018) **362** 185-189. DOI: 10.1126/science.aat0474
73. Tsuboi T.. **Epidemiology of febrile and afebrile convulsions in children in Japan**. *Neurology* (1984) **34** 175-181. DOI: 10.1212/wnl.34.2.175
74. Vezzani A.. **Epilepsy and inflammation in the brain: overview and pathophysiology**. *Epilepsy Curr* (2014) **14** 3-7. DOI: 10.5698/1535-7511-14.s2.3
75. Vezzani A., Aronica E., Mazarati A., Pittman Q. J.. **Epilepsy and brain inflammation**. *Exp Neurol* (2013) **244** 11-21. DOI: 10.1016/j.expneurol.2011.09.033
76. Virta M., Hurme M., Helminen M.. **Increased plasma levels of pro- and anti-inflammatory cytokines in patients with febrile seizures**. *Epilepsia* (2002) **43** 920-923. DOI: 10.1046/j.1528-1157.2002.02002.x
77. Vong L., Ye C., Yang Z., Choi B., Chua S., Lowell B. B.. **Leptin action on GABAergic neurons prevents obesity and reduces inhibitory tone to POMC neurons**. *Neuron* (2011) **71** 142-154. DOI: 10.1016/j.neuron.2011.05.028
78. Waltl I., Kaufer C., Gerhauser I., Chhatbar C., Ghita L., Kalinke U., Loscher W.. **Microglia have a protective role in viral encephalitis-induced seizure development and hippocampal damage**. *Brain Behav Immun* (2018) **74** 186-204. DOI: 10.1016/j.bbi.2018.09.006
79. Wan Y., Feng B., You Y., Yu J., Xu C., Dai H., Hu W.. **Microglial Displacement of GABAergic Synapses Is a Protective Event during Complex Febrile Seizures**. *Cell Rep* (2020) **33** 108346. DOI: 10.1016/j.celrep.2020.108346
80. Wang M., Chen Y.. **Inflammation: A Network in the Pathogenesis of Status Epilepticus**. *Front Mol Neurosci* (2018) **11** 341. DOI: 10.3389/fnmol.2018.00341
81. Wenker I. C., Teran F. A., Wengert E. R., Wagley P. K., Panchal P. S., Blizzard E. A., Patel M. K.. **Postictal Death Is Associated with Tonic Phase Apnea in a Mouse Model of Sudden Unexpected Death in Epilepsy**. *Ann Neurol* (2021) **89** 1023-1035. DOI: 10.1002/ana.26053
82. Winkler A. S., Tluway A., Schmutzhard E.. **Febrile seizures in rural Tanzania: hospital-based incidence and clinical characteristics**. *J Trop Pediatr* (2013) **59** 298-304. DOI: 10.1093/tropej/fmt022
83. Wu W., Li Y., Wei Y., Bosco D. B., Xie M., Zhao M. G., Wu L. J.. **Microglial depletion aggravates the severity of acute and chronic seizures in mice**. *Brain Behav Immun* (2020) **89** 245-255. DOI: 10.1016/j.bbi.2020.06.028
84. Yokoi S., Kidokoro H., Yamamoto H., Ohno A., Nakata T., Kubota T., Natsume J.. **Hippocampal diffusion abnormality after febrile status epilepticus is related to subsequent epilepsy**. *Epilepsia* (2019) **60** 1306-1316. DOI: 10.1111/epi.16059
85. Zhan L., Krabbe G., Du F., Jones I., Reichert M. C., Telpoukhovskaia M., Gan L.. **Proximal recolonization by self-renewing microglia re-establishes microglial homeostasis in the adult mouse brain**. *PLoS Biol* (2019) **17** e3000134. DOI: 10.1371/journal.pbio.3000134
86. Zhao X. F., Liao Y., Alam M. M., Mathur R., Feustel P., Mazurkiewicz J. E., Huang Y.. **Microglial mTOR is Neuronal Protective and Antiepileptogenic in the Pilocarpine Model of Temporal Lobe Epilepsy**. *J Neurosci* (2020) **40** 7593-7608. DOI: 10.1523/JNEUROSCI.2754-19.2020
|
---
title: Diet-Induced Glial Insulin Resistance Impairs The Clearance Of Neuronal Debris
authors:
- Mroj Alassaf
- Akhila Rajan
journal: bioRxiv
year: 2023
pmcid: PMC10028983
doi: 10.1101/2023.03.09.531940
license: CC BY 4.0
---
# Diet-Induced Glial Insulin Resistance Impairs The Clearance Of Neuronal Debris
## Abstract
Obesity significantly increases the risk of developing neurodegenerative disorders, yet the precise mechanisms underlying this connection remain unclear. Defects in glial phagocytic function are a key feature of neurodegenerative disorders, as delayed clearance of neuronal debris can result in inflammation, neuronal death, and poor nervous system recovery. Mounting evidence indicates that glial function can affect feeding behavior, weight, and systemic metabolism, suggesting that diet may play a role in regulating glial function. While it is appreciated that glial cells are insulin sensitive, whether obesogenic diets can induce glial insulin resistance and thereby impair glial phagocytic function remains unknown. Here, using a Drosophila model, we show that a chronic obesogenic diet induces glial insulin resistance and impairs the clearance of neuronal debris. Specifically, obesogenic diet exposure downregulates the basal and injury-induced expression of the glia-associated phagocytic receptor, Draper. Constitutive activation of systemic insulin release from Drosophila Insulin-producing cells (IPCs) mimics the effect of diet-induced obesity on glial draper expression. In contrast, genetically attenuating systemic insulin release from the IPCs rescues diet-induced glial insulin resistance and draper expression. Significantly, we show that genetically stimulating Phosphoinositide 3-kinase (PI3K), a downstream effector of Insulin receptor signaling, rescues HSD-induced glial defects. Hence, we establish that obesogenic diets impair glial phagocytic function and delays the clearance of neuronal debris.
## Introduction
Obesity significantly increases the risk for developing neurodegenerative disorders [1–3], yet the precise mechanisms underlying this connection remain unclear. Overconsumption of high sugar foods is the leading cause of obesity and its comorbidities including type 2 diabetes [4]. These effects are largely mediated by the breakdown of the insulin signaling pathway. Upon chronic exposure to a high-sugar diet (HSD), circulating insulin levels rise, resulting in insulin resistance, a state characterized by reduced cellular responsiveness to insulin [5]. Although HSD-induced insulin resistance has been well documented to occur in peripheral organs such as fat tissue and the liver, less is known about whether HSD-induced insulin resistance also occurs in the brain. Despite the widespread expression of insulin receptors in the brain, the brain has been historically regarded as insulin insensitive. This is largely because insulin is dispensable for glucose uptake in the brain [6] and its entry is limited by the blood-brain barrier [7, 8]. However, growing evidence suggest that insulin exerts unique regulatory actions on the brain to control cognition, feeding, and systemic metabolism[9, 10] A key feature of neurodegenerative disorders is the diminished clearance of neuronal debris and neuron-secreted toxic proteins [11]. This can lead to inflammation, secondary neuronal death, and impaired axonal regeneration. Microglia are the brain’s resident macrophages, and they play a key role in learning and memory. When activated, they can swiftly mobilize to the site of disease or neuronal injury and initiate phagocytosis [12]. However, chronic activation of microglia can lead to the progressive decline of their phagocytic capacity as seen in the aging brain—the most at risk for neurodegenerative disorders [13]. Interestingly, obese humans and animals also display chronic activation of microglia, which has been shown to contribute to neuroinflammation [14]. However, little is known about the effects of obesity on glial phagocytic function. Uncovering whether and how diet-induced obesity disrupts glial phagocytosis may shed some light on the link between obesity and neurodegenerative disorders.
It was only recently discovered that microglia express insulin receptors indicating that insulin can have a direct regulatory effect on microglial function [15, 16]. While these studies suggest a tight link between glial cell function and insulin, precisely how physiological factors like diet-induced alterations in insulin levels modulate glial function remains unclear.
Glial phagocytosis begins with the recognition of cellular debris via cell-surface receptors. Ablation of these receptors results in impaired clearance of cellular debris, while their overexpression leads to excessive neuronal pruning. Just like in mammals, Drosophila microglia-like cells, ensheathing glia [17], express phagocytic receptors, most prominently the mammalian Multiple EGF Like Domains 10 (MEGF10) homolog, Draper [18]. Several studies have demonstrated that baseline levels of Draper in the uninjured brain determine the phagocytic capacity of ensheathing glia [19, 20]. Upon injury or disease, ensheathing glia upregulate Draper[19, 21–23]. However, low baseline levels may prevent Draper reaching a critical threshold for target detection leading to impaired clearance. Interestingly, Draper’s baseline levels were found to be regulated by Phosphoinositide 3-kinase (PI3K), a downstream effector of Insulin receptor (IR) signaling, while injury-induced Draper upregulation is regulated by another insulin signaling downstream target, the transcription factor Stat92E [19]. While local glial insulin receptor signaling has been shown to be a key regulator of Draper expression [24], it remains unknown whether obesogenic diets disrupt insulin signaling in glia and whether that disrupted signaling affects Draper expression and glial function. Hence, we set out to address whether prolonged obesogenic diets in Drosophila disrupt glial phagocytic function. We have previously shown that prolonged high sugar diet (HSD) treatment causes peripheral insulin resistance in adult flies [25]. Here, we show that chronic HSD exposure leads to insulin resistance in ensheating glia, which results in their impaired ability to clear axotomized olfactory neurons. Genetically inducing insulin release recapitulates HSD-induced Draper downregulation, while attenuating Insulin release rescues HSD-induced Draper downregulation. Importantly, we show that genetically stimulating a downstream effector of Insulin Receptor signaling in ensheathing glia rescues HSD-induced insulin resistance and the downregulation of Draper. Together, this study provides the first in vivo evidence of diet-induced regulation of glial phagocytic function.
## HSD affects the brain’s metabolism and causes lipid droplet accumulation.
Our lab established that adult Drosophila subjected to a prolonged (> 2 weeks) high sugar diet (HSD- see methods) display hallmarks of peripheral insulin resistance including disrupted hunger response [25]. Using the obesogenic diet paradigm we established, we sought to characterize the effects of prolonged HSD treatment on the brain. All cells rely on two main sources of energy: glycolysis (a series of cytosolic biochemical reactions to generate ATP) and mitochondrial oxidative phosphorylation (OxPhos); OxPhos and glycolysis exist in a delicate balance [26, 27] (Figure 1A). To get a broad understanding of the effects of HSD on the brain’s glycolytic state, we sought to examine the expression levels of the glycolytic enzyme, lactate dehydrogenase (Ldh). Given that *Ldh is* responsible for the final step of glycolysis, it has been used as a reliable readout of glycolysis [28], especially using an LDH reporter in flies [29, 30]. We subjected a transgenic fly line that expresses a fluorescent Ldh reporter to either ND or HSD for 3 weeks. For this study, we chose to focus on the antennal lobe region given its accessibility and well-defined histology. The HSD-fed flies had significantly lower levels of Ldh in the antennal lobe region compared to the ND-fed flies (Figure 1B-C) suggesting attenuated glycolysis.
Typically, when glycolysis is stunted, the mitochondria undergo rapid expansion to compensate for the fall in energy levels (Figure 1A) [26, 27, 31]. Given that glycolysis occurs primarily in glia [32–35], we sought to investigate the effects of HSD on glial mitochondria. Specifically, we focused on ensheathing glia, which are functionally similar to microglia and reside within the antennal lobe [22]. As it is technically challenging to measure oxygen consumption rate[36] – a more direct readout for mitochondrial OxPhos- in intact glia without also obtaining readouts from the neurons, we took an imaging-based approach. Mitochondrial morphology can be used as a readout of their activity. The OxPhos capacity of elongated mitochondria is higher than that of circular mitochondria and is used as an accepted measure to indicate OxPhos capacity [31, 37–39]. Hence, we expressed a mitochondrion targeted GFP (mito-GFP) in ensheathing glia and analyzed mitochondrial morphology. We expected that since LDH expression was reduced we would observe elongated mitochondria. Surprisingly, although 3 week of HSD exposure reduced LDH reporter expression in the brain suggesting reduced glycolysis (Figure 1B-C), it did not alter mitochondrial morphology in ensheathing glia as assessed by circularity and elongation (Figure 1D-F). In contrast, in adult Drosophila abdominal adipocytes, we observe a clear shift in mitochondrial morphology to appear more elongated/less circular (Figure I-K) in response to the reduced LDH levels (Figure 1G-H) of HSD-fed flies. We interpret this to mean that the brain and adipocytes respond differently to prolonged HSD at 3-weeks. Based on this, we postulate that the adipocytes maintain their ability, even on HSD, to shift towards OxPhos in response to reduced glycolysis, but the glial cells are unable to do so (See Discussion).
Human and animal studies have shown that microglia accumulate lipid droplets in aging and neurodegenerative disorders, leading to impaired function [40, 41]. We have previously shown that prolonged HSD treatment increases the number and size of lipid droplets in adipocytes [25]. Thus, we asked whether HSD has a similar effect on glial lipid storage. To answer this, we visualized the lipid droplets in the brains of ND and HSD-fed flies using LipidTOX, a neutral lipid stain, and drove membrane-tagged GFP expression specifically in ensheathing glia. We found that prolonged HSD treatment markedly increased the number and size of lipid droplets in ensheathing glia surrounding the antennal lobes (Figure 1I) indicating possible glial dysfunction.
## HSD causes insulin resistance in glia.
Lipid droplet accumulation and reduced glycolysis are tightly associated with impaired insulin signaling [25, 42–45]. Though obesogenic diets have been established by us and others to cause peripheral insulin resistance [25, 46, 47], whether obesogenic diets lead to central brain insulin resistance remains unclear. In a previous study, our lab demonstrated that chronic HSD treatment of 2 weeks or more causes insulin resistance in the adult Drosophila fat tissue [25]. To determine whether HSD treatment causes insulin resistance in the brain, we first compared the levels of the Drosophila Insulin-like peptide 5 (Dilp5) retained in the insulin producing cells (IPCs) of ND and HSD-fed flies. Dilp5 is primarily produced by the IPCs and its secretion is dictated by nutrient abundance [48, 49]. Dilp5’s retention in the IPCs is often used as a readout of its secretion [25, 50–55]. As expected, the HSD-fed flies showed reduced Dilp5 accumulation in their IPCs suggesting increased Dilp5 secretion (Figure 2A-B).
To investigate whether the increase in insulin signaling led to glial insulin resistance, we used a transgenic line that expresses a fluorescent reporter (tGPH) for Phosphoinositide 3-kinase (PI3K) [56, 57], a downstream effector of Insulin receptor (IR) signaling. Under normal conditions, the IR autophosphorylates upon interacting with insulin, which leads to activation of the PI3K pathway. However, excessive levels of circulating insulin can attenuate IR sensitivity leading to reduced PI3K activation [58, 59] (Figure 2C). To this end, we subjected the tGPH flies to 3 weeks of either ND or HSD and measured tGPH fluorescence in the area surrounding the antennal lobe where ensheathing glia reside (Figure 2D-E). We found that HSD caused a significant downregulation of PI3K activity indicating insulin resistance (Figure 2D-F). Notably, we did not observe any gross morphological defects in response to the prolonged HSD treatment (Figure S1). To determine whether the HSD-induced attenuation of PI3K signaling is specifically due to excessive systemic insulin secretion, we genetically induced insulin secretion from the IPCs by expressing the neuronal activator TrpA1 under the control of an IPC-specific Gal4 driver [51, 55]. Remarkably, we found that forced activation of the IPCs for 1 week mimics the effects of a 3-week HSD exposure on PI3K activity in ensheathing glia (Figure 2G-I). In contrast, attenuating the release of insulin in the HSD-fed flies by expressing a genetically modified potassium channel (EKO) that inhibits neuronal activation [60, 61] under the control of an IPC-specific driver increases glial insulin signaling (Figure 2J-L). Together, these findings demonstrate that obesogenic diets directly cause glial insulin resistance through excess systemic insulin.
## HSD downregulates basal Draper levels.
As the brain’s resident macrophages, microglia are crucial to the survival and function of the nervous system through their phagocytic activity [62]. Just like the mammalian microglia, the phagocytic activity of ensheathing glia is governed by the engulfment receptor, Draper (The Drosophila ortholog to the mammalian MEGF10) [22, 23]. Given that PI3K signaling has been shown to regulate basal Draper levels [19], we reasoned that HSD treatment and the ensuing downregulation of PI3K signaling (Figure 2D-F) would result in reduced basal Draper levels. To test this, we measured Draper immunofluorescence within a subset of ensheathing glia in ND and HSD-flies. We found that 3 weeks of HSD treatment caused a substantial reduction in basal Draper levels (Figure 3A-C).
It is possible that HSD treatment results in non-insulin dependent downregulation of Draper signaling. Therefore, we reasoned that if glial insulin resistance (Figure 2) is responsible for Draper downregulation in HSD-fed flies, then forced systemic insulin release from the IPCs would also result in Draper downregulation. Indeed, we find that expressing the neuronal activator TrpA1 under the control of an IPC-specific Gal4 driver leads to reduced Draper levels under ND conditions (Figure 3D-F). In contrast, attenuating systemic insulin release from the IPCs by expressing a genetically modified potassium channel (EKO) that inhibits neuronal activation paradoxically increases Draper expression in the HSD-fed flies (Figure 3G-I).
Next, we reasoned that if glial insulin resistance underlies the downregulation of Draper in HSD-fed flies, then stimulating PI3K signaling, a downstream arm of insulin signaling, will upregulate Draper expression. To address this, we expressed a constitutively active form of PI3K that is fused to a farnesylation signal (CAAX) [63] under the control of an ensheathing glia promoter. We found that stimulating ensheathing glia PI3K signaling increases Draper expression under HSD compared to the HSD-fed controls (Figure 3J-L). Together, this suggests that HSD downregulates Draper expression by inducing glial insulin resistance.
## HSD delays the clearance of degenerating axons by inhibiting injury-induced Draper and STAT upregulation.
Normally, neuronal injury triggers the upregulation of Draper in ensheathing glia that peaks one day after injury and persists until neuronal debris has been cleared [23]. Therefore, we asked whether HSD treatment prevents the upregulation of Draper after neuronal injury. To answer this, we took advantage of the accessibility of the olfactory neurons. We performed unilateral ablation of the third antennal segment, which houses the cell bodies of olfactory neurons. This results in the Wallerian degeneration of olfactory neurons’ axons that project to the antennal lobe, which induces ensheathing glia to phagocytose axonal debris [22, 23, 64]. Then, we immunostained for Draper one day post antennal ablation (Figure 4A). As expected, Draper levels increased significantly in the ND-fed flies, whereas the HSD-fed flies showed no upregulation in Draper (Figure 4B-C).
While baseline levels of Draper are regulated by PI3K [19], it has been shown that Stat92E, a transcription factor that acts downstream of both Draper and Insulin signaling is essential for the injury-induced Draper upregulation [19, 24]. Draper-dependent activation of Stat92E creates a positive autoregulatory loop in which Stat92E upregulates the transcription of the *Draper* gene [19]. Given that HSD treatment causes a diminished injury-induced Draper response (Figure 4B-C), we reasoned that Stat92E signaling would be attenuated in the antennal lobe region of the HSD-fed flies. To address this, we used a transgenic reporter line with ten Stat92E binding sites that drive the expression of a destabilized GFP [19]. We found that while ND-fed flies exhibited the expected post-injury Stat92E upregulation, HSD treatment caused a reduction in Stat92E levels at baseline and inhibited post injury upregulation (Figure 4D-E). Together, these data indicate that chronic HSD attenuate the Stat92E/Draper signaling pathway leading to impaired glial phagocytotic function.
To understand how disrupted Draper signaling in the HSD-fed flies affect glial phagocytic function, we subjected flies that express membrane tagged GFP in a subset of olfactory neurons (Odorant receptor 22a) to either 3 weeks of ND or HSD, then performed unilateral antennal ablation and examined the rate of GFP clearance over time (Figure 4F). By normalizing GFP fluorescence on the injured side to the uninjured side of the same animal, we were able to establish an endogenous control (Figure 4G). We found that HSD-fed flies had higher levels of GFP florescence at every timepoint indicating a delay in axonal clearance (Figure 4G-H). Together this data indicate that diet-induced glial insulin resistance impairs the clearance of neuronal debris by downregulating Draper.
## Discussion
With increased life expectancy, age-related neurodegenerative disorders are expected to rise, placing a tremendous burden on the healthcare system. Large-scale epidemiological studies have found that mid-life obesity is an independent risk factor for developing neurodegenerative disorders [1–3]. However, the mechanism underlying this connection remains unclear. Here, using a Drosophila in vivo model, we draw a causal link between diet-induced obesity and impaired glial phagocytic function, a major contributor to the pathology of age-related neurodegenerative disorders [65]. We show that excessive systemic insulin signaling leads to glial insulin resistance (Figure 2), which dampens the expression of the engulfment receptor, Draper (Figure 3), resulting in impaired glial clearance of degenerating axons (Figure 4). Together, our study provides a strong mechanistic insight into how diet-induced obesity alters glial function, thereby increasing the risk of neurodegenerative disorders.
## HSD causes glial insulin resistance.
Insulin signaling is critical for cell metabolism and function. However, excess systemic insulin can lead to insulin resistance, which results in diminished cellular response. Obesogenic diets are well known to cause insulin resistance in peripheral tissues, including fat, which rely on insulin to regulate glucose uptake[58, 66]. However, the evidence for diet-induced insulin resistance in the brain is scarce. This is mainly due to the belief that the brain is insulin independent [6], though evidence suggests that it acts on neurons and glia in a glucose-independent manner [67].
Our previous work has demonstrated that Drosophila can serve as an excellent diet-induced obesity model as its adipose tissue also undergoes insulin resistance when subjected to prolonged HSD treatment [25]. Thus, we used the same dietary regimen to investigate whether obesogenic diets lead to brain insulin resistance. Microglia play a significant role in maintaining nervous system homeostasis and their dysfunction is implicated in a myriad of metabolic and neurodegenerative disorders [62, 68, 69]. Although microglia express the insulin receptor [15], it remains unknown whether obesogenic diets can result in glial insulin resistance. In this study, we showed that prolonged HSD treatment attenuated glial PI3K signaling, a downstream arm of the insulin receptor signaling pathway (Figure 2D-F). This coincided with reduced Ldh levels indicating depressed glycolysis, which is tightly regulated by insulin signaling [42, 44, 45] (Figure 1B-C). Consistent with this, insulin resistance is associated with reduced glycolysis [70, 71]. Interestingly, a prerequisite to microglial activation is the metabolic switch from oxidative phosphorylation to glycolysis [72, 73] Intriguingly, even though glycolysis appears to be disrupted in the brains of HSD-fed flies, we did not observe any changes to glial mitochondrial morphology (Figure 1D-F), a biological response we see in the adipose tissue (Figure 1G-K). This suggest that HSD-induced insulin resistance may prevent glial activation by disrupting their OxPhos-Glycolysis balance. Because Ldh expression and mitochondrial morphology are indirect readouts of glycolysis and OxPhos, respectively, more precise metabolic analysis is needed to determine the specific glial metabolic defects caused by HSD. Furthermore, it is important to acknowledge that both the PI3K and Ldh reporters are non-cell specific. Therefore, it is likely that HSD is causing a downregulation of PI3K signaling and glycolysis in both glia and neurons. Given that glia and neurons are intricately connected, it is possible that dysfunctional neuronal metabolism further exacerbates glial dysfunction.
There are seven Drosophila insulin like peptides (Dilps). Dilps 2, 3, and 5 are primarily produced by the IPCs, which reside in the pars intercerebralis (PI) region of the Drosophila brain-the invertebrate equivalent to the mammalian hypothalamus. In the adult fly brain, the IPCs terminate their axons on peripheral targets including the gut and aorta for systemic Dilp release[49, 74, 75]. It is possible that excess circulating Dilps induced by HSD cause glial insulin resistance indirectly by dysregulating peripheral organs, such as muscles, guts, and adipose tissue. Nevertheless, distinct IPC arborizations have been observed in the brain, specifically in the tritocerebrum proximal to the antennal lobes, raising the possibility that the IPCs act in a paracrine manner. Given that genetic activation and inhibition of the IPCs for a short period of time was sufficient to influence glial PI3K signaling (Figure 2), it is possible that IPC-released Dilps act directly on ensheathing glia. In the future, it would be interesting to untangle the IPC-ensheathing glia insulin signaling circuit.
## HSD-induced glial dysfunction resembles that caused by aging.
A hallmark of neurodegenerative disorders is the failure to clear neuronal debris and cytotoxic proteins, triggering a cascade of devastating effects that include inflammation, cell death, and impaired regeneration. Therefore, it is not surprising that microglial dysfunction is implicated in driving the pathogenesis of many neurodegenerative disorders [12]. Although it is known that microglia express the insulin receptors and respond to insulin treatment in vitro [16, 76], it remained unknown whether they experience insulin resistance and whether that impacts their phagocytic activity. Both obesity and age-related neurodegenerative disorders are associated with dysfunctional insulin signaling [58, 66, 77, 78], suggesting a potential link. Here, we show that diet-induced insulin resistance disrupts glial phagocytic activity (Figure 4G-H) by downregulating the phagocytic receptor, Draper (Figures 3 & 4). We were able to demonstrate the direct effects of insulin signaling by showing that Dilp release alone mimicked HSD-induced glial defects (Figures 2 & 3D-L). While other groups have shown that local glial insulin receptor activity regulates Draper expression [24], we found that systemic insulin signaling directly regulates glial Draper expression.
It has been demonstrated by other groups as well as us that physiological factors affect glial function in Drosophila. Stanhope and colleagues[79] found that sleep plays a crucial role in glial phagocytosis. Similarly to HSD treatment, sleep loss causes Draper to downregulate, which leads to a failure to clear neuronal debris after injury. As with obesity and type 2 diabetes, dysregulated sleep has been associated with neurodegenerative disorders, further supporting the link between dysfunctional glial phagocytosis and neurodegenerative disorders. It is interesting to draw parallels between the results of this study and a recent study by Purice and colleagues [20] that found that aged flies exhibit delayed axonal clearance because of impaired Draper and PI3K activity. The connection between HSD exposure and aging holds true for humans and mammals as well. Age and neurodegenerative disorders are associated with reductions in MEGF10 [80], the mammalian ortholog of Draper, impaired glial phagocytosis [13], and stunted glycolysis [81]. One study showed that Insulin infusion into young rats activated microglia, but this effect was not observed in older rats suggesting that microglia’s insulin sensitivity is age dependent [82]. Furthermore, similar to our findings, which show an accumulation of lipids in the brains of HSD-fed flies, aged microglia also accumulate lipid droplets leading to impaired phagocytic function [41]. As a result, it could be argued that chronic HSD exposure may “accelerate” aging in flies. It would be interesting to explore this connection in more detail by comparing the transcriptomes of aged and HSD-treated flies in the future.
## Fly husbandry
The following Drosophila strains were used: Or22a-mCD8GFP (BDSC #52620), Ensheathing glia-Gal4 (BDSC # 39157), 10xSTAT92E-GFP (Perrimon Lab), LDH-GFP (a generous gift from Dr. Tennessen), tGPH (Gift of Bruce Edgar), Dilp2-Gal4 (a gift from P.Shen), UAS-TrpA1 (BDSC #26263), UAS-EKO22 (BDSC #40974), UAS-PI3K-CAAX (BDSC #8294). Flies were housed in 25°C incubators and all experiments were done on adult male flies. To induce the expression of the temperature sensitive TrpA1, flies were moved to a 29°C incubator 1 week post eclosion. Following 1 week after eclosion, the flies were placed on either a normal diet containing 15 g yeast, 8.6 g soy flour, 63 g corn flour, 5 g agar, 5 g malt, 74 mL corn syrup per liter, or a HSD, which consists of an additional 300 g of sucrose per liter ($30\%$ increase).
## Antennal nerve injury
As adapted from [19–21, 23, 24, 64, 79, 83, 84], flies were anesthetized using CO2 and antennal nerve injury was accomplished by unilaterally removing the third antennal segment of anesthetized adult flies using forceps. Flies were then placed back into either ND or HSD until they were dissected 24 hours after injury or as indicated otherwise in the figure legends.
## Immunostaining
Immunostaining of adult brains and fat bodies was performed as previously described[25, 50, 51]. Tissues were dissected in ice-cold PBS. Brains were fixed overnight in $0.8\%$ paraformaldehyde (PFA) in PBS at 4°C. The fixed brains were washed five times in PBS with $0.5\%$ BSA and $0.5\%$ Triton X-100 (PAT), blocked for 1 hour in PAT + $5\%$ NDS, and then incubated overnight at 4°C with the primary antibodies. Following incubation, the brains were washed five times in PAT, re-blocked for 30 min, then incubated in secondary antibody in block for 4 hr at room temperature. Finally, the brains were washed five times in PAT, then mounted on slides in Slow fade gold antifade. Primary antibodies were as follows: rabbit anti-Dilp5 (1:500; this study); Chicken anti-GFP (1:500; Cat# ab13970, RRID:AB_300798); And Mouse anti-Draper (1:50; DSHB 5D14 RRID:AB_2618105). Secondary antibodies from Jackson ImmunoResearch (1:500) include donkey anti-Chicken Alexa 488 (Cat# 703-545-155, RRID: AB_2340375); donkey anti-mouse Alexa 594 (Cat# 715-585-150, RRID:AB_2340854). Lipid droplets were stained with lipidtox (1:500, Thermo Fisher Cat#H34477) overnight at 4C.
## Image analysis
Images were acquired with a Zeiss LSM 800 confocal system and analyzed using ImageJ[85]. All images within each experiment were acquired with the same confocal settings. Z-stack summation projections that spanned the depth of the antennal lobes at 0.3 um intervals were generated and a region of interest (indicated on the figure) was used to measure the fluorescent intensity of GFP or Draper. Dilp 5 was measured using z-stack summation projections that included the full depth of the IPCs. A region of interest around the IPCs was manually drawn using the free hand tool and the integrated density values were acquired. To measure mitochondrial morphology, A maximum-intensity projection of Z-stacks that covered the full depth of the antennal lobe was used for ImageJ analysis. The maximum-intensity projection was first inverted then automatically thresholded before the ‘analyze particles’ function was used to measure the average mitochondrial circularity, major axis, and minor axis.
## Statistics
All the statistical tests were done using GraphPad PRISM. ( GraphPad Software Incorporated, La Jolla, Ca, USA, RRID:SCR_002798). The assumption of normality was tested using Shapiro-Wilk’s test. A t test was performed for Figures 1–3, while a 2-way ANOVA was used for Figures 4–5. Data are presented as means ± standard deviation. Significance was set at $p \leq 0.05.$ N for each experiment is detailed in the figures.
## References
1. Xu W.L.. **Midlife overweight and obesity increase late-life dementia risk: a population-based twin study.**. *Neurology* (2011) **76** 1568-74. PMID: 21536637
2. Hassing L.B.. **Overweight in midlife and risk of dementia: a 40-year follow-up study.**. *Int J Obes (Lond)* (2009) **33** 893-8. PMID: 19506566
3. Allen A.N.. **Adiposity in middle and old age and risk of death from dementia: 40-year follow-up of 19,000 men in the Whitehall study.**. *Age Ageing* (2019) **48** 247-253. PMID: 30624572
4. Magriplis E.. **Dietary Sugar Intake and Its Association with Obesity in Children and Adolescents.**. *Children (Basel)* (2021) **8**
5. Amatruda J.M., Livingston J.N., Lockwood D.H.. **Cellular mechanisms in selected states of insulin resistance: human obesity, glucocorticoid excess, and chronic renal failure.**. *Diabetes Metab Rev* (1985) **1** 293-317. PMID: 3915256
6. McEwen B.S., Reagan L.P.. **Glucose transporter expression in the central nervous system: relationship to synaptic function.**. *Eur J Pharmacol* (2004) **490** 13-24. PMID: 15094070
7. Margolis R.U., Altszuler N.. **Insulin in the cerebrospinal fluid.**. *Nature* (1967) **215** 1375-6. PMID: 6055448
8. Woods S.C.. **Insulin and the blood-brain barrier.**. *Curr Pharm Des* (2003) **9** 795-800. PMID: 12678878
9. Kleinridders A.. **Insulin action in brain regulates systemic metabolism and brain function.**. *Diabetes* (2014) **63** 2232-43. PMID: 24931034
10. Agrawal R.. **Insulin action in the brain regulates both central and peripheral functions.**. *Am J Physiol Endocrinol Metab* (2021) **321** E156-E163. PMID: 34056920
11. Wilson D.M.. **Hallmarks of neurodegenerative diseases.**. *Cell* (2023) **186** 693-714. PMID: 36803602
12. Hickman S.. **Microglia in neurodegeneration.**. *Nat Neurosci* (2018) **21** 1359-1369. PMID: 30258234
13. Gabande-Rodriguez E., Keane L., Capasso M.. **Microglial phagocytosis in aging and Alzheimer’s disease.**. *J Neurosci Res* (2020) **98** 284-298. PMID: 30942936
14. Mendes N.F.. **Hypothalamic Microglial Activation in Obesity: A Mini-Review.**. *Front Neurosci* (2018) **12** 846. PMID: 30524228
15. Myhre C.L.. **Microglia Express Insulin-Like Growth Factor-1 in the Hippocampus of Aged APP(swe)/PS1(DeltaE9) Transgenic Mice.**. *Front Cell Neurosci* (2019) **13** 308. PMID: 31417357
16. Brabazon F.. **The effects of insulin on the inflammatory activity of BV2 microglia.**. *PLoS One* (2018) **13** e0201878. PMID: 30148836
17. Freeman M.R.. **Drosophila Central Nervous System Glia.**. *Cold Spring Harb Perspect Biol* (2015) **7**
18. Raiders S.. **Engulfed by Glia: Glial Pruning in Development, Function, and Injury across Species.**. *J Neurosci* (2021) **41** 823-833. PMID: 33468571
19. Doherty J.. **PI3K signaling and Stat92E converge to modulate glial responsiveness to axonal injury.**. *PLoS Biol* (2014) **12** e1001985. PMID: 25369313
20. Purice M.D., Speese S.D., Logan M.A.. **Delayed glial clearance of degenerating axons in aged Drosophila is due to reduced PI3K/Draper activity.**. *Nat Commun* (2016) **7** 12871. PMID: 27647497
21. Ray A., Speese S.D., Logan M.A.. **Glial Draper Rescues Abeta Toxicity in a Drosophila Model of Alzheimer’s Disease.**. *J Neurosci* (2017) **37** 11881-11893. PMID: 29109235
22. Doherty J.. **Ensheathing glia function as phagocytes in the adult Drosophila brain.**. *J Neurosci* (2009) **29** 4768-81. PMID: 19369546
23. MacDonald J.M.. **The Drosophila cell corpse engulfment receptor Draper mediates glial clearance of severed axons.**. *Neuron* (2006) **50** 869-81. PMID: 16772169
24. Musashe D.T.. **Insulin-like Signaling Promotes Glial Phagocytic Clearance of Degenerating Axons through Regulation of Draper.**. *Cell Rep* (2016) **16** 1838-50. PMID: 27498858
25. Kelly K.P.. **Fat Body Phospholipid State Dictates Hunger Driven Feeding Behavior.**. *bioRxiv* (2022)
26. Li T.. **PKM2 coordinates glycolysis with mitochondrial fusion and oxidative phosphorylation.**. *Protein Cell* (2019) **10** 583-594. PMID: 30887444
27. Xu Y.. **Why All the Fuss about Oxidative Phosphorylation (OXPHOS)?**. *J Med Chem* (2020) **63** 14276-14307. PMID: 33103432
28. Gandara L.. **A genetic toolkit for the analysis of metabolic changes in Drosophila provides new insights into metabolic responses to stress and malignant transformation.**. *Sci Rep* (2019) **9** 19945. PMID: 31882718
29. Wong K.K.L., Liao J.Z., Verheyen E.M.. **A positive feedback loop between Myc and aerobic glycolysis sustains tumor growth in a Drosophila tumor model.**. *Elife* (2019) **8**
30. Krejcova G.. **Drosophila macrophages switch to aerobic glycolysis to mount effective antibacterial defense.**. *Elife* (2019) **8**
31. Seo B.J.. **Role of mitochondrial fission-related genes in mitochondrial morphology and energy metabolism in mouse embryonic stem cells.**. *Redox Biol* (2020) **36** 101599. PMID: 32521505
32. Volkenhoff A.. **Glial Glycolysis Is Essential for Neuronal Survival in Drosophila.**. *Cell Metab* (2015) **22** 437-47. PMID: 26235423
33. Herrero-Mendez A.. **The bioenergetic and antioxidant status of neurons is controlled by continuous degradation of a key glycolytic enzyme by APC/C-Cdh1.**. *Nat Cell Biol* (2009) **11** 747-52. PMID: 19448625
34. Dienel G.A., Hertz L.. **Glucose and lactate metabolism during brain activation.**. *J Neurosci Res* (2001) **66** 824-38. PMID: 11746408
35. Bittner C.X.. **High resolution measurement of the glycolytic rate.**. *Front Neuroenergetics* (2010) **2**
36. Plitzko B., Loesgen S.. **Measurement of Oxygen Consumption Rate (OCR) and Extracellular Acidification Rate (ECAR) in Culture Cells for Assessment of the Energy Metabolism.**. *Bio Protoc* (2018) **8** e2850
37. Yao C.H.. **Mitochondrial fusion supports increased oxidative phosphorylation during cell proliferation.**. *Elife* (2019) **8**
38. Kuznetsov A.V.. **Structural and functional remodeling of mitochondria as an adaptive response to energy deprivation.**. *Biochim Biophys Acta Bioenerg* (2021) **1862** 148393. PMID: 33549532
39. Mishra P., Chan D.C.. **Metabolic regulation of mitochondrial dynamics.**. *J Cell Biol* (2016) **212** 379-87. PMID: 26858267
40. Farmer B.C.. **Lipid Droplets in Neurodegenerative Disorders.**. *Front Neurosci* (2020) **14** 742. PMID: 32848541
41. Marschallinger J.. **Lipid-droplet-accumulating microglia represent a dysfunctional and proinflammatory state in the aging brain.**. *Nat Neurosci* (2020) **23** 194-208. PMID: 31959936
42. Beitner R., Kalant N.. **Stimulation of glycolysis by insulin.**. *J Biol Chem* (1971) **246** 500-3. PMID: 5542017
43. Dimitriadis G.. **Insulin effects in muscle and adipose tissue.**. *Diabetes Res Clin Pract* (2011) **93** S52-9. PMID: 21864752
44. Wu C., Khan S.A., Lange A.J.. **Regulation of glycolysis-role of insulin.**. *Exp Gerontol* (2005) **40** 894-9. PMID: 16157461
45. Noguchi R.. **The selective control of glycolysis, gluconeogenesis and glycogenesis by temporal insulin patterns.**. *Mol Syst Biol* (2013) **9** 664. PMID: 23670537
46. Musselman L.P.. **A high-sugar diet produces obesity and insulin resistance in wild-type Drosophila.**. *Dis Model Mech* (2011) **4** 842-9. PMID: 21719444
47. Na J.. **A Drosophila model of high sugar diet-induced cardiomyopathy.**. *PLoS Genet* (2013) **9** e1003175. PMID: 23326243
48. Ikeya T.. **Nutrient-dependent expression of insulin-like peptides from neuroendocrine cells in the CNS contributes to growth regulation in Drosophila.**. *Curr Biol* (2002) **12** 1293-300. PMID: 12176357
49. Nassel D.R., Liu Y., Luo J.. **Insulin/IGF signaling and its regulation in Drosophila.**. *Gen Comp Endocrinol* (2015) **221** 255-66. PMID: 25616197
50. Rajan A.. **A Mechanism Coupling Systemic Energy Sensing to Adipokine Secretion.**. *Dev Cell* (2017) **43** 83-98. PMID: 29017032
51. Brent A.E., Rajan A.. **Insulin and Leptin/Upd2 Exert Opposing Influences on Synapse Number in Fat-Sensing Neurons.**. *Cell Metab* (2020) **32** 786-800. PMID: 32976758
52. Delanoue R.. **Drosophila insulin release is triggered by adipose Stunted ligand to brain Methuselah receptor.**. *Science* (2016) **353** 1553-1556. PMID: 27708106
53. Ziegler A.B., Maniere G., Grosjean Y.. **JhI-21 plays a role in Drosophila insulin-like peptide release from larval IPCs via leucine transport.**. *Sci Rep* (2018) **8** 1908. PMID: 29382949
54. Enell L.E.. **Insulin signaling, lifespan and stress resistance are modulated by metabotropic GABA receptors on insulin producing cells in the brain of Drosophila.**. *PLoS One* (2010) **5** e15780. PMID: 21209905
55. Liu Y.. **Serotonin and insulin-like peptides modulate leucokinin-producing neurons that affect feeding and water homeostasis in Drosophila.**. *J Comp Neurol* (2015) **523** 1840-63. PMID: 25732325
56. Britton J.S.. **Drosophila’s insulin/PI3-kinase pathway coordinates cellular metabolism with nutritional conditions.**. *Dev Cell* (2002) **2** 239-49. PMID: 11832249
57. Read R.D.. **A Drosophila model for EGFR-Ras and PI3K-dependent human glioma.**. *PLoS Genet* (2009) **5** e1000374. PMID: 19214224
58. Huang X.. **The PI3K/AKT pathway in obesity and type 2 diabetes.**. *Int J Biol Sci* (2018) **14** 1483-1496. PMID: 30263000
59. Khalid M.. **Insulin Signal Transduction Perturbations in Insulin Resistance.**. *Int J Mol Sci* (2021) **22**
60. White B.H.. **Targeted attenuation of electrical activity in Drosophila using a genetically modified K(+) channel.**. *Neuron* (2001) **31** 699-711. PMID: 11567611
61. Rajan A., Perrimon N.. **Drosophila cytokine unpaired 2 regulates physiological homeostasis by remotely controlling insulin secretion.**. *Cell* (2012) **151** 123-37. PMID: 23021220
62. Li Q., Barres B.A.. **Microglia and macrophages in brain homeostasis and disease.**. *Nat Rev Immunol* (2018) **18** 225-242. PMID: 29151590
63. Leevers S.J.. **The Drosophila phosphoinositide 3-kinase Dp110 promotes cell growth.**. *EMBO J* (1996) **15** 6584-94. PMID: 8978685
64. Lu T.Y.. **Axon degeneration induces glial responses through Draper-TRAF4-JNK signalling.**. *Nat Commun* (2017) **8** 14355. PMID: 28165006
65. Xu Y.. **Microglia in neurodegenerative diseases.**. *Neural Regen Res* (2021) **16** 270-280. PMID: 32859774
66. Adeva-Andany M.M.. **Effect of diet composition on insulin sensitivity in humans.**. *Clinical Nutrition ESPEN* (2019) **33** 29-38. PMID: 31451269
67. Lee S.H.. **Insulin in the nervous system and the mind: Functions in metabolism, memory, and mood.**. *Mol Metab* (2016) **5** 589-601. PMID: 27656397
68. Afridi R., Lee W.H., Suk K.. **Microglia Gone Awry: Linking Immunometabolism to Neurodegeneration.**. *Front Cell Neurosci* (2020) **14** 246. PMID: 32903682
69. Lee J.W.. **The Role of Microglia in the Development of Neurodegenerative Diseases.**. *Biomedicines* (2021) **9**
70. Lee Y.. **Hyperglycemia- and hyperinsulinemia-induced insulin resistance causes alterations in cellular bioenergetics and activation of inflammatory signaling in lymphatic muscle.**. *FASEB J* (2017) **31** 2744-2759. PMID: 28298335
71. Friesen M.. **Development of a physiological insulin resistance model in human stem cell-derived adipocytes.**. *Sci Adv* (2022) **8** eabn7298. PMID: 35714187
72. Gimeno-Bayon J.. **Glucose pathways adaptation supports acquisition of activated microglia phenotype.**. *J Neurosci Res* (2014) **92** 723-31. PMID: 24510633
73. Cheng J.. **Early glycolytic reprogramming controls microglial inflammatory activation.**. *J Neuroinflammation* (2021) **18** 129. PMID: 34107997
74. Nassel D.R.. **Factors that regulate insulin producing cells and their output in Drosophila.**. *Front Physiol* (2013) **4** 252. PMID: 24062693
75. Nassel D.R., Vanden Broeck J.. **Insulin/IGF signaling in Drosophila and other insects: factors that regulate production, release and post-release action of the insulin-like peptides.**. *Cell Mol Life Sci* (2016) **73** 271-90. PMID: 26472340
76. Spielman L.J.. **Insulin Modulates In Vitro Secretion of Cytokines and Cytotoxins by Human Glial Cells.**. *Curr Alzheimer Res* (2015) **12** 684-93. PMID: 26159205
77. Baranowska-Bik A., Bik W.. **Insulin and brain aging.**. *Prz Menopauzalny* (2017) **16** 44-46. PMID: 28721128
78. Sedzikowska A., Szablewski L.. **Insulin and Insulin Resistance in Alzheimer’s Disease.**. *Int J Mol Sci* (2021) **22**
79. Stanhope B.A.. **Sleep Regulates Glial Plasticity and Expression of the Engulfment Receptor Draper Following Neural Injury.**. *Curr Biol* (2020) **30** 1092-1101. PMID: 32142708
80. Lee S.Y., Chung W.S.. **The roles of astrocytic phagocytosis in maintaining homeostasis of brains.**. *J Pharmacol Sci* (2021) **145** 223-227. PMID: 33602502
81. Goyal M.S.. **Loss of Brain Aerobic Glycolysis in Normal Human Aging.**. *Cell Metab* (2017) **26** 353-360. PMID: 28768174
82. Haas C.B.. **Insulin activates microglia and increases COX-2/IL-1beta expression in young but not in aged hippocampus.**. *Brain Res* (2020) **1741** 146884. PMID: 32422127
83. Hilu-Dadia R.. **Draper-mediated JNK signaling is required for glial phagocytosis of apoptotic neurons during Drosophila metamorphosis.**. *Glia* (2018) **66** 1520-1532. PMID: 29520845
84. Doherty J.. **Ensheathing glia function as phagocytes in the adult Drosophila brain.**. *J Neurosci* (2009) **29** 4768-81. PMID: 19369546
85. Schneider C.A., Rasband W.S., Eliceiri K.W.. **NIH Image to ImageJ: 25 years of image analysis.**. *Nat Methods* (2012) **9** 671-5. PMID: 22930834
|
---
title: Pharmacologic Activation of a Compensatory Integrated Stress Response Kinase
Promotes Mitochondrial Remodeling in PERK-deficient Cells
authors:
- Valerie Perea
- Kelsey R. Baron
- Vivian Dolina
- Giovanni Aviles
- Jessica D. Rosarda
- Xiaoyan Guo
- Martin Kampmann
- R. Luke Wiseman
journal: bioRxiv
year: 2023
pmcid: PMC10029010
doi: 10.1101/2023.03.11.532186
license: CC BY 4.0
---
# Pharmacologic Activation of a Compensatory Integrated Stress Response Kinase Promotes Mitochondrial Remodeling in PERK-deficient Cells
## SUMMARY
The integrated stress response (ISR) comprises the eIF2α kinases PERK, GCN2, HRI, and PKR, which induce translational and transcriptional signaling in response to diverse insults. Deficiencies in PERK signaling lead to mitochondrial dysfunction and contribute to the pathogenesis of numerous diseases. We define the potential for pharmacologic activation of compensatory eIF2α kinases to rescue ISR signaling and promote mitochondrial adaptation in PERK-deficient cells. We show that the HRI activator BtdCPU and GCN2 activator halofuginone promote ISR signaling and rescue ER stress sensitivity in PERK-deficient cells. However, BtdCPU induces mitochondrial depolarization, leading to mitochondrial fragmentation and activation of the OMA1-DELE1-HRI signaling axis. In contrast, halofuginone promotes mitochondrial elongation and adaptive mitochondrial respiration, mimicking regulation induced by PERK. This shows halofuginone can compensate for deficiencies in PERK signaling and promote adaptive mitochondrial remodeling, highlighting the potential for pharmacologic ISR activation to mitigate mitochondrial dysfunction and motivating the pursuit of highly-selective ISR activators.
## INTRODUCTION
Endoplasmic reticulum (ER) stress and mitochondrial dysfunction are inextricably linked in the onset and pathogenesis of etiologically diverse diseases including cancer, diabetes, and many neurodegenerative disorders. This has led to considerable interest in defining the biological mechanisms responsible for regulating mitochondria during ER stress. The PERK arm of the unfolded protein response (UPR) has emerged as an important stress-responsive signaling pathway for adapting mitochondria in response to pathologic ER insults (Almeida et al, 2022; Rainbolt et al, 2014). PERK is an ER-localized kinase that is activated in response to ER stress through a mechanism involving autophosphorylation and dimerization (Gardner et al, 2013; Hetz et al, 2020; Walter & Ron, 2011). Once activated, PERK primarily functions through selective phosphorylation of the α subunit of eukaryotic initiation factor 2 (eIF2α). This leads to both a reduction in new protein synthesis and the selective activation of stress-responsive transcription factors such as ATF4 that regulate expression of genes involved in many adaptive pathways including cellular redox, amino acid biosynthesis, and cellular proteostasis (Han et al, 2013; Harding et al, 2000; Wek & Cavener, 2007).
PERK localizes to ER-mitochondrial contact sites, positioning this ER stress sensor to coordinate ER and mitochondria function in response to pathologic insults (Munoz et al, 2013; Verfaillie et al, 2012). Consistent with this, PERK-regulated transcriptional and translational signaling are both implicated in the adaptive remodeling of mitochondrial morphology and function (Almeida et al., 2022; Rainbolt et al., 2014). PERK signaling promotes both protective mitochondrial elongation and mitochondrial cristae formation in response to ER stress through mechanisms including adaptive remodeling of mitochondrial phospholipids and regulated import of the MICOS subunit MIC19 (Barad et al, 2022; Latorre-Muro et al, 2021; Lebeau et al, 2018; Perea et al, 2022). This organellar and ultrastructural remodeling functions to regulate mitochondrial bioenergetics and prevent premature mitochondrial fragmentation during ER stress. Similarly, ATF4-dependent upregulation of SCAF1 downstream of PERK increases assembly of respiratory chain supercomplexes to further adapt mitochondrial bioenergetics during ER stress (Balsa et al, 2019). PERK signaling also regulates mitochondrial proteostasis through multiple mechanisms including ATF4-dependent expression of mitochondrial proteostasis factors (e.g., HSPA9, LON) and reductions in the core TIM23 import subunit TIM17A downstream of translation attenuation (Han et al., 2013; Harding et al, 2003; Hori et al, 2002; Rainbolt et al, 2013). Apart from these adaptive functions, PERK signaling regulates mitochondria-derived apoptotic signaling following prolonged, severe ER stress through multiple mechanisms primarily involving sustained upregulation of the transcription factor CHOP (Hetz & Papa, 2018). Thus, PERK serves a central role in dictating mitochondrial adaptation and cellular survival in response to varying levels of ER stress.
Deficiencies in PERK activity induced by genetic, environmental, or aging-related factors are implicated in the onset and pathogenesis of numerous diseases (Almeida et al., 2022). Loss-of-function mutations in EIF2AK3, the gene that encodes PERK, are causatively associated with Wolcott-Rallison syndrome – a rare autosomal-recessive disorder that involves multi-organ failures including prominent neonatal or early-childhood insulin-dependent diabetes, kidney and liver dysfunction, and cardiac abnormalities (Delepine et al, 2000; Julier & Nicolino, 2010; Mann et al, 2022). Hypomorphic EIF2AK3 alleles are also implicated in neurodegenerative diseases including the tauopathy progressive supranuclear palsy (PSP) (Hoglinger et al, 2011; Park et al, 2022; Yuan et al, 2018). Further, deficiencies in PERK signaling have been implicated in the pathogenesis of many diseases including PSP and Huntington’s disease (HD) (Almeida et al., 2022; Bruch et al, 2017; Ganz et al, 2020; Shacham et al, 2021). Intriguingly, the pathogenesis of these diseases all involves mitochondrial dysfunction, suggesting that impaired PERK signaling could contribute to disease pathology through deficient regulation of mitochondria.
The above results suggest that pharmacologic enhancement of PERK signaling offers a potential opportunity to mitigate pathologic cellular and mitochondrial dysfunction caused by PERK deficiency. Multiple compounds have been identified to activate PERK signaling including CCT020312 and MK28 (Ganz et al., 2020; Grandjean & Wiseman, 2020; Stockwell et al, 2012). Both of these compounds have been shown to be beneficial in cellular and mouse models of neurodegenerative diseases including PSP and HD (Bruch et al., 2017; Ganz et al., 2020). However, the mechanism by which these compounds activate PERK, their selectivity for PERK signaling over other stress-responsive signaling pathways, and the specific dependence of the observed protection on PERK activity remain poorly defined. Further, PERK activating compounds are likely to be limited in their ability to promote protective PERK-dependent signaling in cells expressing inactive or hypomorphic PERK variants. Thus, different strategies are likely required to fully access the therapeutic potential for enhancing adaptive PERK signaling in the context of disease.
An alternative strategy to promote cellular and mitochondria remodeling in PERK-deficient cells is to activate compensatory kinases that induce similar signaling to that regulated by PERK. The integrated stress response (ISR) comprises four stress-activated eIF2α kinases activated in response to diverse pathologic insults (Costa-Mattioli & Walter, 2020; Pakos-Zebrucka et al, 2016) (Fig. 1A). Apart from PERK, these include GCN2 (activated by amino acid deprivation), HRI (activated by heme deficiency, oxidative stress, and mitochondrial dysfunction), and PKR (activated by double stranded RNA). The activation of these kinases leads to eIF2α phosphorylation and similar translational and transcriptional signaling to that observed upon PERK activation. This suggests that pharmacologic activation of these other ISR kinases could compensate for deficiencies in PERK signaling and rescue pathologic cellular and mitochondrial dysfunction induced by reduced PERK activity. Consistent with this idea, HRI is activated in Perk-deficient neurons, but not Perk-deficient astrocytes, to promote translational attenuation and ATF4 activation downstream of eIF2α during ER stress (Wolzak et al, 2022). While this may reflect mitochondrial stress-dependent activation of HRI signaling induced by ER stress in neurons lacking Perk (Fessler et al, 2020; Guo et al, 2020), these results support the potential for pharmacologically enhancing alternative eIF2α kinases to promote ISR signaling in Perk-deficient cells.
Numerous small molecules have been identified to activate these other ISR kinases. BtdCPU and related N,N’-diarylureas are activators of HRI signaling, although the mechanism by which these compounds activate HRI are poorly defined (Chen et al, 2011; Zhang et al, 2020). Alternatively, GCN2 is activated by the glutamyl-prolyl tRNA synthetase inhibitor halofuginone through a mechanism involving the accumulation of uncharged proline tRNA (Keller et al, 2012). Tyrosine kinase inhibitors such as erlotinib and sunitinib were also shown to activate GCN2 through a mechanism potentially involving direct binding to this ISR kinase (Tang et al, 2022). Like PERK activators, these other ISR kinase activators, most notably BtdCPU and halofuginone, have been shown to be protective in cellular and in vivo models of numerous disorders, further highlighting the potential for pharmacologic ISR activation in disease (Chen et al., 2011; Ishii et al, 2009; Juarez et al, 2012; Keller et al., 2012; Tian et al, 2021). However, the capacity for these compounds to promote adaptive mitochondrial remodeling in wild-type cells or cells deficient in PERK activity is currently unknown.
Here, we define the potential for pharmacologic, stress independent activation of ISR kinases to promote adaptive cellular and mitochondrial remodeling in wild-type and Perk-deficient cells. We identify two prioritized ISR activators, BtdCPU and halofuginone, that restores ISR signaling and ER stress sensitivity in Perk-deficient cells. However, we find that these compounds differentially impact mitochondria. We show that BtdCPU induces mitochondrial uncoupling, which in turn leads to mitochondrial fragmentation and ISR activation through the OMA1-DELE1-HRI mitochondrial stress signaling axis (Fessler et al., 2020; Guo et al., 2020). In contrast, halofuginone induces ISR-dependent mitochondrial elongation and adaptive remodeling of mitochondrial respiratory chain activity, mimicking mitochondrial adaptations induced by PERK signaling. These results demonstrate the potential for pharmacologic, stress-independent activation of compensatory ISR kinases to promote adaptive mitochondrial remodeling in wild-type and PERK-deficient cells, motivating the development of highly-selective ISR kinase activating compounds for the treatment of diseases associated with PERK deficiency and/or mitochondrial dysfunction.
## Activity profiling of ISR activating compounds.
The ISR comprises four eIF2α kinases – PERK, HRI, GCN2, and PKR – that are activated in response to diverse stimuli (Costa-Mattioli & Walter, 2020; Pakos-Zebrucka et al., 2016) (Fig. 1A). We initially probed the activity of previously reported compounds predicted to activate specific ISR kinases. These include the PERK activators CCT020312 and MK28 (Ganz et al., 2020; Stockwell et al., 2012), the HRI activator BtdCPU (Chen et al., 2011), and the GCN2 activators halofuginone, erlotinib, and sunitinib (Keller et al., 2012; Tang et al., 2022) (Fig. 1B). Initially, we tested the ability of these compounds to activate a luciferase-based reporter of ATF4 translation (ATF4-FLuc) as an indicator of ISR activity in HEK293T cells (Fig. S1A) (Yang et al, 2022). We confirmed that thapsigargin (Tg), a SERCA inhibitor that activates the ER stress responsive PERK ISR kinase, robustly increases ATF4-FLuc activity (Fig. S1B). All tested ISR activating compounds activated the ATF4-FLuc reporter with varying potency and efficacy that was consistent with previous reports (Fig. 1C) (Chen et al., 2011; Ganz et al., 2020; Keller et al., 2012; Stockwell et al., 2012; Tang et al., 2022). Next, to determine the specific kinase responsible for compound-dependent ISR activation, we used HEK293T cells expressing an ATF4-mApple translational reporter of ISR activation (Fig. S1A) and CRISPRi-depleted of individual ISR kinases (Guo et al., 2020). We confirmed the effectiveness of this approach by showing that Tg-dependent ATF4-mApple activation was inhibited in cells deficient in PERK, but not cells deficient in other ISR kinases, confirming that ER stress-dependent ATF4 activation is mediated through PERK (Fig. S1C). Treatment with CCT020312 or MK28 for 8 h increases ATF4-mApple fluorescence in all cell lines, indicating that, under these conditions, these two compounds activate the ISR reporter through a mechanism not solely dependent on a single kinase (Fig. S1C). HRI-depletion selectively blocked BtdCPU-dependent ATF4-mApple fluorescence, demonstrating this compound activates the ISR through HRI (Fig. 1D). Similarly, GCN2-depletion selectively inhibited ATF4-mApple fluorescence induced by halofuginone, erlotinib, and sunitinib, confirming that these compounds activated ISR signaling through GCN2 (Fig. 1D). These results show that BtdCPU, halofuginone, erlotinib, and sunitinib activate the ISR through the activity of specific ISR kinases, confirming previously published results (Chen et al., 2011; Keller et al., 2012; Tang et al., 2022).
## Compensatory ISR kinase activation restores ISR signaling in Perk-deficient cells
We next sought to identify compounds that could restore ISR signaling and mitigate pathologic phenotypes linked to deficiencies in PERK activity. Initially, we treated Perk+/+ and Perk−/− MEFs with ISR activating compounds for 3 h and monitored expression of the ISR target protein ATF4. Tg-dependent increases of ATF4 were blocked in Perk−/− MEFs, further confirming that ER stress induces ATF4 through a PERK-regulated mechanism (Fig. 2A,B). In contrast, BtdCPU, halofuginone, erlotinib, sunitinib, CCT020312, and MK28 increased ATF4 in both Perk+/+ and Perk−/− MEFs, although Perk-deficient cells did show lower levels of ATF4 induction for many compounds, as compared to wild-type cells (Fig. 2A,B). Considering other activities of erlotinib (e.g., EGFR inhibition) and sunitinib (e.g., receptor tyrosine kinase inhibition) and the promiscuity for ISR kinase activation observed for CCT020312 and MK28 (Fig. S1C), we prioritized BtdCPU and halofuginone for further study. We showed that BtdCPU and halofuginone did not increase ATF4 in knockin MEFs expressing the non-phosphorylatable eIF2α mutant S51A (MEFA/A) (Scheuner et al, 2001), confirming that the observed increase in ATF4 afforded by these compounds could be attributed to ISR signaling (Fig. S2A).
To further probe the potential for BtdCPU and halofuginone to induce ISR signaling, we monitored gene expression by RNAseq in Perk+/+ and Perk−/− MEFs treated with these compounds in the presence or absence of Tg (Table S1). Initially, we compared the expression of 16 established ISR target genes (Grandjean et al, 2019) to that observed in Tg-treated Perk+/+ cells across all conditions (Fig. 2C,D, Table S2). As expected, Perk-deficient cells treated with Tg showed no significant induction of ISR target genes. However, ER stress-responsive genes regulated by the IRE1/XBP1s and ATF6 arms of the UPR are efficiently induced in Tg-treated, Perk-deficient cells (Fig. 2D, Table S2). Treatment with BtdCPU or halofuginone increased ISR target gene expression in both Perk+/+ and Perk−/− MEFs in the presence or absence of Tg (Fig. 2C,D). We confirmed these results for the ISR target genes Ddit3/Chop and Chac1 by qPCR (Fig. S2B,C) These results indicate that BtdCPU and halofuginone both induce ISR signaling in wild-type cells and restore ISR signaling in Perk-deficient cells during conditions of ER stress.
Analysis of the transcriptome-wide changes indicate that BtdCPU is a more selective activator of the ISR than halofuginone. Gene set enrichment analysis (GSEA) showed that treatment with BtdCPU (Table S3) significantly induced expression of genes related to the unfolded protein response in both Perk+/+ and Perk−/− MEFs (Fig. 2E, Fig. S2D). This reflects the increased expression of ISR target genes such as Atf3, Chac1, Asns and Atf4 induced by this compound, as markers of other UPR pathways (i.e., IRE1/XBP1s or ATF6) are not significantly induced upon BtdCPU treatment in these cells (Fig. 2D). GSEA also identified increased inflammatory and hypoxic signaling in BtdCPU-treated cells (Fig. 2E, Fig. S2D). However, this appears to be primarily driven by the expression of established ISR target genes such as Atf3 and PPP1R15A, which are included in all these genesets (Table S3). By profiling selective transcriptional targets of these and other stress-responsive signaling pathways (Grandjean et al., 2019), we found that BtdCPU did not broadly induce activation of other stress-responsive signaling pathways (Fig. S2E, Table S4). Thus, our results indicate that BtdCPU is a preferential activator of ISR signaling in these cells.
In contrast, numerous pathways are impacted by halofuginone treatment in Perk+/+ and Perk−/− MEFs (Fig. 2F, Fig. S2F). These include the UPR (reflecting ISR activation) and NFκB-mediated inflammatory signaling. Consistent with this, we found that genesets comprised of ISR and NFκB targets are increased by halofuginone treatment in both genotypes (Fig. S2E, Table S4). However, halofuginone did not significantly induce expression of genes regulated by other stress-responsive signaling pathways such as the IRE1/XBP1s or ATF6 arms of the UPR, heat shock response (HSR), or oxidative stress response (OSR). Thus, while halofuginone robustly induces stress-responsive signaling through the ISR in both Perk+/+ and Perk−/− MEFs, our results indicate that this activity is not selective transcriptome-wide.
## BtdCPU and halofuginone reduce ER stress sensitivity of Perk-deficient cells
Perk-deficiency leads to increased cellular sensitivity to ER stress (Harding et al., 2000). Consistent with this Perk−/− MEFs show reduced proliferation following Tg treatment, as compared to Perk+/+ MEFs, when measured by crystal violet staining (Fig. S3A). Re-overexpression of PERKWT rescues the increased ER stress sensitivity of Perk−/− MEFs, confirming this effect can be attributed to PERK. Co-treatment with BtdCPU or halofuginone improved proliferation of Perk-deficient cells treated with Tg (Fig. 3A, Fig. S3B). Other GCN2 activators including erlotinib and sunitinib also showed improved proliferation in Tg-treated Perk−/− MEFs (Fig. S3B). In contast, the putative PERK activator CCT020312 did not increase viability in Tg-treated cells deficient in Perk. Further, MK28 showed toxicity in both Perk+/+ and Perk−/− MEFs, likely reflecting a PERK-independent, off-target activity of this compound.
We next monitored the activity of the pro-apoptotic caspases 3 and 7 in Perk+/+ and Perk−/− MEFs co-treated with Tg and the ISR activators BtdCPU or halofuginone using Caspase-Glo $\frac{3}{7.}$ Tg-treated Perk−/− MEFs showed higher caspase $\frac{3}{7}$ activity, as compared to Tg-treated Perk+/+ MEFs (Fig. 3B). Treatment with BtdCPU or halofuginone, on their own, did not significantly influence caspase $\frac{3}{7}$ activity in either Perk+/+ or Perk−/− MEFs. However, co-treatment with Tg and either BtdCPU or halofuginone reduced Tg-dependent caspase activity in both cell types (Fig. 3B). Similar results were observed when caspase activity was monitored by immunoblotting for active, cleaved caspase 3 or cleaved PARP, a caspase substrate (Fig. 3C-E) These results indicate that pharmacologic ISR activators such as BtdCPU or halofuginone restores ER stress sensitivity in Perk-deficient cells.
## BtdCPU and halofuginone differentially impact mitochondrial morphology.
PERK-dependent ISR activation promotes protective mitochondrial elongation in response to ER stress (Lebeau et al., 2018; Perea et al., 2022). Genetic depletion of Perk increases mitochondrial fragmentation and ablates ER stress-dependent increases in elongated mitochondria (Lebeau et al., 2018; Perea et al., 2022). Thus, we sought to determine whether compensatory ISR kinase activation could rescue mitochondrial elongation in Perk-deficient cells. We monitored mitochondrial morphology in Perk+/+ and Perk−/− MEFs transiently expressing mitochondrial-targeted GFP (mtGFP) treated with Tg, BtdCPU, or halofuginone. As reported previously (Lebeau et al., 2018; Perea et al., 2022), Perk−/− MEFs showed increased basal mitochondrial fragmentation and were refractory to Tg-induced mitochondrial elongation. ( Fig. 4A,B). Further, the increase in elongated mitochondria observed in Tg-treated Perk+/+ MEFs was inhibited by co-treatment with ISRIB – an ISR inhibitor that binds to eIF2B and desensitizes cells to eIF2α phosphorylation (Schoof et al, 2021; Sidrauski et al, 2013; Tsai et al, 2018; Zyryanova et al, 2018). These results further confirm that the mitochondria elongation observed in Tg-treated cells is attributed to PERK-dependent ISR signaling.
Treatment with BtdCPU did not increase mitochondria elongation (Fig. 4A,B). Instead, BtdCPU increased populations of fragmented mitochondria in both Perk+/+ and Perk−/− MEFs (Fig. 4A,B). Similar results were observed in MEF cells stably expressing mtGFP (MEFmtGFP; Fig. S4A). This increase in fragmentation was not inhibited by co-treatment with ISRIB, indicating that this effect is not attributed to ISR signaling (Fig. 4A,B, Fig. S4A). In contrast, halofuginone increased mitochondrial elongation in both Perk+/+ and Perk−/− MEFs to levels similar to that observed in Tg-treated Perk+/+ cells (Fig. 4A,B). Halofuginone-dependent increases in elongated mitochondria were also observed in MEFmtGFP cells (Fig. S4B,C). Co-treatment with ISRIB inhibited halofuginone-dependent increases in mitochondrial elongation in both genotypes, indicating that the observed elongation results from ISR signaling (Fig. 4A,B). Consistent with this, halofuginone treatment demonstrated impaired mitochondrial elongation in MEFA/A cells expressing the non-phosphorylatable S51A eIF2α mutant (Scheuner et al., 2001) (Fig. S4D,E). These results show that halofuginone and BtdCPU differentially impact mitochondrial morphology. While BtdCPU increases mitochondrial fragmentation, halofuginone promotes adaptive mitochondrial elongation in both Perk+/+ and Perk−/− MEFs through activation of the ISR, mimicking the mitochondrial elongation induced by PERK activation (Lebeau et al., 2018; Perea et al., 2022).
## BtdCPU induces mitochondrial depolarization to promote OMA1-DELE1-HRI signaling
Mitochondrial fragmentation is a marker of mitochondrial dysfunction. Thus, the increased fragmentation observed in cells treated with BtdCPU suggests that this compound disrupts mitochondrial activity. To probe this, we monitored mitochondrial respiratory chain activity in Perk+/+ MEFs treated with BtdCPU using Seahorse (Fig. 5A). Treatment with BtdCPU did not impact basal respiration, but reduced ATP-linked respiration (Fig. 5A,B). This corresponded with an increase of proton leakage, suggesting that BtdCPU impacted mitochondrial membrane potential. Consistent with this, we observed reductions of mitochondrial membrane potential in Perk+/+ MEFs treated with BtdCPU, as measured by TMRE staining (Fig. 5C). Similar results were observed in other cell types including HEK293T and SHSY5Y cells (Fig. S5A,B). Co-treatment with ISRIB did not influence BtdCPU-dependent mitochondrial depolarization (Fig. S5B). This indicates that BtdCPU promotes mitochondrial depolarization independent of ISR activation.
Mitochondrial depolarization can activate ISR signaling through a mechanism involving stress-induced activation of the mitochondrial protease OMA1, cytosolic accumulation of full-length or a C-terminal cleavage products of DELE1, and subsequent DELE1-dependent HRI activation (Fessler et al., 2020; Fessler et al, 2022; Guo et al., 2020). Our results showing that BtdCPU depolarizes mitochondria suggests that this compound could activate ISR signaling through this OMA1-DELE1-HRI signaling axis. Consistent with this, deletion of HRI or DELE1 reduces BtdCPU-dependent ATF4 expression in HEK293T cells, mirroring results observed with the mitochondrial uncoupler CCCP (Fig. 5D,E). Similarly, OMA1-deletion also inhibited BtdCPU-dependent ATF4 expression (Fig. 5F). Re-overexpression of OMA1 restored ATF4 induction in BtdCPU-treated OMA1-deficient cells, confirming that this effect can be attributed to OMA1.
OMA1 is a stress-activated protease localized to the inner mitochondrial membrane that is activated in response to mitochondrial stressors such as membrane depolarization (Zhang et al, 2014). Once activated, OMA1 induces proteolytic processing of substrates including DELE1 and the inner membrane GTPase OPA1, which induces HRI activation and mitochondrial fission, respectively (Fessler et al., 2020; Guo et al., 2020; MacVicar & Langer, 2016). We found that BtdCPU increased proteolytic processing of full-length DELE1-mClover expressed in HEK293T cells (Fig. 5G,H). Further, OMA1-dependent processing of OPA1 from long to short isoforms was increased in BtdCPU-treated HEK293T cells (Fig. 5F). Similar results were observed in MEFA/A cells expressing the non-phosphorylatable S51A eIF2α mutant (Scheuner et al., 2001), indicating that this increase of OMA1-dependent OPA1 processing is independent of ISR activity (Fig. 5I). We also found that OMA1-deletion did not influence BtdCPU-dependent mitochondria depolarization, indicating that OMA1 activation was downstream of mitochondrial uncoupling (Fig. S5C). Collectively, these results indicate that BtdCPU-dependent mitochondria depolarization activates OMA1 proteolytic activity to promote mitochondrial fragmentation and ISR activation through a mechanism involving the OMA1-DELE1-HRI signaling pathway, revealing new insights into the mechanism of action for ISR activation afforded by this compound.
## Halofuginone promotes adaptive remodeling of mitochondria respiration in Perk-deficient cells.
Mitochondrial elongation is an adaptive mechanism that regulates mitochondrial bioenergetics during conditions of stress (Gomes & Scorrano, 2011; Yao et al, 2019). Thus, our results that show PERK signaling impacts mitochondrial morphology both in the absence and presence of acute ER stress suggest an important role for PERK in regulating mitochondrial energy production. To test this, we monitored mitochondrial respiration in Perk-deficient MEFs using Seahorse. Cells deficient in Perk show increased basal respiration, ATP-linked respiration, spare respiratory capacity, and proton leak, as compared to Perk+/+ MEFs (Fig 6A,B). This is in contrast to experiments showing Perk-depletion in other cell types including HEK293, myeloid leukemia, and brown adipocyte cells reduce basal respiration (Bassot et al, 2023; Grenier et al, 2022; Kato et al, 2020). However, we demonstrate that re-overexpression of PERKWT in Perk−/− MEFs restores basal respiratory chain activity to the same levels observed in Perk+/+ MEFs, indicating the observed changes in Perk-deficient MEFs can be attributed to the presence of PERK. This likely highlights cell-type specific compensatory regulation of mitochondrial respiration associated with Perk-deficiency.
Regardless, we show that Perk-deficient MEFs show impaired regulation of respiratory chain activity in response to acute ER stress. In Perk+/+ MEFs, treatment with Tg for 3 h reduces both basal mitochondrial respiration and ATP-linked respiration, while enhancing spare respiratory capacity (Fig. 6C,D; Fig. S6A). However, these changes were not observed in Perk-deficient cells. These results highlight an important role for PERK signaling in regulating mitochondrial respiration in the presence of acute ER stress.
Since halofuginone increases mitochondrial elongation in both Perk+/+ and Perk−/− MEFs (Fig. 4A,B) and mitochondrial elongation is linked to adaptive regulation of mitochondrial respiration (Gomes & Scorrano, 2011; Yao et al., 2019), we anticipated that treatment with this compound should promote similar changes to mitochondrial respiration to those observed in wild-type cells treated with PERK-activating ER stressors (e.g., Tg). Consistent with this, treatment of Perk+/+ MEFs with halofuginone induced identical changes of mitochondrial respiration to those observed in Tg-treated cells (Fig. 6C). This includes reductions in basal respiration and ATP-linked respiration, and increases in spare respiratory capacity (Fig. 6D, Fig. S6A,B). Halofuginone also reduced basal respiration and ATP-linked respiration in Perk−/− MEFs, restoring these parameters to levels similar to those observed in wild-type cells (Fig. 6C,D). However, halofuginone did not alter spare respiratory capacity or proton leakage in Perk-deficient cells (Fig. S6A,B), likely reflecting a chronic consequence of deficient PERK activity on mitochondrial function that cannot be rescued by this short treatment. These results indicate that halofuginone can induce adaptive remodeling of mitochondrial respiration in both wild-type and Perk-deficient cells, mimicking changes observed upon PERK activation during ER stress.
## DISCUSSION
Pharmacologic ISR activation has emerged as a promising strategy to mitigate pathologies implicated in etiologically-diverse diseases including many types of cancers, ischemic diseases, and neurodegenerative disorders such as PSP (Almeida et al., 2022; Chen et al., 2011; Ganz et al., 2020; Hughes & Mallucci, 2019; Rosarda et al, 2021; Zhang et al, 2022). Here, we sought to further define the potential for pharmacologic ISR activation to promote adaptive mitochondrial remodeling and prevent pathologic mitochondrial dysfunction induced under conditions such as Perk-deficiency (Almeida et al., 2022; Hoglinger et al., 2011; Stutzbach et al, 2013). By probing the activity of established ISR activating compounds, we demonstrate that treatment with halofuginone restores ER stress sensitivity and promotes adaptive mitochondrial remodeling in both wild-type and Perk-deficient cells. These results underscore the potential for pharmacologic ISR kinase activation to mitigate mitochondrial dysfunction associated with diverse disorders.
Numerous compounds have been reported to selectively activate ISR kinases independent of cellular stress; however, each of these compounds has liabilities that limit their ability to probe the biological and therapeutic potential for ISR signaling in cellular and in vivo models of disease. The PERK activator CCT020312 has been used to activate PERK signaling in multiple models, including mouse models of PSP where CCT020312 was shown to promote clearance of tau aggregates (Bruch et al., 2017). However, recent results suggest this increased tau clearance reflects an off-target activity of this compound that increases autophagy independent of the ISR, questioning the dependence of this effect on PERK activation (Yoon et al, 2022). Similarly, the PERK activator MK28 reduces toxic huntingtin aggregation in mouse models of HD (Ganz et al., 2020). However, we found this compound to be cytotoxic in Perk+/+ and Perk−/− MEFs (Fig. S3B), likely reflecting a PERK-independent, off-target activity. Further, the mechanism of both CCT020312- and MK28-dependent PERK activation is currently poorly understood. While elucidation of the mechanism of PERK activation for these compounds could enable the establishment of next generation compounds with improved selectivity and potency, the use of CCT020312 and MK28 as PERK activators should be approached with care and include use of both pharmacologic or genetic controls to confirm that specific phenotypes can be attributed to PERK activation, as opposed to off-target activities.
BtdCPU and related N,N’ diarylureas have largely been developed to increase apoptotic signaling downstream of HRI for the treatment of hematologic cancers including multiple myeloma and acute leukemia (Burwick et al, 2017; Smith et al, 2021). However, the mechanism of BtdCPU-dependent HRI activation was previously undefined. We demonstrate that BtdCPU disrupts the mitochondrial membrane potential to activate the stress-activated mitochondrial protease OMA1. This leads to ISR activation through the OMA1-DELE1-HRI mitochondrial stress-signaling axis and mitochondrial fragmentation through OMA1-dependent proteolytic processing of OPA1. Since mitochondrial uncoupling is associated many pathologic conditions, this mechanism of action limits the application of BtdCPU and related compounds to promote protective ISR signaling and mitochondrial adaptation in context of other diseases.
Unlike other ISR activator compounds discussed above, the mechanism of halofuginone-dependent ISR activation is known to be attributed to its inhibition of the glutamyl-prolyl tRNA synthetase (Keller et al., 2012). This leads to accumulation of uncharged proline tRNAs and subsequent GCN2 activation. Halofuginone has been shown to be protective in models of etiologically-diverse diseases including ischemic disorders, cardiovascular disease, and many cancers (Ishii et al., 2009; Juarez et al., 2012; Pines & Spector, 2015). However, apart from ISR activation, halofuginone can also promote protection through other mechanisms including inhibition of TGFβ signaling (Pines & Spector, 2015). In addition, our RNAseq transcriptional profiling indicates that halofuginone does not show transcriptome-wide selectivity for the ISR, as it also induces expression of genes regulated by other pathways including NFκB-mediated inflammatory signaling. Further, higher doses of halofuginone can inhibit global translation independent of the ISR, owing to its glutamyl-prolyl tRNA synthetase inhibition (Keller et al., 2012; Pitera et al, 2022). These effects may limit the translational potential for halofuginone-dependent ISR activation as a strategy to mitigate mitochondrial dysfunction in disease.
Regardless of these limitations, we show that halofuginone-dependent ISR activation both promotes ER stress sensitivity and adaptive mitochondrial remodeling in wild-type and Perk-deficient cells. This demonstrates the potential for pharmacologic activation of compensatory ISR kinases to promote adaptive mitochondria remodeling, even in cells deficient in PERK. However, our results also demonstrate the need for continued development of highly selective ISR kinase activators that can be used to further probe the potential for this approach to mitigate cellular and mitochondrial dysfunction in disease. While pharmacologic ISR activation offers unique opportunities to promote protective signaling through this pathway, a challenge for the translational development of pharmacologic ISR kinase activators is the pro-apoptotic signaling that could result from chronic ISR activation (Hetz & Papa, 2018). Previous results have shown that optimization of compound PK/PD can allow transient, in vivo activation of similar stress-responsive signaling pathways such as the IRE1/XBP1s signaling arm of the UPR to induce protective, adaptive signaling without the pathologic consequences of chronic pathway activation (Madhavan et al, 2022). Similar strategies could be applied for ISR kinase activating compounds to allow protective, adaptive ISR signaling without inducing pro-apoptotic signaling associated with chronic ISR activity. As we, and others, continue identifying ISR kinase activating compounds, the therapeutic potential and optimized activity of this class of compounds will continue to be defined, revealing new insights into the translational potential of pharmacologic ISR kinase activation to mitigate mitochondrial dysfunction for diverse disorders.
## Cell Culture, Transfections, Lentiviral Transduction
Perk+/+ and Perk−/− MEFs (kind gifts from David Ron, Cambridge)(Harding et al., 2000) and MEFA/A cells (kind gifts from Randal Kaufman; Sanford-Burnham-Prebys)(Scheuner et al., 2001) were cultured as previously described at 37oC and $5\%$ CO2 in DMEM (Corning-Cellgro) supplemented with $10\%$ fetal bovine serum (FBS; Omega Scientific), 2 mM L-glutamine (GIBCO), 100 Units/mL penicillin, 100 mg/mL streptomycin (GIBCO), non-essential amino acids (GIBCO), and 2-mercaptoethanol (ThermoFisher). HEK293T cells (purchased from ATCC) and SHSY5Y cells (purchased from ATCC) were cultured at 37°C and $5\%$ CO2 in DMEM (Corning-Cellgro) supplemented with $10\%$ fetal bovine serum (FBS; Omega Scientific), 2 mM L-glutamine (GIBCO), 100 Units/mL penicillin, and 100 mg/mL streptomycin (GIBCO). OMA1-deficient HEK293T cells, DELE1-deficient HEK293T cells, and HRI-deficient HEK293T cells were described previously (Guo et al., 2020) and cultured at 37°C and $5\%$ CO2 in DMEM (Corning-Cellgro) supplemented with $10\%$ fetal bovine serum (FBS; Omega Scientific), 2 mM L-glutamine (GIBCO), 100 Units/mL penicillin, and 100 mg/mL streptomycin (GIBCO). HEK293T cells stably expressing ATF4-FLuc or ATF4-mAPPLE CRISPRi-depleted of HRI, GCN2, PERK, or PKR were described previously (Guo et al., 2020; Yang et al., 2022) and cultured at 37°C and $5\%$ CO2 in DMEM (Corning-Cellgro) supplemented with $10\%$ fetal bovine serum (FBS; Omega Scientific), 2 mM L-glutamine (GIBCO), 100 Units/mL penicillin, and 100 mg/mL streptomycin (GIBCO). MEFmtGFP cells (kind gift from Peter Schultz, TSRI)(Wang et al, 2012) were cultured at 37°C and $5\%$ CO2 in DMEM (Corning-Cellgro) supplemented with $10\%$ fetal bovine serum (FBS; Omega Scientific), 2 mM L-glutamine (GIBCO), 100 Units/mL penicillin, and 100 mg/mL streptomycin (GIBCO). MEF cells were transfected with MEF Avalanche Transfection Reagent (EZ Biosystems) according to the manufacturer’s protocol.
## Plasmids, compounds, and reagents
All compounds used in this study were purchased: thapsigargin (Tg; 50-464-295, Fisher Scientific),, ISRIB (SML0843, Sigma), CCCP (C2759, Sigma), BtdCPU (32-489-210MG, Fisher), halofuginone (50-576-30001, Sigma), erlotinib (S7786, Selleckchem), sunitinib (SU11248, Selleckchem), MK-28 (HY-137207, MedChemExpress), CCT020312 (HY-119240, Fisher), staurosporine (S1421, Selleckchem). PerkWT overexpression plasmid was a kind gift from Jonathan Lin (Stanford) (Yuan et al., 2018). The mitochondrial-targeted GFP plasmid (Lebeau et al., 2018) and the DELE1L-mClover plasmid (Guo et al., 2020) were described previously.
## Measurements of ISR activation in ATF4-reporter cell lines
ATF4-FLuc reporter cells were seeded at a density of 15,000 cells per well in Greiner Bio-One CELLSTAR flat 384-well white plates with clear bottoms. The following day, cells were treated with the indicated compound in triplicate 10-point dose response format for eight hours. After treatment, an equal volume of Promega Bright-Glo substrate was added to the wells and allowed to incubate at room temperature for 10 minutes. Luminescence was then measured in an Infinite F200 PRO plate reader (Tecan) with an integration time of 1000 ms.
ATF4-mApple reporter cells were seeded at a density of 300,000 cells per well in 6-well TC-treated flat bottom plates (Genesee Scientific). The following day, cells were treated for eight hours with the indicated concentrations. Following treatment, cells were washed twice with phosphate-buffered saline (PBS) and dissociated using TrypLE Express (Thermo Fisher). The enzymatic reaction was neutralized through addition of flow buffer containing PBS and five percent fetal bovine serum (FBS). Flow cytometry was performed on a Bio-Rad ZE5 Cell Analyzer. mApple ($\frac{568}{592}$ nm) was measured using the 561 nm green-yellow laser in combination with the $\frac{577}{15}$ filter. Analysis was performed using FlowJo™ Software (BD Biosciences).
## Fluorescence Microscopy
MEF or HeLa cells transiently transfected with mtGFP or MEFmtGFP cells were seeded at a density of 100,000 cells/well on glass-bottom dishes (MatTek) coated with poly-D-lysine (Sigma) or rat tail collagen 1 (GIBCO). Cells were then treated as indicated and images were recorded with an Olympus IX71 microscope with 60x oil objective (Olympus), a Hamamatsu C8484 camera (Hamamatsu Photonics), and HCI image software (Hamamatsu Photonics). Quantification was performed by blinding the images and then scoring cells based on the presence of primarily fragmented, tubular, or elongated mitochondria, as before (Lebeau et al., 2018). At least three different researchers scored each set of images and these scores were averaged for each individual experiment and all quantifications shown were performed for at least 3 independent experiments quantifying a total of >60 cells/condition across all experiments. The data were then analyzed in PRISM (GraphPad, San Diego, CA) and plotted on a stacked bar plot to show the average morphology and standard error of the mean across all experiments. Statistical comparisons were performed using a 2-way ANOVA in PRISM, comparing the relative amounts of fragmented, tubular, or elongated mitochondria across different conditions.
## Immunoblotting and Antibodies
Whole cells were lysed on ice in HEPES lysis buffer (20 mM Hepes pH 7.4, 100 mM NaCal, 1 mM EDTA, $1\%$ Triton X100) supplemented with 1x Pierce protease inhibitor (ThermoFisher). Total protein concentrations of lysates were then normalized using the Bio-Rad protein assay and lystaes were combined with 1x Laemmli buffer supplemented with 100 mM DTT and boiled for 5 min. Samples (100 µg) were then separated by SDS-PAGE and transferred to nitrocellulose membranes (Bio-Rad). Membranes were blocked with $5\%$ milk in tris-buffered saline (TBS) and then incubated overnight at 4°C with the indicated primary antibody. The next day, membranes were washed in TBS supplemented with Tween, incubated with the species appropriate secondary antibody conjugated to IR-Dye (LICOR Biosciences), and then imaged using an Odyssey Infrared Imaging System (LICOR Biosciences). Quantification was then carried out using the LICOR Imaging Studio software. Primary antibodies were acquired from commercial sources and used in the indicated dilutions in antibody buffer (50mM Tris [pH 7.5], 150mM NaCl supplemented with $5\%$ BSA (w/v) and $0.1\%$ NaN3 (w/v)): ATF4 (Cell Signaling, 1:500), Tubulin [B-5-1-2] (Sigma, 1:5000), HSP60 [LK1] (Thermo Scientific, 1:1000), PERK (C33E10) (Cell Signaling, 1:1000), HA [Clone: 16B12] (Biolegend, 1: 1000), GFP (B2) (Santa Cruz, 1:1000), OPA1 (BD Transduction Labs, 1:2000), OMA1 (Cell Signaling, 1:1000), beta-actin (Cell Signaling 1:5,000), cleaved Caspase 3 (Cell Signaling, 1:1000), PARP (Cell Signaling, 1:1000), GAPDH (Fisher, 1:1000).
## RNA sequencing and analysis
Perk+/+ and Perk−/− MEFs cells were treated for 6 h with respective compounds as noted. Cells were rinsed with DPBS, lysed, and total RNA was collected using the QuickRNA mini kit (Zymo) according to the manufacturer′s instructions. Transcriptional profiling using whole transcriptome RNA sequencing was conducted via BGI Americas on the DNBseq platform with three biological replicates for each condition. All samples were sequenced to a minimum depth of 20 M PE 150 bp stranded reads. Alignment of reads was performed using DNAstar Lasergene SeqManPro to the mouse genome GRCm39 assembly. Aligned reads were imported into ArrayStar 12.2 with Qseq (DNAStar Inc.) to quantify the gene expression levels. Differential expression analysis and statistical significance comparisons were assessed using DESeq 2 v. 1.34 in R compared to vehicle-treated Perk+/+ cells. Mouse annotations were converted to human orthologs using BiomaRt v. 2.50.3 prior to functional gene set enrichment with the R package fast gene set enrichment (fgsea) v. 1.20.0 using the Hallmark Pathways Geneset v7.5.1 from MsigDB. Code can be provided upon request. The complete RNA-seq data is deposited in gene expression omnibus (GEO) as GSE227134.
## Quantitative Polymerase Chain Reaction (qPCR)
The relative mRNA expression of target genes was measured using quantitative RT-PCR. Cells were treated as indicated and then washed with phosphate buffered saline (PBD; Gibco). RNA was extracted using Quick-RNA MiniPrepKit (Zymo Research) according to the manufacturers protocol. RNA (500 ng) was then converted to cDNA using the QuantiTect Reverse Transcription Kit (Qiagen). qPCR reactions were prepared using Power SYBR Green PCR Master Mix (Applied Biosystems), and primers (below) were obtained from Integrated DNA Technologies. Amplification reactions were run in an ABI 7900HT Fast Real Time PCR machine with an initial melting period of 95 °C for 5 min and then 45 cycles of 10 s at 95 °C, 30 s at 60 °C.
## Measurement of Mitochondrial Membrane Potential
Cells were seeded at a density of 200,000 cells per well in a 6-well plate and treated with 500 nM Tg for 3h prior to collection. CCCP (10 µM) was added 50 min before collection, followed by 200 nM TMRE (Thermofisher) 20 min before collection. Following treatments, cells were washed twice with phosphate-buffered saline (PBS) and dissociated using TrypLE Express (Thermo Fisher). The enzymatic reaction was neutralized through addition of flow buffer containing PBS and five percent fetal bovine serum (FBS). Fluorescence intensity of TMRE ($\frac{552}{574}$ nm) for 20,000 cells/condition was measured on a Bio-Rad ZE5 Cell Analyzer using the 561 nm green-yellow laser in combination with the $\frac{577}{15}$ filter. Analysis was performed using FlowJo™ Software (BD Biosciences). Data are presented as geometric mean of the fluorescence intensity from three experiments normalized to vehicle-treated cells.
## Cell Proliferation and Apoptosis Assays
Perk+/+ and Perk−/− MEFs were treated for 6 h with either vehicle (DMSO) or the indicated drugs. Cells were washed with room temperature PBS and then trypsinized. Fresh media was added and cells were counted using a Countess Automated Cell Counter (ThermoFisher). Cells were seeded in a 6-well plate at three different dilutions: 30,000, 15,000 and 7,500 and then allowed to proliferate for five days while incubated at 37°C. At five days, media was aspirated off and cells were rinsed with PBS. Cells were then fixed with crystal violet staining solution ($0.5\%$ crystal violet (w/v), $20\%$ methanol) for 10 minutes in room temperature. The crystal violet staining solution was then removed and cells were washed 3x with PBS and allowed to dry overnight. Images were then taken of the stained plates. Other parameters of ER Stress induced cell death were measured through immunoblotting or with the Caspase $\frac{3}{7}$ Glo Assay Kit (Promega) according to the manufacturer’s protocol. In brief, cells were plated at 10k/well in a white 96-well plate. The next day, cells were treated with the indicated drugs for 24 h ($$n = 5$$ for each condition). After treatment, an equal volume of Caspase $\frac{3}{7}$ substrate was added to the wells and allowed to incubate at room temperature for 1 h at 37oC and $5\%$ CO2. Luminescence was then measured in an Infinite F200 PRO plate reader (Tecan) with an integration time of 1000 ms.
## Mitochondrial Respiration
Mitochondrial respiration parameters were measured using a Mito Stress Test Kit and XF96 Extracellular Flux Analyzer (Seahorse Bioscience) according to the manufacturer’s protocol. In brief, 15,000 cells/well were plated on a 96-well plate on Cell-Tak (Corning) coated wells in their standard growing media and cultured overnight. The next day, cells were treated with drugs and incubated at 37°C, $5\%$ CO2 for the indicated treatment time. Seahorse XF assay media (Agilent) was used to wash and remove the standard growing media from the cell plate and the sensor plate was prepped with standard Mito Stress Test drugs. After calibration of the sensor plate on the XF96 Extracellular Flux Analyzer, the cell plate was inserted into the machine and each parameter was calculated with 4 measurements separated by 5 min intervals following injections of drugs. Basal respiration measurement was followed by injections of oligomycin (2 µM), FCCP (0.5 µM) and Rotenone/Antimycin (1µM each), respectively. Individual respiratory parameters were then calculated according to the manufacturer’s protocol.
## References
1. Almeida LM, Pinho BR, Duchen MR, Oliveira JMA. **The PERKs of mitochondria protection during stress: insights for PERK modulation in neurodegenerative and metabolic diseases.**. *Biol Rev Camb Philos Soc* (2022) **97** 1737-1748. PMID: 35475315
2. Balsa E, Soustek MS, Thomas A, Cogliati S, Garcia-Poyatos C, Martin-Garcia E, Jedrychowski M, Gygi SP, Enriquez JA, Puigserver P. **ER and Nutrient Stress Promote Assembly of Respiratory Chain Supercomplexes through the PERK-eIF2alpha Axis.**. *Mol Cell* (2019) **74** 877-890. PMID: 31023583
3. Barad BA, Medina M, Fuentes D, Wiseman RL, Grotjahn DA. **A surface morphometrics toolkit to quantify organellar membrane ultrastructure using cryo-electron tomography.**. *bioRxiv* (2022)
4. Bassot A, Chen J, Takahashi-Yamashiro K, Yap MC, Gibhardt CS, Le GNT, Hario S, Nasu Y, Moore J, Gutierrez T. **The endoplasmic reticulum kinase PERK interacts with the oxidoreductase ERO1 to metabolically adapt mitochondria.**. *Cell Rep* (2023) **42** 111899. PMID: 36586409
5. Bruch J, Xu H, Rosler TW, De Andrade A, Kuhn PH, Lichtenthaler SF, Arzberger T, Winklhofer KF, Muller U, Hoglinger GU. **PERK activation mitigates tau pathology in vitro and in vivo.**. *EMBO Mol Med* (2017) **9** 371-384. PMID: 28148553
6. Burwick N, Zhang MY, de la Puente P, Azab AK, Hyun TS, Ruiz-Gutierrez M, Sanchez-Bonilla M, Nakamura T, Delrow JJ, MacKay VL. **The eIF2-alpha kinase HRI is a novel therapeutic target in multiple myeloma.**. *Leuk Res* (2017) **55** 23-32. PMID: 28119225
7. Chen T, Ozel D, Qiao Y, Harbinski F, Chen L, Denoyelle S, He X, Zvereva N, Supko JG, Chorev M. **Chemical genetics identify eIF2alpha kinase heme-regulated inhibitor as an anticancer target.**. *Nat Chem Biol* (2011) **7** 610-616. PMID: 21765405
8. Costa-Mattioli M, Walter P. **The integrated stress response: From mechanism to disease.**. *Science* (2020) **368**
9. Delepine M, Nicolino M, Barrett T, Golamaully M, Lathrop GM, Julier C. **EIF2AK3, encoding translation initiation factor 2-alpha kinase 3, is mutated in patients with Wolcott-Rallison syndrome.**. *Nat Genet* (2000) **25** 406-409. PMID: 10932183
10. 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) **579** 433-437. PMID: 32132706
11. Fessler E, Krumwiede L, Jae LT. **DELE1 tracks perturbed protein import and processing in human mitochondria.**. *Nat Commun* (2022) **13** 1853. PMID: 35388015
12. Ganz J, Shacham T, Kramer M, Shenkman M, Eiger H, Weinberg N, Iancovici O, Roy S, Simhaev L, Da’adoosh B. **A novel specific PERK activator reduces toxicity and extends survival in Huntington’s disease models.**. *Sci Rep* (2020) **10** 6875. PMID: 32327686
13. Gardner BM, Pincus D, Gotthardt K, Gallagher CM, Walter P. **Endoplasmic reticulum stress sensing in the unfolded protein response.**. *Cold Spring Harb Perspect Biol* (2013) **5** a013169. PMID: 23388626
14. Gomes LC, Scorrano L. **Mitochondrial elongation during autophagy: a stereotypical response to survive in difficult times.**. *Autophagy* (2011) **7** 1251-1253. PMID: 21743300
15. Grandjean JMD, Plate L, Morimoto RI, Bollong MJ, Powers ET, Wiseman RL. **Deconvoluting Stress-Responsive Proteostasis Signaling Pathways for Pharmacologic Activation Using Targeted RNA Sequencing.**. *ACS Chem Biol* (2019) **14** 784-795. PMID: 30821953
16. Grandjean JMD, Wiseman RL. **Small molecule strategies to harness the unfolded protein response: where do we go from here?**. *J Biol Chem* (2020) **295** 15692-15711. PMID: 32887796
17. Grenier A, Poulain L, Mondesir J, Jacquel A, Bosc C, Stuani L, Mouche S, Larrue C, Sahal A, Birsen R. **AMPK-PERK axis represses oxidative metabolism and enhances apoptotic priming of mitochondria in acute myeloid leukemia.**. *Cell Rep* (2022) **38** 110197. PMID: 34986346
18. Guo X, Aviles G, Liu Y, Tian R, Unger BA, Lin YT, Wiita AP, Xu K, Correia MA, Kampmann M. **Mitochondrial stress is relayed to the cytosol by an OMA1-DELE1-HRI pathway.**. *Nature* (2020) **579** 427-432. PMID: 32132707
19. Han J, Back SH, Hur J, Lin YH, Gildersleeve R, Shan J, Yuan CL, Krokowski D, Wang S, Hatzoglou M. **ER-stress-induced transcriptional regulation increases protein synthesis leading to cell death.**. *Nat Cell Biol* (2013) **15** 481-490. PMID: 23624402
20. Harding HP, Zhang Y, Bertolotti A, Zeng H, Ron D. **Perk is essential for translational regulation and cell survival during the unfolded protein response.**. *Mol Cell* (2000) **5** 897-904. PMID: 10882126
21. Harding HP, Zhang Y, Zeng H, Novoa I, Lu PD, Calfon M, Sadri N, Yun C, Popko B, Paules R. **An integrated stress response regulates amino acid metabolism and resistance to oxidative stress.**. *Mol Cell* (2003) **11** 619-633. PMID: 12667446
22. Hetz C, Papa FR. **The Unfolded Protein Response and Cell Fate Control.**. *Mol Cell* (2018) **69** 169-181. PMID: 29107536
23. Hetz C, Zhang K, Kaufman RJ. **Mechanisms, regulation and functions of the unfolded protein response.**. *Nat Rev Mol Cell Biol* (2020) **21** 421-438. PMID: 32457508
24. Hoglinger GU, Melhem NM, Dickson DW, Sleiman PM, Wang LS, Klei L, Rademakers R, de Silva R, Litvan I, Riley DE. **Identification of common variants influencing risk of the tauopathy progressive supranuclear palsy.**. *Nat Genet* (2011) **43** 699-705. PMID: 21685912
25. Hori O, Ichinoda F, Tamatani T, Yamaguchi A, Sato N, Ozawa K, Kitao Y, Miyazaki M, Harding HP, Ron D. **Transmission of cell stress from endoplasmic reticulum to mitochondria: enhanced expression of Lon protease.**. *J Cell Biol* (2002) **157** 1151-1160. PMID: 12082077
26. Hughes D, Mallucci GR. **The unfolded protein response in neurodegenerative disorders - therapeutic modulation of the PERK pathway.**. *FEBS J* (2019) **286** 342-355. PMID: 29476642
27. Ishii H, Choudhuri R, Mathias A, Sowers AL, Flanders KC, Cook JA, Mitchell JB. **Halofuginone mediated protection against radiation-induced leg contracture.**. *Int J Oncol* (2009) **35** 315-319. PMID: 19578745
28. Juarez P, Mohammad KS, Yin JJ, Fournier PG, McKenna RC, Davis HW, Peng XH, Niewolna M, Javelaud D, Chirgwin JM. **Halofuginone inhibits the establishment and progression of melanoma bone metastases.**. *Cancer Res* (2012) **72** 6247-6256. PMID: 23002206
29. Julier C, Nicolino M. **Wolcott-Rallison syndrome.**. *Orphanet J Rare Dis* (2010) **5** 29. PMID: 21050479
30. Kato H, Okabe K, Miyake M, Hattori K, Fukaya T, Tanimoto K, Beini S, Mizuguchi M, Torii S, Arakawa S. **ER-resident sensor PERK is essential for mitochondrial thermogenesis in brown adipose tissue.**. *Life Sci Alliance* (2020) **3**
31. Keller TL, Zocco D, Sundrud MS, Hendrick M, Edenius M, Yum J, Kim YJ, Lee HK, Cortese JF, Wirth DF. **Halofuginone and other febrifugine derivatives inhibit prolyl-tRNA synthetase.**. *Nat Chem Biol* (2012) **8** 311-317. PMID: 22327401
32. Latorre-Muro P, O’Malley KE, Bennett CF, Perry EA, Balsa E, Tavares CDJ, Jedrychowski M, Gygi SP, Puigserver P. **A cold-stress-inducible PERK/OGT axis controls TOM70-assisted mitochondrial protein import and cristae formation.**. *Cell Metab* (2021) **33** 598-614. PMID: 33592173
33. Lebeau J, Saunders JM, Moraes VWR, Madhavan A, Madrazo N, Anthony MC, Wiseman RL. **The PERK Arm of the Unfolded Protein Response Regulates Mitochondrial Morphology during Acute Endoplasmic Reticulum Stress.**. *Cell Rep* (2018) **22** 2827-2836. PMID: 29539413
34. MacVicar T, Langer T. **OPA1 processing in cell death and disease - the long and short of it.**. *J Cell Sci* (2016) **129** 2297-2306. PMID: 27189080
35. Madhavan A, Kok BP, Rius B, Grandjean JMD, Alabi A, Albert V, Sukiasyan A, Powers ET, Galmozzi A, Saez E. **Pharmacologic IRE1/XBP1s activation promotes systemic adaptive remodeling in obesity.**. *Nat Commun* (2022) **13** 608. PMID: 35105890
36. Mann JP, Lenz D, Stamataki Z, Kelly D. **Common mechanisms in pediatric acute liver failure.**. *Trends Mol Med* (2022)
37. Munoz JP, Ivanova S, Sanchez-Wandelmer J, Martinez-Cristobal P, Noguera E, Sancho A, Diaz-Ramos A, Hernandez-Alvarez MI, Sebastian D, Mauvezin C. **Mfn2 modulates the UPR and mitochondrial function via repression of PERK.**. *EMBO J* (2013) **32** 2348-2361. PMID: 23921556
38. Pakos-Zebrucka K, Koryga I, Mnich K, Ljujic M, Samali A, Gorman AM. **The integrated stress response.**. *EMBO Rep* (2016) **17** 1374-1395. PMID: 27629041
39. Park G, Xu K, Chea L, Kim K, Safarta L, Song KH, Wu J, Park S, Min H, Hiramatsu N. **Neurodegeneration risk factor, EIF2AK3 (PERK), influences tau protein aggregation.**. *J Biol Chem* (2022) **299** 102821. PMID: 36563857
40. Perea V, Cole C, Lebeau J, Dolina V, Baron KR, Madhavan A, Kelly JW, Grotjahn DA, Wiseman RL. **PERK Signaling Promotes Mitochondrial Elongation By Remodeling Membrane Phosphatidic Acid.**. *bioRxiv* (2022)
41. Pines M, Spector I. **Halofuginone - the multifaceted molecule.**. *Molecules* (2015) **20** 573-594. PMID: 25569515
42. Pitera AP, Szaruga M, Peak-Chew SY, Wingett SW, Bertolotti A. **Cellular responses to halofuginone reveal a vulnerability of the GCN2 branch of the integrated stress response.**. *EMBO J* (2022) **41** e109985. PMID: 35466425
43. Rainbolt TK, Atanassova N, Genereux JC, Wiseman RL. **Stress-regulated translational attenuation adapts mitochondrial protein import through Tim17A degradation.**. *Cell Metab* (2013) **18** 908-919. PMID: 24315374
44. Rainbolt TK, Saunders JM, Wiseman RL. **Stress-responsive regulation of mitochondria through the ER unfolded protein response.**. *Trends Endocrinol Metab* (2014) **25** 528-537. PMID: 25048297
45. Rosarda JD, Baron KR, Nutsch K, Kline GM, Stanton C, Kelly JW, Bollong MJ, Wiseman RL. **Metabolically Activated Proteostasis Regulators Protect against Glutamate Toxicity by Activating NRF2.**. *ACS Chem Biol* (2021) **16** 2852-2863. PMID: 34797633
46. Scheuner D, Song B, McEwen E, Liu C, Laybutt R, Gillespie P, Saunders T, Bonner-Weir S, Kaufman RJ. **Translational control is required for the unfolded protein response and in vivo glucose homeostasis.**. *Mol Cell* (2001) **7** 1165-1176. PMID: 11430820
47. Schoof M, Boone M, Wang L, Lawrence R, Frost A, Walter P. **eIF2B conformation and assembly state regulate the integrated stress response.**. *Elife* (2021) **10**
48. Shacham T, Patel C, Lederkremer GZ. **PERK Pathway and Neurodegenerative Disease: To Inhibit or to Activate?**. *Biomolecules* (2021) **11**
49. Sidrauski C, Acosta-Alvear D, Khoutorsky A, Vedantham P, Hearn BR, Li H, Gamache K, Gallagher CM, Ang KK, Wilson C. **Pharmacological brake-release of mRNA translation enhances cognitive memory.**. *Elife* (2013) **2** e00498. PMID: 23741617
50. Smith KH, Budhraja A, Lynch J, Roberts K, Panetta JC, Connelly JP, Turnis ME, Pruett-Miller SM, Schuetz JD, Mullighan CG. **The Heme-Regulated Inhibitor Pathway Modulates Susceptibility of Poor Prognosis B-Lineage Acute Leukemia to BH3-Mimetics.**. *Mol Cancer Res* (2021) **19** 636-650. PMID: 33288732
51. Stockwell SR, Platt G, Barrie SE, Zoumpoulidou G, Te Poele RH, Aherne GW, Wilson SC, Sheldrake P, McDonald E, Venet M. **Mechanism-based screen for G1/S checkpoint activators identifies a selective activator of EIF2AK3/PERK signalling.**. *PLoS One* (2012) **7** e28568. PMID: 22253692
52. Stutzbach LD, Xie SX, Naj AC, Albin R, Gilman S, Lee VM, Trojanowski JQ, Devlin B, Schellenberg GD. **The unfolded protein response is activated in disease-affected brain regions in progressive supranuclear palsy and Alzheimer’s disease.**. *Acta Neuropathol Commun* (2013) **1** 31. PMID: 24252572
53. Tang CP, Clark O, Ferrarone JR, Campos C, Lalani AS, Chodera JD, Intlekofer AM, Elemento O, Mellinghoff IK. **GCN2 kinase activation by ATP-competitive kinase inhibitors.**. *Nat Chem Biol* (2022) **18** 207-215. PMID: 34949839
54. Tian X, Zhang S, Zhou L, Seyhan AA, Hernandez Borrero L, Zhang Y, El-Deiry WS. **Targeting the Integrated Stress Response in Cancer Therapy.**. *Front Pharmacol* (2021) **12** 747837. PMID: 34630117
55. Tsai JC, Miller-Vedam LE, Anand AA, Jaishankar P, Nguyen HC, Renslo AR, Frost A, Walter P. **Structure of the nucleotide exchange factor eIF2B reveals mechanism of memory-enhancing molecule.**. *Science* (2018) **359**
56. Verfaillie T, Rubio N, Garg AD, Bultynck G, Rizzuto R, Decuypere JP, Piette J, Linehan C, Gupta S, Samali A. **PERK is required at the ER-mitochondrial contact sites to convey apoptosis after ROS-based ER stress.**. *Cell Death Differ* (2012) **19** 1880-1891. PMID: 22705852
57. Walter P, Ron D. **The unfolded protein response: from stress pathway to homeostatic regulation.**. *Science* (2011) **334** 1081-1086. PMID: 22116877
58. Wang D, Wang J, Bonamy GM, Meeusen S, Brusch RG, Turk C, Yang P, Schultz PG. **A small molecule promotes mitochondrial fusion in mammalian cells.**. *Angew Chem Int Ed Engl* (2012) **51** 9302-9305. PMID: 22907892
59. Wek RC, Cavener DR. **Translational control and the unfolded protein response.**. *Antioxid Redox Signal* (2007) **9** 2357-2371. PMID: 17760508
60. Wolzak K, Nolle A, Farina M, Abbink TE, van der Knaap MS, Verhage M, Scheper W. **Neuron-specific translational control shift ensures proteostatic resilience during ER stress.**. *EMBO J* (2022) **41** e110501. PMID: 35791631
61. Yang J, Baron KR, Pride DE, Schneemann A, Guo X, Chen W, Song AS, Aviles G, Kampmann M, Wiseman RL. **DELE1 oligomerization promotes integrated stress response activation.**. *bioRxiv* (2022)
62. Yao CH, Wang R, Wang Y, Kung CP, Weber JD, Patti GJ. **Mitochondrial fusion supports increased oxidative phosphorylation during cell proliferation.**. *Elife* (2019) **8**
63. Yoon L, Botham RC, Verhelle A, Sanz-Martinez P, Xu J, Cole CM, Tan EP, Chou CC, Cuoco CA, Massey LA. **mTOR inhibitor-independent Autophagy Activator Ameliorates Tauopathy and Prionopathy Neurodegeneration Phenotypes.**. *bioRxiv* (2022)
64. Yuan SH, Hiramatsu N, Liu Q, Sun XV, Lenh D, Chan P, Chiang K, Koo EH, Kao AW, Litvan I. **Tauopathy-associated PERK alleles are functional hypomorphs that increase neuronal vulnerability to ER stress.**. *Hum Mol Genet* (2018) **27** 3951-3963. PMID: 30137327
65. Zhang G, Wang X, Rothermel BA, Lavandero S, Wang ZV. **The integrated stress response in ischemic diseases.**. *Cell Death Differ* (2022) **29** 750-757. PMID: 34743204
66. Zhang K, Li H, Song Z. **Membrane depolarization activates the mitochondrial protease OMA1 by stimulating self-cleavage.**. *EMBO Rep* (2014) **15** 576-585. PMID: 24719224
67. Zhang Q, Du R, Reis Monteiro Dos Santos GR, Yefidoff-Freedman R, Bohm A, Halperin J, Chorev M, Aktas BH. **New activators of eIF2alpha Kinase Heme-Regulated Inhibitor (HRI) with improved biophysical properties.**. *Eur J Med Chem* (2020) **187** 111973. PMID: 31881453
68. Zyryanova AF, Weis F, Faille A, Alard AA, Crespillo-Casado A, Sekine Y, Harding HP, Allen F, Parts L, Fromont C. **Binding of ISRIB reveals a regulatory site in the nucleotide exchange factor eIF2B.**. *Science* (2018) **359** 1533-1536. PMID: 29599245
|
---
title: Identifying metabolic features of colorectal cancer liability using Mendelian
randomization
authors:
- Caroline J. Bull
- Emma Hazelwood
- Joshua A. Bell
- Vanessa Y. Tan
- Andrei-Emil Constantinescu
- Maria Carolina Borges
- Danny N. Legge
- Kimberly Burrows
- Jeroen R. Huyghe
- Hermann Brenner
- Sergi Castellví-Bel
- Andrew T Chan
- Sun-Seog Kweon
- Loic Le Marchand
- Li Li
- Iona Cheng
- Rish K. Pai
- Jane C. Figueiredo
- Neil Murphy
- Marc J. Gunter
- Nicholas J. Timpson
- Emma E. Vincent
journal: medRxiv
year: 2023
pmcid: PMC10029059
doi: 10.1101/2023.03.10.23287084
license: CC BY 4.0
---
# Identifying metabolic features of colorectal cancer liability using Mendelian randomization
## Abstract
### Background:
Recognizing the early signs of cancer risk is vital for informing prevention, early detection, and survival.
### Methods:
To investigate whether changes in circulating metabolites characterise the early stages of colorectal cancer (CRC) development, we examined associations between a genetic risk score (GRS) associated with CRC liability (72 single nucleotide polymorphisms) and 231 circulating metabolites measured by nuclear magnetic resonance spectroscopy in the Avon Longitudinal Study of Parents and Children ($$n = 6$$,221). Linear regression models were applied to examine associations between genetic liability to colorectal cancer and circulating metabolites measured in the same individuals at age 8, 16, 18 and 25 years.
### Results:
The GRS for CRC was associated with up to $28\%$ of the circulating metabolites at FDR-$P \leq 0.05$ across all time points, particularly with higher fatty acids and very-low- and low-density lipoprotein subclass lipids. Two-sample reverse Mendelian randomization (MR) analyses investigating CRC liability (52,775 cases, 45,940 controls) and metabolites measured in a random subset of UK Biobank participants ($$n = 118$$,466, median age 58y) revealed broadly consistent effect estimates with the GRS analysis. In conventional (forward) MR analyses, genetically predicted polyunsaturated fatty acid concentrations were most strongly associated with higher CRC risk.
### Conclusions:
These analyses suggest that higher genetic liability to CRC can cause early alterations in systemic metabolism, and suggest that fatty acids may play an important role in CRC development.
### Funding:
This work was supported by the Elizabeth Blackwell Institute for Health Research, University of Bristol, the Wellcome Trust, the Medical Research Council, Diabetes UK, the University of Bristol NIHR Biomedical Research Centre, and Cancer Research UK. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. This work used the computational facilities of the Advanced Computing Research Centre, University of Bristol - http://www.bristol.ac.uk/acrc/.
## Introduction
Colorectal cancer (CRC) is the third most frequently diagnosed cancer worldwide and the fourth most common cause of death from cancer.1,2 *There is* a genetic component to risk of the disease, which is thought to explain up to $35\%$ of variability in CRC risk.3–5 In addition, modifiable lifestyle factors, including obesity, consumption of processed meat, and alcohol are thought to increase CRC risk.2,6–9 However, the underlying biological pathways remain unclear, which limits targeted prevention strategies. Whilst CRC has higher mortality rates when diagnosed at later stages, early-stage CRC or precancerous lesions are largely treatable, meaning colorectal cancer screening programmes have the potential to be highly effective.10,11 Due to the lack of known predictive biomarkers for CRC, wide-scale screening (if implemented at all) is expensive and often targeted crudely by age range. Identifying biomarkers predictive of CRC, or with causal roles in disease development, is therefore vital.
One potential source of biomarkers for CRC risk is the circulating metabolome, which offers a dynamic insight into cellular processes and disease states. It is increasingly clear from mechanistic studies that both systemic and intracellular tumour metabolism play an important role in CRC development and progression.12,13 Interestingly, several major risk factors for CRC are known to have profound effects on metabolism.14 For instance, obesity has been shown via conventional observational and Mendelian randomization (MR) analyses to strongly alter circulating metabolite levels.9,15–17 This suggests that the circulating metabolome may play a mediating role in the relationship between at least some common risk factors, such as obesity, and CRC – or at least might be a useful biomarker for disease or intermediates thereof. In particular, previous work has highlighted polyunsaturated fatty acids (PUFA) as potentially having a role in colorectal cancer development. The term PUFA includes omega-3 and −6 fatty acids. Recent MR work has highlighted a possible link between PUFAs, in particular omega 6 PUFAs, and colorectal cancer risk.18 Further investigating the relationship between CRC and circulating metabolites may therefore provide powerful insights into the causal pathways underlying disease risk, or alternatively may be valuable in prediction and early diagnosis.
MR is a genetic epidemiological approach used to evaluate causal relationships between traits.19,20 This method uses genetic variation as a proxy measure for traits in an instrumental variable framework to assess the causal relevance of the traits in disease development. As germline genetic variants are theoretically randomised between generations and fixed at conception, this approach should be less prone to bias and confounding than conventional analyses undertaken in an observational context. Conventionally, MR is used to investigate the effect of an exposure on a disease outcome. In reverse MR, genetic instruments proxy the association between liability to a disease and other traits.21 This approach can identify biomarkers which cause the disease, are predictive for the disease, or have diagnostic potential.21 Given the suspected importance of the circulating metabolome in CRC development, employing both reverse MR and conventional forward MR for metabolites in the same study may be an efficient approach for revealing causal and predictive biomarkers for CRC. Although previous observational studies have investigated associations between the circulating metabolome and CRC risk, these studies may have been influenced by confounding bias which should be less relevant to MR analyses.22–31 Additionally, these studies focussed on adults, who commonly take medications which may confound metabolite associations, further complicating interpretations.
Here, we applied a reverse MR framework to identify circulating metabolites which are associated with CRC liability across different stages of the early life course (spanning childhood to young adulthood, when use of medications and CRC are both rare) using data from a birth cohort study. We then attempted to replicate these results using reverse two-sample MR in an independent cohort of middle-aged adults (UK Biobank). We then performed conventional ‘forward’ MR of metabolites onto CRC risk using large-scale cancer consortia data to identify metabolites which may have a causal role in CRC development.
## Study populations
This study uses data from 2 cohort studies: the Avon Longitudinal Study of Parents and Children (ALSPAC) offspring (generation 1) cohort (individual-level data) and the UK Biobank cohort (summary-level data); plus summary-level data from a genome-wide association study (GWAS) meta-analysis of CRC comprising the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), Colorectal Transdisciplinary Study (CORECT), and Colon Cancer Family Registry (CCFR).
ALSPAC is a population-based birth cohort study in which 14,541 pregnant women with an expected delivery date between 1 April 1991 and 31 December 1992 were recruited from the former Avon County of southwest England.32 Since then, 13,988 offspring alive at one year have been followed repeatedly with questionnaire- and clinic-based assessments.33,34 Sufficient information was available on 6,221 of these individuals to be included in our analysis, as metabolomics was not performed for all individuals in the ALSPAC study. Study data were collected and managed using REDCap electronic data capture tools hosted at the University of Bristol.35 REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies. Offspring genotype was assessed using the Illumina HumanHap550 quad chip platform. Quality control measures included exclusion of participants with sex mismatch, minimal or excessive heterozygosity, disproportionately missing data, insufficient sample replication, cryptic relatedness, and non-European ancestry. Imputation was performed using the Haplotype Reference Consortium (HRC) panel. Offspring were considered for the current analyses if they had no older siblings in ALSPAC (203 excluded) and were of white ethnicity (based on reports by parents, 604 excluded) to reduce the potential for confounding by genotype. The study website contains details of all available data through a fully searchable data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/our-data/).
UK *Biobank is* a population-based cohort study based in 22 centres across the UK.36 The cohort is made up of around 500,000 adults aged 40–80 years old, who were enrolled between 2006 and 2010. Genotyping data is available for 488,377 participants.37 Participants were genotyped using one of two arrays – either the Applied Biosystems UK BiLEVE Axiom Array by Affymetrix (now part of Thermo Fisher Scientific), or the closely related Applied Biosystems UK Biobank Axiom Array. Approaches based on Principal Component Analysis (PCA) were used to account for population structure. Individuals were excluded: if reported sex differed from inferred sex based on genotyping data; if they had sex chromosome karyotypes which were not XX or XY; if they were outliers in terms of heterozygosity and missing rates; or if they had high relatedness to another participant. Multiallelic SNPs or those with a minor allele frequency of below $1\%$ were removed. Imputation was performed using the UK10K haplotype and HRC reference panels.
The GWAS meta-analysis for CRC included up to 52,775 cases and 45,940 controls.38,39 This sample excluded cases and controls from UK Biobank to avoid potential bias due to sample overlap which may be problematic in MR analyses.40 Cases were diagnosed by a physician and recorded overall and by site (colon, 28,736 cases; proximal colon, 14,416 cases; distal colon, 12,879 cases; and rectal, 14,150 cases). Colon cancer included proximal colon (any primary tumour arising in the cecum, ascending colon, hepatic flexure, or transverse colon), distal colon (any primary tumour arising in the pleenic flexure, descending colon or sigmoid colon), and colon cases with unspecified site. Rectal cancer included any primary tumour arising in the rectum or rectosigmoid junction.38 Approximately $92\%$ of participants in the overall CRC GWAS were white-European (~$8\%$ were East Asian). All participants included in site-specific CRC analyses were of European ancestry. Imputation was performed using the Michigan imputation server and HRC r1.0 reference panel. Regression models were further adjusted for age, sex, genotyping platform, and genomic principal components as described previously.38
## Assessment of CRC genetic liability
Genetic liability to CRC was based on single nucleotide polymorphisms (SNPs) associated with CRC case status at genome-wide significance ($P \leq 5$×10−8). 108 independent SNPs reported by two major GWAS meta-analyses were eligible for inclusion in a CRC genetic risk score (GRS).38,41 The set of SNPs was filtered, excluding 36 SNPs that were in linkage disequilibrium based on R2>0.001 using the TwoSampleMR package (SNPs with the lowest P-values were retained).42 This left 72 SNPs independently associated with CRC (Supplementary File 1a), 65 of which were available in imputed ALSPAC genotype data post quality control. As GWAS of site-specific CRC have identified marked heterogeneity,43 GRS describing site-specific CRCs were constructed for sensitivity analyses using the same process outlined above. The GRS for colon cancer, rectal cancer, proximal colon cancer and distal colon cancer were comprised of 38, 25, 20 and 24 variants, respectively (Supplementary File 1a). For overall CRC and site-specific CRC analyses, sensitivity analyses excluding any SNPs in the FADS cluster (i.e. within the gene regions of FADS1, FADS2, or FADS3) (Supplementary File 1a) were performed given a likely role for these SNPs in influencing circulating metabolite levels directly, in particular via lipid metabolism (i.e., not primarily due to CRC).44–50
## Assessment of circulating metabolites
Circulating metabolite measures were drawn from ALSPAC and UK Biobank using the same targeted metabolomics platform. In ALSPAC, participants provided non-fasting blood samples during a clinic visit while aged approximately 8y, and fasting blood samples from clinic visits while aged approximately 16y, 18y, and 25y. Proton nuclear magnetic resonance (1H-NMR) spectroscopy was performed on Ethylenediaminetetraacetic acid (EDTA) plasma (stored at or below −70 degrees Celsius pre-processing) to quantify a maximum of 231 metabolites.51 Quantified metabolites included the cholesterol and triglyceride content of lipoprotein particles; the concentrations and diameter/size of these particles; apolipoprotein B and apolipoprotein A-1 concentrations; as well as fatty acids and their ratios to total fatty acid concentration, branched chain and aromatic amino acids, glucose and pre-glycaemic factors including lactate and citrate, fluid balance factors including albumin and creatinine, and the inflammatory marker glycoprotein acetyls (GlycA). This metabolomics platform has limited coverage of fatty acids. In UK Biobank, EDTA plasma samples from 117,121 participants, a random subset of the original ~500,000 who provided samples at assessment centres between 2006 and 2013, were analysed between 2019 and 2020 for levels of 249 metabolic traits (168 concentrations plus 81 ratios) using the same high-throughput 1H-NMR platform. Data pre-processing and QC steps are described previously.51–53 To allow comparability between MR and GRS estimates all metabolite measures were standardised and normalised using rank-based inverse normal transformation. For descriptive purposes in ALSPAC, body mass index (BMI) was calculated at each time point as weight (kg) divided by squared height (m2) based on clinic measures of weight to the nearest 0.1 kg using a Tanita scale and height measured in light clothing without shoes to the nearest 0.1 cm using a Harpenden stadiometer.
CRC liability variants were combined into a GRS using PLINK 1.9, specifying the effect (risk raising) allele and coefficient (logOR) with estimates from the CRC GWAS used as external weights.38,41 GRSs were calculated as the number of effect alleles (or dosages if imputed) at each SNP (0, 1, or 2) multiplied by its weighting, summing these, and dividing by the total number of SNPs used. Z-scores of GRS variables were calculated to standardize scoring.
## Statistical approach
An overview of the study design is presented in Figure 1. To estimate the effect of increased genetic liability to CRC on circulating metabolites we conducted a GRS analysis in ALSPAC and reverse two-sample MR analyses in UK Biobank. Estimates were interpreted within a ‘reverse MR’ framework,54 wherein results are taken to reflect ‘metabolic features’ of CRC liability which could capture causal or predictive metabolite-disease associations. To clarify the direction of metabolite-CRC associations, we additionally performed conventional ‘forward’ two-sample MR analyses to estimate the effect of circulating metabolites on CRC risk using large-scale GWAS data on metabolites and CRC.
## Associations of CRC liability with circulating metabolites in early life
Separate linear regression models with robust standard errors were used to estimate coefficients and $95\%$ confidence intervals for associations of GRSs with each metabolite as a dependent variable measured on the same individuals at age 8y, 16y, 18y, and 25y, adjusted for sex and age at the time of metabolite assessment. To aid interpretations, estimates were multiplied by 0.693 (loge2) to reflect SD-unit differences in metabolites per doubling of genetic liability to CRC.55 The Benjamini-Hochberg method was used to adjust P-values for multiple testing and an adjusted P-value of <0.05 was used as a heuristic for evidence for association given current sample sizes.56 At the time the ALSPAC blood samples were taken, the mean age of participants was 7.5y ($$n = 4$$,767), 15.5y ($$n = 2$$,930), 17.8y ($$n = 2$$,613), and 24.5y ($$n = 2$$,559) for the childhood, early adolescence, late adolescence and young adulthood time points respectively. The proportion of participants which were male were $50.5\%$, $47.4\%$, $44.5\%$, and $39.1\%$ and mean BMI was 16.2, 21.4, 22.7, and 24.8 kg/m2 for each time point respectively. The socio-demographic profile of ALSPAC offspring participants has been reported previously.69 Mean and standard deviation (SD) values for metabolites on each measurement occasion in ALSPAC are shown in Supplementary File 1b.
In the GRS analysis, there was no strong evidence of association of CRC liability with metabolites at age 8y (Supplementary File 1c). At age 16y, there was evidence for association with several lipid traits including higher cholesteryl esters to total lipids ratio in large low-density lipoprotein (LDL) (SD change per doubling CRC liability = 0.06, $95\%$ CI = 0.02 to 0.10) and higher cholesterol in very small very low-density lipoprotein (VLDL) (SD change per doubling CRC liability = 0.06, $95\%$ CI = 0.03 to 0.10). There was strong evidence for association with several traits at age 18y including higher non-high-density lipoprotein (non-HDL) lipids, e.g., a 1 doubling CRC liability was associated with higher levels of total cholesterol (SD change = 0.05 $95\%$ CI = 0.01 to 0.09), VLDL-cholesterol (SD change = 0.05, $95\%$ CI = 0.01 to 0.09), LDL-cholesterol (SD change = 0.06, $95\%$ CI = 0.02 to 0.09)), apolipoproteins (apolipoprotein B (SD change = 0.06, $95\%$ CI = 0.02 to 0.09)), and fatty acids (omega-3 (SD change = 0.08, $95\%$ CI = 0.04 to 0.11), docosahexaenoic acid (DHA) (SD change = 0.05, $95\%$ CI = 0.02 to 0.09)) (Supplementary File 1c). Figure 2(-figure supplements 1–6) shows results for all clinically validated metabolites. At age 25y, there was no strong evidence of association of CRC liability with metabolites. In anatomical site-specific analyses, there was strong evidence for association of liability to colon cancer with omega-3 (SD change = 0.07, $95\%$ CI = 0.03 to 0.11) and DHA (SD change = 0.07, $95\%$ CI = 0.03 to 0.10) at age 18y. There was little evidence for any associations at any other CRC site or age (Supplementary File 1c). When SNPs in the FADS cluster gene regions were excluded due to possible horizontal pleiotropy given the role of FADS in lipid metabolism, there was a reduction in strength of evidence for an association of liability to CRC with any metabolite measured, although estimates were in a largely consistent direction with the prior analysis (Supplementary File 1d).
## Reverse MR of the effects of CRC liability on circulating metabolites in middle adulthood
“Reverse” MR analyses54 were conducted using UK Biobank for outcome datasets in two sample MR to examine the effect of CRC liability on circulating metabolites. SNP-outcome (metabolite) estimates were obtained from a GWAS of metabolites in UK Biobank.57,58 Prior to GWAS, all metabolite measures were standardised and normalised using rank-based inverse normal transformation. Genetic association data for metabolites were retrieved using the MRC IEU UK Biobank GWAS pipeline.59 Full summary statistics are available via the IEU Open GWAS project.54,60 Up to 3 statistical methods were used to generate reverse MR estimates of the effect of CRC liability on circulating metabolites using the TwoSampleMR package61: random-effects inverse variance weighted (IVW), weighted-median, and weighted-mode, which each make differing assumptions about directional pleiotropy and SNP heterogeneity.62–64 The IVW MR model will produce biased effect estimates in the presence of horizontal pleiotropy, i.e. where one or more genetic variant(s) included in the instrument affect the outcome by a pathway other than through the exposure. In the weighted median model, each genetic variant is weighted according to its distance from the median effect of all genetic variants. Thus, the weighted median model will provide an unbiased estimate when at least $50\%$ of the information in an instrument comes from genetic variants that are not horizontally pleiotropic. The weighted mode model uses a similar approach but weights genetic instruments according to the mean effect. In this model, over $50\%$ of the weight of the genetic instrument can be contributed to by genetic variants which are horizontally pleiotropic, but the most common amount of pleiotropy must be zero (known as the Zero Modal Pleiotropy Assumption (ZEMPA))65. As above, estimates were multiplied by 0.693 (loge2) to reflect SD-unit differences in metabolites per doubling of genetic liability to CRC.66 All instrument sets from the reverse MR analysis had an F-statistic greater than 10 (minimum F-statistic = 36, median = 40), suggesting our analyses did not suffer from weak instrument bias (Supplementary File 1e). There was little evidence of an association of CRC liability (overall or by anatomical site) on any of the circulating metabolites investigated, including when the SNP in the FADS gene region was excluded, based on our pre-determined cut-off of FDR-$P \leq 0.05$; however, the direction of effect estimates was largely consistent with those seen in ALSPAC GRS analyses, with higher CRC liability weakly associated with higher non-HDLs, lipoproteins and fatty acid levels (Supplementary File 1f–g). Figure 3(-figure supplements 1–3) shows the results for clinically validated metabolites. In subsite stratified analyses, there was strong evidence for a causal effect of genetic liability to proximal colon cancer on several traits, including total fatty acids (SD change per doubling of liability = 0.02, $95\%$ CI = 0.01 to 0.04) and omega-6 fatty acids (SD change per doubling of liability = 0.03, $95\%$ CI = 0.01 to 0.05).
## Forward MR of the effects of metabolites on CRC
Forward MR analyses were conducted using summary statistics from UK Biobank for the same NMR-measured metabolites (SNP-exposure) and from GECCO/CORECT/CCFR as outlined above (SNP-outcome). We identified SNPs that were independently associated (R2<0.001 and $P \leq 5$×10−8) with metabolites from a GWAS of 249 metabolites in UK Biobank described above. As before, we used up to 3 statistical methods to generate MR estimates of the effect of circulating metabolites on CRC risk (overall and site-specific): random-effects IVW, weighted-median, and weighted-mode. The Benjamini-Hochberg method was used to adjust P-values for multiple testing and an adjusted P-value of <0.05 was used as a heuristic for nominal evidence for a causal effect.56 MR outputs are beta coefficients representing the logOR for CRC per SD higher metabolite, exponentiated to reflect the OR for CRC per SD metabolite.
MR analyses were performed in R version 4.0.3.67 and GRS analyses in Stata 16.1 (StataCorp, College Station, Texas, USA). The ggforestplot R package was used to generate results visualisations.68
## Ethics
Written informed consent was obtained for all study participants. Ethical approval was obtained from the ALSPAC Law and Ethics Committee and the local research ethics committee (proposal B3538). Consent for biological samples has been collected in accordance with the Human Tissue Act [2004]. Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time. Ethics for the CRC GWAS were approved by respective institutional review boards.
## Forward MR for the effects of metabolites on CRC risk
All instrument sets from the forward MR analysis had an F-statistic greater than 10 (minimum F-statistic = 54, median = 141), suggesting that our analyses were unlikely to suffer from weak instrument bias (Supplementary File 1h–i). There was strong evidence for an effect of several fatty acid traits on overall CRC risk, including of omega-3 fatty acids (CRC OR = 1.13, $95\%$ CI = 1.06 to 1.21), DHA (OR CRC = 1.76, $95\%$ CI = 1.08 to 1.28), ratio of omega-3 fatty acids to total fatty acids (OR CRC = 1.18, $95\%$ CI = 1.11 to 1.25), ratio of DHA to total fatty acids (CRC OR = 1.20, $95\%$ CI = 1.10 to 1.31), and ratio of omega-6 fatty acids to omega-3 fatty acids (CRC OR = 0.86, $95\%$ CI = 0.80 to 9.13) (Supplementary File 1j, Figure 4-figure supplements 1–3). These estimates were overlapping with variable precision in MR sensitivity models. When SNPs in the FADS gene region were excluded, there was little evidence for a causal effect of any metabolite investigated on CRC risk based on the predetermined FDR-P cut of off < 0.05, although the directions of effect estimates were consistent with previous analyses (Supplementary File 1k).
In anatomical subtype stratified analyses evidence was strongest for an effect of fatty acid traits on higher CRC risk, and this appeared specific to the distal colon, e.g., omega-3 (distal CRC OR = 1.20, $95\%$ CI = 1.09 to 1.32), and ratio of DHA to total fatty acids (distal colon OR = 1.29, $95\%$ CI = 1.16 to 1.43). There was also evidence of a negative effect of ratio of omega-6 to omega-3 fatty acids (distal CRC OR = 0.80, $95\%$ CI = 0.74 to 0.88) and a positive effect of ratio of omega-3 fatty acids to total fatty acids (distal CRC = 1.24, $95\%$ CI = 1.15 to 1.35; seen also for proximal CRC OR = 1.15, $95\%$ CI = 1.07 to 1.23) (Supplementary File 1j). These estimates were also directionally consistent in MR sensitivity models.
## Discussion
Here, we used a reverse MR framework to identify circulating metabolites which are associated with genetic CRC liability across different stages of the early life course and attempted to replicate results in an independent cohort of middle-aged adults. We then performed forward MR to characterise the causal direction of the relationship between metabolites and CRC. Our GRS analysis provided evidence for an association of genetic liability to CRC with higher circulating levels of lipoprotein lipids (including total cholesterol, VLDL-cholesterol, and LDL-cholesterol), apolipoproteins (including apolipoprotein B), and fatty acids (including omega-3 and DHA) in young adults. These results were largely consistent in direction (though smaller in magnitude and weaker in strength of evidence) in a two-sample MR analysis in an independent cohort of middle-aged adults. Results were attenuated, but consistent in direction, when potentially pleiotropic SNPs in the FADS gene regions were excluded. However, it should be noted that use of a narrow window for exclusion based on being within one of the three FADS genes may mean that some pleiotropic SNPs remain. Our subsequent forward MR analysis highlighted polyunsaturated fatty acids as potentially having a causal role in the development of CRC.
Our analyses highlight a potentially important role of polyunsaturated fatty acids in colorectal cancer liability. However, these analyses may be biased by substantial genetic pleiotropy among fatty acid traits. SNPs which are associated with levels of one fatty acid are generally associated with levels of many more fatty acid (and non-fatty acid) traits.70,71 For instance, genetic instruments within the FADS cluster of genes will likely affect both omega-3 and omega-6 fatty acids, given FADS1 and FADS2 encode enzymes which catalyse the conversion of both from shorter chain into longer chain fatty acids.71 In addition, the NMR metabolomics platform utilised in the analyses outlined here has limited coverage of fatty acids, meaning many putative causal metabolites for CRC, for example arachidonic acid, could not be investigated. Therefore, although our results indicate that polyunsaturated fatty acids may be important in colorectal cancer risk, given the pleiotropic nature of the fatty acid genetic instruments and the limited coverage of the NMR platform, we are unable to determine with any certainty which specific classes of fatty acids may be driving these associations.
Our analyses featured evaluating the effect of genetic liability to CRC on circulating metabolites across repeated measures in the ALSPAC cohort. The mean ages at the time of the repeated measures were 8y, 16y, 18y, and 25y, representing childhood, early adolescence, late adolescence, and young adulthood respectively, and therefore individuals in this cohort are unlikely to be taking metabolite-altering medication such as statins, and unlikely to have CRC. The strongest evidence for an effect of liability to CRC on metabolite levels was seen in late adolescence. The reason for this remains unclear. It is possible that this represents a true biological phenomenon if late adolescence is a critical window in CRC development or metabolite variability, which may be likely given the limited variance in metabolite levels at the later age of 25y (Supplementary File 1b). The lack of an effect at the younger ages could be explained by the fact that the CRC GRS may capture many key life events or experiences which could impact the metabolome (e.g., initiation of smoking, higher category of BMI reached, educational attainment level set, etc) but may not have yet happened at younger ages, thus obscuring an effect of genetic liability to CRC on the metabolome. Our results suggest that puberty could be important, with an effect seen seemingly particularly at the end of puberty. Repeating our analysis with sex-stratified data may aid in determining whether this is likely to be the case; sex-stratified GWAS for metabolites are not currently available to replicate such analyses. An alternative explanation is selection bias due to loss of follow-up, leading to a change in sample characteristics over time.
Another key finding in the reverse MR analysis was that genetic liability to CRC was associated with increased levels of total cholesterol, VLDL-cholesterol, LDL-cholesterol, and apolipoprotein B, though we find little evidence for a causal effect of these traits on risk of CRC in the forward MR, replicating previous forward MR analyses for total and LDL-cholesterol.9,72–74 This suggests that these traits may either be only predictive of (i.e., non-causal for) later CRC development, or may be influenced by the development of CRC and could have diagnostic or predictive potential. Given that the participants in the ALSPAC cohort are many decades younger than the average age of diagnosis for CRC (mean age 25 years in the latest repeated measure analysed in ALSPAC; whereas the median age at diagnosis of CRC is 64 years),75 the former seems the most likely scenario. Previous conventional observational studies have presented conflicting results when investigating the association between measures of cholesterol and CRC risk with some finding an inverse association and others a positive association, possibly reflecting residual confounding in conventional observational analyses.76–82 Previous MR studies have had similar findings to our forward MR analysis, in that there seems to be little evidence for a causal effect of cholesterol on CRC development.72–74 One possible explanation for how circulating levels of total cholesterol, VLDL-cholesterol, LDL-cholesterol and apolipoprotein B could predict (without necessarily causing) future CRC development could be linked to diet. A previous MR analysis suggested an effect of increased BMI on several measures of circulating cholesterol.9 *Consuming a* diet which is high in fat may increase CRC risk both through and possibly independently of adiposity, alongside increasing levels of circulating cholesterol.83–88 The potential for lipoprotein or apolipoprotein lipid measures in future CRC risk prediction should be further investigated.
Our analyses stratified by anatomical subsite highlighted fatty acids as being affected by genetic liability to colon and proximal colon cancer, with the forward MR confirming that fatty acid traits may be particularly important in the development of these subsites of CRC as well as distal colon cancer.
In our forward MR analyses we were unable to replicate the findings of three previous MR studies which found evidence for a causal effect of circulating linoleic acid levels on CRC development in terms of strength of evidence, though the direction of the effect estimate was similar to previous studies.89–91 *This is* surprising as all three previous analyses had a much smaller sample size than that included in our analysis (the largest had sample size of 24,748 for exposure vs 118,466 presently; and 11,016 cases and 13,732 controls for outcome vs 52,775 cases and 45,940 controls presently). Our analysis using updated genetic instruments to proxy fatty acids may be more successful in accurately instrumenting heterogenous phenotypes such as metabolite levels compared with previous analyses. All other findings in our forward MR analysis are consistent with previous MR studies where they exist.72–74
## Limitations
The limitations of this study include firstly the relatively small sample size included in the ALSPAC analysis, which may have implications for power and precision. Secondly, mostly due to the longitudinal nature of the ASLAPC study, our sample at each time point is composed of slightly different individuals. This could be influencing our results, and should be taken into account when comparing across time points. Thirdly, our analyses involving genetic instruments for CRC liability may have suffered from horizontal pleiotropy, even after excluding genetic variants in or near the FADS gene. Fourthly, our analyses were mostly restricted to white Europeans, which limits the generalisability of our findings to other populations. Fifthly, our analysis would benefit from being repeated with sex-stratified data. Although such GWAS results for metabolites are not currently available, the data to perform such GWAS are available in UK Biobank for future analyses. Sixthly, for our forward MR analysis, we used the UK Biobank for our exposure data. The UK Biobank has a median age of 58 at the time these measurements were taken, meaning statin use may be widespread in this population, which could be attenuating our effect estimates. Future work could attempt to replicate our analysis in a population with lower prevalence of statins intake. Finally, we included only metabolites measured using NMR. Confirming whether our results replicate using metabolite data measured with an alternative method would strengthen our findings.
## Conclusions
Our analysis provides evidence that genetic liability to CRC is associated with altered levels of metabolites at certain ages, some of which may have a causal role in CRC development. Further investigating the role of polyunsaturated fatty acids in CRC risk and circulating cholesterol in CRC prediction may be promising avenues for future research.
## Author funding
JAB is supported by the Elizabeth Blackwell Institute for Health Research, University of Bristol and the Wellcome Trust Institutional Strategic Support Fund (204813/Z/16/Z) and works in a Unit funded by the Medical Research Council (MC_UU_$\frac{00011}{1}$) and the University of Bristol. EEV, DNL and CB are supported by Diabetes UK ($\frac{17}{0005587}$). NJT is a Wellcome Trust Investigator (202802/Z/16/Z), is the PI of the Avon Longitudinal Study of Parents and Children (MRC & WT $\frac{102215}{2}$/$\frac{13}{2}$), is supported by the University of Bristol NIHR Biomedical Research Centre (BRC-1215-20011), the MRC Integrative Epidemiology Unit (MC_UU_$\frac{00011}{1}$) and works within the CRUK Integrative Cancer Epidemiology Programme (C18281/A19169). EH is supported by a Cancer Research UK Population Research Committee Studentship (C18281/A30905), is supported by the CRUK Integrative Cancer Epidemiology Programme (C18281/A29019) and is part of the Medical Research Council Integrative Epidemiology Unit at the University of Bristol which is supported by the Medical Research Council (MC_UU_$\frac{00011}{4}$) and the University of Bristol. AC acknowledges funding from grant MR/N$\frac{0137941}{1}$ for the GW4 BIOMED MRC DTP, awarded to the Universities of Bath, Bristol, Cardiff and Exeter from the Medical Research Council (MRC)/UKRI. MCB was supported by the UK Medical Research Council (MRC) Skills Development Fellowship (MR/P$\frac{014054}{1}$), University of Bristol Vice-Chancellor’s Fellowship and MRC Integrative Epidemiology Unit (MC_UU_$\frac{00011}{6}$). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. This work used the computational facilities of the Advanced Computing Research Centre, University of Bristol - http://www.bristol.ac.uk/acrc/.
## Study funding
ALSPAC: The UK Medical Research Council and Wellcome (Grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors who will serve as guarantors for the contents of this paper. A comprehensive list of grants funding is available on the ALSPAC website (http://www.bristol.ac.uk/alspac/external/documents/grant-acknowledgements.pdf); This research was specifically funded by The UK Medical Research Council (Grant ref: MC_UU_$\frac{12013}{1}$). GWAS data was generated by Sample Logistics and Genotyping Facilities at Wellcome Sanger Institute and LabCorp (Laboratory Corporation of America) using support from 23andMe.
Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO): National Cancer Institute, National Institutes of Health, U.S. Department of Health and Human Services (U01 CA137088, R01 CA059045, R01 201407). Genotyping/Sequencing services were provided by the Center for Inherited Disease Research (CIDR) contract number HHSN268201700006I and HHSN268201200008I. This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA015704. Scientific Computing Infrastructure at Fred Hutch funded by ORIP grant S10OD028685.
ASTERISK: a Hospital Clinical Research Program (PHRC-BRD09/C) from the University Hospital Center of Nantes (CHU de Nantes) and supported by the Regional Council of Pays de la Loire, the Groupement des Entreprises Françaises dans la Lutte contre le Cancer (GEFLUC), the Association Anne de Bretagne Génétique and the Ligue Régionale Contre le Cancer (LRCC).
The ATBC *Study is* supported by the Intramural Research Program of the U.S. National Cancer Institute, National Institutes of Health, Department of Health and Human Services.
CLUE II funding was from the National Cancer Institute (U01 CA086308, Early Detection Research Network; P30 CA006973), National Institute on Aging (U01 AG018033), and the American Institute for Cancer Research. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government.
Maryland Cancer Registry (MCR): *Cancer data* was provided by the Maryland Cancer Registry, Center for Cancer Prevention and Control, Maryland Department of Health, with funding from the State of Maryland and the Maryland Cigarette Restitution Fund. The collection and availability of cancer registry data is also supported by the Cooperative Agreement NU58DP006333, funded by the Centers for Disease Control and Prevention. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the Centers for Disease Control and Prevention or the Department of Health and Human Services.
ColoCare: This work was supported by the National Institutes of Health (grant numbers R01 CA189184 (Li/Ulrich), U01 CA206110 (Ulrich/Li/Siegel/Figueiredo/Colditz, 2P30CA015704– 40 (Gilliland), R01 CA207371 (Ulrich/Li)), the Matthias Lackas-Foundation, the German Consortium for Translational Cancer Research, and the EU TRANSCAN initiative.
The Colon Cancer Family Registry (CCFR, www.coloncfr.org) is supported in part by funding from the National Cancer Institute (NCI), National Institutes of Health (NIH) (award U01 CA167551). Support for case ascertainment was provided in part from the Surveillance, Epidemiology, and End Results (SEER) Program and the following U.S. state cancer registries: AZ, CO, MN, NC, NH; and by the Victoria Cancer Registry (Australia) and Ontario Cancer Registry (Canada). The CCFR Set-1 (Illumina 1M/1M-Duo) and Set-2 (Illumina Omni1-Quad) scans were supported by NIH awards U01 CA122839 and R01 CA143237 (to GC). The CCFR Set-3 (Affymetrix Axiom CORECT Set array) was supported by NIH award U19 CA148107 and R01 CA81488 (to SBG). The CCFR Set-4 (Illumina OncoArray 600K SNP array) was supported by NIH award U19 CA148107 (to SBG) and by the Center for Inherited Disease Research (CIDR), which is funded by the NIH to the Johns Hopkins University, contract number HHSN268201200008I. Additional funding for the OFCCR/ARCTIC was through award GL201–043 from the Ontario Research Fund (to BWZ), award 112746 from the Canadian Institutes of Health Research (to TJH), through a Cancer Risk Evaluation (CaRE) Program grant from the Canadian Cancer Society (to SG), and through generous support from the Ontario Ministry of Research and Innovation. The SFCCR Illumina HumanCytoSNP array was supported in part through NCI/NIH awards U01/U24 CA074794 and R01 CA076366 (to PAN). The content of this manuscript does not necessarily reflect the views or policies of the NCI, NIH or any of the collaborating centers in the Colon Cancer Family Registry (CCFR), nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government, any cancer registry, or the CCFR.
COLON: The COLON study is sponsored by Wereld Kanker Onderzoek Fonds, including funds from grant $\frac{2014}{1179}$ as part of the World Cancer Research Fund International Regular Grant Programme, by Alpe d’Huzes and the Dutch Cancer Society (UM 2012–5653, UW 2013–5927, UW2015–7946), and by TRANSCAN (JTC2012-MetaboCCC, JTC2013-FOCUS). The Nqplus study is sponsored by a ZonMW investment grant (98–10030); by PREVIEW, the project PREVention of diabetes through lifestyle intervention and population studies in Europe and around the World (PREVIEW) project which received funding from the European Union Seventh Framework Programme (FP$\frac{7}{2007}$–2013) under grant no. 312057; by funds from TI Food and Nutrition (cardiovascular health theme), a public–private partnership on precompetitive research in food and nutrition; and by FOODBALL, the Food Biomarker Alliance, a project from JPI Healthy Diet for a Healthy Life.
COLO2&3: National Institutes of Health (R01 CA060987) Colorectal Cancer Transdisciplinary (CORECT) Study: The CORECT Study was supported by the National Cancer Institute, National Institutes of Health (NCI/NIH), U.S. Department of Health and Human Services (grant numbers U19 CA148107, R01 CA081488, P30 CA014089, R01 CA197350; P01 CA196569; R01 CA201407; R01 CA242218), National Institutes of Environmental Health Sciences, National Institutes of Health (grant number T32 ES013678) and a generous gift from Daniel and Maryann Fong.
CORSA: The CORSA study was funded by Austrian Research Funding Agency (FFG) BRIDGE (grant 829675, to Andrea Gsur), the “Herzfelder’sche Familienstiftung” (grant to Andrea Gsur) and was supported by COST Action BM1206.
CPS-II: The American Cancer Society funds the creation, maintenance, and updating of the Cancer Prevention Study-II (CPS-II) cohort. The study protocol was approved by the institutional review boards of Emory University, and those of participating registries as required.
CRCGEN: Colorectal Cancer Genetics & Genomics, Spanish study was supported by Instituto de Salud Carlos III, co-funded by FEDER funds –a way to build Europe– (grants PI14–613 and PI09–1286), Agency for Management of University and Research Grants (AGAUR) of the Catalan Government (grant 2017SGR723), Junta de Castilla y León (grant LE22A10–2), the Spanish Association Against Cancer (AECC) Scientific Foundation grant GCTRA18022MORE and the Consortium for Biomedical Research in Epidemiology and Public Health (CIBERESP), action Genrisk. Sample collection of this work was supported by the Xarxa de Bancs de Tumors de Catalunya sponsored by Pla Director d’Oncología de Catalunya (XBTC), Plataforma Biobancos PT$\frac{13}{0010}$/0013 and ICOBIOBANC, sponsored by the Catalan Institute of Oncology. We thank CERCA Programme, Generalitat de Catalunya for institutional support.
Czech Republic CCS: This work was supported by the Grant Agency of the Czech Republic (21–04607X, 20–03997S), by the Grant Agency of the Ministry of Health of the Czech Republic (grants AZV NU21–07-00247 and AZV NU21–03-00506), and Charles University Research Fund (Cooperation 43-Surgical disciplines) DACHS: This work was supported by the German Research Council (BR $\frac{1704}{6}$–1, BR $\frac{1704}{6}$–3, BR $\frac{1704}{6}$–4, CH $\frac{117}{1}$–1, HO $\frac{5117}{2}$–1, HE $\frac{5998}{2}$–1, KL $\frac{2354}{3}$–1, RO $\frac{2270}{8}$–1 and BR $\frac{1704}{17}$–1), the Interdisciplinary Research Program of the National Center for Tumor Diseases (NCT), Germany, and the German Federal Ministry of Education and Research (01KH0404, 01ER0814, 01ER0815, 01ER1505A and 01ER1505B).
DALS: National Institutes of Health (R01 CA048998 to M. L. Slattery).
EDRN: This work is funded and supported by the NCI, EDRN Grant (U01-CA152753).
EPIC: The coordination of EPIC is financially supported by International Agency for Research on Cancer (IARC) and also by the Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London which has additional infrastructure support provided by the NIHR Imperial Biomedical Research Centre (BRC). The national cohorts are supported by: Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l’Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM) (France); German Cancer Aid, German Cancer Research Center (DKFZ), German Institute of Human Nutrition Potsdam- Rehbruecke (DIfE), Federal Ministry of Education and Research (BMBF) (Germany); Associazione Italiana per la Ricerca sul Cancro-AIRC-Italy, Compagnia di SanPaolo and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund (WCRF), Statistics Netherlands (The Netherlands); Health Research Fund (FIS) - Instituto de Salud Carlos III (ISCIII), Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, and the Catalan Institute of Oncology - ICO (Spain); Swedish Cancer Society, Swedish Research Council and and Region Skåne and Region Västerbotten (Sweden); Cancer Research UK (14136 to EPIC-Norfolk; C8221/A29017 to EPIC-Oxford), Medical Research Council (1000143 to EPIC-Norfolk; MR/M$\frac{012190}{1}$ to EPIC-Oxford). ( United Kingdom).
EPICOLON: This work was supported by grants from Fondo de Investigación Sanitaria/FEDER (PI$\frac{08}{0024}$, PI$\frac{08}{1276}$, PS$\frac{09}{02368}$, PI$\frac{11}{00219}$, PI$\frac{11}{00681}$, PI$\frac{14}{00173}$, PI$\frac{14}{00230}$, PI$\frac{17}{00509}$, $\frac{17}{00878}$, PI$\frac{20}{00113}$, PI$\frac{20}{00226}$, Acción Transversal de Cáncer), Xunta de Galicia (PGIDIT07PXIB9101209PR), Ministerio de Economia y Competitividad (SAF07–64873, SAF 2010–19273, SAF2014–54453R), Fundación Científica de la Asociación Española contra el Cáncer (GCB13131592CAST, PRYGN211085CAST), Beca Grupo de Trabajo “Oncología” AEG (Asociación Española de Gastroenterología), Fundación Privada Olga Torres, FP7 CHIBCHA Consortium, Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR, Generalitat de Catalunya, 2014SGR135, 2014SGR255, 2017SGR21, 2017SGR653, 2021SGR00716, 2021SGR01185), Catalan Tumour Bank Network (Pla Director d’Oncologia, Generalitat de Catalunya), PERIS (SLT$\frac{002}{16}$/00398, Generalitat de Catalunya), Marató TV3 (202008–10), CERCA Programme (Generalitat de Catalunya) and COST Actions BM1206 and CA17118. CIBERehd is funded by the Instituto de Salud Carlos III.ESTHER/VERDI. This work was supported by grants from the Baden-Württemberg Ministry of Science, Research and Arts and the German Cancer Aid.
Harvard cohorts: HPFS is supported by the National Institutes of Health (P01 CA055075, UM1 CA167552, U01 CA167552, R01 CA137178, R01 CA151993, and R35 CA197735), NHS by the National Institutes of Health (P01 CA087969, UM1 CA186107, R01 CA137178, R01 CA151993, and R35 CA197735), and PHS by the National Institutes of Health (R01 CA042182).
Hawaii Adenoma Study: NCI grants R01 CA072520.
HCES-CRC: the Hwasun Cancer Epidemiology Study–Colon and Rectum Cancer (HCES-CRC; grants from Chonnam National University Hwasun Hospital, HCRI15011–1).
Kentucky: This work was supported by the following grant support: Clinical Investigator Award from Damon Runyon Cancer Research Foundation (CI-8); NCI R01CA136726.
LCCS: The Leeds Colorectal Cancer Study was funded by the Food Standards Agency and Cancer Research UK Programme Award (C588/A19167).
MCCS cohort recruitment was funded by VicHealth and Cancer Council Victoria. The MCCS was further supported by Australian NHMRC grants 509348, 209057, 251553 and 504711 and by infrastructure provided by Cancer Council Victoria. Cases and their vital status were ascertained through the Victorian Cancer Registry (VCR) and the Australian Institute of Health and Welfare (AIHW), including the National Death Index and the Australian Cancer Database. BMLynch was supported by MCRF18005 from the Victorian Cancer Agency.
MEC: National Institutes of Health (R37 CA054281, P01 CA033619, and R01 CA063464).
MECC: This work was supported by the National Institutes of Health, U.S. Department of Health and Human Services (R01 CA081488, R01 CA197350, U19 CA148107, R01 CA242218, and a generous gift from Daniel and Maryann Fong.
MSKCC: The work at Sloan Kettering in New York was supported by the Robert and Kate Niehaus Center for Inherited Cancer Genomics and the Romeo Milio Foundation. Moffitt: This work was supported by funding from the National Institutes of Health (grant numbers R01 CA189184, P30 CA076292), Florida Department of Health Bankhead-Coley Grant 09BN-13, and the University of South Florida Oehler Foundation. Moffitt contributions were supported in part by the Total Cancer Care Initiative, Collaborative Data Services Core, and Tissue Core at the H. Lee Moffitt Cancer Center & Research Institute, a National Cancer Institute-designated Comprehensive Cancer Center (grant number P30 CA076292).
NCCCS I & II: We acknowledge funding support for this project from the National Institutes of Health, R01 CA066635 and P30 DK034987.
NFCCR: This work was supported by an Interdisciplinary Health Research Team award from the Canadian Institutes of Health Research (CRT 43821); the National Institutes of Health, U.S. Department of Health and Human Serivces (U01 CA074783); and National Cancer Institute of Canada grants (18223 and 18226). The authors wish to acknowledge the contribution of Alexandre Belisle and the genotyping team of the McGill University and Génome Québec Innovation Centre, Montréal, Canada, for genotyping the Sequenom panel in the NFCCR samples. Funding was provided to Michael O. Woods by the Canadian Cancer Society Research Institute.
NSHDS: The research was supported by Biobank Sweden through funding from the Swedish Research Council (VR 2017–00650, VR 2017–01737), the Swedish Cancer Society (CAN $\frac{2017}{581}$), Region Västerbotten (VLL-841671, VLL-833291), Knut and Alice Wallenberg Foundation (VLL-765961), and the Lion’s Cancer Research Foundation (several grants) and Insamlingsstiftelsen, both at Umeå University.
OSUMC: OCCPI funding was provided by Pelotonia and HNPCC funding was provided by the NCI (CA016058 and CA067941).
PLCO: Intramural Research Program of the Division of Cancer Epidemiology and Genetics and supported by contracts from the Division of Cancer Prevention, National Cancer Institute, NIH, DHHS. Funding was provided by National Institutes of Health (NIH), Genes, Environment and Health Initiative (GEI) Z01 CP 010200, NIH U01 HG004446, and NIH GEI U01 HG 004438.
SEARCH: The University of Cambridge has received salary support in respect of PDPP from the NHS in the East of England through the Clinical Academic Reserve. Cancer Research UK (C490/A16561); the UK National Institute for Health Research Biomedical Research Centres at the University of Cambridge.
SELECT: Research reported in this publication was supported in part by the National Cancer Institute of the National Institutes of Health under Award Numbers U10 CA037429 (CD Blanke), and UM1 CA182883 (CM Tangen/IM Thompson). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
SMS and REACHS: This work was supported by the National Cancer Institute (grant P01 CA074184 to J.D.P. and P.A.N., grants R01 CA097325, R03 CA153323, and K05 CA152715 to P.A.N., and the National Center for Advancing Translational Sciences at the National Institutes of Health (grant KL2 TR000421 to A.N.B.-H.) The Swedish Low-risk Colorectal Cancer Study: The study was supported by grants from the Swedish research council; K2015–55X-22674–01-4, K2008–55X-20157–03-3, K2006–72X-20157–01-2 and the Stockholm County Council (ALF project).
Swedish Mammography Cohort and Cohort of Swedish Men: This work is supported by the Swedish Research Council /Infrastructure grant, the Swedish Cancer Foundation, and the Karolinska Institutés Distinguished Professor Award to Alicja Wolk.
UK Biobank: This research has been conducted using the UK Biobank Resource under Application Number 8614 VITAL: National Institutes of Health (K05 CA154337).
The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts 75N92021D00001, 75N92021D00002, 75N92021D00003, 75N92021D00004, 75N92021D00005
## Data availability:
Individual-level ALSPAC data are available following an application. This process of managed access is detailed at www.bristol.ac.uk/alspac/researchers/access. Cohort details and data descriptions for ALSPAC are publicly available at the same web address. Summary-level GWAS data used in this study from UK Biobank are publicly available without the need for application through the MR-Base platform, which is accessible at http://www.mrbase.org/. The summary-level GWAS data for CRC used in this study are available following an application to GECCO (managed access). All data generated by this study are available in the manuscript and supporting material. R scripts used in this study have been made publicly available on GitHub at: https://github.com/cb12104/adiposity_metabolites_crc.
## References
1. Ferlay J, Soerjomataram I, Dikshit R. *Int J Cancer [Internet]* (2015.0) **136** E359-E386. DOI: 10.1002/ijc.29210
2. 2.World Cancer Research Fund/American Institute for Cancer Research. Continuous Update Project Expert Report. Diet, nutrition, physical activity and colorectal cancer. 2018;. *Continuous Update Project Expert Report. Diet, nutrition, physical activity and colorectal cancer* (2018.0)
3. Huyghe JR, Bien SA, Harrison TA. *Nature Genetics 2018 51:1 [Internet]* (2018.0) **51** 76-87
4. Czene K, Lichtenstein P, Hemminki K. *Int J Cancer [Internet]* (2002.0) **99** 260-266. DOI: 10.1002/ijc.10332
5. Aul P, Ichtenstein L, Olm IVH. *Environmental and Heritable Factors in the Causation of Cancer — Analyses of Cohorts of Twins from Sweden, Denmark, and Finland* (2000.0) **343** 78-85. DOI: 10.1056/NEJM200007133430201
6. Händel MN, Rohde JF, Jacobsen R. *European Journal of Clinical Nutrition 2020 74:8 [Internet]* (2020.0) **74** 1132-1148
7. McNabb S, Harrison TA, Albanes D. *Int J Cancer [Internet]* (2020.0) **146** 861-873. DOI: 10.1002/ijc.32377
8. Lauby-Secretan B, Scoccianti C, Loomis D, Grosse Y, Bianchini F, Straif K. **Body Fatness and Cancer--Viewpoint of the IARC Working Group**. *N Engl J Med [Internet]. N Engl J Med* (2016.0) **375** 794-798. PMID: 27557308
9. Bull CJ, Bell JA, Murphy N. *BMC Med [Internet]* (2020.0) **18** 1-16. PMID: 31898501
10. Ladabaum U, Dominitz JA, Kahi C, Schoen RE. *Gastroenterology* (2020.0) **158** 418-432. PMID: 31394083
11. Cardoso R, Guo F, Heisser T. *Lancet Oncol* (2021.0) **22** 1002-1013. PMID: 34048685
12. Qiu Y, Cai G, Su M. *J Proteome Res [Internet]* (2009.0) **8** 4844-4850. DOI: 10.1021/pr9004162
13. Ward PS, Thompson CB. **Metabolic reprogramming: a cancer hallmark even warburg did not anticipate**. *Cancer Cell [Internet]. Cancer Cell* (2012.0) **21** 297-308. PMID: 22439925
14. Rattray NJW, Charkoftaki G, Rattray Z, Hansen JE, Vasiliou V, Johnson CH. *Current Pharmacology Reports 2017 3:3 [Internet]* (2017.0) **3** 114-125
15. Singla P, Bardoloi A, Parkash AA. *World J Diabetes [Internet]* (2010.0) **1** 76. PMID: 21537431
16. Papandreou C, García-Gavilán J, Camacho-Barcia L. *Metabolites 2021, Vol 11, Page 317 [Internet]* (2021.0) **11** 317
17. Ahmad S, Hammar U, Kennedy B. *Diabetes [Internet]* (2022.0) **71** 329-339. PMID: 34785567
18. Haycock PC, Carolina Borges M, Burrows K. *The association between genetically elevated polyunsaturated fatty acids and risk of cancer* (2023.0) **91**
19. Yarmolinsky J, Wade KH, Richmond RC. *Causal Inference in Cancer Epidemiology: What Is the Role of Mendelian Randomization?*
20. GD S S E. **‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?**. *Int J Epidemiol [Internet]. Int J Epidemiol* (2003.0) **32**
21. Holmes M v, Smith GD. *Clin Chem [Internet]* (2019.0) **65** 363-366. PMID: 30692117
22. Gao R, Wu C, Zhu Y. *Gastroenterology* (2022.0) **163** 1024-1037.e9. PMID: 35788345
23. Rattner JI, Kopciuk KA, Vogel HJ. *Oncotarget [Internet]* (2022.0) **13** 61. PMID: 35028011
24. Leichtle AB, Nuoffer JM, Ceglarek U. *Metabolomics [Internet]* (2012.0) **8** 643-653. PMID: 22833708
25. Ritchie SA, Ahiahonu PWK, Jayasinghe D. **Reduced levels of hydroxylated, polyunsaturated ultra long-chain fatty acids in the serum of colorectal cancer patients: Implications for early screening and detection**. *BMC Med [Internet]. BioMed Central* (2010.0) **8** 1-20
26. Nishiumi S, Kobayashi T, Ikeda A. *PLoS One [Internet]* (2012.0) **7** e40459. PMID: 22792336
27. Ma Y, Zhang P, Wang F, Liu W, Yang J, Qin H. **An integrated proteomics and metabolomics approach for defining oncofetal biomarkers in the colorectal cancer**. *Ann Surg [Internet]* (2012.0) **255** 720-730. PMID: 22395091
28. Bertini I, Cacciatore S, Jensen B v. *Cancer Res [Internet]* (2012.0) **72** 356-364. PMID: 22080567
29. Zhang A, Sun H, Yan G, Wang P, Han Y, Wang X. **Metabolomics in diagnosis and biomarker discovery of colorectal cancer**. *Cancer Lett. Elsevier* (2014.0) **345** 17-20
30. Farshidfar F, Weljie AM, Kopciuk KA. *British Journal of Cancer 2016 115:7 [Internet]* (2016.0) **115** 848-857
31. Farshidfar F, Weljie AM, Kopciuk K. **Serum metabolomic profile as a means to distinguish stage of colorectal cancer**. *Genome Med [Internet]. BioMed Central* (2012.0) **4** 1-13
32. Boyd A, Golding J, Macleod J. **Cohort Profile: The ‘Children of the 90s’—the index offspring of the Avon Longitudinal Study of Parents and Children**. *Int J Epidemiol* (2013.0) **42** 111-127. PMID: 22507743
33. Fraser A, Macdonald-Wallis C, Tilling K. **Cohort Profile: The Avon Longitudinal Study of Parents and Children: ALSPAC mothers cohort**. *Int J Epidemiol* (2013.0) **42** 97-110. PMID: 22507742
34. Northstone K, Lewcock M, Groom A. *Wellcome Open Res [Internet]* (2019.0) **4**
35. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. *J Biomed Inform [Internet]* (2009.0) **42** 377. PMID: 18929686
36. Sudlow C, Gallacher J, Allen N. **UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age**. *Public Library of Science* (2015.0) **12** e1001779. DOI: 10.1371/journal.pmed.1001779
37. Bycroft C, Freeman C, Petkova D. *Nature 2018 562:7726 [Internet]* (2018.0) **562** 203-209
38. Huyghe JR, Bien SA, Harrison TA. **Discovery of common and rare genetic risk variants for colorectal cancer**. *Nat Genet* (2019.0) **51** 76-87. PMID: 30510241
39. Huyghe JR, Harrison TA, Bien SA. *Gut [Internet]* (2021.0) **70** 1325-1334. PMID: 33632709
40. Burgess S, Davies NM, Thompson SG. **Bias due to participant overlap in two-sample Mendelian randomization**. *Genet Epidemiol* (2016.0) **40** 597-608. PMID: 27625185
41. Law PJ, Timofeeva M, Fernandez-Rozadilla C. *Association analyses identify 31 new risk loci for colorectal cancer susceptibility*
42. Hemani G, Zheng J, Elsworth B. **The MR-Base platform supports systematic causal inference across the human phenome**. *Elife* (2018.0) **7**
43. Huyghe JR, Harrison TA, Bien SA. **Genetic architectures of proximal and distal colorectal cancer are partly distinct**. *Gut. Gut* (2021.0) **70** 1325-1334. PMID: 33632709
44. Lu T, Sun L, Wang Z, Zhang Y, He Z, Xu C. *Onco Targets Ther [Internet]* (2019.0) **12** 3339. PMID: 31118685
45. Zaytseva YY, Rychahou PG, Le AT. *Oncotarget [Internet]* (2018.0) **9** 24787. PMID: 29872506
46. 46.TVB 2640 for Resectable Colon Cancer Other Resectable Cancers; a Window Trial. - Full Text View - ClinicalTrials.gov [Internet]. [cited 2022 Nov 22]. Available from: https://clinicaltrials.gov/ct2/show/NCT02980029. *TVB 2640 for Resectable Colon Cancer Other Resectable Cancers; a Window Trial. - Full Text View*
47. Fhu CW, Ali A. **Fatty Acid Synthase: An Emerging Target in Cancer**. *Molecules [Internet]. Multidisciplinary Digital Publishing Institute (MDPI)* (2020.0) **25**
48. Chen HY, Cairns BJ, Small AM. *JAMA Cardiol [Internet]* (2020.0) **5** 694-702. PMID: 32186652
49. Kathiresan S, Willer CJ, Peloso GM. *Nature Genetics 2008 41:1 [Internet]* (2008.0) **41** 56-65
50. Tanaka T, Shen J, Abecasis GR. **Genome-Wide Association Study of Plasma Polyunsaturated Fatty Acids in the InCHIANTI Study**. *Public Library of Science* (2009.0) **5** e1000338
51. Würtz P, Kangas AJ, Soininen P, Lawlor DA, Davey Smith G, Ala-Korpela M. *Am J Epidemiol* (2017.0) **186** 1084-1096. PMID: 29106475
52. Julkunen H, Cichońska A, Slagboom PE, Würtz P. *Elife* (2021.0) **10**
53. Bycroft C, Freeman C, Petkova D. *Nature 2018 562:7726* (2018.0) **562** 203-209
54. Holmes M V., Smith GD. *Clin Chem* (2019.0) 363-366
55. Burgess S, Labrecque JA. **Mendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates**. *Eur J Epidemiol [Internet]. Eur J Epidemiol* (2018.0) **33** 947-952. PMID: 30039250
56. Benjamini Y, Hochberg Y. *Journal of the Royal Statistical Society: Series B (Methodological) [Internet]* (1995.0) **57** 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x
57. Clayton GL, Borges MC, Lawlor DA. *From menarche to menopause: the impact of reproductive factors on the metabolic profile of over 65,000 women* (2022.0)
58. Borges MC, Haycock PC, Zheng J. **Role of circulating polyunsaturated fatty acids on cardiovascular diseases risk: analysis using Mendelian randomization and fatty acid genetic association data from over 114,000 UK Biobank participants**. *BMC Med [Internet]. BMC Med* (2022.0) **20**
59. 59.MRC IEU UK Biobank GWAS pipeline version 2 - Datasets - data.bris [Internet]. [cited 2022 Oct 14]. Available from: https://data.bris.ac.uk/data/dataset/pnoat8cxo0u52p6ynfaekeigi. *MRC IEU UK Biobank GWAS pipeline version 2 - Datasets - data.bris [Internet]*
60. Elsworth B, Lyon M, Alexander T. *The MRC IEU OpenGWAS data infrastructure*
61. Wade Gibran Hemani Kaitlin H, Laurin Charles, Elsworth Benjamin, Burgess Stephen, Bowden Jack, Langdon Ryan, Tan Vanessa, Yarmolinsky James, Shihab Hashem A., Timpson Nicholas, Evans David M, Relton Caroline, Martin Richard M, Smith George Davey, Tom R G JZ. **MR-Base: a platform for systematic causal inference across the phenome using billions of genetic associations**. *Biorxiv* (2016.0)
62. 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. PMID: 27061298
63. Hartwig FP, Smith GD, Bowden J. **Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption**. *Int J Epidemiol* (2017.0) **46** 1985-1998. PMID: 29040600
64. Hartwig FP, Smith GD, Bowden J. **Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption**. *Int J Epidemiol [Internet]. Int J Epidemiol* (2017.0) **46** 1985-1998. PMID: 29040600
65. Hartwig FP, Smith GD, Bowden J. *Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption*
66. Burgess S, Labrecque JA. *Mendelian randomization with a binary exposure variable: interpretation and presentation of causal estimates*
67. 67.R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.; 2020.. *R: A language and environment for statistical computing* (2020.0)
68. **Forestplots of Measures of Effects and Their Confidence Intervals**. *ggforestplot [Internet]*
69. Boyd A, Golding J, Macleod J. **Cohort Profile: The ‘Children of the 90s’—the index offspring of the Avon Longitudinal Study of Parents and Children**. *Int J Epidemiol [Internet]. Oxford Academic* (2013.0) **42** 111-127
70. Borges MC, Haycock PC, Zheng J. **Role of circulating polyunsaturated fatty acids on cardiovascular diseases risk: analysis using Mendelian randomization and fatty acid genetic association data from over 114,000 UK Biobank participants**. *BMC Med [Internet]. BioMed Central Ltd* (2022.0) **20** 1-14. DOI: 10.1186/s12916-022-02399-w
71. Borges MC, Haycock P, Zheng J. *Hum Mol Genet [Internet]* (2022.0) **31** 4034-4054. PMID: 35796550
72. Rodriguez-Broadbent H, Law PJ, Sud A. **Mendelian randomisation implicates hyperlipidaemia as a risk factor for colorectal cancer**. *Int J Cancer [Internet]. Int J Cancer* (2017.0) **140** 2701-2708. PMID: 28340513
73. Cornish AJ, Law PJ, Timofeeva M. **Modifiable pathways for colorectal cancer: a mendelian randomisation analysis. Lancet Gastroenterol Hepatol [Internet]**. *Lancet Gastroenterol Hepatol* (2020.0) **5** 55-62. PMID: 31668584
74. Luo X, Tu Z, Chen H, Ding J. **Blood lipids and risk of colon or rectal cancer: a Mendelian randomization study**. *J Cancer Res Clin Oncol [Internet]. J Cancer Res Clin Oncol* (2021.0) **147** 3591-3599. PMID: 34490583
75. Granados-Romero JJ, Valderrama-Treviño A, Hazzel E, Flores C, Herrera M. **Colorectal cancer: a review**. *Article in International Journal of Research in Medical Sciences [Internet]* (2017.0). DOI: 10.18203/2320-6012.ijrms20174914
76. Park SK, Joo JS, Kim DH, Kim YE, Kang D, Yoo KY. **Association of serum lipids and glucose with the risk of colorectal adenomatous polyp in men: a case-control study in Korea**. *J Korean Med Sci [Internet]. Korean Academy of Medical Science* (2000.0) **15** 690-695. DOI: 10.3346/jkms.2000.15.6.690
77. Yang Z, Tang H, Lu S, Sun X, Rao B. *BMJ Open [Internet]* (2022.0) **12** e052373
78. Tian Y, Wang K, Li J. *Public Health Nutr [Internet]* (2015.0) **18** 3355-3370. PMID: 25776573
79. Yao X, Tian Z. *Cancer Causes and Control [Internet]* (2015.0) **26** 257-268. DOI: 10.1007/s10552-014-0507-y
80. Mayengbam SS, Singh A, Pillai AD, Bhat MK. **Influence of cholesterol on cancer progression and therapy**. *Transl Oncol. Elsevier* (2021.0) **14** 101043
81. Mannes GA, Maier A, Thieme C, Wiebecke B, Paumgartner G. **Relation between the Frequency of Colorectal Adenoma and the Serum Cholesterol Level**. *Internet]. Massachusetts Medical Society* (2009.0) **315** 1634-1638. DOI: 10.1056/NEJM198612253152602
82. Mamtani R, Lewis JD, Scott FI. *PLoS Med [Internet]* (2016.0) **13** e1002007. DOI: 10.1371/journal.pmed.1002007
83. Ocvirk S, Wilson AS, Appolonia CN, Thomas TK, O’Keefe SJD. *Curr Gastroenterol Rep [Internet]* (2019.0) **21** 1-7. DOI: 10.1007/s11894-019-0725-2
84. Tang FY, Pai MH, Chiang EPI. *J Nutr Biochem* (2012.0) **23** 1302-1313. PMID: 22221675
85. Bell AE, Culp PA. **Reduction in Saturated Fat Intake for Cardiovascular Disease**. *Am Fam Physician [Internet]. NLM* (2022.0) **105**. DOI: 10.1002/14651858.CD011737.pub2/full
86. Chiu S, Williams PT, Krauss RM. *PLoS One [Internet]* (2017.0) **12** e0170664. DOI: 10.1371/journal.pone.0170664
87. Han Q, Yeung SC, Ip MSM, Mak JCW. *Dysregulation of cardiac lipid parameters in high-fat high-cholesterol diet-induced rat model*. DOI: 10.1186/s12944-018-0905-3
88. Vincent MJ, Allen B, Palacios OM, Haber LT, Maki KC. *Am J Clin Nutr [Internet]* (2019.0) **109** 7-16. PMID: 30596814
89. May-Wilson S, Sud A, Law PJ. **Pro-inflammatory fatty acid profile and colorectal cancer risk: A Mendelian randomisation analysis**. *Eur J Cancer [Internet]. Eur J Cancer* (2017.0) **84** 228-238. PMID: 28829991
90. Khankari NK, Banbury BL, Borges MC. **Mendelian Randomization of Circulating Polyunsaturated Fatty Acids and Colorectal Cancer Risk**. *Cancer Epidemiol Biomarkers Prev [Internet]. Cancer Epidemiol Biomarkers Prev* (2020.0) **29** 860-870. PMID: 32051193
91. Liyanage UE, Ong JS, An J, Gharahkhani P, Law MH, MacGregor S. **Mendelian Randomization Study for Genetically Predicted Polyunsaturated Fatty Acids Levels on Overall Cancer Risk and Mortality**. *Cancer Epidemiol Biomarkers Prev [Internet]. Cancer Epidemiol Biomarkers Prev* (2019.0) **28** 1015-1023. PMID: 30948449
|
---
title: 'Prospective Bidirectional Relations Between Depression and Metabolic Health:
30 Year Follow-up from the NHLBI CARDIA Study'
authors:
- Nicholas R. Moorehead
- Jeffrey L. Goodie
- David S. Krantz
journal: medRxiv
year: 2023
pmcid: PMC10029061
doi: 10.1101/2023.03.08.23286983
license: CC0 1.0
---
# Prospective Bidirectional Relations Between Depression and Metabolic Health: 30 Year Follow-up from the NHLBI CARDIA Study
## Body
Metabolic syndrome (MetS), defined by the presence of obesity, hyperglycemia, dyslipidemia, and hypertension, is an important risk factor for the development of cardiovascular disease (CVD; Huang, 2009). Research has also shown that MetS is associated with depression (Akbaraly et al., 2009; Goldbacher et al., 2009; Pan et al., 2012), and depression is also a risk factor for the development of CVD (Ogunmoroti et al., 2022). Further, it has been suggested that the relationship between depression and MetS is bidirectional, with both depression preceding the onset of MetS and MetS predicting the onset of depression (Pan et al., 2012). Research involving multiple longitudinal assessments of depression and MetS in the same sample is needed to test directional associations between depression and MetS.
## Abstract
### Objective:
This study investigated prospective bidirectional relationships between depression and metabolic syndrome (MetS), and the moderating effects of race, sex, and health behaviors in a diverse cohort followed for 30 years.
### Methods:
Data were analyzed from the NHLBI CARDIA study, a 30 year-prospective study of young adults ($$n = 5113$$; M age = 24.76 (SD = 3.63) at baseline; $45\%$ male) who were tested every 5 years between 1985–2015. Measures included biological assessments of MetS components, and self-reported depressive symptoms based on the Center for Epidemiologic Studies Depression (CESD) scale. Data analyses included bi-directional general estimating equations analyses of time-lagged associations between depressive symptoms and MetS.
### Results:
There was a consistent, bi-directional relationship between depressive symptoms and MetS over time. Individuals with more CESD depressive symptoms were more likely to develop MetS over time compared to those reporting fewer symptoms (Wald Chi-Square = 7.09 [1], $p \leq 0.008$), and MetS was similarly predictive of CESD. MetS more consistently predicted depressive symptoms at each 5-year exam than depressive symptoms predicted MetS. Race and sex moderated relationships between depression and MetS, with White females, White individuals overall, and females overall demonstrating significant relationships. Health behaviors were not related to depression-MetS associations.
### Conclusion:
In a diverse young adult population prospectively followed into late middle age, MetS more consistently predicted depression over time than depression predicted MetS. The relation between MetS and depressive symptoms was moderated by race and sex, but not health behaviors.
## Relationships between Depressive Symptoms and MetS
The predictive relationship of depression for metabolic syndrome is well-established. For example, a recent meta-analysis of 49 studies (Moradi et al., 2021) with a total sample size of nearly 400,000, observed a greater likelihood of developing MetS among depressed individuals compared to those who were non-depressed. These relationships held for both cross-sectional and cohort studies, and there was evidence of a stronger relationship between depression and MetS in women, compared to men. Evidence further indicates that associations between depression and MetS may differ by race as well as gender (Womack et al., 2016). However, because the incidence of MetS increases with age, research is needed to determine whether there are gender and race differences in the relationships between depression and MetS over the lifespan (Moradi et al., 2021; Pan et al., 2012; Womack et al., 2016).
Mechanisms suggested for a causal relationship of depressive symptoms to MetS include associations between depression and biological and behavioral changes that promote the development of MetS. Neuroendocrine, autonomic, and inflammatory effects of depression can have effects on factors contributing to MetS such as abdominal fat, metabolism, and blood pressure (Akbaraly et al., 2009; Hryhorczuk et al., 2013; Liaw et al., 2015; Pan et al., 2012).
Depression is also associated with behavioral risk factors that increase the development of MetS. These include reduced exercise (Belvederi Murri et al., 2019), obesity (Hryhorczuk et al., 2013; Luppino et al., 2010; Needham et al., 2010), poor diet (Ljungberg et al., 2020), smoking (Bakhshaie et al., 2015; Fluharty et al., 2017; Mathew et al., 2017), sleep (Murphy & Peterson, 2015), and reduced medication compliance and health behaviors (Akbaraly et al., 2009; Grenard et al., 2011; Liaw et al., 2015; Womack et al., 2016).
## Bi-directional Relationships Between Depression and MetS
The overlap between physiological, social, behavioral, and emotional sequalae of both MetS and depression has led researchers to investigate the extent to which this relationship may be bi-directional (i.e., the presence of metabolic syndrome may be predictive of depression as well). However, evidence for metabolic syndrome as a predictor of depression is equivocal, with both positive (Akbaraly et al., 2009), and negative findings (Foley et al., 2010). Summarizing this literature, a meta-analysis of 29 cross-sectional and 11 cohort studies of the relation of depression and MetS included 11 studies that used MetS as a predictor of depression and 12 studies examining depression as a predictor of MetS (Pan et al, 2012). These authors concluded that the aggregate data supported a bidirectional relationship between depression and MetS. The relation between depression and MetS appeared to be stronger in cross-sectional studies that measured depression using self-reported symptom scales compared to clinical interviews and formal clinical diagnosis (Pan et al., 2012).
There are plausible mechanisms that may account for a possible predictive relationship of MetS for depression. MetS is associated with increased levels of inflammatory cytokines such as C-reaction protein, interleukin, as well as leptin resistance which are also associated with depressed mood (Akbaraly et al., 2009). Behavioral factors that are associated with MetS (e.g., sedentary behavior, poor diet, obesity, and sleep disturbance) might increase an individual’s likelihood of developing depression as a reduction in behavioral activation can lead to increased negative thoughts and emotions (Akbaraly et al., 2009; Gozal et al., 2016; Hryhorczuk et al., 2013; Liaw et al., 2015; Pan et al., 2012). In addition, many of the treatments recommended for MetS (e.g., physical activity, eat a healthy diet, improve sleep) are similar to behavioral treatments effective in reducing depression (Hiles et al., 2016; Pan et al., 2012).
## Relations of Sex, Age, and Race with Depression and Metabolic Syndrome
Demographic factors such as sex, age, and race are associated with the incidence and prevalence of both depression and metabolic syndrome (Bailey et al., 2019; Gurka et al., 2014; Hargrove et al., 2020; Womack et al., 2016). Women experience twice the risk of depression when compared to men (Kessler & Bromet, 2013; Sutin et al., 2013). With regard to age, longitudinal research indicates that depressive symptoms are highest in young adulthood, decrease in middle age, and subsequently increase again in older adults (Sutin et al, 2013). Metabolic syndrome increases in prevalence with age, with the age-related increase being greater in women across the lifespan (Vishram et al., 2014). However, it is not known whether there are bi-directional relationships between depression and metabolic syndrome across the lifespan for extended periods from young adulthood to older ages (Pan et al., 2012). Reviews and meta-analyses of multiple studies have been able to indirectly assess the presence of a bidirectional relation between depression and MetS, but little is known about a possible bidirectional relation in the same individuals (Moradi et al., 2021; Pan et al., 2012; Womack et al., 2016).
A recent report from the NHLBI CARDIA study examined relationships between MetS and depression utilizing 15-year follow-up (Womack et al., 2016). The ages at follow-up in that study were between 33 and 45 years. The prevalence of MetS was highest among Black women, followed by White men, Black men, and White women. As in prior research (Womack et al., 2016), incidence of MetS was higher for those who reported more depressive symptoms. In addition, depressive symptoms were more strongly associated with the development of MetS in White men and White women, with weak or no associations among Black men and women (Womack et al., 2016).
## Current Study
The presence, strength, and consistency of a possible bi-directional relationship over the lifespan is not well understood. Therefore, the primary goal of the present study is to examine possible bi-directional relationships in a 30-year longitudinal follow-up with assessments made every 5 years in the same sample of participants. In addition, this study also extended the findings of Womack et al. [ 2016] in the CARDIA study, using a longer follow-up (30 vs. 15 years) from young adulthood (ages 18–30) to late middle age (ages 48 – 60). Another study goal was to determine whether health behaviors and demographic factors influenced these relationships. Finally, we further extended CARDIA findings by examining these relationships among Black and White men and women using this longer follow-up. We hypothesized that there would be a bidirectional association between metabolic health and depression, and that race, sex, and health behaviors would moderate the relation between depression and metabolic health.
## METHODS
This study utilized data from the limited access dataset from National Heart, Lung, and Blood Institute (NHLBI) Coronary Artery Disease in Young Adults (CARDIA) study (NHLBI, 2021) of a diverse population of young adults followed up for 30 years between 1985 and 2015. Participants in were 5115 Black ($$n = 2637$$) and White ($$n = 2462$$) men ($$n = 2328$$) and women ($$n = 2785$$) age 18–30 at intake who were re-tested at follow-ups during 1987 (Year 2), 1990 (Year 5), 1992 (Year 7), 1995 (Year 10), 2000 (Year 15), 2005 (Year 20), 2010 (Year 25), and most recently 2015 (Year 30). Detailed methodology for CARDIA is presented elsewhere (Friedman et al., 1988). Due to the low number of individuals identifying as Hispanic ($$n = 14$$), these individuals were excluded from analyses. The present analyses use assessments from 1990 (Year 5), 1995 (Year 10), 2000 (Year 15), 2005 (Year 20), 2010 (Year 25), and 2015 (Year 30).
## Demographic and Clinical Data.
At baseline and every follow-up, medical and demographic measures were collected. Variables of interest for the present analyses included: anthropomorphic and sociodemographic variables such as body mass index, age, race, and sex; smoking status; alcohol consumption; depression/depressive symptoms; and biological variables including MetS and its components.
## Physical Activity.
To measure physical activity, we utilized the CARDIA self-report physical activity history measured at each examination (Camhi et al., 2013; Jacobs et al., 1989). This questionnaire measure was developed specifically for use in the CARDIA study to provide a suitable reliable and valid brief, yet comprehensive, questionnaire to measure different types of physical activity (Camhi et al., 2013; Jacobs et al., 1989). Questions pertain to the types of physical activity, perceived level of difficulty/ rigor, and the total amount in hours/ minutes at either daily or weekly intervals. A physical activity score based on the time, intensity, and frequency of each documented activity was calculated. Total hours per week of moderate, hard, very hard activity are also derived from the physical activity recall.
## Center for Epidemiological Studies Depression Scale (CESD).
This 20-item questionnaire is widely used to assess depression and depressive disorder in population studies. CESD items consist of Likert scales ranging from “rarely or none of the time” to “most or all of the time,” with scale scores ranging from 0 to 60. Although not intended for clinical diagnosis, a score ≥16 has been associated with suspected depression diagnosis (Radloff, 1977). The CESD has good sensitivity and specificity for detecting depression, is useful across the lifespan, and is reliable when compared to other depression scales (Cosco et al., 2017; Radloff, 1977; Shafer, 2006; Vilagut et al., 2016). In CARDIA, The CESD was administered at baseline and at follow-up years 5, 10, 15, 20, 25, and 30. Therefore, the final depression measure was 30 years following intake, and the first measure of depression used was at year 5.
## Bloodwork and Identification of MetS.
Before each collection of blood, participants fasted for at least 12 hours in accordance with standardized protocols (Friedman et al., 1988). At each exam, participants had their blood pressure taken three times with the average of the final two being recorded. Waist circumference was determined by the average of two waist circumference measurements (Friedman et al., 1988; Womack et al., 2016).
For this study, we used the NCEP-ATP III criteria to define MetS (Grundy, Brewer, et al., 2004; Grundy, Hansen, et al., 2004). This criterion indicates that no single variable threshold is required for MetS, but subjects must meet 3 of the 5 following criteria: [1] Obesity defined by a waist circumference greater than 40 in. ( 101.6 cm) for men or greater than 35 in. ( 88.9 cm) for women; [2] Hyperglycemia indicated by fasting glucose greater than or equal to 100 mg/dl or by a prescription; [3] Dyslipidemia indicated by any of the following: triglyceride levels ≥ 150 mg/dl; medication treatment for dyslipidemia; HDL-C < 40 mg/dl for men and < 50 mg/dl for women [4] Hypertension indicated by systolic blood pressure >130 mmHg, diastolic blood pressure >85 mmHg, or use of prescribed antihypertensive medications. MetS was coded as a binary variable because the presence of MetS is considered to be a diagnostic category without an established criteria for severity of the syndrome (Grundy, Brewer, et al., 2004; Grundy, Hansen, et al., 2004; Huang, 2009).
## Relations Between Depression and MetS.
To determine longitudinal relationships between the predictor variable depression to the MetS outcome over time, and the moderating effects of health behaviors on that relationship, data were analyzed using general estimating equations (GEE) in SPSS. This analysis allowed for both a cross-sectional and longitudinal output in a single GEE model accounting for fixed effects. This means a single analysis provided between-groups (between-subjects) relationships at time points baseline, Year 5, Year 10, Year 15, Year 20, Year 25, and Year 30, longitudinal (within-subjects) results over the entire study period, and allowed for pairwise comparisons to be included to compare groups at various time points. The GEE approach also has the benefit of being robust enough to include participants even if data are missing at one or several timepoints, and significantly reduce likelihood of Type 1 error by reducing the total numbers of analyses/ models conducted.
Depression was coded as binary (CESD < 16 = “not depressed”; CESD ≥ 16 = “depressed”). MetS was coded as binary (MetS = ≥3 criterion variables [i.e., hypertension, obesity, dyslipidemia by triglycerides, dyslipidemia by HDL-C, or hyperglycemia]; vs No MetS (<3 or more criterion variables) at each time point. This allowed determination of an odds ratio (OR) for developing MetS relative to the reference (not depressed) group. Covariates (age, sex, race, physical activity, smoking, alcohol consumption, and time) were included in the analyses.
## Bi-directional Relations Between Depression and MetS.
To examine bi-directional associations between depression and MetS, two sets of lagged time-series logistic regressions were utilized. First, to determine if depression predicts MetS, binary coding of depression at each exam year served as the predictor variable and MetS at the subsequent exam year was the outcome variable. Five logistic regressions were conducted: depression at Year 5, 10, 15, 20, and 25 predicting MetS status at Year 10, 15, 20, 25, and 30, respectively. To determine whether depression predicted future MetS, 5 logistic regressions were also conducted with depression at each exam year used to predict MetS status at the next exam year. Covariates (age, sex, race, physical activity, smoking, alcohol consumption, and time) were included in the analyses.
## Depression and MetS Relations with Sex and Race.
To examine whether relationships between depression and MetS outcomes differ by sex (male vs. female) and race (Black vs. White), Depression and MetS were again coded as binary variables, and sex or race were included as possible moderators of relationships between depression and MetS. Three GEE models were conducted, one for testing the possible moderating effects of sex, one for testing the moderating effects of race, and on for testing the moderating effects of race and sex together.
## Demographics.
As shown in Table 1, the baseline sample had an approximately even racial Black/White split ($51\%$ identifying as Black), and an even male/female split ($55\%$ females). Over the course of the study the total number of participants decreased from 5113 at baseline to 3357 at Year 30 demonstrating a $66\%$ retention rate. At Year 30, the split across racial and sex lines differed ($57\%$ females), and $48\%$ Black.
## Health Behaviors.
Those who reporting “never smoked” stayed relatively consistent, ($56\%$ at baseline vs. $63\%$ of the retained sample at Year 30; see Table 2). Comparing baseline to Year 30: current smokers decreased (approximately $31\%$ at baseline vs. $14\%$); reported alcohol consumed stayed relatively constant; and physical activity slightly declined.
## MetS.
Table 3 presents rates of those meeting overall MetS diagnostic criteria, and rates with MetS components of hypertension, obesity, dyslipidemia, hyperglycemia; fasting glucose was not measured at Year 5. The number of positive MetS components, and number of individuals meeting criteria for MetS increases over the course of the study.
## Depression.
Table 3 summarizes depression data. Participants scoring in the depressed range was highest at Year 5 (i.e., $24\%$ of the sample), dropped to approximately $17\%$ of the sample at Year 15, and remained stable until Year 30. The CESD was not given used at baseline.
## Relations Between Depression, MetS, and Health Behaviors
As summarized in Table 4, depression status was significantly related to MetS status (Wald Chi-Square = 7.09 [1], $p \leq 0.008$) over the course of the study. The effect of time by itself was significant ($p \leq 0.0001$). The time by depression interaction was not significant ($$p \leq 0.45$$), suggesting that the relation between depression and MetS diagnosis does not vary over time.
Physical activity, smoking, and alcohol consumption were each examined as possible moderators of the predictive relationships between depression and MetS. For CESD depression as the independent variable and MetS diagnosis as the dependent variable, in cross-sectional, longitudinal, and/or lagged-times series analyses, no individual health behavior had a significant moderating effect on the relationship between depression and MetS diagnosis. Although there were significant effects for depression and for physical activity on MetS in the overall model ($p \leq 0.05$), physical activity did not moderate these relationships. Therefore, the relationship to MetS to depression did not vary by physical activity or any other health behavior.
## Bi-directional Relations Between Depression and MetS
Results for the first series of lagged logistic regressions examining whether depression predicts MetS status are presented in Figure 1A. Three of the 5 time-lagged results for depression as a predictor of MetS (years 10–15, 15–20, and 25–30) were significant ($p \leq 0.05$). The OR’s stay consistent across time, ranging diagnosis of depression at Year 10 preceding development of MetS at Year 15 (OR = 1.38, $95\%$ CI [1.08, 1.75], $p \leq 0.009$), to (OR = 1.32, $95\%$ CI [1.05, 1.66], $p \leq 0.016$) for depression at year 25 predicting MetS at Year 30.
Figure 1B presents results for the series of lagged logistic regressions examining MetS as a predictor of depression. Although the relationships were significant at all time points, the OR’s were somewhat more variable in their magnitude than in analyses of depression predicting MetS. At Year 5, if a person was diagnosed with MetS vs. no MetS, they were 1.61 times more likely to have depression at Year 10 (OR = 1.61, $95\%$ CI [1.05, 2.50], $p \leq 0.030$). For Year 25 MetS predicting Year 30 depression, the OR was 1.30, $95\%$ CI [1.02. 1.65], $p \leq 0.032.$
## Sex, Race, and Relationships between Depression and MetS
Table 5 presents results of three separate GEE models examining whether relationships between depression and MetS outcomes differ by race (Black vs. White), by sex (male vs. female), and their combination. In the first model, race did not have significant relation with MetS diagnosis; nor did it moderate the relation between depression and MetS status ($$p \leq 0.076$$). In the second model, Sex had a significant relation with MetS diagnosis ($p \leq 0.001$), and sex moderated the relation between depression and MetS diagnosis ($p \leq 0.004$), with effects stronger in women. In the third model, Sex and Race significantly interact ($p \leq 0.001$), and together moderate the relation between depression and MetS ($p \leq .002$). Figure 2 displays the ORs and $95\%$ confidence intervals for comparative effects of the Sex and Race groups as possible moderators of depression as a predictor of MetS. As indicated in the figure, the only significant sex/race groups are White females, females, and White individuals.
## DISCUSSION
This study is unique in examining bi-directional, longitudinal relations between depression and MetS in a young adult cohort as they progressed to middle-age over a 30-year period. The present findings of significant relationships between depression and MetS over time are consistent with prior findings from several studies and meta-analyses (Akbaraly et al., 2009; Foley et al., 2010; Goldbacher et al., 2009; Pan et al., 2012). In addition, this study found bi-directional relationships between depression and MetS over the 30-year time period.
The present results suggest that weaker findings found in cross-sectional comparisons in prior studies may reflect stronger relationships between depression and MetS at certain points in the lifespan, and/or that cross-sectional assessments at only a single time point may contribute to weaker associations than longitudinal measurements at multiple time points. Interestingly, the present findings were obtained despite the fact that the NCEP ATP III criteria were used to define MetS in the present study. Pan et al. [ 2012] noted that in other studies using NCEP ATP III criteria to define MetS, the relation was found to be weaker, likely due to the NCEP ATP III requiring a lower threshold to meet diagnostic criteria compared to other diagnostic criteria for MetS (Pan et al., 2012). Also consistent with prior research are our findings over 30-years that the relation between MetS and depression is weak or not significant in males, and more consistently found in females (Pan et al., 2012). These findings are addressed later in the discussion section.
Prior studies did not assess bidirectional relationships between depression and MetS in the same study cohort. In this study, unique analyses using lagged time series models were made possible by the multiple assessments over time in CARDIA, and the predictive relation between depression and MetS was slightly stronger and more consistent for MetS prospectively predicting depression than the opposite (Akbaraly et al., 2009; Foley et al., 2010; Goldbacher et al., 2009; Hiles et al., 2016; Moradi et al., 2021; Pan et al., 2012; Womack et al., 2016). This finding of a more consistent effect for MetS predicting depression is supported by some studies (Akbaraly et al., 2009; Pan et al., 2012) but not others (Hiles et al., 2016). However, compared to prior studies not looking at both relationships in the same cohort, the present findings based on multiple longitudinal datapoints in the same cohort over 30 years are convincing. However, further study is needed to assess the consistency of these bidirectional relationships in other cohorts.
## Possible Mechanisms for Links Between Depression and MetS
Several causal mechanisms have been suggested to explain the predictive relation of depression for MetS; however, the predictive relationship of MetS for depression might seem less intuitive. Factors contributing to both directional relationships may include biological factors (e.g., elevated stress levels causing HPA axis dysregulation, changes in levels of cortisol and glucocorticoids, and increased pro-inflammatory responses), psychological factors (e.g., increased social isolation associated with depression), and behavioral changes (e.g., health-impairing behaviors such as smoking, alcoholism, decreased exercise levels, poor diet).
In the present study, moderating factors examined were physical activity, smoking, and alcohol consumption. Of these, only physical activity demonstrated a moderating effect on MetS. The reasons for null findings for smoking and alcohol consumption, which are known to increase risk of MetS (Boyle et al., 2018; Carroll et al., 2019; Sun et al., 2012), are not clear. Although previous reports indicate that neither smoking nor alcohol consumption were significantly related to MetS in the CARDIA sample (Carnethon et al., 2004; Duffey et al., 2010), one possible explanation is that those who engage in these behaviors are more likely to drop out of CARDIA. As evidence for this, current smokers accounted for a smaller proportion of the sample at later time points, whereas there was no change in the proportion of former smokers over time. This may indicate a higher dropout rate for those who smoked; had they remained in the study, this might have revealed a relationship between smoking and increased risk of MetS.
For alcohol consumption, it also appears the relationship with metabolic syndrome is more strongly related to drinking high caloric beverages rather than specifically to alcohol use (Duffey et al., 2010). However, the relationship between depression and metabolic syndrome is evident regardless of whether individuals smoke or consume alcohol.
Although not examined in the present study, inflammation has been suggested as a particularly important common biological and bidirectional link between depression and MetS because of its association with both conditions (Capuron et al., 2008; Frank et al., 2021; Sumner et al., 2020). Specifically, studies indicate that an increased inflammatory response caused by a variety of conditions may affect mood and predict depression (Capuron et al., 2008; Frank et al., 2021; Sumner et al., 2020), and that inflammation also increases the likelihood MetS because of the effects of inflammation on of C-reactive protein and interleukin (Capuron et al., 2008; Frank et al., 2021; Sumner et al., 2020). If inflammation can serve as both a causal mechanism and a consequence of both MetS and depression, and this has the potential to create a “vicious cycle” in vulnerable individuals. Specifically, MetS could lead to depression, which in turn, would then worsen the individual’s MetS, and so on.
## Depression and MetS Relations with Sex and Race
In the present study, the relation between depression and MetS was moderated by sex and race (Blacks vs. Whites). Sex independently moderated the relation, but race by itself did not moderate the relation. Furthermore, the relations between depression with MetS were significant for White females, females, and White individuals, but not for other race/sex subgroups. Other studies also support the present results (Beydoun et al., 2020; Cooper et al., 2013; Womack et al., 2016; Yu et al., 2020). Further, the present findings replicate the prior 15-year findings (Womack et al., 2016) from CARDIA over a 30-year period later in the lifespan. One exception is that in the prior study, White males did not demonstrate a significant relationship between depression and MetS. These results may be due to a shorter follow-up, and suggest that some differences emerge in older ages.
It has been suggested that MetS has a “unique phenotype” in Black vs. White adults (Womack et al., 2016). Specifically, this may be partially explained by the fact that dyslipidemia has a significant relation with depression. Black-identified individuals may be less likely to have dyslipidemia as a contributor for MetS, whereas dyslipidemia is often a stronger contributor to MetS in White-identified individuals (Womack et al., 2016). Because dyslipidemia also has a significant relation with depression, it may help to explain the absence of a significant relation between depression and MetS for Black-identified individuals (Womack et al., 2016). As a result, this may show that dyslipidemia is serving as a possible differentiating factor in the variation of the association between MetS and depression along racial lines.
Biological mechanisms alone may not fully explain the differences among race/sex groups in relationships between MetS and depression. There may also be issues with assessment, measurement, and diagnosis of symptoms to identify Black-identified individuals with depression/ depression symptomology as there are known variations in presentations along racial lines (Womack et al., 2016). This may most likely manifest in the form of under-reported depression as Black-identified individuals who experience more socioeconomic stress are less likely to report psychological symptoms (Bailey et al., 2019). Furthermore, minority populations are less likely to suffer from acute depression and more likely to suffer from chronic depression than White-identified individuals (Bailey et al., 2019). The higher likelihood that minority populations suffer from chronic depression is a problem that is compounded by the fact that there is evidence to suggest Black-identified individuals are more likely to experience health disparities and access to healthcare issues. ( Colen et al., 2018; Noonan et al., 2016).
Also consistent with the present study are prior findings that systemic inflammation might be more pronounced among the subgroup of White-identified individuals (Beydoun et al., 2020). Another study indicates a sex difference in that major depressive disorder is associated with MetS among women but not among men (Yu et al., 2020). This may be due to differences that may exist in the metabolic processes between men and women. However, this may also be due to the ways in depression is reported, recorded, and treatment is sought out which may be different across sex lines similar to differences seen in race (Yu et al., 2020).
It should be noted that there are some contradictory findings that indicate a link between depression and MetS among those identified as Black, particularly Black women (Cooper et al., 2013). In contrast to our findings, the latter study found that hypertension, obesity, and HDL-C contributed most to the link between depression and MetS (Cooper et al., 2013) whereas we did not find an association between depression and either hypertension or obesity. Since causal factors (i.e., hypertension and obesity) were not related to depression MetS itself would not be significant for this sample either.
## Study Limitations and Strengths
Although this study is prospective, a limitation is that other than examining health behaviors, it did not evaluate possible causal mechanisms or explanatory links for the relation between depression and MetS. There are also limits to the generalizability of the findings from this study. The study cohort included individuals who identified as White or Black, but the study did not include other racial or ethnic groups. Additionally, this study’s sample population was aged 18–30 at intake and 48–60 at the latest available timepoint, so the results may not be generalizable to other age groups. There may also be limitations to these conclusions, as the CESD measure for depression may not be sensitive enough to identify clinical depression, and may be subject to racial differences in depression symptoms (Bailey et al., 2019; Gurka et al., 2014; Womack et al., 2016). Results may also be specific to use of the NCEP-ATP III criteria to define MetS (Huang, 2009), which requires that individuals subjects must meet three of five components for MetS.
There are several notable strengths of the study. The CARDIA study cohort allowed us to examine relationships between depression and MetS in a bi-racial cohort with equal numbers of males and females allowing for a robust test of race and sex as moderators. Furthermore, this study looked at a cohort over the course of 30 years comparing longitudinal and cross-sectional data across those 30 years. Additionally, because of the CARDIA study design with multiple assessments of both variables, we were able to evaluate predictive bi-directional relationships between depression and MetS in the same cohort using lagged time series analyses.
The present findings have several implications for future research on depression, MetS, and relations between the two. First, having established a bidirectional relation between depression and metabolic health in the same cohort, there is a need to determine mechanisms that might mediate these effects in both directions. These potential causal mechanisms may be evaluated in a future study by looking evaluating a potential mediator such as cortisol or inflammatory cytokines. Another issue needing investigation is whether these effects are evident in individuals over the age of 60 years. To better understand the moderating effects of sex and race, further research should determine whether differences are evident with other measures of depression (e.g., clinical interview or clinician ratings) that are not solely based on self-report and that may be less subject to biases due to sex and race. In addition, examining the underlying mechanisms of how MetS and depression are evaluated and assessed across race and sex lines may allow for better diagnosis of conditions in different populations.
## References
1. Akbaraly T. N., Kivimaki M., Brunner E. J., Chandola T., Marmot M. G., Singh-Manoux A., Ferrie J. E.. **Association between metabolic syndrome and depressive symptoms in middle-aged adults: Results from the Whitehall II Study**. *Diabetes Care* (2009) **32** 499-504. DOI: 10.2337/dc08-1358
2. Bailey R., Mokonogho J., Kumar A.. **Racial and ethnic differences in depression: current perspectives**. *Neuropsychiatric Disease and Treatment* (2019) **15** 603-609. DOI: 10.2147/ndt.s128584
3. Bakhshaie J., Zvolensky M. J., Goodwin R. D.. **Cigarette smoking and the onset and persistence of depression among adults in the United States: 1994–2005**. *Comprehensive Psychiatry* (2015) **60** 142-148. DOI: 10.1016/j.comppsych.2014.10.012
4. Belvederi Murri M., Ekkekakis P., Magagnoli M., Zampogna D., Cattedra S., Capobianco L., Serafini G., Calcagno P., Zanetidou S., Amore M.. **Physical exercise in major depression: Reducing the mortality gap while improving clinical outcomes**. *Frontiers in Psychiatry* (2019) **9** 1-10. DOI: 10.3389/fpsyt.2018.00762
5. Beydoun M. A., Obhi H. K., Weiss J., Canas J. A., Beydoun H. A., Evans M. K., Zonderman A. B.. **Systemic inflammation is associated with depressive symptoms differentially by sex and race: A longitudinal study of urban adults**. *Molecular Psychiatry* (2020) **25** 1286-1300. DOI: 10.1038/s41380-019-0408-2
6. Boyle M., Masson S., Anstee Q. M.. **The bidirectional impacts of alcohol consumption and the metabolic syndrome: Cofactors for progressive fatty liver disease**. *Journal of Hepatology* (2018) **68** 251-267. DOI: 10.1016/j.jhep.2017.11.006
7. Camhi S. M., Katzmarzyk P. T., Broyles S., Church T. S., Hankinson A. L., Carnethon M. R., Sternfeld B., Lewis C. E.. **Association of metabolic risk with longitudinal physical activity and fitness: Coronary artery risk development in young adults (CARDIA)**. *Metabolic Syndrome and Related Disorders* (2013) **11** 195-204. DOI: 10.1089/met.2012.0120
8. Capuron L., Su S., Miller A. H., Bremner J. D., Goldberg J., Vogt G. J., Maisano C., Jones L., Murrah N. V., Vaccarino V.. **Depressive symptoms and metabolic syndrome: Is inflammation the underlying link?**. *Biological Psychiatry* (2008) **64** 896-900. DOI: 10.1016/j.biopsych.2008.05.019
9. Carnethon M. R., Loria C. M., Hill J. O., Sidney S., Savage P. J., Liu K.. **Risk factors for the metabolic syndrome**. *Diabetes Care* (2004) **27** 2707-2715. DOI: 10.2337/diacare.27.11.2707
10. Carney R. M., Freedland K. E.. **Depression and coronary heart disease**. *Nature Reviews Cardiology* (2017) **14** 145-155. DOI: 10.1038/nrcardio.2016.181
11. Carroll A. J., Huffman M. D., Zhao L., Jacobs D. R., Stewart J. C., Kiefe C. I., Liu K., Hitsman B.. **Evaluating longitudinal associations between depressive symptoms, smoking, and biomarkers of cardiovascular disease in the CARDIA study**. *Psychosomatic Medicine* (2019) **81** 372-379. DOI: 10.1097/psy.0000000000000667
12. Colen C. G., Ramey D. M., Cooksey E. C., Williams D. R.. **Racial disparities in health among nonpoor African Americans and Hispanics: The role of acute and chronic discrimination**. *Social Science & Medicine* (2018) **199** 167-180. DOI: 10.1016/j.socscimed.2017.04.051
13. Cooper D. C., Trivedi R. B., Nelson K. M., Reiber G. E., Zonderman A. B., Evans M. K., Waldstein S. R.. **Sex differences in associations of depressive symptoms with cardiovascular risk factors and metabolic syndrome among African Americans**. *Cardiovascular Psychiatry and Neurology* (2013) **2013** 1-10. DOI: 10.1155/2013/979185
14. Cosco T. D., Prina M., Stubbs B., Wu Y. T.. **Reliability and validity of the center for epidemiologic studies depression scale in a population-based cohort of middle-aged U.S. adults**. *Journal of Nusing Measurement* (2017) **25** 476-485. DOI: 10.1891/1061-3749.25.3.476
15. Duffey K. J., Gordon-Larsen P., Steffen L. M., Jacobs D. R., Popkin B. M.. **Drinking caloric beverages increases the risk of adverse cardiometabolic outcomes in the Coronary Artery Risk Development in Young Adults (CARDIA) Study**. *The American Journal of Clinical Nutrition* (2010) **92** 954-959. DOI: 10.3945/ajcn.2010.29478
16. Fluharty M., Taylor A. E., Grabski M., Munafò M. R.. **The association of cigarette smoking with depression and anxiety: A systematic review**. *Nicotine & Tobacco Research* (2017) **19** 3-13. DOI: 10.1093/ntr/ntw140
17. Foley D. L., Morley K. I., Madden P. A. F., Heath A. C., Whitfield J. B., Martin N. G.. **Major depression and the metabolic syndrome**. *Twin Research and Human Genetics* (2010) **13** 347-358. DOI: 10.1375/twin.13.4.347
18. Frank P., Jokela M., Batty G. D., Cadar D., Steptoe A., Kivimaki M.. **Association between systemic inflammation and individual symptoms of depression: A pooled analysis of 15 population-based cohort studies**. *The American Journal of Psychiatry* (2021) **178** 1107-1118. DOI: 10.1176/appi.ajp.2021.20121776
19. Friedman G. D., Cutter G. R., Donahue R. P., Hughes G. H., Hulley S. B., Jacobs D. R., Liu K., Savage P. J.. **CARDIA: Study design, recruitment, and some characteristics of the examined subjects**. *Journal of Clinical Epidemiology* (1988) **41** 1105-1116. DOI: 10.1016/0895-4356(88)90080-7
20. Goldbacher E. M., Bromberger J., Matthews K. A.. **Lifetime history of major depression predicts the development of the metabolic syndrome in middle-aged women**. *Psychosomatic Medicine* (2009) **71** 266-272. DOI: 10.1097/psy.0b013e318197a4d5
21. Gozal D., Dumin M., Koren D.. **Role of sleep quality in the metabolic syndrome**. *Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy* (2016) **9** 281-310. DOI: 10.2147/dmso.s95120
22. Grenard J. L., Munjas B. A., Adams J. L., Suttorp M., Maglione M., McGlynn E. A., Gellad W. F.. **Depression and medication adherence in the treatment of chronic diseases in the United States: A meta-analysis**. *Journal of General Internal Medicine* (2011) **26** 1175-1182. DOI: 10.1007/s11606-011-1704-y
23. Grundy S. M., Brewer H. B., Cleeman J. I., Smith S. C., Lenfant C.. **Definition of metabolic syndrome**. *Arteriosclerosis, Thrombosis, and Vascular Biology* (2004) **24** e13-e18. DOI: 10.1161/01.atv.0000111245.75752.c6
24. Grundy S. M., Hansen B., Smith S. C., Cleeman J. I., Kahn R. A.. **Clinical management of metabolic syndrome**. *Arteriosclerosis, Thrombosis, and Vascular Biology* (2004) **24** e19-e24. DOI: 10.1161/01.atv.0000112379.88385.67
25. Gurka M. J., Lilly C. L., Oliver M. N., Deboer M. D.. **An examination of sex and racial/ethnic differences in the metabolic syndrome among adults: A confirmatory factor analysis and a resulting continuous severity score**. *Metabolism* (2014) **63** 218-225. DOI: 10.1016/j.metabol.2013.10.006
26. Hargrove T. W., Halpern C. T., Gaydosh L., Hussey J. M., Whitsel E. A., Dole N., Hummer R. A., Harris K. M.. **Race/ethnicity, gender, and trajectories of depressive symptoms across early- and mid-life among the add health cohort**. *Journal of Racial and Ethnic Health Disparities* (2020) **7** 619-629. DOI: 10.1007/s40615-019-00692-8
27. Hiles S. A., Révész D., Lamers F., Giltay E., Penninx B. W. J. H.. **Bidirectional prospective associations of metabolic syndrome components with depression, anxiety, and antidepressant use**. *Depression and Anxiety* (2016) **33** 754-764. DOI: 10.1002/da.22512
28. Hryhorczuk C., Sharma S., Fulton S. E.. **Metabolic disturbances connecting obesity and depression**. *Frontiers in Neuroscience* (2013) **7**. DOI: 10.3389/fnins.2013.00177
29. Huang P. L.. **A comprehensive definition for metabolic syndrome**. *Disease Models & Mechanisms* (2009) **2** 231-237. DOI: 10.1242/dmm.001180
30. Jacobs D. R., Hahn L. P., Haskell W. L., Pirie P., Sidney S.. **Validity and reliability of short physical activity history: CARDIA and the Minnesota heart health program**. *Journal of Cardiopulmonary Rehabilitation* (1989) **9** 448-459. DOI: 10.1097/00008483-198911000-00003
31. Liaw F.-Y., Kao T.-W., Hsueh J.-T., Chan Y.-H., Chang Y.-W., Chen W.-L.. **Exploring the link between the components of metabolic syndrome and the risk of depression**. *BioMed Research International* (2015) **2015** 1-8. DOI: 10.1155/2015/586251
32. Ljungberg T., Bondza E., Lethin C.. **Evidence of the importance of dietary habits regarding depressive symptoms and depression**. *International Journal of Environmental Research and Public Health* (2020) **17** 1-18. DOI: 10.3390/ijerph17051616
33. Luppino F. S., De Wit L. M., Bouvy P. F., Stijnen T., Cuijpers P., Penninx B. W. J. H., Zitman F. G.. **Overweight, obesity, and depression**. *Archives of General Psychiatry* (2010) **67** 220-229. DOI: 10.1001/archgenpsychiatry.2010.2
34. Mathew A. R., Hogarth L., Leventhal A. M., Cook J. W., Hitsman B.. **Cigarette smoking and depression comorbidity: systematic review and proposed theoretical model**. *Addiction* (2017) **112** 401-412. DOI: 10.1111/add.13604
35. Moradi Y., Albatineh A. N., Mahmoodi H., Gheshlagh R. G.. **The relationship between depression and risk of metabolic syndrome: a meta-analysis of observational studies**. *Clinical Diabetes and Endocrinology* (2021) **7** 1-12. DOI: 10.1186/s40842-021-00117-8
36. Murphy M. J., Peterson M. J.. **Sleep disturbances in depression**. *Sleep Medicine Clinics* (2015) **10** 17-23. DOI: 10.1016/j.jsmc.2014.11.009
37. Needham B. L., Epel E. S., Adler N. E., Kiefe C.. **Trajectories of change in obesity and symptoms of depression: The CARDIA study**. *American Journal of Public Health* (2010) **100** 1040-1046. DOI: 10.2105/ajph.2009.172809
38. NHLBI. (2021). CARDIA Study Overview. https://www.cardia.dopm.uab.edu/cardia-overview/overview-more. *CARDIA Study Overview* (2021)
39. Noonan A. S., Velasco-Mondragon H. E., Wagner F. A.. **Improving the health of African Americans in the USA: An overdue opportunity for social justice**. *Public Health Reviews* (2016) **37**. DOI: 10.1186/s40985-016-0025-4
40. Ogunmoroti O., Osibogun O., Spatz E. S., Okunrintemi V., Mathews L., Ndumele C. E., Michos E. D.. **A systematic review of the bidirectional relationship between depressive symptoms and cardiovascular health**. *Preventive Medicine* (2022) **154**. DOI: 10.1016/j.ypmed.2021.106891
41. Pan A., Keum N., Okereke O. I., Sun Q., Kivimaki M., Rubin R. R., Hu F. B.. **Bidirectional association between depression and metabolic syndrome: A systematic review and meta-analysis of epidemiological studies**. *Diabetes Care* (2012) **35** 1171-1180. DOI: 10.2337/dc11-2055
42. Radloff L. S.. **The CES-D Scale**. *Applied Psychological Measurement* (1977) **1** 385-401. DOI: 10.1177/014662167700100306
43. Shafer A. B.. **Meta-analysis of the factor structures of four depression questionnaires: Beck, CES-D, Hamilton, and Zung**. *Journal of Clinical Psychology* (2006) **62** 123-146. DOI: 10.1002/jclp.20213
44. Sumner J. A., Nishimi K. M., Koenen K. C., Roberts A. L., Kubzansky L. D.. **Posttraumatic stress disorder and inflammation: Untangling issues of bidirectionality**. *Biological Psychiatry* (2020) **87** 885-897. DOI: 10.1016/j.biopsych.2019.11.005
45. Sun K., Liu J., Ning G.. **Active smoking and risk of metabolic syndrome: A meta-analysis of prospective studies**. *PLoS ONE* (2012) **7** 1-9. DOI: 10.1371/journal.pone.0047791
46. Vilagut G., Forero C. G., Barbaglia G., Alonso J.. **Screening for depression in the general population with the center for epidemiologic studies depression (CES-D): A systematic review with meta-analysis**. *PLoS ONE* (2016) **11** 1-17. DOI: 10.1371/journal.pone.0155431
47. Vishram J. K., Borglykke A., Andreasen A. H., Jeppesen J., Ibsen H., Jorgensen T., Palmieri L., Giampaoli S., Donfrancesco C., Kee F., Mancia G., Cesana G., Kuulasmaa K., Salomaa V., Sans S., Ferrieres J., Dallongeville J., Soderberg S., Arveiler D., Wagner A., Tunstall-Pedoe H., Drygas W., Olsen M. H., Project M.. **Impact of age and gender on the prevalence and prognostic importance of the metabolic syndrome and its components in Europeans. The MORGAM Prospective Cohort Project**. *PLoS ONE* (2014) **9** 1-12. DOI: 10.1371/journal.pone.0107294
48. Womack V. Y., De Chavez P. J., Albrecht S. S., Durant N., Loucks E. B., Puterman E., Redmond N., Siddique J., Williams D. R., Carnethon M. R.. **A longitudinal relationship between depressive symptoms and development of metabolic syndrome**. *Psychosomatic Medicine* (2016) **78** 867-873. DOI: 10.1097/psy.0000000000000347
49. Yu S., Guo X., Li G. X., Yang H., Zheng L., Sun Y.. **Metabolic syndrome associated with the onset of depressive symptoms among women but not men in rural Northeast China**. *BMC Psychiatry* (2020) **20** 1-10. DOI: 10.1186/s12888-020-02668-z
|
---
title: Impact of Adrenomedullin on Mitochondrial Respiratory Capacity in Human Adipocyte
authors:
- Yuanlin Dong
- Vidyadharan Alukkal Vipin
- Chellakkan Selvanesan Blesson
- Chandrasekhar Yallampalli
journal: Research Square
year: 2023
pmcid: PMC10029071
doi: 10.21203/rs.3.rs-2600140/v1
license: CC BY 4.0
---
# Impact of Adrenomedullin on Mitochondrial Respiratory Capacity in Human Adipocyte
## Abstract
For metabolic homeostasis adequate mitochondrial function in adipocytes is essential. Our previous observation showed that circulating levels of adrenomedullin (ADM) and mRNA and protein for ADM in omental adipose tissue were higher in patients with gestational diabetes mellitus (GDM) compared with normal pregnancy, and these alterations are accompanied by glucose and lipid metabolic dysregulation, but the impact of ADM on mitochondrial biogenesis and respiration in human adipocyte remain elusive. In this study we demonstrated that: [1] Increasing doses of glucose and ADM inhibit human adipocyte mRNA expressions of mitochondrial DNA (mtDNA)-encoded subunits of electron transport chain (ETC), including nicotinamide adenine dinucleotide dehydrogenase (ND) 1 and 2, cytochrome (CYT) b, as well as ATPase 6; [2] ADM significantly increases human adipocyte mitochondrial reactive oxygen species (ROS) generation and this increase is reversed by ADM antagonist, ADM22–52, but does not significantly affect adipocyte mitochondrial contents; [3] Adipocyte basal and maximal oxygen consumption rate (OCR) are dose-dependently suppressed by ADM, and results in impaired mitochondrial respiratory capacity. We conclude that elevatedADM observed in diabetic pregnancy may be involved in glucose and lipid dysregulation through compromising adipocyte mitochondrial function, and blockade of ADM actions in adipocytes may improve GDM-related metabolic complications.
## Introduction
Adipose tissue is not only an energy reservoir for lipid droplets, but also an important endocrine organ, secreting pro-inflammatory cytokines, reactive oxygen species, and adipokines, including adrenomedullin (ADM)1,2. Adipose tissue dysregulation contributes to the pathophysiology of a variety of metabolic disorders, including cardiovascular diseases, obesity, polycystic ovary syndrome (PCOS), and diabetes mellitus3,4 Recent studies in the rat model show that ADM and its receptors are expressed in adipose tissue5, administration of ADM induces hyperglycemia, which can be reversed by an ADM neutralizing antibody6. In humans, plasma ADM concentrations are elevated in obese individuals7 and patients with T2DM8. Our previous studies have shown that both circulating ADM and mRNA and protein of ADM in omental adipose tissue were increased in GDM patients9,10, indicating the involvement of ADM in the impaired metabolic homeostasis. However, the underlying mechanisms of ADM contributing to GDM-related metabolic dysregulation remain unclear.
Emerging evidence indicated that mitochondria play central roles in energy homeostasis, metabolism, pathway signaling, and cellular apoptosis11, and mitochondrial dysfunction in adipocytes is tightly related with insulin resistance in obese and diabetic individuals12,13. Patients with type 2 diabetes show that mitochondrial functions are declined, which are associated with a reduction of both mitochondrial DNA (mtDNA) copy numbers and key factors regulating mitochondrial biogenesis14. Impaired mitochondrial biogenesis and functions in adipose tissue are also observed in animal models of type 2 diabetes15. Furthermore, a decrease in mitochondrial mass and function has been found in adipose tissue of obese ob/ob mice16. However, the impact of excessive ADM found in GDM patients on adipocyte mitochondrial function remains unclear. In the present study, we hypothesized that excessive ADM may induce adipocyte mitochondrial respiratory dysfunction and contributes to adipocyte-related metabolic complications. To address this hypothesis, we studied the impact of ADM on mRNA expression of mitochondrial DNA (mtDNA)-encoded subunits of electron transport chain, mitochondrial content, reactive oxygen species (ROS) generation, and mitochondrial respiratory capacity in human adipocytes.
## Glucose suppresses mRNA expression for mtDNA-encoded subunits ND1 and ND2 in electron transport chains.
To investigate the effect of increasing doses of glucose on mtDNA-encoded subunits in the electron transport chain in human adipocytes, we measured the gene expression for mtDNA-encoded subunits of the electron transport chain using q-PCR. As shown in Fig. 1, the expressions of ND1 were significantly inhibited in the adipocytes by glucose in a dose dependent manner ($P \leq 0.01$), and ND2 was down regulated at higher dose of glucose ($P \leq 0.05$). However, the alterations in mRNA expression for CYTb, CO1, and ATPase 6 were not significant compared with controls ($P \leq 0.05$). These results indicate that increased glucose concentration, mimicking the hyperglycemia environment in GDM patients, is associated with reduced mtDNA-encoded subunits of the electron transport chain in human adipocytes.
## ADM inhibits mRNA expression for mtDNA-encoded subunits ND1, ND2, CYTb, and ATPase in the adipocytes.
To investigate the impact of ADM on mtDNA-encoded subunits in human adipocytes, we treated the cells with increasing dose of ADM for 24 hours. As shown in Fig. 2, mRNA levels for ND1, ND2, CYTb, and ATPase ($P \leq 0.05$ or $P \leq 0.01$), but not CO1 ($P \leq 0.05$), were inhibited by ADM in the adipocytes. Moreover, coincubation with ADM antagonist, ADM22–52, blocked the effects of ADM indicating the specificity of ADM effects. This finding suggests that excessive ADM expression seen in adipose tissue from GDM patients may induce a decrease in mRNA levels for mtDNA-encoded subunits of the electron transport chain.
## ADM does not significantly affect adipocyte mitochondrial content.
The number of copies of mtDNA per cell is a general marker of mitochondrial fitness. To provide further evidence of the mitochondria regulation by ADM in adipocytes, we determined the mitochondrial content by staining the cells with Mitochondrial-specific fluorescence dye MitoTracker green. As shown in Fig. 3, adipocytes treated with ADM were weakly stained with MitoTracker compared with controls, and ADM antagonist ADM22–52 partially reverse the reduced staining, but no significant differences were detected between group, implying that ADM does not significantly affect mitochondrial content in human adipocytes.
## ADM induces ROS generation in adipocytes.
ROS, the by-products of mitochondrial respiration, are produced normally by the adipocytes, and overproduction of the ROS may damage various components in the cells. To evaluate the effect of ADM on ROS generation, we measured ROS levels in adipocytes by MitoTracker Red, a mitochondrion-specific dye. As shown in Fig. 4, ADM stimulates ROS production in adipocytes as compared to controls, and this increase was reversed by ADM antagonist, ADM22–52. These results suggest that ADM increases adipocyte ROS generation, and this increase is specific to ADM, which may contribute to adipocyte systemic mitochondrial dysfunction.
## ADM disrupts mitochondrial respiratory capacity.
To further test the effects of ADM on mitochondrial function, we assessed mitochondrial respiratory capacity using the Seahorse Biosciences XF-96 Analyzer. We used a typical bioenergetic profile, involved in a four-step analysis: [1] basal OCR, adipocytes were incubated in normal medium; [2] ATP synthesis turnover, oligomycin (2.0 mM) was supplemented to the medium to inhibit ATP synthase; [3] maximal mitochondrial respiratory capacity, cells were motivated with FCCP (1.0 mM); and [4] non-mitochondrial respiration, rotenone (1.0 mM) was introduced to inhibit complex I. As shown in Fig. 5, increasing doses of ADM (−10 to −8M) exerted negative effects on the mitochondrial respiratory function of human adipocytes. Specifically, the basal mitochondrial respiration was inhibited by ADM starting from concentration of −10 M, and further reduced by ADM at a concentration of −9M and − 8M ($P \leq 0.01$). Moreover, ADM also inhibited the maximal mitochondrial and non-mitochondrial OCR of adipocytes in a dose-dependent manner ($P \leq 0.01$). However, there was no significant changes in the ATP linked OCR of the cells treated with ADM ($P \leq 0.05$). These results indicate that ADM significantly suppresses mitochondrial respiratory function, denoting reduced ability of mitochondria to respond to increased energy requirements, but ATP linked OCR was not significantly affected.
## Discussion
Mitochondria are important organelles participating in the regulation of numerous cellular activities, including thermogenesis, ROS generation, redox and Ca2+ homeostasis, and cell apoptosis. Mitochondrial dysfunction in adipocytes can affect whole-body energy homeostasis as well as insulin resistance17. In the present study, we performed a comprehensive set of experiments to test a hypothesis that ADM impaired mitochondrial function in human adipocytes. Our data revealed that both glucose and ADM inhibit human adipocytes mRNA expressions of mtDNA-encoded subunits of electron transport chain, including ND 1 and 2, CYTb, and ATPase 6. Furthermore, ADM stimulates mitochondrial ROS generation, but does not affect the mitochondrial contents in the adipocytes. In addition. ADM suppresses adipocyte basal and maximal oxygen consumption rate in a dose-dependent manner, leading to compromised mitochondrial respiratory capacity. Therefore, excessive ADM seen in GDM patients may contribute to lipid metabolic dysregulation through disrupting adipocyte mitochondrial function. Thus, these data bring new insights into GDM-related adipose tissue dysfunction.
Mammalian mitochondria possess their own genome, which consists of a single, circular double-stranded mtDNA molecule18. mtDNA encodes essential components of complexes of the electron transport chain, including [1] Complex I: seven nicotinamide adenine dinucleotide dehydrogenase subunits involved (ND1, ND2, ND3, ND4L, ND5 and ND6) of NADH dehydrogenase; [2] Complex III: the cytochrome b (CYTb) subunit of the ubiquinol-cytochrome c oxidoreductase involved; [3] Complex IV: three subunits (COI, COII and COIII) of cytochrome c oxidase involved, and [4] Complex V: the ATPase 6 and 8 subunits, which are necessary for protein production within the mitochondria19. It has been reported that decreased expression of the genes in complexes I and IV leads to adipocyte dysfunction20, and reduced mRNA for complex I, III and V can induce triglyceride (TG) accumulation in 3T3-L1 cells21. Present study demonstrated that ADM dose-dependently inhibited human adipocyte mRNA expressions of mtDNA-encoded subunits of electron transport chain, including ND1 and 2, CYTb, and ATPase 6, suggesting the negative impact of ADM on adipocyte mitochondrial function. Considering mitochondrial mtDNA impairment is associated with reduced fatty acid-oxidation and increased cytosolic free fatty acid accumulation in adipocytes that alters glucose uptake22, our results may reveal a novel molecular mechanism linking adipocyte-ADM and mitochondrial dysfunction in the pathogenesis of diabetic pregnancy.
The new mitochondria generation involves complete replication of mitochondrial DNA. Mitochondrial biogenesis is driven by the transcriptional activator of NRF-1, NRF-2, PGC-1α, which is activated by various pathways such as receptor tyrosine kinases, natriuretic peptide receptors and nitric oxide through the generation of cGMP23. It has been reported that both mitochondrial mass and respiratory chain activity are decreased in adipocytes in diabetic mice15, implying impaired mitochondrial biogenesis by glucose dysregulation. Our data showed that ADM does not significantly alter the content of mitochondria in human adipocytes, thus the impaired mitochondrial function in the adipocyte is unlikely resulted from a lower number of mitochondria mass, at least in our present study.
Mitochondrial ROS are generated by the respiratory chain, and thus indirectly associated with the status of mitochondrial activity. Evidence have proven that low concentrations of ROS functions as secondary messengers, playing a role in cell signaling inside and outside mitochondria16. However, excessive mitochondrial ROS generation in adipocytes by chronic oxidative stress may contribute to the development of insulin resistance and the progression of various metabolic diseases, including GDM. Particularly, increased ROS production in 3T3-L1 preadipocytes has been demonstrated to be associated with inhibited cell proliferation24, and elevated intracellular ROS levels impair adipocyte function, which is accompanied by glucose intolerance and insulin resistance25. GDM is associated with higher ROS generation compared with normal pregnancies26. In the present study, we used cultured adipocytes to assess the effects of ADM on ROS generation. The MitoSOX™ Red staining, the superoxide indicator of mitochondria, were significantly enhanced in ADM treated adipocytes compared with controls, indicating that oxidative stress was induced by ADM in adipocytes, thus, increased circulating ADM in GDM patients may contribute to the metabolic complications, including glucose intolerance and insulin resistance.
Mitochondrial respiratory capacity is vital to the functionality and viability of the adipocytes, and cellular oxygen consumption is a fundamental indicator of mitochondrial function. Specifically, mitochondrial basic respiration includes coupled as well as uncoupled mitochondrial oxygen consumption27. The coupled oxygen consumption produces ATP, and the uncoupled oxygen consumption forms ROS, which is involved in multiple physiological and pathological activities. In addition, the maximal OCR is an indicator which represents the ability of mitochondria to reserve energy28, and mitochondrial stress often leads to excessive ROS generation and mitochondrial dysfunction. The present study revealed that ADM induces mitochondrial stress by inhibiting basal and maximal mitochondrial OCR in a dose-dependent manner. Accordingly, non-mitochondrial respiratory capacity, roughly displaying adaptation to metabolic changes, was also reduced by ADM in adipocytes. On the contrary, no significant differences in ATP linked OCR were detected between groups, indicating that oligomycin addition had no significant impact on ADM treated adipocytes. It has been reported that mitochondrial ATP is generated from reduced equivalent electron carrier nicotinamide adenine dinucleotide (NADH or NAD + H+) (complex I, NADH dehydrogenase) and reduced flavin adenine dinucleotide (FADH2) (complex II, succinate dehydrogenase), and finally through oxidative phosphorylation at the F0F1-ATP synthase (complex V)27. In the present study, although we found the expression of ND1, ND2, CYTb, and ATPase 6 were inhibited by ADM in adipocytes, but the role of other parts of Complex I and V in the balance of the ATP production and consumption remains unclear. Thus, further study focusing on the mRNA, proteins, and activity for mtDNA-encoded subunits of electron transport chain, including but not limited to ND3, ND4, ND4L, ND5 and ND6 of NADH dehydrogenase and ATPase 6 and 8, are apparently warranted.
In conclusion, our findings provide evidence of ADM treatment resulted in mitochondrial dysfunction in human adipocytes, and excessive ADM found in GDM patients may act as a circulating factor linking energy generation and consumption and contribute to impaired adipocyte mitochondrial metabolism in diabetic pregnancy. Therefore, the new concept that ADM regulates mitochondrial functions may have therapeutic potential for the treatment of important pathophysiological conditions related to glucose/lipid metabolism.
## Limitation:
The influence of ADM on the activity of complex I, III, IV, and V in the mtDNA needs to be clarified. In addition, the specific downstream signaling underlying ADM effects on mitochondrial biogenesis and the ex vivo effects of ADM and its antagonist on the mitochondrial biogenesis and function in adipose tissue from GDM patients remain to be explored.
## Human pre-adipocyte culture
Primary normal human pre-adipocytes (ATCC PCS-210–010, American Type Culture Collection, Manassas, VA, USA) were differentiated into mature adipocytes in wells of 24-well-plates containing adipocyte differentiation medium (Cell Applications, Inc. San Diego, CA) in a $5\%$ CO2 atmosphere at 370 C29. These cells can be expanded in an undifferentiated state for future differentiation to mature adipocytes and show higher efficiency of adipogenesis compared to mesenchymal stem cells. In this study, the cells were cultured in adipocyte differentiation medium with increasing doses of glucose (8.4mM to 19.3mM, Sigma-Aldrich, St. Louis, MO), or ADM (1×10− 10M to 1×10− 8M, Sigma-Aldrich) for 24 hours. Total RNA was isolated from the cells using TRIzol (Life Technologies, Grand Island, NY) and RT was performed for further Quantitative Real-time-PCR analysis.
## The mRNA expression for mitochondrial DNA (mtDNA)-encoded subunits of the electron transport chain
Quantitative Real-time-PCR was performed using Taq universal SYBR Green Supermix (Bio-Rad). PCR primers used for amplification for mitochondrial DNA (mtDNA)-encoded subunits of the electron transport chain were purchased from Integrated DNA Technologies (IDT) and the primer sequences were listed in Table 1. Amplification of 18S and GAPDH served as endogenous controls. PCR conditions for SYBR *Green* gene expression were 10 min at 95°C for 1 cycle, then 15 sec at 94°C, 30 sec at 60°C and 15 sec at 72°C for 39 cycles. All experiments were performed in triplicate. The average CT value was used to calculate the results using the 2–ΔΔCT method and expressed in fold increase/decrease of the gene of interest.
## Measurement of mitochondrial contents
Human pre-adipocytes were seeded onto 8 chamber glass slides containing adipocyte differentiation medium, differentiated adipocytes were treated with ADM (1×10− 8M) with or without ADM22–52 (1×10−7M) for 48 h. The cells were then loaded with Mitochondrial-specific fluorescence dye MitoTracker green (100 nM, Invitrogen) for 45 min at 37°C. The slides were then mounted with mounting-medium containing 4′, 6-diamidino-2-phenylindole (DAPI; Vector Laboratories Inc., Burlingame, CA) and viewed under an Olympus BX51 microscope. The intensity of the immunofluorescence was measured by using CellSence software (Olympus Scientific, Walthan MA, USA), and the relative densities of the immunofluorescence to the number of nuclei were calculated and compared between groups.
## Assessment of mitochondrial reactive oxygen species (ROS)
Human pre-adipocytes were seeded onto 8 chamber glass slide containing adipocyte differentiation medium. Differentiated adipocytes were treated with ADM (1×10− 8M) with or without ADM22–52 (1×10− 7M) for 48 h. The cells were then loaded with MitoSOX red probe (5 μM, Invitrogen) for 10 min at 37°C. The slides were then mounted with mounting-medium containing 4′, 6-diamidino-2-phenylindole (DAPI; Vector Laboratories Inc., Burlingame, CA) and viewed under an Olympus BX51 microscope. The intensity of the immunofluorescence was measured by using CellSence software (Olympus Scientific, Walthan MA, USA), and the relative densities of the immunofluorescence to the number of nuclei were calculated and compared between groups.
## Determination of the mitochondrial oxygen consumption rate (OCR)
Ten thousand preadipocytes per well were seeded in 96-well XF assay plates containing adipocyte growth medium and differentiated into mature adipocytes in the presence or absence of ADM (1×10− 10M to 1×10− 8M). The cells were then subjected to real-time measurements of oxygen consumption rate (OCR) using Seahorse Biosciences XF-96 Analyzer (Agilent, CA). For mitochondrial stress tests, mitochondrial complex inhibitors were injected to all the following treatments sequentially in the following order: oligomycine (1.5 μM), carbonyl cyanide-ptrifluoromethoxyphenylhydrazone (FCCP; 0.5 μM), antimycine A/rotenone (0.5 μM each), and 3 readings were taken after each injection. OCR was automatically recorded by XF-96 software provided by the manufacturer. Calculations of proton leak, coupling efficiency, and maximal respiration were performed according to the manufacturer’s instructions.
## Statistics
All data were presented as mean ± SEM. Data were calculated and analyzed by GraphPad Prism (La Jolla, CA). Repeated measures ANOVA (treatment and time as factors) with a Bonferroni post hoc test were used for comparisons between groups. mRNA and protein expression were compared between control and treatment groups using unpaired Student t test. Statistical significance was defined as $p \leq 0.05.$
## Grant Support:
This work was supported by National Institutes of Health (Grants HD091503 and HL58144 to C.Y.)
## Funding:
This work was supported by National Institutes of Health (Grants DK114689 and HL58144 to C.Y.)
## Data Availability:
The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.
## References
1. Recinella L.. **Adipokines: New Potential Therapeutic Target for Obesity and Metabolic, Rheumatic, and Cardiovascular Diseases**. *Front Physiol* (2020) **11** 578966. DOI: 10.3389/fphys.2020.578966
2. McGown C., Birerdinc A., Younossi Z. M.. **Adipose tissue as an endocrine organ**. *Clin Liver Dis* (2014) **18** 41-58. DOI: 10.1016/j.cld.2013.09.012
3. Chait A., den Hartigh L. J.. **Adipose Tissue Distribution, Inflammation and Its Metabolic Consequences, Including Diabetes and Cardiovascular Disease**. *Front Cardiovasc Med* (2020) **7** 22. DOI: 10.3389/fcvm.2020.00022
4. Mintziori G.. **The effect of excess body fat on female and male reproduction**. *Metabolism* (2020) **107** 154193. DOI: 10.1016/j.metabol.2020.154193
5. Fukai N.. **Concomitant expression of adrenomedullin and its receptor components in rat adipose tissues**. *Am J Physiol Endocrinol Metab* (2005) **288** E56-62. DOI: 10.1152/ajpendo.00586.2003
6. Martinez A.. **Is adrenomedullin a causal agent in some cases of type 2 diabetes?**. *Peptides* (1999) **20** 1471-1478. DOI: 10.1016/s0196-9781(99)00158-8
7. Kato J.. **Plasma levels of adrenomedullin and atrial and brain natriuretic peptides in the general population: their relations to age and pulse pressure**. *Hypertens Res* (2002) **25** 887-892. DOI: 10.1291/hypres.25.887
8. Hayashi M.. **Plasma adrenomedullin in diabetes**. *Lancet* (1997) **350** 1449-1450. DOI: 10.1016/s0140-6736(05)64211-0
9. Dong Y., Betancourt A., Belfort M., Yallampalli C.. **Targeting Adrenomedullin to Improve Lipid Homeostasis in Diabetic Pregnancies**. *J Clin Endocrinol Metab* (2017) **102** 3425-3436. DOI: 10.1210/jc.2017-00920
10. Dong Y.. **Circulating Adrenomedullin Is Elevated in Gestational Diabetes and Its Role in Impaired Insulin Production by beta-Cells**. *J Clin Endocrinol Metab* (2019) **104** 697-706. DOI: 10.1210/jc.2018-01119
11. Taylor R.. **Causation of type 2 diabetes – the Gordian knot unravels**. *N Engl J Med* (2004) **350** 639-641. DOI: 10.1056/NEJMp038239
12. Hartman M. L.. **Relation of mitochondrial oxygen consumption in peripheral blood mononuclear cells to vascular function in type 2 diabetes mellitus**. *Vasc Med* (2014) **19** 67-74. DOI: 10.1177/1358863X14521315
13. Schrauwen P., Hesselink M. K.. **Oxidative capacity, lipotoxicity, and mitochondrial damage in type 2 diabetes**. *Diabetes* (2004) **53** 1412-1417. DOI: 10.2337/diabetes.53.6.1412
14. Szendroedi J.. **Impaired mitochondrial function and insulin resistance of skeletal muscle in mitochondrial diabetes**. *Diabetes Care* (2009) **32** 677-679. DOI: 10.2337/dc08-2078
15. Choo H. J.. **Mitochondria are impaired in the adipocytes of type 2 diabetic mice**. *Diabetologia* (2006) **49** 784-791. DOI: 10.1007/s00125-006-0170-2
16. Wilson-Fritch L.. **Mitochondrial biogenesis and remodeling during adipogenesis and in response to the insulin sensitizer rosiglitazone**. *Mol Cell Biol* (2003) **23** 1085-1094. DOI: 10.1128/MCB.23.3.1085-1094.2003
17. Newton B. W.. **Proteomic analysis of 3T3-L1 adipocyte mitochondria during differentiation and enlargement**. *J Proteome Res* (2011) **10** 4692-4702. DOI: 10.1021/pr200491h
18. Anderson S.. **Sequence and organization of the human mitochondrial genome**. *Nature* (1981) **290** 457-465. DOI: 10.1038/290457a0
19. Bohr V. A.. **Repair of oxidative DNA damage in nuclear and mitochondrial DNA, and some changes with aging in mammalian cells**. *Free Radic Biol Med* (2002) **32** 804-812. DOI: 10.1016/s0891-5849(02)00787-6
20. Vernochet C.. **Adipose-specific deletion of TFAM increases mitochondrial oxidation and protects mice against obesity and insulin resistance**. *Cell Metab* (2012) **16** 765-776. DOI: 10.1016/j.cmet.2012.10.016
21. Vankoningsloo S.. **Mitochondrial dysfunction induces triglyceride accumulation in 3T3-L1 cells: role of fatty acid beta-oxidation and glucose**. *J Lipid Res* (2005) **46** 1133-1149. DOI: 10.1194/jlr.M400464-JLR200
22. Lowell B. B., Shulman G. I.. **Mitochondrial dysfunction and type 2 diabetes**. *Science* (2005) **307** 384-387. DOI: 10.1126/science.1104343
23. Whitaker R. M., Corum D., Beeson C. C., Schnellmann R. G.. **Mitochondrial Biogenesis as a Pharmacological Target: A New Approach to Acute and Chronic Diseases**. *Annu Rev Pharmacol Toxicol* (2016) **56** 229-249. DOI: 10.1146/annurev-pharmtox-010715-103155
24. Carriere A., Fernandez Y., Rigoulet M., Penicaud L., Casteilla L.. **Inhibition of preadipocyte proliferation by mitochondrial reactive oxygen species**. *FEBS Lett* (2003) **550** 163-167. DOI: 10.1016/s0014-5793(03)00862-7
25. Wang C. H., Wang C. C., Huang H. C., Wei Y. H.. **Mitochondrial dysfunction leads to impairment of insulin sensitivity and adiponectin secretion in adipocytes**. *FEBS J* (2013) **280** 1039-1050. DOI: 10.1111/febs.12096
26. Cheang W. S.. **Metformin protects endothelial function in diet-induced obese mice by inhibition of endoplasmic reticulum stress through 5’ adenosine monophosphate-activated protein kinase-peroxisome proliferator-activated receptor delta pathway**. *Arterioscler Thromb Vasc Biol* (2014) **34** 830-836. DOI: 10.1161/ATVBAHA.113.301938
27. Junge W., Nelson N.. **ATP synthase**. *Annu Rev Biochem* (2015) **84** 631-657. DOI: 10.1146/annurev-biochem-060614-034124
28. Owen S. C.. **Colloidal drug formulations can explain “bell-shaped” concentration-response curves**. *ACS Chem Biol* (2014) **9** 777-784. DOI: 10.1021/cb4007584
29. Dong Y., Chauhan M., Betancourt A., Belfort M., Yallampalli C.. **Adipose Tissue Inflammation and Adrenomedullin Overexpression Contribute to Lipid Dysregulation in Diabetic Pregnancies**. *J Clin Endocrinol Metab* (2018) **103** 3810-3818. DOI: 10.1210/jc.2018-00905
|
---
title: 'Resolving the soluble-to-toxic transformation of amyloidogenic proteins: A
method to assess intervention by small-molecules'
authors:
- Jyoti Ahlawat
- Daisy L. Wilson
- Ana Carreon
- Mahesh Narayan
journal: Research Square
year: 2023
pmcid: PMC10029074
doi: 10.21203/rs.3.rs-2631727/v1
license: CC BY 4.0
---
# Resolving the soluble-to-toxic transformation of amyloidogenic proteins: A method to assess intervention by small-molecules
## Abstract
The soluble-to-toxic transformation of intrinsically disordered amyloidogenic proteins such as amyloid beta (Aβ), α-synuclein, mutant Huntingtin Protein (mHTT) and islet amyloid polypeptide (IAPP) among others is associated with disorders such as Alzheimer’s disease (AD), Parkinson’s disease (PD), Huntington’s disease (HD) and Type 2 Diabetes (T2D), respectively. Conversely, the dissolution of mature fibrils and toxic amyloidogenic intermediates including oligomers remains the holy grail in the treatment of neurodegenerative disorders. Yet, methods to effectively, and quantitatively, report on the interconversion between amyloid monomers, oligomers and mature fibrils fall short. For the first time, we describe the use of gel electrophoresis to address the transformation between soluble monomeric amyloid proteins and mature amyloid fibrils. The technique permits rapid, inexpensive and quantitative assessment of the fraction of amyloid monomers that form intermediates and mature fibrils. In addition, the method facilitates the screening of small molecules that disintegrate oligomers and fibrils into monomers or retain amyloid proteins in their monomeric forms. Importantly, our methodological advance diminishes major existing barriers associated with existing (alternative) techniques to evaluate fibril formation and intervention.
## Introduction
A hallmark feature of neurodegenerative disorders such as AD, PD, HD and T2D is the soluble-to-toxic conversion of disease-associated prion-like amyloidogenic proteins such as Aβ, α-synuclein, mHTT, and IAPP, respectively (1–7). The formation of mature fibrils from their soluble, monomeric counterparts is often the “end-point” of the amyloid-forming (amyloidogenic) trajectory. Fibril formation is essentially irreversible. Mature fibrils, which are rich in β-sheet content, are insoluble and therefore not easily amenable to structural studies.
Amyloid monomers are converted to mature fibrils via a sequential process that first results in the formation of dimers and/or neurotoxic oligomers (Scheme 1; 7).
Oligomers form proto-fibrils prior to the formation mature fibrils, which is a terminal process as aforementioned. A comparison of the kinetics of monomer consumption relative the fibril formation is important. A difference in the rate of monomer consumption relative to fibril formation suggests the presence of intermediates. A lag in the time to form mature fibrils is indicative of kinetically-trapped conformations [7]. Quantifying the loss of monomers is essential for a detailed biophysical understanding of the amyloidogenic trajectory. After all, it is the most experimentally tractable of all species along the amyloid-fibril-forming pathway. The rate of monomer consumption informs us whether the ambient conditions are biased towards retaining the monomeric conformation or towards fibril formation. Measurement of the rate of monomer loss can be used to fine-tune ambient (fibril-forming) conditions either to intervene in the fibrillation or to promote it (say, for biophysical studies) [8]. Comparison of the rate of monomer consumption with that of mature fibril formation facilitates the generation of a kinetic and quasi-structural roadmap of the process(es) by which soluble amyloids form insoluble aggregates.
Conversion of mature fibrils to their soluble monomeric counterparts is also indispensable for qualitative and quantitative evaluation of the efficacy by which small molecules may intervene (therapeutically or prophylactically) in amyloid-forming trajectories. Molecules such as tanshinone, brazilin and other aromatics along with specific carbon nano materials known as carbon quantum dots and graphene quantum dots have been instrumental in passivating amyloid monomers, remodeling oligomers, and dissolving mature fibrils (9–14). W.r.t. small molecule intervention, the ability to revert all non-monomeric intermediates including mature fibrils, to their soluble monomeric counterpart is key. Also critical is the ability to localize where along the fibril-forming trajectory that a small molecule intervenes is important for further advancing the candidacy of the said molecule [7].
Existing techniques to identify fibrils include dynamic light-scattering (DLS), fluorescence spectroscopy, advanced microscopy (AFM, TEM, HR-TEM, etc.), x-ray fiber diffraction, solid-state NMR, and EPR among others (15–20). While each technique offers specific advantages towards the detection of fibrils, they also require equipment that is not easily accessible, is expensive, and/or requires extensive sample preparation. Furthermore, the quantified conversion of mature fibrils to monomers by small-molecules is not easily realized using the aforementioned techniques.
Here, we demonstrate the use of gel electrophoresis to determine whether select small molecules revert mature fibrils to their soluble monomeric counterparts.
The advantages of our method over existing techniques is discussed.
## Gel Electrophoresis 12%
Gels were prepared as described elsewhere [21, 22]. Briefly, for the running buffer, 1650 uL of water, 2000 uL of $30\%$ acrylamide, 1250 uL of 1.5 M Tris (pH 8.8), 50 uL $10\%$ ammonium persulfate and 2ul TEMED was combined in a 15 mL falcon tube and transferred to the slides. Later, the layering was completed using tertiary butanol. The gels were allowed to polymerize for about 20 minutes. The stacking solution containing 1550 uL of water, 250 uL of $30\%$ acrylamide, 190 uL of 1.5 M Tris (pH 6.8), 15 uL ammonium persulfate and 1.5 ul TEMED in a 15 mL falcon tube was introduced into the gel on top of the running gel. The stacking gel was left to polymerize for 15 minutes and then stored at −4 °C until further use (using wet Kim-wipes covered with the aluminum foil).
## Preparation of Lysozyme solutions
2 mg/mL of Hen-Egg White Lysozyme (HEWL; Sigma) solution in freshly made potassium phosphate buffer (20 mM, pH = 6.3, 3M Guanidinium Hydrochloride) was prepared in a 5 mL glass vial and kept in an incubator-shaker at 550 rpm for 6 hours at 58 °C. After 6 hours, (the contents of the glass vial were turbid), mature fibrils were visualized using Transmission Electron Microscopy [7].
## Loading of amyloid samples onto the gel
The aforementioned solution was dialyzed and added into 1.5 mL Eppendorf tubes and centrifuged (12,400 rpm for 15 minutes). The supernatant was collected in 1.5 mL Eppendorf tubes and DI water was added to the pellet and mixed well. 30 uL of the solution (including supernatant and pellet) was then transferred in separate 0.5 mL pre-labelled Eppendorf tubes. Later, 10 uL of 4X loading dye was added to 30 uL of supernatant and pellet solution. Monomeric solution of Lysozyme (2mg/mL) was prepared as a control and 30 uL was mixed with 10 uL of 4X loading dye. The samples were heated at 95 °C for 5 minutes and 20 uL of this solution was then loaded into the wells of the gel. The gel was then run for 85 minutes at 120V and 400 A. For staining-destaining, gels were removed from the glass slides and rinsed with water. Later, the gels were submerged in Coomasie staining solution overnight. The next day, destaining was performed s using 1:1:0.2 ratio of water:methanol: acetic acid. Destaining was repeated thrice for 20 minutes each. After the third destaining wash, the gel was submerged in water to and an image was subsequently obtained using the Invitrogen iBright Imaging system.
## Imaging of HEWL fibrils
For Transmission electron microscopy analysis, samples were suspended in deionized water and sonicated for 5–10 minutes before adsorption to carbon-coated Cu grids (Electron Microscopy Sciences, Hatboro, PA) followed by negative staining with $2.5\%$ uranyl acetate. Excess stain was adsorbed with Whatman #1 filter paper and grids were air dried and viewed in a model H-7650 transmission electron microscope operated at 80 kV (Hitachi High-Technologies, Dallas, TX). Digital images were collected with an AMT XR 60 CCD camera system (Advanced Microscopy Techniques, Woburn, MA).
## Fluorescence assays
Lysozyme samples were aliquoted for analysis after 0, 1, 2, 3, and 4 hours of incubation. Thioflavin T fluorescence (20 μM) was used to determine the fibril content of each sample in a DM45 Olis Spectrofluorometer using 450 nm and 480 nm as excitation and emission wavelengths, respectively.
## Data Analysis
The obtained images of the gel using the iBright imaging system were analyzed using the Image J software. The data obtained from Image J were transferred to Origin Pro software and mean and standard deviation values are calculated for each band. The bar graph is plotted against Integrated Density vs Sample name.
## Results
Figure 1A is a representative TEM image of mature HEWL fibrils. The fibrils are needle-form and well-delineated in nature. The mature fibrils appear to be interspersed with smaller, potentially, proto-fibrillary aggregates. The data are in good agreement with previous literature [23]. 1B shows the increase in fluorescence emission that results when ThT is added (@ 20s) to a solution containing mature fibrils (black curve). The sharp and rapid increase in fluorescence intensity upon introduction of the fluorophore is indicative of ThT binding to fibrils [7, 23]. The plateauing of the curve suggests that all fibril is either ThT bound or that there is no free ThT in solution even though there may be unbound fibrils. By contrast, the introduction of ThT to monomeric lysozyme (red curve) did not elicit any increase in fluorescence as anticipated.
We determined whether gel electrophoresis could be used to qualitatively discriminate between HEWL mature amyloid fibrils and its monomeric counterpart. Figure 2A is an image of a PAGE experiment where HEWL monomers (2: 1 a and b), the supernatant (2A: 2 a and b) from a centrifuged solution containing mature HEWL fibrils and a resuspended HEWL fibril pellet (2A: 3 a and b) were loaded onto the gel. The location of the bands correspond to the molecular weight of monomeric HEWL. Inspection of the band intensities reveals that compared to the sample exclusively containing HEWL monomers (2A: 1 a and b), there is a decrease in the intensity of HEWL monomers in sample (2A: 2 a and b) and a further attrition in its concentration when sampled from the fibril pellet (2A: 3 a and b). The mature fibrils do not enter the gel due to size-exclusion.
Figure 2B shows quantified results from the aforementioned experiment. Statistical significance was found between samples 2B:1 and 2B:2 and samples 2B:1 and 2B:3 indicating that PAGE can be used to quantify the soluble-to-fibril transformation of amyloid-fibril-forming proteins.
We tested whether small molecules and carbon nano materials revert HEWL fibrils to their soluble monomers. Dimethyl sulfoxide (DMSO) is known to dissociate amyloid fibrils [24, 25]. Figure 3A shows an increase in the concentration of HEWL monomer, relative to untreated fibrils, when mature HEWL fibrils are exposed to DMSO. Furthermore, the difference in monomeric HEWL concentrations between DMSO-treated fibrils and untreated fibrils is statistically significant. The data indicate that the DMSO-driven reconversion of mature fibrils to their monomeric counterpart can easily be detected and quantified use gel electrophoresis. The (statistically significant) difference in monomeric HEWL concentration between the monomer control and the DMSO-treated fibrils is also notable. The fraction of monomer released from DMSO-treated HEWL fibrils reflects the small-molecule-driven fibril-to-soluble reconversion (at the small-molecule concentration). In principle, a small-molecule dose-response curve can be constructed to screen and rank candidate molecules.
Figure 3B shows results from HEWL fibril exposure to carbon quantum dots (CQD1: citric; CQD2:gelatinized carbon). Although there appears to be a CQD-dependent increase in soluble monomers relative to the untreated fibrils, the results were not statistically significant at the CQD dose used.
## Discussion
The soluble-to-toxic conversion of amyloid proteins such as Aβ, α-synuclein, mHTT among others is a critical milestone in the onset and pathogenesis of amyloid-specific neurodegeneraive disorders. Efforts to develop an understanding of this biophysical transformation are driven by spectroscopic and immunohistochemical tools. Nevertheless, access to instruments such as solid-state NMR, microscopes (TEM, HR-TEM, SEM, AFM), ATR-IR, DLS instruments and biochemical kits precludes routine studies of the process for many laboratories and investigators.
Even if high-resolution microscopes are accessible, extensive sample preparation protocols, analyses times and availability of very specific technical/instrumentation expertise are barriers that still need to be overcome. Finally, and critically, higher-resolution structural techniques are not amenable to quantification and kinetics measurements. As previously noted, quantification of oligomers and fibrils formed from soluble monomers and, perhaphs more importantly, the reverse process is important for advancing biomedical intervention. The in vitro screening of small-molecules that intervene in amyloidogenesis precedes testing in preclinical models.
Optical methods such as DLS or fluorescence using ThT or Congo red to identify fibrils are frequently confounded by interference from small-molecule fluorescence [26]. Others techniques such as solid-state NMR, are not amenable to easy use, lack access, and fail to satisfactorily quantify the interconversion between the monomeric amyloid, its intermediates and the mature fibril. Often, necessary sample preparation conditions do not recapitulate solution conditions.
Through several inroads, the method described here reduce barriers towards the study of amyloidogenesis which has traditionally involved elaborate sample preparation, mounting of “dried” samples, expensive instrumentation and protracted sample analyses times [16, 17]. Even though the technique is chemically and structurally “low-resolution” in nature, it provides a rapid, facile and inexpensive mechanism by which to quantify the loss of monomers (via their conversion to dimers, oligomers, proto-fibrils and fibrils), starting from a known concentration of the amyloid nomer. Importantly, by quantifying the intensity of the bands on the gel, it permits the user to build a kinetic profile of the consumption of monomers, formation of dimers, oligomers and finally the transformation of the amyloid protein into mature fibrils. From a biomedical perspective, the use of PAGE to establish a quantitative and dose-dependent profile of small-molecule efficiency in dissolving fibrils and oligomeric aggregates to their monomeric counterparts is highly desired.
In conclusion, we demonstrate that a readily existing method and easily accessible appartatus can be used to obtain rich biophysical (kinetic) data about amyloid forming trajectories and the interplay between intermediates therein. Equally importantly, it can be used to screen small-molecules and also determine, via size analysis, where along the trajectory that the small-molecule intervenes. It provides undergdatuates, graduate students and advanced biomedical researchers in an insittituion with a powerful, affordable, facile method, which is already widely availble, to study an important neurodegeneration-associated process.
## References
1. Adamcik J., Mezzenga R.. **Study of amyloid fibrils via atomic force microscopy**. *Current Opinion in Colloid & Interface Science* (2012) **17** 369-376. DOI: 10.1016/j.cocis.2012.08.001
2. Biancalana M., Koide S.. **Molecular mechanism of Thioflavin-T binding to amyloid fibrils**. *Biochimica Et Biophysica Acta (BBA) - Proteins and Proteomics* (2010) **1804** 1405-1412. DOI: 10.1016/j.bbapap.2010.04.001
3. Chan S. W. S., Yau J., Ing C., Liu K., Farber P., Won A., Bhandari V., Kara-Yacoubian N., Seraphim T.V., Chakrabarti N., Kay L. E., Yip C. M., Pomès R., Sharpe S., Houry W. A.. **Mechanism of amyloidogenesis of a bacterial AAA+ chaperone**. *Structure* (2016) **24** 1095-1109. DOI: 10.1016/j.str.2016.05.002
4. Doig A. J., Derreumaux P.. **Inhibition of protein aggregation and amyloid formation by small molecules**. *Current Opinion in Structural Biology* (2015) **30** 50-56. DOI: 10.1016/j.sbi.2014.12.004
5. Fitzpatrick A. W. P., Saibil H. R.. **Cryo-EM of amyloid fibrils and cellular aggregates**. *Current Opinion in Structural Biology* (2019) **58** 34-42. DOI: 10.1016/j.sbi.2019.05.003
6. Guerrero E. D., Lopez-Velazquez A. M., Ahlawat J., Narayan M.. **Carbon quantum dots for treatment of amyloid disorders**. *ACS Applied Nano Materials* (2021) **4** 2423-2433. DOI: 10.1021/acsanm.0c02792
7. Ippel J. H., Olofsson A., Schleucher J., Lundgren E., Wijmenga S. S.. **Probing solvent accessibility of amyloid fibrils by solution NMR spectroscopy**. *Proceedings of the National Academy of Sciences* (2002) **99** 8648-8653. DOI: 10.1073/pnas.132098999
8. Jucker M., Walker L. C.. **Pathogenic protein seeding in alzheimer disease and other neurodegenerative disorders**. *Annals of Neurology* (2011) **70** 532-540. DOI: 10.1002/ana.22615
9. Kim D., Yoo J. M., Hwang H., Lee J., Lee S. H., Yun S. P., Park M. J., Lee M. J., Choi S., Kwon S. H., Lee S., Kwon S.-H., Kim S., Park Y. J., Kinoshita M., Lee Y.-H., Shin S., Paik S. R., Lee S. J., Ko H.S.. **Graphene quantum dots prevent α-synucleinopathy in parkinson’s disease**. *Nature Nanotechnology* (2018) **13** 812-818. DOI: 10.1038/s41565-018-0179-y
10. LAEMMLI U. K.. **Cleavage of structural proteins during the Assembly of the head of bacteriophage T4**. *Nature* (1970) **227** 680-685. DOI: 10.1038/227680a0
11. Loksztejn A., Dzwolak W.. **Noncooperative dimethyl sulfoxide-induced dissection of insulin fibrils: Toward soluble building blocks of amyloid**. *Biochemistry* (2009) **48** 4846-4851. DOI: 10.1021/bi900394b
12. Mahmoudi M., Akhavan O., Ghavami M., Rezaee F., Ghiasi S. M.. **Graphene oxide strongly inhibits amyloid beta fibrillation**. *Nanoscale* (2012) **4** 7322. DOI: 10.1039/c2nr31657a
13. Markwell J.. **Fundamental laboratory approaches for biochemistry and biotechnology, 2nd edition**. *Biochemistry and Molecular Biology Education* (2009) **37** 317-318. DOI: 10.1002/bmb.20321
14. Morris K. L., Serpell L. C.. **X-ray fibre diffraction studies of amyloid fibrils**. *Methods in Molecular Biology* (2012) 121-135. DOI: 10.1007/978-1-61779-551-0_9
15. Pujols J., Peña-Díaz S., Lázaro D. F., Peccati F., Pinheiro F., González D., Carija A., Navarro S., Conde-Giménez M., García J., Guardiola S., Giralt E., Salvatella X., Sancho J., Sodupe M., Outeiro T. F., Dalfó E., Ventura S.. **Small molecule inhibits α-synuclein aggregation, disrupts amyloid fibrils, and prevents degeneration of dopaminergic neurons**. *Proceedings of the National Academy of Sciences* (2018) **115** 10481-10486. DOI: 10.1073/pnas.18041981151
16. Sengupta U., Nilson A. N., Kayed R.. **The role of amyloid-β oligomers in toxicity, propagation, and immunotherapy**. *EBioMedicine* (2016) **6** 42-49. DOI: 10.1016/j.ebiom.2016.03.035
17. Sengupta U., Nilson A. N., Kayed R.. **The role of amyloid-β oligomers in toxicity, propagation, and immunotherapy**. *EBioMedicine* (2016) **6** 42-49. DOI: 10.1016/j.ebiom.2016.03.035
18. Singh P., Bhat R.. **Binding of noradrenaline to native and intermediate states during the fibrillation of α-synuclein leads to the formation of stable and structured cytotoxic species**. *ACS Chemical Neuroscience* (2019) **10** 2741-2755. DOI: 10.1021/acschemneuro.8b00650
19. Sreenivasan S., Narayan M.. **Learnings from protein folding projected onto amyloid misfolding**. *ACS Chemical Neuroscience* (2019) **10** 3911-3913. DOI: 10.1021/acschemneuro.9b00445
20. Sreeprasad S., Narayan M.. **Nanoscopic portrait of an amyloidogenic pathway visualized through tip-enhanced Raman spectroscopy**. *ACS Chemical Neuroscience* (2019) **10** 3343-3345. DOI: 10.1021/acschemneuro.9b00353
21. Stetefeld J., McKenna S. A., Patel T. R.. **Dynamic light scattering: A practical guide and applications in biomedical sciences**. *Biophysical Reviews* (2016) **8** 409-427. DOI: 10.1007/s12551-016-0218-6
22. Streets A. M., Sourigues Y., Kopito R. R., Melki R., Quake S. R.. **Simultaneous measurement of amyloid fibril formation by dynamic light scattering and fluorescence reveals complex aggregation kinetics**. *PLoS ONE* (2013) **8**. DOI: 10.1371/journal.pone.0054541
23. Subedi L., Gaire B. P.. **Tanshinone IIA: A phytochemical as a promising drug candidate for Neurodegenerative Diseases**. *Pharmacological Research* (2021) **169** 105661. DOI: 10.1016/j.phrs.2021.105661
24. Tycko R.. **Solid-state NMR studies of amyloid fibril structure**. *Annual Review of Physical Chemistry* (2011) **62** 279-299. DOI: 10.1146/annurev-physchem-032210-103539
25. Wang M., Sun Y., Cao X., Peng G., Javed I., Kakinen A., Davis T. P., Lin S., Liu J., Ding F., Ke P. C.. **Graphene quantum dots against human IAPP aggregation and toxicity in vivo**. *Nanoscale* (2018) **10** 19995-20006. DOI: 10.1039/c8nr07180b
26. Watanabe-Nakayama T., Sahoo B. R., Ramamoorthy A., Ono K.. **High-speed atomic force microscopy reveals the structural dynamics of the amyloid-β and amylin aggregation pathways**. *International Journal of Molecular Sciences* (2020) **21** 4287. DOI: 10.3390/ijms21124287
|
---
title: 'Delphi panel to obtain clinical consensus about using long-acting injectable
antipsychotics to treat first-episode and early-phase schizophrenia: Treatment goals
and approaches to functional recovery'
authors:
- Celso Arango
- Andrea Fagiolini
- Philip Gorwood
- John M. Kane
- Sergio Diaz-Mendoza
- Navdeep Sahota
- Christoph U Correll
journal: Research Square
year: 2023
pmcid: PMC10029086
doi: 10.21203/rs.3.rs-2594278/v1
license: CC BY 4.0
---
# Delphi panel to obtain clinical consensus about using long-acting injectable antipsychotics to treat first-episode and early-phase schizophrenia: Treatment goals and approaches to functional recovery
## Abstract
### Background
Schizophrenia is mostly a chronic disorder whose symptoms include psychosis, negative symptoms and cognitive dysfunction. Poor adherence is common and related relapse can impair outcomes. Long-acting injectable antipsychotics (LAIs) may promote treatment adherence and decrease the likelihood of relapse and rehospitalization. Using LAIs in first-episode psychosis (FEP) and early-phase (EP) schizophrenia patients could benefit them, yet LAIs have traditionally been reserved for chronic patients.
### Methods
A three-step modified Delphi panel process was used to obtain expert consensus on using LAIs with FEP and EP schizophrenia patients. A literature review and input from a steering committee of five experts in psychiatry were used to develop statements about patient population, adverse event management, and functional recovery. Recruited Delphi process psychiatrists rated the extent of their agreement with the statements over three rounds (Round 1: paper survey, 1:1 interview; Rounds 2–3: email survey). Analysis rules determined whether a statement progressed to the next round and the level of agreement deemed consensus. Measures of central tendency (mode, mean) and variability (interquartile range) were reported back to help panelists assess their previous responses in the context of those of the overall group.
### Results
The Delphi panelists were 17 psychiatrists experienced in treating schizophrenia with LAIs, practicing in seven countries (France, Italy, US, Germany, Spain, Denmark, UK). Panelists were presented with 73 statements spanning three categories: patient population; medication dosage, management, and adverse events; and functional recovery domains and assessment. Fifty-five statements achieved ≥ $80\%$ agreement (considered consensus). Statements with low agreement ($40\%$−$79\%$) or very low agreement (< $39\%$) concerned initiating dosage in FEP and EP patients, and managing loss of efficacy and breakthrough episodes, reflecting current evidence gaps. The panel emphasized benefits of LAIs in FEP and EP patients, with consensus that LAIs can decrease the risk of relapse, rehospitalization, and functional dysfunction. The panel supported links between these benefits and multidimensional longer-term functional recovery beyond symptomatic remission.
### Conclusions
Findings from this Delphi panel support the use of LAIs in FEP and EP schizophrenia patients regardless of disease severity, number of relapses, or social support status. Gaps in clinician knowledge make generating evidence on using LAIs in FEP and EP patients critical.
## Introduction
Schizophrenia is one of the most burdensome and costly illnesses worldwide due to its common onset in adolescence or early adulthood and its high rate of disability [1]. Symptoms of psychosis, such as hallucinations, delusions, and disordered behavior and speech [2, 3], can affect all areas of the patient’s life, including personal, familial, social, educational, and occupational functioning [3]. Furthermore, family members can suffer as a result of the shifting of care from hospital to families [4].
Antipsychotic medication is the primary modality for the treatment of schizophrenia and should not be delayed [3]. When diagnosis and treatment are delayed, patients are at greater risk for poorer outcomes, such as a lesser response to treatment and discontinuation of it, which can exacerbate illness and increase chances of relapses.[5] Rates of non-adherence may be higher for schizophrenia patients than for patients with other chronic illnesses due to adverse events, a lack of efficacy, challenges in symptom control, poor insight into illness, cognitive dysfunction, substance abuse, stigma, and social drift [6–8]. Poor adherence to antipsychotic medication can lead to relapse and rehospitalization, as well as functional decline [9, 10].
The goals of schizophrenia management are not only symptom reduction in the short-term, but also relapse prevention, maintenance of physical and mental well-being, improved health related quality of life (HRQoL), and full functional recovery [11, 12]. Long-acting injectable antipsychotics (LAIs) have been historically used for schizophrenia patients, who have exhibited non-adherence; however, there is evidence that they are potentially beneficial for all patients with schizophrenia, as they improve treatment adherence, decrease treatment discontinuation, and may reduce the risk of relapse and rehospitalization [9]. In addition, use of LAIs can improve adherence to medications addressing cardiometabolic risk, facilitate functioning, and decrease all-cause as well as specific-cause mortality [5, 6, 12–16]. However, LAIs are frequently underused in current practice, with a highly heterogeneous pattern of use among countries due to perceived stigma as being coercive, service barriers, and lack of clinician knowledge [5, 17, 18].
Because LAIs are typically regarded as a last resort, they are usually reserved for more seriously ill, chronic patients. However, they may benefit first-episode psychosis (FEP) or early-phase (EP) patients, who arguably have the most to gain if treated early, and the most to lose if not, as use of these formulations can prevent relapses, and functional decline [17]. Aside from describing their use in patients with obvious non-adherence, the literature on the clinical benefits of LAIs is sparse, and their benefits in clinical populations, such as early-stage schizophrenia patients or those without multiple hospitalizations, should be explored [19, 20].
To explore some of the underlying aspects of underutilization and to better evaluate the potential benefits of using LAIs in FEP and EP schizophrenia patients, we aimed to obtain expert consensus on the use of LAIs for the treatment of patients with FEP or EP schizophrenia.
## Study design
A Delphi panel is an iterative technique characterized by repeated rounds of controlled feedback until consensus is achieved; in this manner, it allows the systematic collection and aggregation of informed judgments from experts [21, 22]. To meet the study objective, a three-step modified Delphi technique was used consisting of a single 1:1 interview round and two survey rounds (Fig. 1). The first round survey was developed from a targeted literature review and discussion with a steering committee (SC) rather than from an initial open-ended round of statements as would happen with a classical Delphi panel [23]. Five psychiatrists (CA, AF, PG, JK, CC) with expertise in schizophrenia and LAIs formed the SC, providing input into the study design, potential panelists, and survey development.
## Panelist selection
To reach the target sample of 18, as per the recommended sample size of 5–20 individuals [24], the SC were asked to recommend Delphi panelists from the United States (US), United Kingdom (UK) and Europe.
Panelists were screened according to the following inclusion criteria: Practicing psychiatristSeen a minimum of 10 patients with schizophrenia in the last two yearsHas published in the area of schizophrenia Forty-two clinicians were invited by email to participate in the Delphi panel (38 recommendations from the SC and four from Sponsors). Twenty clinicians accepted the invitation, and 17 took part in the Round 1 interview.
A meeting was held with the SC in February 2021 with the following agenda: It was agreed to use the following sections to address the objective with subsections added after initial review of the final dataset (Table.1):
## Procedure
The Delphi panel was conducted between April and November 2021. Potential panelists were invited via email to participate. Written consent was obtained from those invited who agreed to participate, and their Round 1 interviews scheduled. Round 2 and 3 surveys were emailed to panelists, who were typically given 14 days to return their responses (reminders were sent at regular intervals).
## Survey development
During each of the three rounds, panelists were asked to rate the extent to which they agreed with each statement using either Likert scales or binary responses. In Round 1, all statements were presented with a 5-point Likert response scale (1 = Completely Disagree, 2 = Disagree, 3 = Neutral, 4 = Agree, 5 = Completely Agree). Open-ended questions were asked only during the first round to generate further statements for Round 2. Areas for comments under each statement were also incorporated, allowing panelists to provide additional qualitative insights. After the first round, 3-point Likert scale (1 = Disagree, 2 = Neutral, 3 = Agree) or binary response options (Disagree, Agree) were selected by the researchers and approved by the SC depending on the percentage frequencies a statement achieved in the prior round, as per the analysis rules (Table.2).
## Round 1
The Round 1 (April–July 2021) survey was completed by panelists during a 1:1 audio-recorded teleconference interview with a researcher, to facilitate discussion of the statements. Panelists were provided a structured list of 40 statements and 15 open-ended questions to collect both quantitative and qualitative data. Panelists’ qualitative data were used to generate further statements for the Round 2 survey (Fig. 1).
## Rounds 2 and 3
According to the modified Delphi methodology, all open-ended questions and the three statements that did not achieve the minimum response threshold ($41\%$) during Round 1 were removed from the survey. Twenty-two new statements were generated following analysis of panelists’ qualitative responses to the open-ended questions and their comments on pre-existing statements, resulting in a 67-item Round 2 survey.
The Round 2 survey was customized for each panelist, presenting the panelist’s individual responses and the group mode, mean, and interquartile range (IQR) for statements brought forward from Round 1. Round 2 was conducted between August and September 2021. Following quantitative analysis of the Round 2 survey data, nine statements were removed (seven achieving consensus, two not achieving the minimum response threshold) as per the analysis rules, leaving 58 items in the Round 3 survey (Fig. 1). Similar to Round 2, individual and group responses were reported to panelists in the Round 3 survey, which was sent to panelists in October 2021.
## Data analysis and definition of consensus
Qualitative comments and answers from the panelists’ Round 1 interview were reviewed and addressed either to refine existing statements or to create new statements for the Round 2 survey. After each round, quantitative survey responses were extracted for each statement into a Microsoft Excel database and were assigned a score/code (i.e., 1–5, 1–3, or 1, 2) corresponding to the appropriate Likert or binary response scale. The IQR was calculated and used to summarize the extent of the spread of the data. Central tendencies (mean, median, and mode) were calculated to present the group’s responses back to panelists, and percentage response frequencies for each statement were calculated to determine whether consensus had been achieved. The consensus definition was determined a priori with the SC and was later refined and standardized into the following set of analysis rules (Table.2).
## Participation in the survey
Out of the 42 invited,17 clinicians based in seven countries (Italy, France, US, Germany, Spain, UK, Denmark) accepted the invitation to participate. All 17 participated in Rounds 1 and 2, and 16 completed Round 3.
## Overview of results
Overall, panelists were presented with 70 statements over three rounds. A total of 53 statements reached the minimum level of agreement (≥ $80\%$) to be considered a consensus. Two statements reached the minimum level of disagreement (≥ $80\%$) to be considered a dissensus (Fig. 2, Table.3).
## Generic LAI use
In the first section, four statements were presented to the panel on generic LAI use in schizophrenia patients, of which $100\%$ reached consensus. All patients with schizophrenia should be evaluated clinically to be considered for LAI treatment. The treatment goals and clinical management for schizophrenia patients do not differ whether they are treated with LAI or oral antipsychotic (AP) medication. Clinical management for LAIs does not differentiate between FEP and EP patients. When using LAIs, adverse events, stability, and the therapeutic alliance should be taken into consideration.
## Potential exclusions
Eight statements on potential reasons to exclude FEP or EP patients for LAI treatment were presented to the panel in Round 1, and some additional exclusions were added in Round 2 following the qualitative analysis of open-ended questions. The panel reached dissensus ($80\%$ disagree) on $25\%$ of statements in this section: 1) a history of drug abuse and 2) patient obesity was ≥ not considered reasons to exclude FEP or EP patients from LAI treatment. There was a consensus on $25\%$ of statements: 3) LAI treatment can be used for breastfeeding FEP and EP patients if the risk vs benefit is carefully considered, and 4) if they are monitored regularly. There was low agreement ($40\%$−$79\%$ agreement) on $50\%$ of statements: 5) FEP and EP patients should avoid breastfeeding while on oral or LAI treatment. Panelists did not reach consensus (≤ $39\%$ agreement) on whether 6) pregnancy, 7) needle phobia and 8) known cardiac issues were reasons to exclude a FEP or EP schizophrenia patient from LAIs.
## Appropriateness of LAIs for FEP or EP patients
The panel were presented six statements in this section, with $100\%$ reaching consensus.
LAI treatment is appropriate for even those FEP and EP patients who: 1) have a high level of insight into their illness, 2) have good social support, 3) are currently adherent to medication, 4) and have not had a relapse Panelists also reached consensus that 5) FEP or EP patients should be fully informed before switching to LAI from oral AP, and 6) it is appropriate to switch when tolerability and efficacy to the same oral AP are established.
## Adverse event management
The four statements in this subsection were developed from the Round 1 open-ended questions regarding steps to take if an FEP or EP patient has adverse events in the first half of the injection interval (within two weeks of a four-week interval).
Seventy-five percent of the statements reached consensus: the treating psychiatrist should 1) prescribe a counteracting medication depending on the adverse event, 2) observe to see if the adverse event resolves, unless urgent action is required, and 3) decrease the dose at the next interval if the adverse event does not resolve, depending on severity.
There was low agreement that 4) the use of LAIs in FEP or EP patients required any specific monitoring compared with the same or other oral APs.
## Psychotic episodes on LAI
The three statements in this subsection were developed from qualitative analysis in Round 1 on how to manage an FEP or EP patient who has a breakthrough episode while on an LAI.
No statements reached consensus on actions to take when a psychotic episode occurs when the patient is on an LAI: 1) raising the dose without switching to a different drug by adding the oral version of that LAI (depending on the current dose), 2) increasing the dose of the LAI at the next injection interval if the psychosis improves after adding the oral version of the LAI, or 3) switching to a new dose of a different oral AP and commencing with the LAI once efficacy and tolerability are established.
## Loss of efficacy
Three statements on actions to take in the case of the LAI losing efficacy were developed for Round 1 and rephrased for round 2 following feedback from both Delphi panelists and the SC.
Consensus was reached for one statement ($33\%$): 1) if LAI treatment starts to lose efficacy for a FEP or EP schizophrenia patient, the dose can be increased depending on what dose is already being used and after ruling out other factors, such as drug use and medical or psychiatric comorbidities.
Panelists had low agreement that 2) the injection interval can be decreased when loss of efficacy occurs; 3) changing the injection site from gluteal to deltoid injections if LAI treatment starts to lose efficacy for a FEP or EP schizophrenia patient did not reach consensus.
## LAI dosing & intervals for FEP and EP patients
Panelists reviewed four statements designed to address potential concerns physicians may have using LAIs in FEP or EP patients.
The panel reached a consensus on $50\%$ of statements: 1) that FEP and EP patients can be on the equivalent dose of their oral AP providing they are monitored for the first three months and 2) in order to establish tolerability and efficacy when switching from oral to LAIs, panelists agreed that they would use different doses of LAIs in FEP and EP patients.
There was low agreement that 3) FEP and EP patients need to commence LAIs at a lower dose because they are younger and have less medication exposure and that 4) for a FEP or EP schizophrenia patient, a monthly injection interval would be used.
## Long-term treatment goals
Panelists were presented with a definition of functional recovery to ensure they were all aligned. The definition highlighted that it goes beyond symptomatic remission and encompasses multiple aspects of the patient’s life.
All five statements in this section reached consensus: 1) functional recovery is multidimensional, exists on a spectrum, and is not a binary state for a FEP or EP schizophrenia patient; 2) the assessment for functional recovery in FEP or EP schizophrenia patients involves both the use of patient-specific tools (patient-reported outcomes) in addition to 3) conversations with the patient, informants, and the clinical team; 4) functional recovery is a reachable treatment goal for schizophrenia patients, particularly if treated with medication during the FEP or EP; and 5) the patient’s attitude toward treatment and patient psychoeducation are important to achieve functional recovery in FEP or EP schizophrenia patients.
## LAI links to functional recovery
Three statements linking the use of LAIs in FEP and EP patients were presented to the panel over the three rounds, which all achieved consensus.
The use of LAIs results in a better treatment outcome and promotes functional recovery by 1) increasing adherence, 2) reducing treatment burden, and 3) reducing functional decline.
## Functional recovery domain
Nine statements about the dimensions of functional recovery were developed during the first round from the literature and SC input, with $100\%$ reaching consensus. The following dimensions are important to assess when aiming for functional recovery in FEP and EP patients: 1) depression, 2) aggressive behavior, 3) social interaction, 4) family functioning, 5) education and/or employment, 6) sexual functioning, 7) leisure activities, and 8) self-care. In addition to the specific domains, 9) functional recovery should include an element that is meaningful to the FEP or EP schizophrenia patient. Qualitative analysis in Round 1 did not yield additional statements to be used in subsequent rounds.
## Functional recovery assessment
Thirteen statements on aspects of assessing FEP and EP patients for functional recovery were presented to the Delphi panelists, $85\%$ reaching consensus. They were that 1) functional recovery is multidimensional, operating on a spectrum and not a binary state for a patient with FEP or EP schizophrenia; 2) the consideration of specific dimensions (e.g., depression, social functioning) is useful when assessing the extent of functional recovery in FEP or EP schizophrenia patients; 3) as a part of the assessment for functional recovery in FEP or EP schizophrenia, patient-specific tools (e.g., patient-reported outcomes) can be used; 4) conversations with the patient, informants, and the clinical team can be used as a part of the assessment for functional recovery in FEP or EP schizophrenia patients; and 5) a FEP or EP schizophrenia patient can have partial functional recovery if some of the dimensions (e.g., depression, social interaction) are either not improved or only partially improved.
In the first round, the panelists were asked to provide an example of a question they would ask about each dimension to initiate a conversation with the patient about it. In the following round, panelists were asked if they would discuss each dimension of functional recovery at every encounter with the FEP or EP patient. Panelists reached consensus they would ask about 6) depression, 7) aggressive behavior, 8) social interaction, 9) family functioning, 10) education and/or employment, and 11) self-care at every encounter. There was a low level of agreement on $15\%$ of statements; panelists did not reach a consensus on asking about 12) leisure activities or 13) sexual functioning at every encounter with the patient.
## Achieved functional recovery
There was consensus on $85\%$ of the 8 statements in this section: the dimensions 1) depression, 2) aggressive behavior, 3) social interaction, 4) family functioning, 5) education and/or employment, 6) leisure activities, and 7) self-care should be minimally impaired to consider the FEP or EP patient as having achieved functional recovery. Panelists did not meet the minimum threshold for consensus on the statement that 8) sexual functioning should be only minimally impaired to consider the patient to have attained functional recovery.
## Discussion
Delphi panels are widely used in healthcare research and are proven to be a rigorous and feasible way to obtain consensus, allowing for anonymous, expert input through several rounds of controlled feedback. Using a modified Delphi panel, we obtained expert consensus about using LAIs to treat patients with FEP or EP schizophrenia and promote functional recovery, yielding the following main results: 1) 55 statements achieved consensus (i.e., ≥ $80\%$ agreement); 2) Statements with low agreement ($40\%$−$79\%$) or very low agreement (< $39\%$) concerned antipsychotic initiation dosage in FEP and EP patients, and managing loss of efficacy and breakthrough episodes; 3) benefits of LAIs in FEP and EP patients include decreasing the risk of relapse, rehospitalization, and functional dysfunction, supporting the use of LAIs in FEP and EP schizophrenia patients regardless of disease severity, number of relapses, or social support status; and 4) links between these benefits and multidimensional longer-term functional recovery beyond symptomatic remission were supported.
## Current treatment guidelines
Several, but not all, current treatment guidelines suggest that LAIs should be reserved for patients who have had multiple relapses, have exhibited non-adherence to oral or LAIs in the past, have poor social support, and have previous LAI experience [12, 18]. While there is support from some guidelines, others are neutral or silent on LAI use in FEP and EP patients, or indeed some recommend against use in this population [12]. FEP patients are unlikely to have experienced a relapse, and it is not clear whether non-adherence is going to be an issue at this stage; thus, in current clinical practice, they are rarely considered for LAI treatment [19]. This expert panel agreed to statements pertaining to the appropriateness of LAIs for all patients with schizophrenia, including FEP and EP patients. This recommendation is despite the fact that those patients’ have not yet experienced the other issues that more chronically ill, multi-episode patients may have who are usually considered for LAIs, such as poor social support. Further support for the use of LAIs was demonstrated, as the panel found that other than severe adverse eventadverse events on the same oral AP that the LAI would be considered as, no other contraindications to starting patients on LAIs exist, providing they are appropriately monitored. This finding is supported by recent evidence that there is not any greater risk with LAIs compared to oral APs for the potentially fatal reaction neuroleptic malignant syndrome [25–27].
## Gaps in clinician knowledge
The lack of published guidance on how to initiate and maintain FEP and EP schizophrenia patients with LAIs has uncovered gaps in knowledge and concerns for clinicians using LAIs in this patient population [5, 17, 19]. To address these barriers, this Delphi panel indicated expert agreement that treatment goals and clinical management remain the same across patient subgroups and oral vs LAIs. However, whether there should be specific monitoring for FEP or EP patients on LAIs compared to the same oral AP was not determined in this panel. This finding indicates that the difference in formulation of the AP does not affect general recommendations about strategies for monitoring efficacy and safety, nor is there any specific difference in monitoring requirements between FEP/EP and chronic patients.
Most statements on managing adverse events in the first half of the injection interval achieved consensus, as did statements about establishing and maintaining efficacy. Feedback from the panelists suggested that reducing the injection interval to manage insufficient/loss of efficacy is an off-label strategy or against country-specific guidance. Incorporating advice on the management of adverse events and inefficacy for FEP and EP patients on LAIs into current guidelines could still be beneficial to support less experienced clinicians, particularly considering different countries’ regulations.
While statements describing raising the dose by adding the oral version of the AP and then implementing a higher dose to counteract loss of efficacy at the next interval received moderate agreement ($69\%$ and $75\%$, respectively), the panelists remained overall neutral ($50\%$ agreement) regarding switching to a new oral AP and commencing LAIs once tolerability and efficacy are established as the next strategy. Additionally, statements on initiating FEP or EP patients on a lower LAI dose than for chronic patients received a low consensus. This lack of consensus reflects the fact that schizophrenia and its clinical manifestations are heterogeneous, such that no simple categorization is possible; some patients with chronic illness may only tolerate or respond to lower AP doses whereas some FEP patients may require and tolerate higher doses.
## Barriers to LAIs
Barriers to LAIs include the stigma associated with injections (being possibly perceived as coercive) and the sense that LAIs are to be reserved for patients with severe illness [17]. Many FEP and EP patients are unaware that LAIs are an option for them; however, if they are included in guidelines and patients are informed of the benefits, this could increase acceptance [5, 9, 28]. There is recent evidence to suggest that some FEP and EP patients are open to LAI treatment and may even prefer it [29]. Offering LAIs as an option and eliciting patient’s preferences before switching to a specific antipsychotic agent could encourage patients and their families to consider the LAI treatment option. Additionally, a recent Delphi panel also found that experts agreed that LAIs can reduce the stigma of having to take daily medication [19].
## Long-term functional recovery
Functional recovery is a complex concept, and there remains a lack of clarity on its definition [30]. However, in this Delphi panel, recovery was conceptualized as going beyond symptomatic remission and encompassing multiple aspects of a patient’s life [31,32]. There is evidence that functional outcomes are especially improved with LAIs, particularly when offered earlier rather than later in the illness [33–35]. This finding is supported by the results from this current Delphi panel, which attained the agreement among experts that the known benefits of LAIs (increasing adherence and reducing functional decline, rehospitalization, and treatment burden) lead to a better long-term treatment outcomes and fuller functional recovery, which has been implied in other research [9, 17, 19, 29].
Because LAIs need to be administered in a patient care setting, such as a clinic, the patient may be seen more frequently by clinicians than patients not treated with LAIs [5]. In a clinic, it is easier to evaluate how the patient is progressing towards functional recovery. The expert panel in this Delphi reached consensus that functional recovery and HRQoL are often linked, which has been further supported, for example by a long-term trial follow-up for schizophrenia patients on the LAI aripiprazole once monthly [36].
## Functional recovery approach
While there are existing scales to assess functional remission, symptom improvement, and HRQoL (e.g., Functional Remission of General Schizophrenia, Positive and Negative Syndrome Scale, and Quality of Life Scale), this current Delphi panel sought expert consensus on an approach to functional recovery. Less than full functional recovery in some domains (sexual functioning and leisure activities) is considered acceptable. Furthermore, there are aspects of functional recovery that should be individualized, to take account of the patient’s personal goals and aspirations, attitude toward treatment, and receipt of appropriate psychoeducation [32].
## Limitations and further research
Limitations of the current study include the uneven distribution of Delphi panelists across countries, such that the group was Euro-centric. Thus, these results may not be fully generalizable to other locales. Additionally, further information on the demographics of the panelists could have been captured (e.g., years in practice), allowing for a better understanding of the effect of different experiences/expertise among panelists. Another limitation is the lack of a pilot study to further affirm the compressibility of the questionnaire and usefulness of the response options [37]. The lack of a pilot study was mitigated by the 1:1 interview in the first round to determine comprehensibility and to clarify any statements. Finally, the response option “I don’t know,” which has been used in other Delphi panels [38], was not used in this current study, which could have skewed the results.
## Summary And Conclusion
In summary, this *Delphi consensus* panel regarding the potential value of LAIs for treating patients with FEP or EP schizophrenia, with a particular focus on functional recovery identified many areas of broad consensus, as well as areas with low or very low agreement, which concerned antipsychotic initiation dosage in FEP and EP patients, and best managing practices when facing inefficacy and breakthrough episodes. However, there was broad consensus that FEP and EP patients could benefit from LAIs regarding decreasing the risk of relapse, rehospitalization, and functional dysfunction, supporting the use of LAIs in FEP and EP schizophrenia patients regardless of disease severity, number of relapses, or social support status, ultimately improving opportunities for achieving multidimensional functional recovery.
## Funding
This Delphi panel was funded by Lundbeck Otsuka Alliance
## Author Disclosures:
Celso Arango has received support by the Spanish Ministry of Science and Innovation, Instituto de Salud Carlos III (ISCIII), co-financed by the European Union, ERDF Funds from the European Commission, “A way of making Europe”, financed by the European Union - NextGenerationEU (PMP$\frac{21}{00051}$), PI$\frac{19}{01024.}$ CIBERSAM, Madrid Regional Government (B2017/BMD-3740 AGES-CM-2), European Union Structural Funds, European Union Seventh Framework Program, European Union H2020 Program under the Innovative Medicines Initiative 2 Joint Undertaking: Project PRISM-2 (Grant agreement No.101034377),Project AIMS-2-TRIALS (Grant agreement No 777394), Horizon Europe, the National Institute of Mental Health of the National Institutes of Health under Award Number 1U01MH12463901 (Project ProNET) and Award Number 5P50MH11584603 (project FEP-CAUSAL), Fundación Familia Alonso, and Fundación Alicia Koplowitz. Dr. Arango has been a consultant to or has received honoraria or grants from Acadia, Angelini, Biogen, Boehringer, Gedeon Richter, Janssen Cilag, Lundbeck, Medscape, Menarini, Minerva, Otsuka, Pfizer, Roche, Sage, Servier, Shire, Schering Plough, Sumitomo Dainippon Pharma, Sunovion and Takeda.
Philip Gorwood received during the last 5 years fees for presentations at congresses or participation in scientific boards from Angelini, EISI, Janssen, Lundbeck, Otsuka, Richter, Merk and Viatris.
John M. Kanehas been a consultant for or received honoraria from Alkermes, Allergan, Boehringer-Ingelheim, Cerevel, Dainippon Sumitomo, H. Lundbeck, HealthRhythms, HLS, Indivior, Intracellular Therapies, Janssen Pharmaceutical, Johnson & Johnson, LB Pharmaceuticals, Merck, Minerva, Neurocrine, Newron, Novartis, Otsuka, Roche, Saladax, Sunovion, and Teva. Dr. Kane has received grant support from Otsuka, Lundbeck, Sunovion and Janssen. Dr. Kane is a Shareholder in Vanguard Research Group, North Shore Therapeutics, Health Rhythms, MedinCell, and LB Pharmaceuticals, Inc.
Andrea Fagiolini is/has been a consultant and/or a speaker and/or has received research grants from Angelini, Apsen, Boheringer Ingelheim, Daiichi Sankyo, Glaxo Smith Kline, Italfarmaco, Lundbeck, Janssen, Mylan, Otsuka, Pfizer, Recordati, Sanofi Aventis, Sunovion, Viatris, Vifor Christoph U Correll has been a consultant and/or advisor to or has received honoraria from: AbbVie, Acadia, Alkermes, Allergan, Angelini, Aristo, Boehringer-Ingelheim, Cardio Diagnostics, Cerevel, CNX Therapeutics, Compass Pathways, Darnitsa, Gedeon Richter, Hikma, Holmusk, IntraCellular Therapies, Janssen/J&J, Karuna, LB Pharma, Lundbeck, MedAvante-ProPhase, MedInCell, Merck, Mindpax, Mitsubishi Tanabe Pharma, Mylan, Neurocrine, Newron, Noven, Otsuka, Pharmabrain, PPD Biotech, Recordati, Relmada, Reviva, Rovi, Seqirus, SK Life Science, Sunovion, Sun Pharma, Supernus, Takeda, Teva, and Viatris. He provided expert testimony for Janssen and Otsuka. He served on a Data Safety Monitoring Board for Compass, Lundbeck, Relmada, Reviva, Rovi, Supernus, and Teva. He has received grant support from Janssen and Takeda. He received royalties from UpToDate and is also a stock option holder of Cardio Diagnostics, Mindpax, LB Pharma and Quantic.
## Availability of data and materials
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.
## References
1. Charlson FJ, Ferrari AJ, Santomauro DF, Diminic S, Stockings E, Scott JG. **Global epidemiology and burden of schizophrenia: Findings from the global gurden of disease study 2016**. *Schizophr Bull* (2018) **44** 1195-203. PMID: 29762765
2. Kahn RS, Sommer IE, Murray RM, Meyer-Lindenberg A, Weinberger DR, Cannon TD. *Schizophrenia Nat Rev Dis Primers* (2015) **12** 1-23
3. Giordano G, Tomassini L, Cuomo I, Amici E, Perrini F, Callovini G. **Aripiprazole long-acting injection during first episode schizophrenia—an exploratory analysis**. *Front Psychiatry* (2020) **8** 10
4. Lippi G.. **Schizophrenia in a member of the family: Burden, expressed emotion and addressing the needs of the whole family**. *S Afr J Psychiatr* (2016) **22** 922. PMID: 30263163
5. Kane JM, McEvoy JP, Correll CU, Llorca PM. **Controversies surrounding the use of long-acting injectable antipsychotic medications for the treatment of patients with schizophrenia**. *CNS Drugs* (2021) **35** 1189-205. PMID: 34636025
6. Kane JM, Kishimoto T, Correll CU. **Non-adherence to medication in patients with psychotic disorders: epidemiology, contributing factors and management strategies**. *World Psychiatry* (2013) **12** 216-26. PMID: 24096780
7. Haddad PM, Brain C, Scott J. **Nonadherence with antipsychotic medication in schizophrenia: challenges and management strategies**. *Patient Relat Outcome Meas* (2014) **5** 43-62. PMID: 25061342
8. Higashi K, Medic G, Littlewood KJ, Diez T, Granström O, De Hert M. **Medication adherence in schizophrenia: factors influencing adherence and consequences of nonadherence, a systematic literature review**. *Ther Adv Psychopharmacol* (2013) **3** 200-18. PMID: 24167693
9. Correll CU, Citrome L, Haddad PM, Lauriello J, Olfson M, Calloway SM. **The use of long-acting injectable antipsychotics in schizophrenia: evaluating the evidence**. *J Clin Psychiatry* (2016) **77** 1-24. PMID: 27732772
10. Leijala J, Kampman O, Suvisaari J, Eskelinen S. **Daily functioning and symptom factors contributing to attitudes toward antipsychotic treatment and treatment adherence in outpatients with schizophrenia spectrum disorders**. *BMC Psychiatry* (2021) **21** 37. PMID: 33441112
11. Brugnoli R, Rapinesi C, Kotzalidis GD, Marcellusi A, Mennini FS, Filippis SD. **Model of Management (Mo.Ma) for the patient with schizophrenia: crisis control, maintenance, relapse prevention, and recovery with long-acting injectable antipsychotics (LAIs)**. *Riv Psichiatr* (2016) **51** 47-59. PMID: 27183509
12. Correll CU, Martin A, Patel C, Benson C, Goulding R, Kern-Sliwa J. **Systematic literature review of schizophrenia clinical practice guidelines on acute and maintenance management with antipsychotics**. *NPJ Schizophr* (2022) **24** 5
13. Correll CU, Solmi M, Croatto G, Schneider LK, Rohani-Montez SC, Fairley L. **Mortality in people with schizophrenia: a systematic review and meta‐analysis of relative risk and aggravating or attenuating factors**. *World Psychiatry* (2022) **21** 248-71. PMID: 35524619
14. Rubio JM, Taipale H, Tanskanen A, Correll CU, Kane JM, Tiihonen J. **Long-term continuity of antipsychotic treatment for schizophrenia: a nationwide study**. *Schizophr Bull* (2021) **21** 1611-20
15. Solmi M, Tiihonen J, Lähteenvuo M, Tanskanen A, Correll CU, Taipale H. **Antipsychotics use is associated with greater adherence to cardiometabolic medications in patients with schizophrenia: results from a nationwide, within-subject design study**. *Schizophr Bull* (2022) **48** 166-75. PMID: 34286338
16. Kishimoto T, Hagi K, Kurokawa S, Kane JM, Correll CU. **Long-acting injectable versus oral antipsychotics for the maintenance treatment of schizophrenia: a systematic review and comparative meta-analysis of randomised, cohort, and pre–post studies**. *The Lancet Psychiatry* (2021) **8** 387-404. PMID: 33862018
17. Bosanac P, Castle DJ. **Why are long-acting injectable antipsychotics still underused?**. *BJPsych advances* (2015) **21** 98-105
18. Arango C, Baeza I, Bernardo M, Cañas F, de Dios C, Díaz-Marsá M. **Antipsicóticos inyectables de liberación prolongada para el tratamiento de la esquizofrenia en España**. *Revista de Psiquiatría y Salud Mental* (2019) **12** 92-105. PMID: 29954707
19. Sajatovic M, Ross R, Legacy SN, Correll CU, Kane JM, DiBiasi F. **Identifying patients and clinical scenarios for use of long-acting injectable antipsychotics - expert consensus survey part 1**. *Neuropsychiatr Dis Treat* (2018) **14** 1463-74. PMID: 29922062
20. Kane JM, Schooler NR, Marcy P, Correll CU, Achtyes ED, Gibbons RD, Robinson DG. **Effect of long-acting injectable antipsychotics vs usual care on time to first hospitalization in early-phase schizophrenia: a randomized clinical trial**. *JAMA Psychiatry* (2020) **77** 1217-122421. PMID: 32667636
21. Dalkey N, Helmer O. **An experimental application of the DELPHI method to the use of experts**. *Manage Sci* (1963) **9** 458-67
22. Fink A, Kosecoff J, Chassin M, Brook RH. **Consensus methods: characteristics and guidelines for use.Am**. *J Public Health* (1984) **74** 979
23. McKenna H, Keeney S, Hasson F. *The Delphi Technique in Nursing and Health Research* (2011)
24. Rowe G, Wright G. *Expert opinions in forecasting: the role of the Delphi technique* (2001) 125-44
25. Guinart D, Misawa F, Rubio JM, Pereira J, de Filippis R, Gastaldon C. **A systematic review and pooled, patient-level analysis of predictors of mortality in neuroleptic malignant syndrome**. *Acta Psychiatr Scand* (2021) **144** 329-41. PMID: 34358327
26. Guinart D, Taipale H, Rubio JM, Tanskanen A, Correll CU, Tiihonen J. **Risk factors, incidence, and outcomes of neuroleptic malignant syndrome on long-acting injectable vs oral antipsychotics in a nationwide schizophrenia cohort**. *Schizophr Bull* (2021) **47** 1621-30. PMID: 34013325
27. Guinart D, Misawa F, Rubio JM, Pereira J, Sharma H, Schoretsanitis G. **Outcomes of neuroleptic malignant syndrome with depot versus oral antipsychotics: a systematic review and pooled, patient-level analysis of 662 rase reports**. *J Clin Psychiatry* (2020) **82** 20r13272
28. Lindenmayer JP, Glick ID, Talreja H, Underriner M. **Persistent barriers to the use of long-acting injectable antipsychotics for the treatment of schizophrenia**. *J Clin Psychopharmacol* (2020) **40** 346-9. PMID: 32639287
29. Kane JM, Schooler NR, Marcy P, Achtyes ED, Correll CU, Robinson DG. **Patients with early-phase schizophrenia will accept treatment with sustained-release medication (long-acting injectable antipsychotics): results from the recruitment phase of the PRELAPSE trial**. *J Clin Psychiatry* (2019) **80** 0-0
30. Correll CU, Kishimoto T, Nielsen J, Kane JM. **Quantifying clinical relevance in the treatment of schizophrenia**. *Clin Ther* (2011) **33** B16-39. PMID: 22177377
31. Correll CU, Ismail Z, McIntyre RS, Rafeyan R, Thase ME. **Patient functioning, life engagement, and treatment goals in schizophrenia**. *J Clin Psychiatry* (2022) **83**
32. Lahera G, Gálvez JL, Sánchez P, Martínez-Roig M, Pérez-Fuster JV, García-Portilla P. **Functional recovery in patients with schizophrenia: recommendations from a panel of experts**. *BMC Psychiatry* (2018) **18** 176. PMID: 29871616
33. Hargarter L, Bergmans P, Cherubin P, Keim S, Conca A, Serrano-Blanco A. **Once-monthly paliperidone palmitate in recently diagnosed and chronic non-acute patients with schizophrenia**. *Expert Opin Pharmacother* (2016) **17** 1043-53. PMID: 27042990
34. Brown B, Turkoz I, Mancevski B, Mathews M. **Evaluation of paliperidone palmitate long-acting injectable antipsychotic therapy as an early treatment option in patients with schizophrenia**. *Early Interv Psychiatry* (2020) **14** 428-38. PMID: 31515972
35. Kim S, Kim S, Koh M, Choi G, Kim JJ, Paik IH. **Effects of long-acting injectable paliperidone palmitate on clinical and functional outcomes in patients with schizophrenia based on illness duration**. *J Clin Psychiatry* (2021) **82** 5842
36. Naber D, Baker RA, Eramo A, Forray C, Hansen K, Sapin C. **Long-term effectiveness of aripiprazole once-monthly for schizophrenia is maintained in the QUALIFY extension study**. *Schizophr Res* (2018) **1** 192
37. Jünger S, Payne S, Brearley S, Ploenes V, Radbruch L. **Consensus building in palliative care: a europe-wide Delphi study on common understandings and conceptual differences**. *J Pain Symptom Manag* (2012) **44** 192-205
38. Vogel C, Zwolinsky S, Griffiths C, Hobbs M, Henderson E, Wilkins E. **A Delphi study to build consensus on the definition and use of big data in obesity research**. *Int J Obes (Lond)* (2019) **43** 2573-86. PMID: 30655580
|
---
title: 'Risk Factors and Predictive Modeling for Post-Acute Sequelae of SARS-CoV-2
Infection: Findings from EHR Cohorts of the RECOVER Initiative'
authors:
- Chengxi Zang
- Yu Hou
- Edward Schenck
- Zhenxing Xu
- Yongkang Zhang
- Jie Xu
- Jiang Bian
- Dmitry Morozyuk
- Dhruv Khullar
- Anna Nordvig
- Elizabeth Shenkman
- Russel Rothman
- Jason Block
- Kristin Lyman
- Yiye Zhang
- Jay Varma
- Mark Weiner
- Thomas Carton
- Fei Wang
- Rainu Kaushal
journal: Research Square
year: 2023
pmcid: PMC10029117
doi: 10.21203/rs.3.rs-2592194/v1
license: CC BY 4.0
---
# Risk Factors and Predictive Modeling for Post-Acute Sequelae of SARS-CoV-2 Infection: Findings from EHR Cohorts of the RECOVER Initiative
## Abstract
### Background
Patients who were SARS-CoV-2 infected could suffer from newly incidental conditions in their post-acute infection period. These conditions, denoted as the post-acute sequelae of SARS-CoV-2 infection (PASC), are highly heterogeneous and involve a diverse set of organ systems. Limited studies have investigated the predictability of these conditions and their associated risk factors.
### Method
In this retrospective cohort study, we investigated two large-scale PCORnet clinical research networks, INSIGHT and OneFlorida+, including 11 million patients in the New York *City area* and 16.8 million patients from Florida, to develop machine learning prediction models for those who are at risk for newly incident PASC and to identify factors associated with newly incident PASC conditions. Adult patients aged 20 with SARS-CoV-2 infection and without recorded infection between March 1st, 2020, and November 30th, 2021, were used for identifying associated factors with incident PASC after removing background associations. The predictive models were developed on infected adults.
### Results
We find several incident PASC, e.g., malnutrition, COPD, dementia, and acute kidney failure, were associated with severe acute SARS-CoV-2 infection, defined by hospitalization and ICU stay. Older age and extremes of weight were also associated with these incident conditions. These conditions were better predicted (C-index >0.8). Moderately predictable conditions included diabetes and thromboembolic disease (C-index 0.7–0.8). These were associated with a wider variety of baseline conditions. Less predictable conditions included fatigue, anxiety, sleep disorders, and depression (C-index around 0.6).
### Conclusions
This observational study suggests that a set of likely risk factors for different PASC conditions were identifiable from EHRs, predictability of different PASC conditions was heterogeneous, and using machine learning-based predictive models might help in identifying patients who were at risk of developing incident PASC.
## Plain Language Summary
Can we predict those who are at risk of newly incident post-acute sequelae of SARS-CoV-2 infection (PASC) using EHR data and machine learning predictive models? What are the factors associated with newly incident PASC conditions?*In this* retrospective cohort study of 35,275 [22,341] adults with SARS-CoV-2 infection and 326,126 [177,010] adults without SARS-CoV-2 infection from INSIGHT (OneFlorida+), we found that several incident PASC conditions (e.g., malnutrition, COPD, dementia, and acute kidney failure) were associated with severe acute SARS-CoV-2 infection and older age, and these conditions were better predicted based on information collected before baseline and during the acute phase. However, a variety of PASC conditions were moderately predictable (e.g., diabetes and thromboembolic disease) or less predictable (e.g, fatigue, anxiety, sleep disorders, and depression).These findings suggest that different PASC conditions exhibited heterogeneous predictability and association patterns and using machine learning-based predictive models and EHR datasets could help in identifying patients who were at risk of developing incident PASC conditions and managing the heterogeneous PASC conditions.
## Introduction
The global COVID-19 pandemic starting in late 2019 has led to more than 557 million infections and 6.4 million deaths as of July 14, 2022.1 Growing scientific and clinical evidence has demonstrated the existence of potential post-acute and long-term effects of COVID-19, which affect multiple organ systems2 and are referred to as post-acute sequelae of SARS-CoV-2 infection (PASC). Recently there have been several retrospective cohort analyses identifying potential PASC using real-world patient data3(p19),4,5(p19). However, research on the predictability of PASC and their associated risk factors is still limited, and mixed results have been reported. Such predictive modeling research can help patients and healthcare professionals to recognize the risk of PASC early and inform effective actions. Several studies found older age, higher severities in the acute phase of SARS-CoV-2 infection6, and pre-existing conditions (e.g., hypertension, obesity) may be associated with a higher risk of developing PASC.7–12 By contrast, some studies also reported that baseline clinical characteristics or demographics were not associated with PASC.10 Two main challenges may explain these seemingly conflicting findings: 1) Prior studies have typically been conducted using patient cohorts with small sample sizes including only a few hundred13 or thousand8 patients, limiting the significance and generalizability of conclusions derived; and 2) PASC conditions are highly heterogeneous14–16, thus their predictabilities and associated risk factors could be heterogeneous as well.
To fill in the knowledge gap and address these challenges, we conducted a systematic study on the predictability of a broad spectrum of incident PASC conditions and their associated risk factors. We used two large electronic health records (EHR) cohorts from the PCORnet clinical research networks (CRN)17, namely INSIGHT18, covering patients in the New York City (NYC) area, and OneFlorida +19, including patients from Florida. The INSIGHT and OneFlorida + were used as primary analyses and validation respectively. A wide range of PASC conditions were selected based on our previous findings using a rigorous data-driven analysis pipeline14 and other existing evidence or clinical knowledge (See the method section for a detailed list of PASC diagnoses). We developed machine learning-based prediction models to identify patients who were more likely to develop particular incident PASC conditions with their baseline characteristics and acute severity according to medical utilization. We compared the performance of machine learning models with different levels of complexity, including regularized Cox proportional hazard model, regularized logistic regression, gradient boosting machine, and deep neural network in both the survival analysis setting and binary classification setting, as well as examined the potential risk factors for different PASC conditions after removing background associations. We observed that within the broad range of PASC, a pattern of post-severe acute disease-associated conditions was reliably predictable. Decreasing the burden of severe disease will likely improve these outcomes. However, a variety of PASC conditions were less predictable and were less associated with upfront disease severity. This lack of predictability may represent a challenge as the burden of severe disease decreases. This study is part of the NIH Researching COVID to Enhance Recovery (RECOVER) Initiative, which seeks to understand, treat, and prevent the post-acute sequelae of SARS-CoV-2 infection (PASC).
## Data
This study used two large-scale de-identified real-world EHR datasets from the INSIGHT Clinical Research Network (CRN)18 and the OneFlorida + CRN19. The INSIGHT CRN contained longitudinal clinical data of approximately 12 million patients in the New York City metropolitan area, and the OneFlorida + CRN contained the EHR data of nearly 15 million patients from Florida and selected cities in Georgia and Alabama. The use of the INSIGHT data was approved by the Institutional Review Board (IRB) of Weill Cornell Medicine following NIH protocol 21–10-95–380 with protocol title: Adult PCORnet-PASC Response to the Proposed Revised Milestones for the PASC EHR/ORWD Teams (RECOVER). The use of the OneFlorida + data for this study was approved under the University of Florida IRB number IRB202001831.
## Definition of Post-acute Sequelae of SARS-CoV-2 (PASC)
We examined a list of potential PASC conditions as outcomes, including depressive disorders, anxiety disorder, general PASC symptoms and signs with ICD codes U099/B948, fever, malaise and fatigue, dizziness, malnutrition, fluid disorders, diabetes mellitus, edema, pressure ulcers, hair loss, paresthesia, dermatitis, chronic obstructive pulmonary disease (COPD), atelectasis, pulmonary fibrosis, dyspnea, acute pharyngitis, acute bronchitis, dementia, myopathies, cerebral ischemia, encephalopathy, cognitive problems, sleep disorders, headache, muscle weakness, fibromyalgia, joint pain, acute kidney failure, cystitis, genitourinary problems, constipation, gastroparesis, abdominal pain, gastroesophageal reflux disease (GERD), heart failure, hypotension, pulmonary embolism, thromboembolism, abnormal heartbeat, chest pain, and anemia. We compiled this list based on both our previous study and evidence from other literature.3,4,14 Any incident condition was defined in the SARS-CoV-2 infected patients who had the condition from 31 days to 180 days after but not having the condition three years to seven days before their acute infection. See Supplementary Table 1 for the detailed code list.
## Eligibility criteria and study population
We included adult patients aged 20 years or older with at least one SARS-CoV-2 polymerase chain reaction (PCR) or antigen laboratory test from March 1st, 2020, to November 30th, 2021. We further required at least one diagnosis code within three years to seven days before the index date (referred to as the baseline period), and at least one diagnosis code from 31 days to 180 days after the index date (referred to as the post-acute phase or follow-up period), to ensure that patients were connected to the healthcare system and were being observed during the study period. We followed each patient from 31 days after his/her index date until the day of the first target outcome, documented death, the latest date of any documented records in the database, 180 days after the baseline, or the end of our observational window (December 31, 2021), whichever came first. Two exposure groups were the SARS-CoV-2 infected group and the non-infected group. The SARS-CoV-2 infected group included patients with a positive SARS-CoV-2 PCR or antigen laboratory test. The index date of the infected group was defined as the date of the first documented positive PCR or antigen test. The non-infected group included patients whose SARS-CoV-2 PCR or Antigen tests were all negative throughout the entire study period with no documented COVID-19-related diagnoses at any time. The index date for patients in the non-infected group was defined as the date of the first negative PCR or antigen test. The association study and predictive modeling were conducted on the infected group. The non-infected group was used to rule out background associations that were not specific to PASC. The patient inclusion and exclusion cascades were illustrated in Fig. 1.
## Covariates
We collected clinical features in the baseline period (3 years to 1 week before lab-confirmed SARS-CoV-2 infection) and the severity of acute infection (1 week before to 30 days after lab-confirmed SARS-CoV-2 infection). Age was categorized into 20–39 years, 40–54 years, 55–64 years, 65–74 years, and 75 years and older groups. We set 55–64 as the reference group. Gender was grouped into female and male (reference). Only three patients in INSIGHT were identified as other/missing gender who were excluded. The race was categorized into Asian, Black or African American, White (reference), other or missing. Ethnicity was grouped into Hispanic, not Hispanic (reference), and other/missing. We used the national-level area deprivation index (ADI) to capture the socioeconomic disadvantage of patients’ residential neighborhood.20 Larger ADI values indicate mode socioeconomically deprived status. Missing ADI value was imputed with median ADI per site. The ADI is a ranking from 1 to 100 with 1 and 100 representing the lowest and the highest level of disadvantage, respectively. We grouped ADI into five categories and set the ADI category 1–20 as the reference group. Baseline healthcare utilization up to three years before the index date was measured according to their care setting. For each inpatient, outpatient, and emergency department encounter, we categorized each setting into 0 visit (reference group), 1 or 2 visits, and 3 or more visits, respectively. We also considered the infection periods, which were grouped into March 2020 – June 2020, July 2020 – October 2020, November 2020 - February 2021, March 2021 – June 2021, and July 2021 – November 2021. We set the first wave of the pandemic from March 2020 to June 2020 in INSIGHT as the reference group. Of note, the third wave from July 2021 – November 2021 period was dominated by the Delta variant. Body mass index (BMI) was grouped according to the WHO classification, BMI < 18.5 as underweight, BMI 18.5–24.9 as normal weight (reference), BMI 25–29.9 as overweight, BMI > = 30 obese, and set missing value as a separate group. Smoking status was also considered, and categorized into never (reference), current, former, and missing. There is a significant missingness regarding smoking status ($90.2\%$ in COVID-positive patients and $49.8\%$ in patients with at least one PASC) and we grouped these patients into the missing category.
A wide range of baseline clinical comorbidities were collected, based on a revised list of the Elixhauser comorbidities, conditions recommended by our clinician group, and related medications. Patients were ascertained as having a condition if they had at least two corresponding diagnoses documented during the baseline period. We also counted the number of pre-existing conditions and grouped them into no comorbidity (reference), 1, 2, 3, 4, and 5 or more. A detailed list of these pre-existing conditions and defined reference categories were summarized in Extended Data Table 1.
## Statistical analysis
For each potential PASC condition, we performed statistical analysis on its association with various covariates including the following steps. ( Step I) We built a separate multivariate Cox proportional hazard model on the EHR of patients who were infected by SARS-CoV-2 to assess associations of covariates and time to the first incident event or censoring in the follow-up period (31–180 days after COVID-19 confirmation). The censoring event is defined as the earliest event of documented death, loss of follow-up in the database, 180 days after the baseline, or the end of our observational window (December 31, 2021). Fully adjusted hazard ratios (aHR) of each covariate and target PASC event were estimated.(Step II) We built another multivariate Cox proportional hazard model on the EHR of all patients regardless of their SARS-CoV-2 infection status. The model inputs include two parts. One is the set of covariates. The other is the set of interaction terms defined as the product of each covariate and SARS-CoV-2 infection status (1 for SAR-CoV-2 infected patients and 0 for non-infected control patients) on the outcome condition. In this way, the coefficient of a particular covariate captured its association with the outcome condition for patients who were not infected by SARS-CoV-2, and the coefficient of its corresponding interaction term captured the “quantitative modifications” of such association for patients who were infected by SARS-CoV-2. Fully adjusted hazard ratios of each covariate and interaction term were estimated on both infected and non-infected patients.
We identified a covariate to be a potential risk factor of a particular PASC condition if it satisfied the following three criteria: The adjusted hazard ratio (aHR) estimated from the infected patients in Step I were greater than 1;The p-value of the above aHR was smaller than 0.000562, which was corrected by the Bonferroni method for multiple testing;The aHR of the interaction term of the corresponding covariate should also be greater than 1 in Step II.
To build predictive models for each PASC condition, we examined different machine learning models in both survival analysis and binary classification settings. For the survival analysis setting, we used a multivariate Cox proportional hazard model with L2 norm regularization to predict the time to the outcome event. For the binary classification setting, the occurrence of the target event in the follow-up period was labeled as 1 and 0 otherwise. We used logistic regression with L2 norm regularization, gradient boosting machine with random forest base learner, and deep feed-forward neural network. For each of the abovementioned models, the best model was selected by grid search (see details in the following sensitivity analysis paragraph) a predefined hyper-parameter space through repeated cross-validation (ten times, five folds). The concordance index (C-index) and the area under the receiver operating characteristic curve (AUROC) were used to evaluate survival prediction performance and binary prediction performance respectively. Both two measures range from 0 to 1 with 0.5 indicating random guess and 1 indicating perfect prediction. The $95\%$ confidence interval of the final performance was estimated by 1000-times bootstrapping performance on each of the testing datasets in repeated cross-validation.
## Stratified analysis
The stratified analysis was conducted by stratifying patients by their severity in the acute infection phase (hospitalized or non-hospitalized) and then performing statistical analysis within each stratum. The noninfected control patients were also stratified according to the hospitalized or non-hospitalized during the 1 week before to 30 days after their index date, to capture background associations within each subgroup population.
## Sensitivity analysis
To get robust conclusions, we conducted the following sensitivity analyses. For association analysis, we also used a univariate Cox model for each covariate adjusted for age, sex, and acute severity. We further tested the impact of lifting Step II of the statistical analysis on the identified risk associations.
For the predictive modeling, we also tested a different feature engineering method, which used the first 3-digits of ICD-10 codes and medication at the ingredient level to test to what extent PASC can be predicted in a data-driven manner. We selected 1,593 ICD-10 diagnosis codes, 2,309 drugs and 1,698 ICD-10 diagnosis codes, and 4,366 drugs from the INSIGHT dataset and OneFlorida + data, respectively. These ICD-10 diagnosis codes and medications were selected to construct the input feature vectors of the prediction model based on the significant difference (P-value less than 0.0001) between patients with positive and negative PASC conditions results. After the feature selection process, the selected ICD-10 diagnosis codes, medication, and collected baseline covariates were constructed to represent every PASC condition.
We also tested different machine learning predictive models in both the survival analysis setting and binary classification setting to validate the predictability of each PASC potentially impacted by different models. For the survival analysis setting, we tested Cox proportional model with L2-norm regularization. For the binary classification setting, we investigated three machine-learning models with different complexity. The first one is the regularized logistic regression. We adopted the L2-norm penalty and searched the inverse of regularization strength from 10^−3 to 10^3 with 0.5 as the sampling step size. The second one is the gradient boosting machine with a random forest as the base learner. We searched hyperparameters from maximum depth [3,4,5], max number of leaves in one tree [10,20,30], and a minimal number of data in one leaf [30]. The third one is the deep forward neural network. We used ReLU (Rectified Linear Unit) activation function for the hidden layer, and search the hidden layers ([32,], [64,], [128,], [32, 32], [64, 64], [128, 128]), and learning rate (0.001, 0.01, 0.1). For each of the above-mentioned models, the best model was selected by grid search of the corresponding hyperparameter space through repeated cross-validation (ten times, five folds). In the repeated cross-validation process, we set one of the folds as the test set and the rest of the data as the training set. The C-index and the area under the receiver operating characteristic curve (AUROC) were used to measure the predictive performance in the survival setting and binary classification setting, respectively.
We have examined the impact of the criterion on requiring the identified association to be with a higher risk in SARS-CoV-2 infected patients compared to non-infected patients. Extended Data Fig. 1 depicted the identified associations after we lifted this requirement, i.e., we only require these associations to satisfy the adjusted hazard ratio and statistical significance constraints. From the figure, we observed that more associations have been identified compared to Fig. 3, and many of these associations may not be relevant to SARS-CoV-2 infection. Taking patients with pre-existing cancer as an example, they were associated with a higher risk of being diagnosed with fluid disorders, acute kidney failure, thromboembolism, encephalopathy, edema, malaise, and fatigue in the post-acute period after SARS-CoV-2 infection. However, these associations can also be identified for non-infected cancer patients. Therefore, the excessive risk criterion is necessary for filtering out the associations that are not specific to SARS-CoV-2 infection.
We also tested to what extent the predictability of incident potential PASC conditions is affected by different machine learning models. We investigated a range of machine learning models with different complexities, including regularized logistic regression models, gradient boosting machines, and feed-forward deep neural networks. As shown in Extended Data Fig. 2, we observed little difference across the performance of these different models, and the heterogeneous predictability patterns were still observed, i.e., conditions that were difficult to predict in Fig. 2 were still with low predictive performance despite using more complex models.
Lastly, we studied if different feature engineering can impact the prediction results of different PASC conditions. Instead of using pre-defined baseline comorbidities, we leveraged a data-driven approach by using the first three digits of ICD-10 codes of all diagnoses and all medications in RxNorm codes at the ingredient level in the baseline period to predict PASC. We reported the predictive performance of different machine learning models using this large set of features in Extended Data Fig. 3, which does not show big differences compared to the performance in Extended Data Fig. 2 or Fig. 2, and the heterogeneous predictability patterns remain the same.
## Validation analysis and generalizability
To get a generalizable conclusion, we further replicated the abovementioned association analyses and predictive analyses to the OneFlorida + cohort. The cohort selection and modeling strategies were the same as our primary analyses on the INSIGHT cohort.
## Prediction Performance
We developed our primary results on the INSIGHT cohort and used the OneFlorida + cohort as a validation cohort. Both cohorts were collected from patients who has at least one PCR/antigen test for SARS-CoV-2 infection from March 2020 to November 2021, and the inclusion-exclusion cascade was provided in Fig. 1. The INSIGHT cohort included 35,275 adult patients with lab-confirmed SARS-CoV-2 infection and 326,126 non-infected control patients.
The current definition of PASC in the RECOVER protocols is ongoing, relapsing, new symptoms, or other health effects occurring four or more weeks after the acute phase of SARS-CoV-2 infection.21 We compiled a broad list of potential PASC conditions in terms of Clinical Classifications Software Refined (CCSR) categories22 based on our previous findings14 and evidence from other literature.3,4 Here we studied incident PASC conditions ascertained from 31 days to 180 days after the start of the acute SARS-CoV-2 infection date, denoted as the index date, but not existed one week to three years prior to the index date (See Method for the approach on how the list was compiled and Supplementary Table 1 for the detailed information).
We built a list of 89 covariates that are potentially associated with PASC based on a revised list of Elixhauser comorbidities, recommendations of our RECOVER clinician team, and the severity of acute infection of SARS-CoV-2. These covariates included basic demographics (e.g., age, gender, race, ethnicity), social-economic status in terms of Area Deprivation Index (ADI)21, healthcare utilization history, body mass index, the period of infection, comorbidities, and the care settings in acute phase including hospitalization and ICU admission. For each of the categorical covariates, we defined its reference group the same as prior studies for acute SARS-CoV-2 infection (details see Method covariates section).6 We built different machine learning models to predict the individual risk of encountering each incident condition using these covariates. The prediction performance of a regularized Cox model measured by the Concordance index (C-index)23 with a $95\%$ confidence interval was shown in Fig. 2 (results for other machine learning models are provided in the Sensitivity Analysis section).
Figure 2 shows that different incident conditions were associated with heterogeneous predictive performance. Conditions such as dementia, malnutrition, stroke, non-specific PASC (U099/B948), and kidney failure had a C-index > 0.8, in addition to other conditions such as myopathy, and pressure sores. We noted that diabetes, thromboembolic disease, and COPD were moderately predictable, with a C-index > 0.7, and other conditions such as fatigue, anxiety disorders, and sleep disorders were less predictable, with a C-index < 0.6.
## Associations between risk factors and specific PASC conditions.
Furthermore, we analyzed the associations between the covariates and the risk of developing any incident condition from our list. The unadjusted hazard ratio (HR) and fully adjusted hazard ratio (aHR) for each covariate. A covariate was identified as a potential risk factor for developing a particular condition if it satisfied the following three criteria: [1] the corresponding aHR of the covariate with respect to the target condition is larger than 1 when compared with the reference group (Method covariates section and Extended Data Table 1); [2] the association was statistically significant after multiple testing correction (p-Value < 0.000562); and [3] the associated risk was higher in SARS-CoV-2 infected patient population compared to the non-infected population. Note that criterion [3] is to guarantee the risk association we identified is not a common one that widely exists in patients without COVID-19, and the technical details on implementing this have been provided in Methods. Overall, among 35,275 enrolled SARS-CoV-2 infected patients in the INSIGHT cohort, 17,571 ($49.8\%$) of them had at least one incident potential PASC condition (Table 1). The associations between the covariates and the risk of getting at least one PASC were summarized in Extended Data Table 1. Figure 3 depicted the associations between the identified risk factors and specific PASC conditions, which we would further elaborate on as follows.
## The severity of acute infection.
Increased severity of the acute SARS-CoV-2 infection (according to the care settings) was associated with a higher risk of being diagnosed with new incident conditions in the post-acute period. Overall, a higher risk of getting any incident diagnosis was observed in patients who were hospitalized during the acute phase (1.29 (1.24–1.33)) or in ICU (1.40 (1.32–1.49)) compared to patients who were not hospitalized during the acute phase (as a reference group, see the Extended Data Table 1). Figure 3 further showed the associations between the acute phase severity and a range of potential PASC conditions. Specifically, compared to non-hospitalized patients, the ICU patients showed a 4.7-fold higher risk of being diagnosed with myopathy, 2.5-fold higher risk of being diagnosed with pressure ulcers, 2.3-fold higher risk of being diagnosed with thromboembolism, 2.1-fold higher risk of being diagnosed with malaise and fatigue. In addition, patients who were hospitalized or admitted to ICU during the acute phase had a higher risk of being diagnosed with general PASC codes U099/B948, with 4.3- and 2.2- fold increases compared to non-hospitalized patients.
## Age.
Patients aged 75 or older showed an increased risk of being diagnosed with a wide range of potential PASC conditions in the post-acute infection phase, including dementia (5.8-fold higher), COPD (2.2-fold), cerebral ischemia (2.1-fold), malnutrition (1.8-fold), pressure ulcer (1.8-fold), anemia (1.6-fold), cognitive problems (1.6-fold) compared to patients were 55–64 years old (as reference). Patients with 65 to 74 years old showed an increased risk of being diagnosed with dementia (2.7-fold), heart failure (1.6-fold), and diabetes mellitus (1.6-fold) compared to reference patients. By contrast, younger patients aged 20–39 years old exhibited an increased risk of getting milder potential PASC conditions including acute pharyngitis (1.7-fold), headache (1.4-fold), and anxiety disorder (1.4-fold) than patients in the reference group.
## Gender and Race.
Female patients exhibited a 4.3- and 1.3-fold increased risk of being diagnosed with incident hair loss and anxiety disorder in the post-acute infection period compared to male patients. Black patients exhibited a 1.9-fold increased risk of being diagnosed with incident diabetes mellitus than white patients.
## Body Mass Index.
Patients who were underweight (BMI < 18.5 kg/m2) or obese (BMI ≥ 30 kg/m2) were at higher risk of being diagnosed with certain potential PASC conditions than those with normal BMI (BMI from 18.5 to 24.9 kg/m2). Specifically, underweight patients were at a 1.6-fold-increased risk of being diagnosed with heart failure, and diabetes mellitus, and a 1.4-fold-increased risk of being diagnosed with malnutrition than patients with normal BMI. Obese patients showed a 1.8-fold-increased risk of being diagnosed with diabetes mellitus and a 1.3-fold-increased risk of being diagnosed with a sleep disorder.
## Period of infection.
We observed that patients who got infected from July 2021 to November 2021, which was dominated by the Delta variant of SARS-CoV-224, showed an increased risk of being diagnosed with incident pharyngitis (3.2-fold), chest pain (1.9-fold), abdominal pain (1.7-fold), dyspnea (1.6-fold), as well as being diagnosed with general PASC symptoms and signs with the U099/B948 ICD codes (5-fold) in the post-acute infection period compared to patients got infected during March 2020 to June 2020 (the 1st wave) as the reference period.
## Pre-existing conditions.
As shown in Fig. 3, having one or more baseline conditions was associated with a higher risk of potential PASC diagnosis including malnutrition, fluid disorders, anemia, and chest pain. Specifically, cancer patients showed increased risk in a broad list of post-acute conditions including malnutrition, atelectasis, fever, anemia, pulmonary fibrosis, constipation, and fibromyalgia compared to those without cancer diagnoses at baseline. Patients having baseline chronic kidney disease showed an increased risk of being diagnosed with heart failure and anemia. Those with baseline cirrhosis showed a 3-fold-increased risk of gastroparesis, a 2-fold-increased risk of atelectasis, and a 1.8-fold-increased risk of anemia. Those with baseline coagulopathy showed a higher risk of thromboembolism and cognitive problems. Patients with end-stage renal disease showed a higher risk of COPD and malnutrition. Those with baseline mental health disorders exhibited a higher risk of dementia and anxiety disorders in the post-acute period. Parkinson’s disease patients showed a 2.2-fold-increased risk of encephalopathy. Pregnant females showed a 2.4-fold increased risk of anemia in the post-acute period. Those with baseline pulmonary circulation disorder showed a 3.3-fold-increased risk of pulmonary embolism and a 1.9-fold-increased risk of heart failure. Patients with weight loss at baseline were at a higher risk of being diagnosed with pressure ulcers, COPD, constipation, and general PASC (with U099/B948) in the post-acute phase.
## Stratified Risk analysis
We further conducted stratified analyses to examine the associations between baseline factors and incident potential PASC conditions according to the care settings in the acute phase (hospitalized versus non-hospitalized). The same criteria on adjusted hazard ratio and statistical significance as we used in Fig. 3 were adopted here to identify potential risk associations, which were demonstrated in Fig. 4. Overall, certain demographic characteristics including older age, female, and black race, as well as baseline conditions including obesity and chronic kidney disease, were associated with increased risk of begin diagnosed with PASC in both non-hospitalized and hospitalized patients. There were also differences in these identified associations across the two different settings. Specifically, for patients who were not hospitalized during acute infection, we observed that baseline arrythmia was associated with a 1.9-fold increased risk of an incident diagnosis of heart failure in the post-acute period, pregnancy was associated with 3.4-fold-increased risk of incident anemia, and patients with baseline sickle cell disease showed a 3.2-fold-increased risk of being diagnosed with anxiety disorder. However, these associations were not identified among patients who were hospitalized in the acute phase. For these patients, we observed that baseline pulmonary circulation disorder was associated with an 11.4-fold-increased risk of being diagnosed with pulmonary embolism, and baseline multiple sclerosis was associated with a 2.7-fold-increased risk of being diagnosed with malaise and fatigue in the post-acute phase.
## Validation by the OneFlorida + Cohort
To assess the generalizability of our findings, we replicated our analyses on the OneFlorida + cohort as an independent validation. The OneFlorida + cohort included 22,341 adult patients with lab-confirmed SARS-CoV-2 infection and 177,010 non-infected as control patients (See inclusion cascade in Fig. 1). We summarized the prediction performance of different potential PASC conditions with regularized Cox model in Extended Data Fig. 4 and the identified risk associations in Extended Data Fig. 5. From Extended Data Fig. 4 we again observed the heterogeneous predictability of different conditions as has been observed in Fig. 2, and the more predictable conditions (with c-index > 0.8, such as malnutrition, COPD, dementia, and acute kidney failure) and less predictable (with c-index around or below 0.6, such as fatigue, anxiety, sleep disorders, and depression) remained the same. Similarly, the risk associations shown in Extended Data Fig. 4 are consistent with the risk associations shown in Fig. 3. Hospitalization and ICU admission in the acute infection phase were associated with a diverse set of incident diagnoses in the post-acute infection phase. We still observed the risk associations between older age and dementia (5.4-fold increased risk), female and hair loss (2.2-fold increased risk), black race, and diabetes (1.5-fold increased risk). Infection confirmation from July to November 2021 was associated with a 1.7-fold increased risk of being diagnosed with general PASC symptoms and signs (the U099/B948 ICD code).
## Discussion
In this paper, we conducted a systematic study on the predictability of a wide range of potential PASC conditions as well as their associated risk factors using the EHR data from two large-scale PCORnet clinical research networks, INSIGHT, and OneFlorida+. Compared with existing research on this topic which was mostly based on patient-reported symptoms12,25, our study was based on routinely collected EHR data from large patient populations.
We investigated the predictability of different potential PASC diagnoses using patient demographics, prior conditions, and care settings in the acute phase. Different types of machine learning models, including linear models, tree-based models, and deep learning models were tested. Following prior research on PASC3,26, we focused on newly incident conditions in the post-acute infection period in this study. We have built a comprehensive list of diagnoses based on a prior study by Al-Aly et al.3, with further refinements from our clinician team. Different from existing relevant studies that treated PASC as a holistic concept3,26, we have explored the predictability and potential risk factors of each individual condition, as there had been abundant evidence suggesting PASC was a highly heterogeneous condition affecting multiple organ systems3,14.
The results from regularized Cox regression were summarized in Fig. 2, which suggested that different conditions were associated with different predictabilities in the INSIGHT cohort. Conditions such as stroke, heart failure, and kidney failure were more predictable. These conditions are with clear diagnostic criteria according to the underlying disease etiologies and are more likely to be severe COVID complications. Pressure ulcers was also highly predictable, but it was more likely due to prolonged hospital stay during the acute phase of SARS-CoV-2 infection27. General PASC symptoms and signs with the U099/B948 codes were also associated with good prediction performance, which is consistent with prior studies28. One potential reason was that these codes were relatively new, and the clinicians might be cautiously using them only when the symptoms and signs were typical. Conditions such as headache, dizziness, chest pain, joint pain, anxiety, and depressive disorders, were more difficult to predict. These conditions are most subjective to diagnose, more similar to patient-reported symptoms, and cannot be explained by alternative disease etiologies. The prediction performance obtained from more complex machine learning models did not make such differences, as evidenced by Extended Data Fig. 2. In addition, we have replicated the predictive modeling analysis on the OneFlorida + cohort, and the results summarized in Extended Data Fig. 4 were highly consistent with the conclusions obtained from the INSIGHT cohort.
With fully adjusted analysis, we examined the statistical associations between a broad list of covariates including demographics, pre-existing conditions, and severities in the acute phase of SARS-CoV-2 infection according to care settings and each potential PASC condition. For a covariate to be considered as a potential risk factor of a specific condition, we required its corresponding adjust hazard ratio (aHR) to be larger than 1 and statistically significant. We further required the estimated aHR value to be larger in patients who were infected by SARS-CoV-2 compared to the non-infected patients, in this way associations that may not be attributed to COVID-19 can be filtered. Figure 3 and Extended Data Fig. 5 summarized the identified risk associations from the INSIGHT and OneFlorida + cohorts. Both figures showed that hospitalizations and admissions to ICU during the acute infection phase were associated with a broad set of incident conditions in the post-acute infection phase, including pressure ulcers, heart failure, acute kidney failure, COPD, etc., which suggested that these conditions could be related to either severe acute COVID complications or acute care processes. Older age (> = 75 years) was also found to be a potential risk factor for many of these conditions. Black patients were at higher risk of being diagnosed with incident diabetes. These discoveries were consistent with the conclusions from prior studies29,30. Other notable risk associations consistently identified from both cohorts include higher baseline comorbidity burden and fluid disorder, baseline obesity and sleep disorder, as well as baseline end-stage renal disease and malnutrition. Some associations should be interpreted more cautiously. For example, baseline pulmonary circulation disorder was consistently identified as a risk pulmonary embolism, but the two conditions are highly correlated with each other, and this association could just be due to the ICD coding and grouping. Another example was baseline pregnancy and anemia, as anemia is the most common hematologic problem in pregnancy31. However, there were also studies suggesting that SARS-CoV-2 infection during pregnancy can further exacerbate iron deficiency anemia due to hyperinflammation during the acute infection phase32.
There were several strengths of our study. First, we studied a comprehensive set of associations between 89 factors and 44 incident PASC conditions in two large EHR cohorts. To our knowledge, this is one of the largest studies on predictive modeling and risk factor analysis for PASC. Second, we derived our primary results from INSIGHT and did a validation study on OneFlorida+, which validated the generalizability of our findings. Third, we have tested the prediction performance of different machine learning models on both a narrow and broad list of covariates, which further validates the robustness of our conclusions. Finally, we ruled out potential background associations by requiring the adjusted hazard ratio value of the identified association estimated from the patients who were infected by SARS-CoV-2 to be larger than the value estimated from patients who are not infected by SARS-CoV-2.
Our study had several limitations. Our analysis was based on EHR data, which would miss the information from patients who did not visit the hospitals within the CRNs. We only considered newly incident individual conditions in the post-acute period but did not explore conditions that were prolonged, worsened, or relapsed before and after COVID-19 infection, as well as condition clusters or subphenotypes. Vaccine information was not incorporated in our study due to its incompleteness in the EHR records, and we are working on adding other data sources (e.g., state registry data) to make the information more robust. In addition, our analyses did not cover the recent Omicron wave due to the availability of the data.
In conclusion, we used two large-scale clinical research networks, INSIGHT and OneFlorida + to identify risk factors associated with newly incident PASC conditions and to develop predictive models to identify those who are at risk of these conditions. Our results highlight that several predictive PASC diagnoses are associated with severity in the acute phase. However, less predictable PASC diagnoses represent an ongoing challenge that may not respond to other measures that decrease the severity of acute COVID.
## Data Availability
The INSIGHT data can be requested through https://insightcrn.org/. The OneFlorida+ data can be requested through https://onefloridaconsortium.org. Both the INSIGHT and the OneFlorida+ data are HIPAA-limited. Therefore, data use agreements must be established with the INSIGHT and OneFlorida+ networks.
## References
1. 1.WHO Coronavirus (COVID-19) Dashboard. Accessed
July 19, 2022. https://covid19.who.int. *WHO Coronavirus (COVID-19) Dashboard*
2. Nalbandian A, Sehgal K, Gupta A. **Post-acute COVID-19 syndrome**. *Nat Med* (2021.0) **27** 601-615. DOI: 10.1038/s41591-021-01283-z
3. Al-Aly Z, Xie Y, Bowe B. **High-dimensional characterization of post-acute sequelae of COVID-19**. *Nature* (2021.0) **594** 259-264. DOI: 10.1038/s41586-021-03553-9
4. Xie Y, Xu E, Bowe B, Al-Aly Z. **Long-term cardiovascular outcomes of COVID-19**. *Nat Med* (2022.0) 1-8. DOI: 10.1038/s41591-022-01689-3
5. Xie Y, Xu E, Al-Aly Z. **Risks of mental health outcomes in people with covid-19: cohort study**. *BMJ* (2022.0) **376** e068993. DOI: 10.1136/bmj-2021-068993
6. Williamson EJ, Walker AJ, Bhaskaran K. **Factors associated with COVID-19-related death using OpenSAFELY**. *Nature* (2020.0) **584** 430-436. DOI: 10.1038/s41586-020-2521-4
7. Antonelli M, Penfold RS, Merino J. **Risk factors and disease profile of post-vaccination SARS-CoV-2 infection in UK users of the COVID Symptom Study app: a prospective, community-based, nested, case-control study**. *Lancet Infect Dis* (2022.0) **22** 43-55. DOI: 10.1016/S1473-3099(21)00460-6
8. Asadi-Pooya AA, Akbari A, Emami A. **Risk Factors Associated with Long COVID Syndrome: A Retrospective Study**. *Iran J Med Sci* (2021.0) **46** 428-436. DOI: 10.30476/ijms.2021.92080.2326
9. Crook H, Raza S, Nowell J, Young M, Edison P. **Long covid—mechanisms, risk factors, and management**. *BMJ* (2021.0) n1648. DOI: 10.1136/bmj.n1648
10. **Post-acute COVID-19 syndrome**. *Incidence and risk factors: A Mediterranean cohort study - ScienceDirect*
11. Aminian A, Bena J, Pantalone KM, Burguera B. **Association of obesity with postacute sequelae of COVID-19**. *Diabetes Obes Metab* (2021.0) **23** 2183-2188. DOI: 10.1111/dom.14454
12. Sudre CH, Murray B, Varsavsky T. **Attributes and predictors of long COVID**. *Nat Med* (2021.0) **27** 626-631. DOI: 10.1038/s41591-021-01292-y
13. Moy FM, Hairi NN, Lim ERJ, Bulgiba A. *Long COVID and its associated factors among COVID survivors in the community from a middle-income country – an online cross-sectional study* (2022.0). DOI: 10.1101/2022.03.09.22272168
14. Zang C, Zhang Y, Xu J. *Understanding Post-Acute Sequelae of SARS-CoV-2 Infection through Data-Driven Analysis with Longitudinal Electronic Health Records: Findings from the RECOVER Initiative* (2022.0). DOI: 10.1101/2022.05.21.22275420
15. Zhang H, Zang C, Xu Z. *Machine Learning for Identifying Data-Driven Subphenotypes of Incident Post-Acute SARS-CoV-2 Infection Conditions with Large Scale Electronic Health Records: Findings from the RECOVER Initiative* (2022.0). DOI: 10.1101/2022.05.21.22275412
16. Zhang H, Zang C, Xu Z. **Data-driven identification of post-acute SARS-CoV-2 infection subphenotypes**. *Nat Med* (2022.0) 1-10. DOI: 10.1038/s41591-022-02116-3
17. Fleurence RL, Curtis LH, Califf RM, Platt R, Selby JV, Brown JS. **Launching PCORnet, a national patient-centered clinical research network**. *J Am Med Inform Assoc* (2014.0) **21** 578-582. DOI: 10.1136/amiajnl-2014-002747
18. Kaushal R, Hripcsak G, Ascheim DD. **Changing the research landscape: the New York City Clinical Data Research Network**. *J Am Med Inform Assoc* (2014.0) **21** 587-590. DOI: 10.1136/amiajnl-2014-002764
19. Shenkman E, Hurt M, Hogan W. **OneFlorida Clinical Research Consortium: Linking a Clinical and Translational Science Institute With a Community-Based Distributive Medical Education Model**. *Acad Med* (2018.0) **93** 451-455. DOI: 10.1097/ACM.0000000000002029
20. Kind AJH, Buckingham WR. **Making Neighborhood-Disadvantage Metrics Accessible — The Neighborhood Atlas**. *N Engl J Med* (2018.0) **378** 2456-2458. DOI: 10.1056/NEJMp1802313
21. Plummer J. **(NIH/OD) [E]**. *NIH-Post-Acute Sequelae of SARS-CoV-2 Infection Initiative: SARS-CoV-2 Recovery Cohort Studies*
22. 22.Clinical Classifications Software Refined (CCSR). Accessed
April 3, 2022. https://www.hcup-us.ahrq.gov/toolssoftware/ccsr/ccs_refined.jsp
23. Brentnall AR, Cuzick J. **Use of the concordance index for predictors of censored survival data**. *Stat Methods Med Res* (2018.0) **27** 2359-2373. DOI: 10.1177/0962280216680245
24. Antonelli M, Pujol JC, Spector TD, Ourselin S, Steves CJ. **Risk of long COVID associated with delta versus omicron variants of SARS-CoV-2**. *The Lancet* (2022.0) **399** 2263-2264. DOI: 10.1016/S0140-6736(22)00941-2
25. Su Y, Yuan D, Chen DG. **Multiple early factors anticipate post-acute COVID-19 sequelae**. *Cell* (2022.0) **185** 881-895.e20. DOI: 10.1016/j.cell.2022.01.014
26. Xie Y, Bowe B, Al-Aly Z. **Burdens of post-acute sequelae of COVID-19 by severity of acute infection, demographics and health status**. *Nat Commun* (2021.0) **12** 6571. DOI: 10.1038/s41467-021-26513-3
27. Gefen A, Ousey K. **COVID-19: pressure ulcers, pain and the cytokine storm**. *J Wound Care* (2020.0) **29** 540-542. DOI: 10.12968/jowc.2020.29.10.540
28. 28.Identifying who has long COVID in the USA: a machine learning approach using N3C data – The Lancet Digital Health. Accessed
November 16, 2022. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(22)00048-6/fulltext. *Identifying who has long COVID in the USA: a machine learning approach using N3C data – The Lancet Digital Health*
29. Cohen K, Ren S, Heath K. **Risk of persistent and new clinical sequelae among adults aged 65 years and older during the post-acute phase of SARS-CoV-2 infection: retrospective cohort study**. *BMJ* (2022.0) **376** e068414. DOI: 10.1136/bmj-2021-068414
30. 30.Risks and burdens of incident diabetes in long COVID: a cohort study - The Lancet Diabetes & Endocrinology. Accessed
April 3, 2022. https://www.thelancet.com/journals/landia/article/PIIS2213-8587(22)00044-4/fulltext. *Risks and burdens of incident diabetes in long COVID: a cohort study - The Lancet Diabetes & Endocrinology*
31. Sifakis S, Pharmakides G. **Anemia in Pregnancy**. *Ann N Y Acad Sci* (2000.0) **900** 125-136. DOI: 10.1111/j.1749-6632.2000.tb06223.x
32. 32.COVID-19 and iron deficiency anemia: relationships of pathogenesis and therapy | Gromova | Obstetrics, Gynecology and Reproduction. Accessed
November 16, 2022. https://www.gynecology.su/jour/article/view/831?locale=en_US. *COVID-19 and iron deficiency anemia: relationships of pathogenesis and therapy | Gromova | Obstetrics, Gynecology and Reproduction*
|
---
title: Chemical pancreatectomy in non-human primates ablates the acini and ducts and
enhances beta-cell function
authors:
- Ranjeet S. Kalsi
- Alexander M. Kreger
- Mohamed Saleh
- Shiho Yoshida
- Kartikeya Sharma
- Joseph Fusco
- Jami L. Saloman
- Ting Zhang
- Madison Thomas
- Anuradha Sehrawat
- Yan Wang
- Jason Reif
- Juliana Mills
- Sarah Raad
- Bugra Zengin
- Ana Gomez
- Aatur Singhi
- Sameh Tadros
- Adam Slivka
- Farzad Esni
- Krishna Prasadan
- George Gittes
journal: Research Square
year: 2023
pmcid: PMC10029118
doi: 10.21203/rs.3.rs-2618133/v1
license: CC BY 4.0
---
# Chemical pancreatectomy in non-human primates ablates the acini and ducts and enhances beta-cell function
## Abstract
Chronic pancreatitis is a debilitating disease affecting millions worldwide. These patients suffer from bouts of severe pain that are minimally relieved by pain medications and may necessitate major surgeries with high morbidity and mortality. Previously, we demonstrated that “chemical pancreatectomy,” a pancreatic intraductal infusion of dilute acetic acid solution, ablated the exocrine pancreas while preserving the endocrine pancreas. Notably, chemical pancreatectomy resolved chronic inflammation, alleviated allodynia in the cerulein pancreatitis model, and improved glucose homeostasis. Herein, we extensively tested the feasibility of a chemical pancreatectomy in NHPs and validated our previously published pilot study. We did serial computed tomography (CT) scans of the abdomen and pelvis, analyzed dorsal root ganglia, measured serum enzymes, and performed histological and ultrastructural assessments and pancreatic endocrine function assays. Based on serial CT scans, chemical pancreatectomy led to the loss of pancreatic volume. Immunohistochemistry and transmission electron microscopy demonstrated exocrine pancreatic ablation with endocrine islet preservation. Importantly, chemical pancreatectomy did not increase pro-nociceptive markers in harvested dorsal root ganglia. Also, chemical pancreatectomy improved insulin secretion to supranormal levels in vivo and in vitro. Thus, this study may provide a foundation for translating this procedure to patients with chronic pancreatitis or other conditions requiring a pancreatectomy.
## Introduction
Chronic pancreatitis affects over 250,000 people in the United States and millions worldwide1–3. The global incidence of chronic pancreatitis is ten cases per 100,000 individuals of the general population annually, with a prevalence of 42 per 100,000 persons. Longstanding chronic pancreatitis leads to debilitating chronic pain, exocrine pancreatic insufficiency, and a brittle form of diabetes4. These long-term sequelae of chronic pancreatitis are challenging to treat and expensive to manage over a patient’s disease progression. Chronic pain often requires treatment with opioids, frequently leading to dependence and addiction5,6. In addition, patients often have exocrine pancreatic insufficiency necessitating lifelong pancreatic enzyme supplementation. Pancreatogenic diabetes occurs in approximately $80\%$ of patients with chronic pancreatitis, and it is difficult to treat7. The inflammatory environment of chronic pancreatitis is thought to lead to the loss of both insulin-producing beta-cells and glucagon-producing alpha-cells. Typically, glucagon reverses hypoglycemia; however, in patients with pancreatogenic diabetes, a small insulin overdose resulting in hypoglycemia may be life-threatening8. Other severe complications of chronic pancreatitis include pseudocyst formation, pancreatic ascites, metabolic bone disease, splenic vein thrombosis, and biliary obstruction9. Furthermore, patients with chronic pancreatitis have approximately an 8-fold increased risk of developing pancreatic ductal adenocarcinoma10–12.
In addition to managing the sequelae of chronic pancreatitis, more invasive options for treatment are often necessary, including major surgeries such as partial or total pancreatectomy that ultimately lead to a loss of islets and altered endocrine function13. Total pancreatectomy with islet auto-transplantation (TPIAT) has emerged as a treatment that addresses chronic pancreatitis-associated pain while attempting to preserve endocrine function, but it remains a procedure with high morbidity, substantial islet loss during islet isolation, and a $5\%$ 30-day mortality14–16. Unfortunately, many continue to have pain after surgery. Approximately 20–$40\%$ of adults continue to require analgesics, and 59–$70\%$ of these patients are insulin dependent after surgery17.
Our lab recently developed a novel therapeutic option to treat chronic pancreatitis by using an infusion of acetic acid (AcA) into the pancreatic duct. In mice, this procedure led to near-complete ablation of the exocrine pancreas while preserving the islets18. This technique, termed a “chemical pancreatectomy,” also relieved chronic pancreatitis-associated pain behaviors in mice. Remarkably, it improved glucose tolerance and insulin secretion in vivo to supranormal levels. Chronic pancreatitis-associated diabetes is thought to occur due to the surrounding inflammation. Thus, removing the offending exocrine pancreas may protect islets from further damage, preserving and even improving existing endocrine function.
To translate this therapeutic approach to one day treat chronic pancreatitis in humans, we analyzed the histological, anatomical, and islet functional changes accompanying chemical pancreatectomy in normal non-human primates (NHPs). Anatomically, the primate pancreas is similar to the human pancreas. In humans, the pancreatic duct is routinely accessed via a non-surgical endoscopic technique called endoscopic retrograde cholangiopancreatography (ERCP). However, given the small size of the primates, our attempts at ERCP were unsuccessful in selectively cannulating the pancreatic duct, which necessitated a laparotomy similar to that required in mice. In NHPs, the chemical pancreatectomy demonstrated excellent ablation of the exocrine pancreas. There was neither overt pain nor nerve injury, and islets were preserved, exhibiting improved (supranormal) glucose homeostasis and insulin secretion like the mice.
## Gross evaluation and imaging of the pancreas in NHPs after chemical pancreatectomy
We have previously demonstrated successful ablation of the exocrine pancreas with preservation of the endocrine pancreas in mice with chronic pancreatitis. Herein, we further showed similar success in NHPs. The NHPs underwent intrapancreatic ductal infusion of normal saline (NS) or $2\%$ AcA (Fig. 1A). Iodine contrast infusion pancreatography ensured correct positioning of the catheter, allowing for infusion down the pancreatic duct without duodenal spillage or flow beyond our bulldog clamp on the common bile duct (Fig. 1B, Supplementary Figs. 1A-1B, Supplemental Video 1). Six months after the chemical pancreatectomy, pancreatography showed proximal filling of a short segment of the pancreatic duct with the remainder of the duct not visualized (Fig. 1C), indicating ductal obliteration. Six months after surgery, AcA-treated pancreata were atrophied compared to saline-infused pancreata, which appeared normal in size and looked pink to tan (Fig. 1D). The AcA-treated pancreas appeared as a scant membrane that laid as a thin veil over the splenic vein with grossly apparent clusters of islets, which were histologically confirmed (Fig. 1E).
A baseline CT of the NHP abdomen and pelvis was obtained before surgery to identify pancreatic or intraabdominal abnormalities and to calculate the pancreatic volume. *We* generated 3D reconstructions of the pancreas from the CT scans before surgery and at two weeks, eight weeks, and six months after surgery (Figs. 1F–1H, Supplemental videos 2–4). In the saline-treated NHPs, the pancreatic volume slightly increased two weeks after the infusion, which is consistent with edema of the pancreas following infusion (Fig. 1I–1J, Supplementary Fig. 2A). In the AcA-treated NHPs, pancreatic volume was reduced at two weeks. The CT-based volumetric analysis showed a significant loss of pancreatic volume in the AcA-treated pancreata compared to controls beginning eight weeks after the infusion. This volume loss persisted six months after the infusion (Figs. 1I–1K, Supplementary Fig. 2B). Compared to baseline, there was a $70\%$ reduction in the volume of the AcA-treated pancreas at eight weeks and six months (Fig. 1L and Supplementary Fig. 2C). At the six-month time point, the saline-treated and AcA-treated NHPs had comparable body weights (Supplementary Fig. 2D). However, pancreatic volume assessed by volumetric displacement and pancreatic volume standardized to body weight revealed a statistically significant decrease in pancreatic volume in AcA-treated NHPs compared to saline-treated NHPs (Figs. 1M–1N). Similarly, pancreatic weight after chemical pancreatectomy was significantly lower than the saline-treated group (Fig. 1O). Also, the CT scan-based pancreatic volume calculations were comparable to the actual volumes of the harvested pancreata as measured by volumetric displacement at pancreatic harvest (Supplementary Fig. 2E). A table with individual NHP body weights and pancreatic measurements is included separately (Supplementary Table 1).
## Histological changes of NHP pancreas six months after pancreatic intraductal infusion
H&E evaluation of the saline-treated NHPs after surgery showed normal exocrine and endocrine pancreas with normal neurovasculature and ducts (Figs. 2A–2B). However, histology after chemical pancreatectomy demonstrated fatty replacement of the pancreas, minimal residual exocrine pancreas, intact neurovasculature, and intact islets of Langerhans (Figs. 2A–2B). The saline-treated NHP pancreas demonstrated normal IHC of amylase, insulin, and glucagon throughout the head, body, and tail (Fig. 2C). After chemical pancreatectomy, immunostaining showed a small amount of amylase staining was present in the pancreatic head, but it was otherwise absent in the body and tail. Normal expression of insulin and glucagon was found throughout the pancreas (Fig. 2C), consistent with ablation of the exocrine pancreas and preservation of the endocrine pancreas. Histology-based quantification of the residual exocrine pancreas demonstrated ablation of approximately $85\%$ of the exocrine pancreas (Fig. 2D). In addition, Masson’s Trichrome staining and histological evaluation revealed less fibrosis in the AcA-treated NHPs at six months than the to AcA-treated at eight weeks (Figs. 2E–2F).
## EM imaging of the pancreas after chemical pancreatectomy in NHPs
EM evaluation of the AcA-treated NHP pancreas correlates with our IHC data. The saline-treated NHP pancreas showed normal ultrastructural features with normal-appearing exocrine pancreas containing zymogen granules, beta-cells containing beta granules, and normal mitochondria (Figs. 3A–3B). Two days after the chemical pancreatectomy, there was extensive damage to the exocrine pancreas consistent with cell injury and cell death, including enlargement and disruption of the endoplasmic reticulum, karyorrhexis and karyolysis (Figs. 3C–3D), vacuolization of the exocrine cells (Fig. 3E), mitochondrial swelling (Fig. 3F), and changes in nuclear morphology (Fig. 3F). Additionally, two days after AcA, the pancreas body demonstrated an absence of zymogen granules, while Islets of Langerhans appeared normal throughout the pancreas except for the periphery of some islets with dysmorphic nuclei (Fig. 3E). Six months after the chemical pancreatectomy, there were neither zymogen granules nor any evidence of tissue of exocrine origin in the body and tail, and there was extensive fatty replacement with prevalent adipocytes surrounding endocrine cells (Figs. 3G–3H).
Lastly, we evaluated the endothelial cells within the islets of Langerhans. The dense fenestrations characteristic of islet endothelial cells (thought to facilitate the exchange of glucose and hormones between the blood and islet endocrine cells) were intact when comparing the saline-treated and AcA-treated NHPs (Figs. 3I–3K).
Chemical Pancreatectomy does not alter pain, induce persistent nerve injury, or cause persistent pancreatitis in NHPs Based on observation and serial evaluations with the onsite veterinarian, the NHP pain behavioral assessment demonstrated no difference in the saline-treated NHPs compared to the AcA-treated NHPs. NHP pain assessment included evaluating NHP activity, food consumption, guarding of the surgical site, posture, changes in respiration, facial expressions, restlessness, and vocalizations19,20. For one to two weeks following surgery, all NHPs appeared to have discomfort associated with surgical pain. However, all NHPs required the same amount of pain medications (a combination of buprenorphine (opioid), ketoprofen (NSAID), and acetaminophen). Moreover, the AcA-treated NHPs did not require more pain medication when compared to the saline-treated NHPs (Fig. 4A). To further evaluate markers of nerve injury (Atf3) and neurogenic inflammation (Tac1), dorsal root ganglia (DRG) were harvested from saline-treated and AcA-treated NHPs six months after surgery. There was no statistical difference in the Atf3 or Tac1 mRNA levels when comparing the saline-treated and AcA-treated NHPs. In collaboration with another lab that studies acute kidney injury (AKI), we studied the effects of a laparotomy and nephrectomy on nerve injury and neurogenic inflammation. A nephrectomy results in the axotomy of afferent neurons located in the same dorsal root ganglia as those that innervate the pancreas. Thus, the AKI NHPs served as a positive control. Furthermore, when comparing the saline-treated and AcA-treated NHPs to the AKI NHPs, the AKI NHPs demonstrated higher levels of Atf3 and Tac1 mRNA, confirming the presence of nerve injury (Figs. 4B–4C) and indicating the absence of nociception and nerve injury in the saline-treated and AcA-treated NHPs.
Bloodwork obtained from the AcA NHPs revealed an elevation of the white blood cell count and an elevation of the serum amylase and lipase levels one day following surgery. However, pancreatic enzymes and the WBC count normalized within one week, indicating resolution of the pancreatic inflammation, and remained normal up to six months after surgery (Figs. 4D–4E, Supplementary Table 2). Significantly, lipase levels eight weeks and six months after chemical pancreatectomy were statistically lower compared to preoperative levels; this is presumably due to the acinar cells being the primary source of serum lipase. Lipase was also lower than those of corresponding saline-treated NHPs at eight weeks and six months; however, amylase levels did not differ between AcA-treated and saline-treated NHPs (Figs. 4F–4G). As expected, consistent with a good ablation of the exocrine pancreas, the NHPs developed some degree of pancreatic insufficiency. Upon evaluating serum levels of vitamin D and vitamin E (Fig. 4H), there were significantly lower levels of vitamin E in the AcA-treated NHPs compared to the saline-treated NHPs. Similarly, the average vitamin D level in AcA-treated NHPs was lower, but this was not statistically different from the saline-treated NHPs. These data may suggest a lack of absorption of lipids.
Consistent with the biochemical changes, two days after AcA treatment, there was an increased inflammatory infiltrate consisting of neutrophils and macrophages (Fig. 4I). Over time, this inflammation was markedly reduced, and there was no evidence of chronic inflammation, as denoted by a lack of T cells assessed via CD3 staining (Fig. 4I).
## Chemical pancreatectomy improves glucose homeostasis in NHPs
By the time of sacrifice, all NHPs had gained weight (Fig. 5A). We have previously shown an improvement in glucose tolerance following chemical pancreatectomy in rodents and NHPs18. To investigate the cause of the improved glucose tolerance in NHPs, we performed a hyperglycemic clamp to assess the dynamics of insulin secretion. During this study, we monitored the m-rate (“m” for metabolism), which represents the amount of glucose (mg) infused per NHP body weight (kg) per time (min). Thus, m-rate is an indirect measure of insulin levels/secretion. The hyperglycemic clamp six months after surgery revealed that the m-rate was significantly higher in AcA-treated NHPs than in saline-treated NHPs, indicating increased insulin secretion (Figs. 5B–5C). Additionally, the actual serum insulin levels measured during the hyperglycemic clamp were significantly higher in the chemical pancreatectomy group compared to saline-treated NHPs (Figs. 5D–5E). Next, islets were isolated from NHPs six months post-surgery for in vitro glucose-stimulated insulin secretion (GSIS) and perifusion studies. There was no difference in the islet weight between the two groups (Fig. 5F). We found that the islets isolated from the AcA-treated group had increased insulin content compared to those from saline-treated (Fig. 5G). In vitro, the GSIS assay did not reveal a statistical difference between the saline-treated and AcA-treated islets, but there was a statistical trend ($$P \leq .0759$$) upon exposing islets to high-glucose during the GSIS (Fig. 5H). Here, a statistical significance was likely not found because of the small number of animals. The islet perifusion study, a more comprehensive study that illustrates ex-vivo dynamic insulin secretion, demonstrated increased insulin secretion in the chemical pancreatectomy islets compared to saline-treated islets (Figs. 5I–5J). Next, we calculated HOMA-IR as a marker for insulin sensitivity (Fig. 5K). There was no difference in insulin sensitivity between the AcA-treated NHPs compared to saline-treated NHPs, indicating that the improved glucose tolerance in the AcA-treated NHPs is mainly due to improved beta-cell function. Lastly, we evaluated fasting serum c-peptide levels in saline-treated NHPs and AcA-treated NHPs preoperatively and postoperatively. We found no difference between any of the groups (Fig. 5L).
## Feasibility of chemical pancreatectomy via ERCP in cadavers
To assess the feasibility of this approach in humans, we performed intrapancreatic AcA infusion via ERCP in human cadavers. First, the common channel was cannulated, followed by selective cannulation of the pancreatic duct under fluoroscopic guidance (Supplementary Fig. 3A-C, Supplemental Videos 5–7). Upon verification of correct positioning, we infused a solution of methylene blue and $2\%$ AcA, with a volume of 1.2cc/mL of pancreatic volume (estimating the average pancreas is 70mL). We immediately harvested the pancreas and found it uniformly blue and edematous, consistent with perfusion of the entire pancreas (Supplementary Fig. 3D). Importantly, we found that there was no leakage from the minor duct into the duodenum.
## Discussion
Chronic pancreatitis plagues many individuals, and it remains challenging to treat. Current therapeutic options are limited and often sacrifice endocrine function for pain control21. Herein, we explored using a chemical pancreatectomy as a potential new therapy to ablate the exocrine pancreas, preserve the endocrine pancreas, and improve glucose homeostasis.
In this study, we found that chemical pancreatectomy occludes the pancreatic ductal system (Fig. 1C). This likely reflects distal ablation of the pancreatic duct. The proximal pancreatic duct may remain patent because of persistent drainage from the proximal, unablated pancreas, or because of the larger caliber of the proximal pancreatic duct. Ablation of the distal pancreatic duct could explain the failure of the exocrine pancreas to regenerate.
Grossly and histologically, we see that the degree of fibrosis is significantly decreased by six months after chemical pancreatectomy compared to eight weeks. Interestingly, we previously saw no significant fibrosis in the mice18. This discrepancy likely reflects the difference in the pancreatic architecture and consistency, with the primate pancreas normally containing a more collagenous interstitium, rendering it firmer than the mouse pancreas. Over time, however, we see a significant reduction in the fibrosis, with the six-month time-point showing a similar degree of ablation and minimal residual fibrosis in the NHP pancreas as we previously saw with the mouse pancreas.
One day after the pancreatic intraductal infusion, we observed a rise in the white blood cell count, amylase, and lipase with subsequent normalization within one week. Lipase is a more specific marker for pancreatic injury than amylase because several other organs produce amylase22,23; thus, seeing a decrease in the lipase below baseline in the AcA-treated NHPs could be used as a marker to assess the degree of exocrine pancreas ablation. Additionally, in view of our data, postoperative serial pancreatic imaging with CT or MRI could also be used as a tool to evaluate the ablation of the exocrine pancreas after chemical pancreatectomy.
Given that the primary purpose of the chemical pancreatectomy is to one day alleviate pain in patients with chronic pancreatitis, we confirmed that the AcA-treated NHPs returned to their baseline behavior and were not different from saline-treated NHPs. In brief, we showed that an AcA infusion did not cause or exacerbate pain. Additionally, the NHPs maintained a normal appetite and continued to gain weight after surgery. Importantly, we are hopeful that this therapy will translate well to those with pain secondary to chronic pancreatitis and prevent the development of widespread central sensitization, which affects 21–$28\%$ of patients24,25.
We envision that chemical pancreatectomy may not only prevent chronic pancreatitis-related diabetes (CPRD) in chronic pancreatitis patients but may even reverse it. The pathophysiology of CPRD is thought to be related to the toxic milieu of the pancreas combined with extensive fibrosis, which then damages islets and inhibits their function26. Previously, we have found improved glucose homeostasis in mice and NHPs following a chemical pancreatectomy18. To further study the improved glucose homeostasis, we employed the hyperglycemic clamp, the gold standard for evaluating insulin secretion27, which showed a significant improvement in insulin secretion in AcA-treated NHPs compared to saline-treated NHPs. We saw this same pattern of improved islet function six months after surgery in AcA-treated NHP islets in vitro using GSIS and islet perifusion. Thus, the improvement in glucose homeostasis associated with the loss of the exocrine pancreas is possibly secondary to the loss of an inhibitory factor secreted by the exocrine pancreas, which functions in a paracrine manner. This substance may normally limit endocrine function, and once the exocrine tissue is removed may then allow the islets to have enhanced function. Given the longstanding improvement in glucose homeostasis observed in mice (8 weeks was the latest time-point evaluated) and NHPs (six months was the latest time-point evaluated), chemical pancreatectomy may improve glucose tolerance in patients with CPRD.
In NHPs, chemical pancreatectomy was associated with few complications. One complication we saw was duodenal narrowing within four weeks of surgery. In NHPs, the head of the pancreas is typically in contact with 50–$60\%$ of the circumference of the duodenum, significantly more than in humans. Once this portion of the pancreas becomes fibrotic and edematous, there can be a significant narrowing of the duodenum. This problem is less likely to occur in humans because the human pancreas is in contact with a smaller arc around the duodenum. Additionally, we found that one of 21 ($5\%$) NHPs developed a distal common bile duct stricture and was managed with IV fluids and prophylactic antibiotics. This complication was likely related to infusing into the common channel, where a portion of our infusion inevitably travels into the common bile duct. In humans, this should not happen because a proceduralist will selectively cannulate the pancreatic duct at the time of ERCP, and we would envision a sphincterotomy at the time of the ERCP.
Postoperatively, all NHPs (saline-treated and AcA-treated) develop elevation of pancreatic enzymes; however, optimal postoperative care, including pain medication, anti-emetics, and IV fluids, has generally facilitated clinical recovery. In humans, intensive care unit level monitoring and treatment would allow expeditious treatment of any periprocedural pancreatitis. Interestingly, the AcA-treated NHPs had a much higher rise of their pancreatic enzymes, likely reflecting a more global uncontrolled release as opposed to a localized release secondary to injury as in the saline-treated NHPs. This likely represents a sudden “fixation” and near immediate death of the majority of the acinar and duct cells.
Another nonspecific surgical complication that we encountered was infection. In three of 21 animals, the CT scans demonstrated an intraabdominal abscess. These NHPs were taken back to the operating room for a repeat laparotomy and abdominal washout. Also, two of 21 NHPs developed central line infections, requiring removal of the central lines. Both sets of NHPs did well after their respective interventions, and both sets of complications seem to be related to the laparotomy and duodenotomy and the less-than-ideal aseptic conditions of the animal surgery.
Ideally, this procedure will be translated into humans and carried out via an ERCP. To evaluate the feasibility of this translation, we performed an ERCP in human cadavers. Importantly, we could infuse a large volume entirely through an ERCP approach without surgical intervention; given the blue and edematous appearance of the pancreas post-ERCP, we believe we successfully perfused the entire pancreas. Of note, ERCP-associated pancreatitis occurs in approximately $7\%$ of patients28. Because we are immediately “fixing” the tissue upon infusion, we expect the incidence of clinically significant pancreatitis to be much lower.
Upon translation to humans, patients would undergo preoperative imaging in the form of a secretin-enhanced magnetic resonance cholangiopancreatography (MRCP) and possibly ERCP. MRCP would allow for the calculation of preoperative pancreatic volume and provide an overview of ductal anatomy, which would be essential for identifying areas of strictures and stones. ERCP would allow for intervention in the form of lithotripsy or stent placement, which may be necessary before a chemical pancreatectomy. This procedure may require surgical laparoscopic assistance due to, for example, an inability to selectively cannulate the pancreatic duct, requiring infusion through the common channel. Thus, laparoscopically, we would temporarily clamp the common bile duct. Additionally, we have seen that some infusate permeates through the pancreas into the peritoneal cavity and the retroperitoneum. We have found that this permeated infusate is diluted quickly based on pH analysis (Supplementary table 3). However, one may consider performing a laparoscopic washout of the infusate. Patients with prior pancreatic surgery or any setting where an ERCP could not be performed would be a relative contraindication, considering the same procedure could be performed through a surgical approach as we have described in mice and NHPs.
Approximately 20–$60\%$ of chronic pancreatitis patients develop exocrine pancreatic insufficiency and typically require enzyme replacement in the form of pancrelipase, which would have to be continued after chemical pancreatectomy29. Consistent with pancreatic insufficiency, we found that the AcA-treated NHPs had lower vitamin D and vitamin E levels compared to saline-treated NHPs. Those not previously taking pancrelipase would likely require initiation after chemical pancreatectomy, and they may require additional vitamin supplementation. Our NHPs remained on a regular diet and required neither enzyme supplementation nor placement on an elemental diet. We found that in most of the NHPs that underwent a chemical pancreatectomy, a small amount of normal exocrine pancreas remained in the head of the pancreas. This is likely due to NHP pancreatic ductal anatomy differing from human pancreatic ductal anatomy. NHPs may have small ductal tributaries draining directly into the duodenum separate from the main ductal system such that they were unaffected by the acetic acid infusion. Notably, there was no regeneration of the exocrine pancreas by six months after acetic acid as evidenced by the histology and immunostaining of the pancreas.
A limitation of our study is that we did not perform this therapy in a model of chronic pancreatitis in NHPs. Unfortunately, to the best of our knowledge, there is no published model of chronic pancreatitis in NHPs, forcing us to carry out our experiments in normal NHPs. Additionally, given the small size of the NHPs, we were unable to perform the procedure via an ERCP. However, given our success in human cadavers, we do not expect there to be significant difficulties in translating this procedure to humans.
## Conclusion
Here, we demonstrated the efficacy of chemical pancreatectomy in ablating the exocrine pancreas in NHPs, validating our pilot study in NHPs. Furthermore, we tested the feasibility of translating this procedure to humans via an ERCP. We envision that one day in the near future, clinicians will treat chronic pancreatitis differently, utilizing a chemical pancreatectomy, and these patients may prevent or improve their pancreatogenic diabetes.
## Animal manipulation: NHPs
Cynomolgus macaques from both sexes (4–6 years old, weighing 4.5–7.0kg) were purchased from Alpha Genesis Inc. NHPs were quarantined for 30 days and allowed to acclimate 1–2-weeks thereafter. NHPs were housed with a 12-hour light/12-hour dark cycle with lights on from 7 am-7 pm. Animals had free access to water and a diet of biscuits, forage mix, fruits, and vegetables. Endocrine assays were performed on males.
## Pancreatic ductal infusion in NHPs
Surgeries were performed as previously described18. Animals were put in a jacket and tether system for two weeks before surgery to allow the NHP to acclimate to the system. Before surgery, the NHP was sedated using ketamine (10 mg/kg intramuscular), intubated, and administered inhalational anesthesia with isoflurane. Following a tunneled central venous catheter placement into the right internal jugular vein (Supplementary Fig. 1C), we performed a midline laparotomy and placed a temporary clamp on the common bile duct to prevent infusion into the liver and gallbladder. Next, a duodenotomy was performed to allow for the identification of the major papilla. We then cannulated the common pancreaticobiliary channel using an umbilical catheter and secured the catheter using a bulldog clamp, preventing backflow into the duodenum. Subsequently, we infused methylene blue solution, identified any solution leakage, and controlled this with additional bulldog clamps. The contrast was injected through the catheter, and fluoroscopy was used to confirm adequate placement of the catheter and patency of the pancreatic duct. Next, we infused a volume of 1.7mL/kg of $2\%$ AcA or normal saline (NS) via an umbilical catheter at a rate of 2 mL/min. Acetic acid was then allowed to dwell within the pancreas for 10 minutes before removing the catheter and clamps and closing the duodenotomy. The laparotomy was then closed, and the jacket and tether system reapplied. NHPs received pain medication, ondansetron, famotidine, and IV fluids postoperatively until adequate oral intake was re-established.
## Acute Kidney Injury (AKI)
In collaboration with a separate lab that studies AKI in NHPs, using age-matched Cynomolgus macaques, ischemia-reperfusion was performed by clamping the right kidney for approximately one hour while a left nephrectomy was performed. These NHPs were sacrificed after seven days to evaluate the kidney for histology. The lab allowed us to evaluate the pancreata of these NHPs and evaluate them for markers of nerve injury. Additionally, these pancreata were used as controls for in vitro islet studies.
## Hyperglycemic clamp
The hyperglycemic clamp was completed in NHPs as previously described30,31. Following overnight fasting, NHPs were sedated (ketamine, 10mg/kg intramuscular). Two vascular catheters were placed: one for the continuous infusion of the $20\%$ dextrose solution; the second to collect blood at serial time points for serum insulin measurements. Capillary blood glucose was measured from the NHP’s tail tip serially. The capillary blood glucose concentration was raised to 125 mg/dL above baseline and maintained at that level for the duration of this procedure through adjustments of the infusion rate every five minutes. The glucose solution was continuously infused throughout the experiment, with adjustments in the rate calculated every five minutes to maintain the desired glucose level. Serial blood samples were collected for the designated time points for serum insulin measurements.
## NHP islet isolation
The islet isolation protocol was adapted for NHP pancreata32. Total pancreaticoduodenectomy was performed following NHP sacrifice. Harvested pancreata were transferred to cold University of Wisconsin (UW) solution with cold ischemia time averaging 20 minutes. After removing non-pancreatic tissue, the pancreatic tissue was washed in an antibiotic solution and weighed. The pancreatic neck was transected, and catheters were placed proximally and distally. Both catheters were secured with suture. We infused a blend of exogenous enzymes [collagenases and neutral proteases (Vitacyte)] into the pancreas and then transferred them into the Ricordi digestion chamber (Biorep Technologies Inc.) for mechanical disruption by shaking. After digestion, islets and pancreatic cells were washed in a cold RPMI solution supplemented with human serum albumin. The islets were purified using a polysucrose discontinuous gradient, washed, hand-picked, and cultured at 37°C overnight before performing functional analyses.
## In vitro Glucose-stimulated insulin secretion (GSIS)
Isolated islets (AKI NHP $$n = 2$$, control NHP $$n = 1$$, AcA NHP $$n = 2$$) recovered overnight at 37°C in CMRL-1066 media containing $10\%$ fetal bovine serum and 2mmol/l L-glutamine. GSIS was performed as previously described18. Briefly, groups of 30 islets per NHP were incubated in 2.8 mM glucose for 30 min at 37°C to establish a stable basal insulin secretion and were then washed with Krebs buffer twice. The islets were transferred into a new well containing 2 ml of 2.8 mM glucose solution for 30 min at 37°C, and 100 μl of media was collected for time point 1. The islets were then transferred into a new well containing 2 ml of 20 mM glucose solution for 30 min at 37°C, and 100 μl of media was collected for time point 2. The islets were then recovered for protein quantification. Insulin levels in the collected media were measured using the human insulin ELISA kit (ALPCO) and were normalized to the protein content.
## Islet perifusion assay
Isolated islets (from the same groups as above) recovered overnight in CMRL-1066 medium (Gibco) containing $10\%$ fetal bovine serum and 2 mmol/l L-glutamine at 37°C. Thirty islets per NHP were placed in a dynamic perifusion system (Amersham Biosciences AKTA FPLC System) as previously described33. To summarize, the perifusion was performed using Krebs buffer with 2.8 mM glucose at a flow rate of 1 ml/min for 30 minutes to establish stable basal insulin secretion. Next, the islets were perifused with 2.8 mM glucose for 10 minutes, and fractions of 500 μl were collected every 30 seconds. Then, the glucose concentration was increased to 20 mM, and fractions of 500 μl were collected every 30 seconds for 20 minutes. Finally, the islets were perifused with 30 mM KCl, and fractions of 500 μl were collected every 30 seconds for 10 minutes. After the perifusion, the islets were recollected from the column for protein quantification. The insulin in the effluent was measured as described above. The fractional insulin secretion rate was calculated as secreted insulin per minute normalized to the protein content.
## Measuring insulin content from isolated islets
Thirty equal-sized islets per NHP were incubated in 100μL acid/ethanol ($75\%$ ethanol and 0.15M HCl) at 4°C overnight as previously described33. Gentle rotation was used to extract insulin, followed by centrifugation at 14,000 rpm for 10 minutes. The supernatant was diluted to a 1:50 ratio, and the insulin content was measured as above and normalized to the total islet protein content.
## Assessment of pancreatic volume via computed tomography (CT) scan
NHPs were sedated and prepared as before surgery18, oral contrast (gastrografin) was administered via an orogastric tube, and IV contrast was administered via an IV catheter. CT scans were performed preoperatively, two weeks, four weeks, eight weeks, and six months after surgery. Two separate investigators measured pancreatic volume by creating a 3-D reconstruction of the pancreas using Vitrea software as described34. A radiologist verified these measurements.
## Measurement of serum pancreatic enzymes, comprehensive metabolic profile (CMP) and complete blood count (CBC), vitamin D levels, vitamin E levels
Blood was collected to measure pancreatic enzymes, CMP, and CBC as previously described18. Blood was collected from the central line or peripherally from the saphenous vein after the placement of a peripheral angiocatheter. Pancreatic enzymes and comprehensive metabolic profile were measured using the Dimension Vista 500 chemistry analyzer (Siemens Medical Solutions USA Inc.). Serum vitamin D and vitamin E levels were measured by IDEXX BioAnalytics.
## Tissue processing and histology
Pancreatic tissues were processed and evaluated for IHC as described18. Pancreas samples were fixed with $4\%$ paraformaldehyde (PFA) for 24 hours at 4°C. The tissues were then embedded into paraffin, and 5 μm sections were cut. For IHC, antigen retrieval was performed as appropriate using heat and citrate buffer as described. Slides were incubated with the following primary antibodies: Guinea pig insulin (Abcam ab195956, $\frac{1}{500}$), Rabbit amylase (Sigma a8273, $\frac{1}{300}$), Mouse glucagon (Abcam ab10988, $\frac{1}{1000}$) at 4°C overnight. On the subsequent day, slides were incubated with fluorescent-conjugated (CY2, CY3, CY5) secondary antibodies at a concentration of $\frac{1}{300}$ (Jackson ImmunoResearch Labs) for 1 hour at room temperature. Mounting and nuclear staining were performed using Fluoroshield with DAPI (Sigma-Aldrich). Additionally, five μm sections of paraffin-embedded pancreata were stained with H&E and Masson’s Trichrome stain for histological assessment.
## Transmission electron microscopy
We performed transmission electron microscopy (TEM) as previously described18. The tissue is fixed in MJK solution and rinsed three times for fifteen minutes each in a 0.1 M Sodium Cacodylate buffer. The tissue is then placed in a $1\%$ osmium tetroxide for 1 hour. The tissue is then rinsed with $50\%$ ethanol, and then sequentially dehydrated in $50\%$ ethanol, $75\%$ ethanol, and $95\%$ ethanol for 15 minutes each. Lastly, the tissue is placed in $100\%$ EtOH (15 minutes × 2). Next, the tissue is placed in propylene oxide for (15 minutes × 2). Then the tissue was preinfiltrated with a 2:1 propylene oxide resin for 1 hour, then 1:2 propylene oxide resin for 2 hours, and then pure resin for 1 hour. Lastly, the tissue is embedded and polymerized at 60 degrees C overnight. The embedded tissues were sectioned with a Leica EM UC6 ultramicrotome at a thickness of 90 nm and collected on copper mesh grids. The following day, the tissue is stained with alcoholic uranyl acetate for five minutes and then Reynolds lead citrate for five minutes. TEM imaging was performed on a Philips EM 208 microscope at 60 kV using a Hamamatsu digital camera.
## Isolation of RNA and qPCR
DRG (T8–T12) were harvested and snap-frozen. RNA was isolated using Trizol reagent (Invitrogen). 0.5mg RNA was used to synthesize cDNA using iScript cDNA synthesis kit (BioRad). Primers were designed using NCBI and targeted to *Macaca fascicularis* Gapdh (accession no. KM491710), Atf3 (XM_005540783), and Tac1 (AB220474). Primers were validated using Monkey Universal Reference cDNA (Zyagen). Gel electrophoresis was used to confirm that PCR products were of the predicted size without dimerization or nonspecific bands. Sybr green PCR amplification (Universal Sybr Green Supermix, Bio-rad) was performed using a Bio-Rad CFX Connect real-time system. After amplification, a dissociation curve was plotted against the melting temperature to ensure the amplification of a single product. All samples were run in duplicate. The negative control was the amplification of a control reaction product (product of reverse transcription of master mix in the absence of template cDNA). The relative fluorescence of SYBR green was compared with a passive reference for each cycle. Cycle time (Ct) values were determined via regression and recorded as a measure of initial template concentration. Relative changes (ΔCt) in mRNA levels were calculated using Gapdh as a reference standard. Fold expression was determined using the ΔΔCt method.
## Endoscopic retrograde cholangiopancreatography (ERCP) in cadavers
We inserted a side-viewing endoscope into the oropharynx of the supine cadaver, advanced through the esophagus, and into the stomach. The scope was then passed through the pylorus and into the duodenum. The ampulla was identified and cannulated. The cadaver was then repositioned into the prone position, and fluoroscopy was used to ensure proper position. Next, under fluoroscopic guidance, contrast was used to identify the pancreatic ductal system. We then infused a $20\%$ methylene blue $2\%$ AcA solution into the pancreas. We chose to infuse a volume of 1.2cc/mL of the pancreas (estimating the average pancreas is 70mL).
## Graphics
Gross images were taken with a Sony Macro Lens and camera. Figure 1A was created with BioRender.com. IHC images were captured using Leica STELLARIS 5 confocal system (using the 10x objective) with Leica Application Suite X (LAS X) at the CHP Rangos imaging core facility. Brightfield imaging was obtained using the Thermo Fisher EVOS M7000 system (using the 10x and 20x objectives) with EVOS M7000 Software revision 2.0.2094.0. Whole H&E slides were scanned using the Leica Aperio AT2 system (using the 40x objective) and subsequently evaluated using Aperio ImageScope 12.4.3. Images were optimized using Adobe Photoshop (version 23.3.1). Figures were compiled using Adobe Illustrator (version 26.2.1).
## Quantification of the Exocrine Pancreas
NHP pancreas slides were scanned using the EVOS m7000 Software using the 20x objective. ImageJ was then employed to measure the total pancreatic area and the area of the exocrine pancreas, evaluating the head, body, and tail of saline-treated and AcA-treated NHPs.
## Quantification of Fibrosis
Masson’s Trichrome stained NHP pancreas slides were scanned using EVOS m7000 Software using the 20x objective. ImageJ was then utilized to quantify the areas of fibrosis across the body and tail of multiple saline-treated and Aca-treated NHPs.
## Study approval
The Animal Research and Care Committee at the Children’s Hospital of Pittsburgh and the University of Pittsburgh Institutional Animal Care and Use Committee approved all NHP experiments, which were carried out in accordance with their respective guidelines and regulations. All methods are reported in accordance with ARRIVE guidelines. The Humanity Gifts Registry Program provided the cadavers, and all cadaveric experimental protocols were approved by the University of Pittsburgh, and all methods involving the cadavers were carried out in accordance with the University of Pittsburgh guidelines and regulations. Informed consent was obtained from all donors or the next of kin by the Humanity Gifts Registry.
All authors had access to the study data as well as reviewed and approved the final manuscript.
## Calculations and statistics
Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) [(Fasting Glucose × Fasting Insulin)/22.5] was calculated35. AUC was calculated by the trapezoidal method. Data are displayed as mean±SD. Comparisons between groups were made using unpaired, paired 2-tailed t-tests, or 1- or 2-way ANOVA as indicated, followed by the indicated post hoc test for multiple comparisons. P ≤.05 was considered statistically significant. Statistical tests were conducted using GraphPad Prism, version 9.3.1.
## Funding:
This work was partially supported by the NIH (R01DK120698 to GKG).
## Availability of data and materials:
Data, analytic methods, and study materials available upon request. Please contact the corresponding author for additional information.
## References
1. Xiao A. Y.. **Global incidence and mortality of pancreatic diseases: a systematic review, meta-analysis, and meta-regression of population-based cohort studies**. *Lancet Gastroenterol Hepatol* (2016) **1** 45-55. DOI: 10.1016/S2468-1253(16)30004-8
2. Yadav D., Timmons L., Benson J. T., Dierkhising R. A., Chari S. T.. **Incidence, prevalence, and survival of chronic pancreatitis: a population-based study**. *Am J Gastroenterol* (2011) **106** 2192-2199. DOI: 10.1038/ajg.2011.328
3. Levy P.. **Estimation of the prevalence and incidence of chronic pancreatitis and its complications**. *Gastroenterol Clin Biol* (2006) **30** 838-844. DOI: 10.1016/s0399-8320(06)73330-9
4. Hart P. A., Conwell D. L.. **Chronic Pancreatitis: Managing a Difficult Disease**. *Am J Gastroenterol* (2020) **115** 49-55. DOI: 10.14309/ajg.0000000000000421
5. Dunbar E. K., Saloman J. L., Phillips A. E., Whitcomb D. C.. **Severe Pain in Chronic Pancreatitis Patients: Considering Mental Health and Associated Genetic Factors**. *J Pain Res* (2021) **14** 773-784. DOI: 10.2147/JPR.S274276
6. Adejumo A. C.. **Predictors, rates, and trends of opioid use disorder among patients hospitalized with chronic pancreatitis**. *Ann Gastroenterol* (2021) **34** 262-272. DOI: 10.20524/aog.2021.0579
7. Malka D.. **Risk factors for diabetes mellitus in chronic pancreatitis**. *Gastroenterology* (2000) **119** 1324-1332. DOI: 10.1053/gast.2000.19286
8. Makuc J.. **Management of pancreatogenic diabetes: challenges and solutions**. *Diabetes Metab Syndr Obes* (2016) **9** 311-315. DOI: 10.2147/DMSO.S99701
9. Ramsey M. L., Conwell D. L., Hart P. A.. **Complications of Chronic Pancreatitis**. *Dig Dis Sci* (2017) **62** 1745-1750. DOI: 10.1007/s10620-017-4518-x
10. Kandikattu H. K., Venkateshaiah S. U., Mishra A.. **Chronic Pancreatitis and the Development of Pancreatic Cancer**. *Endocr Metab Immune Disord Drug Targets* (2020) **20** 1182-1210. DOI: 10.2174/1871530320666200423095700
11. Lowenfels A. B.. **Pancreatitis and the risk of pancreatic cancer. International Pancreatitis Study Group**. *N Engl J Med* (1993) **328** 1433-1437. DOI: 10.1056/NEJM199305203282001
12. Kirkegard J., Mortensen F. V., Cronin-Fenton D.. **Chronic Pancreatitis and Pancreatic Cancer Risk: A Systematic Review and Meta-analysis**. *Am J Gastroenterol* (2017) **112** 1366-1372. DOI: 10.1038/ajg.2017.218
13. Bouwense S. A. W.. **Surgery in Chronic Pancreatitis: Indication, Timing and Procedures**. *Visc Med* (2019) **35** 110-118. DOI: 10.1159/000499612
14. Bellin M. D.. **How Durable Is Total Pancreatectomy and Intraportal Islet Cell Transplantation for Treatment of Chronic Pancreatitis?**. *J Am Coll Surg* (2019) **228** 329-339. DOI: 10.1016/j.jamcollsurg.2018.12.019
15. Rickels M. R., Robertson R. P.. **Pancreatic Islet Transplantation in Humans: Recent Progress and Future Directions**. *Endocr Rev* (2019) **40** 631-668. DOI: 10.1210/er.2018-00154
16. Magistri P.. **Autotransplantation of pancreatic islets. A single-center first experience**. *Ann Ital Chir* (2016) **87** 268-270. PMID: 27345954
17. Kempeneers M. A.. **Efficacy of total pancreatectomy with islet autotransplantation on opioid and insulin requirement in painful chronic pancreatitis: A systematic review and meta-analysis**. *Surgery* (2019) **166** 263-270. DOI: 10.1016/j.surg.2019.03.014
18. Saleh M.. **Chemical pancreatectomy treats chronic pancreatitis while preserving endocrine function in preclinical models**. *J Clin Invest* (2021) **131**. DOI: 10.1172/JCI143301
19. Miyabe-Nishiwaki T., Gris V. N., Muta K., Nishimura R., Mills D. S.. **Primate veterinarians’ knowledge and attitudes regarding pain in macaques**. *J Med Primatol* (2021) **50** 259-269. DOI: 10.1111/jmp.12537
20. Paterson E. A., Turner P. V.. **Challenges with Assessing and Treating Pain in Research Primates: A Focused Survey and Literature Review**. *Animals (Basel)* (2022) **12**. DOI: 10.3390/ani12172304
21. Treacy J.. **Evaluation of amylase and lipase in the diagnosis of acute pancreatitis**. *ANZ J Surg* (2001) **71** 577-582. DOI: 10.1046/j.1445-2197.2001.02220.x
22. Ismail O. Z., Bhayana V.. **Lipase or amylase for the diagnosis of acute pancreatitis?**. *Clin Biochem* (2017) **50** 1275-1280. DOI: 10.1016/j.clinbiochem.2017.07.003
23. Pieper-Bigelow C., Strocchi A., Levitt M. D.. **Where does serum amylase come from and where does it go?**. *Gastroenterol Clin North Am* (1990) **19** 793-810. PMID: 1702756
24. Faghih M.. **Pancreatic QST Differentiates Chronic Pancreatitis Patients into Distinct Pain Phenotypes Independent of Psychiatric Comorbidities**. *Clin Gastroenterol Hepatol* (2022) **20** 153-161 e152. DOI: 10.1016/j.cgh.2020.10.036
25. Phillips A. E.. **A clinically feasible method for the assessment and characterization of pain in patients with chronic pancreatitis**. *Pancreatology* (2020) **20** 25-34. DOI: 10.1016/j.pan.2019.11.007
26. Hart P. A.. **Type 3c (pancreatogenic) diabetes mellitus secondary to chronic pancreatitis and pancreatic cancer**. *Lancet Gastroenterol Hepatol* (2016) **1** 226-237. DOI: 10.1016/S2468-1253(16)30106-6
27. Stumvoll M., Fritsche A., Stefan N., Haring H. A. **60 minute hyperglycemic clamp is sufficient to assess both phases of insulin secretion**. *Horm Metab Res* (2000) **32** 230-232. DOI: 10.1055/s-2007-978626
28. Kang X.. **The Incidence and Severity of Post-ERCP Pancreatitis in Patients Receiving Standard Administration of NSAIDs: a Systematic Review and Meta-analysis**. *J Gastrointest Surg* (2022) **26** 2380-2389. DOI: 10.1007/s11605-022-05399-6
29. Machicado J. D., Chari S. T., Timmons L., Tang G., Yadav D.. **A population-based evaluation of the natural history of chronic pancreatitis**. *Pancreatology* (2018) **18** 39-45. DOI: 10.1016/j.pan.2017.11.012
30. Mather K. J.. **Precision and accuracy of hyperglycemic clamps in a multicenter study**. *Am J Physiol Endocrinol Metab* (2021) **320** E797-E807. DOI: 10.1152/ajpendo.00598.2020
31. Xiaoli Wang B. W., Guofeng Sun, Wu Jing, Liu Yongqiang, Wang Yixin (Jim), Xiao Yong-Fu. **Dysglycemia and Dyslipidemia Models in Nonhuman Primates: Part I. Model of Naturally Occurring Diabetes**. *J Diabetes Metab* (2015) **13** 13. DOI: 10.4172/2155-6156.S13-010
32. Linetsky E.. **Improved human islet isolation using a new enzyme blend, liberase**. *Diabetes* (1997) **46** 1120-1123. DOI: 10.2337/diab.46.7.1120
33. Saleh M.. **beta-cell Smad2 null mice have improved beta-cell function and are protected from diet-induced hyperglycemia**. *J Biol Chem* (2021) **297** 101235. DOI: 10.1016/j.jbc.2021.101235
34. Saisho Y.. **Pancreas volumes in humans from birth to age one hundred taking into account sex, obesity, and presence of type-2 diabetes**. *Clin Anat* (2007) **20** 933-942. DOI: 10.1002/ca.20543
35. Matthews D. R.. **Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man**. *Diabetologia* (1985) **28** 412-419. DOI: 10.1007/BF00280883
|
---
title: Neddylation is required for perinatal cardiac development through stimulation
of metabolic maturation
authors:
- Jianqiu Zou
- Wenjuan Wang
- Yi Lu
- Juan Ayala
- Kunzhe Dong
- Hongyi Zhou
- Jinxi Wang
- Weiqin Chen
- Neal L. Weintraub
- Jiliang Zhou
- Jie Li
- Huabo Su
journal: Cell reports
year: 2023
pmcid: PMC10029150
doi: 10.1016/j.celrep.2023.112018
license: CC BY 4.0
---
# Neddylation is required for perinatal cardiac development through stimulation of metabolic maturation
## SUMMARY
Cardiac maturation is crucial for postnatal cardiac development and is increasingly known to be regulated by a series of transcription factors. However, post-translational mechanisms regulating this process remain unclear. Here we report the indispensable role of neddylation in cardiac maturation. Mosaic deletion of NAE1, an essential enzyme for neddylation, in neonatal hearts results in the rapid development of cardiomyopathy and heart failure. NAE1 deficiency disrupts transverse tubule formation, inhibits physiological hypertrophy, and represses fetal-to-adult isoform switching, thus culminating in cardiomyocyte immaturation. Mechanistically, we find that neddylation is needed for the perinatal metabolic transition from glycolytic to oxidative metabolism in cardiomyocytes. Further, we show that HIF1α is a putative neddylation target and that inhibition of neddylation accumulates HIF1α and impairs fatty acid utilization and bioenergetics in cardiomyocytes. Together, our data show neddylation is required for cardiomyocyte maturation through promoting oxidative metabolism in the developing heart.
## Graphical abstract
## In brief
Zou et al. investigate the role of neddylation, a protein modification process, in the development and maturation of the heart during the perinatal stage. The results suggest that neddylation plays a crucial role in heart cell maturation and postnatal cardiac development by regulating the switch between glycolysis and oxidative metabolism.
## INTRODUCTION
During development, the heart undergoes a series of profound structural, morphological, and functional changes until it ultimately develops into a functionally competent adult organ. From midgestation to adulthood, terminally differentiated cardiomyocytes (CMs) undergo progressive maturation processes, including sarcomeric protein isoform switching, transverse-tubule network development, electrophysiological maturation, cell-cycle withdrawal, hypertrophic growth, and metabolic reprogramming.1,2 Disruption of these critical maturation processes can lead to congenital heart disease or predisposition to cardiomyopathies and heart failure in adult life,3–6 emphasizing the significance of CM maturation in cardiac physiology and disease. Importantly, knowledge gained from cardiac maturation research might help to optimize approaches to promote the maturation of induced pluripotent stem cell (iPSC)-derived CMs used in cardiac regenerative medicine.7 Therefore, there is broad interest in understanding the regulatory mechanisms that underpin CM maturation.
Cardiac maturation involves crucial metabolic reprogramming during the transition from the fetal to the postnatal stage. Fetal hearts rely predominantly on glycolysis to generate ATP due to the hypoxic environment. In response to increased oxygen tension and altered nutrition sources postpartum, perinatal hearts primarily utilize fatty acid oxidation to meet their rapidly increased energy demands, which persists throughout life.8 This metabolic maturation is accompanied by the expansion of mitochondrial number and size,9 densification of mitochondrial cristae,10 upregulation of genes involved in fatty acid metabolism and oxidative phosphorylation,11 and downregulation of glycolytic genes.12 Perturbation of the developmental metabolic shift has been shown to cause heart failure and perinatal lethality in mouse models13–17 and in humans,18 highlighting a critical need to understand the upstream mechanisms governing this process. Over the past decade, considerable progress has been made in uncovering transcriptional regulators crucial for CM metabolic maturation, including hypoxia-inducible factor 1α (HIF1α),6 estrogen-related receptors α and γ (ERRα and ERRγ),19 and the peroxisome proliferator-activated receptor γ coactivator 1α/β/peroxisome proliferator-activated receptor (PGC1α/β/PPAR) axis.5,14 In particular, downregulation of HIF1α-mediated hypoxic signaling appears to be an important physiological switch driving metabolic reprogramming during cardiac maturation.6,20 While HIF1α is highly expressed in fetal hearts and is essential for embryonic heart development by sustaining CM proliferation,21 its level rapidly decreases from day 14.5 in murine embryonic hearts6 and after birth,17,22 and persistent activation of HIF1α in perinatal hearts impairs metabolic reprogramming, leading to heart failure and perinatal lethality.6,20 In addition to transcriptional mechanisms, posttranslational modifications (PTMs) provide another layer of regulation for cells to regulate diverse cellular processes. In eukaryotic cells, proteins are subjected to over 300 PTMs in response to external physical or chemical stimuli, which significantly expand the diversity of the proteome. PTMs result in rapid changes in protein conformation, subcellular localization, assembly/disassembly of multiple protein complexes, and binding affinity to DNA and proteins. By doing so, PTMs can subtly or dramatically alter protein activity and function without triggering de novo protein synthesis at the transcriptional level, thereby saving energy and material resources. Despite the increasing appreciation of PTMs, such as phosphorylation, methylation, acetylation, O-GlcNAcylation, ubiquitination, and SUMOylation, in cardiac development and pathophysiology,23–26 many more known PTMs remain to be investigated in heart tissues.
Neural precursor cell expressed, developmentally downregulated 8 (NEDD8) is a ubiquitin-like protein that covalently modifies target proteins in a way analogous to ubiquitination.27 Conjugation of NEDD8 to target proteins, termed neddylation, is catalyzed by a NEDD8-specific E1-E2-E3 enzyme system and can be reversed by NEDD8 proteases.28–30 By modulating the function of its various substrates, such as cullin proteins, neddylation regulates diverse cellular processes and multiple pathophysiological states, such as tumorigenesis,31 metabolic disorders,32 liver dysfunction,33 and neural development.34,35 We previously demonstrated that an intact NEDD8 pathway is essential for cardiac homeostasis in adulthood,36–39 and dysregulation of neddylation is associated with cardiomyopathies in human and murine hearts.40 More recently, we reported that neddylation is downregulated in the developing mouse heart after postnatal day 7, a time window when CMs start exiting the cell cycle.41 Deletion of NEDD8-activating enzyme 1 (NAE1), a regulatory subunit of the only NEDD8 E1 that is essential for NEDD8 activation and conjugation, in the developing heart via αMHCCre causes ventricular non-compaction and heart failure, which is at least in part attributable to inactivation of Yes-associated protein (YAP) signaling and CM proliferation arrest.41 While these findings demonstrate a critical role for neddylation in cardiac chamber development in embryonic hearts, the downregulation of neddylation in postnatal hearts drove us to determine whether neddylation is dispensable for postnatal cardiac development and whether neddylation has a role beyond regulating CM proliferation.
In this report, we describe a series of studies designed to determine the role of neddylation in postnatal hearts. Using strategies to conditionally delete the gene encoding NAE1 in mice, we demonstrate that neddylation is necessary for proper control of CM metabolic transition and for normal peri- and postnatal cardiac maturation. Furthermore, we identify that neddylation is required to suppress HIF1α signaling by directly modifying HIF1α and its ubiquitin ligase cullin 2 (Cul2).
## Postnatal deletion of NAE1 induces dilated cardiomyopathy and heart failure
Perinatal lethality of αMHCCre-driven NAE1 knockout41 prevents using this mouse line to investigate the role of neddylation in postnatal cardiac development. We employed adeno-associated virus 9 (AAV9) expressing Cre recombinase under the control of the cardiac troponin T (cTnT) promoter42 to generate mosaic cardiac-specific NAE1 knockout mice. The AAV-Cre was injected into neonatal NAE1F/+ and NAE1F/F mice carrying a lineage-tracing reporter Rosa26mTmG allele43 (Figure 1A), which labels Cre-expressing cells with membrane-bound GFP (mG) and non-expressing cells with membrane-bound Tomato (mT) (Figures 1B and S1A). High-dose AAV-Cre (5 × 1011 viral genome copies [GC]/pup) transduced about $80\%$ of CMs and significantly reduced NAE1 transcripts and proteins, leading to a significant decrease in total neddylated proteins and neddylated Cul2, a well-known NEDD8 substrate, in NAE1F/F hearts compared with NAE1F/+ littermates (Figures 1D–1F), confirming inhibition of neddylation. Cardiac phenotypes of these mice were characterized by temporal echocardiography and morphological and gravimetric analyses. Compared with littermate NAE1F/+ mice, high-dose AAV-Cre induced progressive dilated cardiomyopathy and eventually heart failure in NAE1F/F mice by 6 weeks of age, evidenced by enlarged heart size, increased heart weight-to-body weight and lung weight-to-body weight ratios, increased left-ventricular chamber size, reduced left-ventricular wall thickness, and gradually deteriorating left-ventricular ejection fraction and fractional shortening (Figures 1G–1K). Quantitative real-time PCR demonstrated the upregulation of the cardiac stress markers Nppa and Nppb and the downregulation of Serca2a (Figure 1D). Histological analyses showed enlarged heart size and prominent cardiac fibrosis in NAE1F/F hearts (Figures S1B and S1C). Terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) staining, Evans blue dye (EBD) infiltration assay, and immunoblotting of cleaved caspase 3 did not identify evidence of pronounced CM cell death in 4- and 6-week-old NAE1F/F hearts (Figures S1D–S1F). Taken together, these data demonstrate that neddylation is indispensable for normal postnatal cardiac development.
## Deletion of NAE1 impairs CM maturation
Since the deletion of NAE1 occurred in the critical time window of CM maturation, we hypothesized that disruption of CM maturation underlies the pathogenesis of cardiomyopathy in NAE1F/F hearts receiving high-dose AAV-Cre. To rule out confounding effects secondary to pathological cardiac remodeling, we titrated AAV-Cre doses, which led to detection of Cre activity in ~$60\%$ and $40\%$ of CMs at medium (0.5 × 1011 viral GC/pup) and low (0.25 × 1011 viral GC/pup) doses, respectively (Figure 1C). Neither dose caused discernible cardiac remodeling in NAE1F/F mice, evidenced by comparable fibrotic areas, cardiac morphometric parameters, ejection fraction, and left-ventricular mass measured by echocardiography between NAE1F/F and NAE1F/+ mice at 4 weeks of age (Figures S1G and S1H). Thus, these hearts were used to assess the impact of NAE1 deletion on CM maturation.
To visualize T-tubule organization in live CMs in situ, we labeled T tubules with the plasma membrane dye FM 4-64FX through retrograde heart perfusion and performed in situ confocal imaging.44 We observed the T tubules to be drastically disorganized in mG+ CMs in NAE1F/F hearts infected with low-dose AAV-Cre, while those in neighboring mT+ CMs remained well aligned (Figure 2A). Quantification of FM 4-64FX fluorescence intensity showed a significant reduction in T-tubule contents in mG+ CMs vs. mT+ CMs (Figure 2C). Interestingly, we noted that the mG and mT signals perfectly matched those of FM 4-64FX (Figure 2B), likely due to the membrane-targeting sequence. Immunostaining of the T-tubule protein junctophilin-2 (JPH2) further confirmed that both mT and mG colocalize with JPH2 (Figures S2A and S2B), suggesting that the mT/mG reporter faithfully traces the T-tubule network. Examining mT/mG patterns in 4-week-old NAE1F/+ and NAE1F/F hearts receiving low-dose AAV-Cre revealed a remarkable disruption of T tubules in mG+ CMs from NAE1F/F hearts, but not in mT+ CMs from NAE1F/F hearts or in mG+ or mT+ CMs from NAE1F/+ hearts (Figure 2D). Quantification of T-tubule patterns showed a significant reduction of T-tubule integrity and regularity in mG+ CMs from NAE1F/F hearts (Figures 2E and 2F). Together these data suggest that neddylation is essential for T-tubule organization.
In addition to T-tubule development, CM maturation is accompanied by CM growth and switch of fetal to adult gene expression.1 We isolated CMs from low-dose AAV-treated NAE1F/F:mTmG hearts and found that mG+ CMs exhibited decreased cell area and increased length-to-width ratio compared with mT+ CMs, indicating an important role for neddylation in CM growth (Figure 2G). We further assessed the expression of the fetal and adult isoforms of genes involved in sarcomere and electrophysiological maturation. Quantitative real-time PCR analysis of medium-dose AAV-treated hearts revealed a significant downregulation of sarcomeric adult isoforms, including Myh6, Tnni3, and Myl2, and potassium channel gene Kcnj2 in NAE1F/F hearts compared with NAE1F/+ hearts. In contrast, the fetal isoforms of these genes, such as Myh7, Tnni1, Myl7, and Hcn4, were drastically upregulated (Figure 2H). Immunoblotting confirmed these changes at the protein level in medium-dose AAV-treated hearts (Figures 2I and 2J). Thus, these data identify an essential role for neddylation in CM maturation.
## Transcriptome analysis reveals defective maturation and metabolic dysregulation in NAE1-deficient hearts
To gain insights into how neddylation regulates CM maturation, we sought to conduct transcriptional profiling. Analysis of mouse hearts with AAV-Cre-mediated mosaic NAE1 deletion may underestimate the transcriptional changes, and isolation and sorting of viable adult mG+ and mT+ CMs from NAE1F/F hearts is challenging. Therefore, we chose to analyze the transcriptomes of αMHCCre-driven neonatal NAE1 knockout hearts (NAE1CKO) (Figure 3A). Principal-component analysis revealed patterns distinguished between NAE1CKO and littermate control groups (Figure 3B). *Differential* gene expression analysis identified 959 downregulated and 735 upregulated genes, with Nae1 being one of the top downregulated genes (Figure 3C, Table S1). Consistent with the findings in AVV-infected hearts, the ratios of fetal to mature myosin heavy chain isoforms (Myh7 to Myh6) and troponin I isoforms (Tnni1 to Tnni3) were also significantly increased in NAE1CKO hearts (Figure 3D). To further analyze the global impact of NAE1 deletion on the CM maturation status, gene set enrichment analysis (GSEA) was performed to assess the enrichment of differentially expressed genes among the previously identified mature (293 genes) and immature (354 genes) gene sets common in human and mouse.45 Genes upregulated in NAE1-deficient hearts were highly enriched for immature genes (normalized enrichment score [NES] = −2.32; Figure 3E), while downregulated genes were highly enriched for mature genes (NES = 3.38; Figure 3F). GSEA of the differentially expressed genes among the top 100 neonatal- or adult-specific genes identified in mouse heart3 also revealed a similar trend: upregulated genes were enriched for neonatal genes (NES = −1.04; Figures S3A and S3C), while downregulated genes were enriched for adult genes (NES = 1.31; Figures S3B and S3D). These data suggest that inhibition of neddylation results in persistent expression of genes associated with immaturity and inactivation of the transcriptional network required for CM maturation.
We next analyzed the RNA-sequencing (RNA-seq) data to identify biological processes enriched within the differentially expressed genes in NAE1-deficient hearts. Interestingly, the downregulated genes were highly enriched for oxidative phosphorylation and fatty acid metabolism, while the upregulated genes were enriched for more diverse cellular processes, including hypoxia signaling and glycolysis (Figure 3G). Specifically, many genes involved in fatty acid oxidation, mitochondrial fatty oxidative oxidation, and electron transportation chain were significantly downregulated in NAE1-deficient hearts (Figures 3H and S3E), whereas genes involved in glucose transport and glycolysis were either up- or downregulated (Figure S3F). Interestingly, many of these dysregulated metabolic genes were enriched in PPARα and HIF1α signaling (Figure 3I), suggesting their potential involvement in metabolic dysregulation. Screening of a list of key metabolic transcriptional regulators showed a significant downregulation of PERM1 and PPARα, but not others, such as HIF1α, in NAE1-deficient hearts (Figure S3G). These data identify a link between neddylation and cardiac metabolic maturation.
## Defective oxidative metabolism in NAE1-deficient hearts
Perinatal metabolic transition from glycolysis to oxidative metabolism is essential for CM maturation and postnatal cardiac development.8,46 Quantitative real-time PCR analysis showed that several genes involved in fatty acid utilization, such as Dgat2, Hadha, Mlycd, and Eci1, were significantly downregulated in medium-dose AAV-Cre-infected NAE1F/F hearts compared with NAE1F/+ hearts, whereas Igf1, a gene that stimulates glucose utilization, was markedly upregulated (Figure 4A). In contrast, we did not observe a significant change in stress markers (Nppa, Nppb, and Myh6) (Figure S4A), consistent with the lack of pronounced cardiac dysfunction in these hearts (Figure S1C). Moreover, western blotting revealed a decrease in the enzyme regulating fatty acid utilization (acyl-CoA synthetase long chain family member 1, ACSL1) and proteins involved in mitochondrial biogenesis (PGC1α and PERM1)47 in medium-dose AAV-Cre-infected NAE1F/F hearts (Figure 4B and 4C). Measurement of respiration function of isolated cardiac mitochondria showed a substantial reduction in basal and maximal mitochondrial respiration in NAE1F/F hearts (Figures 4D and 4E). In addition, the ATP content in NAE1F/F hearts also decreased to nearly half of the levels in NAE1F/+ hearts (Figure 4F). Since medium-dose AAV-Cre deletes NAE1 in ~$60\%$ of CMs, these metabolic changes may be underestimated, highlighting the impact of neddylation on cardiac metabolism.
Cardiac metabolism switching from glycolysis to oxidative metabolism starts at midgestation.6 We sought to determine whether NAE1CKO affects cardiac metabolic function. Quantitative real-time PCR analysis showed that NAE1CKO led to dramatically decreased mRNA levels of fatty acid oxidation genes, including Acadm, Cpt1b, and Atgl (Figure 4G). Ultra-structural analysis revealed a significant increase in lipid droplets in mutant CMs (Figures S4B and S4C). Oil red O (ORO) staining revealed increased accumulation of lipid droplets in NAECKO hearts compared with littermate control hearts (Figure S4D). Meanwhile, NAE1-deficient CMs also displayed a remarkable increase in degenerating/immature mitochondria with loose and fragmented cristae and mitophagic vesicles (Figures 4H and 4I), suggesting mitochondrial dysfunction. Thus, the in vivo findings from NAE1-deficient hearts support a critical role for neddylation in instigating cardiac oxidative metabolism.
## Inhibition of neddylation impairs fatty acid utilization in cultured CMs
To study whether neddylation regulates fatty acid utilization in a cell-autonomous manner, neonatal rat ventricular CMs (NRVCs) were treated with MLN4924 (MLN), a potent and specific NAE1 inhibitor with minimal effect on Ub or other Ub-like proteins,48 or transfected with siRNAs against both subunits (NAE1 and UBA3) of NAE to inhibit neddylation (Figures 5A and S5A). Since cultured CMs are prone to using glucose as a major energy source, NRVCs were treated with oleic acid (OA) to boost fatty acid utilization. Silencing of NAE and MLN treatment led to greater accumulation of lipid as measured by LipidTOX and BODIPY labeling and quantification of triglyceride contents (Figures 5C, 5D, and S5C–S5F). Mitochondrial stress test revealed that both treatments significantly impaired mitochondrial bioenergetics and resulted in significantly diminished basal and maximal respiration (Figures 5E and 5F). Palmitate oxidation stress test further revealed that silencing of NAE suppressed fatty acid utilization (Figures 5G and 5H). MLN treatment significantly blunted OA-induced upregulation of genes crucial for fatty acid utilization, such as Cpt1b and Atgl, and mitochondrial biogenesis, such as Pgc1a and Pgc1b, and attenuated OA-induced downregulation of the glycolytic gene Pkm, although it had little effect on Pparg and Hk2 expression (Figure 5I). MLN also significantly decreased the expression of PGC1α (a master regulator of mitochondrial biogenesis), CPT1B (carnitine palmitoyltransferase 1B; the rate-controlling enzyme of long-chain fatty acid β-oxidation), and ACSL1 (acyl-CoA synthetase long-chain family member 1; an isozyme of the long-chain fatty acid-coenzyme A ligase) (Figures S5A and S5B). Together, these data suggest that inhibition of neddylation represses fatty acid oxidative metabolism in CMs.
## Metabolomics analysis revealed altered metabolic intermediates in MLN-treated NRVCs
To understand the broader impact of neddylation on CM metabolism, we performed untargeted metabolomics analysis of cultured CMs with and without MLN treatment. Principal-component analysis (PCA) of the detected metabolites clearly distinguished between vehicle (Veh)- and MLN-treated cells (Figure S6A). Among 1,436 annotated metabolites (Table S2), 205 were significantly downregulated, while 208 were upregulated, in MLN-treated CMs (fold change [FC] > 1.5, adjusted $p \leq 0.05$; Figure S6B). Integrated pathway enrichment analysis of the significantly altered metabolites identified many metabolic pathways significantly affected by neddylation inhibition (Figure S6C and Table S2). Pathways with top enrichment included critical cellular glucose and fatty acid metabolism, such as fatty acid biosynthesis, carnitine synthesis, Warburg effect, and glycolysis, as well as multiple interconnected amino acid metabolic pathways. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of these metabolites also identified similar enriched pathways (Figure S6D). Individual metabolites whose levels were altered by treatment with MLN are shown in Figure S6E. Taken together, the results of the metabolomics analysis suggest that inhibition of neddylation has a global impact on the metabolic profile, especially in the pathways related to glucose and fatty acid metabolism in CMs.
## Persisting HIF1α signaling in neddylation-deficient CMs
Activation of HIF1α promotes glycolysis, suppresses fatty acid oxidation, and inhibits mitochondrial biogenesis and oxidative phosphorylation.49–51 Developmental downregulation of HIF1α in the heart is essential for perinatal metabolic transition and cardiac chamber development.6 The enrichment of dysregulated genes in HIF1α signaling (Figure 3I) prompted us to explore a possible link between neddylation and HIF1α signaling. Analysis of chromatin immunoprecipitation sequencing (ChIP-seq) data of HIF1α in embryonic day 12.5 mouse hearts21 showed that 155 ($9.1\%$) of the dysregulated genes in NAE1-deficient hearts were putative HIF1α targets (Figure 6A). Gene ontology (GO) enrichment analysis demonstrated that these 155 genes were mostly associated with fructose and pyruvate metabolism and mitochondria, as well as cardiac muscle contraction and morphogenesis (Figure 6A). Specifically, many HIF1α-regulated glycolytic, fatty acid oxidative, and mitochondrial genes were downregulated in NAE1-deficient hearts (Figure 6B). We confirmed that deletion of NAE1 caused a substantial accumulation of HIF1α proteins without altering its transcripts (Figures 6C and 7D). Consistent with previous reports,6,21 HIF1α was abundant in E9.5 mouse hearts but remarkably downregulated in E14.5 mouse hearts and mainly resided in the cytosol of CMs (Figures S7A and 6E). However, HIF1α accumulated in the nucleus of NAE1-deficient CMs at E14.5 (Figures 6E and 6F), a time point when cardiac morphology was comparable between control and mutant hearts.41 Moreover, Glut1, a glucose transporter that is a known HIF1α downstream target, was upregulated in mG+ (NAE1-deleted) CMs, but not in mT+ CMs, in E14.5 mouse hearts (Figures 7G and 7H), supporting the activation of HIF1α. Similarly, depletion of NAE1 via AAV-Cre also led to accumulation of HIF1α in postnatal hearts (Figures 6I and 6J) and downregulation of HIF1α target genes involved in fatty acid metabolism (Dgat2, Acaa2) and mitochondrial function (Ech1, Slc25a33, Coq9) (Figure 6K). HIF1α reporter assay showed that both silencing of NAE and MLN treatment increased HIF1α activity in H9C2 and HEK293 cells (Figures S7B–S7D). Taken together, these results suggest that inhibition of neddylation promotes accumulation and activation of HIF1a, which in turn represses the establishment of oxidative metabolism in the developing heart.
## Neddylation regulates HIF1α expression in Cul2-dependent and -independent manners
HIF1α expression is largely controlled at the posttranscriptional level. Under normoxic conditions, HIF1α is ubiquitinated by Cul2-von Hippel-Lindau (VHL) ubiquitin ligase and subsequently degraded by the proteasome.52 *Neddylation is* essential for cullin-RING ubiquitin ligase activity,53 and inhibition of neddylation via MLN or NAE1CKO robustly increased HIF1α in CMs (Figures 6D, 6I, 7A, and S7E). Interestingly, however, MLN and bortezomib (BZM; a 20S proteasome inhibitor) had synergistic effects in stabilizing HIF1α in CMs under both normoxic and hypoxic conditions (Figure 7A), suggesting that neddylation functions independent of Cul2 to control HIF1α expression. Immunoprecipitation of HIF1α under denatured conditions, which prevents non-covalent protein-protein interactions, identified neddylated species, which were absent in cells expressing conjugation-deficient NEDD8 mutant and reduced by MLN treatment (Figure 7B). Moreover, overexpression of NEDD8 E2 enzyme UBC12 also increased neddylated HIF1α (Figure 7C), further supporting HIF1α as a putative NEDD8 target. Furthermore, MLN remained effective in stabilizing HIF1α in Cul2-deficient CMs (Figures 7D and 7E). These data suggest that neddylation of HIF1α promotes its degradation in a manner that is independent of Cul2 ubiquitin ligase. Inhibition of HIF1α with echinomycin (Ech), a compound that inhibits HIF1α binding to DNA and thus its transcriptional activity,54 substantially reduced MLN-induced lipid accumulation in CMs (Figure 7F), and this was further confirmed by quantification of triglyceride (TG) content (Figure 7G). Thus, we propose that neddylation has dual roles in the regulation of HIF1α expression and that upregulation of HIF1α affects fatty acid utilization in neddylation-deficient CMs.
## DISCUSSION
In this study, we demonstrated that neddylation is required for CM maturation and perinatal cardiac development by ensuring developmental metabolic transition. Our findings support a model in which neddylation modifies HIF1α and Cul2, and neddylation of both proteins has a synergistic effect in promoting efficient HIF1α ubiquitination and degradation, which is essential for the establishment of oxidative metabolism in the developing heart (Figure 7H). Consequently, inhibition of neddylation leads to accumulation of HIF1α proteins and persistent HIF1α signaling, which represses fatty acid utilization in late gestational and postnatal heart, leading to defects in CM maturation and eventually heart failure. Thus, our study identifies neddylation as a crucial posttranslational mechanism regulating cardiac maturation and metabolism.
The biological functions of neddylation have been primarily described in the context of its impact on cell differentiation and proliferation,31–35 but its significance in terminally differentiated, postmitotic organs has not been investigated. Using AAV-TnT-Cre to achieve mosaic, postnatal gene deletion in the heart, our study defines a role for neddylation in CM maturation. This is evidenced by a disrupted T-tubule network, decreased CM cell size, and significantly dysregulated expression of maturation genes in neddylation-deficient CMs in the absence of overt cardiac dysfunction (Figures 2 and 3). Consistent with an arrest in cardiac maturation, mice with high-level deficiency of neddylation developed cardiomyopathy and heart failure before adolescence (Figure 1), and genetic deletion of NAE1 in embryonic heart at midgestational stage via αMHCCre induces heart failure and perinatal lethality.41 Notably, deletion of CSN8, a subunit of the deneddylase COP9 signalosome (CSN), in postnatal hearts via αMHCCre causes dilated cardiomyopathy and heart failure by 3 weeks of age.38 Transient inhibition of neddylation with MLN4924 during the first week after birth predisposes the heart to isoproterenol-induced pathological remodeling.55 Since perturbations of deneddylation or neddylation in these studies occurred in the critical time window of CM maturation, disruption of CM maturation may contribute to the observed cardiac phenotypes. ASB2, a Cul5-RING Ub ligase whose activity is regulated by neddylation, was shown to regulate CM maturation by facilitating sarcomere organization and formation of cell-cell junctions.56 Together, recent findings from our studies and others support a previously unrecognized mechanism regulating CM maturation and perinatal cardiac development.
Increasing evidence has suggested a critical role for a metabolic shift from glycolysis to oxidative metabolism in CM maturation. Perturbations of PGC1/PPAR signaling, an important pathway regulating mitochondrial biogenesis and maturation, disrupts CM maturation, whereas PGC1/PPAR activation improves the maturation of pluripotent stem cell-derived CMs.5 Similarly, disruption of ERRα and ERRγ, which function as critical transcriptional activators of metabolic genes in adult CM, impairs cardiac maturation and results in heart failure in the developing heart.19 Findings from this study establish neddylation as a crucial mechanism in the regulation of cardiac oxidative metabolism. We present in vitro and in vivo evidence showing that inhibition of neddylation disrupts the expression of metabolic genes, inhibits mitochondrial maturation and respiration, suppresses fatty acid utilization, and significantly alters metabolite profiles in CMs (Figures 3, 4, 5, and S3–S6), which collectively culminate in the arrest of CM maturation. Consistent with our results, inhibition of neddylation via deletion of NEDD8 or UBA3 in the liver suppressed mitochondrial oxidative phosphorylation and fatty acid oxidation, leading to hepatic steatosis.33 Interestingly, MLN4924 treatment and UBA3 knockdown were shown to enhance basal and maximal oxidative phosphorylation in cancer cell lines,57 suggesting that the effect of neddylation in metabolism could be cell-type dependent.
The spatiotemporal downregulation of HIF1α in the developing heart is crucial for the metabolic transition from glycolysis to oxidative metabolism and cardiac maturation. Despite its essential role in stimulating CM proliferation in hypoxic embryonic hearts,21 persistent HIF1α activation in postnatal hearts has been linked to heart failure and premature death, at least in part due to defective energy metabolism.17,58–61 Deletion of the E3 ubiquitin ligase VHL in the developing heart results in HIF1α hyperactivation, abrogating the developmental metabolic shift and impairing cardiac maturation and function, which can be rescued by concomitant deletion of HIF1α.6 Moreover, chronic perinatal hypoxia is sufficient to delay cardiac maturation.62 We observed accumulation of HIF1α (Figures 6 and 7) and deficits in oxidative metabolism (Figures 4 and 5) in neddylation-deficient hearts and cultured CMs, and inhibition of HIF1α attenuated lipid accumulation in neddylation-deficient CMs (Figure 7G). Thus, our data suggest that neddylation regulates developmental metabolic reprogramming, at least in part, by repressing HIF1α signaling the heart.
Identification of NEDD8 substrates is of utmost importance to elucidate the biological functions of neddylation. Consistent with the findings in non-cardiac cells,63–66 we confirmed that HIF1α is degraded by Cul2-VHL Ub ligase and is also a putative NEDD8 target in CMs (Figure 7). Interestingly, inhibition of neddylation in the absence of Cul2 remains effective in stabilizing HIF1α, suggesting that neddylation of HIF1α promotes its degradation. Thus, neddylation appears to regulate HIF1α by directly targeting Cul2 and HIF1α, respectively. While neddylation itself does not directly target the modified substrates for proteasomal degradation under basal conditions, it is reported that NEDD8 incorporates into the existing Ub chain under stress conditions to facilitate the degradation of ubiquitinated proteins,67,68 which may otherwise prevent the exhaustion of the ubiquitin machinery. Further experiments are needed to identify the neddylation site on HIF1α and elucidate the biological functions of HIF1α neddylation.
As a protein modification that targets diverse cellular proteins,69,70 neddylation may have multiple downstream effectors that coordinate CM metabolism. Other than HIF1α, mitochondrial electron transfer flavoproteins A and B (ETFA and ETFB), two key proteins responsible for relaying electrons in the electron transport chain, were identified as neddylated substrates, and neddylation increased the stability of ETFA and ETFB by preventing their ubiquitination and degradation in hepatocytes.33 PPARγ neddylation was reported to control the expression of genes involved in fatty acid storage and is essential for adipogenesis.32 Whether neddylation of these proteins regulates cardiac developmental metabolic reprogramming remains to be determined. It should be pointed out that the NEDD8 proteome is likely cell-type specific. Mapping the NEDD8 proteome landscape will provide mechanistic insights into how neddylation regulates cardiac homeostasis.
## Limitations of the study
There are still some limitations in this study. First, we cannot exclude the possibility that neddylation inhibition perturbs the Hippo-YAP pathway, as we reported previously,41 which may have effects on CM metabolism71 and the last round of proliferation and division.72 Second, single-cell or single-nucleus RNA-seq of cells isolated from hearts infected with medium- or low-dose AAV-Cre will provide precise transcriptomic changes in neddylation-deficient CMs, which may define the metabolic phenotype independent of ongoing cardiac remodeling. Third, neddylation may have a direct impact on multiple pivotal cellular processes beyond cardiac metabolism and maturation, such as CM contractility, T-tubule remodeling, and Ca2+ handling. Defects in these pathways may contribute to the cardiac phenotype observed in NAE1-deficient hearts.
## Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Huabo Su (hsu@augusta.edu).
## Materials availability
All reagents generated in this study will be made available on request from the lead contact with a completed Materials Transfer Agreement.
## Animals
All animal experiments were approved by the Augusta University Institutional Animal Care and Use Committee. A transgenic mouse line bearing a NAE1Flox allele was crossed with αMHCCre mice (the Jackson Laboratory, strain # 011038) to generate CM-restricted NAE1 knockout (NAE1CKO) mice. A transgenic mouse line bearing the Rosa26mTmG allele (the Jackson Laboratory, strain # 007676) was used for lineage tracing. These mice were maintained in the C57BL/6J inbred background for our studies.
During all experiments involving transgenic mice, experimental mice were randomly assigned into all groups, including sex (male and female), age (8–12 weeks for breeding/4, 6 weeks or neonatal P1 as specified in each figure), unless specified. The total estimate number of mice used in this article is ~180 mice (including embryonic stage mice) and ~120 rat pups for NRVC isolation. All mice were at healthy status before experiments. No subjects were involved in previous procedures except for AAV injections indicated in specific figures at age of P1. The housing conditions were well maintained at room temperature in the animal facility of Augusta University. The influence of sex was not determined for 4 or 6 weeks sacrificed mice. The influence of sex of P1 pups was considered as not significant.
## Ethical statement
Our studies did not include human participants, human data, or human tissue. All animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at Augusta University.
## Culture of neonatal cardiomyocytes and cell lines
Neonatal rat ventricular cardiomyocytes (NRVCs) were isolated using Neonatal Cardiomyocyte Isolation System (Worthington) following the manufacturer’s protocol.40 Briefly, neonatal rat hearts were minced into ~1 mm diameter and digested in $0.05\%$ Trypsin in 4°C overnight on a slow speed rocker. The other day, digested heart tissue was washed 4 times and subject to Collagenase digestion at 37°C for 40 min. NRVCs were next separated by pipet flushing and through 70 mm filter, pre-plated for 2 h to exclude cardiac fibroblast, and finally plated in 60-mm dishes in DMEM containing $10\%$ FBS, $1\%$ P/S and 1x BrdU for two days. The culture media was next changed to $2\%$ FBS, which could be cultured in 37°C incubator with $5\%$ CO2 for at most 3 days prior to use. Hypoxia condition was created by N2 humidified chamber with $5\%$ CO2 and $1\%$ O2 supplied at 37°C for 24 h.
HEK293 cells were cultured in DMEM supplemented with $10\%$ fetal bovine serum at 37°Cand $5\%$ CO2. pShuttle-CMV-SF-NEDD8 (Strep- and FLAG-tagged NEDD8) was generated as described.40 The plasmids were transfected into the cells using X-tremeGene HP DNA transfection reagent (Roche) according to the manufacturer’s protocol. Cells were harvested for analysis 48-72 h after transfection or as indicated.
Some of the cells were treated with vehicle (DMSO or BSA), 1 μM MLN4924 (Active Biochem), 100 nM bortezomib (Enzolife), or 1 nM Echinomycin (Thermo Fisher) where applicable.
## Adeno-associated virus infection
AAV9 virus expressing Cre under the control of a mouse cardiac troponin T promoter (AAV9-cTnT-Cre)42 (Vigene Biosciences) was subcutaneously injected into mice at age P1 at the indicated doses (high dose: 5 x 1011 GC/pup; medium dose: 0.5 × 1011 GC/pup; low dose: 0.25 x 1011 GC/pup). Genotyping of mouse pups were blind to injection performer.
## In situ t-tubule labeling, imaging, and quantification
The hearts from AAV-infected mice were isolated and perfused with FM4-64FX and subjected to in situ confocal microscopy. The t-tubule content, integrity and regularity were next analyzed in ImageJ following previously described STAR Methods.79 Briefly, intact mouse hearts were Langendorff-perfused at room temperature with Tyrode’s solution (NaCl 137, KCl 5.4, HEPES 10, Glucose 10, MgCl2 1, NaH2PO4 0.33, pH adjusted to 7.4 with NaOH, oxygenated with $95\%$ O2 and $5\%$ CO2 during experiments), containing 10 μM FM4-64FX, a lipophilic fluorescence indicator of membrane structure (Thermo Fisher), for 20 min. The membrane structure of epicardial myocytes was analyzed in situ with confocal microscope (STELLARIS 8, Leica Microsystems). T-tubule images were next analyzed with IDL image analysis program (ITT VIS Inc., Colorado). Background noise in confocal images was filtered with a threshold value retrieved from image intensity histogram. T-tubule two-dimensional images were converted to frequency domain using the Fast Fourier Transformation function in IDL, so that it could be determined whether repeating patterns occur (T-tubule regularity) and how strong the repeating patterns are (T-tubule power).
## RNA-seq analysis
Mouse ventricles were minced and treated with RNAlater (Thermo Fisher) according to the manufacturer’s protocol at −80°C. The isolated RNA was subjected to RNA-sequencing analysis performed by Genome Technology Access Center (GTAC) at Washington *University via* Next Generation Sequencing. Genotyping information of the mice was pre-coded and blind during RNA-seq analysis and data analysis.
## Lipid droplets staining assays
Cells were stained with LipidTOX kit or BODIPY staining following the manufacturer’s protocol (Thermo Fisher #H34475 or #D3922, respectively). Briefly, after indicated treatment, NRVCs grown on coverslips were fixed in $4\%$ paraformaldehyde at room temperature for 10 min and subjected to LipidTOX labeling (1: 1000 in PBS, RT for 30 min) or BODIPY labeling (1: 5000 in PBS, RT for 30 min). The cells were counterstained with cardiac Troponin-T (TnT)/Phalloidin (Thermo Fisher) and DAPI (diamidino-2-phenylindole, Thermo Fisher) to label cardiomyocytes and nuclei, respectively. Confocal images were quantified by ImageJ. LipidTOX/BODIPY intensity was normalized with DAPI intensity. Eight views/sample, 3 samples/group were quantified.
## Triglyceride (TG) measurement
Cells were lysed in $1\%$ Triton X-100 in PBS. The concentrations of TG were measured using a TG assay kit (Infinity Triglycerides kit) following the manufacturer’s protocol. The colormetric readings were collected with a microplate reader at OD 570 nm and the data were normalized to total proteins.
## Metabolomics analysis
Cell pellets (~2.4X107 cells/biological replicate) were collected and snap frozen in liquid nitrogen for untargeted metabolomics analysis (Creative Proteomics, NY). A total of 4 control samples and 4 MLN-treated samples were included in this analysis. No sample was excluded since sample submission. Briefly, samples were thawed in 800 μL of $80\%$ methanol, sonicated at 4°C for 30 min, kept at −40°C for 1 h, vortexed for 30 s, and centrifuged at 12000 rpm at 4°C for 15 min. Finally, 200 μL of supernatant and 5 μL of DL-o-chlorophenylalanine (140 μg/mL) were used for LC-MS analysis. Quality control (QC) samples were prepared using the same sample preparation procedure. Samples were separated by Ultimate 3000LC combined with Q Exactive MS (Thermo) and screened with ESI-MS. The LC system is comprised of an ACQUITY UPLC HSS T3 (100 × 2.1 mm × 1.8 μm) with Ultimate 3000LC. The mobile phase was composed of solvent A ($0.05\%$ formic acid water) and solvent B (acetonitrile) with a gradient elution. The flow rate of the mobile phase was 0.3 mL/min. The column temperature was maintained at 40°C, and the sample manager temperature was set at 4°C. Mass spectrometry parameters in ESI- mode are as follows: heater temperature 300°C, sheath gas flow rate, 45 arb; auxiliary gas flow rate, 15arb; sweep gas flow rate, 1 arb; spray voltage, 3.2 kV; capillary temperature,350°C; S-Lens radiofrequency level, $60\%$. Sample information of specific treatments was blind during metabolomics analysis and data collection.
The raw data were acquired and aligned using MetaboAnalyst 5.079 based on the m/z value and the retention time of the ion signals. Principal component analysis, statistical analysis, enrichment analysis, pathway analysis and networking analysis were also performed under the MetaboAnalyst 5.0 platform. A value of $p \leq 0.05$ was considered statistically significant unless specifically defined.
## Embryo isolation
Female mice crossed with their mating partners were checked for plug formation. Mouse embryos at different embryonic ages were dissected and washed in cold PBS. Embryo heads and lower dorsal parts were removed prior to fixation. For immunostaining, the embryos were fixed in cold $4\%$ paraformaldehyde overnight at 4°C, $70\%$ ethanol overnight at 4°C, and subjected to either OCT embedding for cryosection or paraffin processing and embedding for paraffin sections at thickness of 5 μm. Approximately 60 embryos within 8 litters were used in this article. The grouping was determined by the genotyping of yolk sac of individual embryo. Sex was not considered at this stage of embryonic development.
## Echocardiography
Perinatal mice were anesthetized by inhalation of isoflurane ($2.5\%$ for induction and $1.5\%$ for maintenance) via a nose cone. The adequacy of anesthesia was monitored by toe pinch. Cardiac images and loops were recorded using a VEVO 2100 echocardiography system with a 30MHz transducer (Visual Sonics). The LV morphometric and functional parameters were analyzed offline using VEVO 2100 software. Echocardiography of conscious neonates was performed by gently securing the mice on the station with tapes. Mice information including genotyping and treatment was blind to echocardiography performer.
## Histology and immunohistochemistry analysis
For histology analysis, 5-μm OCT-embedded cryosections were subjected to Fast Green/Sirius Red staining. For immunohistochemistry analysis, cryosections were subjected to antigen retrieval in preheated sodium citrate buffer (pH 6.0, 98°C) for 10 min with the PT Link system (Dako). For cryo-sectioned tissues, deparaffinizing and antigen retrieval procedure were replaced by treatment of $1\%$ Triton X-100 in PBS for 10 min at room temperature. After pre-incubation with $10\%$ non-immune goat serum (Thermo Fisher Scientific) to prevent non-specific binding, tissue sections were incubated with primary antibodies at 4°C overnight and subsequently with appropriate Alexa-Fluor conjugated secondary antibodies (Thermo Fisher Scientific) for 1 h at room temperature. Finally, sections were stained with DAPI (Sigma) and mounted in VECTASHIELD antifade mounting medium (Vector Laboratories). Images were captured with Olympus BX41 (Olympus) or Zeiss Upright 780 confocal microscope (Zeiss). Section information was pre-coded and blind to procedure performers including embedding, sectioning, staining and imaging.
## Protein extraction and Western blot analysis
Protein was extracted from ventricular myocardium tissues or cultured cells, concentration determined with BCA reagents (Thermo Fisher Scientific), and SDS-PAGE, immunoblotting, and densitometry analysis were performed as previously described.38 Briefly, frozen tissue or cell was homogenized in lysis buffer (50 mm Tris-HCl pH 6.8 containing $1\%$ SDS, $10\%$ glycerol, and complete protease inhibitor mixture), sonicated and spin down at 14,000 rpm. The supernatant was collected and boiled for 10 min. After protein concentration was determined, the protein lysate was mixed with half the volume of 3x SDS loading buffer with $15\%$ β-mercaptoethanol. The mixed sample was next boiled for 5 min and subjected to SDS-PAGE gel isolation, transferred to PVDF membrane, and blotted with specific primary antibody followed by secondary antibody conjugated with HRP and film development. No unique technique of Western blot was conducted. Primary antibodies used in this study are listed in key resources table and detailed in Table S3.
All uncropped WB figures can be found in Supplemental Materials.
## Immunoprecipitation
For denaturing IP, cells were lysed in TSD buffer (50 mM Tris pH 7.5, $1\%$ SDS, 5 mM DTT) containing a cocktail of phosphatase and protease inhibitors (Sigma). The lysates were then diluted with 10-fold volume of TNN buffer (50 mM Tris pH 7.5, 250 mM NaCl, 5 mM EDTA, $0.5\%$ NP-40), and incubated with primary antibody and protein A Sepharose beads (ThermoFisher Scientific) with rotation at 4°C overnight. For Biotin-IP, cells were lysed by RIPA buffer (NaCl 150mM, Tris-HCl pH 7.4 10mM, EDTA 1mM, Triton X-100 $1\%$, SDS $0.1\%$, Sodium deoxycholate $0.1\%$) and incubated with NeutrAvidin agarose resin (Thermo Scientific, #29201) with rotation at 4°C overnight. Immunoprecipitates were eluted with SDS-PAGE sampling buffer at 95°C for 5 min, and subjected to SDS-PAGE and immunoblotting.
## RNA preparation and real-time PCR
Isolation of total RNA and reverse transcription into single-stranded cDNA was performed as previously described.41 Briefly, Total RNA was isolated from heart tissue or NRVCs using the TRIzol Reagent (Invitrogen) following the manufacturer’s protocol. Gene expression levels were measured in at least triplicate per sample by real-time quantitative PCR (StepOnePlus Real-Time PCR system, Thermo Fisher Scientific) using the SYBR-Green assay with gene-specific primers at a final concentration of 200 nM. *Relative* gene expression was calculated using the 2−ΔΔct method against a rat house-keeping gene acidic ribosomal phosphoprotein P0 (RPLP0) or a mouse house-keeping gene hypoxanthine guanine phosphoribosyl transferase 1 (Hprt) as appropriate. The primers used for qPCR are listed in Key Resources Table and detailed in Table S4.
## siRNA transfection
siRNAs against rat Cul2 (5’ – TTCGAGCGACCAGTAACCTTA-3′) and luciferase (5’ -AACGTACGCGGAATACTTCGA-3′) were used. Briefly, NRVCs were transfected with siRNAs (100 pmol per 2×106 cells) using Lipofectamine RNAimax (Thermo Fisher Scientific) following the manufacturer’s protocol at 24–48 h after plating. Six hours after the transfection, the siRNA-containing medium was replaced with fresh medium containing $2\%$ FBS. In some experiments, a second round of siRNA transfection may be performed 2 days after the first transfection to achieve sustained gene silencing.
## ATP content measurement
Cells were lysed in cold $10\%$ trichloroacetic acid (TCA) and diluted 100-fold in Tris-acetate (pH 7.8). ATP content was measured by ATP Bioluminescent Assay Kit (Sigma, #FL-AA) on a luminescent microplate reader and normalized to total protein levels.
## Dual luciferase assay
The dual luciferase assay was performed following the manufacture protocol of Dual-Luciferase Reporter Assay System (Promega). Specifically, HRE-firefly-luciferase and Renilla luciferase were co-transfected in H9C2 cells and cultured for 72 h. Next, the firefly and Renilla luciferase activities of the same sample were sequentially measured by a luminometer after addition of specific substrates, with the ratio of firefly to Renilla luciferase activity (Fluc/Rluc) as readout of HIF-1α activity.
## Plasmids
HA-HIF1α (#18949), HRE-firefly-luciferase (#26731), Renilla luciferase (#118016), 5HRE-GFP (#46926) plasmids were obtained from Addgene. Received bacteria were amplified and correlating plasmids were extracted with Plasmid Maxi kit (Qiagen) following manufactures instructions. pShuttle-CMV-FLAG-NEDD8 and pShuttle-CMV-FLAG-NEDD8-dGG plasmids were generated as previously described.41 Briefly, NEDD8 or NEDD8-dGG sequence was synthesized and cloned in pShuttle-CMV-FLAG construct. CAGΔX-bioNEDD8-BirAOV5-g2A-puro and CAGΔX -bioNEDD8-BirAOV5-g2A -UBC12 plasmids were a kind gift from James Sutherland at CIC bioGUNE.77
## QUANTIFICATION AND STATISTICAL ANALYSIS
Results are shown as mean ± SEM. Paired data were evaluated using a two-tailed Student’s t-test. For multiple comparisons, one-way analysis of variance (ANOVA) or when appropriate, two-way ANOVA, followed by post hoc test was performed.
A p-value < 0.05 was considered statistically significant. In figures, different p value was labeled as follow: NS, $p \leq 0.05$; *, 0.05 > p ≥ 0.01; **, 0.01 > p ≥ 0.001, ***, 0.001 > p ≥ 0.0001; ****, $p \leq 0.0001.$ Specific statistical analysis method and n number of individual experiment can be found in related figure legend. All centers of bar graph in this article represent mean value of that group. All error bars in this article represent ±SEM unless stated otherwise in figure legends.
## References
1. Guo Y, Pu WT. **Cardiomyocyte maturation: new phase in development**. *Circ. Res* (2020) **126** 1086-1106. DOI: 10.1161/CIRCRESAHA.119.315862
2. Padula SL, Velayutham N, Yutzey KE. **Transcriptional regulation of postnatal cardiomyocyte maturation and regeneration**. *Int. J. Mol. Sci* (2021) **22** 3288. DOI: 10.3390/ijms22063288
3. VanDusen NJ, Lee JY, Gu W, Butler CE, Sethi I, Zheng Y, King JS, Zhou P, Suo S, Guo Y. **Massively parallel in vivo CRISPR screening identifies RNF20/40 as epigenetic regulators of cardiomyocyte maturation**. *Nat. Commun* (2021) **12** 4442. DOI: 10.1038/s41467021-24743-z
4. Guo Y, Cao Y, Jardin BD, Sethi I, Ma Q, Moghadaszadeh B, Troiano EC, Mazumdar N, Trembley MA, Small EM. **Sarcomeres regulate murine cardiomyocyte maturation through MRTF-SRF signaling**. *Proc. Natl. Acad. Sci. USA* (2021) **118** e2008861118. DOI: 10.1073/pnas.2008861118
5. Murphy SA, Miyamoto M, Kervadec A, Kannan S, Tampakakis E, Kambhampati S, Lin BL, Paek S, Andersen P, Lee DI. **PGC1/PPAR drive cardiomyocyte maturation at single cell level via YAP1 and SF3B2**. *Nat. Commun* (2021) **12** 1648. DOI: 10.1038/s41467-02121957-z
6. Menendez-Montes I, Escobar B, Palacios B, Gómez MJ, Izquierdo-Garcia JL, Flores L, Jiménez-Borreguero LJ, Aragones J, Ruiz-Cabello J, Torres M, Martin-Puig S. **Myocardial VHL-HIF signaling controls an embryonic metabolic switch essential for cardiac maturation**. *Dev. Cell* (2016) **39** 724-739. DOI: 10.1016/j.devcel.2016.11.012
7. Karbassi E, Fenix A, Marchiano S, Muraoka N, Nakamura K, Yang X, Murry CE. **Cardiomyocyte maturation: advances in knowledge and implications for regenerative medicine**. *Nat. Rev. Cardiol* (2020) **17** 341-359. DOI: 10.1038/s41569-019-0331-x
8. Lopaschuk GD, Jaswal JS. **Energy metabolic phenotype of the cardiomyocyte during development, differentiation, and postnatal maturation**. *J. Cardiovasc. Pharmacol* (2010) **56** 130-140. DOI: 10.1097/FJC.0b013e3181e74a14
9. Schaper J, Meiser E, Stämmler G. **Ultrastructural morphometric analysis of myocardium from dogs, rats, hamsters, mice, and from human hearts**. *Circ. Res* (1985) **56** 377-391. DOI: 10.1161/01.res.56.3.377
10. Dai DF, Danoviz ME, Wiczer B, Laflamme MA, Tian R. **Mitochondrial maturation in human pluripotent stem cell derived cardiomyocytes**. *Stem Cells Int* (2017) **2017** 5153625. DOI: 10.1155/2017/5153625
11. Sim CB, Ziemann M, Kaspi A, Harikrishnan KN, Ooi J, Khurana I, Chang L, Hudson JE, El-Osta A, Porrello ER. **Dynamic changes in the cardiac methylome during postnatal development**. *FASEB J* (2015) **29** 1329-1343. DOI: 10.1096/fj.14-264093
12. Uosaki H, Cahan P, Lee DI, Wang S, Miyamoto M, Fernandez L, Kass DA, Kwon C. **Transcriptional landscape of cardiomyocyte maturation**. *Cell Rep* (2015) **13** 1705-1716. DOI: 10.1016/j.celrep.2015.10.032
13. Wei Y, Peng S, Wu M, Sachidanandam R, Tu Z, Zhang S, Falce C, Sobie EA, Lebeche D, Zhao Y. **Multifaceted roles of miR-1s in repressing the fetal gene program in the heart**. *Cell Res* (2014) **24** 278-292. DOI: 10.1038/cr.2014.12
14. Lai L, Leone TC, Zechner C, Schaeffer PJ, Kelly SM, Flanagan DP, Medeiros DM, Kovacs A, Kelly DP. **Transcriptional coactivators PGC-1alpha and PGC-lbeta control overlapping programs required for perinatal maturation of the heart**. *Genes Dev* (2008) **22** 1948-1961. DOI: 10.1101/gad.1661708
15. Cossette SM, Gastonguay AJ, Bao X, Lerch-Gaggl A, Zhong L, Harmann LM, Koceja C, Miao RQ, Vakeel P, Chun C. **Sucrose non-fermenting related kinase enzyme is essential for cardiac metabolism**. *Biol. Open* (2014) **4** 48-61. DOI: 10.1242/bio.20149811
16. Hansson A, Hance N, Dufour E, Rantanen A, Hultenby K, Clayton DA, Wibom R, Larsson NG. **A switch in metabolism precedes increased mitochondrial biogenesis in respiratory chain-deficient mouse hearts**. *Proc. Natl. Acad. Sci. USA* (2004) **101** 3136-3141. DOI: 10.1073/pnas.0308710100
17. Breckenridge RA, Piotrowska I, Ng KE, Ragan TJ, West JA, Kotecha S, Towers N, Bennett M, Kienesberger PC, Smolenski RT. **Hypoxic regulation of hand1 controls the fetal-neonatal switch in cardiac metabolism**. *PLoS Biol* (2013) **11** e1001666. DOI: 10.1371/journal.pbio.1001666
18. Schiff M, Ogier de Baulny H, Lombès A. **Neonatal cardiomyopathies and metabolic crises due to oxidative phosphorylation defects**. *Semin. Fetal Neonatal Med* (2011) **16** 216-221. DOI: 10.1016/j.siny.2011.04.002
19. Sakamoto T, Matsuura TR, Wan S, Ryba DM, Kim JU, Won KJ, Lai L, Petucci C, Petrenko N, Musunuru K. **A critical role for estrogen-related receptor signaling in cardiac maturation**. *Circ. Res* (2020) **126** 1685-1702. DOI: 10.1161/CIRCRESAHA.119.316100
20. Neary MT, Ng KE, Ludtmann MHR, Hall AR, Piotrowska I, Ong SB, Hausenloy DJ, Mohun TJ, Abramov AY, Breckenridge RA. **Hypoxia signaling controls postnatal changes in cardiac mitochondrial morphology and function**. *J. Mol. Cell. Cardiol* (2014) **74** 340-352. DOI: 10.1016/j.yjmcc.2014.06.013
21. Guimarães-Camboa N, Stowe J, Aneas I, Sakabe N, Cattaneo P, Henderson L, Kilberg MS, Johnson RS, Chen J, McCulloch AD. **HIF1alpha represses cell stress pathways to Allow proliferation of hypoxic fetal cardiomyocytes**. *Dev. Cell* (2015) **33** 507-521. DOI: 10.1016/j.devcel.2015.04.021
22. Nau PN, Van Natta T, Ralphe JC, Teneyck CJ, Bedell KA, Caldarone CA, Segar JL, Scholz TD. **Metabolic adaptation of the fetal and postnatal ovine heart: regulatory role of hypoxia-inducible factors and nuclear respiratory factor-1**. *Pediatr. Res* (2002) **52** 269-278. DOI: 10.1203/00006450-200208000-00021
23. Backs J, Olson EN. **Control of cardiac growth by histone acetylation/deacetylation**. *Circ. Res* (2006) **93** 15-24. DOI: 10.1161/01.RES.0000197782.21444.8f
24. Smith ZD, Meissner A. **DNA methylation: roles in mammalian development**. *Nat. Rev. Genet* (2013) **14** 204-220. DOI: 10.1038/nrg3354
25. Collins HE, Chatham JC. **Regulation of cardiac O-GlcNAcylation: more than just nutrient availability**. *Biochim. Biophys. Acta, Mol. Basis Dis* (2020) **1866** 165712. DOI: 10.1016/j.bbadis.2020.165712
26. Li J, Johnson JA, Su H. **Ubiquitin and ubiquitin-like proteins in cardiac disease and protection**. *Curr. Drug Targets* (2018) **19** 989-1002. PMID: 26648080
27. Kamitani T, Kito K, Nguyen HP, Yeh ET. **Characterization of NEDD8, a developmentally down-regulated ubiquitin-like protein**. *J. Biol. Chem* (1997) **272** 28557-28562. PMID: 9353319
28. Kandala S, Kim IM, Su H. **Neddylation and deneddylation in cardiac biology**. *Am. J. Cardiovasc. Dis* (2014) **4** 140-158. PMID: 25628956
29. Enchev RI, Schulman BA, Peter M. **Protein neddylation: beyond cullin-RING ligases**. *Nat. Rev. Mol. Cell Biol* (2015) **16** 30-44. DOI: 10.1038/nrm3919
30. Petroski MD, Deshaies RJ. **Function and regulation of cullin-RING ubiquitin ligases**. *Nat. Rev. Mol. Cell Biol* (2005) **6** 9-20. DOI: 10.1038/nrm1547
31. Soucy TA, Dick LR, Smith PG, Milhollen MA, Brownell JE. **The NEDD8 conjugation pathway and its Relevance in cancer biology and Therapy**. *Genes Cancer* (2010) **1** 708-716. DOI: 10.1177/1947601910382898
32. Park HS, Ju UI, Park JW, Song JY, Shin DH, Lee KH, Jeong LS, Yu J, Lee HW, Jeong LS, Yu J, Lee HW, Cho JY. **PPARgamma neddylation essential for adipogenesis is a potential target for treating obesity**. *Cell Death Differ* (2016) **23** 1296-1311. DOI: 10.1038/cdd.2016.6
33. Zhang X, Zhang YL, Qiu G, Pian L, Guo L, Cao H, Liu J, Zhao Y, Li X, Xu Z. **Hepatic neddylation targets and stabilizes electron transfer flavoproteins to facilitate fatty acid beta-oxidation**. *Proc. Natl. Acad. Sci. USA* (2020) **117** 2473-2483. DOI: 10.1073/pnas.1910765117
34. Chen W, Luo B, Gao N, Li H, Wang H, Li L, Cui W, Zhang L, Sun D, Liu F. **Neddylation stabilizes Nav1.1 to maintain interneuron excitability and prevent seizures in murine epilepsy models**. *J. Clin. Invest* (2021) **131** e136956. DOI: 10.1172/JCI136956
35. Zhang L, Jing H, Li H, Chen W, Luo B, Zhang H, Dong Z, Li L, Su H, Xiong WC, Mei L. **Neddylation is critical to cortical development by regulating Wnt/beta-catenin signaling**. *Proc. Natl. Acad. Sci. USA* (2020) **117** 26448-26459. DOI: 10.1073/pnas.2005395117
36. Su H, Li J, Zhang H, Ma W, Wei N, Liu J, Wang X. **COP9 signalosome controls the degradation of cytosolic misfolded proteins and protects against cardiac proteotoxicity**. *Circ. Res* (2015) **117** 956-966. DOI: 10.1161/CIRCRESAHA.115.306783
37. Su H, Li J, Osinska H, Li F, Robbins J, Liu J, Wei N, Wang X. **The COP9 signalosome is required for Autophagy, proteasome-mediated Proteolysis, and cardiomyocyte survival in adult mice**. *Circ. Heart Fail* (2013) **6** 1049-1057. DOI: 10.1161/CIRCHEARTFAILURE.113.000338
38. Su H, Li J, Menon S, Liu J, Kumarapeli AR, Wei N, Wang X. **Perturbation of cullin deneddylation via conditional Csn8 ablation impairs the ubiquitin-proteasome system and causes cardiomyocyte necrosis and dilated cardiomyopathy in mice**. *Circ. Res* (2011) **103** 40-50. DOI: 10.1161/CIRCRESAHA.110.230607
39. Su H, Li F, Ranek MJ, Wei N, Wang X. **COP9 signalosome regulates autophagosome maturation**. *Circulation* (2011) **124** 2117-2128. DOI: 10.1161/CIRCULATIONAHA.111.048934
40. Li J, Ma W, Li H, Hou N, Wang X, Kim IM, Li F, Su H. **NEDD8 ultimate Buster 1 long (NUB1L) protein suppresses atypical neddylation and promotes the proteasomal degradation of misfolded proteins**. *J. Biol. Chem* (2015) **290** 23850-23862. DOI: 10.1074/jbc.M115.664375
41. Zou J, Ma W, Li J, Littlejohn R, Zhou H, Kim IM, Fulton DJR, Chen W, Weintraub NL, Zhou J, Su H. **Neddylation mediates ventricular chamber maturation through repression of Hippo signaling**. *Proc. Natl. Acad. Sci. USA* (2018) **115** E4101-E4110. DOI: 10.1073/pnas.1719309115
42. Lin Z, von Gise A, Zhou P, Gu F, Ma Q, Jiang J, Yau AL, Buck JN, Gouin KA, van Gorp PRR. **Cardiac-specific YAP activation improves cardiac function and survival in an experimental murine MI model**. *Circ. Res* (2014) **115** 354-363. DOI: 10.1161/CIRCRESAHA.115.303632
43. Muzumdar MD, Tasic B, Miyamichi K, Li L, Luo L. **A global double-fluorescent Cre reporter mouse**. *Genesis* (2007) **45** 593-605. DOI: 10.1002/dvg.20335
44. Guo Y, VanDusen NJ, Zhang L, Gu W, Sethi I, Guatimosim S, Ma Q, Jardin BD, Ai Y, Zhang D. **Analysis of cardiac myocyte maturation using CASAAV, a platform for rapid dissection of cardiac myocyte gene function in vivo**. *Circ. Res* (2017) **120** 1874-1888. DOI: 10.1161/CIRCRESAHA.116.310283
45. Uosaki H, Taguchi YH. **Comparative gene expression analysis of mouse and human cardiac maturation**. *Dev. Reprod. Biol* (2016) **14** 207-215. DOI: 10.1016/j.gpb.2016.04.004
46. Ellen Kreipke R, Wang Y, Miklas JW, Mathieu J, Ruohola-Baker H. **Metabolic remodeling in early development and cardiomyocyte maturation**. *Semin. Cell Dev. Biol* (2016) **52** 84-92. DOI: 10.1016/j.semcdb.2016.02.004
47. Cho Y, Tachibana S, Lam K, Arita Y, Khosrowjerdi S, Zhang O, Liang A, Li R, Andreyev A, Murphy AN, Ross RS. **Perm1 promotes cardiomyocyte mitochondrial biogenesis and protects against hypoxia/reoxygenation-induced damage in mice**. *J. Biol. Chem* (2021) **297** 100825. DOI: 10.1016/j.jbc.2021.100825
48. Soucy TA, Smith PG, Milhollen MA, Berger AJ, Gavin JM, Adhikari S, Brownell JE, Burke KE, Cardin DP, Critchley S. **An inhibitor of NEDD8-activating enzyme as a new approach to treat cancer**. *Nature* (2009) **458** 732-736. DOI: 10.1038/nature07884
49. Zhang H, Gao P, Fukuda R, Kumar G, Krishnamachary B, Zeller KI, Dang CV, Semenza GL. **HIF-1 inhibits mitochondrial biogenesis and cellular respiration in VHL-deficient renal cell carcinoma by repression of C-MYC activity**. *Cancer Cell* (2007) **11** 407-420. DOI: 10.1016/j.ccr.2007.04.001
50. Dorn GW. **Mitochondrial dynamics in heart disease**. *Biochim. Biophys. Acta* (2013) **1833** 233-241. DOI: 10.1016/j.bbamcr.2012.03.008
51. Huang Y, Hickey RP, Yeh JL, Liu D, Dadak A, Young LH, Johnson RS, Giordano FJ. **Cardiac myocyte-specific HIF-1alpha deletion alters vascularization, energy availability, calcium flux, and contractility in the normoxic heart**. *FASEB J* (2004) **18** 1138-1140. DOI: 10.1096/fj.04-1510fje
52. Bishop T, Ratcliffe PJ. **HIF hydroxylase pathways in cardiovascular physiology and medicine**. *Circ. Res* (2015) **117** 65-79. DOI: 10.1161/CIRCRESAHA.117.305109
53. Baek K, Scott DC, Schulman BA. **NEDD8 and ubiquitin ligation by cullin-RING E3 ligases**. *Curr. Opin. Struct. Biol* (2021) **67** 101-109. DOI: 10.1016/j.sbi.2020.10.007
54. Kong D, Park EJ, Stephen AG, Calvani M, Cardellina JH, Monks A, Fisher RJ, Shoemaker RH, Melillo G. **Echinomycin, a small-molecule inhibitor of hypoxia-inducible factor-1 DNA-binding activity**. *Cancer Res* (2005) **65** 9047-9055. DOI: 10.1158/0008-5472.CAN-05-1235
55. Zou J, Ma W, Littlejohn R, Li J, Stansfield BK, Kim IM, Liu J, Zhou J, Weintraub NL, Su H. **Transient inhibition of neddylation at neonatal stage evokes reversible cardiomyopathy and predisposes the heart to isoproterenol-induced heart failure**. *Am. J. Physiol. Heart Circ. Physiol* (2019) **316** H1406-H1416. DOI: 10.1152/ajpheart.00806.2018
56. Fukuda R, Gunawan F, Beisaw A, Jimenez-Amilburu V, Maischein HM, Kostin S, Kawakami K, Stainier DYR. **Proteolysis regulates cardiomyocyte maturation and tissue integration**. *Nat. Commun* (2017) **8** 14495. DOI: 10.1038/ncomms14495
57. Zhou Q, Li H, Li Y, Tan M, Fan S, Cao C, Meng F, Zhu L, Zhao L, Guan MX. **Inhibiting neddylation modification alters mitochondrial morphology and reprograms energy metabolism in cancer cells**. *JCI Insight* (2019) **4** e121582. DOI: 10.1172/jci.insight.121582
58. Lei L, Mason S, Liu D, Huang Y, Marks C, Hickey R, Jovin IS, Pypaert M, Johnson RS, Giordano FJ. **Hypoxia-inducible factor-dependent degeneration, failure, and malignant transformation of the heart in the absence of the von Hippel-Lindau protein**. *Mol. Cell Biol* (2008) **28** 3790-3803. DOI: 10.1128/MCB.01580-07
59. Hölscher M, Schäfer K, Krull S, Farhat K, Hesse A, Silter M, Lin Y, Pichler BJ, Thistlethwaite P, El-Armouche A. **Unfavourable consequences of chronic cardiac HIF-1alpha stabilization**. *Cardiovasc. Res* (2012) **94** 77-86. DOI: 10.1093/cvr/cvs014
60. Moslehi J, Minamishima YA, Shi J, Neuberg D, Charytan DM, Padera RF, Signoretti S, Liao R, Kaelin WG. **Loss of hypoxia-inducible factor prolyl hydroxylase activity in cardiomyocytes phenocopies ischemic cardiomyopathy**. *Circulation* (2010) **122** 1004-1016. DOI: 10.1161/CIRCULATIONAHA.109.922427
61. Bekeredjian R, Walton CB, MacCannell KA, Ecker J, Kruse F, Outten JT, Sutcliffe D, Gerard RD, Bruick RK, Shohet RV. **Conditional HIF-1alpha expression produces a reversible cardiomyopathy**. *PLoS One* (2010) **5** e11693. DOI: 10.1371/journal.pone.0011693
62. Romanowicz J, Guerrelli D, Dhari Z, Mulvany C, Reilly M, Swift L, Vasandani N, Ramadan M, Leatherbury L, Ishibashi N, Posnack NG. **Chronic perinatal hypoxia delays cardiac maturation in a mouse model for cyanotic congenital heart disease**. *Am. J. Physiol. Heart Circ. Physiol* (2021) **320** H1873-H1886. DOI: 10.1152/ajpheart.00870.2020
63. Ryu JH, Li SH, Park HS, Park JW, Lee B, Chun YS. **Hypoxia-inducible factor alpha subunit stabilization by NEDD8 conjugation is reactive oxygen species-dependent**. *J. Biol. Chem* (2011) **286** 6963-6970. DOI: 10.1074/jbc.M110.188706
64. Curtis VF, Ehrentraut SF, Campbell EL, Glover LE, Bayless A, Kelly CJ, Kominsky DJ, Colgan SP. **Stabilization of HIF through inhibition of Cullin-2 neddylation is protective in mucosal inflammatory responses**. *Faseb j* (2015) **29** 208-215. DOI: 10.1096/fj.14-259663
65. Heir P, Sufan RI, Greer SN, Poon BP, Lee JE, Ohh M. **DCNL1 functions as a substrate sensor and activator of cullin 2-RING ligase**. *Mol. Cell Biol* (2013) **33** 1621-1631. DOI: 10.1128/mcb.01342-12
66. Ehrentraut SF, Kominsky DJ, Glover LE, Campbell EL, Kelly CJ, Bowers BE, Bayless AJ, Colgan SP. **Central role for endothelial human deneddylase-1/SENP8 in fine-tuning the vascular inflammatory response**. *J. Immunol* (2013) **190** 392-400. DOI: 10.4049/jimmunol.1202041
67. Leidecker O, Matic I, Mahata B, Pion E, Xirodimas DP. **The ubiquitin E1 enzyme Ube1 mediates NEDD8 activation under diverse stress conditions**. *Cell Cycle* (2012) **11** 1142-1150. DOI: 10.4161/cc.11.6.19559
68. Hjerpe R, Thomas Y, Kurz T. **NEDD8 overexpression results in neddylation of ubiquitin substrates by the ubiquitin pathway**. *J. Mol. Biol* (2012) **421** 27-29. DOI: 10.1016/j.jmb.2012.05.013
69. Lobato-Gil S, Heidelberger JB, Maghames C, Bailly A, Brunello L, Rodriguez MS, Beli P, Xirodimas DP. **Proteome-wide identification of NEDD8 modification sites reveals distinct proteomes for canonical and atypical NEDDylation**. *Cell Rep* (2021) **34** 108635. DOI: 10.1016/j.celrep.2020.108635
70. Vogl AM, Phu L, Becerra R, Giusti SA, Verschueren E, Hinkle TB, Bordenave MD, Adrian M, Heidersbach A, Yankilevich P. **Global site-specific neddylation profiling reveals that NEDDylated cofilin regulates actin dynamics**. *Nat. Struct. Mol. Biol* (2020) **27** 210-220. DOI: 10.1038/s41594-019-0370-3
71. Liu J, Wen T, Dong K, He X, Zhou H, Shen J, Fu Z, Hu G, Ma W, Li J. **TEAD1 protects against necroptosis in postmitotic cardiomyocytes through regulation of nuclear DNA-encoded mitochondrial genes**. *Cell Death Differ* (2021) **28** 2045-2059. DOI: 10.1038/s41418-020-00732-5
72. Xin M, Kim Y, Sutherland LB, Murakami M, Qi X, McAnally J, Porrello ER, Mahmoud AI, Tan W, Shelton JM. **Hippo pathway effector Yap promotes cardiac regeneration**. *Proc. Natl. Acad. Sci. USA* (2013) **110** 13839-13844. DOI: 10.1073/pnas.1313192110
73. Kondo K, Klco J, Nakamura E, Lechpammer M, Kaelin WG. **Inhibition of HIF is necessary for tumor suppression by the von Hippel-Lindau protein**. *Cancer Cell* (2002) **1** 237-246. DOI: 10.1016/s1535-6108(02)00043-0
74. Emerling BM, Weinberg F, Liu JL, Mak TW, Chandel NS. **PTEN regulates p300-dependent hypoxia-inducible factor 1 transcriptional activity through Forkhead transcription factor 3a (FOXO3a)**. *Proc. Natl. Acad. Sci. USA* (2008) **105** 2622-2627. DOI: 10.1073/pnas.0706790105
75. Giacomelli AO, Yang X, Lintner RE, McFarland JM, Duby M, Kim J, Howard TP, Takeda DY, Ly SH, Kim E. **Mutational processes shape the landscape of TP53 mutations in human cancer**. *Nat. Genet* (2018) **50** 1381-1387. DOI: 10.1038/s41588-018-0204-y
76. Vordermark D, Shibata T, Brown JM. **Green fluorescent protein is a suitable reporter of tumor hypoxia despite an oxygen requirement for chromophore formation**. *Neoplasia* (2001) **3** 527-534. DOI: 10.1038/sj.neo.7900192
77. Pirone L, Xolalpa W, Sigursson JO, Ramirez J, Pérez C, González M, de Sabando AR, Elortza F, Rodriguez MS, Mayor U. **A comprehensive platform for the analysis of ubiquitin-like protein modifications using in vivo biotinylation**. *Sci. Rep* (2017) **7** 40756. DOI: 10.1038/srep40756
78. Pang Z, Chong J, Zhou G, de Lima Morais DA, Chang L, Barrette M, Gauthier C, Jacques PÉ, Li S, Xia J. **MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights**. *Nucleic Acids Res* (2021) **49** W388-W396. DOI: 10.1093/nar/gkab382
79. Wei S, Guo A, Chen B, Kutschke W, Xie YP, Zimmerman K, Weiss RM, Anderson ME, Cheng H, Song LS. **T-tubule remodeling during transition from hypertrophy to heart failure**. *Circ. Res* (2010) **107** 520-531. DOI: 10.1161/CIRCRESAHA.109.212324
|
---
title: Harnessing protein sensing ability of electrochemical biosensors via a controlled
peptide receptor–electrode interface
authors:
- Ji Hong Kim
- Jae Hwan Shin
- Bumjun Park
- Chae Hwan Cho
- Yun Suk Huh
- Chang-Hyung Choi
- Jong Pil Park
journal: Journal of Nanobiotechnology
year: 2023
pmcid: PMC10029155
doi: 10.1186/s12951-023-01843-0
license: CC BY 4.0
---
# Harnessing protein sensing ability of electrochemical biosensors via a controlled peptide receptor–electrode interface
## Abstract
### Background
Cathepsin B, a cysteine protease, is considered a potential biomarker for early diagnosis of cancer and inflammatory bowel diseases. Therefore, more feasible and effective diagnostic method may be beneficial for monitoring of cancer or related diseases.
### Results
A phage-display library was biopanned against biotinylated cathepsin B to identify a high-affinity peptide with the sequence WDMWPSMDWKAE. The identified peptide-displaying phage clones and phage-free synthetic peptides were characterized using enzyme-linked immunosorbent assays (ELISAs) and electrochemical analyses (impedance spectroscopy, cyclic voltammetry, and square wave voltammetry). Feasibilities of phage-on-a-sensor, peptide-on-a-sensor, and peptide-on-a-AuNPs/MXene sensor were evaluated. The limit of detection and binding affinity values of the peptide-on-a-AuNPs/MXene sensor interface were two to four times lower than those of the two other sensors, indicating that the peptide-on-a-AuNPs/MXene sensor is more specific for cathepsin B (good recovery (86–$102\%$) and %RSD (< $11\%$) with clinical samples, and can distinguish different stages of Crohn’s disease. Furthermore, the concentration of cathepsin B measured by our sensor showed a good correlation with those estimated by the commercially available ELISA kit.
### Conclusion
In summary, screening and rational design of high-affinity peptides specific to cathepsin B for developing peptide-based electrochemical biosensors is reported for the first time. This study could promote the development of alternative antibody-free detection methods for clinical assays to test inflammatory bowel disease and other diseases.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12951-023-01843-0.
## Introduction
Cathepsins have attracted immense clinical attention as potential biomarkers for the early diagnosis of cancer [1–3]. It is the multifunctional enzymes that involve the whole pathogenesis and oncogenic process including differentiation and transition from cancer growth to metastasis. To date, more than 12 cathepsin families (from A to Z) have been found in organisms and rigorously characterized [4]. Among various cathepsin derivatives, cathepsin B (CTSB) is considered a potential biomarker for several types of cancers, such as prostate, pancreatic, colon, and breast cancers, as well as for osteoporosis, rheumatoid arthritis, and infectious diseases, and there is good correlation between tumor progression and the CTSB expression [5–8]. It is known that cathepsin B is a cysteine protease and can be involved in extracellular matrix (ECM) component degradation, cell–cell communication disruption and protease inhibitor expression [9, 10]. According to recent studies, CTSB levels in human fluids (serum or plasma) can be measured and remain within the diagnostic window after the onset of the inflammatory bowel diseases (IBD) including Crohn’s disease and Ulcerative colitis, enabling a non-invasive early detection of these diseases [11, 12]. For this and other purpose, chemical inhibitor and antibodies have been studied to suppress proteolytic activity of proteases in order to impose metastatic infiltration mediated by protease. Some of compound as inhibitor of cathepsin B have been identified in various resources, for example, microorganisms and natural products [13]. It was observed that the suppression of cathepsin B has high therapeutic significance, however, no direct targeting of cathepsin B with alternative binder and/or inhibitor over conventional reagents including antibodies or chemical inhibitor was reported.
The common features of the current antibody-based immunoassay and polymerase chain reaction (PCR) are considered as gold standard detection regime for detection of cancer biomarkers. However, these techniques have some bottlenecks in terms of operating cost, multiple sample preparation and labor-intensive and more effective biosensors have been developed [2, 3, 11, 12].
Phage display is a promising technology for the identification of potential peptide candidates with high affinity and specificity toward various target molecules (organic and inorganic materials) as well as for the development of therapeutic drugs and biosensing applications [14–16]. In phage display, M13 bacteriophages are genetically modified to expose small (generally, 12-mer, or 7-mer) unique peptides that are fused with the minor coat protein pIII (approximately five copies per phage) or major coat protein pVIII (approximately 2700 copies per phage) on their surface randomly, during biopanning [17]. One of the most interesting features of this technique is its ability to obtain large diversified peptide libraries on the phage surfaces in a short time [15, 17–20]. Interestingly, affinity peptides obtained by phage display are relatively smaller than the full-sized antibodies and thus can be cost-effectively mass produced. Further, peptides can be genetically manipulated to form a series of peptide derivatives are easily and reliably conjugated with other motifs, and can be immobilized on various surfaces for biosensor applications.
Electrochemical-based biosensors have attracted among the various analytical techniques because of their high sensitivity, selectivity, portability and diversity of target molecules, for example, drug, virus, and protein [17, 21, 22]. Previously, several advanced electrochemical sensors have been developed by combining with enzymes and nanomaterials, such as layered double hydroxides (LDHs) which is a two-dimensional anionic lamellar clay material with charge compensating anions and solvation molecules for detection of small molecules [23–25] and carbon-based nanocomposites (specifically, MXenes which are two-dimensional transitional metal compounds) [26–30]. In a recent study, Au nanoparticles (AuNPs) and methylene blue were incorporated as signal amplifiers on MXene nanocomposites to fabricate electrochemical biosensors for prostate-specific antigen (PSA) detection [31]. These sensors detected PSA with a limit of detection (LOD) of 0.83 pg/mL and a linear dynamic range of < 10 pico level. MXene-based nanocomposites have recently emerged as viable candidates for electrochemical sensors owing to their favorable physical and chemical properties including high electrical conductivities, biocompatibility, large surface areas, and a facile sensor layer and electrode functionalization [32]. As these common features of detection techniques are further developing combination with functional carbon nanomaterial and unique affinity peptide may be opened the widow as replacement over antibody-based immunoassay in biosensor as well as providing the basis for monitoring progression of cancer.
This paper is the first ever report on the screening and rational design of high-affinity peptides for fabrication as well as the use of peptide-based electrochemical sensors for the accurate detection of cathepsin B. In this study, the M13 phage peptide library was biopanned to screen for high-affinity peptides for cathepsin B. The relative binding affinities of the phage clones and synthetic peptides were investigated by enzyme-linked immunosorbent assays (ELISAs) and electrochemical analyses including electrochemical impedance spectroscopy (EIS), cyclic voltammetry (CV) and square wave voltammetry (SWV). In addition, the sensor electrode was modified with a combination of AuNPs and a MXene material to improve the sensor performance (Fig. 1). *This* general approach could be extended to develop new type of biosensor for the various target molecules. Fig. 1Schematic work flow of the affinity-based electrochemcial sensor used for the detection of cathepsin B. a 1. Phage-display selection, 2. Peptide-displayed whole phage immobilization on bare Au electrode using the MUA-EDC/NHS coupling, 3. Biotin labelled peptide synthesis and synthetic peptide immobilization on bare Au electrode using MUA-EDC/NHS coupling and streptavidin, b 1. AuNP-embedded MXene nanocomposite fabricated on Au electrode for high electrical conductivity, 2. Affinity peptide immobilization on AuNPs–MXene deposited electrode using MUA-EDC/NHS coupling and streptavidin, c Performances of developed sensors for target detection and validation in plama/serum from Crohn’s patients
## Chemicals
Cathepsin B proteins, purified from human liver (~ 27.5 kDa, > $95\%$ purity), were purchased from ENZO (NY, USA). The compounds 2,2ʹ-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid (ABTS), 11-mercaptoundecanoic acid (MUA), N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride (EDC), Tween 20, N-hydroxysuccinimide (NHS), potassium ferricyanide (III), potassium hexacyanoferrate (II) trihydrate, and nafion as well as human plasma were purchased from Sigma-Aldrich (St. Louis, MO, USA). BCA protein assay kits (Pierce™ biotin quantification kit) and streptavidin-coated microwell plates were purchased from Thermo Fisher Scientific (Rockford, IL, USA). The horseradish peroxidase (HRP)-conjugated anti-M13 antibody was purchased from sino biological (Beijing, China). The Ph. D.TM-12 phage display peptide library kit was purchased from New England Biolabs (Ipswich, MA, USA). Lithium fluoride (LiF, − 300 mesh), hydrochloric acid (HCl, ≥ $37\%$), gold (III) chloride trihydrate (HAuCl4·3H2O ≥ $99.9\%$), and trisodium citrate dihydrate (Na3Ctr·2H2O) were purchased from Sigma-Aldrich Chemicals (MO, USA). Titanium aluminum carbide (Ti3AlC2 ≥ $95\%$) in MAX phase was obtained from Y-Cabon Ltd. (Ukraine). Unless otherwise stated, all the chemicals were of analytical grade.
## Preparation of patient samples (plasma and serum)
Blood samples extracted from eight Crohn’s patients were provided by the Biobank of Gyeongsang National University Hospital, a member of the Korea Biobank Network and tested to assess the feasibility of the developed sensor. This study was approved by the ethics review committee of the Institution Review Board, Chung-Ang University (Approval no: 1041078-202107-BR-213-01C). For the analysis, the plasma and serum in the blood samples were separated by centrifugation (3000×g, 20 min) and then stored at − 80 °C until further use. The fractionated plasma and serum were used directly for the electrochemical measurements without further purification. Detailed clinical characteristics of the individuals are shown in Additional file 1: Table S1.
## Biotin labeling of cathepsin B proteins
Biotinylation of the cathepsin B proteins was performed following the manufacturer’s instructions. Cathepsin B (25 µg) was reacted with 10 mM biotin (50-fold molar excess) and then incubated at 4 °C for 24 h. To remove residual reagents, the mixtures were purified using a Zeba desalting spin column (Thermo Fisher Scientific). To estimate the biotin levels quantitatively, biotinylated cathepsin B (100 μL, 250 nM) was coated on streptavidin-coated microplates at 25 °C for 1 h, and streptavidin-conjugated HRP was used to determine the biotinylated protein content. Absorbance was measured at 405 nm using the Multiskan FC microplate photometer (Thermo Fisher Scientific, MA). Further, the Pierce™ biotin quantitation kit was used to determine the biotinylation levels in the labeled cathepsin B.
## Phage display
Biotinylated cathepsin B (99 μL, 500 nM) was pre-reacted with the Ph. D.-12 random phage library (1 μL, 1.0 × 1013 PFU/mL) at 100 rpm for 1 h. Subsequently, 100 µL of the complex mixture was added onto a pre-washed streptavidin-coated microplate, and allowed to react at a shaking speed of 100 rpm for 10 min to facilitate specific binding between avidin and biotin. Then, 0.1 mM biotin was added as a blocking agent and allowed to react for 5 min. After removing the unbound phages and residual biotin, the plate was washed 10 times with $0.1\%$ PBST (0.1 M phosphate-buffered saline (PBS) with $0.1\%$ Tween 20), and the bound phages were eluted using 100 µL of 0.2 M glycine–HCl (pH 2.2) with 1 mg/mL of bovine serum albumin (BSA) solution. Finally, the eluent was neutralized with 15 µL of Tris–HCl (pH 9.1) to prevent deactivation or destruction of the desired phages. Throughout this process, the desired phages were amplified using *Escherichia coli* ER2738 making sufficient copies for the next rounds. The amplified phages were then harvested by polyethylene glycol (PEG)/NaCl precipitation ($20\%$ (v/v) PEG-8000 with 2.5 M NaCl). After every biopanning round, the phages were tittered, and the displayed peptide sequences were analyzed using the − 96 gIII sequencing primer (5ʹ-HOCCC TCA TAG TTA GCG TAA CG-3ʹ).
## ELISA measurements
A 96 well streptavidin-coated microplate was washed three times with PBS, and the biotinylated cathepsin B was immobilized on it with mild shaking at 25 °C for 1 h. The unbound proteins were removed and blocked with a blocking buffer ($5\%$ BSA in NaHCO3) at 25 °C for 1 h. The wells were washed six times with $0.1\%$ PBST, and 100 µL of the amplified phages (1011 PFU/mL), i.e., the capture receptors, were added and incubated for 1 h at 25 °C. After washing six times with PBST again, the HRP-conjugated anti-M13 monoclonal antibodies (1:5000 dilution with blocking buffer), i.e., the detecting antibodies, were added and incubated at 25 °C for 1 h. The residual solution was discarded, and the microplate was washed again with the same buffer. Freshly prepared HRP substrate (ABTS) was added, and the absorbance was measured at 405 nm using a Multiskan FC microplate photometer (Thermo Scientific, Waltham, MA, USA).
## Electrochemical measurements
A conventional three-electrode system including an Au working electrode (diameter: 5 mm, surface area: 19.6 mm2), a Pt counter electrode, and an Ag/AgCl reference electrode, was used for the electrochemical measurements. The EIS, CV, and SWV measurements were conducted in the presence of 2.5 mM ferro/ferricyanide in 1 M KNO3 using a CHI 660E instrument (CH Instruments, Austin, TX, USA) at 25 °C. The EIS measurements were carried out by applying a 10 mV signal with a frequency ranging from 0.01 Hz to 100 kHz. The CV tests were performed between − 0.6 and 0.6 V (scan range) at a scan rate of 100 mVs−1. The SWV tests were carried out in the potential range of − 0.4 to 0.6 V with an amplitude of 5 mV, frequency of 10 Hz, and a scan interval of 4 mV. The relative current change (ΔI%) was obtained by considering that the peak current decreased after the immobilization of the peptides and/or phages and cathepsin B protein interaction using the following equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta I\% = \left({I_0 - I} \right)/I_0 \times 100$$\end{document}ΔI%=I0-I/I0×100where I0 refers to phage-on-a-sensor, peptide-on-a-sensor, or peptide-on-a-AuNPs/MXene sensor and/or peptide oxidation current before cathepsin B introduction, and I refers to Au-protein oxidation current after cathepsin B introduction.
## Statistical analysis
Two-way ANOVA was used to determine the significance of the peptide binding affinities (GraphPad Software Inc, La Jolla, CA). Additionally, the Student’s t-test was used to compare the ΔI% values.
## Identification and characterization of cathepsin B-specific high-affinity peptides
The M13 polyvalent phage display library was biopanned for the selection of dodecamer high-affinity peptides. The phage library used in this study is commercially available and contains random 12-mer peptide sequences that are fused to pIII genes. The labeled biotin level of biotinylated cathepsin B protein was confirmed using HRP-conjugated streptavidin and ABTS at 405 nm (Additional file 1: Fig. S1a). Further, the biotin quantity was analyzed using a biotin quantitation kit. The biotinylated cathepsin B was added to a mixture of 4ʹ-hydroxyazobenzene-2-carboxylic acid, biotinylated HRP, and BSA (biotinylated HRP and BSA were used as the controls) and the biotin ratio of each protein was calculated (Additional file 1: Fig. S1b and c). Four rounds of biopanning were carried out for cathepsin B, and the individual phage clones obtained from each biopanning round were randomly sequenced at the pIII region encoding the phage library peptides; the biopanning yield is shown in Additional file 1: Table S2. Selection was based on the most commonly appearing peptides in each round, whereby three phage clones were chosen for the cathepsin B binding experiments. In two biopanning rounds, only one of the chosen sequences (CTSB 2–3) contained a DG motif at the first and second positions, while two sequences (CTSB 4-1 and CTSB 4-9) in four rounds of biopanning contained PS motifs at the fifth and sixth, and sixth and seventh positions, respectively (Additional file 1: Table S3).
The binding affinities of the selected phage clones for the immobilized cathepsin B were tested by ELISA. For these experiments, 250 nM of cathepsin B or the same amount of BSA as a negative control was immobilized on a well, followed by 1011 PFU/mL of phage addition. The relative binding affinity of each phage was detected using HRP-conjugated M13 antibodies. Two phage clones (CTSB 4-1 and CTSB 4-9) exhibited strong cathepsin B binding, whereas the binding affinity of CTSB 2-3 was much lower (Fig. 2a). Binding to cathepsin B could be related to the PS motif because both the phages (CTSB 4-1, and CTSB 4-9) contained a PS motif in their sequences and could bound to the same epitope on cathepsin B. Interestingly, the CTSB 4-1 phage clones exhibited a dose-dependent binding affinity with increasing cathepsin B concentration from 3.906 to 250 nM, indicating a dose–response binding to cathepsin B (Fig. 2b). The CTSB 4-1 clone was the only phage with a high binding affinity to cathepsin B, motivating further characterization of its binding interaction to confirm its specific binding to cathepsin B with increasing cathepsin B concentration (Fig. 2c).Fig. 2Characterization of the cathepsin B-binding phage particles using ELISA at 405 nm and SWV in 1 M KNO3 with 2.5 mM [Fe(CN)6]3−/4−. a Relative binding affinities of the three newly identified phage clones (~ 1011 PFU/mL). BSA was used as the control. b Effect of CTSB 4-1 concentration on the binding interactions; concentrate range: 108 to 1012 PFU/mL. All the measurements were performed in triplicate, and the error bars represent standard deviations. c Relative binding affinities of the selected phage (CTSB 4-1) clones at different cathepsin B concentrations ranging from 6.25 to 250 nM. d Specificity of the CTSB 4-1 phage (~ 1012 PFU/mL)-based electrochemical sensor using various proteins (BSA, CTSS, CTSL1, CTSB, 125 nM). e Effect of the CTSB 4-1 phage concentration ranging from 108 to 1012 PFU/mL on the binding interactions. f Determination of the binding constant of the selected CTSB 4-1 phage Some researchers have reported M13 peptide-displaying phages instead of antibodies as the recognition elements or chemical scaffolds in biosensing platforms [14, 17] because of their robust, easy, and cost-effective mass production. As proof of concept, a phage sensor was developed to detect cathepsin B. The CTSB 4-1 phage particle identified by phage display was immobilized on an Au electrode pre-functionalized with the MUA-EDC/NHS coupling chemistry, and the detection performance of the fabricated phage sensor was examined by CV, EIS and SWV. Additional file 1: Fig. S2a–c shows the CV, EIS and SWV results obtained for the phage-immobilized electrodes after each step of preparation. In the CV and SWV data, the peak potential shifted after the phage immobilization step because the presence of a biomolecular component can shift the potential of a phage-coated Au electrode [33]. Additionally, the X-ray photoelectron spectroscopy (XPS) survey spectra indicated that the intensities of the peaks of Au (Au 4p3, Au 4d3, Au 4d5, and Au 4f) decreased after the functionalization step performed using the MUA-EDC/NHS coupling, and the peaks almost disappeared after the phage binding. These results were in stark contrast to those of the bare Au electrode (Additional file 1: Fig. S2d). Furthermore, the O 1s, N 1s, and C 1s peak intensities increased significantly upon phage functionalization of the electrode.
The electrochemical response of the phage sensor for cathepsin B detection was evaluated by the specificity analysis of the developed phage. In biosensing, specificity is a critical factor for field testing. Three different cathepsin derivatives, cathepsin S (CTSS), L1 (CTSL1), B and BSA (control) were added to the developed phage sensor, and the binding response was investigated by SWV. As shown in Fig. 2d, among the three cathepsin derivatives, the developed phage sensor exhibited favorable binding to cathepsin B. The high specificity of the peptide-displaying phage-tethered biosensor to cathepsin B indicated an accurate biopanning for isolated affinity peptides against cathepsin B. In the control experiment, the phage sensor did not specifically bind to BSA, as expected. The relative binding affinity of the phage-on-a-sensor with different phage concentrations was measured. A dynamic linear relation was observed with an increase in the phage concentration up to 1012 PFU/mL (Fig. 2e). The developed phage sensor could be an effective electrochemical sensing platform, as it provides an active surface area on the sensor layer. The binding constants of the developed phage sensor and the relative current change (ΔI%) with different cathepsin B concentrations were measured by SWV. The binding constant of the developed phage sensor was 73.73 ± 13.42 nM, and its linear dynamic response was observed in the 0–62.5 nM range (Fig. 2f).
## Chemical synthesis of affinity peptides and their analytical characterization for cathepsin B detection
The CTSB 4-1 phage was selected for sensor development during the biopanning and for the characterization of the affinity peptides using ELISA. The WDMWPSMDWKAE sequence identified in four rounds of the biopanning was used as the chemical motif for rational designing the synthetic peptide derivatives (Additional file 1: Table S4). In CTSB BP1, WDMWPSMDWKAE was attached to the biotin at the C-termini for site-specific immobilization on the streptavidin-coated Au electrodes, which could bind two or three biotin-labelled peptides with each streptavidin. CTSB BP2 was designed to analyze the effect of polar amino acid residues on the binding interactions. Three amino acids in CTSB BP1 were substituted with non-polar residues to generate CTSB BP2 with an amino acid sequence of WGMWPGMGWPAGK-biotin. Specifically, Asp and Ser in the second and sixth positions of CTSB BP1, respectively, were substituted with Gly. CTSB BP3 was synthesized to investigate the effect of polar amino acid residues on the binding interactions with an amino acid sequence of SSTTNSNSTSNTK-biotin. In CTSB BP3, all the CTSB BP1 non-polar amino acid residues were changed to polar residues via substitutions such as the replacement of Trp with Ser at the first position of CTSB BP1. CTSB BP4, with the amino acid sequence of HRHRRHRHRKHHK-biotin, contained all positively charged amino acid residues. In this case, each amino acid on CTSB BP1 was substituted to obtain CTSB BP4. It was used to investigate the effect of positively charged amino acid residues on the binding. To synthesize CTSB BP5 with an amino acid sequence of DDEDDEEDDEDEK-biotin, the amino acids on CTSB BP1 were substituted with negatively charged residues. CTSB BP5 was synthesized to analyze the effect of negatively charged amino acid residues on binding interactions.
To select potential high-affinity peptides from the phage clones, the five synthetic peptides were immobilized on the Au electrode via streptavidin–biotin and MUA-EDC/NHS coupling methods. Each preparation step of the peptide sensor was recorded using CV, EIS, and SWV (Additional file 1: Fig. S3a–c), and the relative binding affinities for cathepsin B were measured by SWV. Further, the intensities of the peaks of O 1s, N 1s, and C 1s increased after the functionalization steps, compared with those observed for bare Au; moreover, the Au peak intensities decreased (Additional file 1: Fig. S3d). The different synthetic peptides (CTSB BP1–BP5, 50 μM) were washed with pure distilled water and immobilized on the streptavidin-modified Au surface layer and incubated with the cathepsin B protein (125 nM) for 1 h. In these experiments, BSA was used as a control, and the three cathepsin derivatives, viz. cathepsin B, CTSS, and CTSL1 were used to determine the binding affinities of the peptides to different proteins. CTSB BP3 exhibited the highest binding affinity to cathepsin B, the primary target protein in the biopanning (Fig. 3a). This result was in agreement with the hypothesis, indicating an accurate identification of the high-affinity peptides under the well-controlled experimental conditions. Unlike CTSB BP3, CTSB BP4 had non-specific binding affinity for BSA, CTSL1, and CTSS with a much lower binding affinity for cathepsin B than for CTSB BP3. This low binding affinity can be attributed to two reasons. First, the positively charged amino acids in the HRHRRHRHRKHHK-biotin amino acid sequence of CTSB BP4 could mediate the peptide–cathepsin B binding interactions. Second, the five Arg residues in CTSB BP4 could be protonated with a higher charge than that of the aliphatic amino acids, leading to weak binding. The relative binding affinity of CTSB BP1 for all the tested proteins was comparable to those of CTSB BP2 and CTSB BP5; however, it did not bind to cathepsin B. Thus, CTSB BP3 was selected for the feasibility tests and characterization studies. Fig. 3Characterization of peptide-based electrochemical sensors using SWV at 1 M KNO3 with 2.5 mM [Fe(CN)6]3−/4−. a Specificity of peptide-on-a-sensor using various proteins (BSA, CTSS, CTSL1, CTSB, 125 nM). b Determination of the binding constant of the peptide-on-a-sensor, along with the results of molecular docking between CTSB BP3 and cathepsin B. c Front and side views of the molecular docking results (docking of the CTSB BP3 peptide with the cathepsin B protein). d Interaction network between the CTSB BP3 peptide and the cathepsin B protein. The arrow indicates the contact distance (Å) The feasibility of the selected peptide (CTSB BP3) tethered on a sensor for cathepsin detection was analyzed by SWV measurements at different peptide concentrations (5–100 μM) (Additional file 1: Fig. S4a). The relative current change (ΔI%) gradually increased up to 50 μM and then slightly decreased at 100 μM peptide concentration (Additional file 1: Fig. S4b). Therefore, 50 μM peptide concentration was selected as the optimal concentration for the characterizations. The change in relative current with increasing cathepsin B concentration, analyzed using SWV (Fig. 3b), exhibited a sigmoidal curve. From these results, the binding constant, Kd was found to be 50.78 ± 7.57 nM.
To identify and elucidate the possible binding sites of the CTSB BP3 peptide with cathepsin B, molecular docking was conducted using the CABS-dock software, and the corresponding docking results are presented in Additional file 1: Table S5. The front and side views of the docking results showed that the CTSB BP3 peptide bound linearly to the cathepsin B protein (Fig. 3c). The binding interfaces of the amino acid residues of the peptide (Ser 2, Thr 3, Asn 5, Ser 10, and Asn 11) interacted with the residue random coil (Glu 98, Ile 99, Asn 301, Asp 306, Asn 307), helix (Lys 97), and β-strand (Glu 288, Phe 309 and Lys 311) under a contact cutoff of 3 Å (Fig. 3d). The interaction between the polar amino acid residues (Ser, Thr, Asn) of the peptide and the electrically charged (Lys, Glu, Asp) and hydrophobic (Ile, Phe) amino acid could be one of the reasons for the high binding affinity. It should be noted that molecular docking is just a simulation without any experimental verification.
## Synthesis and characterization of AuNP–MXene nanocomposites
A phage-on-a-sensor and a peptide-on-a-sensor were developed and their cathepsin B detection performances were evaluated. However, our aim was to develop a more advanced electrochemical sensor with high sensitivity. Efficient and advanced electrochemical sensors can be fabricated by incorporating nanomaterials; they exhibit good electrical conductivity and chemical stability and thus can be operated as biosensors. AuNPs (20 nm in size) in combination with MXene (single-layered form) were used to fabricate the peptide-on-a-AuNPs/MXene sensor, whose cathepsin B detection performance was subsequently evaluated. We expected that the introduction of AuNPs–MXene would significantly improve the electron transfer between the electrode and the electrolyte solutions, compared with other electrode materials because of the large surface area and porosity of the AuNPs. The AuNPs-embedded MXene was synthesized by selective etching of Al from the MAX phases using chemicals depending on the different AuNPs:MXene concentration ratios (0.5–5 mL Au, 50 mg/mL single-layered MXene). The morphology and composition of the synthesized AuNPs–MXene composites were analyzed by scanning electron microscopy and transmission electron microscopy. The AuNPs were attached to the Ti3C2F MXene layers in the AuNPs–MXene composites (Additional file 1: Fig. S5). The AuNPs were synthesized uniformly in spherical shapes through the chemical reduction of AuCl4− with a mean particle diameter of 13.21 ± 1.55 nm (Additional file 1: Fig. S6a and b). Further, a single layer of the 2D Ti3C2F MXene was fabricated in a single step using HCl and LiF salts (Additional file 1: Fig. S6c and d). The atomic ratios of the composites were determined by energy-dispersive X-ray spectroscopic measurements (Additional file 1: Fig. S7), and the Ti:C:F ratio was found to be 3.01:1.88:1.00, which well-matched the stoichiometric ratio of the Ti3C2F MXene (Additional file 1: Fig. S7a). Additionally, the corresponding elemental mapping images confirmed the compositional distribution of the AuNPs–MXene composites (Additional file 1: Fig. S7b–e).
Next, the crystallographic structures and phase purity of the AuNPs–MXene composites were evaluated by X-ray diffraction (XRD). As shown in Additional file 1: Fig. S8a, the peaks at 2θ values of 9.02°, 18.19°, 30.63°, and 60.65° in the XRD patterns could be ascribed to the [002], [006], [008], and [110] planes, respectively, indicating the crystalline structure of the MXene [33]. Additionally, four diffraction peaks at 38.18°, 44.52°, 64.64°, and 77.58°, corresponding to the [111], [200], [220], and [311] planes were detected in the AuNP–MXene composites, indicating the face-centered cubic lattice of the AuNPs [34]. These diffraction peaks indicated that the AuNP attachment had no effect on the MXene crystal structure. The formation of AuNPs–MXene composites was further confirmed by Fourier transform infrared spectroscopy, as shown in Additional file 1: Fig. S8b. The adsorption peaks at 3743.91, 1646.61, 1037.29, and 675.22 cm−1 were associated with the O–H, C=C, C–F, and Ti–O stretching vibrations of the MXene layer [35].
## Verification of the peptide-on-a-AuNPs/MXene sensor performance
After the AuNPs-embedded MXene nanocomposite fabrication, the peptide-on-a-AuNPs/MXene sensor was formed by immobilizing the peptide on the nanocomposite layer. Prior to peptide immobilization, the AuNPs-embedded MXene was attached to the Au electrode sensor layer using $0.1\%$ nafion. As shown in Additional file 1: Fig. S9a and b, 4 mg/mL of the AuNPs-embedded MXene on the Au electrode exhibited a higher conductivity compared with the other concentrations. The active surface areas of bare Au and Au@AuNPs–MXene were extracted from the relationship between the anodic peak current and scan rate of their CV responses using the Randles–Sevick equation (Fig. 4a–c). The active surface area was evaluated to be 0.071 cm2, which is higher than that of the bare Au (0.041 cm2), AuNPs (0.045 cm2) and MXene (0.049 cm2) electrodes at 100 mV/s of scan rate (Additional file 1: Table S6). Furthermore, the electrode charge transfer resistance (Rct) reduced because of the increasing permeability of the redox probes (Fig. 4d). After optimization, CTSB BP3 was immobilized on the functionalized MXene sensor, and its preparation steps were analyzed by electrochemical measurements (Fig. 4e–g) and XPS (Fig. 4h). The results of CV, EIS and SWV for the peptide-on-AuNPs/MXene sensor showed similar responses for each step, and the changes in current and resistance for peptide immobilization were higher than those for the other sensors (phage and only peptide) because of the increased surface area which could bind with the peptide compared with that of the bare Au electrode. Further, the peaks of Au almost disappeared after the fabrication of AuNPs–MXene, whereas the F 1 s and Ti 2p peaks appeared for the Au@AuNPs–MXene electrode. Moreover, the peak intensities of O 1 s, N 1 s, and C 1 s gradually increased according to the peptide immobilization steps. Fig. 4Characterization of AuNPs–MXene fabricated electrode. a–c Effect of scan rate (50 to 400 mVs−1) for bare Au and AuNPs–MXene fabricated Au electrodes determined by CV at 1 M KNO3 with 2.5 mM [Fe(CN)6]3−/4− and Randles-Sevcik plot. In c, all the measurements were done in triplicate, and the error bars represent standard deviations. d Electrical transfer resistance of each electrode (Au, AuNPs–MXene) measured by EIS. e–g CV, EIS, and SWV responses for each preparation step of the peptide-on-a-MXene sensor. h XPS spectra of bare Au, Au@AuNPs–MXene, Au@MUA-EDC/NHS, Au@streptavidin and Au@CTSB BP3 peptide The change in the current for different proteins (BSA, C-reactive protein (CRP), procalcitonin (PCT), CTSS, and CTSL1) which are related to inflammatory response was monitored by SWV to test the cathepsin B detection feasibility of the peptide-on-a-AuNPs/MXene sensor. The sensor exhibited the highest binding affinity for the target protein cathepsin B, and very low binding affinities for CTSS and CTSL1, indicating its specificity and selectivity for cathepsin B. As expected, no specific binding affinity was observed for BSA, CRP, and PCT (Fig. 5a). Additionally, the binding behavior of the sensor dependent on peptide concentration in the 5–100 μM range was analyzed by SWV (Additional file 1: Fig. S10a). The relative current change (ΔI%) gradually increased up to the peptide concentration of 50 μM and then slightly decreased at 100 μM (Additional file 1: Fig. S10), showing a sigmoidal curve as shown in Fig. 5b. This may probably due to the inter/intra-interaction of individual immobilized peptide or steric hindrance on the electrode, resulting in the decrease of binding affinity for cathespsin B. Therefore, 50 μM peptide concentration was selected as the optimal concentration for the subsequent analyses. The binding constant of the peptide-on-a-AuNPs/MXene sensor was found to be 11.64 ± 2.88 nM.Fig. 5Characterization of peptide-on-a-AuNPs/MXene sensor. a Specificity of peptide-on-a-AuNPs/MXene sensor using various proteins (BSA, CRP, CPT, CTSS, CTSL1, CTSB, 125 nM). b Determination of the binding constant of peptide-on-a-AuNPs/MXene sensor. c Comparison of binding affinity between cathepsin B polyclonal antibody (100 μg/mL) and CTSB BP3 peptide (100 μg/mL: ~ 50 μM) on Mxene sensor. d Linear graph of peptide-on-a-AuNPs/MXene sensor. e Comparison of CTSB concentration in Crohn’s disease patient samples (*: $p \leq 0.05$, **, ***: $p \leq 0.05$). f Linear correlation between commercially available ELISA and peptide-on-AuNPs/MXene sensor for CTSB concentration determination in eight human plasma, and serum samples. The SWV measurements were performed in 1 M KNO3 containing 2.5 mM [Fe(CN)6]3−/4−. All the measurements were done in triplicate, and the error bars represent standard deviations. In the case of Fig. 5e, each sample was analyzed three times, resulting in a total of 24 measurements
## Determination of electrochemical response of the sensors for cathepsin B detection
To compare the cathepsin B detection performances, three different sensors were developed, and the relative current change (ΔI%) shown by these sensors with increasing cathepsin B concentration from 3.9 to 125 nM was observed using SWV. The binding affinity of all the developed sensors increased proportionally with increasing protein concentration. The binding constants (Kd) of the three developed sensors, viz. phage-on-a-sensor, peptide-on-a-sensor, and peptide-on-a-AuNPs/MXene sensors were calculated to be 73.73 ± 13.42, 50.78 ± 7.57, and 11.64 ± 2.88 nM, respectively. Furthermore, the cathepsin B detection ability of the peptide-on-a AuNPs/MXene sensor was similar to that of the cathepsin B antibody-immobilized MXene sensor (Fig. 5c).
Their LOD and limit of quantitation (LOQ) were calculated using the following equations:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{LOD}} = 3 \times \rm{\delta }/{\text{s}}$$\end{document}LOD=3×δ/s3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{LOQ}} = 10 \times \rm{\delta }/{\text{s}}$$\end{document}LOQ=10×δ/swhere δ and s are the standard deviation of the y-intercept and slope of the standard curve, respectively. As shown in the Fig. 5d, Additional file 1: Fig. S11 and Table 1, the LOD and LOQ of the peptide-on-a-AuNPs/MXene sensor exhibited the most sensitive cathepsin B detection were 0.18 and 0.59 nM, respectively, which were determined using the equation: $y = 0.261$x + 0.912, where R2 = 0.99, δ = 0.015, and $s = 0.261.$ The LOD and LOQ of the phage-on-a-sensor for cathepsin B detection were 0.62 and 2.07 nM, respectively, obtained using the equation: $y = 0.093$x − 0.5392, where R2 = 0.96, δ = 0.192, and $s = 0.093.$ The corresponding values for the peptide-on-a-sensor were 0.33 and 1.11 nM, respectively, which were derived using the equation: $y = 0.626$x − 0.046, where R2 = 0.97, δ = 0.069, and $s = 0.626.$ Among the three developed sensors, the peptide-on-a-AuNPs/MXene sensor exhibited the most sensitive cathepsin B detection. Table 1Comparison of sensor performances for cathepsin B detectionSensor performancePhage-on-a-sensorPeptide-on-a-sensorPeptide-on-a-AuNPs/MXene sensorLOD (nM)0.620.330.18LOQ (nM)2.071.110.59Kd (nM)73.73 ± 13.4250.78 ± 7.5711.64 ± 2.88Average recovery (%)91.05–93.5896.03–99.4394.52–101.16
## Reproducibility and stability of the cathepsin B sensors
Reproducibility and recovery of biosensors are considered key parameters for commercialization. Therefore, the reproducibility and recovery of the developed sensors were investigated with different cathepsin B concentrations (31.3‒125 nM) spiked in human plasma samples; these analyses were conducted using SWV. The acceptable coefficient of variation and percentage recoveries for the developed sensors were remarkable (Additional file 1: Table S7). The recovery percentage and coefficient of variation (%) were in the range of 96.03–$99.43\%$, and 1.30–6.99, respectively, for the peptide-on-a-sensor, whereas they were in the range of 91.05–$93.58\%$, and 3.85–8.89 for the phage-on-a-sensor. Thus, the recovery percentage and CV (%) values of the peptide-on-a-sensor were much higher than those of the phage-on-a-sensor, indicating that the phage-on-a-sensor could be used as a building block, i.e., as an alternative recognition element. However, its application in clinical testing is not recommended. In contrast, the recovery percent and coefficient of variation (%) values of the peptide-on-a-AuNPs/MXene sensor were in the range of 94.52–$101.16\%$, and 1.68–6.04, respectively, indicating a reproducible and accurate cathepsin B detection ability of this sensor in a complex biological sample (human plasma).
Another vital parameter of a biosensor is its stability. Therefore, the stability of the three different sensors was monitored for 6 days after peptide immobilization on each sensor layer. For the phage-on-a-sensor, the relative current change (ΔI%) decreased by $3.3\%$ on day 1, and gradually dropped by $34.7\%$ from its initial value after the sensor was stored overnight at 4 ℃ (Additional file 1: Fig. S12a). The peptide-on-a-sensor signal was stable on day 1, but its relative current change (ΔI%) decreased by $17.5\%$ on day 2 and further gradually decreased to $43.8\%$ on day 6 (Additional file 1: Fig. S12b). Interestingly, under similar conditions, the relative current change (ΔI%) of the peptide-on-a-AuNPs/MXene sensor remained stable up to day 4, and finally dropped to $29.4\%$ from its initial value (Additional file 1: Fig. S13c), indicating high sensor stability. This could be due to the introduction of peptides on the single-layered MXene surface resulting in an increased electrostatic interaction (attraction or repulsion) which firmly held the peptides between each MXene layer, improving the structural stability and affinity of the peptides. The analytical cathepsin B detection performances of the developed sensors are compared in Additional file 1: Table S8 along with other detection methods.
## Electrochemical sensing with real patient samples
Real testing with liquid biopsy samples such as blood, nucleic acid, and protein derived from human patients is essential for biosensor performance evaluation. Therefore, as proof-of-concept, SWV measurements of diluted crude plasma (1:100 in PBS buffer, $$n = 5$$) and serum (1:100 in PBS buffer, $$n = 3$$) obtained from patients suffering from Crohns’s disease ($$n = 8$$ in total) were performed. The current change observed in the SWV data is shown in Fig. 5e. Interestingly, statistically significant differences in the relative current change (ΔI%) were observed among the mild, moderate, and severe groups using the developed electrochemical sensor (as analyzed with two-way ANOVA). The change in current for both the moderate and severe groups of patients was significantly different ($p \leq 0.05$) from that of the moderate group ($p \leq 0.05$), indicating that the expression level of cathepsin B could be highly patient-dependent. Cathepsin B was up-expressed in both the moderate (up to 16 nM) and severe (up to 32 nM) groups, while the cathepsin B concentration was down to 0.7 nM in the mild group. The following order of cathepsin B levels was obtained: severe group > moderate group > mild group. This result is consistent with those of previous reports [11, 36] on cathepsin B expression in different stages of Crohn’s disease and ulcerative colitis disease. Further, this result also suggests a good correlation between increased cathepsin B concentration and a variety of diseases [37]. As shown in Fig. 5f and Additional file 1: Table S9, the cathepsin B levels measured by the fabricated peptide-on-a-AuNPs/MXene sensor were correlated with those estimated by the ELISA with good recoveries (86–$102\%$) and %RSDs (< $11\%$) in all the tested cases. Although a small number of patient samples were tested, the applicability of the developed peptide sensor was comparable to the reference (commercially available ELISA). Thus, the developed electrochemical peptide sensor exhibited good detection performance with clinical patient samples and could distinguish different stages of Crohn’s disease.
## Conclusion
In this study, a novel biosensor was developed and characterized for the electrochemical detection of cathepsin B based on affinity peptides as alternative recognition elements. The M13 phage library was biopanned to identify high-affinity peptides specific for cathepsin B. The binding affinities and efficacies of the developed sensor were characterized by ELISA and through electrochemical experiments (CV, EIS and SWV). In the phage-on-a-sensor, as proof-of-concept, the phage particles were attached to an Au electrode via the MUA-EDC/NHS chemistry. The particles exhibited high biocompatibility and stability, thereby fulfilling the requirements for potent bioanalytical applications. Through a rational design and chemical synthesis of peptides away from the phage particles selected peptides were used for the construction of the peptide-on-a-sensor system. The developed peptide-on-a-sensor system exhibited good electrochemical sensing performance. Subsequently, AuNPs were introduced on the MXene layer to develop a more advanced electrochemical sensor for cathepsin B detection. The fabricated peptide-on-a-AuNPs/MXene sensor exhibited a low LOD (0.18 nM), low LOQ (0.59 nM), high recovery percentage (94.5–$101.2\%$), and high stability (four days) as well as enabled a highly accurate detection of cathepsin B, even in complex biological samples. The clinical applicability of the peptide-on-a-AuNPs/MXene sensor was evaluated via cathepsin B detection in Crohn’s patient samples. The sensor exhibited an accurate and sensitive cathepsin B detection with a good recovery (86–$102\%$) and %RSD (< $11\%$) and could distinguish between the different stages of Crohn’s disease. Furthermore, cathepsin B concentrations measured by the developed sensor correlated well with those estimated by the commercially available ELISA kit. Although our developed sensor demonstrated a high specificity and good recovery for cathepsin B, further investigations are required before it can be applied to create a peptide sensor for real clinical testing. First, the use of more functional peptides obtained via virtual re-screening with additional modifications and introduction of other nanomaterials is a challenging task. Second, the rational design of a protease-resistant peptide probe is required to minimize the inaccessibility of liquid biopsy or complex solutions in the presence of various proteases. One limitation of our electrochemical techniques including CV, EIS and SWV is the need to operate the electrode in a ferri/ferro cyanide solution (redox mediator) after the target sample is introduced on the peptide-immobilized sensor layer. This may sometimes affect the viscosity change in the electrical measurements. In summary, to the best of our knowledge, this is the first example of a rational design of affinity peptides and the development of a label-free sensing platform for cathepsin B using peptides and MXene nanocomposites. This new sensing system could be useful for the development of peptide-based sensors to detect any desired target by changing the amino acid sequence of the peptides and could be combined with various functional nanomaterials.
## Supplementary Information
Additional file 1: Table S1. Clinical characteristics of individuals in the detection of cathepsin B. Table S2. The yield of biopanning for cathepsin B. Table S3. Screening results of biopanning for identifying cathepsin B specific peptides via phage display. Table S4. The synthetic peptides for detection of cathepsin B. Table S5. Analysis of binding site of CTSB BP3 peptide for cathepsin B using molecular docking. Table S6. The electroactive surface area of bare gold and AuNPs–MXene fabricated electrodes. Table S7. Detection capability of developed sensors at different concentrations of cathepsin B spiked in human plasma. Table S8. Comparison of the analytical performance for the detection of cathepsin B. Table S9. Analytical results between our developed sensor and the reference ELISA with Crohn’s patient samples. Figure S1. The validation assay for biotinylated cathepsin B protein. ( a) The biotin-labeled BSA (250 nM = 16.5 μg/mL) and CTSB (250 nM = 6.875 μg/mL) were immobilized on streptavidin coated plate, and it was confirmed using HRP conjugated streptavidin and ABTS at 405 nm. ( b) To quantitate labelled biotin, the biotinylated cathepsin B was added to the mixture of HABA and avidin and biotinylated HRP and BSA were used as control. ( c) Biotin ratio was calculated from biotin quantitation results, which biotin ratio for cathepsin B was about 3.745. Figure S2. The responses of electrochemical measurements of phage-on-a sensor. ( a–c) CV, EIS and SWV responses for preparation steps of phage sensor (bare Au, Au@1 mM MUA, Au@ 400 mM/100 mM EDC/NHS, Au@4-1 phage). It was conducted at 1 M KNO3 with 2.5 mM [Fe(CN)6]3−/4−. (d) X-ray photoelectron spectroscopy (XPS) spectra of bare Au, Au@MUA-EDC/NHS, Au@4-1 phage. Figure S3. The responses of electrochemical measurements of peptide-on-a sensor. ( a–c) CV, EIS, and SWV responses for preparation steps of phage sensor (bare Au, Au@1 mM MUA, Au@ 400 mM/100 mM EDC/NHS, Au@streptavidin, Au@CTSB BP3 peptide). It was conducted at 1 M KNO3 with 2.5 mM [Fe(CN)6]3−/4−. (d) X-ray photoelectron spectroscopy (XPS) spectra of bare Au, Au@MUA-EDC/NHS, Au@streptavidin, Au@CTSB BP3 peptide. Figure S4. Effect of peptide concentration for peptide-on-a sensor. ( a) The current responses of different CTSB BP3 peptide concentrations (5–100 μM). ( b) The current changes of peptide-on-a sensor in different CTSB BP3 peptide concentrations for cathepsin B. Figure S5. Morphological characterization of AuNPs-MXene composite; SEM images under (a) low and (b) high resolution; TEM images under (c) low and (d) high resolution. Figure S6. ( a) TEM image and (b) size distribution graph of pre-synthesized AuNPs. ( c) TEM image and (d) SEM image of Ti3C2F MXene. Figure S7. ( a) EDS spectrum of AuNPs-MXene composites. ( b-e) Elemental mapping images obtained from EDS for elements Ti, C, F and Au, respectively. Figure S8. ( a) X-ray diffraction spectra and (b) FT-IR spectra of AuNPs-MXene composites (red line) and MXene (black line). Figure S9. Optimization of AuNP-MXene concentration and comparison of the effect of scan rate on gold electrode. ( a) Chronoamperometric response of the sensor with different concentration of AuNP-MXene. ( b) Square wave voltammetry analysis of the sensor with different concentration of AuNP-Mxene. Figure S10. Effect of peptide concentration for peptide-on-a-AuNPs/MXene sensor. ( a) The current responses of different CTSB BP3 peptide concentrations (5–100 μM). ( b) The current changes of peptide-on-a-AuNPs/MXene sensor in different CTSB BP3 peptide concentrations for cathepsin B. Figure S11. Comparison of standard curves of the developed sensors: a) Phage-on-a-sensor, b) Peptide-on-a-sensor. The change of current was measured by SWV with 1 M KNO3 containing 2.5 mM [Fe(CN)6]3−/4−. In phage-on-a-sensor, the titer of phages was 1 × 1012 PFU/mL, while the concentration of peptide used in peptide-on-a-sensor was 50 μM. All measurements were done in triplicate, and error bars represent standard deviations. Figure S12. Comparison of stability of the developed sensor. ( a) Phage on a sensor, (b) peptide on a sensor, (c) peptide on a AuNPs/MXene sensor.
## References
1. Mayeux R. **Biomarkers: potential uses and limitations**. *NeuroRx* (2004) **1** 182-188. DOI: 10.1602/neurorx.1.2.182
2. Turk V, Turk B, Guncar G, Turk D, Kos J. **Lysosomal cathepsins: structure, role in antigen processing and presentation, and cancer**. *Adv Enzyme Regul* (2002) **42** 285-303. DOI: 10.1016/S0065-2571(01)00034-6
3. Berdowska I. **Cysteine proteases as disease markers**. *Clin Chim Acta* (2004) **342** 41-69. DOI: 10.1016/j.cccn.2003.12.016
4. Araujo TF, Cordeiro AV, Vasconcelos DAA, Vitzel KF, Silva VRR. **The role of cathepsin B in autophagy during obesity: a systematic review**. *Life Sci* (2018) **209** 274-281. DOI: 10.1016/j.lfs.2018.08.024
5. Campo E, Munoz J, Miquel R, Palacín A, Cardesa A, Sloane BF, Emmert-Buck MR. **Cathepsin B expression in colorectal carcinomas correlates with tumor progression and shortened patient survival**. *Am J Clin Pathol* (1994) **145** 301-309
6. Yan S, Sameni M, Sloane B. **Cathepsin B and human tumor progression**. *Bio Chem* (1998) **379** 113-123. PMID: 9524062
7. Withana NP, Blum G, Sameni M, Slaney C, Anbalagan A, Olive MB, Bidwell BN, Edgington L, Wang L, Moin K. **Cathepsin B inhibition limits bone metastasis in breast cancer**. *Cancer Res* (2012) **72** 1199-1209. DOI: 10.1158/0008-5472.CAN-11-2759
8. Vizovišek M, Vidak E, Javoršek U, Mikhaylov G, Bratovš A, Turk B. **Cysteine cathepsins as therapeutic targets in inflammatory diseases**. *Expert Opin Ther Targets* (2020) **24** 573-588. DOI: 10.1080/14728222.2020.1746765
9. Turk V, Stoka V, Vasiljeva O, Renko M, Sun T, Turk B, Turk D. **Cysteine cathepsins: from structure, function and regulation to new frontiers**. *BBA-Proteins Proteom* (2012) **1824** 68-88. DOI: 10.1016/j.bbapap.2011.10.002
10. Vidak E, Javoršek U, Vizovišek M, Turk B. **Cysteine cathepsins and their extracellular roles: shaping the microenvironment**. *Cells* (2019) **8** 264. DOI: 10.3390/cells8030264
11. Menzel K, Hausmann M, Obermeier F, Schreiter K, Dunger N, Bataille F, Falk W, Scholmerich J, Herfarth H, Rogler G. **Cathepsins B, L and D in inflammatory bowel disease macrophages and potential therapeutic effects of cathepsin inhibition in vivo**. *Clin Exp Immunol* (2006) **146** 169-180. DOI: 10.1111/j.1365-2249.2006.03188.x
12. Solà-Tapias N, Vergnolle N, Denadai-Souza A, Barreau F. **The interplay between genetic risk factors and proteolytic dysregulation in the pathophysiology of inflammatory bowel disease**. *J Crohn's Colitis* (2020) **14** 1149-1161. DOI: 10.1093/ecco-jcc/jjaa033
13. Soond SM, Kozhevnikova MV, Townsend PA, Zamyatnin AA. **Cysteine cathepsin protease inhibition: an update on its diagnostic, prognostic and therapeutic potential in cancer**. *Pharmaceuticals* (2019) **12** 87. DOI: 10.3390/ph12020087
14. Sidhu SS. **Phage display in pharmaceutical biotechnology**. *Curr Opin Biotechnol* (2000) **11** 610-616. DOI: 10.1016/S0958-1669(00)00152-X
15. Kim JH, Cho CH, Shin JH, Hyun MS, Hwang E, Park TJ, Park JP. **Biomimetic isolation of affinity peptides for electrochemical detection of influenza virus antigen**. *Sens Actuators B Chem* (2021) **343** 130161. DOI: 10.1016/j.snb.2021.130161
16. Wang Y, Selvamani V, Yoo I-K, Kim TW, Hong SH. **A novel strategy for the microbial removal of heavy metals: cell-surface display of peptides**. *Biotechnol Bioprocess Eng* (2021) **26** 1-9. DOI: 10.1007/s12257-020-0218-z
17. Cho CH, Kim JH, Song D-K, Park TJ, Park JP. **An affinity peptide-incorporated electrochemical biosensor for the detection of neutrophil gelatinase-associated lipocalin**. *Biosens Bioelectron* (2019) **142** 111482. DOI: 10.1016/j.bios.2019.111482
18. Hwang HJ, Ryu MY, Park CY, Ahn J, Park HG, Choi C, Ha S-D, Park TJ, Park JP. **High sensitive and selective electrochemical biosensor: label-free detection of human norovirus using affinity peptide as molecular binder**. *Biosens Bioelectron* (2017) **87** 164-170. DOI: 10.1016/j.bios.2016.08.031
19. Lim JM, Kim JH, Ryu MY, Cho CH, Park TJ, Park JP. **An electrochemical peptide sensor for detection of dengue fever biomarker NS1**. *Anal Chim Acta* (2018) **1026** 109-116. DOI: 10.1016/j.aca.2018.04.005
20. Xu P, Ghosh S, Gul AR, Bhamore JR, Park JP, Park TJ. **Screening of specific binding peptides using phage-display techniques and their biosensing applications**. *TrAC-Trends Anal Chem* (2021) **137** 116229. DOI: 10.1016/j.trac.2021.116229
21. Madej M, Kochana J, Bas B. **Selective and highly sensitive voltammetric determination of citalopram with glassy carbon electrode**. *J Electrochem Soc.* (2019) **166** 359-369. DOI: 10.1149/2.1291908jes
22. Kim JH, Shin JH, Cho CH, Hwang J, Kweon D, Park TJ, Choi C, Park JP. **Dual synergistic response for the electrochemical detection of H1N1 virus and viral proteins using high affinity peptide receptors**. *Talanta* (2022) **248** 123613. DOI: 10.1016/j.talanta.2022.123613
23. Asif M, Aziz A, Azeema M, Wang Z, Ashraf G, Xiao F, Chen X, Liu H. **A review on electrochemical biosensing platform based on layered double hydroxides for small molecule biomarkers determination**. *Adv Colloid Interface Sci* (2018) **262** 21-38. DOI: 10.1016/j.cis.2018.11.001
24. Aziz A, Asif M, Azeem M, Ashraf G, Wang Z, Xiao F, Liu H. **Self-stacking of exfoliated charged nanosheets of LDHs and graphene as biosensor with real-time tracking of dopamine from live cells**. *Anal Chim Acta* (2019) **1047** 197-207. DOI: 10.1016/j.aca.2018.10.008
25. Ashraf G, Asif M, Aziz A, Iftikhar T, Liu H. **Rice-spikelet-like copper oxide decorated with platinum stranded in the CNT network for electrochemical in vitro detection of serotonin**. *ACS Appl Mater Interfaces* (2021) **13** 6023-6033. DOI: 10.1021/acsami.0c20645
26. Haghparas Z, Kordrostami Z, Sorouri M, Rajabzadeh M, Khalifeh R. **Fabrication of non-enzymatic electrochemical glucose sensor based on nano-copper oxide micro hollow-spheres**. *Biotechnol Bioprocess Eng* (2020) **25** 528-535. DOI: 10.1007/s12257-020-0058-x
27. Lee SM, Shin S, Kim HY, Won BY, Ahn JK, Park KS, Park HG. **Washing-free electrochemical strategy to detect target DNA utilizing peroxidase mimicking DNAzyme**. *Biotechnol Bioprocess Eng* (2020) **25** 707-714. DOI: 10.1007/s12257-020-0272-6
28. Rajaram R, Kanagavalli P, Senthilkumar S, Mathiyarasu J. **Au nanoparticle-decorated nanoporous PEDOT modified glassy carbon electrode: a new electrochemical sensing platform for the detection of glutathione**. *Biotechnol Bioprocess Eng* (2020) **25** 715-723. DOI: 10.1007/s12257-020-0065-y
29. Wu X, Ma P, Sun Y, Du F, Song D, Xu G. **Application of MXene in electrochemical sensors: a review**. *Electroanalysis* (2021) **33** 1827-1851. DOI: 10.1002/elan.202100192
30. Liu X, Chen L, Yang Y, Xu L, Sun J, Gan T. **MXene-reinforced octahedral PtCu nanocages with boosted electrocatalytic performance towards endocrine disrupting pollutants sensing**. *J Hazard Mater* (2023) **442** 13000. DOI: 10.1016/j.jhazmat.2022.130000
31. Xu Y, Wang X, Ding C, Luo X. **Ratiometric antifouling electrochemical biosensors based on multifunctional peptides and MXene loaded with Au nanoparticles and methylene Blue**. *ACS Appl Mater Inter* (2021) **13** 20388-20396. DOI: 10.1021/acsami.1c04933
32. Naguib M, Kurtoglu M, Presser V, Lu J, Niu J, Heon M, Hultman L, Gogotsi Y, Barsoum MW. **Two-dimensional nanocrystals produced by exfoliation of Ti3AlC2**. *Adv Mater* (2011) **23** 4248-4253. DOI: 10.1002/adma.201102306
33. Sargent A, Loi T, Gal S, Sadik OA. **The electrochemistry of antibody-modified conducting polymer electrodes**. *J Electroanal Chem* (1999) **470** 144-156. DOI: 10.1016/S0022-0728(99)00231-4
34. Krishnamurthy S, Esterle A, Sharma NC, Sahi SV. **Yucca-derived synthesis of gold nanomaterial and their catalytic potential**. *Nanoscale Res Lett* (2014) **9** 1-9. DOI: 10.1186/1556-276X-9-627
35. Li Y, Zhou X, Wang J, Deng Q, Li M, Du S, Han Y-H, Lee J, Huang Q. **Facile preparation of in situ coated Ti**. *RSC Adv* (2017) **7** 24698-708. DOI: 10.1039/C7RA03402D
36. Murnane MJ, Sheahan K, Ozdemirli M, Shuja S. **Stage-specific increases in cathepsin B messenger RNA content in human colorectal carcinoma**. *Cancer Res* (1991) **51** 1137-1142. PMID: 1997157
37. Thangavelu B, Boutté AM. **Single Molecule assay for ultrasensitive detection of cathepsin B in human blood**. *ACS Omega* (2021) **6** 9609-16. DOI: 10.1021/acsomega.1c00180
|
---
title: 'Granulocyte colony-stimulating factor priming improves embryos and pregnancy
rate in patients with poor ovarian reserve: a randomized controlled trial'
authors:
- Masao Jinno
- Yukoku Tamaoka
- Koji Teruya
- Aiko Watanabe
- Naohisa Hatakeyama
- Tomoya Goda
- Hayato Kimata
- Yuichi Jinno
journal: 'Reproductive Biology and Endocrinology : RB&E'
year: 2023
pmcid: PMC10029156
doi: 10.1186/s12958-023-01082-w
license: CC BY 4.0
---
# Granulocyte colony-stimulating factor priming improves embryos and pregnancy rate in patients with poor ovarian reserve: a randomized controlled trial
## Abstract
### Background
Granulocyte colony-stimulating factor (G-CSF) administration increased ovarian preantral follicles and anti-Müllerian hormone (AMH) in animal models with diminished ovarian reserve. We investigated whether G-CSF priming before treatment with assisted reproductive technology (ART) improved embryo development and pregnancy rate while increasing serum AMH in patients with poor ovarian reserve.
### Methods
In this prospective randomized open-label controlled trial, 100 patients 20 to 42 years old with AMH below 2 ng/mL were randomized to priming or control groups (50 patients each). None had over 1 ART failure, day-3 follicle-stimulating hormone (FSH) above 30 IU/L, uterine anomalies, or a partner with azoospermia. All patients initially underwent conventional infertility treatment for 2 consecutive cycles in which the priming group but not controls received a subcutaneous G-CSF priming injection during the early luteal phase. Each group then underwent 1 cycle of in vitro fertilization/intracytoplasmic sperm injection and fresh embryo transfer (IVF/ICSI-fresh ET), followed by cryopreserved ET if needed until live birth or embryo depletion. AMH was measured before and after priming.
### Results
Fertilization rate, embryonic development, and implantation rate by fresh ET were significantly improved by priming. Clinical and ongoing pregnancy rates by IVF/ICSI-fresh ET were significantly higher with priming ($30\%$ and $26\%$ in 47 ART patients; 3 delivered with conventional treatment) than in controls ($12\%$ and $10\%$ in 49 ART patients; 1 dropped out). With priming, significantly more patients achieved cryopreservation of redundant blastocysts. The cumulative live birth rate was $32\%$ in 50 patients with priming, significantly higher than $14\%$ in 49 controls (relative risk, 2.8; $95\%$ confidence interval, 1.04–7.7). Infants derived from priming had no congenital anomalies, while infant weights, birth weeks, and Apgar scores were similar between groups. Among 4 variables (age, day-3 FSH, AMH, and priming), logistic regression significantly associated age and priming with cumulative live birth. Priming significantly increased serum AMH. No adverse effects of priming were observed.
### Conclusion
G-CSF priming improved embryonic development and pregnancy rate during ART treatment and increased AMH in patients with poor ovarian reserve. Enhanced preantral follicle growth likely was responsible.
### Trial registration
UMIN registration in Japan (UMIN000013956) on May 14, 2014. https://www.umin.ac.jp/ctr/index.htm.
## Background
Ten years ago, we treated repeated implantation failure in 10 women with diminished ovarian reserve by administering granulocyte colony-stimulating factor (G-CSF) in association with embryo transfers (ET), based upon previous reports [1, 2]. No short-term outcome improvements resulted, but serendipitously 3 of the 10 spontaneously became pregnant 2 months later. One of these pregnancies involved twins in a 45-year-old woman without ovulation induction. We hypothesized that G-CSF administration might have stimulated preantral follicle growth, with improved ovulation after 2 months.
In diabetic rats, administration of G-CSF consistently decreased ovarian follicular degeneration as well as degeneration and fibrosis of ovarian stroma, while increasing serum anti-Müllerian hormone (AMH) concentrations [3]. G-CSF administration also significantly increased ovarian preantral follicles and serum AMH in rats with diminished ovarian reserve induced by cisplatin [4].
Considering our experience with the 3 patients and these G-CSF effects in animal models, we designed the present prospective randomized clinical trial examining whether G-CSF administration in the early luteal phases of each of 2 cycles (G-CSF priming) preceding the cycle involving assisted reproductive technology (ART) improved embryonic development and pregnancy rate following ART in patients with poor ovarian reserve, as well as whether priming increased serum AMH concentrations. We also investigated associations between effects of G-CSF priming and killer-cell immunoglobulin-like receptor (KIR) types, since some types have been associated with improved implantation following G-CSF administration on the day of ET [1].
## Study design
Our prospective open-label randomized clinical trial investigated whether G-CSF priming preceding ART enhanced preantral follicle growth, thus increasing the ART pregnancy rate in patients with poor ovarian reserve. Between May 19, 2014 and November 26, 2018, a total of 465 patients sought ART treatment at Women’s Clinic Jinno; 111 met study inclusion criteria. Eleven declined participation, leaving 100 to be enrolled and randomly assigned to groups undergoing or not undergoing G-CSF priming prior to standard ART. Randomization involved patients drawing from a box containing group assignments in sealed envelopes mixed 1:1 (Fig. 1). Neither patients nor investigators were blinded to resulting assignments. Fig. 1Participant flow diagram. The rate of live delivery among patients was $32\%$ ($\frac{16}{50}$) in the G-CSF group, significantly higher than $14\%$ ($\frac{7}{49}$) in controls (chi-squared test). G-CSF, granulocyte colony-stimulating factor; OPU, oocyte pick-up; IVF, in vitro fertilization; ICSI, intracytoplasmic sperm injection; ET, embryo transfer. a All 3 of these patients conceived with spontaneous ovulation following menstruation after the initial administration of G-CSF To more clearly detect G-CSF-related improvement of ovarian reserve, we limited our study to patients with mildly to moderately decreased ovarian reserve, excluding patients with severe diminution. We chose a serum AMH concentration lower than 2 ng/mL as an inclusion criterion for our study, given that $93\%$ of infertile women in their forties have been found to have values below 2 ng/mL [5]. An earlier study found 1.9 ng/mL to be the median serum AMH concentration among nulliparous 38-year-old Japanese women, a population on the verge of a precipitous decline in fertility [6].
Based on such considerations, our inclusion criteria were age between 20 and 42 years; no more than 1 prior oocyte retrieval attempt; serum AMH concentration below 2 ng/mL; day-3 serum FSH concentration below 30 IU/L; a medical history free of serious allergic disease, severe hepatic, renal, or heart disease, or uterine infertility; and a male partner without azoospermia.
All enrolled patients initially received conventional infertility treatments in 2 consecutive cycles. ( Since no patient in either group had bilateral tubal occlusion, we believed that a chance of pregnancy using conventional treatment existed.) During these 2 cycles the G-CSF group, but not the control group, underwent subcutaneous administration of G-CSF (100 μg of lenograstim; Neutrogin, Chyugai Pharmaceuticals, Tokyo, Japan) during the early luteal phase, based upon basal body temperature records, vaginal ultrasonographic findings, and, if necessary, serum progesterone determinations (Fig. 1). Controls received no placebo. Conventional infertility treatments included sexual intercourse or intrauterine insemination with or without ovarian stimulation by clomiphene citrate or a recombinant FSH regimen.
Both groups consisted of patients who had failed to conceive with conventional infertility treatments and then underwent ART using controlled ovarian stimulation. Embryos were transferred 2, 3, or 5 days after in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI). Remaining embryos were cryopreserved at the blastocyst stage. When IVF/ICSI and fresh embryo transfer (ET) failed to result in delivery, cryopreserved embryos were thawed and transferred (cryopreserved ET) in subsequent spontaneous cycles. During actual ART treatment using fresh or cryopreserved ET, neither group received G-CSF.
Clinical pregnancy and ongoing pregnancy were diagnosed by ultrasonographic detection of a gestational sac and fetal heart movements, respectively. Abortion and delivery were defined as pregnancy loss before 22 weeks and birth after 22 weeks, respectively. Clinical pregnancy rates by IVF/ICSI-fresh ET (per ovarian stimulation) were compared between groups as the primary outcome measure. Cumulative live delivery rates, follicular growth, fertilization, and embryonic development were compared between groups as secondary outcome measures.
To elucidate stimulatory effects of G-CSF priming on preantral follicle growth, pre- vs. post-treatment changes in serum AMH concentration were compared between groups. Serum samples were obtained within 4 months preceding study enrollment and just before initiating ovarian stimulation for ART (Fig. 1).
To achieve a power of 0.8 and an α error of 0.05, the minimum number of participants required to identify a difference between hypothetical pregnancy rates of $10\%$ and $25\%$ in control and G-CSF groups was 113 patients per group, so we planned to recruit those numbers. However, interim assessment halfway through the expected trial duration already showed statistically significant benefit from G-CSF priming that was greater than expected. We therefore ended our trial with 50 enrollees per group, considering the ethical importance of offering potential benefits of G-CSF priming to control patients in a timely manner.
Informed consent was obtained from all patients. The study was approved by the Ethics Committees of both the Inagi Municipal Hospital and Women’s Clinic Jinno, and registered with the UMIN in Japan (UMIN000013956).
## Assisted reproductive technology
Follicular development was stimulated with the long protocol as described previously [7]. Briefly, buserelin acetate (Buserecur; Fuji Pharmaceuticals, Tokyo, Japan) at 900 µg per day was administered nasally from the mid-luteal phase until hCG administration. Human menopausal gonadotropin, (hMG, 300 IU i.m.; Ferring Pharmaceuticals, Tokyo, Japan) was administered daily from day 3. Human chorionic gonadotropin (hCG, 10 000 IU i.m.; Mochida Pharmaceuticals, Tokyo, Japan) was administered when the dominant follicle reached a diameter of 19 mm.
Oocytes were collected transvaginally 36 h after hCG administration and inseminated as described previously [8]. ICSI was performed when the male partner had severe infertility (sperm count < 5 × 106 per mL and/or motility < $20\%$). Oocytes were considered fertilized when 2 pronuclei were observed 17 to 19 h after insemination or ICSI. At 2, 3, or 5 days after oocyte retrieval, embryos were transferred to the uterus according to number and quality of developing embryos for each patient. Progesterone (25 mg i.m.) was administered daily after ET.
Redundant embryos were cultured for 5 to 6 days after IVF/ICSI to the blastocyst stage and cryopreserved using a vitrification method. Thawed cryopreserved blastocysts were transferred to uteri on luteal day 5 of a spontaneous natural cycle, as described previously [7]. For luteal support, 5000 IU of hCG was administered on luteal days 5, 7, and 9.
## Evaluations of hormones and killer-cell immunoglobulin-like receptor genotypes
Serum concentrations of AMH, follicle-stimulating hormone (FSH), luteinizing hormone (LH), 17β-estradiol (E2), prolactin (PRL), testosterone (T), thyroid-stimulating hormone (TSH), free thyronine (FT3), free thyroxine (FT4), and dehydroepiandrosterone sulfate (DHEA-S) were measured by enzyme chemiluminescent immunoassays on cycle day 3 within 3 months before study enrollment.
Sensitivities and intra- and interassay coefficients of variation were 0.01 ng/mL ($1.4\%$, $0.8\%$) for AMH, 0.06 IU/L ($2.3\%$, $1.0\%$) for FSH, 0.11 IU/L ($6.6\%$, $3.4\%$) for LH, 5.0 pg/mL ($0.7\%$, $0.9\%$) for E2, 0.10 ng/mL ($1.2\%$, $1.4\%$) for PRL, 0.03 ng/mL ($2.5\%$, $3.9\%$) for T, and 2 μg/dL ($6.5\%$, $2.7\%$) for DHEA-S.
DNA was genotyped for 16 KIR genes using PCR-SSOP (sequence-specific oligonucleotide probe) using a commercial kit (LABType KIR SSO Genotyping Test; One Lambda, Canoga Park, CA, USA) and Luminex 100 technology (Austin, TX, USA) as previously described [9].
## Statistical analysis
IBM SPSS Statistics Version 27 (IBM, Tokyo, Japan) was used for statistical analyses. Normality was tested by the Shapiro–Wilk test. If data were not normally distributed, analysis was performed using the Mann–Whitney U test or the Wilcoxon matched-pairs signed rank test as appropriate. If data were normally distributed, unpaired t tests or paired t tests were performed as appropriate. Data also were analyzed using the chi-squared test, Fisher’s exact test, or multiple logistic regression analysis as appropriate. P values below 0.05 were considered to indicate significance. Whenever appropriate, results are presented as the mean ± standard deviation (SD).
## Baseline characteristics of patients
Except for TSH, no significant differences concerning baseline characteristics were evident between the 2 groups completing the study (Table 1). Although TSH was significantly lower in the G-CSF group, these values were within the normal range; neither FT3 nor FT4 differed between groups, making the TSH difference unlikely to be clinically meaningful. Ranges of AMH values were 0.00 to 1.93 ng/mL in the G-CSF group and 0.00 to 1.77 in the control group. Among the 50 G-CSF patients, 10, 50, 4, and 29 respectively had tubal infertility, ovarian dysfunction, endometriosis, and male infertility, as did 9, 49, 4, and 33 of the 49 control patients, showing no significant differences in prevalence of infertility causes (chi-squared test or Fisher’s exact test).Table 1Baseline characteristics of patients completing the study aCharacteristic (unit)G-CSF group ($$n = 50$$)Control group ($$n = 49$$)Age (years)36.6 ± 3.837.5 ± 3.5Infertility duration (years)2.3 ± 2.12.4 ± 3.1Number of previous ART attempts0.3 ± 0.80.2 ± 0.4Gravidity0.8 ± 0.81.0 ± 1.1Parity0.5 ± 0.60.4 ± 0.6Body mass index (kg/m2)20.9 ± 2.321.1 ± 2.8Anti-Müllerian hormone (AMH, ng/mL)0.98 ± 0.540.91 ± 0.49Follicle-stimulating hormone on day 3 (IU/L)9.2 ± 6.48.6 ± 4.7Luteinizing hormone on day 3 (IU/L)4.3 ± 2.74.0 ± 1.8Prolactin (ng/mL)7.3 ± 3.58.8 ± 6.1Estradiol on day 3 (pg/mL)32 ± 1641 ± 28Free testosterone on day 3 (FT, pg/mL) b0.60 ± 0.02, $$n = 240$.60$ ± 0.00, $$n = 25$$Testosterone on day 3 (T, ng/mL) b0.16 ± 0.08, $$n = 260$.17$ ± 0.08, $$n = 24$$Thyroid-stimulating hormone (TSH, μIU/mL)1.88 ± 1.07 c2.46 ± 1.40Free thyronine (pg/mL)2.92 ± 0.322.84 ± 0.43Free thyroxine (FT4, ng/dL)1.23 ± 0.141.23 ± 0.13Fasting plasma glucose (mg/dL)81.9 ± 6.7, $$n = 4982$.9$ ± 6.7, $$n = 47$$*Fasting serum* insulin (μU/mL)4.3 ± 1.9, $$n = 494$.7$ ± 2.0, $$n = 47$$a No significant differences were found between groups, except for TSH (unpaired t test for T, FT4, and AMH; Mann–Whitney U test for other characteristics)b FT and T were measured in the first and second half of this study respectively because production of FT measurement kits was interruptedcP < 0.05 vs. control group, Mann–Whitney U test
## Clinical outcomes
One hundred ART patients were enrolled and randomized to G-CSF or control groups (50 patients each). All G-CSF patients completed the study; 1 control dropped out for unknown reasons, leaving 49 (Fig. 1). No adverse effects of G-CSF were observed. G-CSF and control groups underwent conventional infertility treatments with and without G-CSF priming in the initial and second cycles, resulting in 3 and 0 live deliveries, respectively (Table 2). All 3 patients conceived with spontaneous ovulation after menstruation following the first G-CSF priming. Table 2Clinical outcomesStrategy or outcomeG-CSF groupControl groupPatients completing the study50 patients49 patientsConventional infertility treatments in initial and second cycles with or without G-CSF priming Live deliveries (% per patient)3 a ($6.0\%$)0 ($0\%$)IVF/ICSI and fresh ET Ovarian stimulation (OS)47 patients49 patients No follicular development induced (% per OS)0 ($0\%$)1 ($2.0\%$) Numbers of follicles (≥ 16 mm) on the hCG day4.2 ± 2.93.0 ± 1.7 Serum E2 concentrations (pg/mL) on the hCG day1820 ± 12001350 ± 840 Successful oocyte retrievals (% per OS)47 ($100\%$)48 ($98\%$) No ET for lack of transferrable embryos (% per OS)2 ($4.3\%$)4 ($8.2\%$) Fresh ETs (% per OS)45 ($96\%$)44 ($90\%$) Cryopreservation of redundant blastocysts possible (% per OS)25 ($53\%$) b12 ($24\%$) Clinical pregnancies (% per OS)14 ($30\%$) c6 ($12\%$) Ongoing pregnancies (% per OS)12 ($26\%$) d5 ($10\%$) Live deliveries (% per OS)9 ($19\%$)5 ($10\%$)Cryopreserved ET21 cycles14 cycles Clinical pregnancies (% per cryopreserved ET)6 ($29\%$)4 ($29\%$) Live deliveries (% per cryopreserved ET)4 ($19\%$)2 ($14\%$)Numbers of cumulative live deliveries (% per patient)16 ($32\%$) e7 ($14\%$)a All 3 patients conceived with spontaneous ovulation following initial G-CSF primingbP < 0.01 vs. control group, chi-squared test; relative risk (RR) = 3.5; $95\%$ confidence interval (CI), 1.5–8.3cP < 0.05 vs. control group, chi-squared test; RR = 3.0; $95\%$ CI, 1.1–8.8dP < 0.05 vs. control group, chi-squared test; RR = 3.0; $95\%$ CI, 1.0–9.4eP < 0.05 vs. control group, chi-squared test; RR = 2.8; $95\%$ CI, 1.04–7.7 Forty-seven G-CSF and forty-nine control patients underwent ovarian stimulation, resulting in successful oocyte retrievals in all G-CSF patients and 48 controls (no follicular growth was induced in one control). No transferable embryos were obtained in 2 G-CSF patients and 4 controls, leaving 45 and 44 fresh ET.
G-CSF and control groups respectively achieved 14 and 6 clinical pregnancies, 12 and 5 ongoing pregnancies, and 9 and 5 live deliveries. Rates of clinical and ongoing pregnancy per stimulated patient were significantly higher in the G-CSF group ($30\%$ and $26\%$) than in controls ($12\%$ and $10\%$, Table 2). Numbers of transferred embryos did not differ significantly between G-CSF and control groups (2.0 ± 0.6 and 1.9 ± 0.4 respectively, $$P \leq 0.31$$, Mann–Whitney U test). Moreover, significantly more G-CSF patients achieved cryopreservation of redundant blastocysts than controls.
Subsequently, 21 and 14 cycles of cryopreserved ET were carried out in the G-CSF and control groups, resulting in 6 and 4 clinical pregnancies with 4 and 2 live deliveries. Rates of clinical pregnancy per cryopreserved ET were similar between G-CSF and control groups ($29\%$ and $29\%$, $$P \leq 1.00$$, Fisher’s exact test), as were numbers of transferred embryos per cryopreserved ET (1.9 ± 0.4 and 1.9 ± 0.3, $$P \leq 0.52$$, Mann–Whitney U test).
Live delivery rates tended to be higher in the G-CSF group than in controls for any conventional infertility treatment, IVF/ICSI and fresh ET, and cryopreserved ET ($6\%$ vs. $0\%$, $19\%$ vs. $10\%$, and $19\%$ vs. $14\%$, respectively), although statistical significance was not attained ($$P \leq 0.24$$, Fisher’s exact test; $$P \leq 0.21$$, chi-squared test; and $$P \leq 1.00$$, Fisher’s exact test). However, the rate of cumulative live delivery per patient was $32\%$ in the G-CSF group, significantly higher than $14\%$ in controls ($P \leq 0.05$, RR 2.8, $95\%$ CI 1.04–7.7, chi-squared test; Table 2). Miscarriage rates among all clinical pregnancies were similar between the G-CSF and control groups (respectively $30\%$ [$\frac{7}{23}$] and $30\%$ [$\frac{3}{10}$]).
Associations of 4 major fertility-related factors (age, day-3 FSH, AMH, and G-CSF priming) with achievement of cumulative live delivery were analyzed by logistic regression analysis. A backward stepwise method based on the likelihood ratio test was used for selection of variables. Only age and G-CSF priming significantly correlated with cumulative live delivery ($P \leq 0.05$, odds ratio 0.86, $95\%$ CI 0.75–0.99; and $P \leq 0.05$, odds ratio 2.9, $95\%$ CI 1.0–8.20).
Sixteen G-CSF and seven control patients respectively delivered 20 (10 male, 10 female) and 10 (6 male, 4 female) normal live infants, including 2 and 3 sets of twins and 1 and 0 set of triplets. Considering the 13 and 4 singleton newborns in G-CSF and control groups, no significant difference was evident in body weight (3042 ± 374 vs. 2838 ± 648 g; unpaired t test), gestational age at delivery (38.5 ± 1.6 vs. 37.3 ± 3.1 weeks; unpaired t test) or Apgar scores at 1 and 5 min (8.4 ± 0.9 vs. 8.0 ± 2.0 and 9.3 ± 0.9 vs. 9.3 ± 1.0; Mann–Whitney U test).
## Follicular growth, fertilization and embryonic development
On the day of hCG administration, number of follicles larger than 16 mm and serum E2 concentrations tended to be higher in 47 G-CSF patients than in 48 controls, although statistical significance was not attained (4.2 ± 2.9 vs. 3.0 ± 1.7, $$P \leq 0.06$$ and 1820 ± 1200 pg/mL vs. 1350 ± 840, $$P \leq 0.06$$, respectively; Mann–Whitney U test; Table 2). No significant differences (Mann–Whitney U test) were evident in endometrial thickness (11.4 ± 2.2 mm vs. 11.2 ± 2.6) or total amounts of hMG administered (2800 ± 660 IU vs. 2700 ± 810).
For all patients’retrieved oocytes, developmental outcomes were monitored during the first 2 days of culture. Comparing 47 G-CSF and 48 control patients with successfully retrieved oocytes, numbers of retrieved oocytes did not differ significantly but numbers of fertilized oocytes and day-2 embryos per retrieval were significantly higher in the G-CSF group than in controls (fertilized oocytes, 5.7 ± 3.7 vs. 4.2 ± 2.8; day-2 embryos, 5.3 ± 4.1 vs. 3.7 ± 2.9; Fig. 2A).Fig. 2Numbers of retrieved oocytes, fertilized oocytes, and day-2 embryos per successful oocyte retrieval (A), development to day-5 embryos (B), and oocyte developmental competence (C) were compared between G-CSF and control groups. Significantly more fertilized oocytes and day-2 embryos, a higher rate of blastocyst acquisition, and higher embryo quality were obtained in the G-CSF group. Implantation rate per transferred embryo was defined as (number of gestational sacs / number of transferred embryos) × $100\%$. Serum AMH significantly increased after G-CSF priming; in controls AMH decreased, resulting in higher final concentrations of AMH in the G-CSF group (D). G-CSF, granulocyte colony-stimulating factor; AMH, anti-Müllerian hormone Embryo transfers were carried out at 2, 3, and 5 days after oocyte retrieval in 7, 22, and 18 G-CSF patients and in 9, 27, and 12 control patients, respectively. Distributions of ET days did not differ significantly between groups (chi-squared test). Oocytes from significantly more G-CSF patients proved suitable for embryo culture until 5 days after oocyte retrieval than did oocytes from control patients ($57\%$ [$\frac{27}{47}$] vs. $29\%$ [$\frac{14}{48}$]; Fig. 2B). The rate of blastocyst acquisition per successful oocyte retrieval was significantly higher in the G-CSF group than controls ($53\%$ [$\frac{25}{47}$] vs. $27\%$ [$\frac{13}{48}$]; $P \leq 0.01$, RR 3.1, $95\%$ CI 1.3–7.2, chi-squared test; Fig. 2B). The frequency of retrieved oocytes developing into blastocysts was also significantly higher for G-CSF patients’ than control patients’ oocytes ($21.1\%$ vs. $14.5\%$, Fig. 2B). Consequently, significantly more G-CSF patients achieved cryopreservation of redundant blastocysts than controls (Table 2).
Developmental potentials of oocytes and embryos were significantly greater in the G-CSF group (Fig. 2C). Oocytes were significantly likelier to develop into fertilized oocytes or day-2 embryos in the G-CSF group than in controls. The G-CSF group also showed a significantly higher rate of fertilized oocytes developing into day-2 embryos. Implantation rate per transferred fresh embryo also was significantly higher in the G-CSF group ($21\%$ vs. $9.5\%$, $P \leq 0.05$, RR 2.5, $95\%$ CI 1.03–6.09; chi-squared test; Fig. 2C).
Serum AMH concentrations significantly increased after G-CSF administration but significantly decreased in the same interval for controls (Fig. 2D). Consequently, the latter concentrations of serum AMH were significantly higher in the G-CSF group than in controls (Fig. 2D).
## Genotypes for killer-cell immunoglobulin-like receptor (KIR)
Fisher’s exact test identified no significant differences in frequency of individual KIR genes between patients with and without clinical pregnancies in the G-CSF group (Table 3). Among 2DL5, 2DS1, 2DS5, and 3DS1, none correlated significantly with achievement of clinical pregnancy by G-CSF priming (logistic regression analysis).Table 3Frequency of individual KIR genes in patients with and without clinical pregnancies in the G-CSF groupKIR genesWith clinical pregnancies (18 patients)No. of patients with each gene (%) aWithout clinical pregnancies (26 patients)No. of patients with each gene (%) a2DL118 ($100\%$)26 ($100\%$)2DL23 ($17\%$)6 ($23\%$)2DL318 ($100\%$)26 ($100\%$)2DL418 ($100\%$)26 ($100\%$)2DL55 ($28\%$)5 ($19\%$)2DP118 ($100\%$)26 ($100\%$)2DS15 ($28\%$)4 ($15\%$)2DS23 ($17\%$)6 ($23\%$)2DS32 ($11\%$)2 ($8\%$)2DS417 ($94\%$)25 ($96\%$)2DS54 ($22\%$)3 ($12\%$)3DL117 ($94\%$)25 ($96\%$)3DL218 ($100\%$)26 ($100\%$)3DL318 ($100\%$)26 ($100\%$)3DP118 ($100\%$)26 ($100\%$)3DS14 ($22\%$)4 ($15\%$)a No significant differences in frequency of individual KIR genes were evident between patients with and without clinical pregnancies in the G-CSF group (Fisher’s exact test)
## Discussion
This study suggests a novel, simple, and safe treatment for poor ovarian reserve. In such patients, G-CSF priming in 2 consecutive cycles preceding ART significantly improved fertilization and embryonic development attained by ART, increasing clinical and ongoing pregnancy rates following fresh ET. The cumulative live birth rate was significantly higher in the G-CSF group than in controls. G-CSF priming also significantly increased serum AMH, suggesting enhancement of preantral follicle growth. As G-CSF priming improved oocyte developmental competence without significantly increasing numbers of growing follicles and retrieved oocytes, G-CSF appeared to improve preantral follicle growth in terms of quality rather than quantity. This mechanism clearly differs from those previously suggested for improvement of implantation by G-CSF. We observed no effects of G-CSF priming on miscarriage rates or any association of G-CSF efficacy with KIR genotype.
A variety of clinical effects of G-CSF have been reported. Administration of G-CSF accompanying ET was found to increase implantation rates and clinical pregnancy in ART patients with repeated implantation failure or endometrial thinning [10–14]; such effects have remained uncertain in unselected ART patients [13, 14]. G-CSF also reduced miscarriage rate and increased live birth rate in women with unexplained recurrent miscarriages when its administration was initiated within the implantation window [15], but not when begun following a positive urine pregnancy test [16]. In the absence of 3 activating KIR genes detected particularly frequently in women with unexplained recurrent miscarriage (i.e., lack of 2DS1, 2DS5, and 3DS1) [17], G-CSF has shown high effectiveness in overcoming repeated implantation failure [1]. Intrauterine administration of G-CSF was found to increase endometrial thickness in women with endometrial thinning [2, 18, 19]. Considering such observations, G-CSF administration in the early- and mid-luteal phase may improve endometrial receptivity by immunologic interactions and endometrial growth promotion. G-CSF also can alleviate some forms of ovarian dysfunction; during clomiphene and hCG therapy for infertile patients with luteinized unruptured follicle syndrome, G-CSF administration in the late follicular phase has been found to decrease such follicles [20].
An important difference in our therapeutic use of G-CSF from other reports involves the timing of the result. Previously reported effects of G-CSF occurred promptly, affecting the cycle in which G-CSF was administered. In contrast, we found that G-CSF priming showed novel delayed effects on embryonic development and pregnancy rate in a subsequent cycle. The significant increase in serum AMH and improvement of follicular development in our G-CSF group suggest preantral follicle growth enhancement as an underlying mechanism.
In animal studies, G-CSF attenuated ovarian follicular degeneration and decrements of serum AMH in rats with experimental diabetes [3]. G-CSF also increased numbers of primordial, primary, secondary, and tertiary ovarian follicles in female rats treated with cisplatin [4]. In male mice with acute myeloid leukemia administered chemotherapeutic agents, impaired spermatogenesis and fertility were restored by G-CSF administration [21]. In other experiments, G-CSF administration counteracted apoptosis [22–24], inflammatory states, [3, 21, 24], impaired vascularity [24, 25], growth failure [24, 26], and oxidative stress [3, 24]. Such restoration of a physiologic state might suggest mechanisms applicable to enhancement of preantral follicular growth in our patients with poor ovarian reserve. An autocrine or paracrine role of G-CSF in folliculogenesis might be involved, considering that embryos derived from follicles with higher G-CSF were reported to implant more readily [27].
G-CSF also promotes egress of bone marrow stem cells (BMSC) into peripheral blood [28], potentially aiding tissue regeneration, considering that ovarian transplantation of autologous BMSC collected by apheresis after administration of G-CSF for 5 days was found to improve follicle and oocyte quantity to enable pregnancy in poor ART responders [29]. On the other hand, human plasma derived from apheresis after daily administration of G-CSF for 5 days, which was enriched in BMSC-secreted factors, also improved follicular development and fertility in a mouse model of chemotherapy-induced ovarian damage [30]. Further, through a paracrine action of G-CSF in granulosa cells, human umbilical cord mesenchymal stem cell-derived conditioned medium (hUCMSC-CM) reduced granulosa cell apoptosis and depletion of primordial follicles in cisplatin-treated mice [31]. Thus, an indirect mechanism involving BMSC-secreted factors rather than transdifferentiation of BMSC might be possible.
Clinical safety and tolerance of G-CSF treatment have been established in healthy bone marrow donors treated for 3 to 5 days [26], patients with severe chronic neutropenia treated daily or on alternate days for up to 12 years [32], and patients undergoing ischemic stroke treatment involving G-CSF [24]. In healthy bone marrow donors and patients with repeated implantation failure, unexplained repeated miscarriage, chemotherapy, or severe chronic neutropenia, administration of G-CSF during pregnancy (daily to every 3 days for 1 to 3 trimesters) has shown absence of major maternal or fetal/neonatal adverse effects, including teratogenicity [14–16, 26, 32, 33]. In our study, G-CSF priming had no adverse events in our subjects or their fetuses. All infants born to subjects receiving G-CSF were free of congenital anomalies and had weights similar to those born to controls.
Platelet-rich plasma (PRP) injection into human ovaries has been reported to improve ovarian reserve markers and clinical pregnancy rates [34]. However, ovarian PRP injection with ultrasonographic guidance requires sedation, while subcutaneous G-CSF administration is less invasive, safer, and easier, especially for repeated treatment. We previously administered G-CSF to a 45-year-old perimenopausal ART patient with severely diminished ovarian reserve almost monthly for 3 years (about 31 times), restoring ovulatory cycles without adverse effects. We tentatively chose once-per-cycle administration of G-CSF for our study based upon the serendipitous clinical experience described in the Introduction. However, more frequent administration such as every several days might increase efficacy further; with a single subcutaneous G-CSF administration (100 μg of lenograstim), serum G-CSF concentrations increased from 10.8 ± 4.2 pg/mL on day 0 to 69.6 ± 35.0 on day 1 but fell to 20.9 ± 13.9 on day 6 [20]. In a normal menstrual cycle, serum G-CSF concentrations are lowest in the follicular phase, higher in the luteal phase, and highest in the intervening ovulatory phase [35]. Optimal timing of G-CSF administration during an ovarian cycle remains to be determined.
## Conclusions
In patients with poor ovarian reserve, G-CSF priming in 2 consecutive cycles preceding ART significantly improved fertilization and embryonic development in ART therapy. Consequently, rates of implantation and clinical and ongoing pregnancy by fresh ET were significantly increased. The cumulative live birth rate was significantly higher in the G-CSF group than controls. G-CSF priming also significantly increased serum AMH, consistent with enhancement of preantral follicle growth–a mechanism differing from those previously suggested for implantation improvement by G-CSF. We observed no effects of G-CSF priming on miscarriage rates or any association of its efficacy with KIR genotypes. G-CSF priming showed no adverse events in our subjects or their fetuses. All infants born to subjects receiving G-CSF were free of congenital anomalies and had weights, birth weeks, and Apgar scores similar to those born to controls. This study proposes a novel, simple, and safe treatment for poor ovarian reserve.
## References
1. Würfel W, Santjohanser C, Hirv K, Bühl M, Meri O, Laubert I. **High pregnancy rates with administration of granulocyte colony-stimulating factor in ART-patients with repetitive implantation failure and lacking killer-cell immunglobulin-like receptors**. *Hum Reprod* (2010) **25** 2151-2153. DOI: 10.1093/humrep/deq106
2. Gleicher N, Vidali A, Barad DH. **Successful treatment of unresponsive thin endometrium**. *Fertil Steril* (2011) **95** 2123.e13-e17. DOI: 10.1016/j.fertnstert.2011.01.143
3. Pala HG, Pala EE, Ulkumen BA, Aktug H, Yavasoglu A, Korkmaz HA. **The protective effect of granulocyte colony-stimulating factor on endometrium and ovary in a rat model of diabetes mellitus**. *Gynecol Obstet Invest* (2014) **78** 94-100. DOI: 10.1159/000363239
4. Akdemir A, Zeybek B, Akman L, Ergenoglu AM, Yeniel AO, Erbas O. **Granulocyte-colony stimulating factor decreases the extent of ovarian damage caused by cisplatin in an experimental rat model**. *J Gynecol Oncol* (2014) **25** 328-333. DOI: 10.3802/jgo.2014.25.4.328
5. Seifer DB, Baker VL, Leader B. **Age-specific serum anti-Müllerian hormone values for 17,120 women presenting to fertility centers within the United States**. *Fertil Steril* (2011) **95** 747-750. DOI: 10.1016/j.fertnstert.2010.10.011
6. Segawa T, Omi K, Watanabe Y, Sone Y, Handa M, Kuroda M. **Age-specific values of access anti-Müllerian hormone immunoassay carried out on Japanese patients with infertility: a retrospective large-scale study**. *BMC Women’s Health* (2019) **19** 57. DOI: 10.1186/s12905-019-0752-z
7. Jinno M, Nagai R, Takeuchi M, Watanabe A, Teruya K, Sugawa H. *Reprod Biol Endocrinol* (2021) **19** 149. DOI: 10.1186/s12958-021-00832-y
8. Jinno M, Takeuchi M, Watanabe A, Teruya K, Hirohama J, Eguchi N. **Advanced glycation end-products accumulation compromises embryonic development and achievement of pregnancy by assisted reproductive technology**. *Hum Reprod* (2011) **26** 604-610. DOI: 10.1093/humrep/deq388
9. Cisneros E, Moraru M, Gómez-Lozano N, Muntasell A, López-Botet M, Vilches C. **Haplotype-based analysis of KIR-gene profiles in a South European population—distribution of standard and variant haplotypes, and identification of novel recombinant structures**. *Front Immunol* (2020) **11** 440. DOI: 10.3389/fimmu.2020.00440
10. Würfel W. **Approaches to better implantation**. *J Ass Reprod Genet* (2000) **17** 473
11. Aleyasin A, Abediasl Z, Nazari A, Sheikh M. **Granulocyte colony-stimulating factor in repeated IVF failure, a randomized trial**. *Reproduction* (2016) **151** 637-642. DOI: 10.1530/REP-16-0046
12. Kamath MS, Chittawar PB, Kirubakaran R, Mascarenhas M. **Use of granulocyte-colony stimulating factor in assisted reproductive technology: a systematic review and meta-analysis**. *European J Obstet Gynecol Reprod Biol* (2017) **214** 16-24. DOI: 10.1016/j.ejogrb.2017.04.022
13. Zhang L, Xu W, Fu X, Huang Q, Guo X, Zhang L. **Therapeutic role of granulocyte colony-stimulating factor (G-CSF) for infertile women under in vitro fertilization and embryo transfer (IVF-ET) treatment: a meta-analysis**. *Arch Gynecol Obstet* (2018) **298** 861-871. DOI: 10.1007/s00404-018-4892-4
14. Kamath MS, Kirubakaran R, Sunkara SK. **Granulocyte-colony stimulating factor administration for subfertile women undergoing assisted reproduction**. *Cochrane Database Syst Rev* (2020) **1** CD013226. DOI: 10.1002/14651858.CD013226.pub2
15. Scarpellini F, Sbracia M. **Use of granulocyte colony-stimulating factor for the treatment of unexplained recurrent miscarriage: a randomised controlled trial**. *Hum Reprod* (2009) **24** 2703-2708. DOI: 10.1093/humrep/dep240
16. Eapen A, Joing M, Kwon P, Tong J, Maneta E, De Santo C. **Recombinant human granulocyte-colony stimulating factor in women with unexplained recurrent pregnancy losses: a randomized clinical trial**. *Hum Reprod* (2019) **34** 424-32. DOI: 10.1093/humrep/dey393
17. Hiby SE, Regan L, Lo W, Farrell L, Carrington M, Moffett A. **Association of maternal killer-cell immunoglobulin-like receptors and parental HLA-C genotypes with recurrent miscarriage**. *Hum Reprod* (2008) **23** 972-976. DOI: 10.1093/humrep/den011
18. Gleicher N, Kim A, Michaeli T, Lee H, Shohat-Tal A, Lazzaroni E. **A pilot cohort study of granulocyte colony-stimulating factor in the treatment of unresponsive thin endometrium resistant to standard therapies**. *Hum Reprod* (2013) **28** 172-177. DOI: 10.1093/humrep/des370
19. Xie Y, Zhang T, Tian Z, Zhang J, Wang W, Zhang H. **Efficacy of intrauterine perfusion of granulocyte colony-stimulating factor (G-CSF) for infertile women with thin endometrium: a systematic review and meta-analysis**. *Am J Reprod Immunol* (2017) **78** e12701. DOI: 10.1111/aji.12701
20. Shibata T, Makinoda S, Waseda T, Tomizawa H, Fujii R, Utsunomiya T. **Granulocyte colony-stimulating factor as a potential inducer of ovulation in infertile women with luteinized unruptured follicle syndrome**. *Transl Res* (2016) **171** 63-70. DOI: 10.1016/j.trsl.2015.10.003
21. Michailov Y, AbuMadighem A, Lunenfeld E, Kapelushnik J, Huleihel M. **Granulocyte colony-stimulating factor restored impaired spermatogenesis and fertility in an AML-chemotherapy mice model**. *Int J Mol Sci* (2021) **22** 11157. DOI: 10.3390/ijms222011157
22. Kojima H, Otani A, Oishi A, Makiyama Y, Nakagawa S, Yoshimura N. **Granulocyte colony-stimulating factor attenuates oxidative stress-induced apoptosis in vascular endothelial cells and exhibits functional and morphologic protective effect in oxygen-induced retinopathy**. *Blood* (2011) **117** 1091-1100. DOI: 10.1182/blood-2010-05-286963
23. Harada M, Qin Y, Takano H, Minamino T, Zou Y, Toko H. **G-CSF prevents cardiac remodeling after myocardial infarction by activating the Jak-Stat pathway in cardiomyocytes**. *Nat Med* (2005) **11** 305-311. DOI: 10.1038/nm1199
24. Patel AMR, Apaijai N, Chattipakorn N, Chattipakorn SC. **The protective and reparative role of colony-stimulating factors in the brain with cerebral ischemia/reperfusion injury**. *Neuroendocrinology* (2021) **111** 1029-1065. DOI: 10.1159/000512367
25. Lee ST, Chu K, Jung KH, Ko SY, Kim EH, Shinn DI. **Granulocyte colony-stimulating factor enhances angiogenesis after focal cerebral ischemia**. *Brain Res* (2005) **1058** 120-128. DOI: 10.1016/j.brainres.2005.07.076
26. Würfel W. **Treatment with granulocyte colony-stimulating factor in patients with repetitive implantation failures and/or recurrent spontaneous abortions**. *J Reprod Immunol* (2015) **108** 123-135. DOI: 10.1016/j.jri.2015.01.010
27. Lédée N, Gridelet V, Ravet S, Jouan C, Gaspard O, Wenders F. **Impact of follicular G-CSF quantification on subsequent embryo transfer decisions: a proof of concept study**. *Hum Reprod* (2013) **28** 406-413. DOI: 10.1093/humrep/des354
28. de Kruijf EFM, Fibbe WE, van Pel M. **Cytokine-induced hematopoietic stem and progenitor cell mobilization: unraveling interactions between stem cells and their niche**. *Ann N Y Acad Sci* (2020) **1466** 24-38. DOI: 10.1111/nyas.14059
29. Herraiz S, Romeu M, Buigues A, Martinez S, Diaz-Garcia C, Gómez-Segui I. **Autologous stem cell ovarian transplantation to increase reproductive potential in patients who are poor responders**. *Fertil Steril* (2018) **110** 496-505. DOI: 10.1016/j.fertnstert.2018.04.025
30. Buigues A, Marchante M, de Miguel-Gómez L, Martinez J, Cervelló I, Pellicer A. **Stem cell-secreted factor therapy regenerates the ovarian niche and rescues follicles**. *Am J Obstet Gynecol* (2021) **225** e1-14. DOI: 10.1016/j.ajog.2021.01.023
31. Hong L, Yan L, Xin Z, Hao J, Liu W, Wang S. **Protective effects of human umbilical cord mesenchymal stem cell-derived conditioned medium on ovarian damage**. *J Molecular Cell Biology* (2020) **12** 372-385. DOI: 10.1093/jmcb/mjz105
32. Dale DC, Cottle TE, Fier CJ, Bolyard AA, Bonilla MA, Boxer LA. **Severe chronic neutropenia: treatment and follow-up of patients in the Severe Chronic Neutropenia International Registry**. *Am J Hemtol* (2003) **72** 82-93. DOI: 10.1002/ajh.10255
33. Cruz M, Alecsandru D, García-Valasco JA, Requena A. **Use of granulocyte colony-stimulating factor in ART treatment does not increase the risk of adverse perinatal outcomes**. *RBMO* (2019) **39** 976-980. DOI: 10.1016/j.rbmo.2019.09.0081472-6483
34. Melo P, Navarro C, Jones C, Coward K, Coleman L. **The use of autologous platelet-rich plasma (PRP) versus no intervention in women with low ovarian reserve undergoing fertility treatment: a non-randomized interventional study**. *J Assist Reprod Genet* (2020) **37** 855-863. DOI: 10.1007/s10815-020-01710-z
35. Makinoda S, Mikuni M, Furuta I, Okuyama K, Sagawa T, Fujimoto S. **Serum concentration of endogeneous G-CSF in women during the menstrual cycle and pregnancy**. *Eur J Clin Invest* (1995) **25** 877-879. DOI: 10.1111/j.1365-2362.1995.tb01699.x
|
---
title: Association between raised blood pressure and elevated serum liver enzymes
among active-duty Royal Thai Army personnel in Thailand
authors:
- Boonsub Sakboonyarat
- Jaturon Poovieng
- Sethapong Lertsakulbunlue
- Kanlaya Jongcherdchootrakul
- Phutsapong Srisawat
- Mathirut Mungthin
- Ram Rangsin
journal: BMC Cardiovascular Disorders
year: 2023
pmcid: PMC10029162
doi: 10.1186/s12872-023-03181-3
license: CC BY 4.0
---
# Association between raised blood pressure and elevated serum liver enzymes among active-duty Royal Thai Army personnel in Thailand
## Abstract
### Background
The relationship between hypertension (HT) and serum liver enzymes was reported in a few studies, but the findings were inconsistent. Therefore, the present study aimed to identify the association between elevated serum liver enzymes and raised BP through the use of a large sample of Royal Thai Army (RTA) personnel.
### Methods
The dataset obtained from the annual health examination database of RTA personnel in Thailand was utilized. A total of 244,281 RTA personnel aged 35–60 were included in the current study. Elevated serum liver enzymes were defined as aspartate aminotransferase (AST) or alanine aminotransferase (ALT) ≥ 40 U/L in males and ≥ 35 U/L in females. HT was defined as systolic BP ≥ 140 or diastolic BP ≥ 90 mmHg. A multivariable linear regression model was used to estimate the coefficient and $95\%$ confidence intervals (CI), whereas a multivariable logistic regression model was applied to estimate adjusted odds ratios (AORs) and $95\%$ CI for the association between raised BP and serum liver enzymes.
### Results
Compared to individuals with SBP < 120 and DBP < 80 mmHg, the β coefficients of log-transformed AST and ALT were 0.13 ($95\%$ CI: 0.12–0.13) and 0.11 ($95\%$ CI: 0.11–0.12) in males with HT. Meanwhile, the β coefficients of log-transformed AST and ALT were 0.03 ($95\%$ CI: 0.02–0.04) and 0.07 ($95\%$ CI: 0.05–0.08) in females with HT. In males, HT was associated with elevated AST (AOR: 1.92; $95\%$ CI: 1.85–2.01) and elevated ALT (AOR: 1.43; $95\%$ CI: 1.38–1.48). On the other hand, in females, HT was associated with elevated AST (AOR: 1.42; $95\%$ CI: 1.21–1.66) and elevated ALT (AOR: 1.38; $95\%$ CI: 1.21–1.57).
### Conclusion
Raised BP was positively correlated with elevated AST and ALT in active-duty RTA personnel. Moreover, HT was independently attributed to higher odds of elevated AST and ALT in comparison to optimal BP in both males and females. Furthermore, the relationship between serum liver enzymes and BP was modified by sex.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12872-023-03181-3.
## Background
High blood pressure (BP) is a major cause of cardiovascular diseases (CVD) affecting more than $30\%$ of adults worldwide [1]. Similarly, in Thailand, the National Health Examination Survey VI in 2019 demonstrated that $25\%$ of Thai adults aged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 15 years suffer from hypertension (HT) [2]. Raised BP is a leading cause of end-organ damage, including ischemic heart disease, stroke, and chronic kidney disease [3–7]. Furthermore, a few studies reported the relationship between HT and liver dysfunction [8, 9].
The liver is a vital organ that plays essential roles, including biomolecules' synthesis, storage, degradation, and transformation [10, 11]. The liver enzymes, consisting of aspartate aminotransferase (AST) and alanine aminotransferase (ALT), were suggested to have substantial clinical and convenient surrogate markers that reflect excess fat deposition in the liver and nonalcoholic fatty liver disease (NAFLD) and other related dysfunctions [12–14]. As mentioned above, recently, a few studies reported the connection between liver enzymes and high BP, which may occur through direct partway as insulin resistance resulting in simple steatosis and nonalcoholic steatohepatitis [10].
However, the association between liver enzymes and elevated BP has been reported in limited studies with small sample sizes, in which the findings were conflicting. For instance, a previous study on Bangladeshi adults indicated that only ALT and gamma-glutamyl transferase (GGT) not AST were related to HT [13]. At the same time, a related study in Iran expressed that after adjusting for the potential confounder, AST, ALT, and GGT were not associated with HT [9]. Nevertheless, the evidence from adults in *Thailand is* yet to be available.
In Thailand, nearly 50,000 active-duty Royal Thai Army (RTA) personnel aged at least 35 years participate in yearly health examinations provided by the RTA Medical Department (RTAMED). Raised BP was still a crucial health issue among this population between 2017 and 2021 [15]. Therefore, we aimed to adopt an extensive database of RTA personnel's physical health examinations from 2017 to 2021 so that we can identify the association between elevated serum liver enzymes and raised BP. Furthermore, sex-specific associations between serum liver enzymes and raised BP were assessed among this study population.
## Study design and subjects
The current study employed the dataset obtained from the annual health examination database of RTA personnel in 2017–2021 from the RTAMED in Bangkok, Thailand [15, 16]. The RTAMED provides annual health examinations for active-duty RTA personnel through 37 RTA hospitals nationwide, the Army Institute of Pathology (AIP), and the Armed Forces Research Institute of Medical Sciences (AFRIMS). We included active-duty RTA personnel aged 35–60 who participated in annual health examinations between 2017 and 2021. In the current study, we intended to evaluate the association between blood pressure (BP) and serum liver enzymes. Therefore, individuals without records of BP and serum liver enzymes, carrying AST or ALT, were excluded.
## Data collection
The RTAMED provides health examinations for RTA personnel yearly through RTA hospitals nationwide, the AIP, and AFRIMS. A self-report using a standardized case report form was conducted during the health examination session in order to obtain characteristics data and behavioral factors, such as age, sex, smoking status, alcohol use, and exercise [15, 16]. Furthermore, the annual health examination database of RTA personnel comprised body weight, height, systolic blood pressure (SBP), and diastolic blood pressure (DBP). A trained operator conducted anthropometric measurements. BP was measured through the use of an automatic blood pressure monitor in the standardized technique following the Thai guidelines on the treatment of HT [17]. Body mass index (BMI) was calculated by weight (in kg) divided by height (in meter-squared) [16]. Mean arterial pressure (MAP) was measured by the following formula: DBP + $\frac{1}{3}$(SBP – DBP) [18]. BP was categorized into four groups regarding Thai guidelines on the treatment of HT as follows [17]: [1] SBP < 120 and DBP < 80 mmHg, [2] SBP 120–129 or DBP 80–84 mmHg, [3] SBP 130–139 or DBP 85–89 mmHg, and [4] SBP ≥ 140 or DBP ≥ 90 mmHg. Laboratory data included AST, ALT, fasting plasma glucose (FPG), triglyceride (TG), and total cholesterol (TC). Elevated serum liver enzymes were defined as AST or ALT \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 40 U/L in males and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 35 U/L in females [9].
## Statistical analysis
All statistical analyses were carried out using StataCorp. 2021, Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC. Descriptive statistics were exploited for calculating the distribution of participants' characteristics. Categorical variables were presented as percentages, while continuous variables were displayed as mean, standard deviation (SD), median, and interquartile range (first quartile and third quartile). In order to assess the association between serum liver enzymes and blood pressure (BP), linear regression analysis was utilized. Due to the distribution of serum liver enzymes, the normality assumption may be violated; therefore, the log transformation was performed for serum liver enzymes to improve normality (Supplementary Fig. 1). The difference in the elevated liver enzyme (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 40 U/L in males and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 35 U/L in females) across baseline characteristics was compared by using the Chi-square test or Student’s t-test as appropriate. Moreover, logistic regression analysis was explored for estimating the odds ratio (ORs) and $95\%$ confidence intervals (CIs) to determine the association between elevated liver enzyme and raised BP. The interaction was also tested to explore whether sex modifies the relationship between serum liver enzyme and BP. In order to adjust the potential confounders, sex-specific multivariable analysis was performed, which was coordinated for age, regions, BMI, smoking status, alcohol use, exercise, fasting plasma glucose, total cholesterol, triglyceride, and years. A two-sided p-value less than 0.05 was considered statistically significant.
Although the data in the present study were collected annually and separately, some individuals may repeatedly participate in the physical health examination, which may violate the dependence observation assumption. Therefore, we also conducted a sensitivity analysis to individually assess the association between raised BP and elevated serum liver enzymes each year.
## Ethics consideration
This study was reviewed and approved by the Institutional Review Board, Royal Thai Army Medical Department, following international guidelines including the Declaration of Helsinki, the Belmont Report, CIOMS Guidelines, and the International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use–Good Clinical Practice (ICH–GCP) (approval number S067h/64 & S056h/65). Due to the use of secondary data, a waiver of documentation of informed consent was utilized. The Institutional Review Board, Royal Thai Army Medical Department, approved an informed consent waiver.
## Characteristics of study participants
Table 1 presents the characteristics of 244,281 active-duty RTA personnel included in the study population between 2017 and 2021. The majority (about $90\%$) were males. The mean age of study participants ranged from 46.7 ± 7.7 years to 48.0 ± 7.1 years. Nearly, two-thirds of study participants were current drinkers, while approximately one-fourth were current smokers. The mean SBP of study participants was 130.5 ± 16.9 mmHg in 2017 and increased continuously to 132.2 ± 17.2 mmHg in 2021. However, the mean DBP ranged from 80.8 ± 11.6 mmHg to 81.4 ± 11.6 mmHg over five years. Mean AST ranged from 29.7 ± 25.9 U/L to 30.8 ± 25.7 U/L between 2017 and 2021, while ALT ranged from 31.8 ± 26.4 U/L to 35.4 ± 27.4 U/L over five years. Table 1Characteristics of study participantsYear20172018201920202021Characteristicsn = 42,617n = 47,868n = 54,196n = 54,133n = 45,467Sex, n (%) Male38,614 (90.6)42,630 (89.1)48,553 (89.6)47,682 (88.1)41,012 (90.2) Female4003 (9.4)5238 (10.9)5643 (10.4)6451 (11.9)4455 (9.8)Age (years) Mean ± SD48.0 ± 7.147.5 ± 7.347.4 ± 7.547.4 ± 7.746.7 ± 7.7 Median (Q1-Q3)49 (42—54)48 (41—54)48 (41—54)47 (40—55)46 (40—54)Regions Bangkok7315 (17.2)9730 (20.3)10,840 (20.0)11,085 (20.5)5544 (12.2) Central15,263 (35.8)18,024 (37.7)19,567 (36.1)20,899 (38.6)18,352 (40.4) Northeast8271 (19.4)7478 (15.6)8945 (16.5)9907 (18.3)7881 (17.3) North9953 (23.4)7432 (15.5)9586 (17.7)6872 (12.7)8650 (19.0) South1815 (4.3)5204 (10.9)5258 (9.7)5370 (9.9)5040 (11.1)Current smokers, n (%)10,132 (24.1)12,618 (26.7)14,155 (26.8)14,851 (28.8)12,838 (28.3)Current alcohol use, n (%)27,318 (64.8)30,063 (63.5)34,861 (64.6)34,787 (67.4)28,733 (63.3)Exercise, n (%)39,168 (91.9)44,675 (93.3)51,086 (94.3)50,834 (93.9)41,482 (91.2)Systolic BP (mmHg) Mean ± SD130.5 ± 16.9130.7 ± 17.0131.0 ± 16.8131.3 ± 16.6132.2 ± 17.2 Median (Q1-Q3)130 (120—140)130 (120—140)130 (120—140)130 (120—140)131 (121—140)Diastolic BP (mmHg) Mean ± SD81.4 ± 11.681.3 ± 11.781.0 ± 11.680.8 ± 11.681.3 ± 11.9 Median (Q1-Q3)80 (73—89)81 (73—89)80 (73—88)80 (73—88)81 (73—89)AST (U/L) Mean ± SD30.8 ± 25.730.0 ± 24.929.7 ± 25.929.9 ± 25.530.0 ± 27.5 Median (Q1-Q3)25 (21—32)25 (20—32)24 (20—31)24 (20—31)25 (20—31)ALT (U/L) Mean ± SD35.4 ± 27.434.7 ± 26.633.9 ± 26.231.8 ± 26.433.3 ± 27.8 Median (Q1-Q3)29 (19—44)28 (19—42)27 (18—41)25 (18—37)26 (18—39)BP Blood pressure, SD Standard deviation, Q1 Quartile 1, and Q3 Quartile 3
## Association between raised blood pressure and elevated serum liver enzymes
Effect modification by sex on the association between serum liver enzymes and BP was observed. Table 2 illustrates a sex-specific multivariable linear regression analysis of aminotransferase and blood pressure. A positive relationship was observed between log-transformed AST and SBP, DBP, and MAP in both males and females, with a p-value < 0.001. In addition, the association between log-transformed ALT and SBP, DBP, and MAP among males and females was also marked, with a p-value < 0.001. Consistently, when BP was further assessed as categories, in comparison with the reference group (SBP < 120 and DBP < 80 mmHg), the coefficients of log-transformed AST and ALT were 0.13 ($95\%$ CI: 0.12–0.13) and 0.11 ($95\%$ CI: 0.11–0.12) in males with SBP ≥ 140 or DBP ≥ 90. Meanwhile, the coefficients of log-transformed AST and ALT were 0.03 ($95\%$ CI: 0.02–0.04) and 0.07 ($95\%$ CI: 0.05–0.08) in females with SBP ≥ 140 or DBP ≥ 90 mmHg. Table 2Univariable and multivariable linear regression analysis for the association between raised blood pressure and serum liver enzymesBlood pressureLog-transformed ASTLog-transformed ALT\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\beta}}$$\end{document}β coefficient$95\%$ CIp-value\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{\beta}}$$\end{document}β coefficient$95\%$ CIp-valueMale SBP (mmHg) ¥§ Crude0.0030.003–0.003 < 0.0010.0040.003–0.004 < 0.001 Adjusteda0.0030.003–0.003 < 0.0010.0020.002–0.002 < 0.001 DBP (mmHg) § Crude0.0050.005–0.005 < 0.0010.0070.007–0.007 < 0.001 Adjusteda0.0040.004–0.005 < 0.0010.0040.004–0.004 < 0.001 MAP (mmHg) ¥§ Crude0.0050.005–0.005 < 0.0010.0060.006–0.006 < 0.001 Adjusteda0.0040.004–0.005 < 0.0010.0040.003–0.004 < 0.001 Blood pressure (mmHg) ¥§ Crude SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–840.030.02–0.03 < 0.0010.060.06–0.07 < 0.001 SBP 130–139 or DBP 85–890.060.05–0.06 < 0.0010.110.10–0.11 < 0.001 SBP ≥ 140 or DBP ≥ 900.140.14–0.14 < 0.0010.190.19–0.19 < 0.001 Adjusteda SBP < 120 and DBP < 80 SBP 120–129 or DBP 80–840.030.02–0.03 < 0.0010.040.03–0.04 < 0.001 SBP 130–139 or DBP 85–890.060.05–0.06 < 0.0010.070.06–0.07 < 0.001 SBP ≥ 140 or DBP ≥ 900.130.12–0.13 < 0.0010.110.11–0.12 < 0.001 Female SBP (mmHg) ¥§ Crude0.0020.002–0.003 < 0.0010.0050.005–0.006 < 0.001 Adjusteda0.0010.001–0.0010.0010.0010.001–0.002 < 0.001 DBP (mmHg)§ Crude0.0030.002–0.003 < 0.0010.0080.007–0.009 < 0.001 Adjusteda0.0010.001–0.0010.0030.0020.001–0.003 < 0.001 MAP (mmHg) ¥§ Crude0.0030.003–0.003 < 0.0010.0080.008–0.009 < 0.001 Adjusteda0.0010.001–0.0010.0010.0020.001–0.003 < 0.001 Blood pressure (mmHg) ¥§ Crude SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–840.050.03–0.05 < 0.0010.120.10–0.13 < 0.001 SBP 130–139 or DBP 85–890.070.06–0.08 < 0.0010.190.17–0.20 < 0.001 SBP ≥ 140 or DBP ≥ 900.120.10–0.12 < 0.0010.260.24–0.28 < 0.001 Adjusteda SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–840.010.001–0.0200.0130.040.02–0.05 < 0.001 SBP 130–139 or DBP 85–890.020.01–0.030.0010.060.04–0.07 < 0.001 SBP ≥ 140 or DBP ≥ 900.030.02–0.04 < 0.0010.070.05–0.08 < 0.001SBP Systolic blood pressure, DBP Diastolic blood pressure, MAP Mean arterial pressure, and CI Confidence intervalaAdjusting for age, regions, body mass index, smoking status, alcohol use, exercise, fasting plasma glucose, total cholesterol, triglyceride, and years§P for interaction < 0.05 (sex as an effect modifier on the association between BP and log-transformed AST)¥P for interaction < 0.05 (sex as an effect modifier on the association between BP and log-transformed ALT) Table 3 presents the association between elevated serum liver enzymes (AST or ALT ≥ 40 U/L in males and ≥ 35 U/L in females) and covariates. A higher percentage of elevated AST and ALT with higher BP was observed in both males and females (Figs. 1 and 2); furthermore, sex is the modifier of the association between raised BP and elevated serum liver enzymes (p for heterogeneity < 0.001). The relationship between raised BP and elevated liver enzymes was analyzed through the use of multivariable logistic regression (Table 4). After adjusting for the potential confounders, the association between BP (SBP, DBP, and MAP) and elevated aminotransferase was noticed. In males, HT (SBP ≥ 140 or DBP ≥ 90 mmHg) was associated with elevated AST (adjusted OR: 1.92; $95\%$ CI: 1.85–2.01) and elevated ALT (adjusted OR: 1.43; $95\%$ CI: 1.38–1.48). In females, HT (SBP ≥ 140 or DBP ≥ 90 mmHg) was associated with elevated AST (adjusted OR: 1.42; $95\%$ CI: 1.21–1.66) and elevated ALT (adjusted OR: 1.38; $95\%$ CI: 1.21–1.57). The results of the sensitivity analysis were presented in Supplementary Tables 1 and 2. The annual sensitivity analysis revealed that the association between raised BP and elevated serum liver enzymes followed the same pattern as the primary analysis. Table 3Relationship between baseline characteristics and elevated aminotransferaseBlood pressureMaleFemaleAST < 40 U/LAST ≥ 40 U/Lp-valueALT < 40 U/LALT ≥ 40 U/Lp-valueAST < 35 U/LAST ≥ 35 U/Lp-valueALT < 35 U/LALT ≥ 35 U/Lp-valueAge (years) Mean ± SD47.4 ± 7.547 ± 7.4 < 0.00147.8 ± 7.546.2 ± 7.3 < 0.00147.3 ± 7.649.4 ± 7.3 < 0.00147.3 ± 7.648.6 ± 7.5 < 0.001Regions < 0.001 < 0.001 < 0.001 < 0.001 Bangkok31,607 (90.6)3269 (9.4)27,810 (79.7)7066 (20.3)9139 (94.8)499 (5.2)8822 (91.5)816 (8.5) Central71,642 (85.1)12,519 (14.9)62,865 (74.7)21,296 (25.3)7300 (91.9)644 (8.1)7105 (89.4)839 (10.6) Northeast32,454 (84.0)6162 (16.0)28,560 (74.0)10,056 (26.0)3609 (93.4)257 (6.6)3457 (89.4)409 (10.6) North33,456 (85.3)5785 (14.7)21,431 (54.6)17,810 (45.4)3037 (93.4)215 (6.6)2162 (66.5)1090 (33.5) South18,736 (86.8)2861 (13.2)15,609 (72.3)5988 (27.7)997 (91.5)93 (8.5)953 (87.4)137 (12.6)Current smokers, n (%) < 0.001 < 0.0010.0500.717 No130,770 (87.4)18,840 (12.6)108,267 (72.4)41,343 (27.6)23,428 (93.4)1652 (6.6)21,878 (87.2)3202 (12.8) Yes53,037 (82.7)11,069 (17.3)44,402 (69.3)19,704 (30.7)445 (91.2)43 (8.8)423 (86.7)65 (13.3)Current alcohol use, n (%) < 0.001 < 0.0010.6730.016 No58,948 (88.7)7481 (11.3)48,337 (72.8)18,092 (27.2)17,122 (93.4)1206 (6.6)15,931 (86.9)2397 (13.1) Yes125,957 (84.8)22,564 (15.2)105,270 (70.9)43,251 (29.1)6754 (93.3)487 (6.7)6375 (88.0)866 (12.0)Exercise, n (%)0.2060.0240.6300.327 No12,797 (85.7)2144 (14.3)10,807 (72.3)4134 (27.7)1951 (93.1)144 (6.9)1842 (87.9)253 (12.1) Yes175,098 (86.0)28,452 (14.0)145,468 (71.5)58,082 (28.5)22,131 (93.4)1564 (6.6)20,657 (87.2)3038 (12.8)Body mass index (kg/m2) Mean ± SD25.3 ± 3.625.5 ± 4.1 < 0.00125.0 ± 3.526.4 ± 3.9 < 0.00124.3 ± 4.125.8 ± 4.7 < 0.00124.2 ± 4.125.9 ± 4.4 < 0.001Fasting plasma glucose (mg/dL) Mean ± SD103.6 ± 36.0109.7 ± 40.9 < 0.001102.8 ± 35.8108.9 ± 39 < 0.00195.7 ± 26.1106.5 ± 37.9 < 0.00195.1 ± 25.4107.3 ± 37.4 < 0.001Total cholesterol (mg/dL) Mean ± SD211.0 ± 48.7211.3 ± 55.50.263209.1 ± 48.2215.9 ± 52.8 < 0.001208.2 ± 43.9209.6 ± 46.50.209207.6 ± 43.7212.9 ± 46.9 < 0.001Triglyceride (mg/dL) Mean ± SD165.7 ± 121.2223.5 ± 198.4 < 0.001158.7 ± 117.6211.7 ± 168.5 < 0.001110.9 ± 65.8144.1 ± 93.5 < 0.001109.5 ± 65.9137.7 ± 79.8 < 0.001Systolic BP (mmHg) Mean ± SD131.4 ± 16.4136.1 ± 18.0 < 0.001131.4 ± 16.5133.9 ± 17.0 < 0.001123.0 ± 16.5127.9 ± 17.1 < 0.001122.9 ± 16.5126.4 ± 16.8 < 0.001Diastolic BP (mmHg) Mean ± SD81.4 ± 11.484.9 ± 12.3 < 0.00181.1 ± 11.483.7 ± 11.9 < 0.00174.8 ± 10.677.5 ± 11.4 < 0.00174.6 ± 10.677.7 ± 11.1 < 0.001MAP (mmHg) Mean ± SD98.1 ± 12.1102 ± 13.3 < 0.00197.9 ± 12.2100.5 ± 12.7 < 0.00190.9 ± 11.694.3 ± 12.2 < 0.00190.7 ± 11.593.9 ± 11.9 < 0.001Blood pressure (mmHg) < 0.001 < 0.001 < 0.001 < 0.001 SBP < 120 and DBP < 8036,063 (90.1)3980 (9.9)30,721 (76.7)9322 (23.3)9682 (95.4)467 (4.6)9201 (90.7)948 (9.3) SBP 120–129 or DBP 80–8441,552 (88.6)5369 (11.4)34,917 (74.4)12,004 (25.6)5672 (93.9)367 (6.1)5269 (87.2)770 (12.8) SBP 130–139 or DBP 85–8949,347 (86.9)7465 (13.1)40,931 (72.0)15,881 (28.0)4852 (91.7)437 (8.3)4474 (84.6)815 (15.4) SBP ≥ 140 or DBP ≥ 9060,933 (81.6)13,782 (18.4)49,706 (66.5)25,009 (33.5)3876 (89.9)437 (10.1)3555 (82.4)758 (17.6)SBP Systolic blood pressure, DBP Diastolic blood pressure, MAP Mean arterial pressure, and SD Standard deviationFig. 1Sex-specific percentage and $95\%$ confidence interval of elevated aspartate aminotransferase (AST) stratified by blood pressure categoryFig. 2Sex-specific percentage and $95\%$ confidence interval of elevated alanine aminotransferase (ALT) stratified by blood pressure categoryTable 4Univariable and multivariable logistic regression analysis for the association between raised blood pressure and elevated aminotransferaseBlood pressureElevated ASTElevated ALTOdds ratio$95\%$ CIp-valueOdds ratio$95\%$ CIp-valueMale SBP (mmHg) ¥ Unadjusted model1.021.02–1.02< 0.0011.011.01–1.01< 0.001 Adjusted modela1.021.01–1.02< 0.0011.011.01–1.01< 0.001 DBP (mmHg) ¥ Unadjusted model1.031.02–1.03 < 0.0011.021.02–1.02 < 0.001 Adjusted modela1.021.02–1.02 < 0.0011.011.01–1.01 < 0.001 MAP (mmHg) ¥ Unadjusted model1.021.02–1.03 < 0.0011.021.02–1.02 < 0.001 Adjusted modela1.021.02–1.02 < 0.0011.011.01–1.01 < 0.001 Blood pressure (mmHg) ¥§ Unadjusted model SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–841.171.12–1.22 < 0.0011.131.10–1.17 < 0.001 SBP 130–139 or DBP 85–891.371.32–1.43 < 0.0011.281.24–1.32 < 0.001 SBP ≥ 140 or DBP ≥ 902.051.97–2.13 < 0.0011.661.61–1.70 < 0.001 Adjusted modela§ SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–841.171.11–1.22 < 0.0011.091.05–1.12 < 0.001 SBP 130–139 or DBP 85–891.351.30–1.42 < 0.0011.221.18–1.26 < 0.001 SBP ≥ 140 or DBP ≥ 901.921.85–2.01 < 0.0011.431.38–1.48 < 0.001Female SBP (mmHg) ¥ Unadjusted model1.021.01–1.02 < 0.0011.011.01–1.01 < 0.001 Adjusted modela1.011.01–1.010.0061.011.00–1.01 < 0.001 DBP (mmHg) ¥ Unadjusted model1.021.02–1.03 < 0.0011.031.02–1.03 < 0.001 Adjusted modela1.011.01–1.010.0011.011.01–1.02 < 0.001 MAP (mmHg) ¥ Unadjusted model1.021.02–1.03 < 0.0011.021.02–1.03 < 0.001 Adjusted modela1.011.01–1.010.0011.011.01–1.01 < 0.001 Blood pressure (mmHg) ¥§ Crude model SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–841.341.17–1.54 < 0.0011.421.28–1.57 < 0.001 SBP 130–139 or DBP 85–891.871.63–2.14 < 0.0011.771.60–1.95 < 0.001 SBP ≥ 140 or DBP ≥ 902.342.04–2.68 < 0.0012.071.87–2.29 < 0.001 *Adjusted modela* SBP < 120 and DBP < 80RefRef SBP 120–129 or DBP 80–841.120.97–1.300.1351.221.09–1.380.001 SBP 130–139 or DBP 85–891.341.16–1.55 < 0.0011.361.20–1.54 < 0.001 SBP ≥ 140 or DBP ≥ 901.421.21–1.66 < 0.0011.381.21–1.57 < 0.001SBP Systolic blood pressure, DBP Diastolic blood pressure, MAP Mean arterial pressure, and CI Confidence intervalaAdjusting for age, regions, body mass index, smoking status, alcohol use, exercise, fasting plasma glucose, total cholesterol, triglyceride, and years§P for interaction < 0.05 (sex as an effect modifier on the assocation between BP and elevated AST)¥P for interaction < 0.05 (sex as an effect modifier on the association between BP and elevated ALT)
## Discussion
The associations between raised blood pressure and elevated serum liver enzymes in active-duty RTA personnel in Thailand were identified using a large database of RTA personnel's physical health examinations. After adjusting for baseline covariates, the associations between raised BP, encompassing SBP, DBP, and MAP, and elevated serum liver enzymes, both AST and ALT, were detected. Moreover, it was also found that the odds of elevated AST and ALT were higher in RTA personnel with HT (SBP ≥ 140 or DBP ≥ 90 mmHg) than those with optimal BP (SBP < 120 and DBP < 80 mmHg) in both males and females. To the best of our knowledge, this is the first and largest study examining the relationship between high BP and elevated serum liver enzymes in the Thai population.
In line with the existing literature, the evidence of the associations between raised BP and elevated serum liver enzyme was incompatible. For instance, a small sample size study on Bangladeshi adults exposed the associations between HT and serum liver enzymes, incorporating ALT and GGT, but not AST and alkaline phosphatase (ALP). Conversely, alcohol use behavior was not included in the final model of the Bangladesh study [13]. At the same time, a related study on mild dyslipidemia participants from Korea illustrated that only GGT was associated with higher SBP and DBP [8]. On the contrary, a recent relatively large study in Iran reported that after adjusting for potential confounders, ALP was interrelated with HT in both males and females, while there were no significant associations of AST, ALT, and GGT with HT [9]. Nevertheless, in the current study, considering the secondary database, there was no chance to evaluate the linkage of ALP and GGT with BP. We reported the independent association of log-transformed AST and ALT with increased BP, comprising SBP, DBP, and MAP, among males and females. Likewise, a recent study in Iran consistently exhibited the positive association of log-transformed AST, ALT, and ALP with increasing BP in both sexes [9].
Moreover, we noticed a dose–response relationship with a relatively precise association between raised BP and elevated AST and ALT. After adjusting for baseline variables, we found that the odds for elevated AST and ALT among male RTA personnel with HT were estimated to be $92\%$ and $43\%$ higher than those with optimal BP. Similarly, among female participants, the odds for elevated AST and ALT in individuals with HT were estimated to be 1.42 and 1.38 times higher than those with optimal BP. In the present study, the existing potential confounders were included in the final model. Nonetheless, there is a possibility that unmeasured confounders, involving the information on antihypertensive drug use, the number of drugs used and their types, and other medications for treating their comorbidities, such as dyslipidemia and diabetes, may have an impact on the results of the study.
HT is well-documented to be associated with metabolic syndrome and hyperinsulinemia, which are the key pathways for developing simple steatosis and fatty liver [10, 19]. The most common laboratory-based test reflecting these abnormalities was the elevations in AST and ALT [10]. Our study revealed that HT was independently connected with elevated AST and ALT, which the related evidence in the animal model [20] and clinical study [21] can explain. The animal model suggested that the renin-angiotensin system (RAS), especially angiotensin II (Ang II), played a vital role in activating hepatic stellate cells for liver fibrosis [20, 22]. Moreover, the related clinical study in China manifested that Ang II level was an independent risk factor for patients with NAFLD [21]. Furthermore, Ang II type 1 receptor antagonists can alleviate this progression [23]. On the other hand, HT and elevated serum liver enzymes may be linked by oxidative stress and reactive oxygen species, which play a crucial role in the pathogenesis of HT and also affect the hepatocyte resulting in hepatocellular injuries [24, 25].
A few studies reported the sex-specific association between blood pressure and serum liver enzymes [9, 13, 26]. However, the formal test for an existing interaction in those studies was limited. The present study also found a significant effect modification between sex and BP on elevated liver enzymes. In comparison with females, males showed a stronger association between raised BP and elevated AST and ALT levels. At the same time, contradictory findings from a different study reported that the association between raised DBP and elevated AST was stronger in females, though the association between raised DBP and elevated ALT was stronger in males [9]. Estradiol has an antioxidant effect in females, which may impact serum liver enzyme levels [27, 28]. Thus, one possible mechanism for sex-specific differences in the junction between raised BP and serum liver enzymes could be the effect of sex hormones [26, 27]. However, this concept requires further investigation.
The current study encountered several limitations. Firstly, this was a cross-sectional study; the causal relationship between exposure and outcome could not be presented. Secondly, concerning the secondary database used, we did not have an opportunity to investigate the relationship between raised BP and other serum liver enzymes, containing GGT and ALP. Thirdly, the information on hepatotoxic drug uses and viral hepatitis infection, possibly affecting serum liver enzymes, was not collected; hence, some unmeasured confounders were not included in the adjusted model. Fourthly, although alcohol use was adjusted in the final model, the information on the amount and frequency of alcohol consumption was limited. Thus, residual confounding on the association between BP and elevated liver enzymes may exist. In addition, the information on waist circumference is limited; thus, central obesity, a feature of metabolic syndrome related to high BP and fatty liver, was not included in the final model. Yet, the present study regulated BMI in the multivariable analysis. Next, because the data in the present study were collected each year separately, some individuals may participate in the physical health examination more than once, which may go against the dependent observation assumption. Notwithstanding, the primary analysis results were not altered by the association between raised BP and elevated serum liver enzymes annually obtained from sensitivity analysis. Finally, the current study was carried out among active-duty RTA personnel and comprised a greater proportion of male participants; however, the results reflected an actual situation in this study population.
In addition, the present study encompassed remarkable strengths, combining a large sample size with an adjustment for potential confounders, so that the independent relationship could be assessed. Therefore, our data provided robust evidence supporting the independent association between raised BP and elevated serum liver enzymes, especially AST and ALT. Our results suggest that monitoring serum liver enzyme, a convenient surrogate marker that reflects excess fat deposition in the liver and other related dysfunctions, should be performed, particularly in individuals with raised BP.
## Conclusion
Raised BP was positively associated with elevated AST and ALT in active-duty RTA personnel. In addition, HT was independently associated with higher odds of elevated AST and ALT in comparison with optimal BP in both males and females. It was found that the relationship between serum liver enzymes and BP was modified by sex. These findings supported the evidence of the relationship between BP and serum liver enzymes.
## Supplementary Information
Additional file 1: Figure S1. The log transformation was performed for serum liver enzymes to improve normality. Table S1. Multivariable logistic regression analysis for association between raised blood pressure and elevated aminotransferase in male participant, by year. Table S2. Multivariable logistic regression analysis for association between raised blood pressure and elevated aminotransferase in female participant, by year.
## References
1. Mills KT, Stefanescu A, He J. **The global epidemiology of hypertension**. *Nat Rev Nephrol* (2020.0) **16** 223-37. DOI: 10.1038/s41581-019-0244-2
2. 2.Wichai Aekplakorn. Thai National Health Examination Survey VI (2019-2020). https://online.fliphtml5.com/bcbgj/znee/#p=187. (2019).
3. Chantkran W, Chaisakul J, Rangsin R, Mungthin M, Sakboonyarat B. **Prevalence of and factors associated with stroke in hypertensive patients in Thailand from 2014 to 2018: A nationwide cross-sectional study**. *Sci Rep* (2021.0) **11** 17614. DOI: 10.1038/s41598-021-96878-4
4. Sakboonyarat B, Rangsin R. **Prevalence and associated factors of ischemic heart disease (IHD) among patients with diabetes mellitus: a nation-wide, cross-sectional survey**. *BMC Cardiovasc Disord* (2018.0) **18** 1-7. DOI: 10.1186/s12872-018-0887-0
5. Viwatrangkul P, Lawanwisut S, Leekhaphan P, Prasart-Intara T, Phiensuparp P, Prakiatpongsa S. **Prevalence and associated factors of electrocardiographic left ventricular hypertrophy in a rural community, central Thailand**. *Sci Rep* (2021.0) **11** 7083. DOI: 10.1038/s41598-021-86693-2
6. Hsu CY, McCulloch CE, Darbinian J, Go AS, Iribarren C. **Elevated blood pressure and risk of end-stage renal disease in subjects without baseline kidney disease**. *Arch Intern Med* (2005.0) **165** 923-8. DOI: 10.1001/archinte.165.8.923
7. Chamnanmont C, Chaiyakham N, Kaensingh S, Kaewput W, Rangsin R, Sakboonyarat B. **Prevalence and associated factors in the decline of renal function among outpatients attending a community hospital, centrAL Thailand**. *J Southeast Asian Med Res* (2021.0) **5** 58-66. DOI: 10.55374/jseamed.v5i2.90
8. Park EO, Bae EJ, Park BH, Chae SW. **The associations between liver enzymes and cardiovascular risk factors in adults with mild dyslipidemia**. *J Clin Med* (2020.0) **9** 1147. DOI: 10.3390/jcm9041147
9. Khalili P, Abdollahpoor S, Ayoobi F, Vakilian A, Hakimi H, Rajabi Z. **Evaluation of relationship between serum liver enzymes and hypertension: a cross-sectional study based on data from rafsanjan cohort study**. *Int J Hypertens.* (2022.0) **2022** 5062622. DOI: 10.1155/2022/5062622
10. Brookes MJ, Cooper BT. **Hypertension and fatty liver: Guilty by association?**. *J Human Hyperten* (2007.0) **21** 264-70. DOI: 10.1038/sj.jhh.1002148
11. Corless JK, Middleton HM. **Normal liver function: A basis for understanding hepatic disease**. *Arch Intern Med* (1983.0) **143** 2291-2294. DOI: 10.1001/archinte.1983.00350120085018
12. Clark JM, Diehl AM. **Nonalcoholic fatty liver disease: an underrecognized cause of cryptogenic cirrhosis**. *JAMA* (2003.0) **289** 3000-4. DOI: 10.1001/jama.289.22.3000
13. Rahman S, Islam S, Haque T, Kathak RR, Ali N. **Association between serum liver enzymes and hypertension: A cross-sectional study in Bangladeshi adults**. *BMC Cardiovasc Disord* (2020.0) **20** 128. DOI: 10.1186/s12872-020-01411-6
14. Clark JM, Brancati FL, Diehl AM. **The prevalence and etiology of elevated aminotransferase levels in the United States**. *Am J Gastroenterol* (2003.0) **98** 960-7. DOI: 10.1111/j.1572-0241.2003.07486.x
15. Sakboonyarat B, Poovieng J, Jongcherdchootrakul K, Srisawat P, Hatthachote P, Mungthin M. **Prevalence of hypertriglyceridemia among Royal Thai Army personnel and its related cardiometabolic risk factors, from 2017 to 2021**. *BMC Public Health* (2022.0) **22** 1569. DOI: 10.1186/s12889-022-13992-2
16. Sakboonyarat B, Poovieng J, Jongcherdchootrakul K, Srisawat P, Hatthachote P, Mungthin M. **Rising trends in obesity prevalence among Royal Thai Army personnel from 2017 to 2021**. *Sci Rep* (2022.0) **12** 7726. DOI: 10.1038/s41598-022-11913-2
17. 17.Thai Hypertension Society. 2019 Thai Guidelines on The Treatment of Hypertension. 1st edition. Chiangmai; 2019.
18. 18.Park EO, Bae EJ, Park BH, Chae SW. The associations between liver enzymes and cardiovascular risk factors in adults with mild dyslipidemia. J Clin Med. 2020;9(4):1147.
19. Sakboonyarat B, Rangsin R, Mittleman MA. **Incidence and risk factors of metabolic syndrome among Royal Thai Army personnel**. *Sci Rep* (2022.0) **12** 1-11. DOI: 10.1038/s41598-022-19024-8
20. Yoshiji H, Kuriyama S, Yoshii J, Ikenaka Y, Noguchi R, Nakatani T. **Angiotensin-II type 1 receptor interaction is a major regulator for liver fibrosis development in rats**. *Hepatology* (2001.0) **34** 4 I. DOI: 10.1053/jhep.2001.28231
21. Li Y, Xiong F, Xu W, Liu S. **Increased Serum Angiotensin II Is a Risk Factor of Nonalcoholic Fatty Liver Disease: A Prospective Pilot Study**. *Gastroenterol Res Pract* (2019.0) **2019** 5647161. DOI: 10.1155/2019/5647161
22. Tsuchida T, Friedman SL. **Mechanisms of hepatic stellate cell activation**. *Nat Rev Gastroenterol Hepatol* (2017.0) **14** 397-411. DOI: 10.1038/nrgastro.2017.38
23. Yokohama S, Yoneda M, Haneda M, Okamoto S, Okada M, Aso K. **Therapeutic efficacy of an angiotensin II receptor antagonist in patients with nonalcoholic steatohepatitis**. *Hepatology* (2004.0) **40** 1222-5. DOI: 10.1002/hep.20420
24. Rodrigo R, González J, Paoletto F. **The role of oxidative stress in the pathophysiology of hypertension**. *Hypertension Res* (2011.0) **34** 431-40. DOI: 10.1038/hr.2010.264
25. 25.Fu Y, Chung FL. Oxidative stress and hepatocarcinogenesis. Hepatoma Res. 2018;4:39.
26. Kunutsor SK, Apekey TA, Cheung BMY. **Gamma-glutamyltransferase and risk of hypertension: A systematic review and dose-responsemeta-analysis of prospective evidence**. *J Hypertension* (2015.0) **33** 2373-81. DOI: 10.1097/HJH.0000000000000763
27. Li S, Tan HY, Wang N, Zhang ZJ, Lao L, Wong CW. **The role of oxidative stress and antioxidants in liver diseases**. *Int J Mol Sci* (2015.0) **16** 26087-124. DOI: 10.3390/ijms161125942
28. Wassmann S, Bäumer AT, Strehlow K, van Eickels M, Grohé C, Ahlbory K. **Endothelial dysfunction and oxidative stress during estrogen deficiency in spontaneously hypertensive rats**. *Circulation* (2001.0) **103** 435-41. DOI: 10.1161/01.CIR.103.3.435
|
---
title: TBX3 is dynamically expressed in pancreatic organogenesis and fine-tunes regeneration
authors:
- Michael Karl Melzer
- Silvia Schirge
- Johann Gout
- Frank Arnold
- Dharini Srinivasan
- Ingo Burtscher
- Chantal Allgöwer
- Medhanie Mulaw
- Friedemann Zengerling
- Cagatay Günes
- Heiko Lickert
- Vincent M. Christoffels
- Stefan Liebau
- Martin Wagner
- Thomas Seufferlein
- Christian Bolenz
- Anne M. Moon
- Lukas Perkhofer
- Alexander Kleger
journal: BMC Biology
year: 2023
pmcid: PMC10029195
doi: 10.1186/s12915-023-01553-x
license: CC BY 4.0
---
# TBX3 is dynamically expressed in pancreatic organogenesis and fine-tunes regeneration
## Abstract
### Background
The reactivation of genetic programs from early development is a common mechanism for injury-induced organ regeneration. T-box 3 (TBX3) is a member of the T-box family of transcription factors previously shown to regulate pluripotency and subsequent lineage commitment in a number of tissues, including limb and lung. TBX3 is also involved in lung and heart organogenesis. Here, we provide a comprehensive and thorough characterization of TBX3 and its role during pancreatic organogenesis and regeneration.
### Results
We interrogated the level and cell specificity of TBX3 in the developing and adult pancreas at mRNA and protein levels at multiple developmental stages in mouse and human pancreas. We employed conditional mutagenesis to determine its role in murine pancreatic development and in regeneration after the induction of acute pancreatitis. We found that Tbx3 is dynamically expressed in the pancreatic mesenchyme and epithelium. While Tbx3 is expressed in the developing pancreas, its absence is likely compensated by other factors after ablation from either the mesenchymal or epithelial compartments. In an adult model of acute pancreatitis, we found that a lack of Tbx3 resulted in increased proliferation and fibrosis as well as an enhanced inflammatory gene programs, indicating that Tbx3 has a role in tissue homeostasis and regeneration.
### Conclusions
TBX3 demonstrates dynamic expression patterns in the pancreas. Although TBX3 is dispensable for proper pancreatic development, its absence leads to altered organ regeneration after induction of acute pancreatitis.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12915-023-01553-x.
## Summary statement
TBX3 shows switching expression patterns during embryonic development in the pancreas and leads to fine-tuning of regeneration from acute pancreatitis via limiting proliferation and fibrosis during regeneration.
## Background
The family of T-box transcription factors includes five subfamilies and comprises 17 genes in both humans and mice. They all share highly conserved “T” DNA-binding domains and are essential during embryonic development and tissue homeostasis [1–4].
TBX3, along with TBX2, TBX4, and TBX5, belongs to the TBX2-family [1, 2]. Heterozygous mutations in TBX3 lead to the malformation of multiple structures resulting in ulnar-mammary syndrome in humans [5]. Accordingly, homozygotes for hypomorphic mutations in Tbx3 in murine embryos and adults resulted in lethal maldevelopment of the heart and cardiac arrhythmias [4–6].
TBX3 has multiple molecular functions [7] and participates in transcription factor networks that guide organ patterning in multiple tissues. TBX3 and TBX2 are closely entangled in other regulatory programs, such as Wnt and Hedgehog signaling, and both are essential for proper limb [8], lung [9], and ureter development [10]. Interestingly, redundancy between Tbx3 and Tbx2 to compensate for each other has been reported in several organs [9, 10].
Finally, TBX3 is not only critically involved in pluripotency maintenance but also in lineage-specific exit from the pluripotency circuitry in mouse embryonic stem cells [11–17]. TBX3 is also a putative inhibitor of pancreatic lineage entry in differentiating human pluripotent stem cells [18]. Interestingly, TBX3 was associated with higher aggressiveness in pancreatic cancer via its ability to (i) promote angiogenesis and (ii) induce cancer stem cell properties [19]. Such stem cell-related features may also become relevant during tissue regeneration after organ injury.
Common mechanisms and similar genetic programs between stem cell induction, organ homeostasis, and regeneration from injury have been identified for the pancreas [20, 21]. Among others, augmented Hedgehog, Notch, and Wnt/b-catenin signaling were delineated as mandatory for regeneration in mice after experimental induction of pancreatitis [21–24]. Interestingly, Tbx3 interacts with certain parts of the Wnt and Hedgehog signaling pathways [8–10].
To investigate the functional roles of Tbx3 during pancreatic development, adult pancreatic homeostasis, and pancreatic regeneration after injury, we here re-analyzed publicly available single-cell transcriptomic data sets and employed a Tbx3Venus reporter system as well as a compartment-specific Tbx3 knockout mouse model and a human knockdown induced pluripotent stem cell (iPSC) differentiation model. While TBX3 is expressed in both mesenchyme and epithelium during embryonic development, its deletion does not significantly alter organogenesis. However, loss of Tbx3 leads to overshoot proliferation of acinar cells, accumulation of fibrosis, and enhanced inflammatory stimuli, Il6-Jak-Stat3, and acinar cell-specific NF-κB signaling during pancreatitis.
## TBX3 is dynamically expressed during pancreatic organogenesis and in adult pancreatic stellate cells
We first performed an RNA-based analysis to examine the differential expression patterns of Tbx3 in mice by re-analyzing published datasets of mouse development (GSE101099) [25] as well as of adult murine pancreata (GSE109774) [26] revealing dynamic expression patterns over distinct stages (Fig. 1a–c). At E12.5 (embryonic day 12.5) and E14.5, Tbx3 was expressed in various mesenchymal cells and neural crest cells, as indicated by cluster assignment using Col3a1 or Tlx2 [27, 28] (Fig. 1a,b, Additional file 1: Fig. S1a,b). Interestingly, while Tbx3 expression levels were retained in a few cells of the mesenchymal compartment, only a limited number of endothelial cells expressed Tbx3 at E17.5 as indicated by Pecam1 positivity (Fig. 1c, Additional file 1: Fig. S1c). Of note, a few acinar precursor cells marked by Cpa1 demonstrated expression of Tbx3. In the adult pancreas, Tbx3 was detected only in few stellate (Col3a1) and endothelial (Pecam1) cells but not exocrine (Cpa1, Krt19) or endocrine (Ins1, Gcg, Sst) cells (Fig. 1d, Additional file 1: Fig. S1d).Fig. 1Tbx3 expression during embryonic development of the pancreas. a–d Re-analysis of single-cell transcriptomic data sets [25, 26] from murine pancreata at indicated timepoints. UMAP cell-cluster representation of the re-analyzed single-cell RNA sequencing data (left panels) and expression patterns of Tbx3 in embryonic pancreata as violin plots (middle panel) and featureplots (right panels) at a E12.5 ($$n = 2$$ mice), b E14.5 ($$n = 3$$ mice), c E17.5 ($$n = 3$$ mice), and d in adult pancreata ($$n = 7$$ mice). UMAP, uniform manifold approximation and projection. Expression levels depicts log-normalized counts To substantiate our analysis across another species, we investigated the expression of human TBX3. First, we chose to re-analyze our recently published bulk transcriptomic datasets (GSE131817) [29, 30] to assess T-box gene expression during in vitro differentiation of human PSCs toward pancreatic progenitor cells (Additional file 1: Fig. S2a,b). While other T-box genes (TBX6, TBX19, EOMES) were highly expressed in human pluripotent stem cells and in definitive endoderm, TBX3 was among the most prominent transcripts at the pancreatic endoderm and pancreatic progenitor stage (Additional file 1: Fig. S2b). As mouse pancreatic progenitors start to arise and specify between E9.5 and E12.5 [31–35], this human analysis enabled us to zoom into earlier time points than those from the murine embryonic pancreata (Fig. 1a–c). To investigate TBX3 expression in the more mature pancreatic lineages [36, 37], we investigated our recently published single-cell transcriptome dataset (GSE162547) monitoring, particularly pancreatic ductal differentiation [37]. Here, we noticed that TBX3 expression was heterogeneously, still robustly expressed throughout the arising lineages, including ductal, endothelial, and endocrine cells, indicative of a widely distributed embryonic expression of TBX3 during human development (Additional file 1: Fig. S2c). Next, we analyzed the time-resolved expression patterns of a set of TBX3 interaction partners [38–40] in our published bulk transcriptomic dataset [29, 30] (Additional file 1: Fig. S2d). Here, we observed a decrease in well-known embryonic factors (e.g., NANOG), while several genes demonstrated higher expression levels over time at the pancreatic progenitor stage (e.g., LEF1).
To further elucidate the cell type-specific expression profiles of these T-box genes in adult pancreata, we performed a single-cell transcriptomic in silico analysis from four published human pancreatic islet donors (GSE84133) [41] (Additional file 1: Fig. S2e,f). In addition to endocrine cells (GCG, INS; SST, PPY), also exocrine cell types (marked by KRT19 and CPA1), stellate (COL3A1), and endothelial (PECAM1) cells were identified (Additional file 1: Fig. S2f). As anticipated from the murine adult pancreata, the expression of TBX3 was restricted to the stellate cells (Additional file 1: Fig. S2e).
Thus, TBX3 is dynamically expressed during the differentiation of pancreatic cells at several developmental stages (PSC-derived pancreatic progenitors, duct-like cells, endothelial-like cells, and endocrine-like cells), and its expression shifts from epithelial cells to stromal cells in the adult human pancreas. Thus, expression in the adult human and murine organisms appears comparable in both species.
## TBX3 in site analysis confirms transcriptionally derived expression patterns
Dynamic Tbx3 expression during pancreatic development prompted us to resolve stage-specific protein expression in the pancreatic anlage during embryonic and postnatal development. To visualize the activity of the murine Tbx3 promoter at the protein level, we used a validated Tbx3-Venus (Tbx3tm1(Venus)Vmc) [11, 42–47] reporter system (Fig. 2a). In this model, the production of Venus protein reflects Tbx3 promoter activity as detected with anti-GFP antibodies [42, 45]. By using a global knockout of Tbx3 through the replacement of the first three exons by a Cre-knockin [48], we confirmed TBX3 antibody specificity (Additional file 1: Fig. S3a). Similarly, the anti-GFP antibody was licensed to capture the Venus protein in Tbx3-positive cells of the seminal vesicle serving as a positive control (Additional file 1: Fig. S3b). We chose multiple developmental stages (E12.5, E15.5, E18.5, adult) to visualize Venus expression. In addition, we investigated the pancreata of P7 (postpartum day 7 after birth) newborn mice. These experiments revealed a good correlation between protein and the scRNA-seq data (Fig. 1). Venus was significantly expressed in the pancreatic mesenchyme (Fig. 2b), a structure known to mediate inductive cues to pattern the pancreatic epithelium during organogenesis [49]. Specifically, at E12.5 Venus signal was detected around epithelial buds as indicated by co-staining for transcription factors NKX6-1, PDX1, SOX9, FOXA2, and basal membrane proteins such as laminin (LAM) (Fig. 2c). At E15.5, the expression pattern of Venus was virtually the same as at E12.5, demonstrated by colocalization with mesenchymal markers vimentin and collagen IV (COL IV), but not the epithelial markers NKX6-1, PDX1, CDH1, and CD49f (Fig. 2d). Of note, Venus protein was detected both in the mesenchymal and epithelial compartment of the pancreas at E18.5 and P7 whereas RNA expression of Tbx3 was more abundant in mesenchymal and endothelial cells at E17.5 (Figs. 1d and 2e-g). However, concordant to the few acinar precursor cells expressing Tbx3 at E17.5 (Fig. 1c), the epithelial expression was mostly in acinar cells as proven by colocalization of Venus with amylase (AMY) and not in endocrine cells labeled by insulin (INS) and glucagon (GCG). Furthermore, Venus was detected in mesenchymal cells (collagen IV +), and also detected in PECAM-1 expressing endothelial cells. To further validate the expression of Venus in acinar (precursor) cells at E18.5, we performed staining of Venus and TBX3 on consecutive sections of E18.5 pancreata (Fig. 2f) confirming the expression of both markers in acinar structures (Zoom-ins in Fig. 2f). Immunohistochemistry analysis further licensed the acinar expression pattern (Additional file 1: Fig. S3c). In adult mice, concordant with the RNA data, Venus was not detected in pancreas-specific epithelial (CDH-1, AMY, CK-19, INS, GCG) cell types (Figs. 1e and 2h). Thus, TBX3 protein expression, as assayed with the Venus reporter, presents specific and developmentally dynamic expression patterns in both the pancreatic epithelium and the mesenchyme. These patterns correlate mostly with the RNA findings, except the weak but robust acinar expression around E18.5 which appears underestimated in scRNA sequencing data potentially attributed to technical limitations for weakly expressed genes [50–54].Fig. 2Venus reporter protein expression confirms RNA expression patterns of Tbx3. a Schematic representation of the Tbx3+/Venus [42, 43] reporter mouse model employed to assess TBX3 expression patterns shown in c–h. Venus (VEN) expression is under the control of the Tbx3 promoter and disrupts Tbx3 expression from the knockin allele. b Schematic illustration depicting VEN expression pattern in pancreas during embryonic development and postnatal growth. c Immunofluorescence staining for NKX6-1/LAM (red/white), PDX1/CDH2 (red/white), SOX9/NKX6-1 (red/white), FOXA2/CDH2 (red/white), and Venus (VEN) (green) on E12.5 pancreata ($$n = 3$$ mice). d Immunofluorescence staining for NKX6-1/COL IV (red/white), PDX1/CD49f (red/white), VIM/CDH1 (red/white), COL IV/CD49f (red/white), and VEN (green) on E15.5 pancreata ($$n = 5$$ mice). e Immunofluorescence staining for pancreatic amylase (AMY)/INS (red/white), AMY/GCG (red/white), GCG/CDH2 (red/white), PECAM-1/CDH1 (red/white), and VEN (green) on E18.5 pancreata ($$n = 4$$ mice). f Immunofluorescence staining for VEN (left panels) or TBX3 (right panels), both in green, in combination with CDH1 (red) on consecutive sections in E18.5 pancreata ($$n = 1$$ mouse). g Immunofluorescence staining for VIM/INS (red/white), AMY/GCG (red/white), AMY/PECAM-1 (red/white), COL IV/PECAM-1 (red/white), and VEN (green) on early postnatal (P7) pancreata ($$n = 4$$ mice). h Immunofluorescence staining for AMY/GCG (red/white), CDH1/CK-19 (red/white), INS/PECAM-1 (red/white), and VEN (green) on adult pancreata ($$n = 3$$ mice). Cells were counterstained with DAPI. Scale bars correspond to 25 μm. n≥2 per group. E, exon
## Compartment-specific deletion of Tbx3 leads to virtually unaltered pancreatic development
To study the functional role of TBX3 in the context of pancreatic organogenesis, two different conditional knockout models targeting Tbx3 in either the pancreatic epithelium or mesenchyme were employed. Embryonic deletion of Tbx3 in the epithelial compartment of the pancreas was achieved by crossing Tbx3flox/+ with Ptf1aCre/+ mice, a model known to target all epithelial lineages of the pancreas, including acinar, ductal, and most endocrine cells starting from E10.5 (Fig. 3a) [35]. Specific recombination in epithelial cells of the pancreas was confirmed by immunohistochemical stainings for tdRFP in Ptf1aCre/+ [35] x LSL-tdRFPKI/KI [55, 56] mice (Additional file 1: Fig. S3d) [56]. Histologic morphology was similar between homozygous knockout mice (Tbx3-KO (epi, epithelial)) and control mice (no knockout of Tbx3) (Fig. 3b). Furthermore, immunofluorescence analysis labeling exocrine pancreatic markers (CK-19, AMY2A; Fig. 3c) revealed virtually no difference between Tbx3-KO (epi) and control mice. This was also valid for the endocrine compartment, where only subtle differences with a trend toward smaller islets in Tbx3-KO (epi) mice was documented while maintaining similar percentages of INS+ beta-cells and GCG+ alpha-cells (Fig. 3d–g).Fig. 3Tbx3 is largely dispensable for pancreatic development. a Schematic representation of the Tbx3fl/fl; Ptf1aCre/+ mouse model allowing Tbx3 deletion in pancreatic epithelial tissue (Tbx3-KO (epi)). b Hematoxylin and eosin (H&E)-stained histological sections of control and Tbx3-KO (epi) pancreata. Immunofluorescence stainings for c CK-19/AMY2A (red/green) and d INS/GCG (green/red) on control and Tbx3-KO (epi) pancreata. Quantification of e number of cells per islet based on INS/GCG co-stainings, f percentage of insulin expressing cells in islets, and g percentage of glucagon-expressing cells in Langerhans islets. $$n = 14$$ control (littermates of Tbx3-KO (epi)) pancreata and $$n = 9$$ Ptf1a-Cre-driven Tbx3-KO (epi) pancreata. h Schematic representation depicting Tbx3fl/fl; Nkx3-2Cre/+ mouse model allowing Tbx3 depletion in pancreatic mesenchyme (Tbx3-KO (mes)). i H&E-stained histological sections of control (littermates of Tbx3-KO (mes)) and Nkx3-2-Cre-driven Tbx3-KO (mes) pancreata. Immunofluorescence stainings for j CK-19/AMY2A (red/green) and k INS/GCG (green/red) on control and Tbx3-KO (mes) pancreata. Quantification of l number of cells per islet area based on INS/GCG co-stainings, m percentage of insulin expressing cells in islets, and g percentage of glucagon-expressing cells in Langerhans islets. $$n = 7$$ control (littermates of Tbx3-KO (mes)) pancreata and $$n = 4$$ Nkx3-2-Cre-driven Tbx3-KO (mes) pancreata. Cells were counterstained with DAPI. Scale bars correspond to 50 μm. Data are expressed as individual datapoints and mean ± SEM. Mann-Whitney test was performed to investigate significance levels (no significance reached) Given the essential role of the pancreatic mesenchyme for proper lineage formation [49, 57], Tbx3 was also deleted in this compartment. The Nkx3-2Cre/+ mouse strain was used (Fig. 3h) as it has been employed to demonstrate the relevance of the pancreatic mesenchyme for proper pancreatic development [49]. As with the epithelial compartment, homozygous deletion of Tbx3 (Tbx3-KO (mes, mesenchymal)) in the mesenchyme did not alter pancreatic morphology (assayed by histomorphologic investigation of H&E staining) (Fig. 3i). Additional analysis of exocrine markers (CK-19 and AMY2A) showed no difference between Tbx3-KO (mes) and control mice (Fig. 3j). Interestingly, also the investigation of the endocrine markers insulin and glucagon revealed no significant difference (Fig. 3k–n). The size of islets (Fig. 3l) and the percentages of insulin- and glucagon-expressing cells (Fig. 3m,n) were not significantly different between Tbx3-KO (mes) and control mice. Thus, although Tbx3 displayed specific and dynamic regulation patterns on mRNA and protein levels in both mesenchyme and epithelium, its targeted ablation in either compartment does not relevantly alter proper pancreatic organ formation.
To check whether the well-described homolog TBX2 might compensate for TBX3 loss [9, 10], we performed immunohistochemistry for TBX2 in pancreata of both knockout models (Additional file 1: Fig. S4a,b). However, the overall staining intensity remained similar in control (epi) and Tbx3-KO (epi), as well as control (mes) and Tbx3-KO (mes) pancreata, albeit nuclear and cytoplasmatic protein distribution varied across various animals without a relevant genotype-specific trend.
Finally, we wanted to challenge the relevance of TBX3 in human pancreatic development. Thus, we employed an inducible knockdown of TBX3 by an shRNA in a previously reported iPSC line [12] during in vitro pancreatic progenitor differentiation [58–61]. We assessed trilineage potential (generation of acinar, ductal, and endocrine cells) in a recently described porcine urinary bladder (PUB) organ culture model for the maturation of pancreatic progenitor cells [62, 63] (Additional file 1: Fig. S5a). Of note, we did observe the generation of morphologically similar (Additional file 1: Fig. S5b), ductal (Additional file 1: Fig. S5c), acinar (Additional file 1: Fig. S5d), and endocrine (Additional file 1: Fig. S5e) structures on the PUB scaffold. Thus, the trilineage potential was not significantly altered by TBX3-knockdown during pancreatic progenitor differentiation. We conclude that TBX3 is dispensable for proper pancreatic organ formation in mice and men.
## Loss of Tbx3 results in increased fibrosis and excessive proliferation during recovery from acute pancreatitis
Since pancreatic development shares common regulatory patterns with pancreatic repair, we challenged Tbx3-KO (epi) mice and control mice with caerulein-induced acute pancreatitis to investigate the potential relevance of Tbx3 under exogenous stress conditions (Fig. 4a). Notably, no significant differences between control and Tbx3-KO mice were detected based on the amount of acinar-to-ductal metaplasia (ADM) as well as edema and inflammatory infiltration scores (Fig. 4b–e). As expected, a substantial increase of ADMs at 72h preceded by increased edema and infiltration of inflammatory cells (24h) was documented after induction of caerulein-driven acute pancreatitis in both groups of animals. The damage was largely repaired 7 days after the first injection. However, when assessing the proliferation rates of acinar cells with normal morphological configuration by KI-67 quantification, we noticed a significantly higher amount of KI-67+ cells per field in the Tbx3-KO (epi) mice compared to the control counterpart at 72h (4.3 cells/per field vs 1.7 cells/field) and 168h (3.0 cells/field vs 1.2 cells/field) (Fig. 4f–h). Interestingly, increased acinar proliferation was accompanied by significantly more fibrosis as quantified upon Sirius Red stainings at 72h in Tbx3-KO (epi) mice (Fig. 4i,j). The amount of fibrosis was reflected by the presence of ACTA2-positive cells around ADM structures (Fig. 4k). Interestingly, gene expression analysis of Acta2 at 72h after the onset of pancreatitis confirmed the significant difference between Tbx3-KO (epi) and control mice (Fig. 4l). Finally, we checked if apoptosis at 72h after induction of pancreatitis might play a role in the altered regeneration (Fig. 4m,n). However, no significant differences were observed concerning the rate of apoptotic cells. Thus, despite the fact that regeneration was achieved in Tbx3-KO (epi) mice, prolonged proliferation in acinar cells and accumulation of fibrosis suggest that Tbx3 is involved in proper regeneration from organ injury, but its absence is well compensated by other genes. Fig. 4Tbx3 loss leads to overshoot proliferation of acinar cells and accumulation of fibrosis. a Schematic representation of the caerulein-induced acute pancreatitis assay shown in b–l. Mice were euthanized at the indicated time points (arrows in blue). b Hematoxylin and eosin (H&E)-stained histological sections of control littermates of Tbx3-KO (epi) and Tbx3fl/fl; Ptf1aCre/+ (Tbx3-KO (epi)) pancreata after treatment with caerulein or vehicle. c Quantification of acinar-to-ductal metaplasia (ADM), d edema score, f inflammatory infiltration score per field in pancreata from caerulein-induced acute pancreatitis assay shown in a (n≥6 per group). f Representative immunohistochemistry stainings for KI-67 and g immunofluorescence co-staining of KI-67/AMY2A (red/green) at indicated timepoints. h Quantification of KI-67-positive acinar cells per field in pancreata from caerulein-induced acute pancreatitis assay shown in a (n≥6 per group, detailed in Additional file 3). i Picrosirius red stained histological sections of pancreata from caerulein-induced acute pancreatitis assay shown in a. j Quantification of picrosirius red-positive area as ratio to hematoxylin-based whole pancreas surface in pancreata from caerulein-induced acute pancreatitis assay shown in a (n≥6 per group). k Immunofluorescence staining for ACTA2 (red) in pancreata at 72h after induction of the acute pancreatitis. l Relative mRNA expression of Acta2 at 72h after induction of acute pancreatitis. m Representative immunohistochemistry staining for caspase-3 (CASP3) at the 72h timepoint and n respective quantifications. Cells were counterstained with DAPI for immunofluorescence analysis. Graphs present individual data points and mean with SEM. Two-way ANOVA with Sidak’s post-test was performed for graphs with multiple time points. Mann-Whitney test was performed to assess significance at single time points for apoptosis and Acta2. *, $p \leq 0.05$; **, $p \leq 0.01.$ Scale bars represent 100 μm
## Loss of Tbx3 results in increased susceptibility of acinar cells to fibroinflammatory stimuli
To further specify the subtle differences across the two genotypes during regeneration (Tbx3-KO (epi) vs. control (epi) mice), we performed RNA sequencing for the two most interesting time points (72 and 168h). Interestingly, we noticed a clear difference in the principal component analysis (PCA) of whole transcriptomes from Tbx3-KO (epi) mice compared to their respective controls at 72h, while the differences at 168h were virtually absent (Fig. 5a). We could identify 1152 differentially expressed genes (DEG) at 72h (762 up, 390 down in Tbx3-KO (epi)), while only 1 DEG was detected at 168h (Fig. 5b). Subsequently, we performed gene set enrichment analysis (GSEA) for hallmark (HM) gene sets and one acinar-specific NF-κB response gene set (Additional file 2: Supplementary Table 1) adapted from [64] (Fig. 5c). Interestingly, the significantly enriched gene sets included the Il6-Jak-Stat3-, Il2-Stat5-signaling, Inflammatory-Response, E2F-Targets, G2M-Checkpoint, and the acinar-specific NF-κB response. Notably, most gene sets are related to the immune system. We also checked other well-known gene sets relevant to pancreatic organ regeneration and detected no significant enrichment for the HM-Hedgehog, Wnt, Tgf-β, and Notch signaling. Concerning depleted gene sets, we noticed a non-significant depletion of Pancreatic-Beta-Cell-signature and Protein-Secretion-Signature, the latter being potentially indicative of normal acinar function. Fig. 5Whole transcriptome analysis uncovers a potential role of Tbx3 to fine-tune fibroinflammatory stimuli. a Principal component analysis of $$n = 3$$ mice per genotype (control littermates of Tbx3-KO (epi) and Tbx3fl/fl; Ptf1aCre/+ (Tbx3-KO (epi))) at 72 and 168h after induction of acute pancreatitis. b Differentially expressed genes (DEG) in Tbx3-KO (epi) mice at indicated timepoints. c Gene set enrichment analysis (GSEA) in Tbx3-KO (epi) of different hallmark (HM) gene sets and one acinar-specific upregulated gene set in response to NF-KB (Supplementary Table 1) [64] at 72h. d Volcano plot of the enriched IL-6-JAK-STAT3-Signaling in Tbx3-KO (epi) mice at 72h. e GSEA of immune cell signatures [65] in Tbx3-KO (epi) mice at 72h. f Immunohistochemistry staining for B220+ B cells at 72h after induction of acute pancreatitis. g Immunohistochemistry staining for MPO+ neutrophils at 72h after induction of acute pancreatitis. h Quantification of percentages of B220+ B cell area per field of view in $$n = 7$$ control (littermates of Tbx3-KO (epi)) pancreata and $$n = 6$$ Ptf1a-Cre-driven Tbx3-KO (epi) pancreata. i Quantification of MPO+ neutrophils per field of view in $$n = 8$$ control (littermates of Tbx3-KO (epi)) pancreata and $$n = 9$$ Ptf1a-Cre-driven Tbx3-KO (epi) pancreata. Graphs present individual data points and mean with SEM. Mann-Whitney test was performed to assess significance. *, $p \leq 0.05.$ j Heatmap of log2 fold changes of *Tbx* genes at 72h in Tbx3-KO (epi) pancreata compared to control pancreata. k Heatmap of log2 fold changes of known interaction partners of Tbx3 at 72h in Tbx3-KO (epi) pancreata compared to control pancreata. l DNA footprint analysis of putative DNA binding sites in the promoter of Lef1 at −18bp from TSS. PWM-based score 0.82, p-value=0.0009. m Proposed mechanism for Tbx3-related fine-tuning of acute pancreatitis. *, adjusted $p \leq 0.05.$ Scale bars represent 100 μm. MPO, myeloperoxidase.
Fitting with the enrichment of the Il6-Jak-*Stat3* gene signature, a variety of immune cell-related genes (e.g., Cxcl9, Il2rg) were significantly overexpressed at 72h post pancreatitis (Fig. 5d), which prompted us to investigate specific immune cell signatures (Additional file 2: Supplementary Table 2 [65]; (Fig. 5e). Interestingly, signatures for B cells and neutrophils were significantly upregulated. However, validation on protein level could only confirm an increased abundance of B cells in Tbx3-KO (epi) pancreata as compared to control (epi) counterparts at 72h after induction of pancreatitis (Fig. 5f,h). Neutrophil granulocyte quantification revealed only a trend (Fig. 5g,i). Infiltrating T cells and macrophages remained similar across the two genotypes (Additional file 1: Fig. S6a-d). To identify potential compensators for Tbx3 loss in the epithelial compartment, we analyzed the expression of all expressed *Tbx* genes at 72h (Fig. 5j). Interestingly, most *Tbx* genes remained unaltered, with Tbx18 being significantly downregulated and Tbx21 overexpressed in the Tbx3-KO (epi) mice. Next, a set of Tbx3 target genes were investigated [38] (Fig. 5k). Amongst these, the sole strongly and significantly upregulated gene was Lef1. Accordingly, we calculated the putative binding score and significance using the palindromic TBX3 binding site (18 bp) for the Lef1 promoter consensus sequence. Indeed, the relative similarity score was highly significant (score = 0.82, $$p \leq 0.0009$$), indicating direct TBX3 binding to the Lef1 promoter, most likely triggering its transcriptional repression (Fig. 5l).
## Discussion
In this study, we (i) determined the expression of TBX3 during pancreatic organogenesis, (ii) demonstrated expression switching between mesenchymal and epithelial cells during pancreatic development, and (iii) revealed expression in the stromal compartment of the adult pancreas. Surprisingly, epithelial and mesenchymal knockout of Tbx3 did not lead to statistically significant phenotypic alterations in the murine pancreas, and knockdown of TBX3 in a human iPSC-based system for pancreatic differentiation did not alter ex vivo pancreatic development. It is likely that TBX3 function may be compensated by other genes in these compartments during development. Driven by the fact that developmental programs can become re-activated during regeneration from injury, we investigated the relevance of Tbx3 for regeneration after caerulein-induced acute pancreatitis. Our findings indicate that epithelial deletion of Tbx3 does not significantly impact overall organ regeneration but does prolong proliferation and increase fibrosis. Again, this lack of a major effect in the face of Tbx3 ablation is likely attributable to other factors, while Tbx3 in itself is required for the fine-tuning of the regenerative process. Whole transcriptome analysis finally revealed putative mechanisms that alter the regeneration via enhanced immune cell-acinar interactions, fibroinflammatory stimuli, and dysregulated proliferation (Fig. 5m). Altogether, while Tbx3 is not a master regulator of pancreatic development and organ regeneration, our data show a relevant but subtle contribution of this factor to adult pancreatic organ homeostasis and development.
Interestingly, Tbx3 RNA expression levels demonstrated a virtually lower percentage of Tbx3+ cells than assayed in immunofluorescence analysis. This discrepancy was higher for acinar cells than for mesenchymal or stromal cells, which may be related to either technical limitations of single-cell RNA-seq [51–54] with a bias to filter out genes with low expression [50] and eventually boosted by the fact that the protease- and ribonuclease-rich acinar cells raise difficulties to isolate intact RNA [66–68]. Another potential explanation would be that Tbx3 expression levels in mesenchymal/stromal cells are, in general, higher (also evident in the violin plots of Fig. 1), which eases the detection.
Nevertheless by employing a knockout-validated antibody and a Venus reporter system [11, 42–47], the expression of TBX3, or Venus, respectively, was proven on the protein level. In addition, the evidence of human TBX3 expression in an in vitro iPSC differentiation in the epithelial compartment highlights that TBX3 is indeed expressed during embryonic pancreatic development.
Indeed, important roles for TBX3 have been shown in lung branching morphogenesis and ureter organogenesis [9, 10]. In both cases, TBX3 was located in the mesenchyme, and depletion of Tbx3 (together with Tbx2, which otherwise would have compensated Tbx3) resulted in defective organ patterning [9, 10]. Based on such recent publications, we chose to not only perform a knockout of Tbx3 in the epithelial compartment by Ptf1a-Cre but also in the mesenchymal compartment by Nkx3-2-Cre. Other than our anticipation from the expression level profiles and the recent publications [9, 10], neither of the knockout models led to a strong phenotype. TBX2 and 3 are members of the same T-box subfamily [1]. Their partially redundant roles in lung or ureter development [5, 9, 10] raise the question if this leads to redundant function in pancreatic evolution. However, our solely on gene and protein quantification relying study was underpowered to detect evidence for TBX2 compensating during pancreatic development as TBX2 remained unchanged in both the two knockout models and the pancreatic injury model. To ultimately clarify this hypothesis, a double-knockout of Tbx2 and Tbx3 in both pancreatic compartments would be necessary. However, this was clearly beyond the scope of the current study.
Even though we did not observe a gross impact of Tbx3 on recovery from acute pancreatitis, we observed (i) higher amounts of fibrosis, (ii) prolonged proliferation in acinar cells in its absence, and (iii) higher variance in some investigated parameters of regeneration of Tbx3-KO mice. Of note, we assessed the putative compensation by other *Tbx* genes during pancreatitis. Here we observed an upregulation of Tbx21. However, as Tbx21 (T-bet) and also Eomes (increased expression, though not significant) also play relevant roles in immune cell signaling, we attributed the enhanced expression to the increased inflammatory signature. Considering the downregulation of Tbx18, we pinpoint a recent publication of cardiac pacemaker differentiation where Tbx3 and Tbx18 are co-regulated, implying that a similar program could occur during pancreatic regeneration [69].
Based on the fact that Tbx3 is involved in the regulation of Wnt signaling in embryonic development [9] and Wnt signaling controls the proliferation of acinar cells during organ regeneration [24], Tbx3 may have a regulatory role on Wnt signaling during acinar cell regeneration. By demonstrating that Lef1, a central mediator of the Wnt signaling axis [24, 70–73], is not only significantly upregulated in Tbx3-KO (epi) mice but also possesses a putative DNA binding site for TBX3 in its promoter region, we postulate a previously unknown potential suppressive function of TBX3 on Lef1 building the following hypothetical model: The absence of Tbx3 in the pancreatic epithelium could trigger excessive proliferation eventually involving Lef1, a known member of the Wnt signaling [74, 75], and acinar-specific upregulated NF-KB-pathway members leading to abnormal proliferation of acinar cells [64]. Such proliferation pulse is accompanied by an increased fibroinflammatory response resulting in transient accumulation of B cell-infiltrated fibrotic tissue (Fig. 5m). This suggests an ongoing homeostatic process during pancreatic regeneration from an injury involving only Tbx3 for fine-tuning without the generation of substantial alterations upon genetic loss.
## Conclusions
In summary, Tbx3 shows switching expression patterns in the developing pancreas. Although Tbx3 seems dispensable for proper pancreatic development and lineage entry, its absence results in altered organ regeneration after induction of acute pancreatitis marked by enhanced fibrosis and inflammation.
## Single-cell RNA sequencing re-analysis
Expression patterns of Tbx3 on a single-cell level were derived from two different murine datasets: GSE101099 [25] for embryonic pancreata at E12.5, E14.5, and E17.5, as well as GSE109774 for adult pancreata [26]. Expression of TBX3 in a recently published human pancreatic duct-like cell differentiation approach from pluripotent stem cells was performed as recently described in detail (GSE162547) [37]. The expression of TBX3 in adult pancreata of four human donors was interrogated in a published dataset: GSE84133 [41]. Before expression analysis, preprocessing steps were conducted in RStudio with the version 4.0.4 and the “dplyr”, “Seurat” version 4.0.6 [76–79], and “patchwork” packages. For the murine datasets, cells with less than 1000 expressed genes, more than 4000 genes for E12.5, E14.5, and more than 5000 genes for E17.5 and genes expressed in fewer than 3 cells were filtered out before analysis. For GSE109774, cells expressing more than $5\%$ ERCC RNA spike-ins were additionally removed. For human datasets, cells expressing either less than 300 genes or more than 4000 genes and genes expressed in fewer than 10 cells were filtered out. Data were merged for E12.5 for GSM2699156_E12_B2 and GSM3140915_E12_v2, for E14.5 for GSM2699154_E14_B1, GSM2699155_E14_B2, and GSM3140916_E14_v2, and for E17.5 for GSM2699157_E17_B2, GSM3140917_E17_1_v2, and GSM3140918_E17_2_v2 of the respective dataset. Human datasets were also merged. Batch effects were corrected with the FindIntegrationAnchor() and the IntegrateData() function in the reciprocal PCA approach of the Seurat workflow. Datasets were Log-normalized, and the top 2000 highly variables genes were identified by the “vst” method for the murine datasets. For the human datasets, similar to our recent work [59], the top 4000 highly variable genes were identified. Data were scaled. A standard Seurat workflow was performed. Single-cell neighborhood was calculated with the first 30 principal components and clustering was performed with the Louvain algorithm with a resolution of 0.015 for E12.5, 0.5 for E14.5, 0.5 for E17.5, 0.2 for the adult murine pancreas, and 0.5 for human datasets. Cluster identity was annotated by specific marker genes.
## Bulk RNA expression re-analysis from publicly available RNA sequencing data sets
RNA expression levels of different T-Box transcription factors in embryonic stem cells, definitive endoderm, pancreatic endoderm, and pancreatic progenitors were derived as FPKM (fragments per kilobase of transcript per million mapped reads) from a previously published dataset (GSE131817, wildtype samples) [29, 30]. Data were analyzed in RStudio with the R version 4.0.4. The heatmap was generated with the “pheatmap” package (version 1.0.12), scaling was set to columns, and clustering followed the “ward. D2” method. A list of described TBX3 interaction partners was derived from [38]. A heatmap was generated as indicated above from the same dataset (GSE131817) [29, 30]. The scale was set to the row factor.
## Ethics statement
All animal care and procedure were conducted in compliance with the German legal regulations and were previously approved by the local governmental review board of the state of Baden-Württemberg (Permission no. 1477, O.195-4, O.195-6, O.195-10, and O.195-12) or conducted in compliance with the local government of Bavaria. All mouse work aspects were performed according to acknowledged guidelines of the Society of Laboratory Animals (GV-SOLAS) and of the Federation of Laboratory Animal Science Associations (FELASA).
## Mice
Ptf1aCre (Ptf1atm1(cre)Cvw) mouse strain [35] was previously described [80, 81]. Nkx3-2Cre (Nkx3-2tm1(cre)Wez) was a kind gift of Warren E. Zimmer (Texas A&M University) [82]. Tbx3flox (Tbx3tm3.1Moon) (kind gift of Anne M. Moon) [4], and Tbx3Venus (Tbx3tm1(Venus)Vmc) mouse (kind gift of Vincent M. Christoffels) [42] strain were maintained on a complex C57BL/6×129/*Sv* genetic background. Mice were housed and bred in a conventional health status-controlled animal facility. All animal care and procedures followed German legal regulations and if applicable were previously approved by the governmental review board of the state of Baden-Württemberg. All the aspects of the mouse work were carried out following strict guidelines to insure careful, consistent, and ethical handling of mice.
## Analysis of Ptf1aCre-specific recombination in adult pancreata
Formalin-fixed paraffin-embedded (FFPE) tissue sections of pancreata of Ptf1aCre [35] x R26-LSL-tdRFPKI [55] mice which were previously described [56] were a kind gift of Patrick Hermann (Ulm University Hospital, Department of Internal Medicine I).
## Knockout validation of TBX3-antibodies
TBX3 antibody was analyzed by immunofluorescence in archived Tbx3Cre/Cre at E13.5 or Tbx3wt/wt. [ 48].
## Mouse embryos and adult pancreata preparation
Embryos were collected at 12.5, 15.5, and 18.5 days post-coitum. Newborn mice were euthanized at post natal day 7 (P7). Pancreata were fixed in $4\%$ PFA for 2 h at room temperature or for 16 h at 4°C, cryopreserved, and embedded in optimal cutting temperature compound (OCT compound). Adult mice were at least 8 weeks old. Pancreata were fixed in cold $4\%$ PFA for 16 h at 4°C and embedded in paraffin for histological analysis.
## Model of Caerulein-induced acute pancreatitis
Acute pancreatitis in Tbx3-KO (epi) and control mice was induced in adult animal (≥ 8 weeks) by hourly injections (8 times) of 50 μg caerulein /kg bodyweight (Sigma-Aldrich) dissolved in PBS (vehicle). Mice were sacrificed 24, 72, and 168 h after the first caerulein injection. Corresponding volumes of vehicle were injected in Tbx3-KO (epi) and control mice, and pancreata were isolated 24 h after the first injection.
## Differentiation of pancreatic progenitor cells and organ culture model
Maintenance culture and differentiation to pancreatic progenitors of a well-described iPSC line with a doxycycline-inducible shRNA against TBX3 [12] was performed as described recently in a step-by-step protocol [58, 59] with slight modifications. For shRNA expression, 3 μg/mL doxycycline was added to the cells. The medium was changed on a daily basis. After the differentiation of pancreatic progenitor cells, organ culture of porcine urinary bladders was performed as described recently [62, 63]. Briefly, porcine urinary bladders (PUB) were cleaned, de-epithelialized, and sterilized with $0.1\%$ peroxy-acetic acid. In total, 500,000 cells per ring were seeded in 30 μL of $50\%$ Matrigel and $50\%$ basal medium as in [58, 59], including $5\%$ FCS and 10 μM Y-27632. The medium of PUBs was changed once a week, and doxycycline was added twice a week freshly. After 2 weeks, PUBs were fixed in $3.7\%$ formaldehyde and processed for histology as described recently [62, 63].
## Histology
All histological experiments on FFPE tissue were performed as previously described [20, 59, 62]) following standard procedures. Four-micrometer-thin sections were rehydrated. Heat-mediated antigen retrieval was either performed with citrate-based buffer (pH=6) or TRIS-based buffer (pH=9). Blocking was performed in $5\%$ normal donkey serum in $0.1\%$ Triton X-100 in PBS for 30 min at RT. Primary and secondary antibodies and respective dilution factors are listed in Additional file 2: Supplementary Table 3. Bright-field images were acquired using a Leica DM5500B microscope (Leica) equipped with a Leica DMC5400 camera and Leica Application Suite software (Leica) or by using a Zeiss Axioscope2 microscope (Carl Zeiss) ZEN3.1 imaging software (Carl Zeiss). Immunofluorescence images were obtained with a Zeiss Axioscope2 microscope equipped with an Axiocam 702 (Carl Zeiss). Acquired pictures were subsequently analyzed using ImageJ software (National Institutes of Health).
Immunofluorescence stainings of cryosections were performed following standard protocols. Briefly, 10-μm-thick cryo-sectioned pancreata were rehydrated in PBS for 30 min, washed twice with PBS containing 0.1 % Tween 20 (PBS-T), and permeabilized using 0.1 M glycine (Merck) and $0.1\%$ Triton X-100 (for NKX6.1: $0.5\%$ Triton X-100; Merck) in MilliQ water for 15 min at RT and blocked using $0.1\%$ Tween - 20, $10\%$ heat inactivated fetal calf serum (FCS), $0.1\%$ BSA, and $3\%$ donkey serum in PBS for 1 h at RT. Incubation with primary antibodies diluted in blocking solution occurred overnight at 4°C or 1 h at RT. The slides were washed 3× with PBS-T for 10 min each, and subsequently, secondary antibody solution was added for 2 h at RT. DAPI/PBS solution was added for 20 min before washing of slides 3× with PBS for 10 min each. Finally, the slides were mounted with Elvanol and kept 24 h at RT to dry.
## Histological quantifications
The total number of cells per Langerhans islets, as well as the number of insulin and glucagon-expressing cells, was counted manually from at least 3 different islets per pancreas on one section. Acinar-to-ductal metaplasias (ADMs) were quantified by counting at least ten 200× fields. Edema and immune cell infiltration were scored from 0 to 3 (0, no evident edema to 3, maximal degree of edema; 0, no immune infiltration to 3 maximal degree of immune infiltration) as previously described [20]. Proliferation was quantified by counting KI-67-positive cells in immunohistochemistry from ten 400× fields of each pancreas. KI67-positive cells within only clearly identifiable acinar structures were considered. Immune cells were excluded from the quantification. For picrosirius red-positive area/hematoxylin-based whole pancreas surface ratio quantification, pictures from at least five 100× fields were loaded into ImageJ software (National Institutes of Health) to perform color deconvolution. Areas covered by picrosirus red (red) and hematoxylin (yellow) were quantified automatically. Immune cell populations were quantified by counting the absolute number of positive cells per field in at least ten 200× fields. Caspase-3 (CASP3)-positive cells per field of view were quantified manually by counting the number of positive cells in each pancreas in at least six 100× fields. Monocytes, neutrophils, and T cells were manually quantified in at least five different 200× fields. B cells were quantified in at least 8 different 200× fields by quantifying the B220 positive area per field in ImageJ.
## RNA extraction, cDNA synthesis, qRT-PCR
RNA extraction was performed as described previously [20] with the RNeasy Plus Mini Kit (Qiagen) according to the manufacturer’s instructions. RNA concentration was determined with a NanoDrop. cDNA synthesis was performed with the iScript™ cDNA Synthesis Kit (Bio-Rad) following the manufacturer’s instruction. Briefly, 1 μg of RNA was transcribed. cDNA was diluted 15-fold. PCR was performed at the Rotor-Gene-Q (Qiagen) using 4 μL of diluted cDNA with 5 μL of Green Master Mix (Genaxxon) and 0.4 μM forward and reverse primer (each 0.5μL). Gene expression was normalized to ribosomal protein S18. Primer (Biomers) sequences are as follows: Acta2: fwd 5′-GTTCAGTGGTGCCTCTGTCA-3′, rev 5′-ACTGGGACGACATGGAAAAG-3′, S18: fwd 5′-GTAACCCGTTGAACCCCATT-3′, rev 5′- CCATCCAATCGGTAGTAGCG-3′.
## RNA sequencing
The amount of total RNA was quantified using the Qubit 2.0 Fluorometric Quantitation system (Thermo Fisher Scientific, Waltham, MA, USA), and the RNA integrity number (RIN) was determined using the Experion Automated Electrophoresis System (Bio-Rad, Hercules, CA, USA). RNA-seq libraries were prepared with the NEBNext® Ultra™ II Directional RNA sample preparation kit (New England Biolabs, Inc., Ipswich, MA, USA). Library concentrations were quantified with the Qubit 2.0 Fluorometric Quantitation system (Life Technologies, Carlsbad, CA, USA), and the size distribution was assessed using the Experion Automated Electrophoresis System (Bio-Rad, Hercules, CA, USA). For sequencing, samples were diluted and pooled into NGS libraries in equimolar amounts.
## Next-generation sequencing and raw data acquisition
Expression profiling libraries were sequenced on a HiSeq 3000 instrument (Illumina, San Diego, CA, USA) following a 50-base-pair, single-end recipe. Raw data acquisition (HiSeq Control Software, HCS, HD 3.4.0.38) and base calling (Real-Time Analysis Software, RTA, 2.7.7) was performed on-instrument, while the subsequent raw data processing off the instruments involved two custom programs (https://github.com/epigen/picard/) based on Picard tools (2.19.2) (https://broadinstitute.github.io/picard/). In a first step, base calls were converted into lane-specific, multiplexed, unaligned binary alignment map (BAM) files suitable for long-term archival (IlluminaBasecallsToMultiplexSam, 2.19.2-CeMM). In a second step, archive BAM files were demultiplexed into sample-specific, unaligned BAM files (IlluminaSamDemux, 2.19.2-CeMM).
## Transcriptome analysis
Next-generation sequencing (NGS) reads were mapped to the Genome Reference Consortium GRCm38 assembly via “Spliced Transcripts Alignment to a Reference” (STAR, 2.7.9a) [83] utilising the “basic” Ensembl transcript annotation from version e100 (April 2020) as reference transcriptome. Since the mm10 assembly flavour of the University of California, Santa Cruz (UCSC) Genome Browser was preferred for downstream data processing with Bioconductor packages for entirely technical reasons, Ensembl transcript annotation had to be adjusted to UCSC Genome Browser sequence region names. STAR was run with options recommended by the ENCODE project. NGS read alignments overlapping Ensembl transcript features were counted with the Bioconductor (3.14) GenomicAlignments (1.30.0) package via the summarizeOverlaps function in Union mode, ignoring secondary alignments and alignments not passing vendor quality filtering. Since the NEBNext® Ultra™ II Directional RNA protocol leads to sequencing of the first strand, all alignments needed inverting before strand-specific counting in feature (i.e., gene, transcript, and exon) orientation. Transcript-level counts were aggregated to gene-level counts, and the Bioconductor DESeq2 (1.34.0) [84] package was used to test for differential expression based on a model using the negative binomial distribution.
An initial exploratory analysis included principal component analysis (PCA), multidimensional scaling (MDS), sample distance, and expression heatmap plots, all annotated with variables used in the expression modelling (ggplot2 [85], 3.3.6, and Bioconductor ComplexHeatmap [86], 2.10.0), as well as volcano plots (Bioconductor EnhancedVolcano [87], 1.12.0). Biologically meaningful results were extracted from the model, and log2-fold values were shrunk with the CRAN ashr [88] (2.2.-54) package, while two-tailed p-values obtained from Wald testing were adjusted with the Bioconductor Independent Hypothesis Weighting [89] (IHW, 1.22.0) package.
## Gene set enrichment analysis
GSEA was performed using the hallmark data sets from the Molecular signatures database v7.4 (MSigDB, Broad Institute; http://software.broadinstitute.org/gsea/msigdb) and the GSEA software version 2.4.3 [90, 91]. Additional GSEA was performed for an acinar-specific NF-κB gene set. Genes upregulated with a significant fold change of least 2 from [64] were chosen to generate the reference gene list (Additional file 2: Supplementary Table 1). Immune cell gene sets were derived from [65] (Additional file 2: Supplementary Table 2). Significant enrichments were defined false discovery rate <0.25.
## Putative TBX3-DNA binding analysis
All the analysis was performed using R and Bioconductor (R Core Team, 2022). Lef1 promoter sequence was retrieved from M. musculus UCSC genome version mm10 known Gene database. The position weight matrix (PWM) was generated using the 8 entries of Lef1 promoter regions and was subsequently used for the sequence logo. Mouse TBX3 palindromic binding site was obtained from Uniprot (https://www.uniprot.org/uniprotkb/P70324/entry). Sequence logos, relative similarity score, correlation, Euclidean distance, and p-value were calculated using Transcription Factors binding sites tools in R.
## Statistical analysis
Statistical analysis was performed with GraphPad Prism 9.3.1 software. Data are expressed as individual data points and mean ± SEM (indicated in figure legends). Significance levels for islet size, insulin+, and glucagon+ cells were calculated with a Mann-Whitney test. Significance was evaluated with a two-way ANOVA and a Sidak post-test for the investigation of number of ADM per field, edema score, infiltration, score, proliferation, and fibrosis in the acute pancreatitis model. Significance levels for apoptosis, *Acta2* gene expression, and immune cell infiltrations at 72h after onset of the pancreatitis were assessed by Mann-Whitney test. Significance levels were defined as the following: $p \leq 0.05$ = *, $p \leq 0.01$ = **.
## Supplementary Information
Additional file 1: Fig. S1. Expression patterns of specific marker genes for cluster assignment in murine pancreata. Fig. S2. TBX3 is expressed during human pancreatic differentiation and in stellate cells of the adult pancreas. Fig. S3. Antibody and co-expression validation of TBX3 and Venus markers and validation of pancreatic recombination. Fig. S4. TBX2 expression in Ptf1a-Cre and Nkx3-2-Cre driven Tbx3-KO mice pancreata. Fig. S5. TBX3-knockdown does not impair the formation of human pancreatic tissue. Fig. S6. Tbx3 depletion does not alter T cell and macrophage infiltration during tissue regeneration after acute pancreatitis. Additional file 2: Supplementary Table 1. Upregulated acinar-specific NF-KB response geneset. Supplementary Table 2. Immune cell gene sets. Supplementary Table 3. Antibodies for histology. Additional file 3. Includes individual values for quantitative data depicted in Figs 3, 4 and 5.Additional file 4. Includes individual values for quantitative data depicted in Supplementary Figures 2 and 6.
## References
1. Papaioannou VE. **The t-box gene family: emerging roles in development, Stem cells and cancer**. *Development* (2014.0) **141** 3819-3833. DOI: 10.1242/dev.104471
2. Khan SF, Damerell V, Omar R, Du Toit M, Khan M, Maranyane HM. **The roles and regulation of TBX3 in development and disease**. *Gene.* (2020.0) **726** 144223. DOI: 10.1016/j.gene.2019.144223
3. Bertolessi M, Linta L, Seufferlein T, Kleger A, Liebau S. **A fresh look on T-Box factor action in early embryogenesis (T-Box factors in early development)**. *Stem Cells Dev* (2015.0) **24** 1833-1851. DOI: 10.1089/scd.2015.0102
4. Frank DU, Carter KL, Thomas KR, Burr RM, Bakker ML, Coetzee WA. **Lethal arrhythmias in Tbx3-deficient mice reveal extreme dosage sensitivity of cardiac conduction system function and homeostasis**. *Proc Natl Acad Sci U S A* (2012.0) **109** E154-E163. DOI: 10.1073/pnas.1115165109
5. Singh R, Hoogaars WM, Barnett P, Grieskamp T, Sameer Rana M, Buermans H. **Tbx2 and Tbx3 induce atrioventricular myocardial development and endocardial cushion formation**. *Cell Mol Life Sci* (2012.0) **69** 1377-1389. DOI: 10.1007/s00018-011-0884-2
6. Mesbah K, Harrelson Z, Théveniau-Ruissy M, Papaioannou VE, Kelly RG. **Tbx3 is required for outflow tract development**. *Circ Res* (2008.0) **103** 743-750. DOI: 10.1161/CIRCRESAHA.108.172858
7. Kumar P, Franklin S, Emechebe U, Hu H, Moore B, Lehman C. **TBX3 regulates splicing in vivo: a novel molecular mechanism for ulnar-mammary syndrome**. *PLoS Genet* (2014.0) **10** e1004247. DOI: 10.1371/journal.pgen.1004247
8. Emechebe U, Pavan Kumar P, Rozenberg JM, Moore B, Firment A, Mirshahi T. **T-box3 is a ciliary protein and regulates stability of the Gli3 transcription factor to control digit number**. *Elife.* (2016.0) **5** e07897. DOI: 10.7554/eLife.07897
9. Lüdtke TH, Rudat C, Wojahn I, Weiss AC, Kleppa MJ, Kurz J. **Tbx2 and Tbx3 act downstream of Shh to maintain canonical Wnt signaling during branching morphogenesis of the murine lung**. *Dev Cell* (2016.0) **39** 239-253. DOI: 10.1016/j.devcel.2016.08.007
10. Aydoğdu N, Rudat C, Trowe MO, Kaiser M, Lüdtke TH, Taketo MM. **TBX2 and TBX3 act downstream of canonical WNT signaling in patterning and differentiation of the mouse ureteric mesenchyme**. *Development* (2018.0) **145** dev171827. DOI: 10.1242/dev.171827
11. Russell R, Ilg M, Lin Q, Wu G, Lechel A, Bergmann W. **A dynamic role of TBX3 in the pluripotency circuitry**. *Stem Cell Rep* (2015.0) **5** 1155-1170. DOI: 10.1016/j.stemcr.2015.11.003
12. Klingenstein M, Raab S, Achberger K, Kleger A, Liebau S, Linta L. **TBX3 knockdown decreases reprogramming efficiency of human cells**. *Stem Cells Int* (2016.0) **2016** 6759343. DOI: 10.1155/2016/6759343
13. Weidgang CE, Russell R, Tata PR, Kühl SJ, Illing A, Müller M. **TBX3 directs cell-fate decision toward mesendoderm**. *Stem Cell Reports* (2013.0) **1** 248-265. DOI: 10.1016/j.stemcr.2013.08.002
14. Quarta C, Fisette A, Xu Y, Colldén G, Legutko B, Tseng YT. **Functional identity of hypothalamic melanocortin neurons depends on Tbx3**. *Nat Metab* (2019.0) **1** 222-235. DOI: 10.1038/s42255-018-0028-1
15. Cioffi M, Vallespinos-Serrano M, Trabulo SM, Fernandez-Marcos PJ, Firment AN, Vazquez BN. **MiR-93 controls adiposity via inhibition of Sirt7 and Tbx3**. *Cell Rep* (2015.0) **12** 1594-1605. DOI: 10.1016/j.celrep.2015.08.006
16. Kartikasari AER, Zhou JX, Kanji MS, Chan DN, Sinha A, Grapin-Botton A. **The histone demethylase Jmjd3 sequentially associates with the transcription factors Tbx3 and Eomes to drive endoderm differentiation**. *EMBO J* (2013.0) **32** 1393-1408. DOI: 10.1038/emboj.2013.78
17. Lu R, Yang A, Jin Y. **Dual functions of T-box 3 (Tbx3) in the control of self-renewal and extraembryonic endoderm differentiation in mouse embryonic stem cells**. *J Biol Chem* (2011.0) **286** 8425-8436. DOI: 10.1074/jbc.M110.202150
18. Mukherjee S, French DL, Gadue P. **Loss of TBX3 enhances pancreatic progenitor generation from human pluripotent stem cells**. *Stem Cell Reports* (2021.0) **16** 2617-2627. DOI: 10.1016/j.stemcr.2021.09.004
19. Perkhofer L, Walter K, Costa IG, Carrasco MCR, Eiseler T, Hafner S. **Tbx3 fosters pancreatic cancer growth by increased angiogenesis and activin/nodal-dependent induction of stemness**. *Stem Cell Res* (2016.0) **17** 367-378. DOI: 10.1016/j.scr.2016.08.007
20. Arnold F, Mahaddalkar PU, Kraus JM, Zhong X, Bergmann W, Srinivasan D. **Functional genomic screening during somatic cell reprogramming identifies DKK3 as a roadblock of organ regeneration**. *Adv Sci* (2021.0) **8** 2100626. DOI: 10.1002/advs.202100626
21. Murtaugh LC, Keefe MD. **Regeneration and repair of the exocrine pancreas**. *Annu Rev Physiol* (2015.0) **77** 229-249. DOI: 10.1146/annurev-physiol-021014-071727
22. Fendrich V, Esni F, Garay MVR, Feldmann G, Habbe N, Jensen JN. **Hedgehog signaling is required for effective regeneration of exocrine pancreas**. *Gastroenterology.* (2008.0) **135** 621-631. DOI: 10.1053/j.gastro.2008.04.011
23. Siveke JT, Lubeseder-Martellato C, Lee M, Mazur PK, Nakhai H, Radtke F. **Notch signaling is required for exocrine regeneration after acute pancreatitis**. *Gastroenterology.* (2008.0) **134** 544-555. DOI: 10.1053/j.gastro.2007.11.003
24. Keefe MD, Wang H, De La OJP, Khan A, Firpo MA, Murtaugh LC. **β-catenin is selectively required for the expansion and regeneration of mature pancreatic acinar cells in mice. DMM**. *Dis Model Mech* (2012.0) **5** 503-514. PMID: 22266944
25. Byrnes LE, Wong DM, Subramaniam M, Meyer NP, Gilchrist CL, Knox SM. **Lineage dynamics of murine pancreatic development at single-cell resolution**. *Nat Commun* (2018.0) **9** 3922. DOI: 10.1038/s41467-018-06176-3
26. 26.Tabula Muris Consortium; Overall coordination; Logistical coordination; Organ collection and processing; Library preparation and sequencing; Computational data analysis; Cell type annotation; Writing group; Supplemental text writing group; Principal investigators. Single-cell transcriptomics of 20 mouse organs creates a Tabula Muris. Nature. 2018;562(7727):367–72. 10.1038/s41586-018-0590-4.
27. Hatano M, Iitsuka Y, Yamamoto H, Dezawa M, Yusa S, Kohno Y. **Ncx, a Hox11 related gene, is expressed in a variety of tissues derived from neural crest cells**. *Anat Embryol (Berl)* (1997.0) **195** 419-425. DOI: 10.1007/s004290050061
28. Shirasawa S, Yunker AMR, Roth KA, Brown GA, Horning S, Korsmeyer SJ. **Enx (HOX11L1)-deficient mice develop myenteric neuronal hyperplasia and megacolon**. *Nat Med* (1997.0) **3** 646-650. DOI: 10.1038/nm0697-646
29. 29.Philippi A, Heller S, Costa IG, Senée V, Breunig M, Li Z, et al. Mutations and variants of ONECUT1 in diabetes. Nat Med. 2021. 10.1038/s41591-021-01502-7.
30. Heller S, Li Z, Lin Q, Geusz R, Breunig M, Hohwieler M. **Transcriptional changes and the role of ONECUT1 in hPSC pancreatic differentiation**. *Commun Biol* (2021.0) **4** 1298. DOI: 10.1038/s42003-021-02818-3
31. Burlison JS, Long Q, Fujitani Y, Wright CVE, Magnuson MA. **Pdx-1 and Ptf1a concurrently determine fate specification of pancreatic multipotent progenitor cells**. *Dev Biol* (2008.0) **316** 74-86. DOI: 10.1016/j.ydbio.2008.01.011
32. Haumaitre C, Barbacci E, Jenny M, Ott MO, Gradwohl G, Cereghini S. **Lack of TCF2/vHNF1 in mice leads to pancreas agenesis**. *Proc Natl Acad Sci U S A* (2005.0) **102** 1490-1495. DOI: 10.1073/pnas.0405776102
33. Gu G, Dubauskaite J, Melton DA. **Direct evidence for the pancreatic lineage: NGN3+ cells are islet progenitors and are distinct from duct progenitors**. *Development.* (2002.0) **129** 2447-2457. DOI: 10.1242/dev.129.10.2447
34. Henseleit KD, Nelson SB, Kuhlbrodt K, Hennings JC, Ericson J, Sander M. **NKX6 transcription factor activity is required for α- and β-cell development in the pancreas**. *Development.* (2005.0) **132** 3139-3149. DOI: 10.1242/dev.01875
35. Kawaguchi Y, Cooper B, Gannon M, Ray M, MacDonald RJ, Wright CVE. **The role of the transcriptional regulator Ptf1a in converting intestinal to pancreatic progenitors**. *Nat Genet* (2002.0) **32** 128-134. DOI: 10.1038/ng959
36. 36.Gonçalves CA, Larsen M, Jung S, Stratmann J, Nakamura A, Leuschner M, et al. A 3D system to model human pancreas development and its reference single-cell transcriptome atlas identify signaling pathways required for progenitor expansion. Nat Commun. 2021;12(1):3144. Available from: http://www.nature.com/articles/s41467-021-23295-6.
37. Wiedenmann S, Breunig M, Merkle J, von Toerne C, Georgiev T, Moussus M. **Single-cell-resolved differentiation of human induced pluripotent stem cells into pancreatic duct-like organoids on a microwell chip**. *Nat Biomed Eng* (2021.0) **5** 897-913. DOI: 10.1038/s41551-021-00757-2
38. Washkowitz AJ, Gavrilov S, Begum S, Papaioannou VE. **Diverse functional networks of Tbx3 in development and disease**. *Wiley Interdiscip Rev Syst Biol Med* (2012.0) **4** 273-283. DOI: 10.1002/wsbm.1162
39. Eblaghie MC, Song SJ, Kim JY, Akita K, Tickle C, Jung HS. **Interactions between FGF and Wnt signals and Tbx3 gene expression in mammary gland initiation in mouse embryos**. *J Anat* (2004.0) **205** 1-13. DOI: 10.1111/j.0021-8782.2004.00309.x
40. Davenport TG, Jerome-Majewska LA, Papaioannou VE. **Mammary gland, limb and yolk sac defects in mice lacking Tbx3, the gene mutated in human ulnar mammary syndrome**. *Development.* (2003.0) **130** 2263-2273. DOI: 10.1242/dev.00431
41. Baron M, Veres A, Wolock SL, Faust AL, Gaujoux R, Vetere A. **A single-cell transcriptomic map of the human and mouse pancreas reveals inter- and intra-cell population structure**. *Cell Syst* (2016.0) **3** 346-360. DOI: 10.1016/j.cels.2016.08.011
42. Kunasegaran K, Ho V, Chang THT, De Silva D, Bakker ML, Christoffels VM. **Transcriptional repressor Tbx3 is required for the hormone-sensing cell lineage in mammary epithelium**. *PLoS One* (2014.0) **9** e110191. DOI: 10.1371/journal.pone.0110191
43. van Eif VWW, Stefanovic S, van Duijvenboden K, Bakker M, Wakker V, de Gier-de Vries C. **Transcriptome analysis of mouse and human sinoatrial node cells reveals a conserved genetic program**. *Development* (2019.0) **146** dev173161. PMID: 30936179
44. Mohan RA, Bosada FM, Weerd JHV, van Duijvenboden K, Wang J, Mommersteeg MTM. **T-box transcription factor 3 governs a transcriptional program for the function of the mouse atrioventricular conduction system**. *Proc Natl Acad Sci U S A* (2020.0) **117** 18617-18626. DOI: 10.1073/pnas.1919379117
45. Mohan RA, Mommersteeg MTM, Domínguez JN, Choquet C, Wakker V, de Gier-de Vries C. **Embryonic Tbx3+ cardiomyocytes form the mature cardiac conduction system by progressive fate restriction**. *Development* (2018.0) **145** dev167361. DOI: 10.1242/dev.167361
46. van Weerd JH, Mohan RA, van Duijvenboden K, Hooijkaas IB, Wakker V, Boukens BJ. **Trait-associated noncoding variant regions affect tbx3 regulation and cardiac conduction**. *Elife.* (2020.0) **9** e56697. DOI: 10.7554/eLife.56697
47. van Eif VWW, Stefanovic S, Mohan RA, Christoffels VM. **Gradual differentiation and confinement of the cardiac conduction system as indicated by marker gene expression**. *Biochim Biophys Acta Mol Cell Res* (2020.0) **1867** 118509. DOI: 10.1016/j.bbamcr.2019.07.004
48. Hoogaars WMH, Engel A, Brons JF, Verkerk AO, De Lange FJ, Wong LYE. **Tbx3 controls the sinoatrial node gene program and imposes pacemaker function on the atria**. *Genes Dev* (2007.0) **21** 1098-1112. DOI: 10.1101/gad.416007
49. Landsman L, Nijagal A, Whitchurch TJ, VanderLaan RL, Zimmer WE, MacKenzie TC. **Pancreatic mesenchyme regulates epithelial organogenesis throughout development**. *PLoS Biol* (2011.0) **9** e1001143. DOI: 10.1371/journal.pbio.1001143
50. Svensson V, Natarajan KN, Ly LH, Miragaia RJ, Labalette C, Macaulay IC. **Power analysis of single-cell rnA-sequencing experiments**. *Nat Methods* (2017.0) **14** 381-387. DOI: 10.1038/nmeth.4220
51. Hicks SC, Townes FW, Teng M, Irizarry RA. **Missing data and technical variability in single-cell RNA-sequencing experiments**. *Biostatistics.* (2018.0) **19** 562-578. DOI: 10.1093/biostatistics/kxx053
52. Zheng Y, Zhong Y, Hu J, Shang X. **SCC: an accurate imputation method for scRNA-seq dropouts based on a mixture model**. *BMC Bioinformatics* (2021.0) **22** 5. DOI: 10.1186/s12859-020-03878-8
53. Islam S, Zeisel A, Joost S, La Manno G, Zajac P, Kasper M. **Quantitative single-cell RNA-seq with unique molecular identifiers**. *Nat Methods* (2014.0) **11** 163-166. DOI: 10.1038/nmeth.2772
54. Papalexi E, Satija R. **Single-cell RNA sequencing to explore immune cell heterogeneity**. *Nat Rev Immunol* (2018.0) **18** 35-45. DOI: 10.1038/nri.2017.76
55. Luche H, Weber O, Rao TN, Blum C, Fehling HJ. **Faithful activation of an extra-bright red fluorescent protein in “knock-in” Cre-reporter mice ideally suited for lineage tracing studies**. *Eur J Immunol* (2007.0) **37** 43-53. DOI: 10.1002/eji.200636745
56. Walter K, Tiwary K, Trajkovic-Arsic M, Hidalgo-Sastre A, Dierichs L, Liffers ST. **MEK inhibition targets cancer stem cells and impedes migration of pancreatic cancer cells in vitro and in vivo**. *Stem Cells Int* (2019.0) **2019** 8475389. DOI: 10.1155/2019/8475389
57. Cozzitorto C, Mueller L, Ruzittu S, Mah N, Willnow D, Darrigrand JF. **A specialized niche in the pancreatic microenvironment promotes endocrine differentiation**. *Dev Cell* (2020.0) **55** 150-162. DOI: 10.1016/j.devcel.2020.08.003
58. Breunig M, Merkle J, Melzer MK, Heller S, Seufferlein T, Meier M. **Differentiation of human pluripotent stem cells into pancreatic duct-like organoids**. *STAR Protoc* (2021.0) **2** 100913. DOI: 10.1016/j.xpro.2021.100913
59. Breunig M, Merkle J, Wagner M, Melzer MK, Barth TFE, Engleitner T. **Modeling plasticity and dysplasia of pancreatic ductal organoids derived from human pluripotent stem cells**. *Cell Stem Cell* (2021.0) **28** 1105-1124. DOI: 10.1016/j.stem.2021.03.005
60. Merkle J, Breunig M, Schmid M, Allgöwer C, Krüger J, Melzer MK. **CDKN2A-mutated pancreatic ductal organoids from induced pluripotent stem cells to model a cancer predisposition syndrome**. *Cancers (Basel)* (2021.0) **13** 5139. DOI: 10.3390/cancers13205139
61. Krüger J, Breunig M, Pasquini LP, Morawe M, Groß A, Arnold F. **Functional genomic screening in human pluripotent stem cells reveals new roadblocks in early pancreatic endoderm formation**. *Cells.* (2022.0) **11** 582. DOI: 10.3390/cells11030582
62. Melzer MK, Breunig M, Arnold F, Wezel F, Azoitei A, Roger E. **Organoids at the PUB: the porcine urinary bladder serves as a pancreatic niche for advanced cancer modeling**. *Adv Healthc Mater* (2022.0) **11** e2102345. DOI: 10.1002/adhm.202102345
63. Melzer MK, Breunig M, Lopatta P, Hohwieler M, Merz S, Azoitei A. **Protocol to use de-epithelialized porcine urinary bladder as a tissue scaffold for propagation of pancreatic cells**. *STAR Protoc* (2022.0) **3** 101869. DOI: 10.1016/j.xpro.2022.101869
64. Liou GY, Döppler H, Necela B, Krishna M, Crawford HC, Raimondo M. **Macrophage-secreted cytokines drive pancreatic acinar-to-ductal metaplasia through NF-ΚB and MMPs**. *J Cell Biol* (2013.0) **202** 563-577. DOI: 10.1083/jcb.201301001
65. Nirmal AJ, Regan T, Shih BB, Hume DA, Sims AH, Freeman TC. **Immune cell gene signatures for profiling the microenvironment of solid tumors**. *Cancer Immunol Res* (2018.0) **6** 1388-1400. DOI: 10.1158/2326-6066.CIR-18-0342
66. Assi M, Dauguet N, Jacquemin P. **DIE-RNA: A reproducible strategy for the digestion of normal and injured pancreas, isolation of pancreatic cells from genetically engineered mouse models and extraction of high quality RNA**. *Front Physiol* (2018.0) **9** 129. DOI: 10.3389/fphys.2018.00129
67. Griffin M, Abu-El-Haija M, Abu-El-Haija M, Rokhlina T, Uc A. **Simplified and versatile method for isolation of high-quality RNA from pancreas**. *Biotechniques.* (2012.0) **52** 332-334. DOI: 10.2144/0000113862
68. Li D, Ren W, Wang X, Wang F, Gao Y, Ning Q. **A modified method using TRIzol® reagent and liquid nitrogen produces high-quality RNA from rat pancreas**. *Appl Biochem Biotechnol* (2009.0) **158** 253-261. DOI: 10.1007/s12010-008-8391-0
69. Zhao H, Wang F, Zhang W, Yang M, Tang Y, Wang X. **Overexpression of TBX3 in human induced pluripotent stem cells (hiPSCs) increases their differentiation into cardiac pacemaker-like cells**. *Biomed Pharmacother* (2020.0) **130** 110612. DOI: 10.1016/j.biopha.2020.110612
70. Morris JP, Cano DA, Sekine S, Wang SC, Hebrok M. **β-catenin blocks Kras-dependent reprogramming of acini into pancreatic cancer precursor lesions in mice**. *J Clin Invest* (2010.0) **120** 508-520. DOI: 10.1172/JCI40045
71. Zhang Y, Yu J, Shi C, Huang Y, Wang Y, Yang T. **Lef1 contributes to the differentiation of bulge stem cells by nuclear translocation and cross-talk with the notch signaling pathway**. *Int J Med Sci* (2013.0) **10** 738-746. DOI: 10.7150/ijms.5693
72. Zhou P, Byrne C, Jacobs J, Fuchs E. **Lymphoid enhancer factor 1 directs hair follicle patterning and epithelial cell fate**. *Genes Dev* (1995.0) **9** 700-713. DOI: 10.1101/gad.9.6.700
73. Singhi AD, Lilo M, Hruban RH, Cressman KL, Fuhrer K, Seethala RR. **Overexpression of lymphoid enhancer-binding factor 1 (LEF1) in solid-pseudopapillary neoplasms of the pancreas**. *Mod Pathol* (2014.0) **27** 1355-1363. DOI: 10.1038/modpathol.2014.40
74. Santiago L, Daniels G, Wang D, Deng FM, Lee P. **Wnt signaling pathway protein LEF1 in cancer, as a biomarker for prognosis and a target for treatment**. *Am J Cancer Res* (2017.0) **7** 1389-1406. PMID: 28670499
75. Clevers H. **Wnt/beta-catenin signaling in development and disease**. *Cell.* (2006.0) **127** 469-480. DOI: 10.1016/j.cell.2006.10.018
76. Satija R, Farrell JA, Gennert D, Schier AF, Regev A. **Spatial reconstruction of single-cell gene expression data**. *Nat Biotechnol* (2015.0) **33** 495-502. DOI: 10.1038/nbt.3192
77. Butler A, Hoffman P, Smibert P, Papalexi E, Satija R. **Integrating single-cell transcriptomic data across different conditions, technologies, and species**. *Nat Biotechnol* (2018.0) **36** 411-420. DOI: 10.1038/nbt.4096
78. Stuart T, Butler A, Hoffman P, Hafemeister C, Papalexi E, Mauck WM. **Comprehensive integration of single-cell data**. *Cell.* (2019.0) **177** 1888-1902. DOI: 10.1016/j.cell.2019.05.031
79. Hao Y, Hao S, Andersen-Nissen E, Mauck WM, Zheng S, Butler A. **Integrated analysis of multimodal single-cell data**. *Cell.* (2021.0) **184** 3573-3587. DOI: 10.1016/j.cell.2021.04.048
80. Russell R, Perkhofer L, Liebau S, Lin Q, Lechel A, Feld FM. **Loss of ATM accelerates pancreatic cancer formation and epithelial-mesenchymal transition**. *Nat Commun* (2015.0) **6** 7677. DOI: 10.1038/ncomms8677
81. Perkhofer L, Schmitt A, Romero Carrasco MC, Ihle M, Hampp S, Ruess DA. **ATM deficiency generating genomic instability sensitizes pancreatic ductal adenocarcinoma cells to therapy-induced DNA damage**. *Cancer Res* (2017.0) **77** 5576-5590. DOI: 10.1158/0008-5472.CAN-17-0634
82. Verzi MP, Stanfel MN, Moses KA, Kim BM, Zhang Y, Schwartz RJ. **Role of the homeodomain transcription factor Bapx1 in mouse distal stomach development**. *Gastroenterology.* (2009.0) **136** 1701-1710. DOI: 10.1053/j.gastro.2009.01.009
83. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S. **STAR: ultrafast universal RNA-seq aligner**. *Bioinformatics.* (2013.0) **29** 15-21. DOI: 10.1093/bioinformatics/bts635
84. Love MI, Huber W, Anders S. **Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2**. *Genome Biol* (2014.0) **15** 550. DOI: 10.1186/s13059-014-0550-8
85. Wickham H. *Package ‘ggplot2’: elegant graphics for data analysis* (2016.0)
86. Gu Z, Eils R, Schlesner M. **Complex heatmaps reveal patterns and correlations in multidimensional genomic data**. *Bioinformatics.* (2016.0) **32** 2847-2849. DOI: 10.1093/bioinformatics/btw313
87. Blighe K, Rana S, Lewis M. *EnhancedVolcano: Publication-ready volcano plots with enhanced colouring and labelin. R-Package* (2019.0)
88. Stephens M. **False discovery rates: a new deal**. *Biostatistics.* (2017.0) **18** 275-294. PMID: 27756721
89. Ignatiadis N, Klaus B, Zaugg JB, Huber W. **Data-driven hypothesis weighting increases detection power in genome-scale multiple testing**. *Nat Methods* (2016.0) **13** 577-580. DOI: 10.1038/nmeth.3885
90. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA. **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
91. Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J. **PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes**. *Nat Genet* (2003.0) **34** 267-273. DOI: 10.1038/ng1180
92. Brulet P, Babinet C, Kemler R, Jacob F. **Monoclonal antibodies against trophectoderm-specific markers during mouse blastocyst formation**. *Proc Natl Acad Sci U S A* (1980.0) **77** 4113-4117. DOI: 10.1073/pnas.77.7.4113
|
---
title: Associations of genome-wide and regional autozygosity with 96 complex traits
in old order Amish
authors:
- Megan T. Lynch
- Kristin A. Maloney
- Huichun Xu
- James A. Perry
- Regeneron Genetics Center
- Alan R. Shuldiner
- Braxton D. Mitchell
journal: BMC Genomics
year: 2023
pmcid: PMC10029202
doi: 10.1186/s12864-023-09208-5
license: CC BY 4.0
---
# Associations of genome-wide and regional autozygosity with 96 complex traits in old order Amish
## Abstract
Background: Autozygosity, the proportion of the genome that is homozygous by descent, has been associated with variation in multiple health-related traits impacting evolutionary fitness. Autozygosity (FROH) is typically measured from runs of homozygosity (ROHs) that arise when identical-by-descent (IBD) haplotypes are inherited from each parent. Population isolates with a small set of common founders have elevated autozygosity relative to outbred populations. Methods: *In this* study, we examined whether degree of autozygosity was associated with variation in 96 cardiometabolic traits among 7221 Old Order Amish individuals residing in Lancaster County, PA. We estimated the average length of an ROH segment to be 6350 KB, with each individual having on average 17.2 segments 1.5 KB or larger. Measurements of genome-wide and regional FROH were used as the primary predictors of trait variation in association analysis. Results: In genome-wide FROH analysis, we did not identify any associations that withstood Bonferroni-correction ($$p \leq 0.0005$$). However, on regional FROH analysis, we identified associations exceeding genome-wide thresholds for two traits: serum bilirubin levels, which were significantly associated with a region on chromosome 2 localized to a region surrounding UGT1A10 ($$p \leq 1$$ × 10− 43), and HbA1c levels, which were significantly associated with a region on chromosome 8 localized near CHRNB3 ($$p \leq 8$$ × 10− 10). Conclusions: These analyses highlight the potential value of autozygosity mapping in founder populations.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12864-023-09208-5.
## Introduction
Autozygosity, which is defined as the probability that a region is homozygous due to the inheritance of alleles identical-by-descent (IBD), is determined by the presence of extended homozygosity in that region. Since runs of homozygosity (ROHs) are a recognized signature of recessive inheritance, homozygosity mapping using ROH has commonly been used to map recessive disorders when there is suspicion that parents may share a common haplotype. [ 1].
More recently, studies have evaluated the impact of overall autozygosity on variation in common complex traits, where overall autozygosity is estimated as the proportion of ROH across the entire genome (genome-wide autozygosity). Genetic homogeneity, estimated as overall autozygosity, has previously been linked to adverse health outcomes in multiple traits impacting evolutionary fitness. Populations with increased autozygosity are more likely to experience inbreeding depression, or reduced fitness, which has been linked to a range of phenotypic consequences including cardiovascular disease, [2, 3] shorter stature, [4, 5] lower general cognitive ability, [5] decreased fertility, [6, 7] and higher hip-to-waist ratio. [ 8] In case-control study designs involving outbred populations, higher genome-wide autozygosity has also been associated with coronary artery disease [3] and amyotrophic lateral sclerosis. [ 9] These studies raise the potential for employing ROH mapping to identify hotspots along the genome that contain multiple rare, recessive loci that influence health and disease. Autozygosity mapping may be a particularly powerful tool when applied to founder populations where consanguinity is high. [ 10].
Elevated autozygosity is seen in population isolates that have a small set of common founders. [ 11] Due to limited genetic diversity, recent inbreeding in such populations magnifies the occurrence of mildly deleterious variants, resulting in an increased burden of recessive disorders that are rare in the general population. [ 12] The Amish of Lancaster County, PA are a relatively recent founder population who emigrated from Europe to Lancaster 14–15 generations ago. In this study, we tested for association between degree of autozygosity and variation in 96 different complex traits among 7221 Old Order Amish individuals residing in Lancaster County, PA. We estimated genome-wide levels of autozygosity as the proportion of the autosomal genome in runs of homozygosity > 1.5 Mb. We also estimated the probability of autozygosity at 10 Kb average intervals throughout the genome. We used these measures in association analysis to assess evidence for association of genome-wide and locus-specific association of autozygosity with 96 different phenotypes.
## Participants
Subjects in this study were recruited through multiple protocols carried out between 2003 and 2019 as part of the Amish Research Program. Study participants were recruited through the University of Maryland Amish Research Clinic (ARC) in Lancaster, PA. Studies were designed to assess determinants of cardio-metabolic and bone health in the community, and enrollment was open to volunteers throughout the Lancaster Amish community. [ 13, 14] Recruitment was generally phenotype agnostic; that is, subjects were not recruited for particular diseases or health conditions. This report is based on 7,221 apparently healthy Amish individuals 18 years of age or older recruited from the community and in whom we obtained genotyping data. The average age of participants was 41.8 years and the population was $43\%$ male.
## Phenotyping
Clinical examinations were performed by trained nursing staff at the Amish Research Clinic in Lancaster, PA or in the homes of study participants. For this report, we restricted analysis to a set of phenotypes measured in common across all (or nearly all) protocols. These include basic anthropometrics, blood pressure, fasting blood lipids, glucose and insulin, hemoglobin a1c (glycated hemoglobin), basic blood chemistries, and medical histories. A complete list of the 96 phenotypes analyzed is provided in Supplemental Table 1.
## Genotyping
Genotyping was performed at the Regeneron Genetics Center using the Illumina Global Screening Array, which included 490k single nucleotide polymorphisms (SNPs) that passed quality control parameters. Samples with > $5\%$ missingness were removed from analysis as were variants with > $2\%$ missingness or high levels of Mendelian errors. We also applied a minor allele frequency threshold of 0.01. For these analyses we used only the unimputed genotype data.
## Genome-wide autozygosity estimation
Genome-wide autozygosity was estimated as the proportion of the genome that fell into ROHs. SNPs with more than $3\%$ missingness across individuals and with a minor allele frequency less than $5\%$ were excluded from ROH calculations. ROH estimates were made using the IBD tool implemented in Plink, which has several built-in arguments. [ 15] The key parameters to identify truly autozygous segments are minimum length (kilobase, Kb) needed for a tract to qualify as homozygous, number of contiguous homozygous SNPs, and minimum tract density requirements. To avoid very short and common strands of homozygosity that occur throughout the genome due to linkage disequilibrium (LD), we used a minimal length of ROH of 1.5 Mb. [ 5, 16] The fraction of each autosomal genome in ROH > 1.5 Mb correlates well with pedigree-based estimates of inbreeding. [ 16] All other parameters were the default parameters set by Plink except for a decrease in the number of contiguous homozygous SNPs to 50, consistent with previous studies. [ 5, 11] We used the PLINK default settings to allow each autozygous segment to include up to five missing SNPs and up to one heterozygous SNP. Since study individuals were genotyped with the same assay and processed under the same QC protocol, ROH was defined as the total kilobase (Kb) included in a ROH.
## Regional autozygosity estimation
Regional autozygosity computation was performed using the GARLIC software, which outputs ROH and length class information in UCSC’s plain-text BED format. [ 17] Autozygosity was estimated using a sliding window approach with a step size of 10 kb, a 0.001 genotype error rate, population-specific allele frequencies and a window size of 50 SNPs. For each individual, we extracted ROH inclusion status for each SNP on the genotyping array (i.e., SNP autozygous in the ROH segment or not) for use in the downstream association analysis. Each SNP was assigned a value of 1 indicating that the SNP is included in an ROH or a value of 0 indicating that the SNP is not included in an ROH. Therefore, each SNP now represents the status of regional autozygosity.
## Statistical analysis
We tested for association of the autozygosity estimates (genome-wide and regional) with trait variation using a linear mixed model with phenotype as the outcome and autozygosity as the independent variable. As fixed covariates for each trait analysis, we included age, sex, and genotypes for two large effect SNPs (APOB p.Arg3527Gln and APOC3 p.Arg19Ter) previously identified in this Amish community affecting lipid levels. [ 18, 19] We also included a genomic relationship matrix (GRM) as a random effect to account for covariation among related individuals. [ 20] Genome wide or regional FROH was used as the primary predictor of the traits of interest. The genome wide significance threshold after Bonferroni correction was $$p \leq 0.0005$$ ($\frac{0.05}{96}$ traits). As a secondary analysis, we tested for sex-specific effects of autozygosity on trait variation using an autozygosity*sex interaction term. For regional analysis, an independent association analysis was performed for each of 300k SNPs using ROH status as the primary predictor of the 96 traits of interest. We adjusted the significance threshold to account for 170k independent SNPs among these 300k using an LD threshold of r2 = 0.50. The Bonferroni adjusted significance threshold was $$p \leq 3.1$$ × 10− 9.
## ROH across the genome
On average, each participant had 17.2 segments of > 1.5 Mb, which in aggregate spanned 110 Mb, or ~ $3.7\%$ of the genome (Fig. 1). ROH were present on each autosome and were widely distributed across the genome. We assessed the genome-wide distribution of ROHs by the frequency of entire ROH segments and by the frequency in which SNPs are included in an ROH segment.
Fig. 1Histograms of ROH segment length and number. Participant had an average of 17.2 segments of > 1.5 Mb, which spanned an average of 110 Mba) Frequency distribution of the total length of ROH segments across the genome (Mb) per subjectb) Frequency distribution of the total number of ROH segments present along the genome per subject *In* general, shorter ROH segments occurred more frequently than longer ROH segments, with a mean ROH segment length of 6.3 Mb (Fig. 2A and B). Two short segments had notably high frequencies (Fig. 2B). The highest frequency segment is located at chr1: [145,927,328: 148,353,534] and is 2.4 Mb long, occurring in 184 individuals. The segment with the second highest frequency, occurring in 173 individuals, is located at chr10: [42,113,412: 46,332,633], and is 4.2 Mb long.
Fig. 2Size distribution of ROH segments a) Plot showing the total number and average length of autozygous segments. Each dot corresponds to one individual b) Plot showing the frequency and length of autozygous segments. Each dot corresponds to one autozygous segment identified within at least one member of the study population There are several regions of the genome for which the frequency of individuals with SNPs in ROH segments was particularly high (chr 2: [134600135:136231566]; chr 2: [176,901,247: 178,324,285]; chr 5: [130334251:132873169]; chr 6: [27,512,529: 29,306,571]; chr 11: [47253513:49758165]; chr 20: [34813904:35819776]) (Fig. 3, Supplemental Table 2). For example, $18\%$ of individuals had ROH in a region on chromosome 5 containing genes RAPGEF6, FNIP1, and ACSL6. This region has previously been identified as having enriched ROH frequency for SNPs. [ 21] In chromosome 2, $13\%$ of subjects had ROH in a region that includes LCT, known to have recent positive selection in European ancestry individuals, [22] and genes UBXN4 and R3HDM1, which have not previously been noted in ROH studies. Our finding that a relatively high proportion of subjects ($14\%$) had ROH on chromosome 11 in a region harboring FOLH1 and OR4A47 is consistent with previous findings. [ 21].
Fig. 3Frequency of ROH segments across the autosome. Manhattan plot showing the frequency that a SNP occurred within an autozygous segment in the cohort Almost all SNPs fell into an ROH at least once, except in six instances across the genome in which small clusters of SNPs did not belong in any ROH segment (Fig. 3, Supplemental Table 3). This may indicate that diversity is favored and ROHs are not well tolerated in these regions.
The average Amish individual had autozygous segments spanning ~ $3.7\%$ of their genome. In comparison, individuals with European ancestry from the UK biobank had ~ $0.4\%$.11 When assessing ROHs with a cutoff of > 1 Mb, Amish participants harbored 18.6 ROH segments on average with an average length of 5949 Kb. In contrast, non-founder European populations harbored 8.02 segments on average with an average length of 1421 Kb. [ 25] We were not able to compare distributions of ROHs in Amish with UK biobank because subject level data was not available. However, the observation that ROHs are longer and more numerous in this founder population compared to outbred populations is not surprising given that longer shared haplotypes are inherited from recent common ancestors and reduced effective population size, due to a population bottlenecks, increases the number of ROH present. [ 26].
ROHs were widely distributed across the genome and were present on each autosome. The region with the highest frequency of SNPs in an ROH was identified in chromosome 5. The LCT gene in chromosome 2 had an enrichment of frequency of ROH which is consistent with previous studies. [ 22] Although an enrichment was also seen on chromosome 6, it was not localized to the MHC locus as seen previously. [ 25, 27] Across the genome, there were just six instances in which small clusters of SNPs were not included in any ROH segments. Five of these regions contain pseudogenes and noncoding RNA (Supplemental Table 3). This may indicate that diversity is well tolerated, and the regions are not under selective pressure. The sixth instance is along chromosome 14, in a region containing the gene OR11H12. *This* gene is a member of the olfactory receptor proteins and is known to have copy number variations. [ 28].
## Association of genome-wide ROH with health and disease-related traits
Measurements of FROH were then used as the primary predictor of phenotypic variation. We analyzed 96 traits including basic anthropometrics, blood pressure, fasting blood lipids, glucose, insulin, HbA1c, basic medical blood chemistry measurements, and medical histories. We did not identify any associations that withstood Bonferroni-correction ($p \leq 0.0005$), but the lead association ($$p \leq 0.0036$$, $b = 2.5$ × 10− 5, se = 8.3 × 10− 6) with genome-wide FROH was with electrocardiogram (EKG) QT interval, followed by serum levels of CO2 ($$p \leq 0.03$$, beta = 1.5 × 10− 6, se = 7.3 × 10− 7), urea nitrogen ($$p \leq 0.04$$, beta = 2.5 × 10− 6, se = 1.2 × 10− 6), and thyroid hormone measurement (tsh) ($$p \leq 0.04$$, beta = 4.6 × 10− 6, se = 2.3 × 10− 6). In a sex-stratified analysis, we did not identify any sex-specific effects of FROH.
## Association of regional autozygosity with health and disease-related traits
We analyzed 96 traits for associations with regional autozygosity levels and identified four trait associations at genome-wide significance following Bonferroni correction ($$p \leq 3.1$$ × 10− 9) (Fig. 4a-d). Increased serum bilirubin levels were significantly associated ($$p \leq 1$$ × 10− 43, beta = 0.19, se = 0.013) with increased FROH at a region of chromosome two that includes the UGT1A10 gene. As we [23] and others [24] have previously reported an association of a nearby SNP in UGT1A1, a gene encoding an enzyme that converts bilirubin into a more water-soluble form that the body is better able to excrete, we adjusted for the SNP as an additional fixed effect but found the ROH association to be essentially unchanged (beta = 0.19, $$p \leq 1.1$$ × 10− 43 vs. beta = 0.19, $$p \leq 1.7$$ × 10− 44). Three additional traits were associated with increased FROH, HbA1c ($$p \leq 8$$ × 10− 10, beta = 0.17, se = 0.027) with a region on chromosome eight surrounding the CHRNB3 gene, C-reactive protein ($$p \leq 2.7$$ × 10− 9, beta = 3.17, se = 0.53) with the intergenic region of FBXO33 and thyroid hormone levels ($$p \leq 2.8$$ × 10− 9, beta = 3.14, se = 0.53) with the intergenic region of LRRC3B (Supplemental Fig. 1a-b). To our knowledge, no genes previously linked to these traits are close to these signals.
Fig. 4Autozygosity mapping Manhattan and zoom plots Manhattan plot of (a) serum bilirubin levels and (b) HbA1c levels. Each dot represents a SNP. For statistical analysis, we extracted ROH inclusion for each SNP on the genotyping array Zoom plot of FROH and c) serum bilirubin levels and d) HbA1c levels. Red line indicates the significance threshold of $$P \leq 3.1$$ × 10− 9
## Recessive based SNP analysis
To test whether the regions associated with bilirubin and HbA1c identified through ROH analysis would be detected with other methods, we set up a recessive mode association analysis. We detected a strong association (lowest p value = 1 × 10− 464) between Bilirubin levels and SNPs within the same region that was detected with ROH analysis. HbA1c levels were not strongly associated ($p \leq 9.6$ × 10− 4) with any SNPs in the region when assessing using a recessive model.
## Association of genome-wide ROH with traits of interest
We have tested for associations between autozygosity, as measured by ROH, and a large panel of complex traits in a founder population of Amish individuals. We did not identify associations of genome-wide autozygosity with any traits that withstood Bonferroni-correction. In contrast, a large meta-analysis of 234 cohorts tested the association of genome-wide autozygosity levels with 100 traits and found significant associations with 6 cardiometabolic traits that were included in our analysis. [ 11] These cohorts were mostly non-founder but included the Lancaster Amish cohort from the present study, Hutterites, and several others with high consanguinity.
In the Amish, the lead association was between higher autozygosity measurements and QTc interval. Because the Amish study population is enriched for a known, high effect size, KCNQ1 pathogenic variant, p.Thr244Met, that increases QTc intervals, [29] we removed the KCNQ1 genomic region from the analysis and re-assessed the association. The association, although not statistically significant after Bonferroni correction, was still present and did not diminish.
## Association of regional autozygosity with traits of interest
Previously, ROH mapping has been used to detect loci with a segregating recessive variant. This method is regarded as more powerful than traditional single-marker association studies using a recessive model because there is more certainty of the haplotype on which the two alleles appear and greater viability at extreme minor allele frequencies. [ 9]GARLIC and other ROH prediction tools have been used for case-control studies and whole genome FROH associations with complex phenotypes. Here, we implemented GARLIC to map regional associations for complex trait discovery.
We detected associations of increased FROH with higher levels of serum bilirubin levels at a region on chromosome 2 with increased HbA1c at a region on chromosome 8. The mapping analysis also identified an association with cholesterol clustering around the APOB region on chromosome 2 which contains a high effect size variant p.Arg3527Gln. [ 18] This association was between increased LDL-C levels and decreased autozygosity. Therefore, heterozygotes in this region are driving the association.
## Strengths and limitations
Population isolates have an increased burden of ROH and can uncover genomic regions associated with complex phenotypes that may not have been identified in populations with more distant parental relatedness. Due to low levels of genome-wide homozygosity commonly seen in modern human populations, very large numbers of study subjects are required to provide sufficient statistical power. [ 16] Although no complex traits in our study were associated with genome-wide autozygosity levels, this could potentially be due to limited power given the sample size or because many examined phenotypes were cardio-metabolic which are risk factors for late-onset conditions and may not be under evolutionary selective pressure. [ 5] Also, genome-wide autozygosity measurements may not predict some polygenic traits influenced by dominant alleles at different loci, each influencing the trait in opposite directions. To identify genomic regions with dominant or additive alleles driving trait associations, GWAS are a better method than regional autozygosity mapping. For example, GWAS identified a statistically significant association between UGT1A10 and bilirubin levels ($$p \leq 2$$ × 10− 38 for UGT1A10 rs17854828 in 5830 subjects), although an association between HbA1c levels and chromosome 8 has not previously been reported.
The lead SNP from the recessive model may be too common (MAF = $41.7\%$) to exist on just one haplotype. Therefore, the signal may be diluted when using the ROH method. The causal SNP is found in a large fraction of the population but is not always present on the haplotype with the strongest effect. The SNPs in the region associated with HbA1c levels are rare and are only found in individuals that share a common haplotype.
## Summary
Amish have larger ROH segments and more of the Amish genome included in autozygous regions compared to outbred populations. Genome-wide summed autozygosity was not significantly associated with any of the 96 traits tested in this study. Using regional autozygosity mapping methods, we identified two traits associated with regional levels of autozygosity. We found that increased serum bilirubin levels were associated with increased autozygosity on chromosome two, localized to the UGT1A10 gene and that increased HbA1c levels were associated with increased autozygosity on chromosome eight, localized to the CHRNB3 gene.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Additional file 1: Supplemental Figure 1. Zoom plot showing associations between FROH and a) C-reactive protein and b) thyroid hormone. Red line indicates the significance threshold of $$P \leq 3.1$$ x 10-9.
Additional file 2: Supplemental Table 1. " Panel of complex traits measured in Amish participants that were tested for association with regional and global autozygosity." - this title is incorrect in the current proof.
Additional file 3: Supplemental Table 2. " SNPs with the highest frequency of inclusion in an ROH (frequency = 0.09 – 0.18). The SNPs that are shown were found on chromosome 2, 5, 6, 11, and 20." Additional file 4: Supplemental Table 3. " SNPs with the lowest frequency of inclusion in an ROH (frequency = 0.00 – 0.015)." Additional file 5.
## References
1. Lander E, Botstein D. **Homozygosity mapping: a way to map human recessive traits with the DNA of inbred children**. *Science* (1987.0) **236** 1567-70. DOI: 10.1126/science.2884728
2. Ismail J, Jafar TH, Jafary FH, White F, Faruqui AM, Chaturvedi N. **Risk factors for non-fatal myocardial infarction in young south asian adults**. *Heart* (2004.0) **90** 259-63. DOI: 10.1136/hrt.2003.013631
3. Christofidou P, Nelson CP, Nikpay M, Qu L, Li M, Loley C, Debiec R, Braund PS, Denniff M, Charchar FJ. **Runs of homozygosity: Association with Coronary Artery Disease and Gene expression in Monocytes and Macrophages**. *Am J Hum Genet* (2015.0) **97** 228-37. DOI: 10.1016/j.ajhg.2015.06.001
4. McQuillan R, Eklund N, Pirastu N, Kuningas M, McEvoy BP, Esko T, Corre T, Davies G, Kaakinen M, Lyytikäinen L-P. **Evidence of Inbreeding Depression on Human Height**. *PLoS Genet* (2012.0) **8** e1002655. DOI: 10.1371/journal.pgen.1002655
5. Joshi PK, Esko T, Mattsson H, Eklund N, Gandin I, Nutile T, Jackson AU, Schurmann C, Smith AV. **Directional dominance on stature and cognition in diverse human populations**. *Nature* (2015.0) **523** 459-62. DOI: 10.1038/nature14618
6. Kapurubandara S, Melov S, Shalou E, Alahakoon I. **Consanguinity and associated perinatal outcomes, including stillbirth**. *Aust N Z J Obstet Gynaecol* (2016.0) **56** 599-604. DOI: 10.1111/ajo.12493
7. Assaf S, Khawaja M, DeJong J, Mahfoud Z, Yunis K. **Consanguinity and reproductive wastage in the Palestinian Territories**. *Paediatr Perinat Epidemiol* (2009.0) **23** 107-15. DOI: 10.1111/j.1365-3016.2008.00988.x
8. Yengo L, Zhu Z, Wray NR, Weir BS, Yang J, Robinson MR, Visscher PM. **Detection and quantification of inbreeding depression for complex traits from SNP data**. *Proc Natl Acad Sci USA* (2017.0) **114** 8602-7. DOI: 10.1073/pnas.1621096114
9. McLaughlin RL, Kenna KP, Vajda A, Heverin M, Byrne S, Donaghy CG, Cronin S, Bradley DG, Hardiman O. **Homozygosity mapping in an irish ALS case–control cohort describes local demographic phenomena and points towards potential recessive risk loci**. *Genomics* (2015.0) **105** 237-41. DOI: 10.1016/j.ygeno.2015.01.002
10. Wang S, Haynes C, Barany F, Ott J. **Genome-wide autozygosity mapping in human populations**. *Genet Epidemiol* (2009.0) **33** 172-80. DOI: 10.1002/gepi.20344
11. 11.Clark DW. Associations of autozygosity with a broad range of human phenotypes. 17.
12. Szpiech ZA, Xu J, Pemberton TJ, Peng W, Zöllner S, Rosenberg NA, Li JZ. **Long runs of homozygosity are enriched for deleterious variation**. *Am J Hum Genet* (2013.0) **93** 90-102. DOI: 10.1016/j.ajhg.2013.05.003
13. He S, Ryan KA, Streeten EA, McArdle PF, Daue M, Trubiano D, Rohrer Y, Donnelly P, Drolet M, Newcomer S. **Prevalence, control, and treatment of diabetes, hypertension, and high cholesterol in the amish**. *BMJ Open Diab Res Care* (2020.0) **8** e000912. DOI: 10.1136/bmjdrc-2019-000912
14. 14.Xu H, Ryan KA, Jaworek TJ, Southam L, Reid JG, Overton JD, Baras A, Puurunen MK, Zeggini E, Taylor SI et al. (2017). Familial Hypercholesterolemia and Type 2 Diabetes in the Old Order Amish. Diabetes 66, 2054–2058.
15. 15.Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. (2015). Second-generation PLINK: rising to the challenge of larger and richer datasets. GigaSci 4, 7.
16. McQuillan R, Leutenegger A-L, Abdel-Rahman R, Franklin CS, Pericic M, Barac-Lauc L, Smolej-Narancic N, Janicijevic B, Polasek O, Tenesa A. **Runs of homozygosity in european populations**. *Am J Hum Genet* (2008.0) **83** 359-72. DOI: 10.1016/j.ajhg.2008.08.007
17. 17.Szpiech ZA, Blant A, Pemberton TJ. (2017). GARLIC: Genomic Autozygosity Regions Likelihood-based Inference and Classification. Bioinformatics 33, 2059–2062.
18. Shen H, Damcott CM, Rampersaud E, Pollin TI, Horenstein RB, McArdle PF, Peyser PA, Bielak LF, Post WS, Chang Y-PC. **Familial defective apolipoprotein B-100 and increased low-density lipoprotein cholesterol and coronary artery calcification in the old order amish**. *Arch Intern Med* (2010.0) **170** 1850-5. DOI: 10.1001/archinternmed.2010.384
19. Pollin TI, Damcott CM, Shen H, Ott SH, Shelton J, Horenstein RB, Post W, McLenithan JC, Bielak LF, Peyser PA. **A null mutation in human APOC3 confers a favorable plasma lipid profile and apparent cardioprotection**. *Science* (2008.0) **322** 1702-5. DOI: 10.1126/science.1161524
20. 20.O’Connell J. Mixed Models for Pedigree Analysis (MMAP) software program to account for the relatedness among study subjects. Available from http://edn.som.umaryland.edu/mmap/index.php).
21. Nothnagel M, Lu TT, Kayser M, Krawczak M. **Genomic and geographic distribution of SNP-defined runs of homozygosity in Europeans**. *Hum Mol Genet* (2010.0) **19** 2927-35. DOI: 10.1093/hmg/ddq198
22. Beja-Pereira A, Luikart G, England PR, Bradley DG, Jann OC, Bertorelle G, Chamberlain AT, Nunes TP, Metodiev S, Ferrand N. **Gene-culture coevolution between cattle milk protein genes and human lactase genes**. *Nat Genet* (2003.0) **35** 311-3. DOI: 10.1038/ng1263
23. McArdle PF, Whitcomb BW, Tanner K, Mitchell BD, Shuldiner AR, Parsa A. **Association between bilirubin and cardiovascular disease risk factors: using mendelian randomization to assess causal inference**. *BMC Cardiovasc Disord* (2012.0) **12** 16. DOI: 10.1186/1471-2261-12-16
24. Lin J-P, O’Donnell CJ, Schwaiger JP, Cupples LA, Lingenhel A, Hunt SC, Yang S, Kronenberg F. **Association between the UGT1A1*28 allele, bilirubin levels, and coronary heart disease in the Framingham Heart Study**. *Circulation* (2006.0) **114** 1476-81. DOI: 10.1161/CIRCULATIONAHA.106.633206
25. Howrigan DP, Simonson MA, Davies G, Harris SE, Tenesa A, Starr JM, Liewald DC, Deary IJ, McRae A, Wright MJ. **Genome-wide autozygosity is associated with lower general cognitive ability**. *Mol Psychiatry* (2016.0) **21** 837-43. DOI: 10.1038/mp.2015.120
26. Ceballos FC, Joshi PK, Clark DW, Ramsay M, Wilson JF. **Runs of homozygosity: windows into population history and trait architecture**. *Nat Rev Genet* (2018.0) **19** 220-34. DOI: 10.1038/nrg.2017.109
27. Mukherjee S, Guha S, Ikeda M, Iwata N, Malhotra AK, Pe’er I, Darvasi A, Lencz T. **Excess of homozygosity in the major histocompatibility complex in schizophrenia**. *Hum Mol Genet* (2014.0) **23** 6088-95. DOI: 10.1093/hmg/ddu308
28. Sheth F, Ewers E, Kosyakova N, Weise A, Sheth J, Desai M, Andrieux J, Vermeesch J, Hamid AB, Ziegler M. **A small supernumerary marker chromosome present in a Turner syndrome patient not derived from X- or Y-chromosome: a case report**. *Mol Cytogenet* (2009.0) **2** 22. DOI: 10.1186/1755-8166-2-22
29. Streeten EA, See VY, Jeng LBJ, Maloney KA, Lynch M, Glazer AM, Yang T, Roden D, Pollin TI, Daue M. **KCNQ1 and long QT syndrome in 1/45 amish: the Road from identification to implementation of culturally appropriate Precision Medicine**. *Circ Genom Precis Med* (2020.0) **13** e003133. DOI: 10.1161/CIRCGEN.120.003133
|
---
title: Identification of kukoamine a as an anti-osteoporosis drug target using network
pharmacology and experiment verification
authors:
- Liying Luo
- Zhiyuan Guan
- Xiao Jin
- Zhiqiang Guan
- Yanyun Jiang
journal: Molecular Medicine
year: 2023
pmcid: PMC10029210
doi: 10.1186/s10020-023-00625-6
license: CC BY 4.0
---
# Identification of kukoamine a as an anti-osteoporosis drug target using network pharmacology and experiment verification
## Abstract
### Background
Osteoporosis (OP) is a major and growing public health problem characterized by decreased bone mineral density and destroyed bone microarchitecture. Previous studies found that Lycium Chinense Mill (LC) has a potent role in inhibiting bone loss. Kukoamine A (KuA), a bioactive compound extract from LC was responsible for the anti-osteoporosis effect. This study aimed to investigate the anti-osteoporosis effect of KuA isolated from LC in treating OP and its potential molecular mechanism.
### Method
In this study, network pharmacology and molecular docking were investigated firstly to find the active ingredients of LC such as KuA, and the target genes of OP by the TCMSP platform. The LC-OP-potential *Target* gene network was constructed by the STRING database and network maps were built by Cytoscape software. And then, the anti-osteoporotic effect of KuA in OVX-induced osteoporosis mice and MC3T3-E1 cell lines were investigated and the potential molecular mechanism including inflammation level, cell apoptosis, and oxidative stress was analyzed by dual-energy X-ray absorptiometry (DXA), micro-CT, ELISA, RT-PCR, and Western Blotting.
### Result
A total of 22 active compounds were screened, and we found KuA was identified as the highest active ingredient. Glycogen Phosphorylase (PYGM) was the target gene associated with a maximum number of active ingredients of LC and regulated KuA. In vivo, KuA treatment significantly increased the bone mineral density and improve bone microarchitecture for example increased BV/TV, Tb. N and Tb. Th but reduced Tb. Sp in tibia and lumber 4. Furthermore, KuA increased mRNA expression of osteoblastic differentiation-related genes in OVX mice and protects against OVX-induced cell apoptosis, oxidative stress level and inflammation level. In vitro, KuA significantly improves osteogenic differentiation and mineralization in cells experiment. In addition, KuA also attenuated inflammation levels, cell apoptosis, and oxidative stress level.
### Conclusion
The results suggest that KuA could protect against the development of OP in osteoblast cells and ovariectomized OP model mice and these found to provide a better understanding of the pharmacological activities of KuA again bone loss.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s10020-023-00625-6.
## Introduction
Osteoporosis (OP) is a major and growing public health problem characterized by decreased bone mineral density and destroyed bone microarchitecture (Silverstein et al. 2021). With the inevitable consequence of aging, osteoporotic fractures are becoming more and more common in women over 55 and men over 65, causing huge costs in mortality and health care (Compston et al. 2019). Ironically, despite great advances in the treatment of OP, treatment gaps varied widely for patients at high risk for OP fractures both between and within countries (Khosla and Hofbauer 2017). Considering the complicated mechanisms and different treatment effects of OP, a large and growing body of literature showed that anti-osteoporosis targets have become a promising research field in recent years.
In China, many traditional Chinese medicines are used to treat diabetes and have been proven to be effective. For example, Lycium chinense Mill. ( Lycii Cortex, LC) has great potential in preventing diabetes and glucocorticoid‑induced bone loss (Lee et al. 2021; Park et al. 2019a). However, relatively little information is available on the properties of potential hypoglycemic compounds of LC (Liu et al. 2021). Kukoamine A (KuA), a sperm kaloid, is a critical bioactive component extracted from the root bark of LC. KuA has several pharmacological effects such as anti-inflammatory, anti-pain, antibacterial, neuroprotective, autoimmune enhancing, and hypotensive effects (Hadjipavlou-Litina et al. 2009). Other bioactive component such as betaine, scoplin, Kukoamine B extracted from LC also showed significant effectiveness in the treatment of OP (Lee et al. 2021; Park et al. 2019a, 2020; Yajun et al. 2021). However, a systematic investigation of the main bioactive components of LC that contributed to OP remains unexplored.
To date, network pharmacology integrates pharmacological, bioinformatics, and other scientific analyses into a systematic network and interprets the therapeutic mechanisms of different drug components and the targets of gene delivery. Network pharmacology is a promising approach for understanding multicomponent drug systems such as Traditional Chinese Medicine (TCM) formulae (Li and Zhang 2013). By analyzing the components and targets of diseases, we can provide biological processes and pathways that TCM may play a role in, helping us to analyze the mechanism of TCM treating diseases. Molecular docking is a drug development method that mimics the interaction between receptors and drugs. In recent years, the use of molecular docking to elucidate the appropriate mechanism has become a global trend in drug development (Zhou et al. 2021a). Molecular function and signaling pathways by constructing a “disease-phenotype-genetic’ network can suitably interpret the relationship among different bioactive components in traditional Chinese medicine (Wei et al. 2020; Zhang et al. 2019a).
The current study aimed to investigate the relationship between the potential bioactive components in LC with OP. In this regard, we firstly used network pharmacology to analyze the effective ingredients of LC, then screened the bioactive ingredients in the treatment of OP. These selected targets were evaluated by pathways of action in functional enrichment pharmacology, genetic selection (GO), biological pathways (KEGG), and molecular docking technology (Jiang et al. 2019). Finally, the ovariectomized OP mice with different doses of KuA and gene silencing experiments at the cell level further verified the results of network pharmacologic analysis. This study provided a theoretical basis for investigating the molecular mechanism of LC against OP. The workflow is shown in Additional file 1: Figure S1.
## Ovariectomized OP model mice with different doses of KuA
The experimental protocol was approved by the Department of Laboratory Animal Science of the Shanghai Tenth People’s Hospital of Tongji University (SHDSYY-2021–6420, data:2021.5.6). A total of forty-five C57BL/6N female mice (8 weeks) were housed in the same animal room with a controlled temperature (22 °C) and light cycle (12 h light, 12 h dark) with free access to fresh water and food. The mice were divided randomly into five groups ($$n = 9$$): [1] Sham ($$n = 9$$), [2] Ovariectomized (OVX, $$n = 9$$), [3] OVX + 5 mg/kg/day of KuA (KuA5, $$n = 9$$), [4] OVX + 10 mg/kg/day of KuA (KuA10, $$n = 9$$), [5] OVX + 20 mg/kg/day of KuA (KuA20, $$n = 9$$). The mice treated with osteoporotic intervention measures were administered one week after surgery (Zheng et al. 2020) during 11 weeks of administration. The bone mineral density (BMD) of the right tibia and spine was measured at 0, 6, and 12 weeks respectively after ovariectomy (0 weeks is the time of ovariectomy). All operations were performed to minimize animal suffering and reduce the number of mice used. KuA (purity ≧$98\%$, Liaoning University, Shenyang, China) was dissolved in DMSO and administered by the intravenous route as in previous studies (Liu et al. 2017). All related reagents were of analytical or pharmaceutical grade.
Each mouse used in these studies was euthanized with pentobarbital (50 mg/kg, intraperitoneal injection). Meanwhile, the protocol of mice sham surgery or bilateral oophorectomy was described as before (Inada et al. 2011). All mice underwent intraperitoneal anesthesia with the injection of pentobarbital (50 mg/kg), and the sham group was exposed to both sides of the ovary and raised the fatty tissue around the ovary, leaving the ovary intact, but bilateral ovariectomy for the other eighteen mice performed under the premise of complete ovarian exposure resection. After the surgery, intraperitoneal injection of penicillin was used to prevent infection twice a day for two days.
## Dual-energy X-ray absorptiometry (DXA) and Micro-CT analysis
At the end of 12 weeks after treatment, Lumbar spine (L4) and right tibia bone mineral density (BMD) were determined in small animals using a high-resolution soft X-ray collimator (Faxitron® LX-60 Cabinet radiography system, US). The sample was then placed on the base of the PBS scanner using μ-CT (Inveon, Siemens, Erlangen, Germany) at a spatial resolution of 55 kVp, 145 uA, integration time 300 ms, 720 views, 20 mm voxel resolution. The region of interest (ROI) was 0.36 to 2.1 mm from the right proximal epiphyseal growth plate of the tibia, 1.5 mm long, and 0.5 mm from the L4 proximal growth plate. We analyzed trabecular and cortical bone microarchitecture by measuring bone volume (BV) over total volume (TV), trabecular thickness (Tb. Th), trabecular number (Tb. N), trabecular spacing (Tb. Sp) in the medial tibial trabecular bone and lumbar vertebrae, the total cross‐sectional area inside the periosteal envelope (Tt. Ar), cortical bone area (Ct. Ar), bone marrow area (Ma. Ar), average cortical thickness (Ct. Th) and cortical area fraction (Ct. Ar/Tt. Ar) in the medial tibial cortical bone which has been describing in previous studies (Kalyanaraman et al. 2017; Lei et al. 2018). μ-CT did not analyze surrounding osteophytes.
## Biomechanical analysis
Following radiographic measurements, three-point flexion testing was performed on these right tibias using a mechanical testing system (Landmark, MTS, Inc., Eden Prairie, MN) to determine the mechanical properties. The main support section was 9 mm and the load range was 5 mm. The tibia was placed in a bracket with a stretched medial surface, and the distal portion of the tibiofibular junction was placed directly into the leftmost fixation device. Each tibia should be loaded at 0.01 mm/s until rupture and the load and displacement recorded. Data was automatically recorded by the material testing device. According to the load–displacement curve, the biomechanical properties were evaluated to analyze compressive maximum load [a measure of the maximum force that the sample tibia withstood before fracture (N)], stiffness [the slope on the linear portion of the load-deformation curve related to the tibia's flexural rigidity (N/mm)], displacement of maximum force [a measure of the maximum displacement that the sample tibia withstood before fracture (mm)] and the energy of maximum force (area under the linear portion of the load-deformation curve (mJ) [the slope on the linear portion of the load-deformation curve related to the tibia's flexural rigidity (N/mm)].
## Western Blotting
To analyze the oxidative stress level, we extracted mitochondrial and cytosolic proteins. The mitochondria were isolated from bone tissue using a Mitochondria Isolation Kit (QuadroMACS 130–094-532) according to the manufacturer’s instructions. Other proteins were extracted from bone tissue, and quantitated with a protein assay kit (Bio-Rad, Mississauga, Ontario, Canada). Protein samples (15 µg) were fractionated by SDS-PAGE and transferred to nitrocellulose membranes. Protein concentration was quantified using the BCA Reagent (Thermo Scientific, XH351428).
We performed western blotting analysis with rabbit anti-caspase-3 (1:1000, caspase-3), rabbit anti-cytochrome c (1:500), rabbit anti-Bax (1:1000), and rabbit anti-Bcl-2 (1:1000). The protein load of each channel was detected using an anti-GAPDH antibody (1:5000) and beta-actin antibody (1:2000). Goat anti-rabbit or anti-mouse secondary antibodies (1:12,000) were used before chemiluminescent detection. Immunoblots were visualized using BeyoECL Plus. Results were expressed as a percentage of control.
## ELISA analysis
We collected serum in coagulation tubes and centrifuge (3000 rpm, 15 min), and Collected the plasma and stored it at − 80 °C. Determination of serum osteocalcin (OCN) (Meimian Biotechnology, Yancheng, Jiangsu, China), Tartrated Resistant Acid Phosphatse (TRAP) (Meimian Biotechnology, Yancheng, Jiangsu, China), C-terminal telopeptide II (CTX-II) (Meimian Biotechnology, Yancheng, Jiangsu, China), A Lkaline Phosphatase (ALP) (Meimian Biotechnology, Yancheng, Jiangsu, China), Procollagen I Intact N-Terminal (PINP) (Meimian Biotechnology, Yancheng, Jiangsu, China), Interleukin-6 (IL-6) (Meimian Biotechnology, Yancheng, Jiangsu, China), C-reactive protein (CRP) (Meimian Biotechnology, Yancheng, Jiangsu, China), tumor necrosis factor -α (TNF-α) (Meimian Biotechnology, Yancheng, Jiangsu, China), and Interleukin-1β (IL-1β) (Meimian Biotechnology, Yancheng, Jiangsu, China) were performed using commercial enzyme-linked immunosorbent assay (ELISA) kits.
## Osteoblast cells experiment
Preosteoblast MC3T3-E1 cells were cultured overnight in 48-well plates and treated with 50 g/ml ascorbic acid and 10 mM α-glycerophosphate cells for 3 weeks with or without KuA (5, 10, and 20 M). Cells were fixed with cold $70\%$ ethanol for 10 min at room temperature, then rinsed with water. Calcium precipitation in mineralized cells was determined by staining with Alizarin Red S (Sigma-Aldrich). Alizarin S red staining was positive under light microscopy. To quantify this, cells were extracted with $10\%$ cetylpyridinium chloride for 1 h and seeded into 96-well plates. We measured the absorbance of the extract at 550 nm (BIO-RAD; Hercules, CA, USA) (Park et al. 2019b).
## RT-PCR analysis and oxidative stress
Muscle and connective tissue of the distal left tibia were washed, frozen in liquid nitrogen, and stored at − 80 °C. Frozen tibiae were sprayed with a Bessman tissue sprayer under liquid nitrogen (Spectrum Laboratories, Rancho Dominguez, CA, USA). Total RNA was extracted using Trizol reagent (Invitrogen, Carlsbad, CA, USA). The expression levels of bone metabolism and inflammation-related genes, including OCN, RANKL, OPG, IL-6, and Osterix (Additional file 1: Table S1). The relative change in gene expression was analyzed by the 2-ΔΔCT method.
The levels of MDA, H2O2, cytochrome, and the activities of MnSOD and CuZnSOD were measured using commercially available kits (CAK) according to the manufacturer’s instructions.
## Transient transfections of siRNA molecules
The transient transfection of siRNA molecules PYGM (Life Technologies) was performed using RNAiMAX reagent as instructed by the manufacturer (Life Technologies). Briefly, MC3T3-E1 cells were transfected into 6-well plates containing 10 nM siRNA molecule or RNAiMAX shuffling control. Cells were then maintained in $2\%$ horse serum, differentiated for 3 days, and 1 ml Tri reagent was collected from three RNA extraction wells or 1 ml RIPA buffer (Thermo Scientific) (containing Halt Protease and Phosphatase Inhibitor Cocktail; Thermo Scientific) for RT-PCR analysis (Myers et al. 2013).
The increased differentiation of osteoblasts is closely related to the high expression of the main osteoblast marker gene ALP, OCN, and Osterix. Osteoblastic MC3T3-E1 cell line treated with 20 μM of KuA significantly increased the expression of OCN and Osterix compared to the OVX group. In addition, the PYGM mRNA level was significantly lower in the OVX group than in the control group and after treatment with KuA, the PYGM mRNA level decreased significantly than OVX group (Fig. 9E).
To further explore the PYGM gene functions in osteoblast differentiation, transient transfections of PYGM siRNA molecules were investigated and we found that KuA reduced the PYGM mRNA expression level. After transfection of PYGM siRNA, the OCN and Osterix mRNA levels decreased significantly than the control group. In addition, after treatment with KuA, we found that the KuA with PYGM siRNA group decreased the OCN, and Osterix mRNA levels significantly than the PYGM siRNA group (Fig. 9E).
Finally, to further clarify the influence of KuA intervention on bone loss, we further verified that KuA intervention would significantly improve the progress by principal cause analysis.
## Bioactive ingredients and target genes of LC
Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP, https://tcmspw.com/index.php) was used to analyze the ingredients of LC such as oral bioavailability (OB), drug-likeness (DL), intestinal epithelial permeability, blood–brain barrier penetrability, and water solubility. After screening the TCMSP dataset, 22 bioactive ingredients were obtained (Ru et al. 2014). In preparation for molecular adaptation, constituent SDF files of 13 active molecules (http://pubchem.ncbi.nlm.nih.gov) were downloaded from the PubChem database after calculation in Chemdraw 3D Ultra software.
## Differential Genes analysis of OP
*The* genetic samples (GSM1369766) of patients with low BMD and high BMD were obtained from the GEO dataset. The ‘limma’ package was installed in Perl and the sample values were patched and converted to log2 (logFC). Samples with P-value < 0.005 and ∣log2 fold change∣ > 1 were considered to have statistically significant and selected as differential genes. Then we created a volcano gene map from the sample and selected the top 20 most important up-down corrected genes for the heatmap.
## Protein–Protein Interaction (PPI) Network
In a PPI network, the concentration (DC) of each node is the number of edges per node. The higher the degree, the higher the center position of the node. The relay center (BC) receives the location of the node among other nodes. Specifically, it is the ratio of the number of shortest paths through this node to the total number of shortest paths in the network. DC and BC reflect the influence of individual nodes on the entire network. They describe topological centrality in terms of network connectivity and controllability.
The ‘biogenetic, cytoNAC’ package was installed in Cytoscape 3.8.0 and was used to enter the crossover gene and select the ‘Homo sapiens’ parameter. Data for constructing the PPI network were sourced from six main experimental research databases: Human Protein Reference Database, Biological General Repository for Interaction Datasets, Database of Interacting Proteins, IntAct molecular interaction database, Molecular INTeraction Database, and Biomolecular Interaction Network Database. We selected this method “input nodes and its neighbors” to obtain a PPI network and performed a topology analysis based on the central location of the network.
## LC-OP-Potential target genes network
LC-related target genes were selected from the TCMSP database based on chemical similarity and pharmacophore models. We calibrated LC-related target names to default names using the UniProt database (https://www.uniprot.org/) (Szklarczyk et al. 2016). The TCMSP formula was adapted using Cytoscape web page generation software.
## GO and KEGG enrichment
Go enrichment analysis examines gene function at three levels: biological process (BP), cellular component (CC), and molecular function (MF). BP mainly involves aspects of response to a steroid hormone, response to oxygen levels, and regulation of lipid metabolic process. CC is mainly related to the integral component of the postsynaptic membrane, GABA receptor complex, and GABA-A receptor complex. MF is remarkably linked with neurotransmitter receptor activity, steroid hormone receptor activity, and GABA-A receptor activity.
Firstly, we changed the names of the potential target genes from R-package (org.Hs.eg.db, version 3.8) to entrezID which helps to exclude errors caused by capitalization or abbreviations of the target names. Then, we used the R-package ‘DOSE’, ‘cluster profile’, and ‘pathview’ to visualize the biological functions of graphene oxide and analyze the enrichment of the KEGG pathway for which the p-value was < 0.05 for further analysis.
## Molecular Docking analysis
We used CB-Dock Internet Molecular Docking Technology to select active components of potential LC-OP target gene networks and Docked with PYGM receptors (http://cao.labshare.cn/cbdock//) (Liu et al. 2020). PYGM (protein ID is 6y5o) and the active ingredients were uploaded to the CB-Dock website. After determining the coordinates of the docking pocket, molecular docking and conformational assessment were performed using the CB docking station. The lower the VINA score, the more stable the ligand binding. Finally, receptors are screened for the binding activity of compounds to the target.
## Statistical analysis
All measurements were presented as the mean ± standard deviation (SD) and a P-value of ≤ 0.05 was considered statistically significant. The bodyweight of the time-course study was analyzed by two-way repeated-measures analysis of variance (ANOVA). Data were analyzed for the main effect and timing of the intervention. One-way ANOVA followed by Tukey multiple analysis was performed using GraphPad Prism 8.02 (La Jolla, CA, USA).
## KuA administration protects against OVX-induced bone loss
KuA isolated from LC extracts was identified by magnetic resonance imaging (NMR) and mass spectrometry (Fig. 1A). As expected, the OVX mice showed a significant reduction of BMD in lumber 4 and right tibia at 6 weeks and 12 weeks after surgery. KuA 10 mg and 20 mg administration inhibited the reduction of BMD at 12 weeks (Fig. 1B, C, Additional file 1: Figure S2).Fig. 1The experimental result show that KuA significantly increased the bone mineral density of the spine and tibia in OVX mice. A KuA spectrometry analyses and total experiment flowchart. B Representative figure of bone mineral density in mice. C BMD of L4. D BMD of the tibia. These results found that KuA improves the bone mass of the tibia and spine in ovariectomized mice. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$, ****$P \leq 0.0001.$ KuA: Kukoamine A; OVX: ovariotomy; L4: Lumber 4; BMD: bone mineral density The bone microarchitecture of the right tibia analysis revealed that KuA administration with 5 mg, 10 mg, and 20 mg prevented the tibia-BV/TV than the OVX group at 12 weeks after treatment. Tibia-Tb. N also showed a remarkable improvement by KuA 10 mg compared with the OVX group at 6 weeks and 12 weeks. However, at 12 weeks after treatment, KuA 10 mg can decrease Tibia-Tb. Sp significantly more than the OVX group. In addition, KuA 5 mg, 10 mg, and 20 mg can also increase tibia-Tb. Th significantly than the OVX group 12 weeks after treatment (Fig. 2A–E, Additional file 1: Figure S3).Fig. 2KuA significantly improved the tibia bone microstructure and mechanical properties in OVX mice. A Representative figure in the tibia. B BV/TV of the tibia. C Tb. N of tibia. D Tb. Sp of the tibia. E Tb. Th of the tibia. These results found that KUA improved the bone microstructure of the tibia in ovariectomized mice. F, G *The maximum* force of the tibia. Mechanical results showed that KUA improve the maximum stress of the tibia in ovariectomized mice. The bar is 0.7 mm. * $P \leq 0.05$, **$P \leq 0.01$,***$P \leq 0.001$,****$P \leq 0.0001.$ KuA: Kukoamine A; OVX: ovariotomy; L4: Lumber 4; BMD: bone mineral density. BV/TV: bone volume over total volume; Tb. Th: trabecular thickness; Tb. N: trabecular number, Tb. Sp: trabecular spacing Lumber 4 microarchitecture was also shown in Fig. 3A. For spine Tb. N and BV/TV, KuA 5 mg, 10 mg, and 20 mg improved significantly than the OVX group at 12 weeks, but for spine Tb. Sp and Tb. Th, KuA showed no significant difference in the OVX group at 12 weeks after treatment (Fig. 3B–E).Fig. 3KuA significantly increased the bone microstructure of the spine in OVX mice. A Representative figure in L4. B BV/TV of L4. C Tb. Sp of L 4. D Tb. N of L 4. E Tb. Th of L4. These results found that KuA improved the bone microstructure of the spine in ovariectomized mice. The bar is 0.3 mm. * $P \leq 0.05$, **$P \leq 0.01$,***$P \leq 0.001$,****$P \leq 0.0001.$ KuA: Kukoamine A; OVX: ovariotomy; L4: Lumber 4; BMD: bone mineral density. BV/TV: bone volume over total volume; Tb. Th: trabecular thickness; Tb. N: trabecular number, Tb. Sp: trabecular spacing As shown in Table 1, there was a significant reduction in Ct. Th, Tt. Ar and Ct. Ar/Tt. Ar in the OVX group compared to the control. After KuA intervention, Ct. Th improved significantly in the KuA 20 group than the OVX group. The KuA 10 and KuA 20 groups improved significantly relative to the OVX group in Ct. Ar/Tt. Ar. Table 1Changes in serological markers and cortical bone indicators in KuA-treatment osteoporotic miceCharacteristicsGroupsControlOVXKuA5KuA10KuA20Bone turnover biomarkersOCN (ng/ml)3.22 ± 0.25*^&1.16 ± 0.151.65 ± 0.21**#2.36 ± 0.24**#^3.62 ± 0.28**^&TRAP (ng/ml)10.62 ± 0.698*12.64 ± 1.5911.96 ± 1.85#10.12 ± 2.61*10.06 ± 1.82*CTX-II (ng/ml)0.081 ± 0.001*0.086 ± 0.0020.085 ± 0.0030.084 ± 0.0010.084 ± 0.002*ALP (ng/ml)1.26 ± 0.26*^&0.88 ± 0.140.95 ± 0.08#0.99 ± 0.21#1.11 ± 0.19*^PINP (ng/ml)0.86 ± 0.11*0.61 ± 0.090.67 ± 0.03#0.72 ± 0.120.78 ± 0.15^Cortical bone indicatorsCt. Th (mm)0.14 ± 0.06*0.11 ± 0.030.12 ± 0.05#0.12 ± 0.04#0.13 ± 0.02*Tt. Ar (mm2)1.35 ± 0.36*1.62 ± 0.151.58 ± 0.231.54 ± 0.131.42 ± 0.42*^Ct. Ar (mm2)0.68 ± 0.030.72 ± 0.080.69 ± 0.040.69 ± 0.070.68 ± 0.05Ct. Ar/Tt. Ar (%)62.88 ± 11.62*42.69 ± 12.9551.67 ± 12.86#58.61 ± 14.62*62.59 ± 21.61*^&#compare with the control group. * compare with OVX group. ^ compare with KA5 group. & compare with KA10 group. Tt. Ar, the total cross‐sectional area inside the periosteal envelope; Ct. Ar, cortical bone area; Ct. Th, average cortical thickness; Ct. Ar/Tt. Ar, cortical area fraction; KuA5, OVX + 5 mg/kg/day of KuA; KuA10, OVX + 10 mg/kg/day of KuA; KuA20, OVX + 20 mg/kg/day of KuA; OCN, osteocalcin; ALP, Alkaline Phosphatase; TRAP: Triiodothyronine Receptor Auxiliary Protein; CTX-II: Collagen Type II Alpha 1 Chain; PINP: Procollagen I N-Terminal Propeptide; Tt. Ar: total cross‐sectional area inside the periosteal envelope; Ct. Ar: cortical bone area; Ma. Ar: bone marrow area; Ct. Th: average cortical thickness; Ct. Ar/Tt. Ar:cortical area fraction
For bone turnover biomarkers, KuA treatment can improve the osteogenesis-related indicators including serum OCN and ALP but decrease osteoclast-related indicators TRAP and CTX-II compared with the OVX group (Table 1).
Biomechanical analysis of the right tibia at 12 weeks after treatment was shown in Fig. 6F, KuA 5 mg, 10 mg, and 20 mg administration can significantly improve the maximum force, stiffness, and displacement of than OVX group (Fig. 6G, Additional file 1: Figure S4A, B). however, only KuA 5 mg administration can improve the energy absorption of than OVX group (Additional file 1: Figure S4).
## Identification of KuA as an osteogenic suppressor
LC contains 22 active ingredients totally in the TCMSP dataset. The target genes were then screened from these active ingredients of LC by DrugBank dataset as the conditions. There were 321 target genes obtained for LC active ingredients. The removal of duplicates after verification yielded 242 target genes (Table 2).Table 2The total available compounds of Lycium chinense MillIndexMol IDMolecule nameMWAlogPHdonHaccOB (%)Caco-2BBBDLFASA − HL9MOL002224Aurantiamide acetate444.574.5262658.380790.40909 − 0.221620.588384.57.0359855MOL002218Scopolin354.34 − 0.2884956.44689 − 1.05198 − 1.751180.38714138.823.0065831MOL001552OIN289.411.7211445.970580.431430.091930.1931749.774.46949511MOL002228Kulactone452.746.2330345.438080.855910.16070.8157843.375.52143516MOL000449Stigmasterol412.777.641143.829851.444581.000450.7566520.235.5745952MOL001645Linoleyl acetate308.566.8470242.100771.358261.084130.1984526.37.4785217MOL002221Kukoamine A530.741.7881042.0846 − 0.20726 − 2.115760.56398163.1804MOL001790Linarin592.6 − 0.17871439.84373 − 1.68135 − 2.765210.70925217.9716.0677822MOL000953CLR386.737.3761137.87391.431011.126780.6767720.234.51883414MOL000296Hederagenin414.798.0841136.913911.318760.964280.7507220.235.34751115MOL000358Beta-sitosterol414.798.0841136.913911.324630.985880.7512320.235.3554918MOL002222Sugiol300.484.9871236.113531.140540.699220.2764837.314.619943MOL001689Acacetin284.282.5852534.973570.67146 − 0.046890.2408279.917.248476MOL002219Atropine289.411.9962434.527890.15341 − 0.296610.2141760.773.11774618MOL000472Emodin270.252.4923524.398320.22289 − 0.660960.2391694.83020MOL000008Apigenin270.252.3343523.062160.4256 − 0.61090.2130690.9019MOL000476Physcion284.282.7432522.28640.52191 − 0.402290.2665983.83013MOL000295Alexandrin576.956.3374620.63194 − 0.1993 − 0.809710.6269799.38010MOL002226Lyciumin A874.02 − 0.238102010.07929 − 2.51813 − 2.663370.1977306.7021MOL000880Tricosane324.7110.864008.330481.847181.67990.20880012MOL002229HEPTACOSANE380.8312.689008.180711.879421.800440.361550017MOL000458Campesterol400.767.972115.5686131.596651.408470.7157720.230 By comparing 30 low BMD samples and 30 high BMD samples in the GEO database. A total of 12,548 differential genes were required, including 7167 up-regulated genes and 5381 down-regulated genes. After screening with a P-value < 0.005 and ∣log2 (fold change)∣ > 1, the gene volcano map was analyzed in Fig. 4A, B. PCA analysis, UMAP, and heatmap analysis were shown in Fig. 1C–E. The differential genes in the disease samples were normally distributed, and the number of up-regulated genes was greater than the number of down-regulated genes. Table 3 listed the 20 most important up-and-down-regulated genes. Using Venny 2.1 software to hybridize OP target genes and LC target genes, 24 potential target genes were obtained, as shown in Fig. 5A. We found that PYGM has the highest LogFC value (Table 4).Fig. 4The gene expression analyses in osteoporosis samples from the GEO dataset. A Gene normalization diagram shows this dataset has a good consistency. B Gene volcano map. C *Principal* genetic analysis. D Uniform Manifold Approximation and Projection. D *Heatmap analysis* found that the top 20 genes expressed in GEO dataset. UMAP: Uniform Manifold Approximation and Projection, PCA: *Principal* genetic analysisTable 3The top 20 genes are upregulated and downregulatedGene namesLogFCP valueRegulation directionLCN2 − 0.926060.003277DownCAMP − 0.895230.00297DownOLFM4 − 0.883120.007425DownDEFA4 − 0.882960.00825DownCEACAM8 − 0.829470.024539DownLTF − 0.82210.013364DownCRISP3 − 0.802660.008933DownCD24 − 0.772160.017416DownMMP8 − 0.720180.008758DownELANE − 0.676170.001989DownHAB1 − 0.663182.75E−05DownNCKAP1L − 0.576761.55E−07DownEGR2 − 0.552550.012038DownHP − 0.52320.000704DownARG1 − 0.49470.007518DownPGLYRP1 − 0.491160.000471DownGRINA − 0.490060.001152DownMFSD10 − 0.47826.16E−07DownHSPB1 − 0.47544.22E−05DownCTSD − 0.462512.28E−05DownYIPF50.3083760.00068UpPARP80.3107420.001162UpHIST1H2AC0.3137570.077778UpIFRD10.3158960.006762UpERGIC20.3162510.000205UpFOXO30.3185525.93E−07UpLARP70.3190568.96E−05UpNCOA10.3279335.32E−08UpKIAA15510.3452750.001193UpMAN2A10.3454221.54E−05UpGZMB0.3495380.022463UpTAOK30.350368.64E−06UpRUFY30.3505750.000429UpPPWD10.3605515.56E−05UpPPIG0.3769940.000203UpCACNA2D30.3770528.98E−05UpZNF910.3806845.14E−05UpERAP20.3947090.11964UpDPP80.4013693.81E−07UpFOLR30.594270.001779UpFig. 5Potential target genes and PPI network map of KuA for OP. A The Venn results of potential genes of KuA therapy for OP. B Counts and lists of the top genes of PPI network map. C The PPI network map of 24 target genes. 723 protein nodes and 8743 edges were obtained for intersection genes. After screening with DC > 61 and a BC range of 20–113.2, the first 20 proteins were selected in Table 5 (in descending order of degree), with a total of 322 edges. PPI: Protein–Protein Interaction; KuA: Kukoamine A; OP: osteoporosisTable 4The 20 intersection genes sorted by logFCIntersection geneLogFCP valueBTBD18 − 0.0005556670.983394254DRD30.0022133330.932333004SYCE1L0.0025953330.885291809ATP1A10.0104093330.839478969POLR2D − 0.0109963330.746612769ICAM1 − 0.0209756670.696896549TNFAIP8 − 0.0415413330.569759794RBM47 − 0.0455896670.455734347VNN10.1934436670.337112369TSFM0.0474316670.256808881NRXN10.0395950.148943914FAM186A0.0406780.059556681MFNG − 0.1097430.051251272TMEM212 − 0.0526313330.040565698TPST2 − 0.1549193330.030568761RPS11 − 0.0956483330.014220426GABRD − 0.0967493330.012048201NTS − 0.088130.008618444TMEM120B − 0.0807790.005985507PYGM0.1026823330.000930636
## PPI network and topological analysis
The combination of DC and BC values is an effective method for reliable monitoring of important proteins (Wang et al. 2013). As shown in Fig. 2, 723 protein nodes and 8743 edges were obtained for intersection genes. After screening with DC > 61 and a BC range of 20–113.2, the first 20 proteins were selected in Table 5 (in descending order of degree), with a total of 322 edges. Among the 20 proteins, five proteins were predicted targets of the active ingredients, with their corresponding genes including NTRK1, MCM2, CUL3, NPM1, and FN1 (Fig. 5B, C, Table 5).Table 5Topological analysis results by degree—the first 20 proteinsGene namesAnnotationCloseness centralityDegreeNTRK1Neurotrophic receptor tyrosine kinase 10.582762218MCM2Minichromosome maintenance complex component 20.574324210CUL3Cullin 30.557638197NPM1Nucleophosmin0.555556181FN1Fibronectin 10.555037188HNRNPUHeterogeneous nuclear ribonucleoprotein U0.5494170ESR1Estrogen receptor 10.537975167CDK2Cyclin dependent kinase 20.537975154RPS11Ribosomal protein S110.534591171RPS3ARibosomal protein S3A0.534111148ITGA4Integrin subunit alpha 40.533632164RPS3Ribosomal protein S30.533632150CAND1Cullin associated and neddylation Dissociated 10.533154165RPS4XRibosomal protein S4, X-linked0.530303144COPS5COP9 signalosome subunit 50.524691148RPS14Ribosomal protein S140.524691146RPS16Ribosomal protein S160.524229143VCAM1Vascular cell adhesion molecule 10.523307146CUL1Cullin 10.519197153ICAM1Intercellular adhesion molecule 10.510292161
## Construction and analysis of the LC-OP-potential Target gene network
*The* gene and miRNA prediction network is an important method to predict miRNA and thus has certain significance to analyze the relationship between LC-OP (Fig. 6). The LC-OP-potential *Target* gene network was constructed by Cytoscape software (version 3.7.1). From the network of potential LC-OP target genes, a total of 42 nodes and 266 lines were derived, and coumarin A had the highest level in the process, which also explains its important role in the network (Bai et al. 2021).Fig. 6Topological analysis of the protein–protein interaction network (A) and GO/KEGG enrichment analysis (B, C). GO: genetic selection; KEGG: biological pathways. GO patents were linked with response to stimulus, metabolic process, and biological regulation
## GO and KEGG enrichment analysis
According to the KEGG enrichment results, the mechanism of active ingredients of LC in the treatment of OP mainly focuses on the interaction of neuroactive ligands with receptors, postoperative cancer, and small cell lung cancer (Fig. 6B).
GO enrichment analysis also included heatmap-related select GO patients, GO cell-type signatures, GO disease, GO trust, GO paGenbase and GO transcription factor. Heatmap-related select GO patients were associated with the pathway in cancer and response to the hormone. GO patents were linked with response to stimulus, metabolic process, and biological regulation. SP1, RELA, and NFKB1 pathways also played a key role in GO TRRUST (Bai et al. 2021) (Fig. 6C, Additional file 1: Figure S5). Additionally, the PYGM-related glycogene synthesis and degrade pathway were shown in Additional file 1: Figure S6.
## Molecular Docking
From the network of potential LC-OP target genes, five active ingredients were selected, namely KuA, Emodin, Kulactone, Alexandrine, and Acacetin (Table 6). A low Vina score indicates a stronger and more stable interaction between the compound and the receptor. Molecular docking between Active ingredients and target genes were shown in Table 7. The Vina score results of KuA, Linarin, aurantiamide acetate, and acacetin increased steadily, indicating that KuA has the strongest and most stable binding affinity for PYGM. These results suggest that KuA may be the most suitable starting material for PYGM. 3D images of acacetin, alexandrine, emodin, KuA, and Kulactone to PYGM were shown in Fig. 7C.Table 6Molecular docking parameters and results of seven active ingredients in LC binding with PYGMMolecule nameVina scoresCavity sizeCenterSizexyzxyzOIN − 7.8183752329212127Kukoamine A− 12.31598312430363636Linarin − 11.31598312430242424Aurantiamide acetate − 10.11598312430292323Acacetin − 101598312430292121Apigenin − 9.81598312430292121Kulactone − 8.9148426351242424Emodin − 8.91598312430291919Scopolin − 8.61598312430292121Physcion − 8.61598312430292020Stigmasterol − 8.5148426351252525Lyciumin A − 8.4148426351282828Hederagenin − 8.2148426351252525Alexandrin − 7.9183752329292929Atropine − 7.8183752329212127Sugiol − 7.7148426351262920Beta-sitosterol − 7.7148426351232923Campesterol − 7.7148426351242424Linoleyl acetate − 71598312430292929HEPTACOSANE − 71598312430313131CLR − 6.71598312430363636Tricosane − 6.41598312430282828Table 7Vina score of LC active components to the target gene moleculesActive componentsGenes-Vina scorePYGMTPST2RPS11GABRDAcacetin − 10 − 9 − 7.4 − 7.8Alexandrin − 7.9 − 7.6 − 8.4 − 7.7Apigenin − 9.8 − 8.2 − 7.3 − 7.7Atropine − 7.8 − 8.2 − 6.8 − 7.3Aurantiamide acetate − 10.1 − 7.4 − 8.2 − 7.3Beta-sitosterol − 7.7 − 7.3 − 7.7 − 7.3Campesterol − 7.7 − 7.4 − 7.3 − 7.6CLR − 6.7 − 5.7 − 6.2 − 5.9Emodin − 8.9 − 9.4 − 8.3 − 8.1Hederagenin − 8.2 − 7.6 − 8 − 7.5HEPTACOSANE − 7 − 5.4 − 5.6 − 5.5Kukoamine A − 12.3 − 7.3 − 6.5 − 6.8Kulactone − 8.9 − 7.9 − 8.5 − 9.4Linarin − 11.3 − 9 − 9.1 − 8.8Linoleyl acetate − 7 − 6.1 − 5.9 − 5.4Lyciumin A − 8.4 − 8.5 − 9.8 − 8.7OIN − 7.8 − 8.2 − 6.9 − 7.4Physcion − 8.6 − 9.6 − 7.5 − 8.3Scopolin − 8.6 − 9 − 7.9 − 7.2Stigmasterol − 8.5 − 7.4 − 8.1 − 7.3Sugiol − 7.7 − 6.8 − 7.4 − 7.5Tricosane − 6.4 − 5.3 − 4.6 − 5.4Fig. 7A *Target* genes-miRNA. B TCM compound-disease regulatory network. C The 3D map of binding of KuA. The Vina score results of KuA, Linarin, aurantiamide acetate, and acacetin increased steadily, indicating that KuA has the strongest and most stable binding affinity for PYGM. 3D images of acacetin, alexandrine, emodin, KuA, and Kulactone to PYGM. KuA: Kukoamine A; OVX: ovariotomy; TCM: Traditional Chinese Medicine
## KuA increased mRNA expression of osteoblastic differentiation-related genes in OVX mice
Figure 8A showed the mRNA expression in tibia treated with different concentrations of KuA in OVX mice. We first analyzed the expression of osteogenesis-related genes and found that KuA significantly up-regulated OCN expression compared with the OVX group. What’s more, KuA treatment can down-regulate the expression of osteoclast-related genes such as RNAKL, TRAP, and OPG than the OVX group. Finally, the PYGM was also effectively inhibited in KuA group than the OVX group. Fig. 8KuA protects against cell apoptosis and oxidative stress level in OVX mice. A mRNA expression in the tibia. B Bax/BCL-2 level. C The original membrane of the western blotting. D cytochrome c level. E Caspase-3. F MnSOD. G CuZnSOD. H H2O2 level. * $P \leq 0.05$, **$P \leq 0.01$,***$P \leq 0.001$,****$P \leq 0.0001.$ KuA: Kukoamine A; OVX: ovariotomy; Bax: BCL2 Associated X, Apoptosis Regulator; BCL2: BCL2 Apoptosis Regulator; SOD, Superoxide Dismutase; MDA: malondialdehyde
## KuA protects against OVX-induced inflammation
KuA can also prevent inflammation in hippocampal Neurogenesis (Zhang et al. 2017). So we investigated the anti-inflammatory role of KuA administration in the OVX mice and found that IL-6, CRP, TNF-α, and IL-1b levels were increased significantly in the OVX group than in the control group. Additionally, KuA treatment can inhibit inflammation levels such as IL-6, CRP, TNF-α, and IL-1b than the OVX group (Fig. 8A).
## KuA protects against OVX-induced cell apoptosis and oxidative stress level
According to the previous study, the neuroprotective effects of KuA inhibited oxidative stress in brain injury (Zhang et al. 2016). In this study, we analyzed the effect of KuA administration on cellular processes such as apoptosis and oxidative stress in the OVX mice. The Bax/Bcl-2, cytochrome c, and caspase-3 levels were lower significantly in the control group and KuA administration with 20 mg group than in the OVX group. In addition, we also analyzed the MnSOD activity, CuZnSOD, H2O2, and MDA levels, and it showed that the OVX group can increase MnSOD activity and CuZnSOD levels but reduce H2O2 and MDA levels than the control group. After treatment with KuA, MnSOD activity and CuZnSOD levels increased significantly than OVX group (Figs. 8B–H, 9A).Fig. 9KuA increased the osteoblastic differentiation and mineralized nodule formation of osteoblastic MC3T3-E1 cells. A MDA. B Relative ALP activity. C Alizarin Red SOD. D Relative cell viability. E mRNA level in MC3T3-E1 cells. * $P \leq 0.05$, **$P \leq 0.01$,***$P \leq 0.001$,****$P \leq 0.0001.$ KuA: Kukoamine A; OVX: ovariotomy; ALP, Alkaline Phosphatase; PYGM, Glycogen Phosphorylase, Muscle Associated; OCN, osteocalcin; Osterix: Sp7 Transcription Factor
## KuA increased osteoblastic differentiation and formation of mineralized nodules in MC3T3-E1 cells
We further examined the effects of KuA on osteoblast differentiation such as cell proliferation, matrix maturation, and matrix mineralization in the MC3T3-E1 cell line (Rutkovskiy et al. 2016). Several conventional methods have been used to assess the effects of KuA on osteoblast differentiation, including ALP activity, cell viability, and mineralization. ALP activity plays a key role in new bone mineralization and osteoblast differentiation (Watts 1999; Liu et al. 2014). ALP activity in KuA treatment cells (20 μm) was increased significantly at five days but administration with 5 and 10 μm KuA had no significant difference at five days. After 5 days of incubation with KuA (5, 10, and 20 μM), cell viability improved significantly compared with the control group.
The bone matrix is mineralized by osteoblast differentiation, leading to the induction of calcium and phosphorus-based minerals. Therefore, bone mineralization develops with different matrix proteins. ( Owen and Reilly 2018). Alizarin Red S is a commonly used histochemical method to assess calcium-rich deposits in osteoblast mineralization (Virtanen and Isotupa 1980). After simultaneous treatment with KuA (20 μM) and induction reagent, positive colonies stained with Alizarin Red S were larger than untreated control cells. These results suggested that KuA promotes osteoblast differentiation and the formation of mineralized nodules (Fig. 9).
## Discussion
OP is a systemic bone disease characterized by decreased bone mass and deterioration of bone microarchitecture with associated bone fragility and increased risk of fracture (Ensrud and Crandall 2017). Clinical vertebral and femoral fractures are the most devastating consequences of OP and are associated with morbidity and mortality (Black and Rosen 2016; Rachner et al. 2011). The mechanism of OP is related to multiple factors such as the regulation by the adaptive immune response (Weitzmann and Ofotokun 2016), genetic determination (Yang et al. 2020), oxidative stress, apoptotic mechanisms, sex-steroid deficiency, and macroautophagy (Hendrickx et al. 2015). In our study, the firstly obvious finding to emerge from the analysis was that 22 active ingredients of LC we investigated in treatment with OP were associated with a variety of proteins and signaling pathways, indicating that active ingredients play a potential role in the development of OP. And then we found that KuA plays important role in the treatment of OP through PYGM pathway. In vivo and in vitro experiments, we also found that KuA improves bone loss via inflammation and oxidative stress.
The role of LC has been widely studied and has been confirmed to exert beneficial effects on improving insulin resistance, lipid metabolism, bone metabolism, and tumor progress by inhibiting inflammation and immunity (Jeong et al. 2012; Park et al. 2014; Ye et al. 2008). The bioactive ingredients included KuA and B, scopolin, aurantiamide acetate, and others which have been uploaded into Table 1. The IC50 values of kukoamines A and B were 11.4, 9.5, respectively (Jiang et al. 2020). There are also significant differences in the extraction processes of the two drugs. What’s more, kukoamine A and B, were comparatively investigated for their antioxidant and cytoprotective effects in Fenton-damaged bone marrow-derived mesenchymal stem cells (bmMSCs). When compared with kukoamine B, kukoamine A consistently demonstrated higher IC50 values in PTIO-scavenging (pH 7.4), Cu2+ -reducing, DPPH-scavenging, O2− -scavenging, and OH− scavenging assays. However, in the PTIO-scavenging assay, the IC50 values of each kukoamine varied with pH value. In the Fe2 + -chelating assay, kukoamine B presented greater UV–Vis absorption and darker color than kukoamine A. In the HPLC−ESI−MS/MS analysis, kukoamine A with DPPH produced radical-adduct-formation (RAF) peaks (m/z 922 and 713). The 3- (4,5-Dimethylthiazol-2-yl)-2,5-diphenyl (MTT) assay suggested that both kukoamines concentration-dependently increased the viabilities of Fenton-damaged bmMSCs at 56.5–188.4 μM (Li et al. 2018). In particular, it protects the liver from lipid degeneration (Chen et al. 2018). Administration of aurantiamide acetate suppresses the growth gliomas by blocking autophagic flex (Yang et al. 2015). Scopolin also contains bioactive components used to treat and prevent OP (Park et al. 2020). Anxiolytic and anticonvulsant potential of stigmasterol have the positive modulation of GABA receptors and were considered to be candidates for steroidal drugs in the treatment of neurological disorders (Karim et al. 2021). A total of 22 therapeutic compounds have varying degrees of therapeutic effects on OP, mediated by a variety of cytokines and signaling pathways.
For OP, LC extract prevents OVX-induced BMD loss in mice by promoting osteoblast differentiation (Park et al. 2014). In addition, Kukoamine B has anti-osteoporotic effects in osteoblasts, osteoclasts, and ovariectomized OP mouse models (Park et al. 2019b). What’s more, a combined extract of Lycii Radicis and *Achyranthes japonica* also has an anti-osteoporotic effect (Park et al. 2019a). However, there are few studies to analyze which active components in LC may play an important role in OP. A previous study proved the role of LC-derived substances KuA and KuB in inhibiting amyloid aggregation in Alzheimer’s disease and type II diabetes (Jiang et al. 2020). These findings suggested that several LC active ingredients have synergistic effects in the treatment of OP. The active ingredients in LC such apigenin and scopolin were indeed found to possess anti-osteoporotic effects in the previous studies (Park et al. 2020; Tantowi et al. 2020), but due to there are many active components in LC, further studies are needed to identify the main anti-osteoporotic active ingredients at present and based on this, further analysis of the anti-osteoporosis effect of KuA by Vivo experiment via dual-energy X-ray absorptiometry, Micro-CT analysis, biomechanical analysis, Western blotting, and PCR analysis. Therefore, these integrated, complex methods were better to understand the role of KuA anti-osteoporotic bioactive ingredients in LC. However, which bioactive ingredients were the best ingredients in anti-osteoporotic progress also needs to be further in the future study.
Besides, there are also obvious differences in the anti-osteoporosis effects of different LC components such as Kukoamine A, Kukoamine B, apigenin, and scopolin. Kukoamine B was found to have anti-osteoporotic effects in promoting osteoblast differentiation but did not affect osteoclast differentiation, and ovariectomized OP mouse models in the previous studies (Park et al. 2019b). apigenin is comparable to diclofenac in suppressing inflammation and catabolic proteases for osteoporotic-osteoarthritis prevention (Tantowi et al. 2020). Scopolin treatment enhanced alkaline phosphatase activity and increased mineralized nodule formation in MC3T3-E1 pre-osteoblastic cells. However, osteoclast differentiation in primary-cultured monocytes was reduced by treatment with scopolin. Consistently, scopolin treatment increased osteoblast differentiation in the co-culture of monocytes (osteoclasts) and MC3T3-E1 (osteoblast) cells. Scopolin treatment prevented bone mineral density loss in OVX-induced osteoporotic mice. These results suggest that scopolin could be a therapeutic bioactive constituent for the treatment and prevention of osteoporosis (Park et al. 2020). However, our study also found that Kukoamine A had significant effects on osteoblasts and osteoclasts. Therefore, different LC components have different anti-osteoporosis effects.
Analysis of GO and KEGG enrichment revealed that steroid hormone, oxygen levels, and lipid metabolism may be the mechanism of LC in treating OP, which is according to previous research (Gennari et al. 2007; Domazetovic et al. 2017). The pathways with the best correlation were selected here for a discussion on the mechanism of the LC treatment for OP. Our results indicated a potential mechanism for the treatment of LC with OP.
As can be seen from the network of potential LC-OP target genes, many target genes can be regulated by β bonds, including but not limited to PYGM, RBM47, VNN1, TSFM, and ICAM1. An online meta-analysis examining polygene expression profiles in women finds that PYGM is associated with BMD (He et al. 2016). These results indicated that LC has the biological characteristics of multi-component and multi-target in the treatment of OP. In addition, the PPI results showed that the 89 target proteins were not independent of each other, but were linked and interacted with each other (Zhang et al. 2019b). These results also suggested that LC may participate in the remission and treatment of OP by regulating various proteins, and KuA may be the most critical target.
CB-Dock was designed to perform blind docking at predicted sites, instead of the entire surface of a protein. Therefore, the first step is to detect putative binding sites (Cavity detection). Since the ligand binding sites are usually larger cavities, we select several top cavities according to cavity size for further analysis (Cavity sorting). Then, we calculate the docking center and adjust the docking box size. These parameters are required for molecular docking with AutoDock Vina (Center and Size). After the docking process is finished, the bound poses are reranked according to the docking score (Dock and Rerank). The first conformation is considered the best binding pose and the corresponding site is the optimal binding site for the query ligand (Liu et al. 2020; Cao and Li 2014).
As previously mentioned, KuA is a major bioactive component extracted from the root barks of LC which can upregulate Srebp-1c and inhibit insulin-stimulated glucose uptake and lipid accumulation in hepatic steatosis (Li et al. 2017). It also has an anti-oxidative effect and anti-apoptosis stress in protecting the brain against injury by pMACO oxidative effect and anti-apoptosis stress (Liu et al. 2017). We further investigated the effects of KuA on osteoporotic mice and cell lines. Radiographic results and mechanical tests showed that KuA significantly improved the bone microstructure and mechanical strength of osteoporotic mice and osteoblast activity.
Inflammation status regulated the progress of OP in many previous studies (Cortet et al. 2019; Lee and Kim 2020; Zhou et al. 2021b). In the absence of estrogen, fracture healing is hindered by an increase in proinflammatory cytokines such as IL-6, which may lead to poor healing (Fischer and Haffner-Luntzer 2021). The progressive increase in the secretion of IL-1β and TNF-α contributes to postmenopausal bone loss (Pacifici et al. 1993; Chow et al. 2020). In previous studies, KuA attenuated the pro-inflammatory cytokines such as IL-1β and TNF-a levels in radiation-induced neuroinflammation (Zhang et al. 2017). There are similarities between the attitudes expressed in our study and found that KuA alleviated the inflammation level in treatment OP.
Oxidative stress is thought to be a causative factor in many disease states, possibly including a reduction in bone mineral density in OP (Kimball et al. 2021). Oxidative stress serves as potential biomarkers such as superoxide dismutase (SOD) in erythrocytes, catalase (CAT), total antioxidant status (TAS), hydroperoxides (HY), advanced oxidation protein products (AOPP), malondialdehyde (MDA), and vitamin B12 (VB12) in the etiopathophysiology and clinical course of OP (Zhou et al. 2016). Following the present results, previous studies have demonstrated that KuA has the ability to anti-oxidative stress via attenuated LDH release, ROS production, MDA level, MMP loss, and intracellular Ca2 + overload (Hu et al. 2015). These results are consistent with our studies that KuA reduces the oxidative stress level in OP.
Cell apoptosis was also involved in the progress of OP (Gruver-Yates and Cidlowski 2013). Expression of the pro-apoptotic factor caspase-3 or the anti-apoptotic factor Bcl-2 has been shown to affect osteoblast apoptosis (Akiyama et al. 2003). OP reduces the expression of caspase-3 and Bcl-2 in osteoblasts, thereby preventing bone loss (Jilka et al. 2014). Mitochondrial dysfunction in osteoblasts contributes to glucocorticoid-induced bone loss (Chen et al. 2020). Bcl-2-associated X protein (Bax) is a critical executioner of mitochondrial regulated cell death through its lethal activity of permeabilizing the mitochondrial outer membrane (Spitz and Gavathiotis 2022). The BCL2 family proteins comprise the sentinel network that regulates the mitochondrial or intrinsic apoptotic response (Hata et al. 2015). KuA inhibited apoptosis induction by decreasing the level of Bax, and caspase-3 in human glioblastoma cell growth (Wang et al. 2016). Their results match the observation in our study that KuA attenuated the cell apoptosis level in OP.
Glycogen Phosphorylase (PYGM) is a key enzyme in the first step of glycogenolysis, encoding the muscle-specific glycogen phosphorylase (myophosphorylase). The main role of PYGM is to provide sufficient energy for muscle contraction. However, it is expressed in tissues other than muscle, such as the bone, brain, lymphoid tissues, and blood. PYGM also played an important role in a variety of diseases such as early fatigue, myalgia, and contractures (Villarreal-Salazar et al. 2021; Gomes et al. 2020; Jin and Yang 2019; Nogales-Gadea et al. 2015). PYGM was identified as a candidate gene that may play an important role in BMD regulation in women (He et al. 2016). PYGM plays a potential role in glycogenolysis which affects glycogen metabolism, skeletal muscle, and bone metabolism (He et al. 2016; Tarnopolsky 2018). In our study, PYGM also play important role in KuA treatment of OP. In vivo and in vitro experiments, the PYGM mRNA level was regulated by KuA. What’s more, to further clarify the role of PYGM in KuA in the treatment of OP, we used transient transfections of siRNA molecules to study the role of PYGM in the treatment of OP. Interestingly, PYGM may be a novelty discovered biologically meaningful functional modules in the progress of KuA treatment of OP.
There are some limitations in our study. Kukoamine B was found to have anti-osteoporotic effects in promoting osteoblast differentiation but did not affect osteoclast differentiation, and ovariectomized OP mouse models in previous studies (Park et al. 2019a). These results suggest that Kukoamine B may be a potential therapeutic candidate for the treatment of osteoporosis. However, in our study, Kukoamine A was found to be superior to Kukoamine B in terms of oral bioavailability (OB), drug-likeness (DL), intestinal epithelial permeability, blood–brain barrier penetrability, and water solubility, which needs to be further validated using experiments. Besides, *In this* study, we chose autophagy-related proteins to analyze changes in autophagy levels, but additional autophagy-related assays, including flow cytometry, are needed in future studies.
## Conclusion
Taken together, KuA extract from LC has obvious advantages in the treatment of OP. The biological activity of the active substance of LC and the signal transduction pathway of the target OP gene were studied by network pharmacology method and molecular binding test. Meanwhile, this is the first study to investigate the anti-osteoporotic effect of KuA in vivo and vitro. The results suggest that KuA could be a good candidate for treating and preventing OP.
## Supplementary Information
Additional file 1: Figure S1. The workflow of KuA in treatment of OP. Figure S2. GO enrichment analysis among target genes. ( A) Cell type signature. ( B) Disgenet analysis. ( C) Trrust analysis. ( D) Pagenbase analysis. ( E) Transcription factor. Figure S3. KuA improve the mechanical properties and inflammation level in OVX mice. ( A) Stiffness of tibia. ( B) Displacement of tibia. ( C) Energy absorption. ( D) serum IL-6. ( E)serum CRP. ( F) serum TNF-α. ( G) serum IL-1b. ( H) *Principal* genetic analysis among all variable. Figure S4. The PYGM related signaling pathway of potential target genes of LC in OP. Figure S5. KuA increased significantly the tibia and spine bone microstructure and mechanical properties in OVX mice. Figure S6. KuA increased significantly the spine bone microstructure and mechanical properties in OVX mice. Table S1. Primer Sequences used for RT-QPCR.
## References
1. Akiyama T, Bouillet P, Miyazaki T, Kadono Y, Chikuda H, Chung UI, Fukuda A, Hikita A, Seto H, Okada T. **Regulation of osteoclast apoptosis by ubiquitylation of proapoptotic BH3-only Bcl-2 family member Bim**. *Embo J* (2003) **22** 6653-6664. DOI: 10.1093/emboj/cdg635
2. Bai LL, Chen H, Zhou P, Yu J. **Identification of Tumor Necrosis Factor-Alpha (TNF-α) inhibitor in rheumatoid arthritis using network pharmacology and molecular docking**. *Front Pharmacol* (2021) **12** 690118. DOI: 10.3389/fphar.2021.690118
3. Black DM, Rosen CJ. **Clinical Practice**. *Postmenopausal Osteoporosis N Engl J Med* (2016) **374** 254-262. DOI: 10.1056/NEJMcp1513724
4. Cao Y, Li L. **Improved protein-ligand binding affinity prediction by using a curvature-dependent surface-area model**. *Bioinformatics* (2014) **30** 1674-1680. DOI: 10.1093/bioinformatics/btu104
5. Chen H, Olatunji OJ, Zhou Y. **Ameliorative potential of Cortex Lycii on enzymes involved in carbohydrate metabolism in streptozotocin-nicotinamide induced diabetic rats**. *Pak J Pharm Sci* (2018) **31** 2419-2428. PMID: 30473513
6. Chen L, Hu SL, Xie J, Yan DY, Weng SJ, Tang JH, Wang BZ, Xie ZJ, Wu ZY, Yang L. **Proanthocyanidins-mediated Nrf2 activation ameliorates glucocorticoid-induced oxidative stress and mitochondrial dysfunction in osteoblasts**. *Oxid Med Cell Longev* (2020) **2020** 9102012. DOI: 10.1155/2020/9102012
7. Chow SK, Chim YN, Wang JY, Wong RM, Choy VM, Cheung WH. **Inflammatory response in postmenopausal osteoporotic fracture healing**. *Bone Joint Res* (2020) **9** 368-385. DOI: 10.1302/2046-3758.97.BJR-2019-0300.R2
8. Compston JE, McClung MR, Leslie WD. **Osteoporosis**. *Lancet* (2019) **393** 364-376. DOI: 10.1016/S0140-6736(18)32112-3
9. Cortet B, Lucas S, Legroux-Gerot I, Penel G, Chauveau C, Paccou J. **Bone disorders associated with diabetes mellitus and its treatments**. *Joint Bone Spine* (2019) **86** 315-320. DOI: 10.1016/j.jbspin.2018.08.002
10. Domazetovic V, Marcucci G, Iantomasi T, Brandi ML, Vincenzini MT. **Oxidative stress in bone remodeling: role of antioxidants**. *Clin Cases Miner Bone Metab* (2017) **14** 209-216. DOI: 10.11138/ccmbm/2017.14.1.209
11. Ensrud KE, Crandall CJ. **Osteoporosis**. *Ann Intern Med* (2017) **167** 17-32. DOI: 10.7326/AITC201708010
12. Fischer V, Haffner-Luntzer M. **Interaction between bone and immune cells: Implications for postmenopausal osteoporosis**. *Semin Cell Dev Biol* (2021) **78** 7
13. Gennari L, De Paola V, Merlotti D, Martini G, Nuti R. **Steroid hormone receptor gene polymorphisms and osteoporosis: a pharmacogenomic review**. *Expert Opin Pharmacother* (2007) **8** 537-553. DOI: 10.1517/14656566.8.5.537
14. Gomes CP, da Silva AMS, Zanoteli E, Pesquero JB. **A new mutation in PYGM causing McArdle disease in a Brazilian patient**. *Acta Neurol Belg* (2020) **120** 705-707. DOI: 10.1007/s13760-019-01159-7
15. Gruver-Yates AL, Cidlowski JA. **Tissue-specific actions of glucocorticoids on apoptosis: a double-edged sword**. *Cells* (2013) **2** 202-223. DOI: 10.3390/cells2020202
16. Hadjipavlou-Litina D, Garnelis T, Athanassopoulos CM, Papaioannou D. **Kukoamine A analogs with lipoxygenase inhibitory activity**. *J Enzyme Inhib Med Chem* (2009) **24** 1188-1193. DOI: 10.1080/14756360902779193
17. Hata AN, Engelman JA, Faber AC. **The BCL2 family: key mediators of the apoptotic response to targeted anticancer therapeutics**. *Cancer Discov* (2015) **5** 475-487. DOI: 10.1158/2159-8290.CD-15-0011
18. He H, Cao S, Niu T, Zhou Y, Zhang L, Zeng Y, Zhu W, Wang YP, Deng HW. **Network-based meta-analyses of associations of multiple gene expression profiles with bone mineral density variations in women**. *PLoS ONE* (2016) **11** e0147475. DOI: 10.1371/journal.pone.0147475
19. Hendrickx G, Boudin E, Van Hul W. **A look behind the scenes: the risk and pathogenesis of primary osteoporosis**. *Nat Rev Rheumatol* (2015) **11** 462-474. DOI: 10.1038/nrrheum.2015.48
20. Hu XL, Gao LY, Niu YX, Tian X, Wang J, Meng WH, Zhang Q, Cui C, Han L, Zhao QC. **Neuroprotection by Kukoamine A against oxidative stress may involve N-methyl-D-aspartate receptors**. *Biochim Biophys Acta* (2015) **1850** 287-298. DOI: 10.1016/j.bbagen.2014.11.006
21. Inada M, Matsumoto C, Miyaura C. **Animal models for bone and joint disease. Ovariectomized and orchidectomized animals**. *Clin Calcium* (2011) **21** 164-170. PMID: 21289412
22. Jeong JC, Kim SJ, Kim YK, Kwon CH, Kim KH. **Lycii cortex radicis extract inhibits glioma tumor growth in vitro and in vivo through downregulation of the Akt/ERK pathway**. *Oncol Rep* (2012) **27** 1467-1474. PMID: 22266745
23. Jiang Y, Zhong M, Long F, Yang R, Zhang Y, Liu T. **Network Pharmacology-Based Prediction of Active Ingredients and Mechanisms of Lamiophlomis rotata (Benth) Kudo Against Rheumatoid Arthritis**. *Front Pharmacol* (2019) **10** 1435. DOI: 10.3389/fphar.2019.01435
24. Jiang G, Takase M, Aihara Y, Shigemori H. **Inhibitory activities of kukoamines A and B from Lycii Cortex on amyloid aggregation related to Alzheimer's disease and type 2 diabetes**. *J Nat Med* (2020) **74** 247-251. DOI: 10.1007/s11418-019-01337-0
25. Jilka RL, O'Brien CA, Roberson PK, Bonewald LF, Weinstein RS, Manolagas SC. **Dysapoptosis of osteoblasts and osteocytes increases cancellous bone formation but exaggerates cortical porosity with age**. *J Bone Miner Res* (2014) **29** 103-117. DOI: 10.1002/jbmr.2007
26. Jin Y, Yang Y. **Bioinformatics-based discovery of PYGM and TNNC2 as potential biomarkers of head and neck squamous cell carcinoma**. *Biosci Rep* (2019) **39** 45. DOI: 10.1042/BSR20191612
27. Kalyanaraman H, Ramdani G, Joshua J, Schall N, Boss GR, Cory E, Sah RL, Casteel DE, Pilz RB. **A Novel, Direct NO donor regulates osteoblast and osteoclast functions and increases bone mass in ovariectomized mice**. *J Bone Miner Res* (2017) **32** 46-59. DOI: 10.1002/jbmr.2909
28. Karim N, Khan I, Abdelhalim A, Halim SA, Khan A, Al-Harrasi A. **Stigmasterol can be new steroidal drug for neurological disorders: evidence of the GABAergic mechanism via receptor modulation**. *Phytomedicine* (2021) **90** 153646. DOI: 10.1016/j.phymed.2021.153646
29. Khosla S, Hofbauer LC. **Osteoporosis treatment: recent developments and ongoing challenges**. *Lancet Diabetes Endocrinol* (2017) **5** 898-907. DOI: 10.1016/S2213-8587(17)30188-2
30. Kimball JS, Johnson JP, Carlson DA. **Oxidative stress and osteoporosis**. *J Bone Joint Surg Am* (2021) **103** 1451-1461. DOI: 10.2106/JBJS.20.00989
31. Lee CS, Kim SH. **Anti-inflammatory and Anti-osteoporotic Potential of**. *Probiotics Antimicrob Proteins* (2020) **12** 623-634. DOI: 10.1007/s12602-019-09577-y
32. Lee B, Hong S, Kim M, Kim EY, Park HJ, Jung HS, Kim JH, Sohn Y. *Int J Mol Med* (2021) **48** 2. DOI: 10.3892/ijmm.2021.4988
33. Lei T, Liang Z, Li F, Tang C, Xie K, Wang P, Dong X, Shan S, Jiang M, Xu Q. **Pulsed electromagnetic fields (PEMF) attenuate changes in vertebral bone mass, architecture and strength in ovariectomized mice**. *Bone* (2018) **108** 10-19. DOI: 10.1016/j.bone.2017.12.008
34. Li S, Zhang B. **Traditional Chinese medicine network pharmacology: theory, methodology and application**. *Chin J Nat Med* (2013) **11** 110-120. DOI: 10.3724/SP.J.1009.2013.00110
35. Li G, Zhou F, Chen Y, Zhang W, Wang N. **Kukoamine A attenuates insulin resistance and fatty liver through downregulation of Srebp-1c**. *Biomed Pharmacother* (2017) **89** 536-543. DOI: 10.1016/j.biopha.2017.02.024
36. Li X, Lin J, Chen B, Xie H, Chen D. **Antioxidant and Cytoprotective Effects of Kukoamines A and B: Comparison and Positional Isomeric Effect**. *Molecules* (2018) **23** 4
37. Liu J, Nam HK, Campbell C, Gasque KC, Millán JL, Hatch NE. **Tissue-nonspecific alkaline phosphatase deficiency causes abnormal craniofacial bone development in the Alpl(-/-) mouse model of infantile hypophosphatasia**. *Bone* (2014) **67** 81-94. DOI: 10.1016/j.bone.2014.06.040
38. Liu J, Jiang X, Zhang Q, Lin S, Zhu J, Zhang Y, Du J, Hu X, Meng W, Zhao Q. **Neuroprotective effects of Kukoamine A against cerebral ischemia via antioxidant and inactivation of apoptosis pathway**. *Neurochem Int* (2017) **107** 191-197. DOI: 10.1016/j.neuint.2016.12.024
39. Liu Y, Grimm M, Dai WT, Hou MC, Xiao ZX, Cao Y. **CB-Dock: a web server for cavity detection-guided protein-ligand blind docking**. *Acta Pharmacol Sin* (2020) **41** 138-144. DOI: 10.1038/s41401-019-0228-6
40. Liu Y, Zheng W, Zhong Y, Zhang L, Su T, Liang G, Zhang D, Zhang Y, Chen J, Gong M. **Identification of α-glucosidase inhibitors from Cortex Lycii based on a bioactivity-labeling high-resolution mass spectrometry-metabolomics investigation**. *J Chromatogr A* (2021) **1642** 462041. DOI: 10.1016/j.chroma.2021.462041
41. Myers SA, Nield A, Chew GS, Myers MA. **The zinc transporter, Slc39a7 (Zip7) is implicated in glycaemic control in skeletal muscle cells**. *PLoS ONE* (2013) **8** e79316. DOI: 10.1371/journal.pone.0079316
42. Nogales-Gadea G, Brull A, Santalla A, Andreu AL, Arenas J, Martín MA, Lucia A, de Luna N, Pinós T. **McArdle disease: update of reported mutations and polymorphisms in the PYGM gene**. *Hum Mutat* (2015) **36** 669-678. DOI: 10.1002/humu.22806
43. Owen R, Reilly GC. **In vitro models of bone remodelling and associated disorders**. *Front Bioeng Biotechnol* (2018) **6** 134. DOI: 10.3389/fbioe.2018.00134
44. Pacifici R, Vannice JL, Rifas L, Kimble RB. **Monocytic secretion of interleukin-1 receptor antagonist in normal and osteoporotic women: effects of menopause and estrogen/progesterone therapy**. *J Clin Endocrinol Metab* (1993) **77** 1135-1141. PMID: 8077304
45. Park E, Jin HS, Cho DY, Kim J, Kim MC, Choi CW, Jin Y, Lee JW, Park JH. **Chung YS and others The effect of Lycii Radicis Cortex extract on bone formation in vitro and in vivo**. *Molecules* (2014) **19** 19594-19609. DOI: 10.3390/molecules191219594
46. Park E, Kim J, Yeo S, Lim E, Choi CW, Choi S, Li WY, Lee JW, Park JH, Huh D. **Anti-Osteoporotic Effects of Combined Extract of Lycii Radicis Cortex and Achyranthes japonica in Osteoblast and Osteoclast Cells and Ovariectomized Mice**. *Nutrients* (2019) **11** 11. DOI: 10.3390/nu11112716
47. Park E, Kim J, Kim MC, Yeo S, Kim J, Park S, Jo M, Choi CW, Jin HS, Lee SW. **Anti-Osteoporotic Effects of Kukoamine B Isolated from Lycii Radicis Cortex Extract on Osteoblast and Osteoclast Cells and Ovariectomized Osteoporosis Model Mice**. *Int J Mol Sci* (2019) **20** 11. DOI: 10.3390/ijms20112784
48. Park E, Kim J, Jin HS, Choi CW, Choi TH, Choi S, Huh D, Jeong SY. **Scopolin attenuates osteoporotic bone loss in ovariectomized mice**. *Nutrients* (2020) **12** 11. DOI: 10.3390/nu12113565
49. Rachner TD, Khosla S, Hofbauer LC. **Osteoporosis: now and the future**. *Lancet* (2011) **377** 1276-1287. DOI: 10.1016/S0140-6736(10)62349-5
50. Ru J, Li P, Wang J, Zhou W, Li B, Huang C, Li P, Guo Z, Tao W. **TCMSP: a database of systems pharmacology for drug discovery from herbal medicines**. *J Cheminform* (2014) **6** 13. DOI: 10.1186/1758-2946-6-13
51. Rutkovskiy A, Stensløkken KO, Vaage IJ. **Osteoblast Differentiation at a Glance**. *Med Sci Monit Basic Res* (2016) **22** 95-106. DOI: 10.12659/MSMBR.901142
52. Silverstein WK, Zipursky JS, Shadowitz S. **Closing the osteoporosis care gap: a teachable moment**. *JAMA Intern Med* (2021) **181** 1635-1636. DOI: 10.1001/jamainternmed.2021.5972
53. Spitz AZ, Gavathiotis E. **Physiological and pharmacological modulation of BAX**. *Trends Pharmacol Sci* (2022) **43** 206-220. DOI: 10.1016/j.tips.2021.11.001
54. Szklarczyk D, Santos A, von Mering C, Jensen LJ, Bork P, Kuhn M. **STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data**. *Nucleic Acids Res* (2016) **44** D380-D384. DOI: 10.1093/nar/gkv1277
55. Tantowi N, Mohamed S, Lau SF, Hussin P. **Comparison of diclofenac with apigenin-glycosides rich**. *Daru* (2020) **28** 443-453. DOI: 10.1007/s40199-020-00343-y
56. Tarnopolsky MA. **Myopathies related to glycogen metabolism disorders**. *Neurotherapeutics* (2018) **15** 915-927. DOI: 10.1007/s13311-018-00684-2
57. Villarreal-Salazar M, Brull A, Nogales-Gadea G, Andreu AL, Martín MA, Arenas J, Santalla A, Lucia A, Vissing J, Krag TO. **Preclinical research in McArdle disease: a review of research models and therapeutic strategies**. *Genes (basel)* (2021) **13** 56. DOI: 10.3390/genes13010074
58. Virtanen P, Isotupa K. **Staining properties of alizarin red S for growing bone in vitro**. *Acta Anat (basel)* (1980) **108** 202-207. DOI: 10.1159/000145301
59. Wang J, Peng W, Wu FX. **Computational approaches to predicting essential proteins: a survey**. *Proteomics Clin Appl* (2013) **7** 181-192. DOI: 10.1002/prca.201200068
60. Wang Q, Li H, Sun Z, Dong L, Gao L, Liu C, Wang X. **Kukoamine A inhibits human glioblastoma cell growth and migration through apoptosis induction and epithelial-mesenchymal transition attenuation**. *Sci Rep* (2016) **6** 36543. DOI: 10.1038/srep36543
61. Watts NB. **Clinical utility of biochemical markers of bone remodeling**. *Clin Chem* (1999) **45** 1359-1368. DOI: 10.1093/clinchem/45.8.1359
62. Wei M, Li H, Li Q, Qiao Y, Ma Q, Xie R, Wang R, Liu Y, Wei C, Li B. **Based on Network Pharmacology to Explore the Molecular Targets and Mechanisms of Gegen Qinlian Decoction for the Treatment of Ulcerative Colitis**. *Biomed Res Int* (2020) **2020** 5217405. DOI: 10.1155/2020/5217405
63. Weitzmann MN, Ofotokun I. **Physiological and pathophysiological bone turnover - role of the immune system**. *Nat Rev Endocrinol* (2016) **12** 518-532. DOI: 10.1038/nrendo.2016.91
64. Yajun W, Jin C, Zhengrong G, Chao F, Yan H, Weizong W, Xiaoqun L, Qirong Z, Huiwen C, Hao Z. **Betaine Attenuates Osteoarthritis by Inhibiting Osteoclastogenesis and Angiogenesis in Subchondral Bone**. *Front Pharmacol* (2021) **12** 723988. DOI: 10.3389/fphar.2021.723988
65. Yang Y, Zhang LH, Yang BX, Tian JK, Zhang L. **Aurantiamide acetate suppresses the growth of malignant gliomas in vitro and in vivo by inhibiting autophagic flux**. *J Cell Mol Med* (2015) **19** 1055-1064. DOI: 10.1111/jcmm.12498
66. Yang TL, Shen H, Liu A, Dong SS, Zhang L, Deng FY, Zhao Q, Deng HW. **A road map for understanding molecular and genetic determinants of osteoporosis**. *Nat Rev Endocrinol* (2020) **16** 91-103. DOI: 10.1038/s41574-019-0282-7
67. Ye Z, Huang Q, Ni HX, Wang D. **Cortex Lycii Radicis extracts improve insulin resistance and lipid metabolism in obese-diabetic rats**. *Phytother Res* (2008) **22** 1665-1670. DOI: 10.1002/ptr.2552
68. Zhang Y, Cheng Z, Wang C, Ma H, Meng W, Zhao Q. **Neuroprotective effects of kukoamine a against radiation-induced rat brain injury through inhibition of oxidative stress and neuronal apoptosis**. *Neurochem Res* (2016) **41** 2549-2558. DOI: 10.1007/s11064-016-1967-0
69. Zhang Y, Gao L, Cheng Z, Cai J, Niu Y, Meng W, Zhao Q. **Kukoamine A prevents radiation-induced neuroinflammation and preserves hippocampal neurogenesis in rats by inhibiting activation of NF-κB and AP-1**. *Neurotox Res* (2017) **31** 259-268. DOI: 10.1007/s12640-016-9679-4
70. Zhang R, Zhu X, Bai H, Ning K. **Network pharmacology databases for traditional chinese medicine: review and assessment**. *Front Pharmacol* (2019) **10** 123. DOI: 10.3389/fphar.2019.00123
71. Zhang L, Yu J, Wang C, Wei W. **The effects of total glucosides of paeony (TGP) and paeoniflorin (Pae) on inflammatory-immune responses in rheumatoid arthritis (RA)**. *Funct Plant Biol* (2019) **46** 107-117. DOI: 10.1071/FP18080
72. Zheng W, Ding B, Li X, Liu D, Yokota H, Zhang P. **Knee loading repairs osteoporotic osteoarthritis by relieving abnormal remodeling of subchondral bone via Wnt/β-catenin signaling**. *Faseb j* (2020) **34** 3399-3412. DOI: 10.1096/fj.201902117R
73. Zhou Q, Zhu L, Zhang D, Li N, Li Q, Dai P, Mao Y, Li X, Ma J, Huang S. **Oxidative stress-related biomarkers in postmenopausal osteoporosis: a systematic review and meta-analyses**. *Dis Markers* (2016) **2016** 7067984. DOI: 10.1155/2016/7067984
74. Zhou W, Lai X, Wang X, Yao X, Wang W, Li S. **Network pharmacology to explore the anti-inflammatory mechanism of Xuebijing in the treatment of sepsis**. *Phytomedicine* (2021) **85** 153543. DOI: 10.1016/j.phymed.2021.153543
75. Zhou X, Guan Z, Jin X, Zhao J, Chen G, Ding J, Ren Y, Zhai X, Zhou Q, Guan Z. **Reversal of alopecia areata, osteoporosis follow treatment with activation of Tgr5 in mice**. *Biosci Rep* (2021) **41** 7. DOI: 10.1042/BSR20210609
|
---
title: 'Effect of a four-week isocaloric ketogenic diet on physical performance at
very high-altitude: a pilot study'
authors:
- Nicolas Chiarello
- Bertrand Leger
- Mathieu De Riedmatten
- Michel F. Rossier
- Philippe Vuistiner
- Michael Duc
- Arnaud Rapillard
- Lara Allet
journal: BMC Sports Science, Medicine and Rehabilitation
year: 2023
pmcid: PMC10029223
doi: 10.1186/s13102-023-00649-9
license: CC BY 4.0
---
# Effect of a four-week isocaloric ketogenic diet on physical performance at very high-altitude: a pilot study
## Abstract
### Background
A ketogenic diet (KD) reduces daily carbohydrates (CHOs) ingestion by replacing most calories with fat. KD is of increasing interest among athletes because it may increase their maximal oxygen uptake (VO2max), the principal performance limitation at high-altitudes (1500–3500 m). We examined the tolerance of a 4-week isocaloric KD (ICKD) under simulated hypoxia and the possibility of evaluating ICKD performance benefits with a maximal graded exercise bike test under hypoxia and collected data on the effect of the diet on performance markers and arterial blood gases.
### Methods
In a randomised single-blind cross-over model, 6 recreational mountaineers (age 24–44 years) completed a 4-week ICKD followed or preceded by a 4-week usual mixed Western-style diet (UD). Performance parameters (VO2max, lactate threshold [LT], peak power [Ppeak]) and arterial blood gases (PaO2, PaCO2, pH, HCO3−) were measured at baseline under two conditions (normoxia and hypoxia) as well as after a 4-week UD and 4-week ICKD under the hypoxic condition.
### Results
We analysed data for all 6 participants (BMI 19.9–24.6 kg m−2). Mean VO2max in the normoxic condition was 44.6 ml kg−1 min−1. Hypoxia led to decreased performance in all participants. With the ICKD diet, median values for PaO2 decreased by − $14.5\%$ and VO2max by + $7.3\%$ and Ppeak by + $4.7\%$.
### Conclusion
All participants except one could complete the ICKD. VO2max improved with the ICKD under the hypoxia condition. Therefore, an ICKD is an interesting alternative to CHOs dependency for endurance performance at high-altitudes, including high-altitude training and high-altitude races. Nevertheless, decreased PaO2 with ICKD remains a significant limitation in very-high to extreme altitudes (> 3500 m).
Trial registration Clinical trial registration Nr. NCT05603689 (Clinicaltrials.gov). Ethics approval CER-VD, trial Nr. 2020-00427, registered 18.08.2020—prospectively registered.
## Introduction
High-fat/low-carbohydrate diets or ketogenic diets (KDs) are an innovative strategy to enhance endurance performance if exercise duration is long enough (e.g., > 4 h) and exercise intensity is low enough (50–$60\%$ maximal oxygen uptake [VO2max]) [1]. This strategy restricts daily carbohydrates (CHOs) consumption while maintaining low to moderate protein content, thus replacing most calories with fat. There is no standard definition of KD to achieve ketosis because of interindividual variability. CHOs consumption < 30–50 g d−1 represents an accurate assessment [2, 3].
From a human evolutionary perspective, fat played a dominant role in energy supply [4]. However, the advent of agriculture shifted the major calorie contributor from fat to CHOs [5].
A century ago, modern science tried to determine the ideal human diet to optimize performance, with several historical studies conducted between 1939 and 1967 [6–8]. Performance was first enhanced in 1939 by giving additional CHOs to individuals with low blood sugar [6]. The association of glycogen depletion with the development of fatigue and glycogen resynthesis with a CHOs-rich diet were later shown by the needle biopsy technique [7]. The idea that only a high-CHOs diet could optimize performance gained credence, with a major effort for optimising glycogen storage. Over time, this led to the most commonly accepted dietary recommendation among athletes: high-CHOs, moderate-protein and low-fat diet [9].
However, in the early 1980s, a link between high-fat diets and exercise capacity was demonstrated. Phinney et al. [ 10] showed un-impaired performance in patients with a KD [10]. Several studies challenged the approach of glycogen storage optimisation for enhancing endurance performance. There are some indications that high-CHOs consumption may limit athlete’s performance when competing for an extended time. CHOs stores in muscle tissue (300 g), liver tissue (90 g) and the blood stream (30 g) are sufficient for only 1–3 h of activity for endurance athletes [11]. Peak fat oxidation rate occurs in submaximal exercise intensity between 47 and $64\%$ of VO2max [1, 12]. Also, well above this CHOs threshold (> $80\%$ VO2max), athlete’s performed equally well while eating a high-fat or high-CHOs diet [13]. Empiric observations also indicate well-being with a traditional Inuit diet [14] almost exclusively based on fat and proteins. Furthermore, a growing number of keto-adapted ultra-runner athletes are competing at high levels.
In this context, a new paradigm emerges with the idea to use the virtually unlimited fat store for endurance exercises [1, 5, 15]. The body can adapt to use fat as its primary fuel during submaximal exercise [12] with metabolic adaptation similar to prolonged fasting [4] and increasing the fat oxidation rate (from 0.4–0.6 to 1.2–1.3 g min−1) [16]. This probably functions by affecting the mitochondrial respiratory chain [2, 16–19]. Furthermore, a KD leads to a biological ketosis by forcing the liver to produce ketone bodies (KBs) by diverting acetyl-CoA [2]. KBs may positively affect slow-muscle fibers (type I) and negatively affect fast-muscle fibers (type II), which can potentially also enhance endurance exercises [2, 18]. KBs are also suggested to be more energy-efficient than glucose [15, 20].
The concept of positive effects of keto-adaptation on endurance performance is still strongly challenged. Burke et al. [ 21] investigated the effect of a high-fat diet during a 3-week intensified training. In this study, increased rate of fat oxidation resulted in increased oxygen demand for a given work load, impairing exercise economy [21]. Still today, exercise and sport nutrition guidelines recommend that endurance athletes eat more CHOs (7–10 g kg−1 d−1) than routine CHOs intake (5–7 g kg−1 d−1) to optimise muscle glycogen stores [22–24].
High-altitude mountaineering is said to have been invented in the middle of the eighteenth century by H.B. de Saussure. Originally limited to scientists and conquistadors, mountaineering as a sport that really emerged much later. The first Mount Everest ascent without oxygen in 1978 contributed to this aspect. Since then, a plethora of studies have explored the major performance limitations at high altitude. Indeed, diminished inspiratory oxygen pressure (PIO2) at high altitude [25] is critical for the delivery of oxygen to tissue. The VO2max performance parameter indicates the maximal oxidative metabolic capacity or oxygen supply integrating every step of transport and metabolic capacity of the body [26]. Reduced oxygen delivery at high altitude is responsible for VO2max limitation [27].
A KD influences VO2max by shifting mitochondrial metabolism capacity. Increasing fat rate oxidation requires greater oxygen consumption, thus leading to higher maximal oxygen supply for maintaining a given exercise load [2]. Some evidence suggests a positive effect of a KD on VO2max [21, 28–31] but is contrasted by recent work of Burke et al. [ 21]. At present, these mixed findings are believed to be due to heterogeneity across studies and/or variability among athletes [32]. Nevertheless, this aspect was never investigated under hypoxic conditions. Despite the mixed effect on VO2max, a KD could be a potential performance enhancer in hypoxia. In fact, optimising the fat oxidation rate could give access to the virtually endless fat store and reduce dependence on glycogen. This aspect is particularly important in long hypoxia training such as mountaineering. Furthermore, hypoxia is known to induce a reduction in CHOs oxidation when CHOs are ingested before exercise, thus reinforcing the use of fat at high altitudes [33, 34]. This observation is supported by the subjective benefit of a high-fat diet in high altitude praised by the extreme high-altitude mountaineer Erhard Loretan (e.g., at Everest base camp in 1986).
According to this information, we hypothesized that a 4-week KD would have positive effects on VO2max in healthy, recreational mountaineers during a maximal graded performance test under simulated hypoxic conditions. Various types of KD are described [3]. In our study, we focused on the isocaloric KD (ICKD) in which calories are in line with total energy expenditure.
## Methods
This pilot study was a single, blinded, randomised cross-over clinical trial. The study took place in Switzerland at the Clinique romande de réadaptation (Sion). The protocol was approved by CER-VD (Project-ID: 2020-00427).
## Study aim
This pilot study aimed to [1] assess the tolerance of a non-standardized 4-week ICKD in healthy, recreational mountaineers, [2] assess the possibility of evaluating participants’ ICKD performance benefits under hypoxic conditions by a maximal graded exercise bike test and [3] gather data regarding the benefit of a 4-week ICKD on VO2max during a maximal graded performance test under simulated hypoxic conditions. Furthermore, data concerning lactate threshold (LT) values (PLT, HRLT), peak values (Ppeak, HRpeak), subjective Borg rating of perceived exertion (RPE) and oxygenation status (blood gases) were recorded.
## Participants
Eight recreational mountaineers were recruited to participate. Inclusion criteria were familiarity with altitude (> 2500 m above sea level) and males/females 20–45 years old. Exclusion criteria were high training load (such as professional athletes) or new planned training and dietary restrictions. Participants were enrolled after giving their signed informed consent. Mountaineering level was reported as beginner for participants with no to little experience in mountaineering, medium for those who regularly experienced high altitudes, and experts who regularly mountaineered.
We did not calculate a sample size for this study because this was a pilot study to gather information about diet tolerance/acceptance, the feasibility of the testing procedure and preliminary data on the benefit of an ICKD diet for VO2max in healthy persons.
## Study procedure
Individuals who gave signed informed consent were invited for a consultation during which a medical doctor explained the standardisation of the performance tests and the dietary protocol. In addition, the participant’s ability to perform a maximal graded performance test was assessed with the physical activity aptitude questionnaire (Q-AAP) [35]. We assessed participants’ ability for a ICKD and exposure to hypoxia by two other self-developed questionnaires using known contra-indication to KD [36], previously experienced exposure to hypoxia and predisposing factors to acute mountain sickness [37].
The participant was then asked to perform a maximal graded exercise bike test (performance test) to assess their baseline performance under normoxic conditions (T0N) and, 4 weeks later, under hypoxic conditions (T0H). This test was conducted under supervision by a sport scientist and a medical doctor. The medical doctor was also responsible for collecting an arterial blood sample after a 5-min rest post-exercise.
After this test, participants were randomly assigned to group A or group B with a block size of 4 by a collaborator who was not involved in the study protocol. The RALLOC function of Stata was used. Each participant received a with the group attribution, which allowed for blind the investigators. Group A began with a 4-week UD (T1) followed by a 4-week ICKD (T2), and group B began with a 4-week ICKD (T1) followed by a 4-week UD (T2).
Each 4-week diet period was terminated with a performance test under hypoxic conditions (same procedure as at baseline but under hypoxic conditions). We used a normobaric (940–980 hPa) hypoxic (FIO2 = 12.7–$12.9\%$) room simulating an altitude of ~ 4500 m. The study design is represented in Fig. 1.Fig. 1Study design
## Maximal graded exercise bike test
The initial resistance was 60–90 W depending on training status and sex. The resistance was increased every 3 min by 30 W until exhaustion. Interruption conditions were a clear decrease in cycling frequency (< 70 min−1) or a complete stop of cycling [38]. Oxygen respiratory flow (VO2), carbon dioxide respiratory flow (VCO2), heart rate (HR) and delivered power (P) were measured (Metalyzer 3B, Cortex) until the test’s interruption. Maximal values are defined as peak values. In addition, capillary blood lactate (B-Lac) level was measured 30 s before the end of each increase in resistance. Finally, we used Borg RPE [39] at the end of the test to assess participant subjective perceived exertion of the physical work.
Hypoxia during the performance test was achieved in a simulated altitude facility (hypoxic room) with a hypoxic generator (ATS altitude, Sydney, Australia) by lowering the fraction of inspired oxygen (FIO2), which simulates a very-high-altitude of 4500 m.
## Four-week ICKD
The ICKD definition was based on the works of Sansone et al. [ 2] and Trimboli et al. [ 3]. We defined ICKD as a daily CHOs ingestion < 30–50 g d−1, without any limitation in fat consumption. Participants self-selected their own diet based on a list of advised and forbidden foods developed to fit the definition of ICKD. There were no instructions to limit calories. Regular contact was maintained with all participants during the study with every time available to answer participants’ dietary questions. The maximal graded bike test was planned after an adaptation period of 27 days.
## Four-week UD
There were no limitations on food consumption. Participants were instructed to eat as close as possible to their usual diet.
Participants were instructed to track their 4-week ICKD and UD by using the analysis program MyFoodRepo© (EPFL, Switzerland), a user-friendly smartphone application. The database used by the application is based on Switzerland’s foods and is in constant development. Food intake was manually reported on a daily basis by an investigator for diet monitoring. Participant adherence to ICKD was checked by their daily CHOs intake. If daily CHOs intake was > 50 g and if the β-hydroxybutyrate (β-OHB) level was < 170 μmol L−1, data for the participant were excluded from the statistical data analysis and the study.
## Blood analysis
An arterial blood sample was taken after a 5-min rest at the end of each performance test. Under hypoxic conditions, the participants remained in the hypoxic room for blood collection. The blood samples were analysed on an ABL800 FLEX blood gas analyzer (Radiometer, Denmark) within 10 min for partial arterial oxygen pressure (PaO2 [kPa]), partial arterial carbon dioxide pressure (PaCO2 [kPa], pH, and bicarbonate concentration (HCO3− [mmol L−1]) automatically calculated by using the Henderson–Hasselbalch equation.
Venous blood was sampled after each test. Venous blood was placed into a perchloric acid-tube and frozen at − 80 °C. All samples were analyzed within 3 months for β-OHB and acetoacetate (AcAc) with enzymatic analysis [40]. Reference values (percentile 2.5–97.5) provided by the laboratory (Lausanne university hospital, Switzerland) were for β-OHB, 58–170 μmol L−1, and AcAc, 18–78 μmol L−1.
## Statistical analysis
We used descriptive statistical analysis for [1] diet tolerance, [2] performance values under normoxic and hypoxic conditions and [3] performance values under a UD and ICKD. Performance test results under normoxic and hypoxic conditions are expressed as mean (SD). Because of limited sample size ($$n = 6$$), no confidence intervals or p-values were calculated.
To assess diet tolerance, we analysed the number of dropouts and reported side effects. We assessed diet tolerance by calculating median daily values for CHOs, fat and protein content and energy expenditure for all participants, then calculated median daily CHOs, fat, and protein consumption and energy expenditure for all participants. Further blood analysis of KBs including β-OHB and AcAc were assessed for diet tolerance.
Feasibility of maximal graded exercise was assessed by comparing the values under normoxia and hypoxia. These values were then expressed as percentage difference in “median” (“minimal values” to “maximal value”). We then checked whether these values agreed with previously reported performance decreases under acute hypoxia [41–43].
The effect of ICKD on performance is also expressed as percentage differences in median (range), calculated by comparing the values of performance parameters after the ICKD diet and the UD diet. For Group A, UD values are T1. For Group B, UD values are T0H. VO2max performance parameters after ICKD (hypoxia) compared to UD (hypoxia) were the primary outcome. As a secondary outcome, performance parameters such as LT values (PLT, HRLT), peak values (Ppeak, HRpeak), subjective values (Borg RPE) and oxygenation status (blood gas values) were analysed. LT determination was based on the Dmax model established in 1992 [44], which uses the maximal perpendicular distance from the line connecting the start with the endpoint of the lactate curve. We used the “modified Dmax threshold” (Dmod), which is an updated Dmax model by Bishop et al. [ 45]. This model eliminates the effect of start intensity and maximal effort and determines the LT as the moment when a rapid change in the inclination of the blood lactate curve occurs. This situation matches the maximal lactate steady state reflecting the anaerobic threshold [46].
## Diet tolerance
At the beginning, 8 participants were enrolled (4 males). One participant dropped out after the second performance test because of a schedule mismatch rather than a regimen intolerance. In addition, the data of another participant were excluded from data analysis because of abnormally high daily CHOs intake (> 50 g d−1) and low β-OHB level (< 170 µmol/L), which led to suspecting invalid data (Fig. 2).Fig. 2Flow of participants in the study. ICKD: isocaloric ketogenic diet; UD: usual mixed Western-style diet; CHOs: carbohydrates; β-OHB: β-hydroxybutyrate level Participants were aged 24–44 years (median 29 years) and the median BMI was 22.8 kg m−2.
Frequently reported side effects were weight loss and gastrointestinal disorders. After ICKD, subjective exercise load perception by the RPE scale [39] were similar to reference values for two participants, increased for three participants and decreased for one participant. Nutrition features with the UD and ICKD are presented in Table 1.Table 1Nutrition features of participants with a usual mixed Western-style diet (UD) and isocaloric ketogenic diet (ICKD)UDICKDMedianMinMaxMedianMinMaxCHOs (g d−1)197152321402149Protein (g d−1)82511229751169Fat (g d−1)936314014343164Energy (Kj d−1)7774636612,603849644759841β-OHB (μmol/L)44.041.065.0288.0190.0462.0AcAc (μmol/L)22.56.733.080.059.0132.0Recorded data for UD and ICKD diet characteristics with median (range) showing adherence to the ICKD diet. Reference values (percentile 2.5–97.5) provided by the laboratory (CHUV, Switzerland) are for β-OHB, 58–170 μmol L−1, and for AcAc, 18–78 μmol L−1. CHOs carbohydrates, β-OHB β-hydroxybutyrate, AcAc acetoacetate All participants except one could complete the ICKD. The recorded CHOs values for this participant were beyond the limitation and therefore the data were excluded from analysis. This situation emphasizes the need for a strict diet follow-up and blood KB analysis. Another participant dropped out because of a personal schedule mismatch, which underlines the importance of high motivation and collaboration between researchers and participants in particular because of the high number of maximal graded exercise bike tests with standardized time between the tests. Overall, only minor adverse events such gastrointestinal complaints at the beginning of the diet or weight loss were reported, but none of the participants had to stop the ICKD.
The median energy intake for UD and ICKD were below the typical energy requirements [47], so participants were in negative energy balance.
Moreover, we identified two major difficulties for participants in following the ICKD. First, high exercise load training was difficult during the first weeks of the ICKD diet. Indeed, we found a subjective performance drop at the beginning of the new diet because of ICKD adaptation [48]. Second, participants also reported a subjective social impact during the ICKD.
## UD and ICKD follow-up
The 4-week UD diet was characterized by a median CHOs intake of 197 g d−1. The median post-UD β-OHB level 44.0 μmol/L and AcAc level 22.5 μmol/L were within reference values. With the 4-week ICKD, the median CHOs intake was 40 g d−1. The median post-4-week ICKD β-OHB value 288.0 μmol/L and AcAc value 80.0 μmol/L were above reference values fixed by the laboratory.
One participant showed excess CHOs daily intake and too low β-OHB blood level. His median post-4-week UD CHOs intake was 259 g d−1 and 4-week ICKD intake 67 g d−1. The β-OHB/AcAc blood content was $\frac{41.0}{23.0}$ and $\frac{152.0}{56.0}$ mmol L−1. This participant was not included in the statistical analysis.
## Feasibility of the maximal graded exercise bike test under hypoxia
We used expected and previously characterised performance decreases under hypoxia [41–43] to evaluate the feasibility of the performance test and trustworthiness of the recorded values. We found no adverse events related to the combination of a 4-week ICKD, hypoxia exposure and maximal graded exercise bike test.
## Effect of hypoxia
The results of the performance test under normoxic conditions at baseline are presented in Table 2. The measured effect of hypoxia is expressed as the median difference with normoxic values in percentages. Median performance values decreased for VO2max (− $27.1\%$), PLT (− $28.6\%$) and Ppeak (− $22.4\%$) under hypoxic conditions. Arterial blood samples showed a reduction in median PaO2 by − $50.9\%$. Median values for pH, PaCO2 and HCO3- all remained unaffected. Table 2Effect of hypoxia alone on performance valuesNormoxiaEffect of hypoxia (%)Mean (SD)MedianMinMaxVO2max (ml kg−1 min−1)44.6 (5.8) − 27.1 − 36.5 − 20.3PLT (W)203 [64] − 28.6 − 34.6 − 17.9Ppeak (W)266 [65] − 22.4 − 30.8 − 12.8HRLT (beat min−1)156 [23]2.0 − 7.117.6HRpeak (beat min−1)183 [17] − 5.4 − 15.0 − 1.9B-Lacpeak (mmol L−1)10.7 (2.2) − 0.8 − 2816.1PaO2 (kPa)17.8 (2.2) − 50.9 − 71.1 − 12.2pH7.24 (0.06)0.15 − 0.592.22PaCO2 (kPa)3.8 (0.2) − 9.6 − 19.55.9HCO3− (mmol L−1)11.9 (2.1)0.0 − 0.20.1Maximal graded exercise test values under normoxic conditions for baseline were compared with performance values under hypoxic conditionsNormoxia baseline values (T0N) are compared with hypoxia values (T0H) assessing the effect of hypoxia. VO2max maximal oxygen uptake, LT lactate threshold, P power, HR heart rate, B-Lac blood lactate, PaO2 partial oxygen arterial pressure, PaCO2 partial carbon dioxide arterial pressure For every participant, hypoxia induced a clear decrease in median VO2max by 13 ml kg−1 min−1 (− $27.01\%$), Ppeak by 60 W (− $22.4\%$), and PaO2 by 8.8 kPa (− $50.9\%$) (Table 2). This was expected because the primary limitation for VO2max under hypoxic conditions is oxygen tissue availability [49]. Indeed, significant decreases in VO2max were previously reported with acute exposure to hypoxia [41–43]. This situation results from decrease in barometric pressure with increasing altitude. Therefore PIO2 and, consequently, oxygen transport decrease [41, 50]. A research compilation by Robergs and Robert [51] reported a mean decrease of $8.7\%$ per 1000 m in VO2max. Nevertheless, an average value cannot be expressed because of high inter-individual variation in the reduction in VO2max depending on sea level VO2max, sex, sea level LT and lean body mass [27]. This observation may explain the large range in VO2max decrease (− 36.5 to − $20.3\%$) in our study. Furthermore, we found no significant difference in arterial blood pH, PaCO2 and HCO3.
## ICKD effect on performance
Performance test results under hypoxic conditions at baseline are in Table 3. Keeping the cross-over design, hypoxic baseline values (UD) are at T1 for Group A and T0H for group B. The effect of ICKD on performance test values was assessed by comparing parameter values at T1 with T2 (Group A) and T0H with T1 (Group B). A return to the normal situation could be assessed only in group B by comparing the performance values of T0H and T2.Table 3Effect of ICKD under hypoxic conditions on performance valuesHypoxiaEffect of ICKD (%)Mean (SD)MedianMinMaxVO2max (ml kg−1 min−1)31.6 (3.4)7.3 − 16.825.5PLT (W)147 [47]0.7 − 21.546.6Ppeak (W)205 [48]4.7 − 7.111.1HRLT (beat min−1)160 [22]7.5 − 5.423.8HRpeak (beat min−1)171 [15]3.6 − 2.615.2B-Lacpeak (mmol L−1)10.3 (3.0) − 8.8 − 40.917.1PaO2 (kPa)9.0 (3.4) − 14.5 − 32.1 − 11.5pH7.28 (0.96)0.280.081.62PaCO2 (kPa)3.5 (0.4)2.1 − 23.311.8HCO3− (mmol L−1)12.0 (3.1) − 0.1 − 0.30.1Hypoxic (T0H) performance values were used for baseline and were compared with performance values from a maximal graded exercise test performed after a 4-week ICKD under hypoxic conditionsICKD isocaloric ketogenic diet, UD usual mixed Western diet, VO2max maximal oxygen uptake, LT lactate threshold, P power, HR heart rate, B-Lac blood lactate, PaO2 partial oxygen arterial pressure, PaCO2 partial carbon dioxide arterial pressure
## Effect of ICKD
Median performance values increased for VO2max (+ $7.3\%$), Ppeak ($4.7\%$), HRpeak (+ $3.6\%$) and PLT (+ $0.7\%$) but with large interindividual variability (Table 3). Median bLabase and bLapeak values decreased by − $21.0\%$ and − $8.8\%$. Median PaO2 decreased − $14.5\%$. Values for other parameters, pH, PCO2 and HCO3, could be considered stable.
In Group B, the cross-over design allowed for assessing a return to the normal situation when switching again from a 4-week ICKD to a 4-week UD. So we compared T0H and T2 values for the three participants in group B, which should theoretically be the same values. We expressed results as a percentage difference between T0H and T2. Of note, only median PaO2 (+ $1.4\%$) returned to the initial value. We did not observe a return to initial values for median VO2max (+ $18.1\%$), Ppeak (− $8.3\%$), or PLT (+ $3.3\%$).
## Individual values
Individual performance and blood gas parameters at baseline (T0N) after UD and after ICKD are in Table 4. The 6 participants could be separated as showing a positive effect of ICKD or not (Table 5). Four participants showed an increase in VO2max and Ppeak. PLT was used for further dividing participants. Figure 3 shows B-Lac and HR curves after UD or ICKD, as well as at T0N for every participant. Nr. 6 (Id 6) and Nr. 8 (Id 8) showed improvement in endurance and peak performance parameters. Nr. 1 (Id 1) and Nr. 2 (Id 2) showed improvement in peak performance parameter. Nr. 3 (Id 3) and Nr. 4 (Id 4) showed little to no response or a clear worsening of the performance parameter. One participant (no. 2) showed a particularly large decrease in HRpeak (− $14.9\%$) with hypoxia exposure. ICKD clearly allowed for a return to normality for HRpeak values (+ $13.1\%$) as also confirmed by the cross-over analysis (− $14.2\%$).Table 4Values for individual participants comparing performance after a UD and after an ICKDTestNr. 1Nr. 2Nr. 3Nr. 4Nr. 6Nr. 8SexFMMMFFAge (years)292744242928BMI (kg m−2)22.323.224.623.421.319.9Mountaineering levelMediumMediumMediumExpertMediumExpertVO2max (ml kg−1 min−1)T0N36.848.846.250.738.047.2UD26.631.029.542.828.532.3ICKD28.838.924.538.433.634.4PLT (W)T0N114261266224136215UD8118618215989103ICKD79146168165124151Ppeak (W)T0N180310340300195270UD140240194210123190ICKD150250186195140200HRLT (beat min−1)T0N163175136174119170UD163186129173140147ICKD166176140184159182HRpeak (beat min−1)T0N194194162197162191UD186165152190159184ICKD190190148193167194bLapeak (mmol L−1)T0N9.213.611.811.87.510.1UD8.213.19.312.55.410.7ICKD9.610.75.511.75.59.5RPET0N101717171117UD112015161318ICKD152015171516PaO2 (kPa)T0N1813.61925.51416.9UD15.87.07.711.78.18.4ICKD7.76.06.87.96.07.2pHT0N7.227.147.267.267.347.24UD7.217.217.427.207.387.28ICKD7.257.337.427.227.457.29PaCO2 (kPa)T0N3.63.93.73.54.04.0UD2.93.43.23.73.83.7ICKD3.03.83.52.93.93.1HCO3− (mmol L−1)T0N10.89.51211.315.612.3UD8.4101510.516.512.4ICKD9.414.716.88.520.210.9Only VO2max, PLT and Ppeak values are shown for baseline T0NT0N baseline test under normoxia, UD values after usual mixed Western-style diet under hypoxia, ICKD isocaloric ketogenic diet under hypoxia, VO2max maximal oxygen uptake, LT lactate threshold, P power, HR heart rate, bLa blood lactate, RPE Borg rating of perceived exertion, PaO2 partial oxygen arterial pressure, PaCO2 partial carbon dioxide arterial pressure, Nr. subject numberTable 5Effect of ICKD on performance parametersTypeNrVO2maxPLT and HRLTPpeakRemarkPositive effect of ICKD6;8↑↑↑ICKD conferred improvement regarding endurance and peak performance parameters1;2↑↓↑ICKD had positive effects despite a decrease in LT parameter values. Indeed, Ppeak and VO2max confirmed a benefitNegative effect of ICKD4↓↑↓ICKD conferred little to no response. UD and ICKD values for performance tests can be considered equalNon-assessable3↓↓↓Clear worsening of test performanceICKD isocaloric ketogenic diet, UD usual mixed Western-style diet, VO2max maximal oxygen uptake (ml min−1 kg−1), PLT power at lactate threshold (W), Ppeak peak power (W), HRLT heart rate at lactate threshold (beat min−1), Nr. subject numberFig. 3Effect of ICKD on performance test. ICKD B-Lac and HR curves were drawn for T0N, UD, and ICKD. The first point represents the values at the beginning of the test. The second point is the value at lactate threshold and the third point is the maximal or peak value. ID: participant number
## Discussion
This novel study assessed the tolerance of an ICKD and the feasibility of a bike performance test under hypoxic conditions and gathered preliminary data on the effect of shifting from a UD to an ICKD on VO2max under simulated very-high-altitude conditions. We used a maximal graded exercise bike test to assess endurance parameters and post-exercise arterial blood samples to assess oxygenation and pH status. We hypothesized that reduced CHOs intake would increase VO2max performance values under hypoxia. To our knowledge, these aspects were never investigated before. We analysed data for all 6 participants. Mean VO2max in the normoxic condition was 44.6 ml kg−1 min−1. Hypoxia led to decreased performance in all participants. With the ICKD diet, median values for PaO2 decreased by − $14.5\%$ and VO2max by + $7.3\%$ and Ppeak by + $4.7\%$.
## Evaluating the ICKD benefits on performance under hypoxic conditions
We evaluated ICKD-induced performance benefits under hypoxic conditions by using a maximal graded exercise bike test. The observed decrease in median values for the performance parameters VO2max, PLT, Ppeak and PaO2 when exposed to hypoxia strengthen the reliability of results and the feasibility of our protocol.
## Effect of ICKD on maximal graded exercise test
Improved median VO2max (+ $7.3\%$) and Ppeak (+ $4.7\%$) with a 4-week ICKD under hypoxic conditions in 4 of 6 participants (Tables 3, 4) do not allow for concluding on the effects of the KD. We found no performance marker systematically affected by ICKD. This recorded improvement was similarly reported in part under normoxia [21, 32]. VO2max reflects the cardiorespiratory fitness [52] and is considered a gold standard for measuring aerobic metabolism [53]. A greater VO2max indicates greater endurance capacity. Actual known factors affecting VO2max are age, sex, genetics, body composition, state of training and mode of exercise [54].
The KD has been newly identified as a potential positive factor for VO2max [28, 30, 31] by shifting mitochondrial metabolism [17, 19]. These studies were summarized by Bailey et al. [ 32]. The authors suggested that several factors such as genetics, trainability and or chronic substrate utilization may be affected by KD, which thus might increase VO2max. The positive effect of the ICKD on VO2max under hypoxic conditions we observed strengthens the idea that a high-fat diet might be beneficial for endurance exercise under acute hypoxia exposure.
In addition, other performance LT values were improved in two participants considered ICKD responders (Table 5). As shown in Fig. 3, B-Lac kinetics were lower with the ICKD than UD. Points for LT and maximal value shifted to the right (right curve-shifting). HR kinetics were higher with the ICKD than UD. Points for LT and maximal values were at higher power (left curve-shifting). LTs are performance indicators strongly correlated with endurance performance [46]. They represent the aerobic–anaerobic transition because lactate kinetics are highly related to the metabolic rate and less to oxygen availability [46]. A higher workload to a given blood lactate concentration can be interpreted as improved endurance capacity [55]. Previous studies have reported a shift in the B-Lac curve to higher workloads under KD conditions [56]. This phenomenon is still not fully understood but may be due to an association with decreased glycolysis rate or limited lactate efflux from muscle due to reduced blood buffering capacity [28, 56]. Furthermore, depleted glycogen stores (due to an ICKD, for example) is a known factor leading to lower blood lactate concentration at the same work rate [57].
In this context and considering key performance parameters (VO2max, PLT, HRLT and Ppeak), we separated the 6 participants into two groups: a group showing a positive effect of the ICKD and a group showing no benefits or even negative effects of the ICKD. The group with a positive effect of the ICKD showed a right-shift in B-Lac kinetics (Fig. 3) which can be interpreted as better performance [55]. Peak values (Ppeak and VO2max) also demonstrated better performance. A left-shifting HR curve showed a higher HR work range with the ICKD. For the second group, the intervention had little to no effect, and ICKD and UD tests were considered equal. One participant showed clearly worsened global ICKD performance at every time point, with no clear explanation. Several factors that could explain this performance decrease include a “bad shape day” or the influence of the ICKD on age-related VO2max factors such as a decline in maximal heart rate, stroke volume, fat-free mass and arterio-veinous oxygen differences [58].
## Effect of ICKD on blood gas parameters
With the ICKD, post-exercise PaO2 decreased in all participants (median − $14.5\%$), a response confirmed in the cross-over group B when returning to the UD (+ $14.5\%$). A physiological approach can explain the ICKD-related hypoxemia. PaO2 is determined by alveolar PO2 (PAO2), ventilation, diffusion capacity of the lung and perfusion by the heart. PAO2 depends on the respiratory gas-exchange ratio (RER). RER is the ratio between CO2 pulmonary output (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}CO_{2}$$\end{document}V˙CO2) and O2 uptake (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}O_{2}$$\end{document}V˙O2) expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$RER = \frac{{\dot{V}CO_{2} }}{{\dot{V}O_{2} }}$$\end{document}RER=V˙CO2V˙O2, which can be included in the alveolar air equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$PAO_{2} = PIO_{2} - (\frac{{PaCO_{2} }}{RER})$$\end{document}PAO2=PIO2-(PaCO2RER). RER depends on the steady state (e.g., resting state) for the food metabolized [59]. An ICKD increases fat oxidation and lowers RER (~ 0.7), and consequently PAO2 and PaO2 decrease [59]. Furthermore, the maximal graded test result is not a steady state, and RER varies with exercise. Within a few minutes into recovery, RER decreases to < 0.7 as ventilation declines and the CO2 store re-increases [60]. Decreased ventilation could also affect PaO2, as observed in respiratory failure, a well-known process in diabetic ketoacidosis. *Ketosis* generates a respiratory response in the form of hyperpnea [61], which leads to respiratory muscle fatigue (known as Kussmaul respiration) [62]. An ICKD could lead to a mismatch of the lung maintaining PaO2 by a form of respiratory muscle fatigue due to KBs.
Alternative explanations for the effect of heart perfusion on PaO2 are not relevant. Although KBs can influence heart flow, they increase rather than decrease the hydraulic efficiency of the heart [18]. The higher heart flow rate leads to an increase in pulmonary venous blood admission. Finally, CHOs are known to increase PaO2 at high altitude by increasing the relative production of carbon dioxide and increasing the drive for ventilation [63].
Consistent with a previous study by Hansen et al. [ 60], ICKD worsened the hypoxemia in our simulated very-high to extreme altitude (> 3500 m). This is a relevant limitation of using the ICKD above a very-high-altitude [50, 64], whereas at high-altitude (1500–3500 m), PaO2 is significantly diminished but with only minor impairments in oxygen transport (SaO2 > $90\%$) [50]. Therefore, an ICKD could be used for this altitude.
Arterial blood sampling also showed non-significant effects of the ICKD on pH (+ $0.278\%$) post-exercise, which is below analytical precision. Increased pH could be expected because KBs are acids known to induce ketoacidosis [65, 66]. Nevertheless, Carr et al. contrasted these earlier beliefs by describing the minimal effects of KD on acid base status in elite athletes [30]. They also reported no statistical blood pH differences between high CHOs versus high fat content pre- and post-exercise. Blood pH stability may be due to increased exercise-induced ventilation rate and so would increase the pH.
In conclusion, our preliminary data showed a benefit of ICKD on performance parameters. We found a positive increase in VO2max (primary outcome) and LT performance parameters (secondary outcome). ICKD intervention decreased PaO2, which is consistent with previous findings [60] and may limit the use of an ICKD in very-high-altitude sports.
## Further research
Further characterisation of the ICKD benefits at high-altitude (1500–3500 m) is needed. Despite ICKD-induced hypoxemia, this may not impair oxygen delivery at this altitude. Furthermore, diet modifications (CHOs vs. fat) could be an interesting path for improving acclimation or performance at high-altitude and preventing acute mountain sickness. PaO2 analysis during the whole effort would be needed for a complete assessment of the effect of an ICKD on blood gas values. The cross-over design of the study is pertinent to assess the effectiveness of an ICKD.
The previous experience of participants with KD-like diets seems to be an effective supplementary inclusion criterion for diet tolerance. Moreover, whether subjective perception of KD tolerance matches performance improvements would be interesting and might help predict whether a person would show a positive effect of ICKD or not. This would be possible with a simple assessment by a questionnaire without a maximal graded performance test.
## Strengths and limitations
Our study assessed for the first time the effect of KD implementation in simulated very-high-altitude performance test. We used a maximal graded exercise bike test with a primary outcome of the effect on VO2max. However, with a sample of 6 participants, we focused on individual values, profiling, and trends. The negative energy balance observed without recorded data on body mass at the end of KD limits concluding on its effect. In addition, our design also limits the blood gas kinetics view during the performance test. Moreover, our findings were obtained in acute simulated hypoxia, which limits practical implications for the real-life high-altitude condition.
## Conclusions
The present protocol shows the feasibility of evaluating the benefits of an ICKD on recreational athlete performance by a maximal graded exercise test under hypoxia conditions. Our study successfully combined ICKD, hypoxia and maximal graded exercise, which shows the feasibility of the present protocol. The pilot data showed improved VO2max with the ICKD under hypoxia in 4 participants. Nevertheless, this study does not allow to make any final conclusions about the benefit of ICKD. KD remains an interesting alternative to CHOs dependency for endurance performance. This regimen may be interesting for endurance exercises at high-altitude (1500–3500 m) [67] with only minor impairment in arterial oxygen transport [50]. It could concern typical high-altitude exercises such as high-altitude training [67], high-altitude races [68], trail-running, mountaineering and ski-touring [69].
## References
1. Costa RJS, Hoffman MD, Stellingwerff T. **Considerations for ultra-endurance activities: part 1-nutrition**. *Res Sports Med* (2019.0) **27** 166-181. DOI: 10.1080/15438627.2018.1502188
2. Sansone M, Sansone A, Borrione P, Romanelli F, DiLuigi L, Sgrò P. **Effects of ketone bodies on endurance exercise**. *Curr Sports Med Rep* (2018.0) **17** 444-453. DOI: 10.1249/JSR.0000000000000542
3. Trimboli P, Castellana M, Bellido D, Casanueva FF. **Confusion in the nomenclature of ketogenic diets blurs evidence**. *Rev Endocr Metab Disord* (2020.0). DOI: 10.1007/s11154-020-09546-9
4. Westman EC, Mavropoulos J, Yancy WS, Volek JS. **A review of low-carbohydrate ketogenic diets**. *Curr Atheroscler Rep* (2003.0) **5** 476-483. DOI: 10.1007/s11883-003-0038-6
5. Volek JS, Noakes T, Phinney SD. **Rethinking fat as a fuel for endurance exercise**. *Eur J Sport Sci* (2015.0) **15** 13-20. DOI: 10.1080/17461391.2014.959564
6. Christensen EH, Hansen O. **Arbeitsfähigkeit und Ernährung1**. *Skand Arch Physiol* (1939.0) **81** 160-171. DOI: 10.1111/j.1748-1716.1939.tb01320.x
7. Bergström J, Hultman E. **Muscle glycogen synthesis after exercise : an enhancing factor localized to the muscle cells in man**. *Nature* (1966.0) **210** 309-310. DOI: 10.1038/210309a0
8. Bergström J. **Diet, muscle glycogen and physical performance**. *Acta Physiol Scand* (2021.0). DOI: 10.1111/j.1748-1716.1967.tb03720.x
9. Helge JW. **A high carbohydrate diet remains the evidence based choice for elite athletes to optimise performance**. *J Physiol* (2017.0) **595** 2775. DOI: 10.1113/JP273830
10. Phinney SD, Horton ES, Sims EAH, Hanson JS, Danforth E, Lagrange BM. **Capacity for moderate exercise in obese subjects after adaptation to a hypocaloric, ketogenic diet**. *J Clin Invest* (1980.0) **66** 1152-1161. DOI: 10.1172/JCI109945
11. 11.Ross AC, Caballero BH, Cousins RJ, Tucker KL, Ziegler TR. Modern nutrition in health and disease. Eleventh edn. Wolters Kluwer Health Adis (ESP); 2012. https://jhu.pure.elsevier.com/en/publications/modern-nutrition-in-health-and-disease-eleventh-edition.
12. Achten J, Jeukendrup AE. **Optimizing fat oxidation through exercise and diet**. *Nutrition* (2004.0) **20** 716-727. DOI: 10.1016/j.nut.2004.04.005
13. Prins PJ, Noakes TD, Welton GL, Haley SJ, Esbenshade NJ, Atwell AD. **High rates of fat oxidation induced by a low-carbohydrate, high-fat diet, do not impair 5-km running performance in competitive recreational athletes**. *J Sports Sci Med* (2019.0) **18** 738-750. PMID: 31827359
14. Hu XF, Kenny TA, Chan HM. **Inuit country food diet pattern is associated with lower risk of coronary heart disease**. *J Acad Nutr Diet* (2018.0) **118** 1237-1248.e1. DOI: 10.1016/j.jand.2018.02.004
15. Harvey KL, Holcomb LE, Kolwicz SC. **Ketogenic diets and exercise performance**. *Nutrients* (2019.0) **11** 2296. DOI: 10.3390/nu11102296
16. Venables MC, Achten J, Jeukendrup AE. **Determinants of fat oxidation during exercise in healthy men and women: a cross-sectional study**. *J Appl Physiol* (2005.0) **98** 160-167. DOI: 10.1152/japplphysiol.00662.2003
17. Volek JS, Freidenreich DJ, Saenz C, Kunces LJ, Creighton BC, Bartley JM. **Metabolic characteristics of keto-adapted ultra-endurance runners**. *Metabolism* (2016.0) **65** 100-110. DOI: 10.1016/j.metabol.2015.10.028
18. Sato K, Kashiwaya Y, Keon CA, Tsuchiya N, King MT, Radda GK. **Insulin, ketone bodies, and mitochondrial energy transduction**. *FASEB J* (1995.0) **9** 651-658. DOI: 10.1096/fasebj.9.8.7768357
19. Anderson KA, Hirschey MD. **Mitochondrial protein acetylation regulates metabolism**. *Essays Biochem* (2012.0) **52** 23-35. DOI: 10.1042/bse0520023
20. Cotter DG, Schugar RC, Crawford PA. **Ketone body metabolism and cardiovascular disease**. *Am J Physiol Heart Circ Physiol* (2013.0) **304** H1060-H10766. DOI: 10.1152/ajpheart.00646.2012
21. Burke LM, Ross ML, Garvican-Lewis LA, Welvaert M, Heikura IA, Forbes SG. **Low carbohydrate, high fat diet impairs exercise economy and negates the performance benefit from intensified training in elite race walkers**. *J Physiol (Lond)* (2017.0) **595** 2785-2807. DOI: 10.1113/JP273230
22. Kreider RB, Wilborn CD, Taylor L, Campbell B, Almada AL, Collins R. **ISSN exercise & sport nutrition review: research & recommendations**. *J Int Soc Sports Nutr* (2010.0) **7** 7. DOI: 10.1186/1550-2783-7-7
23. Michalczyk M, Czuba M, Zydek G, Zając A, Langfort J. **Dietary recommendations for cyclists during altitude training**. *Nutrients* (2016.0) **8** 377. DOI: 10.3390/nu8060377
24. Burke LM, Cox GR, Cummings NK, Desbrow B. **Guidelines for daily carbohydrate intake**. *Sports Med* (2001.0) **31** 267-299. DOI: 10.2165/00007256-200131040-00003
25. 25.West JB, Luks AM, Schoene RB, Milleedge JS. 15. Nutrition, metabolisme and intestinal function. In: High altitude medicine and physiology. 5 edn.
26. Wagner PD. **New ideas on limitations to VO2max**. *Exerc Sport Sci Rev* (2000.0) **28** 10-14. PMID: 11131681
27. Robergs RA, Quintana R, Lee Parker D, Frankel CC. **Multiple variables explain the variability in the decrement in VO**. *Med Sci Sports Exerc* (1998.0) **30** 869-879. PMID: 9624645
28. Zajac A, Poprzecki S, Maszczyk A, Czuba M, Michalczyk M, Zydek G. **The effects of a ketogenic diet on exercise metabolism and physical performance in off-road cyclists**. *Nutrients* (2014.0) **6** 2493-2508. DOI: 10.3390/nu6072493
29. Langfort J, Pilis W, Zarzeczny R, Nazar K, Kaciuba-Uściłko H. **Effect of low-carbohydrate-ketogenic diet on metabolic and hormonal responses to graded exercise in men**. *J Physiol Pharmacol* (1996.0) **47** 361-371. PMID: 8807563
30. Carr AJ, Sharma AP, Ross ML, Welvaert M, Slater GJ, Burke LM. **Chronic ketogenic low carbohydrate high fat diet has minimal effects on acid-base status in elite athletes**. *Nutrients* (2018.0) **10** 236. DOI: 10.3390/nu10020236
31. Heatherly AJ, Killen LG, Smith AF, Waldman HS, Seltmann CL, Hollingsworth A. **Effects of Ad libitum low-carbohydrate high-fat dieting in middle-age male runners**. *Med Sci Sports Exerc* (2018.0) **50** 570-579. DOI: 10.1249/MSS.0000000000001477
32. Bailey CP, Hennessy E. **A review of the ketogenic diet for endurance athletes: performance enhancer or placebo effect?**. *J Int Soc Sports Nutr* (2020.0) **17** 33. DOI: 10.1186/s12970-020-00362-9
33. Margolis LM, Wilson MA, Whitney CC, Carrigan CT, Murphy NE, Radcliffe PN. **Acute hypoxia reduces exogenous glucose oxidation, glucose turnover, and metabolic clearance rate during steady-state aerobic exercise**. *Metabolism* (2020.0) **103** 154030. DOI: 10.1016/j.metabol.2019.154030
34. Young AJ, Berryman CE, Kenefick RW, Derosier AN, Margolis LM, Wilson MA. **Altitude acclimatization alleviates the hypoxia-induced suppression of exogenous glucose oxidation during steady-state aerobic exercise**. *Front Physiol* (2018.0). DOI: 10.3389/fphys.2018.00830/full
35. Warburton DE, Jamnik VK, Bredin SS, McKenzie DC, Stone J. **Evidence-based risk assessment and recommendations for physical activity clearance: an introduction**. *Appl Physiol Nutr Metab* (2011.0). DOI: 10.1139/h11-060
36. Stafstrom CE, Rho JM. *Epilepsy and the ketogenic diet* (2004.0) 352
37. Lovis A, Duplain H, Nicod L, Scherrer U, Sartori C. **Maladies liées à l’altitude et consultation de médecine de montagne**. *Forum Med Suisse* (2012.0) **12** 783-789. DOI: 10.4414/fms.2012.01287
38. 38.Dr Micah Gross, Office fédéral du sport OFSPO Haute école fédérale de sport de Macolin HEFSM Département Sport de performance. Manuel de diagnostic de performance. 2015. https://www.swissolympic.ch/dam/jcr:6ea9b202-857b-4561-81de-bbc799b6b690/Diagnostic_de_performance_manual_160201_FR.pdf.
39. Williams N. **The Borg rating of perceived exertion (RPE) scale**. *Occup Med* (2017.0) **67** 404-405. DOI: 10.1093/occmed/kqx063
40. Williamson DH, Mellanby J, Bergmeyer HU. **D-(–)-3-Hydroxybutyrate**. *methods of enzymatic analysis* (1974.0) 1836-1839
41. Lawler J, Powers SK, Thompson D. **Linear relationship between VO2max and VO2max decrement during exposure to acute hypoxia**. *J Appl Physiol* (1988.0) **64** 1486-1492. DOI: 10.1152/jappl.1988.64.4.1486
42. Cymerman A, Reeves JT, Sutton JR, Rock PB, Groves BM, Malconian MK. **Operation Everest II: maximal oxygen uptake at extreme altitude**. *J Appl Physiol* (1989.0) **66** 2446-3253. DOI: 10.1152/jappl.1989.66.5.2446
43. Dill DB, Adams WC. **Maximal oxygen uptake at sea level and at 3,090-m altitude in high school champion runners**. *J Appl Physiol* (1971.0) **30** 854-859. DOI: 10.1152/jappl.1971.30.6.854
44. Cheng B, Kuipers H, Snyder AC, Keizer HA, Jeukendrup A, Hesselink M. **A new approach for the determination of ventilatory and lactate thresholds**. *Int J Sports Med* (1992.0) **13** 518-522. DOI: 10.1055/s-2007-1021309
45. Bishop D, Jenkins DG, Mackinnon LT. **The relationship between plasma lactate parameters, Wpeak and 1-h cycling performance in women**. *Med Sci Sports Exerc* (1998.0) **30** 1270-1275. DOI: 10.1097/00005768-199808000-00014
46. Faude O, Kindermann W, Meyer T. **Lactate threshold concepts: how valid are they?**. *Sports Med* (2009.0) **39** 469-490. DOI: 10.2165/00007256-200939060-00003
47. 47.Allowances NRC (US) S on the TE of the RD. Energy. Recommended Dietary Allowances: 10th Edition. National Academies Press (US); 1989 [cité 10 mars 2022]. https://www.ncbi.nlm.nih.gov/books/NBK234938/.
48. Phinney SD. **Ketogenic diets and physical performance**. *Nutr Metab (Lond)* (2004.0) **1** 2. DOI: 10.1186/1743-7075-1-2
49. Dempsey JA, Wagner PD. **Exercise-induced arterial hypoxemia**. *J Appl Physiol* (1999.0) **87** 1997-2006. DOI: 10.1152/jappl.1999.87.6.1997
50. Paralikar SJ, Paralikar JH. **High-altitude medicine**. *Indian J Occup Environ Med* (2010.0) **14** 6-12. DOI: 10.4103/0019-5278.64608
51. Robergs RA, Roberts S. *Exercise physiology: exercise, performance, and clinical applications* (1997.0)
52. Akalan C, Kravitz L, Robergs RR. **V̇O2max: essentials of the most widely used test in exercise physiology**. *ACSM’s Health Fitness J* (2004.0) **8** 5-9. DOI: 10.1097/00135124-200405000-00004
53. 53.Morton RH, Billat V. Maximal endurance time at VO2max. 9.
54. Wagner PD. **A theoretical analysis of factors determining VO**. *Respir Physiol* (1996.0) **106** 329-343. DOI: 10.1016/S0034-5687(96)00086-2
55. Yoshida T, Udo M, Chida M, Ichioka M, Makiguchi K, Yamaguchi T. **Specificity of physiological adaptation to endurance training in distance runners and competitive walkers**. *Eur J Appl Physiol* (1990.0) **61** 197-201. DOI: 10.1007/BF00357599
56. Helge JW. **Long-term fat diet adaptation effects on performance, training capacity, and fat utilization**. *Med Sci Sports Exerc* (2002.0) **34** 1499-1504. DOI: 10.1097/00005768-200209000-00016
57. McLellan TM, Gass GC. **The relationship between the ventilation and lactate thresholds following normal, low and high carbohydrate diets**. *Eur J Appl Physiol Occup Physiol* (1989.0) **58** 568-576. DOI: 10.1007/BF00418501
58. Rogers MA, Hagberg JM, Martin WH, Ehsani AA, Holloszy JO. **Decline in VO**. *J Appl Physiol* (1985.0) **68** 2195-2199. DOI: 10.1152/jappl.1990.68.5.2195
59. Bergman BC, Brooks GA. **Respiratory gas-exchange ratios during graded exercise in fed and fasted trained and untrained men**. *J Appl Physiol* (1999.0) **86** 479-487. DOI: 10.1152/jappl.1999.86.2.479
60. Hansen JE, Hartley LH, Hogan RP. **Arterial oxygen increase by high-carbohydrate diet at altitude**. *J Appl Physiol* (1972.0) **33** 441-445. DOI: 10.1152/jappl.1972.33.4.441
61. Gallo de Moraes A, Surani S. **Effects of diabetic ketoacidosis in the respiratory system**. *World J Diabetes* (2019.0) **10** 16-22. DOI: 10.4239/wjd.v10.i1.16
62. Pierce NF, Fedson DS, Brigham KL, Mitra RC, Sack RB, Mondal A. **The ventilatory response to acute base deficit in humans. Time course during development and correction of metabolic acidosis**. *Ann Intern Med* (1970.0) **72** 633-640. DOI: 10.7326/0003-4819-72-5-633
63. Lawless NP, Dillard TA, Torrington KG, Davis HQ, Kamimori G. **Improvement in hypoxemia at 4600 meters of simulated altitude with carbohydrate ingestion**. *Aviat Space Environ Med* (1999.0) **70** 874-878. PMID: 10503752
64. West JB. **High-altitude medicine**. *Am J Respir Crit Care Med* (2012.0) **186** 1229-1237. DOI: 10.1164/rccm.201207-1323CI
65. Kitabchi AE, Wall BM. **Diabetic ketoacidosis**. *Med Clin N Am* (1995.0) **79** 9-37. DOI: 10.1016/S0025-7125(16)30082-7
66. Greenhaff PL, Gleeson M, Maughan RJ. **The effects of diet on muscle pH and metabolism during high intensity exercise**. *Eur J Appl Physiol* (1988.0) **57** 531-539. DOI: 10.1007/BF00418458
67. Berglund B. **High-altitude training**. *Sports Med* (1992.0) **14** 289-303. DOI: 10.2165/00007256-199214050-00002
68. Praz C, Léger B, Kayser B. **Energy expenditure of extreme competitive mountaineering skiing**. *Eur J Appl Physiol* (2014.0) **114** 2201-2211. DOI: 10.1007/s00421-014-2939-1
69. Praz C, Fasel B, Vuistiner P, Aminian K, Kayser B. **Optimal slopes and speeds in uphill ski mountaineering: a laboratory study**. *Eur J Appl Physiol* (2016.0) **116** 1011-1019. DOI: 10.1007/s00421-016-3371-5
|
---
title: Radiomics nomogram for preoperative differentiation of pulmonary mucinous adenocarcinoma
from tuberculoma in solitary pulmonary solid nodules
authors:
- Junjie Zhang
- Ligang Hao
- MingWei Qi
- Qian Xu
- Ning Zhang
- Hui Feng
- Gaofeng Shi
journal: BMC Cancer
year: 2023
pmcid: PMC10029225
doi: 10.1186/s12885-023-10734-4
license: CC BY 4.0
---
# Radiomics nomogram for preoperative differentiation of pulmonary mucinous adenocarcinoma from tuberculoma in solitary pulmonary solid nodules
## Abstract
### Objective
To develop and validate predictive models using clinical parameters, radiomic features and a combination of both for preoperative differentiation of pulmonary nodular mucinous adenocarcinoma (PNMA) from pulmonary tuberculoma (PTB).
### Method
A total of 124 and 53 patients with PNMA and PTB, respectively, were retrospectively analyzed from January 2017 to November 2022 in The Fourth Affiliated Hospital of Hebei Medical University (Ligang et al., A machine learning model based on CT and clinical features to distinguish pulmonary nodular mucinous adenocarcinoma from tuberculoma, 2023). A total of 1037 radiomic features were extracted from contrast-enhanced computed tomography (CT). The patients were randomly divided into a training group and a test group at a ratio of 7:3. The least absolute shrinkage and selection operator (LASSO) algorithm was used for radiomic feature selection. Three radiomics prediction models were applied: logistic regression (LR), support vector machine (SVM) and random forest (RF). The best performing model was adopted, and the radiomics score (Radscore) was then computed. The clinical model was developed using logistic regression. Finally, a combined model was established based on clinical factors and radiomics features. We externally validated the three models in a group of 68 patients (46 and 22 patients with PNMA and PTB, respectively) from Xing Tai People’s Hospital (30 and 14 patients with PNMA and PTB, respectively) and The First Hospital of Xing Tai (16 and 8 patients with PNMA and PTB, respectively). The area under the receiver operating characteristic (ROC) curve (AUC) value and decision curve analysis were used to evaluate the predictive value of the developed models.
### Results
The combined model established by the logistic regression method had the best performance. The ROC-AUC (also a decision curve analysis) of the combined model was 0.940, 0.990 and 0.960 in the training group, test group and external validation group, respectively, and the combined model showed good predictive performance for the differentiation of PNMA from PTB. The Brier scores of the combined model were 0.132 and 0.068 in the training group and test group, respectively.
### Conclusion
The combined model incorporating radiomics features and clinical parameters may have potential value for the preoperative differentiation of PNMA from PTB.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12885-023-10734-4.
## Introduction
Lung cancer is the leading cause of cancer-related deaths worldwide and one of the most common malignancies in China [1, 2]. In primary lung cancer, non-small cell lung cancer (NSCLC) accounts for almost $85\%$ of the cases, of which adenocarcinoma is the most frequent pathological subtype [3, 4]. According to the International Association for the Study of Lung Cancer (IASLC)/American Thoracic Society (ATS)/European Respiratory Society (ERS) classification system for lung adenocarcinoma published in 2011, as well as the World Health Organization (WHO) published in 2015, primary pulmonary invasive mucinous adenocarcinoma (IMA) was classified as a variant subtype of lung adenocarcinoma, accounting for 2–$5\%$ of the cases of adenocarcinoma [5, 6].
Previous studies have explored the imaging findings of IMA [7, 8]. On the basis of CT findings, IMA is mainly divided into two types: pulmonary nodule-type IMA (PNMA) and pneumonia-type IMA [9, 10]. PNMA nodules, especially nodules with ill-defined margins, internal cavities or vacuoles, and no-mild enhancement on dynamic CT, were similar to benign solid nodules. Pulmonary tuberculoma (PTB) is the most common type of benign solid nodule [11]. The CT characteristics of PTB can also include spiculated signs and pleural indentation, which can make it difficult to distinguish PTB from PNMA. There may be overlap in the morphology and CT features between PNMA and PTB. In fact, many cases of PNMA are misdiagnosed as PTB in clinical practice, which leads to missing the best treatment time, leading to a shortened survival time [12, 13]. Distinguishing PNMA and PTB before treatment in a timely manner is important.
Computed tomography (CT) is widely used for tumor detection, staging and therapeutic response monitoring in clinical practice [14]. There have been few studies concerning the differential diagnosis of PNMA and PTB [12]. The results revealed that satellite lesions were more often observed in the PTB group, and the mean CT attenuation of PTB shown on the plain scan was significantly higher than that of PNMA (35.15 ± 16.00 vs. 24.00 ± 12.67 HU; $P \leq 0.01$). However, the enhanced value of PTB on venous scans was significantly lower than that of PNMA (13.44 ± 13.40 vs. 22.52 ± 14.00 HU; $$P \leq 0.02$$). However, the noninvasive imaging diagnostic information provided by preoperative CT images is still limited and cannot accurately differentiate PNMA from PTB. Radiomics can describe the characteristics of the lesion by high-throughput extraction of a large number of medical image features and is an emerging technology that could enhance clinical decision-making [14, 15]. Radiomics models have been indicated to be favorable for diagnosing lung nodules in clinical applications, such as distinguishing benign from malignant nodules, for the preoperative prediction of nodule type, for prognostic analysis, and for predicting the outcome of a surgery, gene expression pattern and microenvironment of tumor [16–18]. Imaging examination is one of the routine procedures of daily clinical diagnosis, so radiomics research is readily feasible. Based on these former explorations, to better satisfy the needs of precise evaluation of pathology, radiomics features were explored and used to develop a model to differentiate PNMA from PTB in our present study.
## Materials and methods
This study was performed following the Helsinki Declaration and approved by the Ethics Committee of our hospital (Ethics Committee of Hebei Medical University Fourth Affiliated Hospital, reference number: 2022KS017, data 2022.6.27). Ethical approval was obtained from our hospital. And it was approved by the Ethics Committee of our hospital (Ethics Committee of Hebei Medical University Fourth Affiliated Hospital, reference number: 2022KS017, data 2022.6.27) that waivers of consent were granted to the study subjects. As a retrospective study, the exemption from obtaining consent does not affect the rights and interests of the research participant [19]. All methods were performed in accordance with the relevant guidelines and regulations.
## Patient selection
We retrospectively analyzed patients with lung mucinous adenocarcinoma and pulmonary tuberculoma diagnosed from January 2017 to November 2022 in The Fourth Affiliated Hospital of Hebei Medical University. The inclusion criteria of patients with PNMA and PTB were as follows: [1] surgical pathology-confirmed invasive mucinous adenocarcinoma (mucinous adenocarcinoma component > $90\%$) and pulmonary tuberculoma; [2] the maximum diameter of the nodule was less than or equal to 3.0 cm; [3] solitary and solid nodules without calcification, which may contain cavities or vacuoles and do not exhibit a ground glass density; and [4] complete clinical and pathological data, including analyzable plain and enhanced thin-slice CT image data (1.25 mm/slice) and available CT images within 2 weeks before the pathological diagnosis. The exclusion criteria of patients were as follows: [1] multiple pulmonary nodules; [2] antitumor therapy prior to CT examination and pathological diagnosis; [3] other types of cancer or incomplete clinical and imaging data; and [4] lymph node metastases and/or distant metastases [19]. In this retrospective study, a total of 177 patients were enrolled, and their ages ranged from 20 to 81 years. The patients were randomly divided into a training group and a test group at a ratio of 7:3, including 123 patients in the training group (86 patients with PNMA, 37 patients with PTB) and 54 patients in the test group (38 patients with PNMA and 16 patients with PTB). Using retrospective data from the training group, a nomogram was developed, which was internally validated using data from the test group. Additionally, we collected a dataset ($$n = 68$$) from January 2017 to November 2022 in Xing Tai People’s Hospital (30 and 14 patients with PNMA and PTB, respectively) and The First Hospital of Xing Tai (16 and 8 patients with PNMA and PTB, respectively) to validate the nomogram externally. The inclusion and exclusion criteria were the same as for the development cohort. Our hospital's ethical review board approved this retrospective analysis, and the requirement for informed consent was waived.
## CT image acquisition
For all patients, contrast-enhanced chest CT scans were conducted with a 256-multidetector CT scanner (Discovery CT 750 HD Revolution, GE Medical Systems, Milwaukee, Wisconsin, USA). Before scanning, the patients were trained to breathe and hold their breath at the end of inspiration to obtain the scans. The patients were placed in the supine position with both arms raised to reduce scanning artifacts. The locational marker was the sternoclavicular joint, and the range included the thoracic inlet to the lung bases. The scanning parameters were as follows: tube voltage 120 kV, tube current 200 mAs, reconstruction layer thickness 1.25 mm, matrix 512 × 512, and pitch 1.2. The reconstruction algorithm adopts the lung algorithm. After the plain scan was completed, 70–90 ml of nonionic contrast agent iohexol or ioversol (300 mg·I/ml) was injected via a bolus through the cubital vein with a high-pressure syringe at a flow rate of 3 ml/s. Arterial phase and venous phase dual-phase enhanced scans were performed 30 s and 90 s after the injection of the contrast agent, respectively. The other parameters were the same as those for the plain scans. After scanning, the raw data were uploaded to a postprocessing workstation for multiplanar reconstruction (MPR) [19].
Image feature analysis was performed by two board-certified thoracic group radiologists (with 6 and 12 years of experience in chest CT imaging, respectively) who were blinded to the clinical and histological findings. The mediastinal window (window width 400 HU; window level 40 HU) and lung window (window width 1200 HU; window level − 600 HU) were set. The CT image features recorded in the image analysis were as follows: [1] primary tumor location (left and right lungs, upper, middle and lower lobes); [2] tumor size (maximum diameter), mean CT value (plain scan, venous phase), ΔCTV (the difference in the mean CT value of the venous phase and the mean CT value of the plain scan), and edge (lobular, blur); [3] internal features of the tumor: the presence or absence of cavities or vacuoles; and [4] external features of the tumor: the presence or absence of satellite lesions around the nodules. The mean CT values of the nodules on the plain scan and the venous scan were measured. The cavity and vacuole were defined as a gaseous density with a maximum diameter greater than 5 mm and less than 5 mm, respectively. The satellite lesions were defined as ≥ 1 miliary nodule surrounding each nodule (within 3.0 cm), apart from one miliary nodule distal to each nodule (possible obstructive inflammation). All CT image features were independently recorded by two radiologists, and any discrepancies in assessments were consistently resolved [19].
## Segmentation, feature extraction, and selection
The CT images were imported into the open-source software 3D-Slicer (version 5.0.2, http://www.slicer.org) and were read under the lung window (width 1500/-600 HU) and mediastinal window (width $\frac{400}{40}$ HU) settings. The primary lesions of the patients with PNMA or PTB were selected for tumor segmentation after image acquisition. A radiologist without knowledge of the clinical data manually delineated regions of interest (ROIs) layer by layer. The tumor ROI encompassed the entire lesion as much as possible, including cavities or vacuoles within the nodules and excluding bronchi, blood vessels, and normal lung tissue. Tumor segmentation was performed on 40 patients who were randomly selected from the entire cohort for independent segmentation to assess the intraclass agreement one month later. Another radiologist repeated the independent segmentation of the selected 40 patients and evaluated the interclass agreement. Intra- and interclass correlation coefficients (ICCs) were used to assess the intraobserver and interobserver reproducibility of feature extraction.
Pyradiomics in 3D-Slicer software was used to extract the radiomics features. The detailed extraction customization was as follows: 1) Feature classes: All features; 2) Resampling and filtering: Resampled voxel size [3,3,3], LoG kernel sizes [4,5], and Wavelet-based features (√). A total of 1037 features were extracted, including 17 histogram classes, 14 form factor classes, 24 Gy level cooccurrence matrix (GLCM) classes, 16 Gy level run length matrix (GLRLM) classes, 16 Gy level size zone matrix (GLSZM) classes, 5 neighboring gray tone difference matrix (NGTDM) classes, and 14 Gy level dependence matrix (GLDM) classes.
To reduce the dimensionality of the radiomic features to the number of events, we performed three sequential steps for radiomic feature selection. First, we evaluated the interobserver agreement of radiomic features and selected features showing ICC > 0.75. For the next step, we chose radiomic features that showed statistical significance between the PNMA and PTB groups. Finally, the least absolute shrinkage and selection operator (LASSO) logistic regression model was used to choose the most useful predictive features of radiomics for the differentiation of PNMA from PTB in the training group: fivefold cross validation was performed 100 times to avoid overfitting.
## Model development
Three radiomics prediction models, logistic regression (LR), support vector machine (SVM) and random forest (RF), were applied. The best performing model was adopted, and the radiomics score (Radscore) was then computed.
At the same time, we constructed a model based on clinical and CT features for the multivariate logistic regression analysis. The clinical features included sex, age, smoking status, and diabetes history. The CT features are illustrated above.
Finally, three models, the clinical model, radiomics model and the combined model based on the clinical factors and radiomics features, were compared statistically to identify the model with the highest predictability.
## Statistical analysis
All statistical analyses were performed using R version 3.6.3 and Python version 3.7. The patients were randomly divided into a training group and a test group at a ratio of 7:3. All radiomic features were applied with Z score normalization. Baseline data were analyzed by univariate analysis using Python statsmodels 0.11.1. The chi-square test was used for categorical variables, and the t test or Mann‒Whitney U test was used for continuous variables. The factors with significant differences ($P \leq 0.05$) were included in the multivariate logistic regression analysis. The clinical and CT features with significant differences ($P \leq 0.05$) in the multivariate analysis results were selected to construct the clinical prediction model [19].
The main evaluation indicators were the area under the ROC curve, accuracy, sensitivity, specificity, positive predictive value and negative predictive value. Decision curve analysis (DCA) was used to calculate the clinical impact of the three models [19].
## Patient characteristics
In all enrolled patients, there was no significant difference in age between the two groups, and the median ages were 59 and 62 years in the patients with PTB and PNMA, respectively. The proportion of female patients was higher in the patients with PNMA than in the patients with PTB ($50.00\%$ and $24.42\%$, $$P \leq 0.004$$). Smoking history was more prevalent in patients with PTB than in patients with PNMA ($62.26\%$ and $34.68\%$, $P \leq 0.001$). There was a higher proportion of patients with a history of diabetes in the PTB group than in the PNMA group. In the PNMA group and the PTB group, the proportion of lesions located in the lower lobe was $66.94\%$ and $22.64\%$, respectively. Between the PNMA and PTB groups, the proportion of satellite lesions on CT lung window images was statistically significant ($50.94\%$ and $4.03\%$, respectively $P \leq 0.001$). The proportion of cavities or vacuoles was higher in the patients with PNMA than in the patients with PTB ($52.42\%$ and $24.53\%$, respectively). $P \leq 0.001$). The mean CT value on the plain scan of PNMA was significantly lower than that of PTB (17.00 HU vs. 31.00 HU; $P \leq 0.001$). The ΔCTV of PNMA was significantly higher than that of PTB (25.49 HU vs. 4.00 HU; $P \leq 0.001$). The best cost values of plain CT value and ΔCTV were 29 HU and 12 HU, respectively. The characteristics of the patients in the training and testing cohorts are shown in detail in Table 1 [1].Table 1Clinical characteristics of the patientsCharactersTraining cohortTesting cohortPTBPNMApPTBPNMApGender Female25500.07814120.006Male939523Smoking No1058 < 0.00110230.346Yes2431912Diabetes No25790.03711300.022Yes91085Edge clear NO1130.068350.882Yes33761630Lobul No3340.0011130.011Yes31551822Spicul No28630.19112240.687Yes626711Satellite No1685 < 0.0011034 < 0.001Yes18491Cavity No24430.02716160.006Yes1046319Lower lobe No2631 < 0.0011510 < 0.001Yes858425△CTA, median[IQR]2.000[0.602,11.000]18.000[8.000,40.000] < 0.0011.000[1.000,3.000]25.000[12.000,46.000] < 0.001△CTV, median[IQR]5.000[2.000,15.000]23.000[14.000,38.489] < 0.0012.000[1.000,6.000]37.795[16.000,60.712] < 0.001Plain, median[IQR]30.950[21.000,41.000]12.000[-46.288,27.000] < 0.00132.000[23.000,60.000]21.000[-40.000,26.436] < 0.001Diameter, median[IQR]2.000[1.600,2.400]1.700[1.200,2.300]0.1311.700[1.300,2.000]1.400[1.100,2.000]0.224Age, median[IQR]56.000[51.000,64.000]62.000[56.000,67.000]0.03657.474 ± 12.99159.657 ± 10.3700.511
## Feature selection and clinical model construction
The multivariate analysis showed that there were significant differences between PNMA and PTB in smoking, diabetes history, lesion located in the lower lobe, satellite lesions, cavity or vacuole, plain CT value and ΔCTV. The clinical prediction model was established by logistic regression based on the seven features. The AUC values of the clinical models in differentiating PNMA from tuberculoma were 0.918 and 0.888 in the training group and test group, respectively. See Table 2 for details. Table 2Multivariate analysis to identify significant factors for PTB and PNMAPredictorpOdds RatioLowerUpper△CTV0.0061.0261.0081.047Plain0.0040.9780.9610.991Diabetes0.1320.4150.1291.301Smoking0.1270.4640.171.245Satellite0.00.0760.0170.272Cavity0.0114.3991.49315.097Lower lobe0.0143.4651.2999.625
## Radiomics feature selection and model construction
To eliminate redundant features, 688 features that showed no significant difference between PNMA and PTB and 2 highly correlated features with ICC values less than 0.75 were excluded. After screening out the redundant features by LASSO and correlation analysis, the five most robust radiomics features (including flatness, cluster shade, minimum, median and skewness) were retained. The relationships of the five radiomics features were shown to be significantly low (Fig. 1).Fig. 1Radiomics feature selection The machine learning models logistic regression (LR), support vector machine (SVM) and random forest (RF) were developed based on the five radiomics features. The model established by the LR method had the best performance, and the AUC values in the training group and test group were 0.840 and 0.960, respectively (Table 3 and Fig. 4). The Radscore for each patient was then calculated by selected features weighted by their respective coefficients in the LR model, which can be expressed as follows: Radscore = 1.284 + 2.169*Flatness + 0.911*Median-0.613*Cluster Shade-0.482*Mean-0.941* Minimum. The Radscore for each patient in the training group and test group is shown in Fig. 2.Table 3Diagnostic performance of the prediction modelsModelTraining cohortTesting cohortExternal validationAUCSENSPEAUCSENSPEAUCSENSPERADS0.8400.8090.7650.9600.9430.8420.8400.7730.810Clinical0.9200.8200.9120.9500.9140.9470.8900.7730.857Comb0.9400.8990.9120.9900.9431.0000.9600.8640.952Fig. 2Bar charts of the Radscore for each patient in the training cohort (A) and testing cohort (B)
## Combined model construction and validation of performance
Logistic regression was performed to establish a combined model based on seven clinical features and five radiomics features. The results revealed that the combined model and radiomics model were superior to the clinical model and radiomics model, with ROC-AUCs of the combined model, radiomics model and clinical model of 0.940 and 0.990, 0.840 and 0.960 and 0.920 and 0.950 in the training and test groups, respectively (Table 3 and Fig. 3), respectively. The combined model had very high sensitivity and specificity, 0.899 and 0.912, 0.943 and 1.00 in the training and test groups, respectively (Table 3). The decision curve analyses revealed that when the probability of the threshold was over $0\%$, the net benefits of the combined model for preoperative differentiation of PNMA from PTB were higher than those of the clinical model and radiomics model in both the training group and test group (Fig. 4). The Brier scores of the combined model were 0.132 and 0.068 in the training group and test group, respectively, and the calibration plots are shown in Fig. 5.Fig. 3Comparison of receiver operating characteristic (ROC) curves among the clinical model, radiomics model and combined model in the training cohorts (A) and testing cohorts (B)Fig. 4Decision curve analyses for the radiomics-clinical model compared with the radiomics model and clinical model in the training cohort (A) and the testing cohort (B)Fig. 5Calibration plot of the combined model in the training group
## External validation of the three models
The external validation group included 68 patients in the final analysis. Table 1 shows the baseline characteristics of the external group. The AUC values of the clinical model, radiomics model and combined model in differentiating PNMA from PTB were 0.890, 0.840 and 0.960 in the external validation group, respectively (Table 3 and Supplementary Fig. 1).
## Discussion
According to this retrospective study, there were some significant differences between PNMA and PTB patients with solitary pulmonary solid nodules in terms of qualitative and quantitative clinical characteristics. We selected seven clinical variables to build our clinical prediction model to differentiate PNMA from PTB, including smoking history, diabetes history, lesion located in the lower lobe, satellite lesions, cavity or vacuole, plain CT value and ΔCTV [19]. A total of five radiomic features were selected to build the radiomic prediction model, including flatness, cluster shade, minimum, median and skewness. The clinical-radiomics combined model, consisting of the above seven clinical features and five radiomics parameters, demonstrated good predictive ability in both the training and validation sets. Moreover, there were statistically significant differences among the clinical model, radiomics model, and combined clinical-radiomics model. The clinical-radiomics combined model was better than both single models. An external validation group of 68 patients (46 and 22 patients with PNMA and PTB, respectively) was successfully used to validate the clinical-radiomics combined model.
In this study, smoking was an independent factor to distinguish PNMA and PTB. In a previous study, cigarette smoking increases not only the risk of progression to active pulmonary tuberculosis disease but also the risk of acquiring new tuberculosis disease infection [20]. According to another study, there was also a significant linear dose–response relationship between QFT positivity and smoking duration (by years) among ever smokers ($p \leq 0.001$) [21]. This can be explained by the result that lysosomal storage also hinders alveolar macrophage migration to mycobacteria in the lung, suggesting a mechanism for the observed susceptibility of smokers to new tuberculosis disease infection. According to a study specifically relating smoking history to the mucinous phenotype, mucinous adenocarcinoma had little relationship to smoking history [22]. Consequently, we found that smoking history was more prevalent in PTB than in PNMA. Previous studies have shown that diabetes patients' immune systems are impaired, making them more susceptible to tuberculosis [23, 24]. Diabetes mellitus did not significantly affect lung cancer risk in patients with diabetes mellitus (RR: 1.10; $95\%$ CI: 0.99–1.23; $$P \leq 0.087$$) [25]. The present study revealed a similar result: more than half of the PTB patients had diabetes complications, whereas only a few patients with PNMA were complicated with diabetes.
In histology, PNMA is characterized by mucin-rich tumor cells, fibrosis, with central fibrosis and alveolar spaces filled with mucin [26, 27]. PTB is composed of fibrous tissue containing caseous necrotic tissue. It is very easy to misdiagnose one as the other since both are low-density on plain scans and their morphological characteristics are very similar. Based on our multivariate logistic regression analysis, we found that PNMA had a higher tendency to occur in the lower lobe than PTB, whereas patients with PTB showed obvious upper lobe distribution preponderance ($73.58\%$). Our findings were consistent with the reports by Xu [28] and Zhang et al. [ 29]. A possible explanation for the difference is that the tumor cells of PNMA originate from columnar epithelial cells or goblet cells. The cancer cells are relatively well-differentiated, and they can produce more mucus in response to gravity. Therefore, PNMA was much more frequently found in the lower lobe. PTB usually occurs in the apical or posterior segment of the upper lobes and lingular segment on both lower lobes. Tuberculosis bacilli are more likely to colonize and cause disease in cultures with poor blood circulation (the number of macrophages is small) and ventilation (bacteria easily survive). The “satellite lesions” usually refer to small discrete shadows in the immediate vicinity of the main lesion. As a characteristic manifestation of tuberculosis, it has become widely accepted. The pathological basis may be the spreading focus and fibroproliferative focus around the tuberculosis lesion. 26 of the 53 PTB patients had satellite lesions, with a ratio of 1:2. In contrast, 5 of the 124 PNMA had satellite lesions, with a ratio of 1:25. Compared to PNMA, PTB has a high prevalence of satellite lesions, as reported previously [29, 30]. Cavities or vacuoles were found in more than half of the patients with PNMA, but in only a quarter of the patients with PTB. Cavities appear in PNMA due to incomplete obstruction of the bronchioles by the build-up of mucus, resulting in alveolar hyperventilation. On the other hand, vacuoles may be caused by internal necrosis of the tumor, and the necrosis is eliminated through the bronchus. Therefore, cavities or vacuoles were much more common in PNMA.
CT scanning, especially CT dynamic contrast-enhanced scanning, is a valuable tool in the diagnosis of lung cancer and tuberculosis. The CT features of lung adenocarcinoma and pulmonary tuberculoma have received much attention in the literature, but PNMA and PTB have received fewer reports [12, 29, 30]. As a result of this study, we determined that the mean CT value of PTB on plain scan was 31.00 HU and that the CT value of PNMA was 17.00 HU, whereas the ΔCTV was 25.49 HU in the PNMA group and 4 HU in the PTB group. The CT value of the PTB group on plain scan was significantly greater than that of the PNMA group, whereas the ΔCTV in the PTB group was lower than that in the PNMA group, which is inconsistent with a former report [28]. As PNMA and PTB have different pathologic bases, this may explain the difference. The PNMA contains mucin-rich tumor cells, fibrous tissues, and alveolar spaces filled with mucin proteins, resulting in a low CT value. PTB is composed of fibrous tissue that contains caseous necrotic tissue with low density, but calcification can easily occur. There are some calcifications that are fine sand and scattered, which results in a higher CT value. PNMA exhibited a complex pattern of enhancement based on the amount of solid component, fibrous tissue, and mucin present. Additionally, papillary or alveolar components within PNMA increased the difference between CT values. The center of tuberculoma is necrotic tissue with no blood supply, the periphery is a capsule, and the inner layer is granulation tissue with blood supply. As a result, enhancement is either nonenhanced, annular, or in other forms, depending on the degree of necrosis in the caseous region and the presence of granulation tissue.
To explore a much more effective method to differentiate PNMA from PTB, we extracted five independent radiomic features associated with PNMA and PTB, including flatness, cluster shade, minimum, median and skewness. These parameters belong to the Form Factor Parameters, Texture Parameters, and Histogram Parameters. Flatness is a form factor parameter that is independent of the gray-level intensity distribution in the ROI. Cluster Shade, as one of the Texture Parameters, is the task of grouping a group of objects so that objects in the same group (cluster shade) are more similar to each other (in a sense) than objects in other groups (cluster shade). The larger the Cluster Shade value, the more asymmetrical it is. The minimum belongs to the histogram parameter, which represents the minimum pixel value of an image (of the lesion). The median is also a histogram parameter that represents the median pixel value of an image (of the lesion). Another histogram parameter is skewness, which reflects the degree of asymmetry in the histogram distribution, and if the predictive value has been effective, the absolute values of the skewness would have been higher. All features above were obtained via the conversion of images to higher-dimensional data. They allowed high-throughput mining of quantitative imaging features from general medical images, followed by an automated analysis to assist clinical decision-making. Previous studies revealed that radiomics features from CT could preoperatively differentiate lung adenocarcinoma from lung tuberculoma in patients with pulmonary solitary solid nodules and could also distinguish adenocarcinomas from granulomas. To our knowledge, this is the first study to differentiate PNMA from PTB based on radiomic features. In this study, the model based on these radiomic features also illustrated an effective role in preoperatively differentiating PNMA from PTB. According to our ROC analyses, the AUC values for the radiomics model were 0.840 and 0.960 in the training and test groups, respectively. The clinical-radiomics combined model (ROC-AUC: 0.940–0.990) was significantly better than the clinical model and radiomics model. Furthermore, the decision curve analysis also demonstrated that the combined model performed significantly better than the clinical model and radiomic model in predicting outcomes. Decision curve analysis offers important information beyond the standard performance metrics of discrimination and calibration and could be used to evaluate the clinical impact, indicating that there was a higher chance of success. In this study, not only an internal validation was carried out, but an external validation was also carried out on this basis. At the same time, the ROC curve analysis showed that the AUC of the clinical-radiomics combined model was greater than 0.9000 in both the internal validation and external validation. This suggested that the clinical-radiomics combined model had a good discrimination degree and a strong ability to distinguish PNMA from PTB.
Our study had several limitations that must be considered. First, it was a retrospective study with a relatively small sample size. Second, due to the study's inclusion of only patients who had pathologic results after surgery, selection bias cannot be ignored. Additionally, only PNMA and PTB were observed without calcification, so the results should be interpreted cautiously. Third, our study only evaluated the relationship between PNMA and PTB, and other types of lung nodules need to be further explored, such as lung squamous cell carcinoma and other benign granulomatous lesions. To guide clinical practice, the model will be validated in a multicenter, prospective, large-scale study in the future and further optimized.
In conclusion, in our present study, we established a model to differentiate PNMA from PTB by using preoperative clinicopathological features, radiomic features, and clinical-radiomic features for the first time. The clinical-radiomic model established that we established also showed good predictive value and had potential value in clinical practice.
## Supplementary Information
Additional file 1. Additional file 2.
## References
1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. **GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries**. *CA Cancer J Clin* (2018.0) **68** 394-424. DOI: 10.3322/caac.21492
2. 2.Siegel RL, Miller KD, Jemal A: Cancer statistics, 2019. CA: a cancer journal for clinicians 2019, 69(1):7–34.
3. Dong RF, Zhu ML, Liu MM, Xu YT, Yuan LL, Bian J, Xia YZ, Kong LY. **EGFR mutation mediates resistance to EGFR tyrosine kinase inhibitors in NSCLC: From molecular mechanisms to clinical research**. *Pharmacol Res* (2021.0) **167** 105583. DOI: 10.1016/j.phrs.2021.105583
4. Memmott RM, Wolfe AR, Carbone DP, Williams TM. **Predictors of Response, Progression-Free Survival, and Overall Survival in Patients With Lung Cancer Treated With Immune Checkpoint Inhibitors**. *J Thorac Oncol* (2021.0) **16** 1086-1098. DOI: 10.1016/j.jtho.2021.03.017
5. Travis WD, Brambilla E, Nicholson AG, Yatabe Y, Austin JHM, Beasley MB, Chirieac LR, Dacic S, Duhig E, Flieder DB. **The 2015 World Health Organization Classification of Lung Tumors: Impact of Genetic, Clinical and Radiologic Advances Since the 2004 Classification**. *J Thorac Oncol* (2015.0) **10** 1243-1260. DOI: 10.1097/JTO.0000000000000630
6. Travis WD, Brambilla E, Noguchi M, Nicholson AG, Geisinger KR, Yatabe Y, Beer DG, Powell CA, Riely GJ, Van Schil PE. **International association for the study of lung cancer/american thoracic society/european respiratory society international multidisciplinary classification of lung adenocarcinoma**. *J Thorac Oncol* (2011.0) **6** 244-285. DOI: 10.1097/JTO.0b013e318206a221
7. Kim DH, Bae SY, Na KJ, Park S, Park IK, Kang CH, Kim YT. **Radiological and clinical features of screening-detected pulmonary invasive mucinous adenocarcinoma**. *Interact Cardiovasc Thorac Surg* (2022.0) **34** 229-235. DOI: 10.1093/icvts/ivab257
8. Miyata N, Endo M, Nakajima T, Kojima H, Maniwa T, Takahashi S, Isaka M, Kameya T, Ohde Y. **High-resolution computed tomography findings of early mucinous adenocarcinomas and their pathologic characteristics in 22 surgically resected cases**. *Eur J Radiol* (2015.0) **84** 993-997. DOI: 10.1016/j.ejrad.2015.01.014
9. Nie K, Nie W, Zhang YX, Yu H. **Comparing clinicopathological features and prognosis of primary pulmonary invasive mucinous adenocarcinoma based on computed tomography findings**. *Cancer Imaging* (2019.0) **19** 47. DOI: 10.1186/s40644-019-0236-2
10. Watanabe H, Saito H, Yokose T, Sakuma Y, Murakami S, Kondo T, Oshita F, Ito H, Nakayama H, Yamada K. **Relation between thin-section computed tomography and clinical findings of mucinous adenocarcinoma**. *Ann Thorac Surg* (2015.0) **99** 975-981. DOI: 10.1016/j.athoracsur.2014.10.065
11. Niyonkuru A, Chen X, Bakari KH, Wimalarathne DN, Bouhari A, Arnous MMR, Lan X. **Evaluation of the diagnostic efficacy of (18) F-Fluorine-2-Deoxy-D-Glucose PET/CT for lung cancer and pulmonary tuberculosis in a Tuberculosis-endemic Country**. *Cancer Med* (2020.0) **9** 931-942. DOI: 10.1002/cam4.2770
12. Yin YH, Qi YG, Wang B. **Differential diagnosis of pulmonary nodular mucinous adenocarcinoma and tuberculoma with dynamic CT: a retrospective study**. *J Thorac Dis* (2022.0) **14** 1225-1231. DOI: 10.21037/jtd-22-372
13. Dhandapani S, Dheenadayalu NA, Shanmugam S, Amitkumar K, Reddy HK. **Pulmonary tuberculosis masking lung cancer - A case report**. *Respir Med Case Rep* (2015.0) **16** 35-37. PMID: 26744649
14. Wen Q, Yang Z, Zhu J, Qiu Q, Dai H, Feng A, Xing L. **Pretreatment CT-Based Radiomics Signature as a Potential Imaging Biomarker for Predicting the Expression of PD-L1 and CD8+TILs in ESCC**. *Onco Targets Ther* (2020.0) **13** 12003-12013. DOI: 10.2147/OTT.S261068
15. Lee G, Lee HY, Park H, Schiebler ML, van Beek EJR, Ohno Y, Seo JB, Leung A. **Radiomics and its emerging role in lung cancer research, imaging biomarkers and clinical management: State of the art**. *Eur J Radiol* (2017.0) **86** 297-307. DOI: 10.1016/j.ejrad.2016.09.005
16. She Y, Zhang L, Zhu H, Dai C, Xie D, Xie H, Zhang W, Zhao L, Zou L, Fei K. **The predictive value of CT-based radiomics in differentiating indolent from invasive lung adenocarcinoma in patients with pulmonary nodules**. *Eur Radiol* (2018.0) **28** 5121-5128. DOI: 10.1007/s00330-018-5509-9
17. Wu G, Woodruff HC, Sanduleanu S, Refaee T, Jochems A, Leijenaar R, Gietema H, Shen J, Wang R, Xiong J. **Preoperative CT-based radiomics combined with intraoperative frozen section is predictive of invasive adenocarcinoma in pulmonary nodules: a multicenter study**. *Eur Radiol* (2020.0) **30** 2680-2691. DOI: 10.1007/s00330-019-06597-8
18. Zhao W, Xu Y, Yang Z, Sun Y, Li C, Jin L, Gao P, He W, Wang P, Shi H. **Development and validation of a radiomics nomogram for identifying invasiveness of pulmonary adenocarcinomas appearing as subcentimeter ground-glass opacity nodules**. *Eur J Radiol* (2019.0) **112** 161-168. DOI: 10.1016/j.ejrad.2019.01.021
19. 19.Ligang Hao JZ, Yonghui Di, MingWei Qi, Qian Xu, Ning Zhang, Feng Hui, Gaofeng Shi: A machine learning model based on CT and clinical features to distinguish pulmonary nodular mucinous adenocarcinoma from tuberculoma. 2023.
20. Berg RD, Levitte S, O'Sullivan MP, O'Leary SM, Cambier CJ, Cameron J, Takaki KK, Moens CB, Tobin DM, Keane J. **Lysosomal Disorders Drive Susceptibility to Tuberculosis by Compromising Macrophage Migration**. *Cell* (2016.0) **165** 139-152. DOI: 10.1016/j.cell.2016.02.034
21. Zhang H, Xin H, Li X, Li H, Li M, Lu W, Bai L, Wang X, Liu J, Jin Q. **A dose-response relationship of smoking with tuberculosis infection: A cross-sectional study among 21008 rural residents in China**. *PLoS ONE* (2017.0) **12** e0175183. DOI: 10.1371/journal.pone.0175183
22. Krasinskas AM, Chiosea SI, Pal T, Dacic S. **KRAS mutational analysis and immunohistochemical studies can help distinguish pancreatic metastases from primary lung adenocarcinomas**. *Mod Pathol* (2014.0) **27** 262-270. DOI: 10.1038/modpathol.2013.146
23. Al-Rifai RH, Pearson F, Critchley JA, Abu-Raddad LJ. **Association between diabetes mellitus and active tuberculosis: A systematic review and meta-analysis**. *PLoS ONE* (2017.0) **12** e0187967. DOI: 10.1371/journal.pone.0187967
24. McWilliams A, Tammemagi MC, Mayo JR, Roberts H, Liu G, Soghrati K, Yasufuku K, Martel S, Laberge F, Gingras M. **Probability of cancer in pulmonary nodules detected on first screening CT**. *N Engl J Med* (2013.0) **369** 910-919. DOI: 10.1056/NEJMoa1214726
25. Yi ZH, Luther Y, Xiong GH, Ni YL, Yun F, Chen J, Yang Z, Zhang Q, Kuang YM, Zhu YC. **Association between diabetes mellitus and lung cancer: Meta-analysis**. *Eur J Clin Invest* (2020.0) **50** e13332. DOI: 10.1111/eci.13332
26. Sawada E, Nambu A, Motosugi U, Araki T, Kato S, Sato Y, Saito A, Ishihara H, Okuwaki H, Ozawa K. **Localized mucinous bronchioloalveolar carcinoma of the lung: thin-section computed tomography and fluorodeoxyglucose positron emission tomography findings**. *Jpn J Radiol* (2010.0) **28** 251-258. DOI: 10.1007/s11604-009-0414-4
27. Yap CS, Schiepers C, Fishbein MC, Phelps ME, Czernin J. **FDG-PET imaging in lung cancer: how sensitive is it for bronchioloalveolar carcinoma?**. *Eur J Nucl Med Mol Imaging* (2002.0) **29** 1166-1173. DOI: 10.1007/s00259-002-0853-y
28. Xu X, Shen W, Wang D, Li N, Huang Z, Sheng J, Rucker AJ, Mao W, Xu H, Cheng G. **Clinical features and prognosis of resectable pulmonary primary invasive mucinous adenocarcinoma**. *Transl Lung Cancer Res* (2022.0) **11** 420-431. DOI: 10.21037/tlcr-22-190
29. 29.Zhang J, Han T, Ren J, Jin C, Zhang M, Guo Y: Discriminating Small-Sized (2 cm or Less), Noncalcified, Solitary Pulmonary Tuberculoma and Solid Lung Adenocarcinoma in Tuberculosis-Endemic Areas. Diagnostics (Basel, Switzerland) 2021, 11(6).
30. Zhuo Y, Zhan Y, Zhang Z, Shan F, Shen J, Wang D, Yu M. **Clinical and CT Radiomics Nomogram for Preoperative Differentiation of Pulmonary Adenocarcinoma From Tuberculoma in Solitary Solid Nodule**. *Front Oncol* (2021.0) **11** 701598. DOI: 10.3389/fonc.2021.701598
|
---
title: 'Characterization of free fatty acid receptor family in rainbow trout (Oncorhynchus
mykiss): towards a better understanding of their involvement in fatty acid signalisation'
authors:
- Jérôme Roy
- Elodie Baranek
- Lucie Marandel
journal: BMC Genomics
year: 2023
pmcid: PMC10029227
doi: 10.1186/s12864-023-09181-z
license: CC BY 4.0
---
# Characterization of free fatty acid receptor family in rainbow trout (Oncorhynchus mykiss): towards a better understanding of their involvement in fatty acid signalisation
## Abstract
Since 20 years of research, free fatty acids receptors (FFARs) have received considerable attention in mammals. To date, four FFARs (FFAR1, FFAR2, FFAR3 and FFAR4) are especially studied owing to their physiological importance in various biological processes. This ubiquitist group of G protein-coupled receptors (GPCRs) are majors reports in the key physiological functions such as the regulation of energy balance, metabolism or fatty acid sensing. However, up till date, even some studies were interested in their potential involvement in fatty acid metabolism, no genome investigation of these FFARs have been carried out in teleost fish. Through genome mining and phylogenetic analysis, we identified and characterised 7 coding sequences for ffar2 in rainbow trout whereas no ffar3 nor ffar4 gene have been found. This larger repertoire of ffar2 genes in rainbow trout results from successive additional whole-genome duplications which occurred in early teleosts and salmonids, respectively. A syntenic analysis was used to assign a new nomenclature to the salmonid ffar2 and showed a clear conservation of genomic organisation, further supporting the identity of these genes as ffar2. RT-qPCR was then used to examine, firstly during ontogenesis and secondly on feeding response the expression pattern of ffar1 and ffar2 genes in proximal gut and brain of all trout ffar genes. Overall, this study presents a comprehensive overview of the ffar family in rainbow trout.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12864-023-09181-z.
## Introduction
In addition to act as energy sources, free fatty acids (FFAs) are essential nutrients that contribute to various cellular functions and exert biological effects through several signaling pathways [1]. In order to detect FFAs, the free fatty acid receptors (FFAR), a specific family of G protein coupled receptors (GPCRs), has been identified as the main receptors mediate effects of different FFAs [2]. The successful sequencing of the human genome has greatly accelerated the researches in this domain. Indeed, among various research endeavors benefiting from established genomic information, one of the most fruitful areas is the research on GPCRs. Thus, human FFARs sequences were originally identified as a cluster of four intronless genes located at locus 19q13.1 [3]. This discovery has subsequently allowed the emergence of researches on their functions, their ligands and their regulations [4]. Even if more than 800 types of GPCRs are reported in the human genome, only few of them have been identified and characterized in mammals to be activated by FFAs of various chain lengths [5]. To date, four FFARs have received considerable attention owing to their physiological importance in various biological processes, such as facilitation of insulin and incretin hormone secretion, adipocyte differentiation, anti-inflammatory effects, neuronal responses at least in mammals [6]. In particular, FFAR1 (GPR40) and FFAR4 (GPR120) are activated by long-chain (LC) saturated and unsaturated fatty acids, while FFAR3 (GPR41) and FFAR2 (GPR43) are activated by short chain fatty acids (SCFAs), mainly acetate, butyrate, and propionate [6]. FFARs as GPCRs are widely expressed in various tissues and contribute to many important physiological functions that maintain energy balance and immune homeostasis. Also, among their physiological importance in mammals, FFARs play a crucial role in fatty acid sensing and their regulation especially in brain and gut [7, 8].
In the farmed fish sector, the ongoing expansion of aquaculture enforces an urgent need to find alternative ingredients to replace the use of the traditional ingredients in aquafeed composition, meaning fish meal and fish oil (FM/FO). However, for rainbow trout (Oncorhynchus mikyss), one of the most produced species in Europe, despite 20 years of research [9], the total replacement of marine products by plant ingredients has still not been achieved due to the drastic alteration of survival rate and growth performance [10] which could be mainly related to alteration in the regulation of feeding behaviour and feeding efficiency. Indeed, a disadvantage of using plant ingredients is the modification of the nutrient composition of the diet, especially fatty acid composition. By far, the major difference in terms of nutrient replacement in plant-based diets versus commercial diets (containing FM/FO) is the lack of ω-3 long chain polyunsaturated fatty acids (ω-3 LC-PUFA), mainly eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) [11]. Some studies reported that this absence of ω-3 LC-PUFA in plant products, especially DHA, is known to affect feeding behavior during the whole fish life-cycle by reducing feed intake [12] and feeding rhythms [13], and by increasing abnormal feeding behavior [14]. Recently, we revealed that rainbow trout have the fundamental mechanisms (sensory receptors) for nutrient perception related to different diet composition enriched or not with ω-3 LC-PUFA [15, 16] and that ω-3 LC-PUFA controlled its feeding behavior [17]. Furthermore, we observed that a diet rich in ω-3 LC-PUFA impacted in a relatively high proportion the brain function [18] and brain lipid content [19] in rainbow trout. These recent studies disclosed implications of ω-3 LC-PUFA in the modulation and regulation of the control of feeding behaviour particularly the FA sensing pathways.
Thus, the knowledge on fat detection sensing mechanisms by FFARs in fish and their implication in metabolism regulation is still in its infancy. Especially, in farmed fish, the understanding of the mechanisms regulating feed intake remains a major challenge to be studied for the development and maintenance of a sustainable aquaculture. To achieve optimal animal growth, the development of aquaculture have to rely now on the production of cost-effective and nutritionally adequate aquafeed. It is thus essential to improve fish feeding strategies (optimizing food consumption), by reducing economic losses due to non-ingested food and in the same time reducing the part of FM/FO in the composition of aquafeed.
Firstly, identification and characterization of ffar genes can be crucial to better understand nutritional sensing processes in salmonids. Yet, despite the rapid accumulation of genomic and transcriptomic information in teleost fish, coding sequences for ffar have not been widely characterized yet. Moreover, during vertebrates’ evolution several rounds of whole genome duplication (WGD) occured [20], the first one at the emergence of chondrichthyes, the second at the radiation of teleosts and the teleost-specific genome duplication (3R; TGD). Then an additional duplication occured in the salmonids; Salmonid specific genome duplication, SaGD. These WGD duplications led to genes’ duplications, leading to adaptive innovation via the conservation of duplicated genes available for the evolution of new functions [20]. Thus, to assess more comprehensively the functional importance of ffar in rainbow trout, it seems essential to firstly decipher and characterize them within the trout genome by phylogenetic and syntenic analysis. Secondly, to understand how the identified ffar genes could lead to or at least be one factor involved in the fatty acid perception and their regulation in rainbow trout, it is important to consider their transcriptional behaviour during ontogenic period knowing to be a window of metabolic plasticity and during feeding period to observe their regulation by diet changes.
The aim of the present study was thus to: (i) identify and propose new nomenclature of the ffar genes in rainbow trout by phylogenetic and syntenic analysis; (ii) determine their expression pattern at critical developmental stages and in juvenile trout before and after nutritional challenge by commercial-like diet or by plant-based diet in two mains organs involved in the regulation of energy balance and fatty acid metabolism, i.e. the gut and the brain tissue.
## Ethics statement
The animal study was reviewed and approved by French National Consultative Ethics Committee. The study was conducted according to the guidelines of the National Legislation on Animal Care of the French Ministry of Research (Decree No 2013–118, 1 February 2013) and in accordance with EU legal frameworks relating to the protection of animals used for scientific purposes (i.e. Directive $\frac{2010}{63}$/EU). The experiment was conducted at the INRAE NuMeA facilities (https://doi.org/10.15454/GPYD-AM38), and approved by the ethical committee (C2EA-73) of INRAE “Comité d’éthique Aquitain poissons oiseaux (Aquitaine Fish and Bird Ethics Committee)” (agreement number INRAE 21,699, 19th December, 2019). The scientists in charge of the experiments received training and personal authorization.
## Experimental diets design
Diets were manufactured at INRAE experimental facilities at Donzacq using a twinscrew extruder (Clextral). Pellets size were produced between 4 mm diameter and 4 mm length. Details about the ingredients and composition of the experimental diets are given in Table 1 and the proportions of the main FA in the diets in Table 2. The experiment was conducted with one of the two different experimental diets: a control diet containing a mix of FM ($19\%$), FO ($8\%$) and plant ingredients, and a plant-based diet, completely free from FM and FO, which were replaced by a blend of plant ingredients ($8\%$ of rapeseed oil, $6\%$ of linseed oil and $3.6\%$ palm oil). This vegetal oil blend in plant-based diet was chosen in order to provide an overall amount of FA classes in proportion similar to those of FA classes found in control diet. For plant-based diet diet, DHA and EPA (present in FO for control diet) was replaced to the benefit of alpha-linolenic acid (ALA) by adding linseed oil ($6\%$).Table 1Selected fatty acid composition (% total fatty acids)DietCommercial-like dietPlant-based dietC12:00.080.22C14:03.760.65C15:00.230.05C16:012.7813.65C17:00.230.09C18:02.762.81C20:00.350.34C22:00.220.19C24:00.040.09Sum of saturated fatty acids20.4518.09C14:1 ω-70.090.0C16:1 ω-73.760.55C17:1 ω-70.10.04C18:1 ω-933.0638.55C20:1 ω-91.30.61C22:1 ω-90.90.12Sum of MUFAs39.2039.87C16:2 ω-40.760.02C16:3 ω-40.830.12C18:2 ω-40.190.02C18:3 ω-40.10.08Sum of ω-4 PUFAs1.880.24C18:2 ω-6 (LA)15.8621.23C18:3 ω-60.140.0C20:2 ω-60.100.05C20:3 ω-60.060.0C20:4 ω-6 (ARA)0.480.0C22:2 ω-60.00.08C22:4 ω-60.00.0C22:5 ω-60.120.0Sum of ω-6 LC-PUFAs16.7621.36C16:4 ω-30.00.0C18:3 ω-3 (ALA)4.9618.75C18:4 ω-30.930.09C20:3 ω-30.00.0C20:4 ω-30.30.0C20:5 ω-3 (EPA)7.940.77C21:5 ω-30.320.05C22:5 ω-30.960.08C22:6 ω-3 (DHA)4.060.39Sum of ω-3 LC-PUFAs19.4720.13Sum of ω-3 (EPA + DHA)12.01.16ω-3 (DHA + EPA) / ω-60.720.05Table 2Nucleotide sequence of the PCR primers used to evaluate mRNA expression of FFAR transcripts by RT-qPCRTranscriptForward PrimerReverse PrimerDatabaseAccession NumberReference eef1a1TCCTCTTGGTCGTTTCGCTGACCCGAGGGACATCCTGTGEnsemblENSOMYG00000038328FFARs ffar1ACTGTTGCACCTGAGTCTGGGCTGGTCCTGGGTGAAGTTCEnsemblENSOMYG00000041396 ffar2a1aCCGAGTTCCTCTGCTCCATCTAGGTGATGGGGAAGGCAACEnsemblENSOMYG00000004986 ffar2a2GACAACTTCACCCAGGAGCAAGCAGAACACACAGGCCAGEnsemblENSOMYG00000030315 ffar2b1.1TTTTCCACACACAGTTGGCCAGGTAGTGTTGTCGGCATCTEnsemblENSOMYG00000041393 ffar2b1.2GTGTGGCCTTCCCTATCAGAGCAGGGCACAATGTACACAAEnsemblENSOMYG00000041387 ffar2b2aCCCATCCAACACTCGCTGAATGATGACGACGATGCTCAGGEnsemblENSOMYG00000030493 ffar2b2b1TGACCGCAATCAGTGTCGAACCCAGAAGAAGACGCTAGCCEnsemblENSOMYG00000030500 ffar2b2b2GTCCAGTACCATCAACGCCACTGCACACTCTCCAACAGGGTEnsemblENSOMYG00000005604 The two experimental diets contained $23.6\%$ for commercial-like diet and $20.35\%$ for plant-based diet of crude lipids with the same amount of major ω-3 FA; $19.47\%$ in commercial-like diet and $20.13\%$ in plant-based diet. This amount of ω-3 FA class was chosen in order to be close to the proportions of ω-3 FA classes found in marine diet [10]. In order to avoid exceeding anti-nutrient threshold levels, we used a blend of wheat gluten, soybean meal and whole wheat, corn gluten meal, soy protein and fatabean as protein sources (c. $46.11\%$ of total diet). Synthetic L-lysine, L-methionine, dicalciumphosphate and soy-lecithin were added to all diets to correct the deficiency in essential amino acids, phosphorous and phospholipids. Mineral and vitamin premix were added to each diet. Diets were isoenergetic (c. 24.41 kJg-1 of dry diet) and were formulated to cover the nutritional requirements of the rainbow trout [21]. Nutrient compositions of the diets, crude protein and lipids, gross energy, ash and starch content and fatty acid profils were analyzed as previously described [16].
## Experimental design
For ontogenesis analysis, oocytes were fertilized synchronously with neomale sperm and reared in separate tanks at 8 °C in our experimental facilities (INRAE Fish Farm of Lees-Athas, Permit number A64.104.1, vallée d’Aspe, France) as previously described [22]. Rainbow trout were sampled according to spawn origin before fertilization (oocyte) and then during development according to Vernier [23] at stages 5, 6, 7, 8, 10, 12, 15, 22 and 23. Embryos were directly snap-frozen, whereas alevins were killed by terminal anaesthetization by bathing in benzocaine prior to pooling and storage in liquid nitrogen. The samples were stored at - 80 °C until mRNA analysis.
For nutritional analysis, at the beginning of the experiment, female rainbow trout fry come from the same parental stock (INRAE Fish Farm of Lees-Athas, Permit number A64.104.1, vallée d’Aspe, France). The feeding experiment was conducted in a recirculating rearing system at the INRAE facilities of Donzacq, France (authorisation number A40-228.1, Landes). At the beginning of the experiment, female rainbow trout with mean weight of 140 g were randomly distributed among 6 tanks of 100L (50 fish per tank). Water flow was set to ensure an oxygen concentration above $90\%$ saturation. Fish were exposed to natural photoperiod condition and the water temperature was set at 15 ± 1 °C. During the trial, water dissolved oxygen was 9 mg L-1, ammonia < 0.01 mg L-1, nitrite < 0.04 mg L-1, nitrate was about 17 ppm. The quantity of flow was 0.3 l/s by tank, all water of each tank was changed 6 times each hour. During 30 days, all fish were fed by hand twice a day with an interval 8 h with commercial-like diet, until apparent satiety. After 30 days, 5 days of fasted period was realized. After that, trout were fed a single meal with commercial-like diet for three tanks and plant-based diet (Diet composition Table 3) for the three other tank. For sampling, fish was killed by terminal anesthetization by bathing in benzocaine (30 mg/l then a bath at 60 mg/l) and proximal gut and whole brain were sampled before the single meal ($T = 0$), 20mns after the single meal ($T = 0.2$), 4 h after the single meal ($T = 4$) and 24 h after the the single meal ($T = 24$). Tissues were then frozen in liquid nitrogen and stored at - 80 °C until mRNA analysis. Table 3Ingredients and composition of the experimental dietsIngredient (%)DIETCommercial-like dietPlant-based dietFish meal19.00.0Soybean meal11.018.0Extruded whole wheat13.511.5Corn gluten6.015.2Wheat gluten0.08.0Soy protein concentrate18.06.8Fababean protein concentrate9.58.0Soy lecithin2.52.5L-Lysine0.67.16L-methionine0.60.6CaHPO4.2H2O0.11.44Mineral premixa1.42.0Vitamin premixb1.31.2Fish oil8.00.0Rapeseed oil8.58.0Palm oil0.03.6Linseed oil0.06.0Composition (% of dry matter)Dry matter (in % of diet)97.6094.83Crude protein45.2846.95Crude lipid23.620.35Starch17.6611.37Ash4.107.58Energy (kJg-1 DM)24.4924.33aMineral premix: (g or mg kg - 1 diet): calcium carbonate ($40\%$ Ca), 2.15 g; magnesium oxide ($60\%$ Mg), 1.24 g; ferric citrate, 0.2 g; potassium iodide ($75\%$ I), 0.4 mg; zinc sulphate ($36\%$ Zn), 0.4 g; copper sulphate ($25\%$ Cu), 0.3 g; manganese sulphate ($33\%$ Mn), 0.3 g; dibasic calcium phosphate ($20\%$ Ca, $18\%$ P), 5 g; cobalt sulphate, 2 mg; sodium selenite ($30\%$ Se), 3 mg; KCl, 0.9 g; NaCl, 0.4 g (UPAE, INRA)bVitamin premix: (IU or mg kg - 1 diet): DL-a tocopherol acetate, 60 IU; sodium menadione bisulphate, 5 mg; retinyl acetate, 15,000 IU; DL-cholecalciferol, 3000 IU; thiamin, 15 mg; riboflavin, 30 mg; pyridoxine, 15 mg; B12, 0.05 mg; nicotinic acid, 175 mg; folic acid, 500 mg; inositol, 1000 mg; biotin, 2.5 mg; calcium pantothenate, 50 mg; choline chloride, 2000 mg (UPAE, INRA)
## In silico analysis
*Ffars* genes and related protein sequences were identified in the Genomicus software program, version 106.01, 2021–08-15 (https://www.genomicus.bio.ens.psl.eu/genomicus-106.01/cgi-bin/search.pl) and collected from Ensembl (http://www.ensembl.org, Ensembl Release 102; November 2020, RT genome available). The genoscope database (http://www.genoscope.cns.fr/ trout) was used to identify ffar related genes in the rainbow trout genome using BLAST analysis. Sequences are available under the accession numbers reported in Table 3. Ensembl database (http://www.ensembl.org/index.html) was also used to collect amino acids deduced sequences of ffar for all species studied.
Protein alignment and the percentage Identity Matrix established with amino acids deduced sequences were performed using MUSCLE software (https://www.ebi.ac.uk/Tools/msa/muscle/). The Protein sequence encoded by FFAR of trout are presented in supplemental information.
The evolutionary history of the four ffar was inferred using the Neighbor-Joining method [24]. The optimal phylogenetic tree is shown. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (500 replicates) are shown next to the branches [25]. The tree is drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the Poisson correction method [26] and are in the units of the number of amino acid substitutions per site. This analysis involved 57 amino acid sequences. All ambiguous positions were removed for each sequence pair (pairwise deletion option). There were a total of 530 positions in the final dataset. Evolutionary analyses were conducted in MEGA11 [27].
Gene synteny analysis was carried out for all ffar genes between salmonid and other relevant genomes using Genomicus software. To determine the genomic neighbourhood around candidate genes and the conservation of gene order across species, genes were visually examined in NCBI’s genomic region browser and Ensembl Gene Summary databases. The phylogenetic analysis showed that all the gar and teleosts sequences rooted together with FFAR2 tetrapods sequences and were divided in 2 distinct sub-trees that we respectively named ffar2a and ffar2b.
## mRNA levels measurement by real-time quantitative PCR
Total RNA was extracted from the oocytes and embryos ($$n = 3$$ pools), proximal gut and brain, ($$n = 6$$ per fish) using the TRIzol reagent method (Invitrogen, Carlsbad, CA) with Precellys®24 (Bertin technologies, Montigny le Bretonneux, France) following Trizol manufacturer’s instructions. Luciferase control RNA (Promega), 10 pg per 1.9 mg of embryo/alevin or oocyte, was added to each sample for ontogenesis analysis to allow for data normalization during early development as previously described [22, 28]. Total RNA (2 µg) was used for cDNA synthesis. RNA purity was tested by optical density (OD) absorption ratio (OD 260 nm/280 nm) using a NanoDrop 2000c (Thermo, Vantaa, Finland), and only samples with an OD 260 nm/280 nm ratio > 1.8 were used for analysis. The Super-Script III RNAse H-Reverse transcriptase kit (Invitrogen) was used with random primers (Promega, Chartonniéres-les-bains, France) to synthesize cDNA in a final volume reaction of 20 µl, according to the manufacturer’s instructions. QPCR assays were performed with the Roche Lightcycler 480 system (Roche Diagnostics, Neuilly-sur-Seine, France). The reaction mix was 6 µl per sample, including 2 µl of diluted cDNA template (1:10), 0.24 µl of each primer (10 µmol l-1), 3 µl of Light Cycler 480 SYBR® Green I Master mix and 0.52 µl of DNAse/RNAse-free water (5 Prime GmbH, Hamburg, Germany). The qPCR protocol was initiated at 95 °C for 10 min for the initial denaturation of the cDNA and hot-start Taq-polymerase activation, followed by 45 cycles of a two-step amplification program (15 s at 95 °C; 10 s at 60 °C) [16]. Cycle thresholds values superior than 35 cycles were not considered. Melting curves were monitored systematically (temperature gradient 0.11 °C per second from 65 to 97 °C) at the end of the last amplification cycle to confirm the specificity of the amplification reaction. Duplicate wells were used for each sample and negative controls were included in all reactions, consisting in wells containing RNA samples and water instead of cDNA. Efficiency of all qPCR reactions was 96–$100\%$ and R2 was 0.95–1. For ontogenic analysis, data were subsequently normalized to the exogenous luciferase transcript as previously described [22]. The different PCR products were initially checked by sequencing to confirm the nature of the amplicon. For this, primers were tested on a pool of cDNA and amplified PCR products were sequenced. Primer sequences and accession numbers are presented in Table 2. For nutritional analysis, data from mRNA gut and brain were extrapolated from standard curves and normalized to the housekeeping gene after validation; elongation factor 1a gene (eef1a). Relative expression of the target genes was determined by the ΔΔCT method [29]. mRNA sequences of trout found and used in this study are available in NCBI and Ensembl genome browser.
## Statistical analysis
All statistical analyses were performed using R software (version 3.6.1, R development Core Team, 2008)/Commander package. Data are presented as mean ± standard error of the mean (SEM). Analyses were carried out on untransformed data as criteria for normality and homogeneity of variances were fulfilled (Shapiro–Wilk’s and Levene’s test, respectively). If the criteria (normality and homogeneity) were still not met, a non-parametric test was used for the analysis. Diet effect and time effect were analyzed using two-way ANOVA. If interaction was detected (p-value <0.05), data were finally analyzed using one-way ANOVA to test the diet effect and time effect individually. A Tukey’s was used as a post hoc test (p-value < 0.05).
## Phylogenetic analysis of ffar1
By analysing key vertebrate genomes available in Ensembl (Release 102; November 2020), we found one gene related to the sarcopterygii ffar1 in the spotted gar as well as in the analysed teleosts (except for common carp with two encoding genes) including salmonids (Fig. 1). Our phylogenetic analysis showed that both gar and teleost sequences grouped together with tetrapods Ffar1 protein sequences and more generally with actinopterygii selected sequences. Fig. 1Phylogenetic tree of ffar1 in vertebrates. FFAR1 protein sequences were aligned using the Maximum Likelihood Method (with Poisson model). Following alignment, the phylogenetic tree was constructed using the neighbour-joining method in Molecular Evolutionary Genetics Analysis (MEGA) software version 7.0 (Tamura 2013). The branch support values were gained by non-parametric bootstrapping (500 replicates). The scale bar represents the calculated evolutionary distance. Genbank accession numbers (from Ensembl or Genoscope databases) are specifed for each species. Mammalian and teleost FFAR2 protein sequences were used to root the tree
## Phylogenetic and syntenic analysis of ffar2
We first collected annotated Ffar2 amino acids sequences in selected tetrapod species in Ensembl. We found only one annotated sequence for each selected species (i.e. mouse, rat, human, pig, chicken). Using the Blast function in Ensembl and Genomicus software V.106.01, we collected tetrapod-Ffar2 related sequences in selected teleosts species including 3 salmonids (i.e. huchen, atlantic salmon, rainbow trout), species selected to cover the major orders of teleosts. We identified three amino acids sequences related to tetrapods Ffar2 in the spotted gar genome, and between three and nine sequences in teleosts (4 for the zebrafish (Danio rerio), 6 for the *Mexican tetra* (Astyanax mexicanus), 4 for the *Nile tilapia* (Oreochromis niloticus), 3 for the *Japanese medaka* (Oryzias latipes), 3 for the fugu (Takifugu), 9 for the huchen (hucho hucho), 8 for the atlantic salmon (salmo salar) and 7 for rainbow trout).
Regarding the ffar2a sub-tree (Fig. 2), the sequence ENSOMYG00000004986 from the rainbow trout grouped together with other teleosts sequences whereas a subtree was composed of the trout sequence ENSOMYG00000030315 grouping together with additonnal salmonid sequences (i.e. sequences ENSSAG00000053198 in salmon and ENSHHUG00000014048 in huchen). We named these two sequences ffar2a1a and ffar2a2, respectively. This tree configuration is in favour of a ffar2a duplication occurring before or around the teleost radiation with a probable loss of one copy of the gene in non-salmonid species. Such hypothesis seemed to be confirmed by our synthenic analysis (Fig. 3) showing that the syntenic group usf2-ffar2a-ffar2b-lim2.4 containing both ffar2a and ffar2b in the spotted gar on chromosome LG24 is found duplicated in zebrafish on chromosome 16 and 19 but with a loss of ffar2a loci on the lastest chromosome. In addition, the unique trout ffar2a1a sequence grouped together in the phylogenetic tree with 2 sequences in salmon and 2 in huchen (ENSSAG00000068149, ENSSAG00000093514 and ENSHHUG00000032561, ENSHHUG00000014051, respectively) which were included in the same synthenic group (hamp-usf2-etfb-lim2-ffar2a-xrcc1-ethe1) found duplicated on 2 distincts chromosomes in these 2 species (ssa02 and ssa05 in salmon and QNTS01001724.1 and QNTS01000300.1 in huchen; data not shown).Fig. 2Phylogenetic tree of ffar2 in vertebrates. FFAR2 protein sequences were aligned using the Maximum Likelihood Method (with Poisson model). Following alignment, the phylogenetic tree was constructed using the neighbour-joining method in Molecular Evolutionary Genetics Analysis (MEGA) software version 7.0 (Tamura 2013). The branch support values were gained by non-parametric bootstrapping (500 replicates). The scale bar represents the calculated evolutionary distance. Genbank accession numbers (from Ensembl or Genoscope databases) are specifed for each species. Mammalian and teleost FFAR2 protein sequences were used to root the treeFig. 3ffar2 gene synteny in selected vertebrates. The syntenically conserved gene blocks are shown in matching colours. Gene synteny was compiled from up- and down-stream locations relative to each species ffar2 taken from NCBI’s genome browser and using Genomicus software (https://www.genomicus.bio.ens.psl.eu/genomicus-100.01/cgi-bin/search.pl). Species names are displayed at the top of the figure, chromosome number and range (position) are shown above each species gene synteny. Chr., chromosome; LG., liguleless gene The sub-tree that we named ffar2b, was in turn divided into 2 main subtrees respectively called ffar2b1 and ffar2b2 containing respectively 2 (accession number ENSOMYG000000041393 and ENSOMYG000000041387) and 3 (accessions numbers ENSOMYG00000030493, ENSOMYG00000030500, ENSOMYG00000005604) rainbow trout sequences.
In the ffar2b1 subtree, we named these two sequences ffar2b1.1 and ffar2b1.2, respectively.
These two sequences for zebrafish (ENSDARG000000063088 and ENSDARG000000058535) and mexican tetra (ENSAMXG000000042863 and ENSAMXG000000024912) grouped together. The sequence ENSOMYG000000041393 (ffar2b1.1) from the rainbow trout grouped together only with salmonids sequences (i.e. sequences ENSSAG00000069128 in salmon and ENSHHUG00000038647 in huchen). The sequence ENSOMYG000000041387 (ffar2b1.2) from the rainbow trout grouped together with others teleost sequences (i.e. sequences ENSONIG00000043224 in nile tilapia, ENSTRUG00000012836 in fugu, ENSSAG00000069124 in salmon and ENSHHUG00000038645 in huchen). This tree configuration is in a favour of a ffar2b1 tandem duplication ffar2b1 (into b1.1, b1.2) resulting of the duplication in salmonids.
The syntenic analysis (Fig. 3) seemed to confirm this hypothesis showing that this two sequences ffar2b1.1 and ffar2b1.2 were included in the same synthenic group (ago3-olah-tekt2-ffar2b1-usf2-lim2.4) for zebrafish and mexican tetra (data not shown) and same synthenic group (tomm40-ddx6-olah-ago3-ffar2b1-usf2) for rainbow trout, atlantic salmon, huchen (data not shown for salmon and huchen), nile tilapia and fugu. ffar2b1 were found duplicated on same chromosome in this syntenic group for zebrafish (ffar2b1.1 and ffar2b1.2, Chr.19), mexican tetra (APWO02000060.1), rainbow trout (Chr.18), atlantic salmon (ssa27), huchen (QNTS01000383.1; data not shown for salmon and huchen) but with one copy for nile tilapia (LG.22) and fugu (Chr12).
Considering the ffar2b2 subtree, the three sequences were in turn divided into 2 others subtrees respectively called ffar2b2a containing respectively 1 (ENSOMYG000000030493) rainbow trout sequence and ffar2b2b containing respectively 2 (accessions numbers ENSOMYG000000030493) rainbow trout sequences (accessions numbers ENSOMYG00000030500, ENSOMYG00000005604) and named ffar2b2b1 and ffar2b2b2 respectively. The sequences ENSOMYG000000030493 and ENSOMYG000000030500 from the rainbow trout grouped together with two other teleosts sequences respectively (i.e. sequences ENSSAG00000052952 and ENSSAG00000052945 in salmon and ENSHHUG00000022234 and ENSHHUG00000014406 in huchen) and with zebrafish (ENSDARG000000079661), *Mexican tetra* (ENSAMXG000000042613 and ENSAMXG000000025831). This tree configuration is in a favour of a ffar2b2 duplication occurring before or around teleost radiation. The syntenic analysis (Fig. 3) seemed to confirm this hypothesis showing that this two sequences ffar2b2a and ffar2b2b were included in the same synthenic group. Our synthenic analysis (Fig. 3) showed that the syntenic group etfb-usf2-hamp-iglon5-ffar2b2a-ffar2b2b-ago1-ceacam6) containing both ffar2b2a and ffar2b2b in trout on chromosome 2 was found in zebrafish on chromosome 16 containing one sequence of ffar2b2b, in *Mexican tetra* on chromosome 10 containing two sequence of ffar2b2b (ffar2b2b1, ffar2b2b2; data not shown) and in atlantic salmon on ssa05.
For the two rainbow trout ffar2b2b sequences, they grouped together (ENSOMYG000000030500 and ENSOMYG000000004604) in chromosome 2 and chromosome 3 respectively. These two sequences grouped together with two other teleosts sequences respectively (i.e. sequences ENSSAG00000052945 and ENSSAG00000068216 in salmon and ENSHHUG00000014406 and ENSHHUG00000022237 in huchen). These two ffar2b2b sequences for mexican tetra (ENSAMXG000000042613 and ENSAMXG000000043089) grouped together with one copie for zebrafish (ENSDARG000000079661). This tree configuration is in a favour of a ffar2b2b duplication occurring before or around salmonid radiation. Such hypothesis seemed to be confirmed by our synthenic analysis (Fig. 3) showing that the syntenic group etfb-usf2-hamp-iglon5-ffar2b2b-ago1-dhdds is found in trout on chromosome 2 and 3 surrounding ffar2b2b1 and ffar2b2b2, in atlantic salmon on ssa05 and ssa02 (data not shown) and in huchen on QNTS01000300.1 and QNTS01001258.1 (only for iglon5-ffar2b2b-ago1-dhdds (data not shown).
## Phylogenetic analysis of ffar3
Through genome mining and phylogenetic analysis, we identified one gene related to the ffar3 encoding gene in mammals including platypus, but no sequence was found in spotted gar nor in analysed teleosts (Fig. 4). Based on the automatic annotation provided in Ensembl, ffar3 appeared to be lost in actinopterygii. Fig. 4Phylogenetic tree of ffar3 in vertebrates. FFAR3 protein sequences were aligned using the Maximum Likelihood Method (with Poisson model). Following alignment, the phylogenetic tree was constructed using the neighbour-joining method in Molecular Evolutionary Genetics Analysis (MEGA) software version 7.0 (Tamura 2013). The branch support values were gained by non-parametric bootstrapping (500 replicates). The scale bar represents the calculated evolutionary distance. Genbank accession numbers (from Ensembl or Genoscope databases) are specifed for each species
## Phylogenetic analysis of ffar4
Through genome mining and phylogenetic analysis, we identified one gene related to the ffar4 encoding gene in Sarcopterygii (Fig. 5). No sequence was found for spotted gar, and coelacanth. For teleost fish, we found no, one or two genes related to the ffar4 encoding gene in characiphysae (none for zebrafish, one in astyanax or red-bellied piranha and two in channel catfish) depending on the species. For percomorphaceae, one gene was found for pinecone soldierfish, Eupercaria (gilthead seabream), carangaria (greater amberjack, yellowtail amberjack and barramundi perch), cichlidae (nile tilapia), and one for pomacentridae (spiny chromis, clow anemonefish and orange clownfish). *No* gene was found for salmonids fish. The *Phylogenetic analysis* grouped the different Ffar4 protein sequences of teleosts closest to mammalian FFAR4.Fig. 5Phylogenetic tree of ffar4 in vertebrates. FFAR4 protein sequences were aligned using the Maximum Likelihood Method (with Poisson model). Following alignment, the phylogenetic tree was constructed using the neighbour-joining method in Molecular Evolutionary Genetics Analysis (MEGA) software version 7.0 (Tamura 2013). The branch support values were gained by non-parametric bootstrapping (500 replicates). The scale bar represents the calculated evolutionary distance. Genbank accession numbers (from Ensembl or Genoscope databases) are specifed for each species
## mRNA levels of ffar genes during embryonic development.
Real-time PCR was performed to determine the stage-specific mRNA levels of rainbow trout ffar-related genes during embryogenesis and in hatched alevins. This analysis showed (Fig. 6) that all mRNA sequences encoding ffar genes were detected and that mRNA levels increased after stage 10 to reach a maximum level at stages 23 (corresponding to the setting up of primitive organs) before to drastically decrease at alevin stage. Fig. 6Expression pattern of ffar1 and ffar2 transcripts during development. *Relative* gene expression measured by RT-PCR of ffar genes family (ffar1, ffar2a1, ffar2a2, ffar2b1.1 ffar2b1.2, ffar2b2a, ffar2b2b1, ffar2b2b2) during ontogenesis. Embryos were sampled according to Vernier [1969] at stages oocyte [0], 5, 6, 7, 8, 10, 12, 15, 22 and 23, avelin [31]. For all stages, gene expression level was normalized by the abundance of exogenous luciferase RNA. Data are expressed as means ± s.e.m. ( $$n = 3$$ pools of embryos, one pool of 30 embryos per spawn). Different letters indicate significant differences between conditions ($P \leq 0.05$)
## mRNA levels of ffar genes after feeding response in gut and brain of rainbow trout fed by a challenge with plant based diet.
After a single meal following five fasted days, mRNA levels of ffar1 and ffar2 in proximal gut and brain of rainbow trout are presented in Figs. 7 and 8 respectively. All mRNA sequences encoding ffars were detected. Fig. 7Expression pattern of ffar1 and ffar2 transcripts in proximal gut of rainbow trout fed with a commercial and plant-based diet following 5 fasted days. *Relative* gene expression measured by RT-PCR of ffar genes family (ffar1, ffar2a1a, ffar2a2, ffar2b1.1 ffar2b1.2, ffar2b2a, ffar2b2b1, ffar2b2b2) in proximal gut. Values are expressed as group mean ± SEM ($$n = 6$$); two-way ANOVA followed by Tukey post hoc test; if interaction (diet × time), letters indicate a significant difference between conditions as determined by a one-way ANOVA ($p \leq 0.05$)Fig. 8Expression pattern of ffar1 and ffar2 transcripts in brain of rainbow trout fed with a commercial and plant-based diet following 5 fasted days. *Relative* gene expression measured by RT-PCR of ffar genes family (ffar1, ffar2a1a, ffar2a2, ffar2b1.1 ffar2b1.2, ffar2b2a, ffar2b2b1, ffar2b2b2) in brain. Values are expressed as group mean ± SEM ($$n = 6$$); two-way ANOVA followed by Tukey post hoc test; if interaction (diet × time), letters indicate a significant difference between conditions as determined by a one-way ANOVA ($p \leq 0.05$) In gut, mRNA level of ffar1 was significantly affected by the interaction between the diet and the time after the meal with an increase in fish fed with commercial-like compared to those fed the plant-based diet (Fig. 7). Moreover, mRNA levels of ffar2a2 and ffar2b2a were affected by the time after the meal but independently of the diet. The mRNA levels of ffar2a1a, ffar2b1.1, ffar2b1.2, ffar2b2b1 and ffar2b2b2 genes remained stable in time between fish fed with commercial-like and plant-based diet.
In brain, the mRNA levels of ffar1 and ffar2a2 were significantly affected by the diet, the time after the meal and the interaction between these two factors. ffar1 and ffar2a2 mRNA level were increased for fish fed with commercial-like vs plant-based diet 20 min after the meal (Fig. 8). The mRNA levels of ffar2a1a and ffar2b1.1 were significantly affected by the time (increase and decrease respectively) after the meal, independantly of the diet but with the interaction between these two factors. ffar2a1a and ffar2b1.1 mRNA level were increased for fish fed with commercial-like vs plant-based diet 20 min after the meal and decrease 24 h after the meal for same diet.
The mRNA levels of ffar2b2a, ffar2b2b1 and ffar2b2b2 were significantly affected by the interaction between the diet and the time after the meal (with diet effect for ffar2b2b1) with an increase of their mRNA level for fish fed with commercial-like vs plant-based diet 20 min after the meal. The mRNA levels of ffar2b1.2 gene remained stable in time after the meal with commercial-like or plant-based diet.
## Discussion
In the present study, we deciphered a genomic overwiew of the ffar family in rainbow trout by phylogenetic and syntenic methods. We also integrated molecular measures to understand the temporal dynamics during development and in different nutritional conditions of the modulation of these ffars towards a better understanding of their functions in fatty acid regulation. Specifically, taking into consideration the complexity of the trout genome, our aim was to provide original understanding of ffar genes expression during ontogenesis, and to study the time course patterns of their genic regulation in gut and brain of trout fed with commercial or plant-based diet.
## Genomic overwiew of ffar genes in rainbow trout
Although only de-orphanised (discovery of ligands) in the recent past, FFARs responding to FFAs have attracted considerable attention in mammals [30]. Despite the importance of this family in numerous functions, there are still gaps in the knowledge of the teleost-specific FFARs related to evolution, nomenclature and function, which is exacerbated within the salmonids due to the SaGD (4R).
The recent availability of assembled genomes for rainbow trout provide an exhaustive research to establish the ffar genes repertoire and especially new coherent nomenclature of ffar2 in rainbow trout. Hovewer, although well-assembled, the trout genome is not necessarily correctly annotated. This study has therefore carried out a deep phylogenetic and syntenic analysis in order to ensure also these annotations of the different sequences of ffar reported in the trout genome. This study identified no ffar3 nor ffar4, one ffar1 related gene and seven ffar2 genes in trout. These first in silico analyses identified a wrong annotation of ffar sequences in trout where two ffar2 (ffar2b1.1 ffar2b1.2) sequences were initially annotated as ffar3 (ENSOMYG00000041393 and ENSOMYG00000041387) and ffar1 was not annotated in trout genome.
Phylogenetic analysis allowed to classify ffar2 into two subclasses, that we named class a and b. Syntenic analysis showed a clear conservation of genomic organisation, further supporting the identity of these genes as ffar2.
For the first sub-tree, ffar2a, our phylogenetic analysis suggested that the duplication before or around the teleost radiation (TGD, into ffar2a1 and ffar2a2) has resulted with a loss of ffar2a in non-salmonids fish but specie-specific with only one copie is found in zebrafish, mexican tetra and Medaka but two in nile tilapia. Moreover, the duplication before or around the salmonid radiation (SaGD; into ffar2a1a, ffar2a1b and ffar2a2) was followed by a loss of one ffar2a2 sequence. This result is in accordance with Berthelot et al. findings who suggested that [20], when one loss of sequence is observed after the TGD, most of the time one loss is observed in SaGD, by a process termed gene fractionation [31]. Finally, our results suggested an additional loss in rainbow trout, with the presence of ffar2a1b sequence in huchen and atlantic salmon but not in trout.
Concerning ffar2b1 sub-tree, our analysis suggested that the duplication before or around the teleost radiation has resulted with a tandem duplication for ffar2b1 (into ffar2b1.1 and ffar2b1.2) resulting of the duplication in salmonids. Here again, as suggested by Berthelot et al., [ 20], the genes retained in duplicated copies after the successive TGD events that occurred during vertebrate evolution were also more likely to be retained as duplicates following SaGD.
For ffar2b2 sub-tree, our results supported the same conclusion than for ffar2a sub-tree with a duplication (into ffar2b2a and ffar2b2b) before or around the teleost radiation and with another duplication before or around the salmonid radiation. Finally, our results suggested an additional loss in rainbow trout for ffar2b2a with only one copie.
## mRNA levels of ffar genes increase at the formation of the primitive organs during embryo development
The early increase in mRNA levels of ffar-related genes (at stage 23, corresponding to the setting up of primitive organs) followed by an important decrease of these transcript at alevin stage could supported an important role of these receptors in the regulation of fatty acid. Indeed, this regulation would follow a massive production of energy (lipids) at the end of embryo stage. This could also reflect a phase of preparation for the catabolism of dietary nutrients at first feeding, as proposed for digestive enzymes like lipase [32] or previously observed in strickly same pattern of these results for duplicated glucose metabolism-related genes in rainbow trout [22]. Finally, our data may suggest an important role for each of these receptors at an early stage in rainbow trout, even before a role in the regulation of dietary fatty acids.
## Divergence patterns of mRNA levels expression of ffar genes in the fatty acid regulation in gut and brain of rainbow trout after nutritional challenge.
The conservation of duplicated ffar2 genes in the trout genome offered an interesting model to study potential divergences in both the function and the expression of the related paralogues. Indeed, as suggested by Force et al., [ 33] the new duplicated genes can acquire new expression patterns potentially leading to neo- or subfunctionalization which could support the rise of new molecular and cellular functions, and can play an important role in phenotypic variability but could also be either silenced.
FFARs (class 1, 2, 3 and 4) are known to be responsible in part for the various biological and physiological functions of FFAs through their binding to these receptors [34]. Indeed, FFAs are not only an essential energy source, but also function as signaling molecules that regulate various cellular processes and physiological functions according to carbon chain length via FFARs activation through their binding. In mammals, FFARs are expressed in various tissues and influenced many important metabolic functions that maintain energy homeostasis [35]. Moreover, the FFARs-mediated signaling transduction pathways in the regulation of metabolism in intestinal tract is well known as well as the biological role (energy balance, immune responses, fat preference…) of the FFAs via the activation of the FFARs in central nervous system [6]. Our phylogenetic analysis revealed that ffar3 and ffar4 were not present in rainbow trout genome. Knowing the important role of these two FFARs receptor in mammals especially their implication in various biological and physiological functions such as energy regulation, immunological homeostasis, and neuronal functions to the regulation of energy homeostasis [36, 37], this is a surprising finding. However, all FFARs are known to have similar roles even thought the fatty acids (short vs medium or long chain) that bind to it differ. Thus, we could hypothesise that the ffar1 and the seven ffar2 paralogues could assumed the functions of the lost ffar3 and ffar4 reported in others species.
Firstly, the detection of all ffar transcripts in gut and brain tissues of rainbow trout could assumed that ffar1 and ffar2 could have an important role in the digestive tract and central nervous system in rainbow trout. Especially, their modulation by diet changes could assume a role in the regulation of fatty acid that varies according to the diet composition. Considering that FFAR1 are known to be activated by ω-3 LC-PUFA, while FFAR2 are activated by SCFAs in mammals [6], based on our results, it could be also the case in rainbow trout. Indeed, a unique meal of plant-based diet totally devoid of LC-FFAs decreased ffar1 mRNA level 20 min and 6 h after the meal respectively in brain and gut and 20 min for all ffar2 mRNA level except for ffar2b1.2 in brain of trout. Even if no study has been done on teleost fish, these results were consistent with the time course expression of others receptor genes (peroxisome proliferator activated receptor) involved in fatty acid metabolism [38]. Authors concluded that FFAs were able to rapidly induce (less than 1 h with higher expression to 6 h) the expression in muscle cells of mice of key genes involved in their catabolism and that the LC-PUFAs mixture had a positive role increasing the expression of master metabolic regulators and their downstream target gene. Furthermore, Mobraten et al., revealed that LC-PUFAs enhanced the cytosolic concentration of the signaling pathways of FFARs (calcium and MAP kinase ERK$\frac{1}{2}$) with the same efficiency, but with different kinetics (on average 20 min) and intensity in muscle cells [39]. FFAR1 is also reported to be specifically activated by LC-PUFA (DHA) in primary cortical neurons of mouse model, significantly alleviated cognitive functions in mice. This effect was mediated by an increase of intracellular calcium less than 20 min and by the extracellular receptor kinase (ERK) and P38-mitogen-activated protein kinase (MAPK) pathways after FFAR1 activation [40]. All these finding support our present results suggesting that the biological role of LC-PUFAs could be depending in rainbow trout by their binding to the FFAR family then their rapid modulation (than 20 min).
*About* gene function, if we assume that as in mammals the activation of FFAR is associated with increase of genic expression, we can make some assumption to their potential function in trout based on the results. Our results were in agreement with other studies [41], namely an increase of the genic expression of ffar4 (GPR120) just after meal, which could regulate the postprandial mechanisms of fat eating behavior. The increase of ffars expression can indicated an activation of receptors, the FAs released in the mouth cavity activated receptors in brain tissues and after 6 h, the expression of receptors return to a basal condition. The literature shows us that, the early activation of FFAR2 could have a role in the regulation of appetite through a variety of mechanisms related to its activation [42]. Finally, numerous studies concluded that FFAR1 and FFAR4, could plays a critical role in various physiologic homeostasis mechanisms especially the regulation of appetite, eating disorder, or food preference [6, 43, 44]. Moreover, it is known that the activation of FFAR is related to gene expression of ffars increased after a meal. Ozdener et al., demonstrated that, in humans and mice, an incubation of taste bud cells with an ω-3 LC-PUFA (ALA) induced an upregulation of ffar4 [45]. A report of Choo et al. [ 2020], which studied the effect of maternal obesity on the expression of receptors in the offspring, observed an increase of ffar4 expression of female offspring of high-fat diet (HFD) fed mice [46]. These studies are in agreement with our results, which could suggest an activation of FFAR lead to differential physiologic homeostasis mechanisms in trout fed with C diet especially in the regulation of appetite, eating disorder, or food preference. FFAR receptors could have a preponderant role in the growth performance throughout the life cycle of the rainbow trout.
Moreover, these finding demonstrated that gene duplication events of ffar2 offered an interesting model to study potential divergences in both the function and the expression of the related paralogues. Thus, the expression pattern between all ffar2 paralogues were not different in gut tissue where as discrepancies in responses are observed in the brain. In fact, if we consider that all ffar2 (except ffar2a2) were differentially regulated by the diet and specifically at 20 min with up-regulation for trout fed commercial-like diet, many divergent patterns between paralogues were observed. Interestingly for ffar2a genes, ffar2a2 and ffar2a1a genes displayed the same expression pattern. For ffar2b genes, the two paralogues ffar2b2b1, ffar2b2b2 and ffar2b2a genes displayed the same expression pattern where as the tandem duplicated ffar2b1.1 and ffar2b2.2 genes had divergences of their expression after a meal. This finding demonstrated that even if genes are duplicated and therefore very close at the phylogenetic and syntenic level, the modulation of their expression is complex and most of the time results from the combination of different mechanisms. Indeed, gene duplications provide an essential source of genetic redundancy but does not necessarily tenfold the function of the gene. In salmonids, loss of gene functionality is slow, with only $50\%$ of genes loosing function after 50 million years [47] and these duplicated genes may instead undergo neo- or sub-functionalization [33]. All these results here confirmed that ffar2 paralogues in rainbow trout were sub- or neo-functionalized as they were differentially regulated by nutritional statut and/or by the meal.
## Conclusion
Overall, for the first time in rainbow trout, through genome mining and phylogenetic analysis, we identified and characterised 7 coding sequences for ffar2 in salmonid species where as no ffar3 and ffar4 gene have been reported. The differential expression of trout ffar2 genes identified here by nutritional status or feeding may therefore provide evidence of the varying functions of these duplicated genes in rainbow trout. The quantitative assays designed here for individual ffar genes will improve our ability to conduct expression studies as they allow for a more precise characterization of expression and can be utilized to unravel the potential contribution of individual ffar genes in rainbow trout in their various functions. For example, further studies will be necessary to characterize the potential binding (agonist/antagonist) and the role of these individual receptors in the detection and regulation of FFA sensing, metabolism and role in rainbow trout and to elucidate their implication in the regulation of feeding behavior. This knowledge will be important in the aquaculture industry for diversification or substitution of feed ingredients, especially the already expensive and limited FM/FO.
## Supplementary Information
Additional file 1. Normalized and raw RT-PCR datas in all analyzed tissus and conditions. Additional file 2: Supplemental information. Protein sequence encoded by FFAR.
## References
1. Calder PC. **Functional Roles of Fatty Acids and Their Effects on Human Health**. *JPEN J Parenter Enteral Nutr* (2015.0) **39** 18S-32S. DOI: 10.1177/0148607115595980
2. 2.Puebla C, Morselli E, Khan NA, Retamal MA. Editorial: Free Fatty Acids as Signaling Molecules: Role of Free Fatty Acid Receptors and CD36. Front. Physiol. 2022;13. 10.3389/fphys.2022.862458.
3. Sawzdargo M, George SR, Nguyen T, Xu S, Kolakowski LF, O’Dowd BF. **A Cluster of Four Novel Human G Protein-Coupled Receptor Genes Occurring in Close Proximity to CD22 Gene on Chromosome 19q131**. *Biochem Biophys Res Commun.* (1997.0) **239** 543-547. DOI: 10.1006/bbrc.1997.7513
4. Hirasawa A. **Free Fatty Acid Receptor Family: A New Therapeutic Target for Metabolic Diseases**. *Yakugaku Zasshi* (2015.0) **135** 769-777. DOI: 10.1248/yakushi.14-00250-2
5. Miyamoto J, Hasegawa S, Kasubuchi M, Ichimura A, Nakajima A, Kimura I. **Nutritional Signaling via Free Fatty Acid Receptors**. *Int J Mol Sci* (2016.0) **17** 450. DOI: 10.3390/ijms17040450
6. Kimura I, Ichimura A, Ohue-Kitano R, Igarashi M. **Free Fatty Acid Receptors in Health and Disease**. *Physiol Rev* (2020.0) **100** 171-210. DOI: 10.1152/physrev.00041.2018
7. Cartoni C, Yasumatsu K, Ohkuri T, Shigemura N, Yoshida R, Godinot N, le Coutre J, Ninomiya Y, Damak S. **Taste Preference for Fatty Acids Is Mediated by GPR40 and GPR120**. *J Neurosci Off J Soc Neurosci* (2010.0) **30** 8376-8382. DOI: 10.1523/JNEUROSCI.0496-10.2010
8. Galindo MM, Voigt N, Stein J, van Lengerich J, Raguse J-D, Hofmann T, Meyerhof W, Behrens M. **G Protein-Coupled Receptors in Human Fat Taste Perception**. *Chem Senses* (2012.0) **37** 123-139. DOI: 10.1093/chemse/bjr069
9. Gatlin DM, Barrows FT, Brown P, Dabrowski K, Gaylord TG, Hardy RW, Herman E, Hu G, Krogdahl Å, Nelson R. **Expanding the Utilization of Sustainable Plant Products in Aquafeeds: A Review**. *Aquac Res* (2007.0) **38** 551-579. DOI: 10.1111/j.1365-2109.2007.01704.x
10. Lazzarotto V, Médale F, Larroquet L, Corraze G. **Long-Term Dietary Replacement of Fishmeal and Fish Oil in Diets for Rainbow Trout (Oncorhynchus Mykiss): Effects on Growth, Whole Body Fatty Acids and Intestinal and Hepatic Gene Expression**. *PloS One.* (2018.0) **13** e0190730. DOI: 10.1371/journal.pone.0190730
11. Leaver MJ, Villeneuve LA, Obach A, Jensen L, Bron JE, Tocher DR, Taggart JB. **Functional Genomics Reveals Increases in Cholesterol Biosynthetic Genes and Highly Unsaturated Fatty Acid Biosynthesis after Dietary Substitution of Fish Oil with Vegetable Oils in Atlantic Salmon (Salmo Salar)**. *BMC Genomics* (2008.0) **9** 299. DOI: 10.1186/1471-2164-9-299
12. Torstensen BE, Espe M, Sanden M, Stubhaug I, Waagbø R, Hemre G-I, Fontanillas R, Nordgarden U, Hevrøy EM, Olsvik P. **Novel Production of Atlantic Salmon (Salmo Salar) Protein Based on Combined Replacement of Fish Meal and Fish Oil with Plant Meal and Vegetable Oil Blends**. *Aquaculture* (2008.0) **285** 193-200. DOI: 10.1016/j.aquaculture.2008.08.025
13. Izquierdo MS. **Essential Fatty Acid Requirements of Cultured Marine Fish Larvae**. *Aquac Nutr* (1996.0) **2** 183-191. DOI: 10.1111/j.1365-2095.1996.tb00058.x
14. Benítez-Santana T, Atalah E, Betancor MB, Caballero MJ, Hernández-Cruz CM, Izquierdo M. **DHA but Not EPA, Enhances Sound Induced Escape Behavior and Mauthner Cells Activity in Sparus Aurata**. *Physiol Behav* (2014.0) **124** 65-71. DOI: 10.1016/j.physbeh.2013.10.021
15. 15.Roy J, Baranek E, Mercier Y, Larroquet L, Surget A, Ganot A, Sandres F, Lanuque A, Terrier F, Briand L. Involvement of Taste Receptors in the Oro-Sensory Perception of Nutrients in Rainbow Trout (Oncorhynchus Mikyss) Fed Diets with Different Fatty Acid Profiles. Aquac Nutr. 2022;2022:20. Article ID 1152463. 10.1155/2022/1152463.
16. Heraud C, Hirschinger T, Baranek E, Larroquet L, Surget A, Sandres F, Lanuque A, Terrier F, Roy J. **Detection and Modulation of Olfactory Sensing Receptors in Carnivorous Rainbow Trout (Oncorhynchus Mykiss) Fed from First Feeding with Plant-Based Diet**. *Int J Mol Sci* (2022.0) **23** 2123. DOI: 10.3390/ijms23042123
17. Roy J, Mercier Y, Tonnet L, Burel C, Lanuque A, Surget A, Larroquet L, Corraze G, Terrier F, Panserat S. **Rainbow Trout Prefer Diets Rich in Omega-3 Long Chain Polyunsaturated Fatty Acids DHA and EPA**. *Physiol Behav.* (2020.0) **213** 112692. DOI: 10.1016/j.physbeh.2019.112692
18. Roy J, Larroquet L, Surget A, Lanuque A, Sandres F, Terrier F, Corraze G, Chung-Yung Lee J, Skiba-Cassy S. **Impact on Cerebral Function in Rainbow Trout Fed with Plant Based Omega-3 Long Chain Polyunsaturated Fatty Acids Enriched with DHA and EPA**. *Fish Shellfish Immunol* (2020.0) **103** 409-420. DOI: 10.1016/j.fsi.2020.05.044
19. Roy J, Vigor C, Vercauteren J, Reversat G, Zhou B, Surget A, Larroquet L, Lanuque A, Sandres F, Terrier F. **Characterization and Modulation of Brain Lipids Content of Rainbow Trout Fed with 100% Plant Based Diet Rich in Omega-3 Long Chain Polyunsaturated Fatty Acids DHA and EPA**. *Biochimie* (2020.0) **178** 137-147. DOI: 10.1016/j.biochi.2020.06.010
20. Berthelot C, Brunet F, Chalopin D, Juanchich A, Bernard M, Noël B, Bento P, Da Silva C, Labadie K, Alberti A. **The Rainbow Trout Genome Provides Novel Insights into Evolution after Whole-Genome Duplication in Vertebrates**. *Nat Commun* (2014.0) **5** 3657. DOI: 10.1038/ncomms4657
21. 21.The National Research Council [NRC]Nutrient Requirements of Fish and Shrimp2011Washington, DCThe National Academies Press. *Nutrient Requirements of Fish and Shrimp* (2011.0)
22. Marandel L, Véron V, Surget A, Plagnes-Juan É, Panserat S. **Glucose Metabolism Ontogenesis in Rainbow Trout (Oncorhynchus Mykiss) in the Light of the Recently Sequenced Genome: New Tools for Intermediary Metabolism Programming**. *J Exp Biol* (2016.0) **219** 734-743. DOI: 10.1242/jeb.134304
23. Vernier JM. **Table Chronologique Du Développement Embryonnaire de La Truite Arc-En-Ciel**. *Salmo Gardneri Rich Ann Embryol Morphogen* (1969.0) **4** 495-520
24. Saitou N, Nei M. **The Neighbor-Joining Method: A New Method for Reconstructing Phylogenetic Trees**. *Mol Biol Evol* (1987.0) **4** 406-425. DOI: 10.1093/oxfordjournals.molbev.a040454
25. Felsenstein J. **Confidence limits on phylogenies: an approach using the bootstrap**. *Evol Int J Org Evol* (1985.0) **39** 783-791. DOI: 10.1111/j.1558-5646.1985.tb00420.x
26. 26.Zuckerkandl E, Pauling L. Evolutionary divergence and convergence in proteins. 97–166.; Evolving Genes and Proteins, Academic Press, New York, 1965;
27. Tamura K, Stecher G, Kumar S. **MEGA11: Molecular Evolutionary Genetics Analysis Version 11**. *Mol Biol Evol* (2021.0) **38** 3022-3027. DOI: 10.1093/molbev/msab120
28. Desvignes T, Fauvel C, Bobe J. **The Nme Gene Family in Zebrafish Oogenesis and Early Development**. *Naunyn Schmiedebergs Arch Pharmacol* (2011.0) **384** 439-449. DOI: 10.1007/s00210-011-0619-9
29. Pfaffl MW. **A New Mathematical Model for Relative Quantification in Real-Time RT-PCR**. *Nucleic Acids Res.* (2001.0) **29** e45. DOI: 10.1093/nar/29.9.e45
30. Milligan G, Stoddart LA, Brown AJ. **G Protein-Coupled Receptors for Free Fatty Acids**. *Cell Signal* (2006.0) **18** 1360-1365. DOI: 10.1016/j.cellsig.2006.03.011
31. Langham RJ, Walsh J, Dunn M, Ko C, Goff SA, Freeling M. **Genomic Duplication, Fractionation and the Origin of Regulatory Novelty**. *Genetics* (2004.0) **166** 935-945. DOI: 10.1093/genetics/166.2.935
32. Zambonino Infante JL, Cahu CL. **Ontogeny of the Gastrointestinal Tract of Marine Fish Larvae**. *Comp Biochem Physiol Toxicol Pharmacol CBP* (2001.0) **130** 477-487. DOI: 10.1016/s1532-0456(01)00274-5
33. Force A, Lynch M, Pickett FB, Amores A, Yan YL, Postlethwait J. **Preservation of Duplicate Genes by Complementary**. *Degenerative Mutations Genetics* (1999.0) **151** 1531-1545. DOI: 10.1093/genetics/151.4.1531
34. Yonezawa T, Kurata R, Yoshida K, Murayama MA, Cui X, Hasegawa A. **Free Fatty Acids-Sensing G Protein-Coupled Receptors in Drug Targeting and Therapeutics**. *Curr Med Chem* (2013.0) **20** 3855-3871. DOI: 10.2174/09298673113209990168
35. Hara T, Kashihara D, Ichimura A, Kimura I, Tsujimoto G, Hirasawa A. **Role of Free Fatty Acid Receptors in the Regulation of Energy Metabolism**. *Biochim Biophys Acta* (2014.0) **1841** 1292-1300. DOI: 10.1016/j.bbalip.2014.06.002
36. Kimura I, Inoue D, Maeda T, Hara T, Ichimura A, Miyauchi S, Kobayashi M, Hirasawa A, Tsujimoto G. **Short-Chain Fatty Acids and Ketones Directly Regulate Sympathetic Nervous System via G Protein-Coupled Receptor 41 (GPR41)**. *Proc Natl Acad Sci U S A* (2011.0) **108** 8030-8035. DOI: 10.1073/pnas.1016088108
37. Ichimura A, Hasegawa S, Kasubuchi M, Kimura I. **Free Fatty Acid Receptors as Therapeutic Targets for the Treatment of Diabetes**. *Front Pharmacol* (2014.0) **5** 236. DOI: 10.3389/fphar.2014.00236
38. Rodríguez AM, Sánchez J, Tobaruela A, Priego T, Picó C, Palou A. **Time-Course Effects of Increased Fatty Acid Supply on the Expression of Genes Involved in Lipid/Glucose Metabolism in Muscle Cells**. *Cell Physiol Biochem Int J Exp Cell Physiol Biochem Pharmacol* (2010.0) **25** 337-346. DOI: 10.1159/000276566
39. Mobraten K, Haug TM, Kleiveland CR, Lea T. **Omega-3 and Omega-6 PUFAs Induce the Same GPR120-Mediated Signalling Events, but with Different Kinetics and Intensity in Caco-2 Cells**. *Lipids Health Dis* (2013.0) **12** 101. DOI: 10.1186/1476-511X-12-101
40. Sona C, Kumar A, Dogra S, Kumar BA, Umrao D, Yadav PN. **Docosahexaenoic Acid Modulates Brain-Derived Neurotrophic Factor via GPR40 in the Brain and Alleviates Diabesity-Associated Learning and Memory Deficits in Mice**. *Neurobiol Dis* (2018.0) **118** 94-107. DOI: 10.1016/j.nbd.2018.07.002
41. 41.Hichami A, Khan AS, Khan NA. Cellular and Molecular Mechanisms of Fat Taste Perception. In The Pharmacology of Taste; Palmer RK, Servant G. Eds.; Handbook of Experimental Pharmacology; Springer International Publishing: Cham, 2021; Vol. 275, pp. 247–270 ISBN 978–3–031–06449–4.
42. Sleeth ML, Thompson EL, Ford HE, Zac-Varghese SEK, Frost G. **Free Fatty Acid Receptor 2 and Nutrient Sensing: A Proposed Role for Fibre, Fermentable Carbohydrates and Short-Chain Fatty Acids in Appetite Regulation**. *Nutr Res Rev* (2010.0) **23** 135-145. DOI: 10.1017/S0954422410000089
43. Sclafani A, Zukerman S, Ackroff K. **GPR40 and GPR120 Fatty Acid Sensors Are Critical for Postoral but Not Oral Mediation of Fat Preferences in the Mouse**. *Am J Physiol-Regul Integr Comp Physiol* (2013.0) **305** R1490-R1497. DOI: 10.1152/ajpregu.00440.2013
44. Ichimura A, Hirasawa A, Poulain-Godefroy O, Bonnefond A, Hara T, Yengo L, Kimura I, Leloire A, Liu N, Iida K. **Dysfunction of Lipid Sensor GPR120 Leads to Obesity in Both Mouse and Human**. *Nature* (2012.0) **483** 350-354. DOI: 10.1038/nature10798
45. Ozdener MH, Subramaniam S, Sundaresan S, Sery O, Hashimoto T, Asakawa Y, Besnard P, Abumrad NA, Khan NA. **CD36- and GPR120-Mediated Ca2+ Signaling in Human Taste Bud Cells Mediates Differential Responses to Fatty Acids and Is Altered in Obese Mice**. *Gastroenterology* (2014.0) **146** 995-1005. DOI: 10.1053/j.gastro.2014.01.006
46. Choo E, Wong L, Chau P, Bushnell J, Dando R. **Offspring of Obese Mice Display Enhanced Intake and Sensitivity for Palatable Stimuli, with Altered Expression of Taste Signaling Elements**. *Sci Rep* (2020.0) **10** 12776. DOI: 10.1038/s41598-020-68216-7
47. Ferris SD, Whitt GS. **Evolution of the Differential Regulation of Duplicate Genes after Polyploidization**. *J Mol Evol* (1979.0) **12** 267-317. DOI: 10.1007/BF01732026
|
---
title: Human umbilical cord mesenchymal stem cells conditioned medium exerts anti-tumor
effects on KGN cells in a cell density-dependent manner through activation of the
Hippo pathway
authors:
- Wenjing Wan
- Yuyang Miao
- Yuwei Niu
- Kunyuan Zhu
- Yingwan Ma
- Menghao Pan
- Baohua Ma
- Qiang Wei
journal: Stem Cell Research & Therapy
year: 2023
pmcid: PMC10029233
doi: 10.1186/s13287-023-03273-z
license: CC BY 4.0
---
# Human umbilical cord mesenchymal stem cells conditioned medium exerts anti-tumor effects on KGN cells in a cell density-dependent manner through activation of the Hippo pathway
## Abstract
### Background
The conditioned medium from human umbilical cord mesenchymal stem cells (UCMSCs-CM) provides a new cell-free therapy for tumors due to its unique secretome. However, there are many contradictory reports about the effect of UCMSCs-CM on tumor cells. The loss of contact inhibition is a common characteristic of tumor cells. A relationship between the effect of UCMSCs-CM on tumor cells and contact inhibition in tumor cells is rarely concerned. Whether the effect of UCMSCs-CM on tumor cells is affected by cell density? Here, we explored the effect of UCMSCs-CM on granulosa tumor cell line (KGN) cells at low or high density.
### Methods
Growth curve and CCK8 assay were used to assess cell proliferation and viability. Scratch wound and matrigel invasion assay were implicated to detect cell motility of KGN cells. UCMSCs-CM effects on cell cycle, apoptosis and pathway-related proteins were investigated by flow cytometry, TUNEL assay, western blot and immunofluorescence analysis respectively.
### Results
In growth curve analysis, before KGN cells proliferated into confluence, UCMSCs-CM had no effect on cell proliferation. However, once the cells proliferate to contact each other, UCMSCs-CM significantly inhibited proliferation. Meanwhile, when KGN cells were implanted at high density, UCMSCs-CM could induce cell cycle arrest at G1 phase, inhibit cell migration, invasion and promote apoptosis. While it had no similar effect on KGN cells implanted at low density. In mechanism, the UCMSCs-CM treatment activated the Hippo pathway when KGN cells were implanted at high density. Consistently, the MST$\frac{1}{2}$ inhibitor, XMU-MP-1, inhibited the activation of the Hippo pathway induced by UCMSCs-CM treatment and accordingly declined the anti-tumor effect of UCMSCs-CM on KGN cells.
### Conclusions
The effect of UCMSCs-CM on tumor cells is affected by cell density. UCMSCs-CM exerted anti-tumor effect on KGN cells by activating Hippo pathway to restore contact inhibition. Our results suggest that UCMSCs-CM is a promising therapeutic candidate for GCT treatment.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13287-023-03273-z.
## Introduction
Granulosa cell tumor (GCT) is a rare and severe sex cord stromal ovarian tumor, accounting for 5–$8\%$ of all ovarian malignancies [1]. GCTs are associated with significant risk of recurrence [2], which is often nonresponsive to conventional chemotherapies, and $80\%$ of these recurrent cases succumb to their disease [1, 3, 4]. To date, there is no standard treatment for patients with recurrent GCT [5].
Umbilical cord mesenchymal stem cells (UCMSCs) are a promising tool in cell therapies due to their multi-potency, self-renewal and immunomodulatory properties [6]. These characteristics endow UCMSCs with potential for application in the novel tumor therapies [7]. In recent years, because of unique secretome and exosomes, the conditioned medium and cell lysate of UCMSCs were used of cell-free tumor therapy [8]. UCMSCs conditioned medium (CM) contains numbers of special exosomes, growth agents, and cytokines. However, the anti-cancer property of UCMSCs-CM remains largely elusive. The human UCMSCs extracts mediated the inhibition of the proliferation of ovarian cancer cells in vitro [9, 10], UCMSCs-CM inhibited the growth of mammary carcinoma, osteosarcoma, bladder cancer, and lymphoma cells in vitro and in vivo [10–12]. Moreover, UCMSCs-CM suppresses breast cancer cells growth and sensitizes cancer cells to radiotherapy through inhibition of the Stat3 signaling pathway [13]. On the other hand, it is reported that UCMSCs-CM promoted the proliferation and migration of glioblastoma cells, MDA-MB-231 cells, and some subtypes of lung cancer stem cells [14–16]. These contradictory results are usually attributed to the difference in sensibility of various tumor cell types.
The loss of contact inhibition or density-dependent inhibition is one of the key events in oncogenesis [17, 18]. Although there have been many reports about the effect of UCMSCs-CM on tumor cells, a correlation between these effects and contact inhibition on tumor cells is rarely concerned. The Hippo pathway is highly sensitive to cell density and mediate contact inhibition of growth [19, 20]. YAP1 (yes-associated protein 1), as a transcriptional coactivator of pro-proliferation and pro-tumor genes [21, 22], is a main downstream effector of the Hippo pathway [23]. It modulates many of the biological phenotypes of cancer cells, including cell proliferation, invasion and metastasis [24]. Higher YAP1 expression was significantly associated with poorer overall survival and disease-free survival in adrenocortical carcinoma (ACC), brain lower grade glioma (LGG), and pancreatic adenocarcinoma (PAAD) [25]. Notably, evidence also links the Hippo pathway to GCT with YAP being over-activated, which regulated of GCT cell proliferation and migration [26].
In the present study, we evaluated the effects of the UCMSCs-CM against the KGN cells at low and high density in vitro. Our study revealed a new mechanism that UCMSCs-CM inhibited the malignant phenotype of KGN cells related to cell density through activating Hippo pathway. The results obtained supported the utility of UCMSCs-CM as a candidate therapeutic agent for GCT.
## Isolation, culture and identification of hUCMSCs
Human umbilical cords were obtained from mothers who had given birth at Shaanxi maternal and Child Health Hospital and were used in accordance with the ethical guidelines and accepted human studies protocols at Northwest A&F University. Written informed consent forms were obtained from the healthy umbilical cord donors. HUCMSCs were prepared as previously described [27]. The surface markers including positive markers CD44, CD29, CD90, and CD105 and negative markers CD34, CD45 of hUCMSCs were analyzed by flow cytometry and CellQuest software (Becton Dickinson). The antibodies against the above surface markers were purchased from BD Biosciences, USA. The induced differentiation of hUCMSCs was conducted in osteogenic (Gibco, A1007201, USA), adipogenic (Gibco, A1007001, USA), or chondrogenic differentiation medium (Gibco, A1007101, USA) according to manufacturer's instructions. The microscopy images were acquired at 72 × 72 dpi resolution using an inverted microscope (Olympus, DP80, Japan). There was no any processing to enhance the resolution of the microscopy images.
## Cell culture
Human granulosa-like tumor cell line KGN was obtained from Fenghui Biotechnology (Hunan, China). KGN cells were cultured in DMEM/F12 medium (Hyclone, USA) supplemented with $10\%$ FBS (ExCell Bio, China), 100 IU/mL penicillin and 100 mg/mL streptomycin (Gibco, USA), and incubated at 37 °C in a $5\%$ CO2, $95\%$ humidified atmosphere. For the distinction of high and low density, we followed the previous report [20]. In short, at low-density, cells existed as single cells or small colonies (3 × 103 cells/cm2). For high-density conditions allows cells to fuse and contact with each other (1.5 × 105 cells/cm2). The microscopy images were acquired at 72 × 72 dpi resolution using an inverted microscope (Olympus, DP80, Japan). There was no any processing to enhance the resolution of the microscopy images.
## Harvest of hUCMSC conditioned medium
The hUCMSCs at passage 3–6 were cultured in 175 cm2 flask in 20 mL DMEM/F12 supplemented with $10\%$ FBS, then collected medium every 24 h until the cell density reached $90\%$. Subsequently, the conditioned medium collected was filtered through a 0.22 μm syringe filter and stored at − 80 °C until use. HUCMSCs conditioned media represented as UCMSCs-CM in this study.
## Cell growth curve analysis
For growth curve analysis, cells were seeded in 24 well plates with complete growth medium or UCMSCs-CM at a density of 5 × 103 cells per well. Three well cells from each group were counted with a hemocytometer after $0.05\%$ trypsin digestion every 24 h, and then generated a growth curve.
## Cell viability assay
Cell Counting Kit-8 (CCK-8, Abmole, USA) was used to detect the viability of KGN cells. KGN cells were seeded in 96 well plates and cultured for 24 h. The culture medium was changed to 100 μL fresh complete growth medium or UCMSCs-CM and cultured for 48 h, added 10 μL CCK8 solutions in each well and then cultured for 2 h. The optical density values were measured by microplate reader (Bio-Rad, 168–1130 iMark, USA) at 450 nm.
## Flow cytometry analysis
Flow cytometry was performed as previously described with some modifications [28, 29]. Apoptosis was detected by using an Annexin V-FITC apoptosis detection kit (BD Biosciences, USA). Cell cycle was detected by propidium iodide (PI) (50 μg/mL) staining. The results were assessed by FACS Aria flow cytometer (BD Biosciences, BD FACSAriaTMIII 03,141,313, USA).
## TUNEL assay
TUNEL assay was performed as previously described with some modifications [30]. After be treating with UCMSCs-CM for 48 h, the KGN cells were fixed with $4\%$ paraformaldehyde for 30 min at room temperature. According to the TUNEL kit (green fluorescence, C1088, Beyotime, China) manufacturer's instructions, TUNEL test solution was added to the cells devoid of light at 37 °C for 60 min, and then the nuclei were finally stained with DAPI (Sigma, USA) for 10 min. The microscopy images were acquired at 72 × 72 dpi resolution using an inverted microscope (Olympus, DP80, Japan). There was no any processing to enhance the resolution of the microscopy images.
## Scratch wound assay
The motility of KGN cells was assessed by the scratch wound assay. Cells were plated into a 6-well cell culture plate at a density of 4 × 105 cells/well in complete growth medium and grown to $100\%$ confluence. Using a 200 μL pipette tip to make the straight scratch and washed with PBS twice to remove the cell debris. At 0, 24 and 48 h after the scratch, images were captured at 72 × 72 dpi resolution using an inverted microscope (Motic, AE2000, China) at × 100 magnification after incubation at 37˚C with complete growth medium or UCMSCs-CM. There was no any processing to enhance the resolution of the microscopy images.
## Cell invasion assay
The invasion abilities of KGN cells were evaluated by the matrigel invasion assay. Briefly, KGN cells in 100 μL DMEM/F12 medium containing 10 g/L HSA(CSL Behring; Australian) were plated on each 24-well transwell filter upper chamber (8 μm pore size, Corning, USA) coated with matrigel (Corning, 356,234, USA). And the lower chamber was filled with 500 μL complete growth medium or UCMSCs-CM to stimulate cell invasion. After 16 h of incubation, cells on the upper membrane surface were removed from the bottom of the filter, fixed with $4\%$ paraformaldehyde, and stained with $5\%$ crystal violet; 5–8 random fields of each well were captured at 72 × 72 dpi resolution under an inverted microscope (Olympus, DP80, Japan) for quantification of cell invasion. There was no any processing to enhance the resolution of the microscopy images.
## Western blot analysis
Following previous description, western blot was conducted [31]. Primary antibodies specific against cyclin D1 [60,186], p27 [25,614], Bax [50,599], Bcl-2 [12,789], β-actin [20,536] (all from proteintech; USA), YAP1 (ab52771), p-YAP 1(S127) (ab76252) (all from abcam; UK), LATS1(3477 T), p-LATS1 (S909) (9157S) (all from Cell Signaling Technology; USA), all the antibodies were used at the concentration 1:1000. Secondary antibody was HRP-labeled donkey anti-mouse/rabbit IgG (H + L) (1:2000) (Proteintech, USA).
## Immunostaining
Immunofluorescence was performed as previously described with some modifications [32]. Cells were fixed in $4\%$ paraformaldehyde and permeabilized with $0.02\%$ Triton X-100 in PBS, after washing, the cells were blocked with $1\%$ FBS at 25 °C for 20 min and then incubated with the primary antibody, rabbit polyclonal anti-YAP antibody (1:200) at 4 °C for 12 h, after washing, the cells were labeled with a secondary antibody coupled with Aleax-488(1:1000) (Abcam, ab150117, UK) at room temperature for 2 h. Cells were stained with DAPI (Sigma, USA) at room temperature for 10 min. Fluorescent images were captured at 96 × 96 dpi resolution using a confocal microscope (Nikon, TiE-A1 plus, Japan). There was no any processing to enhance the resolution of the microscopy images.
## Statistical analysis
All data were expressed as the Mean ± SEM for at least three independent experiments. Statistical analysis was carried out by GraphPad Prism 8 software (San Diego, USA). Differences among groups were evaluated by the Student's t test. P-values were calculated, and statistical significance is displayed as *$P \leq 0.05$, **$P \leq 0.01.$ *** $P \leq 0.001$, ****$P \leq 0.0001$, NS: not significant, $P \leq 0.05.$
## Characterization of hUCMSCs
The morphology of the isolated hUCMSCs was a class of swirling and fibroblast-like cells under a light microscope (Fig. 1A). The isolated hUCMSCs expressed MSC-related CD surface markers, namely, CD29, CD90, CD44, and CD105. In addition, the hUCMSCs were negative for the hematopoietic markers, including CD34 and CD45 (Fig. 1B). Besides, the isolated hUCMSCs also have differentiation potential into adipocytes, chondrocytes, and osteoblasts (Fig. 1C).Fig. 1Characterization of hUCMSCs. A HUCMSCs exhibited fibroblast-like morphology. Scale bar, 500 μm and 200 μm. B Flow cytometry analysis of CD29, CD90, CD44, CD105, CD34 and CD45 expression in hUCMSCs. C Differentiation of hUCMSCs into adipocytes (top row). Scale bar, 50 μm; Osteocytes (middle row). Scale bar, 50 μm; Chondrocytes (bottom row). Scale bar, 500 μm
## UCMSCs-CM inhibited the proliferation of KGN cells
The effect of UCMSCs-CM on the growth of KGN cells was first examined by growth curve analysis. Within 6 days after the KGN cells were implanted at 5 × 103 cells/well, the cells of the control group and the UCMSCs-CM treatment group proliferated at similar rate. Noticeably, after the seventh day, at which the cells reached confluence, the cell proliferation in UCMSCs-CM treatment group slowed down significantly, while the cells of the control group continue to proliferate (Fig. 2A). To further characterize the growth inhibitory effects of UCMSCs-CM in confluent cultures of KGN cells, the cell viability of KGN cells treated with UCMSCs-CM for 48 h at low and high density were detected by CCK8 assay. The low density ensures that the contact area between cells is as small as possible, while the high density allows cells contact with each other. The result showed that the cell viability of KGN cells treated with UCMSCs-CM was decreased compared with the control group both at low and high density (Fig. 2B). In addition, whether fibroblast conditioned medium (FB-CM) exerted the same effect on high-density of KGN cells was examined, the result showed that the cell viability of KGN cells treated with UCMSCs-CM was obviously decreased compared with the control group and FB-CM group. But there was no statistical difference between the FB-CM group and the control group (Additional file 1: Fig. S1). Meanwhile, the growth and cell morphology of KGN cells was observed. There is a marked difference in the morphologic appearance of control and UCMSCs-CM treated group at high density. Cells in the control group displays a more rounded and small cell shape in dense areas, but cells in the UCMSCs-CM group rest in ellipsoid and larger structure, and cell shrinkage, membrane damage and cell death could be observed. When it at low density, compared with the control group, the density of the UCMSCs-CM group was less dense and the morphologically was altered, nevertheless, the nuclei were normal and no dead cells were observed (Fig. 2C). These data indicated that UCMSCs-CM inhibit the proliferation of KGN cells. Fig. 2Effect of UCMSCs-CM on proliferation and cell viability of KGN cells. KGN cells were implanted at high (High) and low (Low) density. A Growth curve of KGN cells treated with UCMSCs-CM for 8 days. B CCK8 assays on KGN cells treated with UCMSCs-CM for 48 h. C UCMSCs-CM-induced morphological changes in KGN cells, the black arrows represent dead cells. Scale bar, 200 μm. * $P \leq 0.05$, **$P \leq 0.01$
## UCMSCs-CM arrested the cell cycle at the G1 phase of KGN cells at high density
Flow cytometry was conducted to identify changes in the cell cycle of KGN cells treated with UCMSCs-CM for 48 h at low density and high density. The result showed that upon reaching confluence, UCMSCs-CM treated KGN cells underwent a G1 cell cycle arrest (Fig. 3A, B). In order to further explain the G1 arrest in KGN cells in response to the UCMSCs-CM treatments, some cell cycle related proteins were detected by western blot. Compared with the control group, UCMSCs-CM obviously decreased the protein level of cyclin D1 and increased the protein level of p27 in KGN cells at high density (Fig. 3C, D; full-length blots were presented in Additional file 2: Fig. S4A–C). However, there was no significant difference in the expression levels of these proteins between UCMSCs-CM treatment and no treatment when it at low density. These data indicate that UCMSCs-CM inhibits the cell cycle of KGN cells at the G1 phase at high density rather than at low density. Fig. 3Effect of UCMSCs-CM on cell cycle of KGN cells. KGN cells were implanted at high (High) and low (Low) density. A Representative flow cytometry images of KGN cells treated with UCMSCs-CM for 48 h. B The quantitative results of flow cytometry assay. C The expression of cyclin D1 and p27 in KGNs detected by western blotting. Full-length blots are presented in Additional file 2: Fig. S4. D The quantitative results showing the relative expression levels of cyclin D1and p27 proteins. β-actin was used as loading control. * $P \leq 0.05$, **$P \leq 0.01$, ****$P \leq 0.0001.$ NS No statistical difference
## UCMSCs-CM induced apoptosis of KGN cells at high density
The effects of UCMSCs-CM on apoptosis of KGN cells at low and high density were examined. The flow cytometry results presented UCMSCs-CM treatment led to an increase in the proportion of apoptotic cells of KGN cells at high density (10.37 ± $0.6351\%$ vs 3.2 ± $0.5292\%$, $P \leq 0.0001$) (Fig. 4A, C). On the other hand, it had no significant effect at low density (3.3 ± $0.3667\%$ vs 2.6 ± $0.3055\%$, $P \leq 0.05$) (Fig. 4A, C). Consistently, in the TUNEL experiment, KGN cells showed a marked increase in emitted green light fluorescence after treatment with UCMSCs-CM at high density than low density (Fig. 4B). Moreover, the levels of pro-apoptotic factors Bax was upregulated, the expression of anti-apoptotic factor Bcl-2 was downregulated and the ratio of Bax/Bcl-2 upregulated significantly in KGN cells treated with UCMSCs-CM at high density. Yet there was no apparent difference of these protein levels between UCMSCs-CM treatment and no treatment at low density (Fig. 4D–G; full-length blots were presented in Additional file 2: Fig. S5A–C). Based on these data, UCMSCs-CM could induce KGN cells apoptosis at high density while it had no obvious effect when it at low density. Fig. 4Effect of UCMSCs-CM on apoptosis of KGN cells. KGN cells were implanted at high (High) and low (Low) density. A Flow cytometric analysis of apoptosis in KGN cells treated with UCMSCs-CM for 48 h. B Representative TUNEL stain images of KGN cells subjected to UCMSCs-CM treatment. Scale bar, 200 μm. C The quantitative results of flow cytometry assay. D Expression of Bax and Bcl-2 measured by western blot. Full-length blots are presented in Additional file 2: Fig. S5. E–G The quantitative results showing the relative expression levels of Bax and Bcl-2 proteins and the ratio of Bax/Bcl-2. β-actin was used as loading control. * $P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$, ****$P \leq 0.0001.$ NS No statistical difference
## UCMSCs‑CM inhibited the migration and invasion of KGN cells
The migration of KGN cells treated with UCMSCs-CM was next assessed by performing scratch wound assays. Compared with control group, UCMSCs-CM induced a significant decrease in KGN cell migration after 24 h and 48 h, with a reduction of about $15\%$ and about $20\%$ in their migration distance (Fig. 5A, B). To test the effect of UCMSCs-CM on the capability of KGN cells to invade, invasion assays using matrigel-coated transwells was performed. The result showed that compared with control group, UCMSCs-CM treatment impaired KGN cell invasion as shown by the significant reduction of about $50\%$ in the number of invading cells at high density. On the contrary, it had no effect when it at low density (Fig. 5C, D).Fig. 5Effect of UCMSCs-CM on migration and invasion of KGN cells. KGN cells were implanted at high (High) and low (Low) density. A Scratch wound assays of KGN cells treated with UCMSCs-CM for 24 h and 48 h. Scale bar, 200 μm. B Quantification of the closed wound area for KGN cells. C Representative images of invasion assays. Scale bar, 200 μm. D The quantitative results of invading cell numbers. *** $P \leq 0.001$, ****$P \leq 0.0001.$ NS No statistical difference
## UCMSCs-CM activated the Hippo-YAP signaling in KGN cells at high density
Owing to the pivotal role the Hippo pathway plays in GCT development and contact inhibition [19, 26], whether UCMSCs-CM could affect this signaling pathway was examined. KGNs cells with or without treatment of UCMSCs-CM were used to analyze the expression and phosphorylation status of Hippo pathway proteins. Compared with the control group, UCMSCs-CM treatment increased phosphorylation levels of LAST1 (S909), and YAP (S127) in KGN cells at high density (Fig. 6A, C). However, there was no apparent difference in phosphorylation levels of these proteins between UCMSCs-CM treatment and no treatment at low density except LAST1 (Fig. 6A, C; full-length blots were presented in Additional file 2: Fig. S6A–E).To further validate UCMSCs-CM activated the Hippo signaling, KGN cells were pretreated with the XMU-MP-1(5 μM), which is a selective inhibitor of MST$\frac{1}{2}$, the core molecule of the Hippo pathway, for 3 h before exposure to UCMSCs-CM. Western blot analysis revealed that XMU-MP-1 effectively inhibited the phosphorylation of LAST1 and YAP induced by UCMSCs-CM (Fig. 6B, D; full-length blots were presented in Additional file 2: Fig. S7A–E). Since YAP nuclear localization is required for YAP cotranscriptional activity [24, 33–35], YAP localization in KGN cells treated with UCMSCs-CM at low density and high density was observed by confocal microscopy. As expected, when KGN cells were at low density, YAP was preferentially localized in the nucleus regardless of UCMSCs-CM or XMU-MP-1 treatment (Fig. 6E). Oppositely, when KGN cells were at high density, YAP had a dramatic cytoplasmic translocation after treatment with UCMSCs-CM. While when KGN cells were pretreated with XMU-MP-1, YAP still localized in the nucleus even after UCMSCs-CM treatment (Fig. 6E). These results indicated that UCMSCs-CM could activate the Hippo pathway in KGN cells when it at high density. Fig. 6UCMSCs-CM activated the Hippo signaling in KGN cells at high density. KGN cells were implanted at high (High) and low (Low) density. A Expression of p-LAST1 (S909), LAST1, p-YAP (ser127), and YAP detected by western blot analysis. Full-length blots are presented in Additional file 2: Fig. S6. B KGNs pretreated with XMU-MP-1 and cultured with UCMSCs-CM treatment for 48 h. Expression of p-LAST1 (S909), LAST1, p-YAP (ser127), and YAP detected by western blot analysis. Full-length blots are presented in Additional file 2: Fig. S7 C–D The quantitative results showing the ratio of p-LAST1/ LAST1 and p-YAP (ser127)/YAP. E Representative immunofluorescence images of KGN cells stained with anti-YAP (green), DAPI (blue). Scale bar, 50 μm. β -actin was used as loading control.*$P \leq 0.05$, **$P \leq 0.01$, ****$P \leq 0.0001.$ NS No statistical difference
## UCMSCs-CM suppressed the malignant phenotype of KGN cells depended on activating the Hippo pathway
Since UCMSCs-CM activated the Hippo pathway in KGN cells at high density, did UCMSCs-CM inhibit the malignant phenotype of KGN cells depend on the Hippo pathway? KGNs were pretreated with XMU-MP-1 before UCMSCs-CM treatment, the results showed that XMU-MP-1 suppressed UCMSCs-CM-induced decline of cell viability (Fig. 7A). Additionally, XMU-MP-1 treatment also ameliorated the changes of the expression levels of cyclin D1 and p27 induced by UCMSCs-CM (Fig. 7B–D; full-length blots were presented in Additional file 2: Fig. S8A–C). These results indicated that UCMSCs-CM suppress proliferation of KGN cells through the Hippo signaling pathway. Next, whether UCMSCs-CM-mediated promotion of apoptosis in KGN cells could be reversed by XMU-MP-1 was examined by western blot analysis, the result revealed that XMU-MP-1 treatment led to a significant reversal of the expression of apoptosis-related proteins including Bax and Bcl-2 induced by UCMSCs-CM (Fig. 7E, F; full-length blots were presented in Additional file 2: Fig. S9A–C). Furthermore, XMU-MP-1 treatment could rescue KGN cells from UCMSCs-CM-induced invasive inhibition (Fig. 7G, H). These data suggested that the inhibitive effect of UCMSCs-CM on the malignant phenotype of KGN cells was dependent on the activation of the Hippo pathway. Fig. 7UCMSCs-CM suppresses the malignant phenotype depend on activating the Hippo pathway. KGN cells pretreated with 5 μM XMU-MP-1 and cultured with UCMSCs-CM for 48 h at high (High) and low (Low) density. A The cell viability was examined by CCK-8 assay. B Expression of cyclin D1 and p27 detected by western blot. Full-length blots are presented in Additional file 2: Fig. S8. C The quantitative results showing the relative expression levels of cyclin D1. D The quantitative results showing the relative expression levels of p27. E Expression of Bax and Bcl-2 detected by western blot. Full-length blots are presented in Additional file 2: Fig. S9. F The quantitative results showing the ratio of Bax/Bcl-2. G Representative images of invasion assays. Scale bar, 200 μm. H The quantitative results of invading cell numbers. * $P \leq 0.05$, **$P \leq 0.01$, ****$P \leq 0.0001.$ NS No statistical difference
## Discussion
GCTs are characterized by ovarian enlargement, excessive estrogens secretion, long natural history and their tendency to recur years after initial diagnosis [36]. The KGN cell line, which derived from recurrent metastatic GCT, is the only one established cell line, and it is widely used to study GCT [37].
In recent years, MSCs have become the focus of tumor biotherapy research. However, the tumor-suppressing or tumor-promoting effects of MSCs are still the subject of controversy. The main discrepancies may be due to different types of MSCs from different sources [38–40]. One of the factors contributing to oncogenesis is that MSCs could transform into tumor-associated fibroblasts(TAF) which plays an important role in tumor development [41]. It has been reported that hUCMSC do not transform to TAF in breast and ovarian cancer unlike bone marrow mesenchymal stem cells (BMSC) [42]. However, the application of MSCs transplantation is still formations [43, 44]. Hence, MSCs-conditioned medium, which contains numbers soluble molecules including exosomes and cytokines secreted by MSCs could be used as cell-free tumor therapy [45].
In the current study, before KGN cells proliferated into confluence, UCMSCs-CM had no effect on cell proliferation. However, once the cells proliferate to contact with each other, UCMSCs-CM significantly inhibited proliferation (Fig. 2A). We hypothesized that UCMSCs-CM inhibited proliferation of KGN cells related to cell density, and UCMSCs-CM might play an anti-tumor role by restoring tumor cell contact inhibition. Then the CCK8 assay was conducted to detect the cell viability and the morphological were observed at the meanwhile. The results showed that UCMSCs-CM decreased the cell viability of KGN cells both at high and low density (Fig. 2B). However, the morphological changes of KGN cells were quite different between high and low density, the cells of the UCMSCs-CM treatment group at high density showed ellipsoid, larger structure, atrophy, and membrane damage (Fig. 2C), similar morphological changes have been previously reported due to the restoration of contact inhibition [46]. When it at low density, the UCMSCs-CM treated cells showed normal nucleus though the morphological changed a lot. It is suggested that the sensitivities of KGN cells with different densities to UCMSCs-CM were different.
More importantly, when KGN cells were implanted at high density, UCMSCs-CM could induce cell cycle arrest at G1 phase, promote apoptosis, and inhibit cell migration and invasion. However, UCMSCs-CM had no similar effect on KGN cells implanted at low density. The cell cycle of contact inhibited cells is often arrested in G1 phase [47], accompanying by the expression level changes of cyclin D1 and p27, which were reported to take part in mediating contact inhibition[48, 49]. In our study, UCMSCs-CM treatment down-regulated the expression of cyclin D1, up-regulated the expression of p27 in KGN cells when it implanted at high density. In addition, UCMSCs-CM induced the expression of pro-apoptotic factors Bax whereas inhibited the expression of anti-apoptotic factor Bcl-2 when KGN cells were implanted at high density. It is reported similar promotive effect on apoptosis and cycle arrest in human leukemic cell line K562 treated with hUCMSCs and their extracts [50]. Besides, hUCMSC extracts are reported to inhibit ovarian cancer cell lines OVCAR3 and SKOV3 in vitro by inducing cell cycle arrest and apoptosis [10]. Our results consistent with these reports and further confirmed that UCMSCs-CM inhibited proliferation in KGN cells at high density maybe through restoring contact inhibition.
Scratch wound and matrigel invasion assay demonstrated that UCMSCs-CM could strongly inhibit KGN cells migration and invasion (Fig. 5A–D). These data suggest that UCMSCs-CM had the potential to inhibit granulosa cell tumor cell motility which can be helpful to prevent metastasis. It is reported that suppressive effects of dental pulp stem cells and its conditioned medium on development and migration of colorectal cancer cells through MAPKinase pathway [51], decidua parietalis mesenchymal stem/stromal cells (DPMSCs) and their secretome inhibit the invasive characteristics of MDA231 cells in vitro [52]. Our results consistent with these studies using MSCs against cancer cell migration and invasion.
The Hippo pathway plays a critical role in the tumorigenesis of human cancer [53]. YAP is regulated by cell density, and its inactivation plays a role in cell contact inhibition [19]. Especially, YAP highly expressed in human GCT tissues, and overexpression of YAP significantly stimulates the proliferation and migration of the GCT cell line [26]. Whether UCMSCs-CM could affect the Hippo pathway in KGN cells was further evaluated. Strikingly, the phosphorylation levels of Hippo signaling proteins LATS1 (S909), and YAP (S127) were all significantly increased and nuclear localization of YAP protein was significantly decreased in the KGN cells after UCMSCs-CM treatment at high density. However, the phosphorylation levels of these proteins were basically unchanged at low density, which attributed to the different sensitivity of the Hippo pathway to cell density. Moreover, XMU-MP-1, a selective inhibitor of MST$\frac{1}{2}$, could reverse UCMSCs-CM -induced LATS1 and YAP phosphorylation, resulting in restoring cell viability, rescuing UCMSCs-CM -induced apoptosis, cell cycle related protein changes and invasion inhibition at high density. Collectively, these results implicated that UCMSCs-CM activated the Hippo pathway to inhibit the malignant phenotype when KGN cells reach confluence. There had been similar reports that the Hippo pathway were activated in tumor cells when it at high density [20, 54]. It is reported that ferroptosis is regulated by the cellular contact and density [55]. Moreover, the YAP/TAZ activation under low density renders cancer cells sensitivity to ferroptosis [56], which may explain why the cell viability decreased and the morphological changed in the UCMSCs-CM treated KGN cells at low density in our study, this speculation needs further investigation.
There have different opinions about which type of material in the conditioned medium plays the role, Da-Won Choi et al. reported that some cytokine including Dkk-1, Dkk-3 and IGFBP-3 in conditioned medium play an antitumor role [57], Hamidreza Aboulkheyr Es et al. reported that CCL5, which is a cytokine in Human adipose derived MSCs conditioned medium impeded invasiveness and immune-suppressive characteristics of breast cancer cells[58]. On the other hand, many research reported that extracellular vesicles in the conditioned medium show anti‑tumor effect via miRNAs [59, 60], cell density-dependent miRNA such as miR-590-5p, let -7a, miR-10b has been reported to inhibited the tumorigenesis by directly targeting target genes [20, 54, 61]. To confirm what kinds of substances in the medium play the roles, we extracted exosomes derived from umbilical cord mesenchymal stem cells and detected the cell viability, migration and YAP phosphorylation level of KGN cells, the result showed that there was no significant difference between exosomes treatment and no treatment in KGN cell viability, migration and YAP phosphorylation level both when it at low and high density (Additional file 1: Fig. S2A–E; full-length blots were presented in Additional file 2: Fig. S10A–C). Based on these results, the effect of exosomes in the UCMSCs-CM was excluded, we prefer that cytokines secreted by hUCMSCs in conditioned medium play this role, the cytokine array assay was conducted to compare the different cytokine levels in UCMSCs-CM and control medium. Among 104 cytokines evaluated, adiponectin, which has been reported to have antitumor effects, had the highest secrete levels compared to control group (Additional file 1: Fig. S3A, B; full-length blots were presented in Additional file 2: Fig. S11A, B). Interestingly, it has been reported that adiponectin can inhibit tumor progression by regulating tumor cells (including ovarian cancer cells) proliferation and inducing the apoptosis, and low adiponectin have been associated with increased risk of ovarian cancer [62–64]. However, clinical trials based on administration of a single cytokine have been conducted for the treatment of different kinds of diseases, but the results have not been encouraging[65]. Praveen Kumar L et al. supposed that MSC-CM should better orchestrate a “symphony of signals” rather than what can be effected by the ad-ministration of single cytokine [66, 67]. Screen out the cytokines or combinations of cytokines that play the roles warrants further investigation and it is a long and rigorous process.
## Conclusions
In conclusion, UCMSCs-CM could significantly inhibit cell viability, cell migration and invasion, induce cell cycle arrest at G1 phase and promote apoptosis of KGN cells at high density through restoration of contact inhibition of KGN cells by activating the Hippo pathway (Fig. 8). These findings suggest that UCMSCs-CM is a promising therapeutic candidate for GCT treatment. Fig. 8UCMSCs-CM could significantly inhibit cell viability, cell migration and invasion, induce cell cycle arrest at G1 phase and promote apoptosis of KGN cells at high density through restoration of contact inhibition of KGN cells by activating the Hippo pathway
## Supplementary Information
Additional file 1. supplementary materials and methods and supplementary figures S1–S3.Additional file 2. supplementary figures S4–S11.
## References
1. Jamieson S, Fuller PJ. **Molecular pathogenesis of granulosa cell tumors of the ovary**. *Endocr Rev* (2012.0) **33** 109-144. DOI: 10.1210/er.2011-0014
2. Miller K, McCluggage WG. **Prognostic factors in ovarian adult granulosa cell tumour**. *J Clin Pathol* (2008.0) **61** 881-884. DOI: 10.1136/jcp.2008.057604
3. Annunziata CM, Stavnes HT, Kleinberg L, Berner A, Hernandez LF, Birrer MJ, Steinberg SM, Davidson B, Kohn EC. **Nuclear factor kappa B transcription factors are coexpressed and convey a poor outcome in ovarian cancer**. *Cancer* (2010.0) **116** 3276-3284. DOI: 10.1002/cncr.25190
4. Jamieson S, Fuller PJ. **Management of granulosa cell tumour of the ovary**. *Curr Opin Oncol* (2008.0) **20** 560-564. DOI: 10.1097/CCO.0b013e328306316f
5. Bilandzic M, Chu S, Wang Y, Tan HL, Fuller PJ, Findlay JK, Stenvers KL. **Betaglycan alters NFkappaB-TGFbeta2 cross talk to reduce survival of human granulosa tumor cells**. *Mol Endocrinol* (2013.0) **27** 466-479. DOI: 10.1210/me.2012-1239
6. Mebarki M, Abadie C, Larghero J, Cras A. **Human umbilical cord-derived mesenchymal stem/stromal cells: a promising candidate for the development of advanced therapy medicinal products**. *Stem Cell Res Ther* (2021.0) **12** 152. DOI: 10.1186/s13287-021-02222-y
7. Yuan Y, Zhou C, Chen X, Tao C, Cheng H, Lu X. **Suppression of tumor cell proliferation and migration by human umbilical cord mesenchymal stem cells: a possible role for apoptosis and Wnt signaling**. *Oncol Lett* (2018.0) **15** 8536-8544. PMID: 29805590
8. Shin S, Lee J, Kwon Y, Park KS, Jeong JH, Choi SJ, Bang SI, Chang JW, Lee C. **Comparative proteomic analysis of the mesenchymal stem cells secretome from adipose bone marrow, placenta and wharton's jelly**. *Int J Mol Sci* (2021.0) **22** 845. DOI: 10.3390/ijms22020845
9. Kalamegam G, Sait KHW, Anfinan N, Kadam R, Ahmed F, Rasool M, Naseer MI, Pushparaj PN, Al-Qahtani M. **Cytokines secreted by human Wharton's jelly stem cells inhibit the proliferation of ovarian cancer (OVCAR3) cells in vitro**. *Oncol Lett* (2019.0) **17** 4521-4531. PMID: 30944641
10. Kalamegam G, Sait KHW, Ahmed F, Kadam R, Pushparaj PN, Anfinan N, Rasool M, Jamal MS, Abu-Elmagd M, Al-Qahtani M. **Human Wharton's jelly stem cell (hWJSC) extracts inhibit ovarian cancer cell lines OVCAR3 and SKOV3 in vitro by inducing cell cycle arrest and apoptosis**. *Front Oncol* (2018.0) **8** 592. DOI: 10.3389/fonc.2018.00592
11. Gauthaman K, Yee FC, Cheyyatraivendran S, Biswas A, Choolani M, Bongso A. **Human umbilical cord Wharton's jelly stem cell (hWJSC) extracts inhibit cancer cell growth in vitro**. *J Cell Biochem* (2012.0) **113** 2027-2039. DOI: 10.1002/jcb.24073
12. Lin HD, Fong CY, Biswas A, Choolani M, Bongso A. **Human Wharton's jelly stem cells, its conditioned medium and cell-free lysate inhibit the growth of human lymphoma cells**. *Stem Cell Rev Rep* (2014.0) **10** 573-586. DOI: 10.1007/s12015-014-9514-3
13. He N, Kong Y, Lei X, Liu Y, Wang J, Xu C, Wang Y, Du L, Ji K, Wang Q, Li Z, Liu Q. **MSCs inhibit tumor progression and enhance radiosensitivity of breast cancer cells by down-regulating Stat3 signaling pathway**. *Cell Death Dis* (2018.0) **9** 1026. DOI: 10.1038/s41419-018-0949-3
14. Vulcano F, Milazzo L, Ciccarelli C, Eramo A, Sette G, Mauro A, Macioce G, Martinelli A, La Torre R, Casalbore P, Hassan HJ, Giampaolo A. **Wharton's jelly mesenchymal stromal cells have contrasting effects on proliferation and phenotype of cancer stem cells from different subtypes of lung cancer**. *Exp Cell Res* (2016.0) **345** 190-198. DOI: 10.1016/j.yexcr.2016.06.003
15. Li P, Zhou H, Di G, Liu J, Liu Y, Wang Z, Sun Y, Duan H, Sun J. **Mesenchymal stem cell-conditioned medium promotes MDA-MB-231 cell migration and inhibits A549 cell migration by regulating insulin receptor and human epidermal growth factor receptor 3 phosphorylation**. *Oncol Lett* (2017.0) **13** 1581-1586. DOI: 10.3892/ol.2017.5641
16. Vieira de Castro J, Gomes ED, Granja S, Anjo SI, Baltazar F, Manadas B, Salgado AJ, Costa BM. **Impact of mesenchymal stem cells' secretome on glioblastoma pathophysiology**. *J Transl Med* (2017.0) **15** 200. DOI: 10.1186/s12967-017-1303-8
17. Hanahan D, Weinberg RA. **The hallmarks of cancer**. *Cell* (2000.0) **100** 57-70. DOI: 10.1016/S0092-8674(00)81683-9
18. Nelson PJ, Daniel TO. **Emerging targets: molecular mechanisms of cell contact-mediated growth control**. *Kidney Int* (2002.0) **61** S99-105. DOI: 10.1046/j.1523-1755.2002.0610s1099.x
19. Zhao B, Wei X, Li W, Udan RS, Yang Q, Kim J, Xie J, Ikenoue T, Yu J, Li L, Zheng P, Ye K, Chinnaiyan A, Halder G, Lai ZC, Guan KL. **Inactivation of YAP oncoprotein by the Hippo pathway is involved in cell contact inhibition and tissue growth control**. *Genes Dev* (2007.0) **21** 2747-2761. DOI: 10.1101/gad.1602907
20. Mori M, Triboulet R, Mohseni M, Schlegelmilch K, Shrestha K, Camargo FD, Gregory RI. **Hippo signaling regulates microprocessor and links cell-density-dependent miRNA biogenesis to cancer**. *Cell* (2014.0) **156** 893-906. DOI: 10.1016/j.cell.2013.12.043
21. Zhao B, Ye X, Yu J, Li L, Li W, Li S, Yu J, Lin JD, Wang CY, Chinnaiyan AM, Lai ZC, Guan KL. **TEAD mediates YAP-dependent gene induction and growth control**. *Genes Dev* (2008.0) **22** 1962-1971. DOI: 10.1101/gad.1664408
22. Zhao B, Tumaneng K, Guan KL. **The Hippo pathway in organ size control, tissue regeneration and stem cell self-renewal**. *Nat Cell Biol* (2011.0) **13** 877-883. DOI: 10.1038/ncb2303
23. Panciera T, Azzolin L, Cordenonsi M, Piccolo S. **Mechanobiology of YAP and TAZ in physiology and disease**. *Nat Rev Mol Cell Biol* (2017.0) **18** 758-770. DOI: 10.1038/nrm.2017.87
24. Yu FX, Zhao B, Guan KL. **Hippo pathway in organ size control, tissue homeostasis, and cancer**. *Cell* (2015.0) **163** 811-828. DOI: 10.1016/j.cell.2015.10.044
25. Wu B, Tang X, Ke H, Zhou Q, Zhou Z, Tang S, Ke R. **Gene regulation network of prognostic biomarker YAP1 in human cancers: an integrated bioinformatics study**. *Pathol Oncol Res* (2021.0) **27** 1609768. DOI: 10.3389/pore.2021.1609768
26. Fu D, Lv X, Hua G, He C, Dong J, Lele SM, Li DW, Zhai Q, Davis JS, Wang C. **YAP regulates cell proliferation, migration, and steroidogenesis in adult granulosa cell tumors**. *Endocr Relat Cancer* (2014.0) **21** 297-310. DOI: 10.1530/ERC-13-0339
27. Gao LR, Zhang NK, Ding QA, Chen HY, Hu X, Jiang S, Li TC, Chen Y, Wang ZG, Ye Y, Zhu ZM. **Common expression of stemness molecular markers and early cardiac transcription factors in human Wharton's jelly-derived mesenchymal stem cells and embryonic stem cells**. *Cell Transplant* (2013.0) **22** 1883-1900. DOI: 10.3727/096368912X662444
28. Wang L, Li B, Yi X, Xiao X, Zheng Q, Ma L. **Circ_SMAD4 promotes gastric carcinogenesis by activating wnt/β-catenin pathway**. *Cell Prolif* (2021.0) **54** e12981. DOI: 10.1111/cpr.12981
29. Miao Y, Wan W, Zhu K, Pan M, Zhao X, Ma B, Wei Q. **Effects of 4-vinylcyclohexene diepoxide on the cell cycle, apoptosis, and steroid hormone secretion of goat ovarian granulosa cells**. *In Vitro Cell Dev Biol Anim* (2022.0) **58** 220-231. DOI: 10.1007/s11626-022-00663-0
30. Liu H, Wang F, Zhang Y, Xing Y, Wang Q. **Exosomal microRNA-139-5p from mesenchymal stem cells accelerates trophoblast cell invasion and migration by motivation of the ERK/MMP-2 pathway via downregulation of protein tyrosine phosphatase**. *J Obstet Gynaecol Res* (2020.0) **46** 2561-2572. DOI: 10.1111/jog.14495
31. Fontana F, Marzagalli M, Raimondi M, Zuco V, Zaffaroni N, Limonta P. **δ-Tocotrienol sensitizes and re-sensitizes ovarian cancer cells to cisplatin via induction of G1 phase cell cycle arrest and ROS/MAPK-mediated apoptosis**. *Cell Prolif* (2021.0) **54** e13111. DOI: 10.1111/cpr.13111
32. Li Z, Zhang M, Zheng J, Tian Y, Zhang H, Tan Y, Li Q, Zhang J, Huang X. **Human umbilical cord mesenchymal stem cell-derived exosomes improve ovarian function and proliferation of premature ovarian insufficiency by regulating the hippo signaling pathway**. *Front Endocrinol (Lausanne)* (2021.0) **12** 711902. DOI: 10.3389/fendo.2021.711902
33. Camargo FD, Gokhale S, Johnnidis JB, Fu D, Bell GW, Jaenisch R, Brummelkamp TR. **YAP1 increases organ size and expands undifferentiated progenitor cells**. *Curr Biol* (2007.0) **17** 2054-2060. DOI: 10.1016/j.cub.2007.10.039
34. Stein C, Bardet AF, Roma G, Bergling S, Clay I, Ruchti A, Agarinis C, Schmelzle T, Bouwmeester T, Schübeler D, Bauer A. **YAP1 exerts its transcriptional control via TEAD-mediated activation of enhancers**. *PLoS Genet* (2015.0) **11** e1005465. DOI: 10.1371/journal.pgen.1005465
35. Ritchey L, Ha T, Otsuka A, Kabashima K, Wang D, Wang Y, Lowy DR, Tosato G. **DLC1 deficiency and YAP signaling drive endothelial cell contact inhibition of growth and tumorigenesis**. *Oncogene* (2019.0) **38** 7046-7059. DOI: 10.1038/s41388-019-0944-x
36. Jamieson S, Fuller PJ. **Characterization of the inhibitor of kappaB kinase (IKK) complex in granulosa cell tumors of the ovary and granulosa cell tumor-derived cell lines, Horm**. *Cancer* (2013.0) **4** 277-292
37. Nishi Y, Yanase T, Mu Y, Oba K, Ichino I, Saito M, Nomura M, Mukasa C, Okabe T, Goto K, Takayanagi R, Kashimura Y, Haji M, Nawata H. **Establishment and characterization of a steroidogenic human granulosa-like tumor cell line, KGN, that expresses functional follicle-stimulating hormone receptor**. *Endocrinology* (2001.0) **142** 437-445. DOI: 10.1210/endo.142.1.7862
38. Poggi A, Varesano S, Zocchi MR. **How to hit mesenchymal stromal cells and make the tumor microenvironment immunostimulant rather than immunosuppressive**. *Front Immunol* (2018.0) **9** 262. DOI: 10.3389/fimmu.2018.00262
39. Fath MK, Anjomrooz M, Taha SR, Zadeh MS, Sahraei M, Atbaei R, Naghibi AF, Payandeh Z, Rahmani Z, Barati G. **The therapeutic effect of exosomes from mesenchymal stem cells on colorectal cancer: toward cell-free therapy**. *Pathol Res Practice* (2022.0) **237** 154024. DOI: 10.1016/j.prp.2022.154024
40. Zhang X, Lu Y, Wu S, Zhang S, Li S, Tan J. **An overview of current research on mesenchymal stem cell-derived extracellular vesicles: a bibliometric analysis from 2009 to 2021**. *Front Bioeng Biotechnol* (2022.0) **10** 910812. DOI: 10.3389/fbioe.2022.910812
41. Spaeth EL, Dembinski JL, Sasser AK, Watson K, Klopp A, Hall B, Andreeff M, Marini F. **Mesenchymal stem cell transition to tumor-associated fibroblasts contributes to fibrovascular network expansion and tumor progression**. *PLoS ONE* (2009.0) **4** e4992. DOI: 10.1371/journal.pone.0004992
42. Subramanian A, Shu-Uin G, Kae-Siang N, Gauthaman K, Biswas A, Choolani M, Bongso A, Chui-Yee F. **Human umbilical cord Wharton's jelly mesenchymal stem cells do not transform to tumor-associated fibroblasts in the presence of breast and ovarian cancer cells unlike bone marrow mesenchymal stem cells**. *J Cell Biochem* (2012.0) **113** 1886-1895. DOI: 10.1002/jcb.24057
43. Zhao W, Phinney DG, Bonnet D, Dominici M, Krampera M. **Mesenchymal stem cell biodistribution, migration, and homing in vivo**. *Stem Cells Int* (2014.0) **2014** 292109. DOI: 10.1155/2014/292109
44. Putra A, Rosdiana I, Darlan DM, Alif I, Hayuningtyas F, Wijaya I, Aryanti R, Makarim FR, Antari AD. **Intravenous administration is the best route of mesenchymal stem cells migration in improving liver function enzyme of acute liver failure**. *Folia Med (Plovdiv)* (2020.0) **62** 52-58. DOI: 10.3897/folmed.62.e47712
45. Hardiany NS, Yo EC, Ngadiono E, Wanandi SI. **Gene expression of molecules regulating apoptotic pathways in glioblastoma multiforme treated with umbilical cord stem cell conditioned medium, malays**. *J Med Sci* (2019.0) **26** 35-45
46. Schrader J, Deuster O, Rinn B, Schulz M, Kautz A, Dodel R, Meyer B, Al-Abed Y, Balakrishnan K, Reese JP, Bacher M. **Restoration of contact inhibition in human glioblastoma cell lines after MIF knockdown**. *BMC Cancer* (2009.0) **9** 464. DOI: 10.1186/1471-2407-9-464
47. Carter SB. **Tissue homeostasis and the biological basis of cancer**. *Nature* (1968.0) **220** 970-974. DOI: 10.1038/220970a0
48. Chassot AA, Lossaint G, Turchi L, Meneguzzi G, Fisher D, Ponzio G, Dulic V. **Confluence-induced cell cycle exit involves pre-mitotic CDK inhibition by p27(Kip1) and cyclin D1 downregulation**. *Cell Cycle* (2008.0) **7** 2038-2046. DOI: 10.4161/cc.7.13.6233
49. 49.Poon RY, Toyoshima H, Hunter T. Redistribution of the CDK inhibitor p27 between different cyclin. CDK complexes in the mouse fibroblast cell cycle and in cells arrested with lovastatin or ultraviolet irradiation. Molecul Biol Cell 1995;6(9):1197-213.
50. Huwaikem MAH, Kalamegam G, Alrefaei G, Ahmed F, Kadam R, Qadah T, Sait KHW, Pushparaj PN. **Human Wharton's jelly stem cell secretions inhibit human leukemic cell line K562 in vitro by inducing cell cycle arrest and apoptosis**. *Front Cell Dev Biol* (2021.0) **9** 614988. DOI: 10.3389/fcell.2021.614988
51. Nikkhah E, Kalalinia F, Rezaee MA, Tayarani-Najaran Z. **Suppressive effects of dental pulp stem cells and its conditioned medium on development and migration of colorectal cancer cells through MAPKinase pathways**. *Iran J Basic Med Sci* (2021.0) **24** 1292. PMID: 35083017
52. Basmaeil Y, Bahattab E, Al Subayyil A, Kulayb HB, Alrodayyan M, Abumaree M, Khatlani T. **Decidua parietalis mesenchymal stem/stromal cells and their secretome diminish the oncogenic properties of MDA231 cells in vitro**. *Cells* (2021.0) **10** 3493. DOI: 10.3390/cells10123493
53. Ma S, Meng Z, Chen R, Guan KL. **The hippo pathway: biology and pathophysiology**. *Annu Rev Biochem* (2019.0) **88** 577-604. DOI: 10.1146/annurev-biochem-013118-111829
54. Ou C, Sun Z, Li X, Li X, Ren W, Qin Z, Zhang X, Yuan W, Wang J, Yu W, Zhang S, Peng Q, Yan Q, Xiong W, Li G, Ma J. **MiR-590-5p, a density-sensitive microRNA, inhibits tumorigenesis by targeting YAP1 in colorectal cancer**. *Cancer Lett* (2017.0) **399** 53-63. DOI: 10.1016/j.canlet.2017.04.011
55. Wu J, Minikes AM, Gao M, Bian H, Li Y, Stockwell BR, Chen ZN, Jiang X. **Intercellular interaction dictates cancer cell ferroptosis via NF2-YAP signalling**. *Nature* (2019.0) **572** 402-406. DOI: 10.1038/s41586-019-1426-6
56. Sun T, Chi JT. **Regulation of ferroptosis in cancer cells by YAP/TAZ and Hippo pathways: the therapeutic implications**. *Genes Dis* (2021.0) **8** 241-249. DOI: 10.1016/j.gendis.2020.05.004
57. Liu QW, Li JY, Zhang XC, Liu Y, Liu QY, Xiao L, Zhang WJ, Wu HY, Deng KY, Xin HB. **Human amniotic mesenchymal stem cells inhibit hepatocellular carcinoma in tumour-bearing mice**. *J Cell Mol Med* (2020.0) **24** 10525-10541. DOI: 10.1111/jcmm.15668
58. Aboulkheyr Es H, Bigdeli B, Zhand S, Aref AR, Thiery JP, Warkiani ME. **Mesenchymal stem cells induce PD-L1 expression through the secretion of CCL5 in breast cancer cells**. *J Cell Physiol* (2021.0) **236** 3918-3928. DOI: 10.1002/jcp.30135
59. 59.Choi DW, Cho KA, Kim J, Lee HJ, Kim YH, Park JW, Woo SY. Extracellular vesicles from tonsilderived mesenchymal stromal cells show antitumor effect via miR199a3p. Int J Molecul Med 2021;48(6):1
60. Gu ZW, He YF, Wang WJ, Tian Q, Di W. **MiR-1180 from bone marrow-derived mesenchymal stem cells induces glycolysis and chemoresistance in ovarian cancer cells by upregulating the Wnt signaling pathway**. *J Zhejiang Univ Sci B* (2019.0) **20** 219-237. DOI: 10.1631/jzus.B1800190
61. Rani S, Ryan AE, Griffin MD, Ritter T. **Mesenchymal stem cell-derived extracellular vesicles: toward cell-free therapeutic applications**. *Mol Ther* (2015.0) **23** 812-823. DOI: 10.1038/mt.2015.44
62. Perrier S, Jardé T. **Adiponectin, an anti-carcinogenic hormone? A systematic review on breast, colorectal, liver and prostate cancer**. *Curr Med Chem* (2012.0) **19** 5501-5512. DOI: 10.2174/092986712803833137
63. Pu X, Chen D. **Targeting adipokines in obesity-related tumors**. *Front Oncol* (2021.0) **11** 685923. DOI: 10.3389/fonc.2021.685923
64. Gupta RK, Dholariya SJ, Kaushik S, Gupta SK, Tripathi R, Jain SL. **Hyperinsulinemia and hypoadiponectinemia are associated with increased risk for occurrence of ovarian cancer in non-diabetic women of North Indian population, Indian**. *J Clin Biochem* (2021.0) **36** 221-227. DOI: 10.1007/s12291-020-00891-9
65. Beohar N, Rapp J, Pandya S, Losordo DW. **Rebuilding the damaged heart: the potential of cytokines and growth factors in the treatment of ischemic heart disease**. *J Am Coll Cardiol* (2010.0) **56** 1287-1297. DOI: 10.1016/j.jacc.2010.05.039
66. 66.Kumar P, Kandoi S, Misra R, Vijayalakshmi S, Rajagopal K, Verma RS. The mesenchymal stem cell secretome: a new paradigm towards cell-free therapeutic mode in regenerative medicine. Cytokine Growth Factor Rev 2019;46:1-9
67. Bartekova M, Radosinska J, Jelemensky M, Dhalla NS. **Role of cytokines and inflammation in heart function during health and disease**. *Heart Fail Rev* (2018.0) **23** 733-758. DOI: 10.1007/s10741-018-9716-x
|
---
title: Clinical and diagnostic characteristics of non-alcoholic fatty liver disease
among Egyptian children and adolescents with type1 diabetes
authors:
- Hanaa Reyad Abdallah
- Eman Refaat Youness
- Manar Maher Bedeir
- Marwa W. Abouelnaga
- Wafaa M. Ezzat
- Yasser Elhosary
- Hazem Mohamed El-Hariri
- Mona Abd Elmotaleb A. Hussein
- Heba R. Ahmed
- Rasha Eladawy
journal: Diabetology & Metabolic Syndrome
year: 2023
pmcid: PMC10029237
doi: 10.1186/s13098-023-01029-6
license: CC BY 4.0
---
# Clinical and diagnostic characteristics of non-alcoholic fatty liver disease among Egyptian children and adolescents with type1 diabetes
## Abstract
### Background
Type 1 diabetes mellitus (T1DM) patients are at an increased risk for non-alcoholic fatty liver disease (NAFLD). This study aimed to evaluate the clinical criteria associated with the diagnosis of Non-Alcoholic Fatty Liver Disease (NAFLD) among T1DM Egyptian children and adolescents.
### Methods
74 T1DM patients aged 8–18 year were enrolled in this cross sectional study. Assessments of Clinical status, anthropometric measures, lipid profile, glycated haemoglobin (HbA1c) and liver enzymes were done. Abdominal Ultrasound evaluation of hepatic steatosis was done. Accordingly, patients were divided into two groups (NAFLD and normal liver group) and compared together. Assessment of liver fibrosis using acoustic radiation force impulse elastography (ARFI) was done. Statistical analysis included; independent t-test, Chi square and Fisher’s Exact, Pearson and Spearman tests and Logistic regression models for factors associated with fatty liver were used when appropriate.
### Results
In this study; 74 patients were enrolled; 37 males ($50\%$) and 37 females with mean age 14.3 ± 3.0 year. The mean insulin dose was 1.1 ± 0.4 U/kg and mean disease duration was 6.3 ± 3.0 year. NAFLD was detected in 46 cases while 28 cases had normal liver as diagnosed by abdominal ultrasound. Cases with NAFLD had statistically significant higher BMI-Z scores, waist/hip, waist/height and sum of skin fold thicknesses compared to those with normal liver ($P \leq 0.05$). The mean value of HbA1c % was significantly higher in NAFLD group ($$P \leq 0.003$$). Total cholesterol, triglycerides and LDL serum levels were significantly elevated ($p \leq 0.05$), while the HDL level was significantly lower in NAFLD cases ($$p \leq 0.001$$). Although, serum levels of liver enzymes; ALT and AST were significantly higher among cases with NAFLD than in normal liver group ($p \leq 0.05$), their means were within normal. Using the ARFI elastography; NAFLD cases exhibited significant fibrosis (F2, 3 and 4). BMI, patient age and female gender were among risk factors for NAFLD.
### Conclusions
NAFLD represents a serious consequence in type 1 diabetic children and adolescents that deserves attention especially with poor glycemic control. NAFLD has the potential to evolve to fibrosis. This study demonstrated a very high prevalence of NAFLD in T1D children and adolescents using US which was ($62.2\%$) with the percent of liver fibrosis among the NAFLD cases (F2-F4) using ARFI elastography was $26\%$. BMI, age of patients and female gender were detected as risk factors for NAFLD.
## Background
Diabetes mellitus type 1 (T1DM), is an autoimmune disease represents a serious, long-term condition with a major impact on the lives. The International Diabetes Federation (IDF) has reported more than 1 million children and adolescents suffer type 1 diabetes [1].
Non-alcoholic fatty liver disease (NAFLD) is the most common chronic liver disease among pediatric population and adults. Chronic liver diseases comprise a varied spectrum of diseases starting from simple steatosis or NAFLD, then non-alcoholic steatohepatitis (NASH), to cirrhosis. NAFLD should be diagnosed only when other causes of hepatic affection are absent [2, 3]. Throughout the last years, NAFLD has represented an apparently medical and financial burden as a result of increase in obesity and diabetes mellitus prevalence [4]. Furthermore, increased mortality caused by liver diseases was attributed to increase in NAFLD cases. [ 5]. Putting into consideration, that type 1 diabetes is a lifelong disease with extended duration, the range of NAFLD and its long standing consequences are clinically related to type 1 diabetic cases [6]. Inadequate excretion of triglycerides from the liver by VLD lipoproteins or hyperglycemia stimulates hepatic lipogenesis leading to accumulation of fats in the liver in type 1 diabetes mellitus (T1DM) [7]. Liver biopsy is the ‘‘gold standard’’ for the diagnosis of NAFLD. Nevertheless, it exposes patient to hazards of complications as it is an invasive technique [8]. The European guidelines for managing Non-Alcoholic Fatty Liver endorsed the usage of U/S as the first-line imaging in patients with possible Non-Alcoholic Fatty Liver and Non-Alcoholic Steato-Hepatitis [9]. In adults, its sensitivity is $90\%$ and specificity is $95\%$ for moderate to severe steatosis identification, however its sensitivity is decreased if the amount of liver fat is decreased than $33\%$ [10]. In children, hepatic ultrasound can detect fat with 70–$85\%$ sensitivity and 50–$60\%$ specificity [11].
Histological characteristics have shown significantly more severe steatosis in pediatric NAFLD in comparison to adults [12]. When the disease progresses to severe fibrotic phase it is called pediatric non-alcoholic steatohepatitis (NASH) [13]. Accordingly, the necessity for precise diagnosis and staging of this disease is of great significance in people with high risk like pediatric diabetic population using non-invasive methods [14].
Pediatric patients with laboratory measures within normal levels, as well those with normal hepatic U/S or mild steatosis, could actually have had a considerable stage of hepatic fibrosis. Acoustic Radiation Force Impulse Imaging (ARFI) is a promising U/S -based technique for assessing hepatic fibrosis and stiffness with diagnostic accuracy comparable to that of Transient elastography (TE) and could be utilized as a non-invasive method to diagnose pediatric NAFLD particularly in subjects where biopsy is not a preferred technique [15, 16]. So far, there is a paucity in paediatric studies examining the association between liver disease and type 1 DM in children and adolescents.
## Aim of the study
In this study, we aim to investigate the clinical and diagnostic characteristics distinguishing NAFLD associated with type1 diabetes in a cohort of type 1 diabetic Egyptian children and adolescents; using laboratory examination, ultrasound and liver stiffness measurement; acoustic radiation force imaging [ARFI].
## Study design and setting of the study
We conducted a cross sectional observational study enrolled 74 patients with type 1 diabetes, aged 8–18 years, and duration of T1DM > 2 years who were treated and followed in the National Institute of Diabetes and Endocrine Diseases and recruited to the outpatient clinic in the Medical Research Centre of Excellence in National Research Centre during the period from 2019 to 2021. Cases were further subdivided into two groups: patients with NAFLD and patients without NAFLD.
## Participants
Patients’ inclusion criteria included; (8–18) year, confirmed diagnosis of T1DM I according to International Society for Paediatric and Adolescent Diabetes (ISPAD) guidelines [17] and duration of T1DM > than 2 years.
Exclusion criteria included; patients who had a known other system affection like central nervous system, respiratory, renal, cardiovascular and congenital diseases or other endocrine disease like thyroid disorders, Mauriac syndrome, history of chronic liver diseases including history suggestive of viral hepatitis A, B, or C infection, genetic disorders, Wilson’s disease, hemochromatosis, and autoimmune hepatitis, and history of use of drugs causing liver function abnormality (hepatotoxic drugs such as tamoxifen, amiodarone, valproate and methotrexate or prednisone).
## Assessment of participants’ characteristics
All the study cases were subjected to the following; thorough history taking laying stress on; age of onset of diabetes, diabetes duration, diabetes complications, any long term medications other than insulin, diabetic ketoacidosis comas, insulin regimen treatment including; the type and dose of insulin in units (morning and night dose) then the mean of insulin units was taken, glycemic control as indicated by both HbA1C in the last 12 months, frequency of hypoglycemia in the last month (number of times the patient had blood glucose levels ≤ 70 mg/dL, associated diseases, and hospitalization due to any cause in the last year. In addition to gastrointestinal or hepatic symptoms like abdominal pain, nausea, or vomiting, jaundice, pruritus, visceral pain and abdominal distention. Family history of type 1 or 2 Diabetes, obesity and hypertension was also taken.
Complete clinical examination was performed stressing on; signs of hepatic affection including; jaundice, spider naevi, tender liver, hepatomegaly, splenomegaly, and ascites.
Blood pressure was measured according to American Heart Association guidelines; during the patients’ visits to the outpatient clinic, with the use of mercury sphygmomanometer; three consecutive blood pressures were measured for all patients with at least 5 min intervals, in a seated position and through a standard method using an appropriate cuff and sphygmomanometer. Blood pressure (BP) measurements were compared to age-specific percentiles for BP.
Anthropometric assessment was done as follows; Height and weight were measured. The body mass index (BMI) was calculated as weight (in kilograms) divided by height (in square meters). The standard deviation scores (SDS) of BMI were calculated using the WHO ANTHRO Plus softwares [18]. The waist circumference and hip circumference were measured. Waist/Hip ratio and Waist/Height ratio were calculated. The skinfold thicknesses were measured to the nearest 1.0 mm using Holtain skin fold caliber. They included; triceps, biceps, subscapular, suprailiac and abdominal skin folds. Each measurement was taken as the mean of three consecutive measurements, using standardized equipment and following the recommendations of International Biological programs [19].
## Laboratory and biochemical investigations
HbA1c was assessed. In addition, the mean of three readings of glycosylated hemoglobin HbA1c measurements during the last year for each patient was calculated to be representative of long-term metabolic control and patient was considered with poor glycemic control if > $10\%$ regardless of age.
Fasting blood glucose was assessed using enzymatic colorimetric methods using a Hitachi auto analyzer 704 (Roche Diagnostics. Switzerland) [20].
## The diagnosis of NAFLD in our study was based on using these routine noninvasive evaluation including
Biochemical parameters which included complete lipid profile (serum total cholesterol, triglycerides, HDL, LDL) and liver enzymes; aspartate aminotransferase [AST], alanine aminotransferase [ALT] were carried out using automated clinical chemistry analyzer. HBVs Ag and HCV Ab were done using the PRECHECK Kits (USA). Serum Anti smooth muscle antibodies ASMA, Anti-nuclear antibody ANA and anti-liver and kidney microsomal antibodies LKM were measured using ELISA.Abdominal ultrasonography which is the most commonly used imaging modality because it is relatively inexpensive, widely available to detect fatty liver. A routine liver ultrasound was performed by experienced radiologist. Examinations were performed according to a standardized protocol. US evaluation of hepatic steatosis typically consisted of a qualitative visual assessment of hepatic echogenicity, measurements of the difference between the liver and kidneys in echo amplitude, evaluation of echo penetration into the deep portion of the liver, and determination of the clarity of blood vessel structures in the liver. All US was performed by one of the two radiologists involved in the study who were blinded to the blood test-results and clinical history of patients.
## Measurement of liver fibrosis
Acoustic Radiation Force Impulse elastography (ARFI) was done as follows Acoustic radiation force impulse elastography (ARFI) was performed for all subjects with a Siemens Acuson S3000 Virtual Touch ultrasound system (Siemens AG, Erlangen, Germany) with a 6CI transducer. The principle underlying ARFI elastography is that sharing of the examined tissue induces a strain in the tissues. An acoustic “push” pulse is automatically produced by the ultrasound probe and directed to the side of a region of interest (ROI), which is where the speed of the shear wave is measured. The acoustic “push” pulse generates shear waves that propagate into the tissue, perpendicular to the “push” axis. Detection waves are also generated by the transducer to measure the propagation speed of these shear waves, which increases with fibrosis severity [21]. For each patient, 10 valid ARFI measurements were performed under fasting conditions, with the patient in supine position with the right arm in maximum abduction, by the intercostal approach in the right liver lobe, 1–2 cm under the liver capsule. Minimal scanning pressure was applied, and the patient was asked to stop normal breathing for a moment to minimize breathing motion. The mean of 8–10 valid measurements was calculated and considered indicative of the severity of fibrosis.
## Statistical analysis
The collected data were coded, tabulated, and statistically analyzed using IBM SPSS statistics (Statistical Package for Social Sciences) software version 22.0, IBM Corp., Chicago, USA, 2013. Quantitative normally distributed data were described as mean ± SD (standard deviation) after testing for normality using Shapiro–Wilk test, then compared using independent t-test for normally distributed. While Pearson test was used for correlations of normally distributed data and Spearman correlation for ordinal data. Qualitative data were described as number and percentage and compared using Chi square test and Fisher’s Exact test for variables with small expected numbers. Logistic regression models were done for factors associated with fatty liver.
## Results
A total of 74 participants with type 1 diabetes mellitus were enrolled in this study where 37 ($50\%$) were males and 37 were females ($50\%$). The mean age of the studied cases was 14.3 ± 3.0 (8–18) year. The mean age of onset of diabetes was 8.0 ± 3.2 while the mean insulin dose/day was 1.1 ± 0.4 U/kg/day, and the mean disease duration was 6.3 ± 3.0 years. $94.6\%$ of our cases had history of previous DKA attacks, $64.9\%$ had symptoms of diabetic complications and $68.9\%$ had symptoms of liver affection. Demographic data, clinical characteristics of the total studied cases and comparison according to U/S diagnosis of NAFLD are shown in (Tables 1, 2)Table 1Demographic characteristics among the studied cases and comparison according to the presence of NAFLDVariablesAll cases ($$n = 74$$)NAFLDP-valuePresent ($$n = 46$$)Absent ($$n = 28$$)Age (years), mean ± SD14.3 ± 3.015.4 ± 2.312.4 ± 3.1^ < 0.001*Gender (n, %)Male37 ($50.0\%$)17 ($37.0\%$)20 ($71.4\%$)#0.004*Female37 ($50.0\%$)29 ($63.0\%$)8 ($28.6\%$)Family historyType 2 DM47 ($63.5\%$)31 ($67.4\%$)16 ($57.1\%$)#0.374Obesity33 ($44.6\%$)23 ($50.0\%$)10 ($35.7\%$)#0.231Liver disease17 ($23.0\%$)13 ($28.3\%$)4 ($14.3\%$)#0.166Hypertension30 ($40.5\%$)19 ($41.3\%$)11 ($39.3\%$)#0.864Consanguinity13 ($17.6\%$)7 ($15.2\%$)6 ($21.4\%$)§0.539BMI body mass index^Independent t-test#Chi square test§Fishers exact test*Significant ($p \leq 0.050$)Table 2DM characteristics among the studied cases and comparison according to the presence of NAFLDVariablesAll cases ($$n = 74$$)NAFLDP-valuePresent ($$n = 46$$)Absent ($$n = 28$$)Age of onset (years), mean ± SD8.0 ± 3.28.7 ± 3.06.8 ± 3.2^0.015*Duration (years), mean ± SD6.3 ± 3.06.7 ± 3.25.6 ± 2.4^0.110Insulin dose (unit/kg/day) Mean ± SD1.1 ± 0.41.1 ± 0.41.1 ± 0.3^0.520History of DKA (n, %)70 ($94.6\%$)45 ($97.8\%$)25 ($89.3\%$)§0.149DKA frequency (attack/duration), mean ± SD0.7 ± 0.60.7 ± 0.50.6 ± 0.6^0.458Complications, (n, %)Total cases with complications48 ($64.9\%$)35 ($76.1\%$)13 ($46.4\%$)#0.010*Lipodystrophy25 ($33.8\%$)20 ($43.5\%$)5 ($17.9\%$)#0.024*Joint affection25 ($33.8\%$)19 ($41.3\%$)6 ($21.4\%$)#0.080Neuropathy22 ($29.7\%$)18 ($39.1\%$)4 ($14.3\%$)#0.023*Nephropathy7 ($9.5\%$)6 ($13.0\%$)1 ($3.6\%$)§0.242Symptoms of liver affection, (n, %)Total cases with symptoms51 ($68.9\%$)37 ($80.4\%$)14 ($50.0\%$)#0.006*Abdominal pain45 ($60.8\%$)33 ($71.7\%$)12 ($42.9\%$)#0.014*Nausea30 ($40.5\%$)24 ($52.2\%$)6 ($21.4\%$)#0.009*Vomiting24 ($32.4\%$)19 ($41.3\%$)5 ($17.9\%$)#0.037*Pruritis10 ($13.5\%$)7 ($15.2\%$)3 ($10.7\%$)§0.733Jaundice0 ($0.0\%$)0 ($0.0\%$)0 ($0.0\%$)Not applicable^Independent t-test#Chi square test§Fishers exact test*Significant ($p \leq 0.050$) According to the results of abdominal ultrasound, out of the 74 diabetic children; 46 ($62.2\%$) had fatty liver (NAFLD) and the rest of them; 28 ($37.8\%$) had normal liver. According to these U/S findings cases were divided into two groups; patients with NAFLD and patients with normal liver, then they were compared together.
This comparison revealed statistically significant difference in age ($P \leq 0.001$), the NAFLD group patients were older in age than patients with normal liver (15.4 ± 2.3 vs. 12.4 ± 3.1) year. Regarding patients gender the number of females ($63\%$) in cases with NAFLD was significantly more than those ($28.6\%$) in the group without NAFLD ($p \leq 0.05$) as shown in (Table 1).
Concerning the age of onset of diabetes; it was significantly older in cases with NAFLD as compared to cases with normal liver (8.7 ± 3.0 vs. 6.8 ± 3.2) year with ($p \leq 0.05$). Moreover, cases with NAFLD had significantly more frequent diabetic complications than cases with normal liver ($76.1\%$ vs. $46.4\%$) with ($p \leq 0.05$), including; Lipodystrophy and Neuropathy ($p \leq 0.05$) for both. Additionally, cases with liver symptoms were significantly greater in NAFLD patients than those among normal liver group ($80.4\%$ vs. $50.0\%$) with ($p \leq 0.05$) especially cases with Abdominal pain ($71.7\%$ vs. $42.9\%$), Nausea ($52.2\%$ vs. $21.4\%$) and Vomiting ($41.3\%$ vs. $17.9\%$) with ($p \leq 0.05$) for all. These data are demonstrated in (Table 2).
Regarding the anthropometric characteristics; The mean BMI—Z scores ± SD of patients with fatty liver was 0.45 ± 0.77, no patients with fatty liver were wasted while 29 ($63\%$) of them had normal BMI. Among the 28 cases with normal liver U/S, 2 cases ($7.1\%$) were malnourished (BMI-Z scores ≤ -2.0) while 18 ($64.3\%$) of them had normal BMI (BMI—Z score = 0.0).
Cases with fatty liver as compared to cases with normal liver significantly had higher mean BMI-Z score, ($$p \leq 0.001$$). There was significant difference in proportions of patients according to different BMI—Z scores grades between NAFLD cases and those with normal liver ($$p \leq 0.003$$) with more frequent BMI Z-score grade = + 1.0 ($32.6\%$ vs. $7.1\%$) ($p \leq 0.05$). On the other hand, waist/ hip ratio and waist/height ratio were significantly increased in the NAFLD patients ($$p \leq 0.001$$) who also had significantly higher sum of skin fold thicknesses than the normal liver group ($p \leq 0.001$), (Table 3).Table 3Blood pressure and anthropometric measurements among the studied cases and comparison according to NAFLD presenceVariablesTotal cases ($$n = 74$$)NAFLDP-valuePresent ($$n = 46$$)Absent ($$n = 28$$)SBP (mmHg), Mean ± SD109.9 ± 8.4109.3 ± 7.4110.7 ± 9.8^0.499DBP (mmHg), Mean ± SD73.2 ± 5.173.0 ± 5.173.4 ± 5.1^0.776BMI Z-score, Mean ± SD0.16 ± 0.950.45 ± 0.77− 0.31 ± 1.04^0.001*BMI Z-score grades ≤ − 2.02 ($2.7\%$)0 ($0.0\%$)2 ($7.1\%$)#0.003*− 1.010 ($13.5\%$)2 ($4.3\%$)8 ($28.6\%$) ± 0.047 ($63.5\%$)29 ($63.0\%$)18 ($64.3\%$) + 1.017 ($23.0\%$)15 ($32.6\%$)2 ($7.1\%$) ≥ + 2.00 ($0.0\%$)0 ($0.0\%$)0 ($0.0\%$)Waist-hip ratio, Mean ± SD0.85 ± 0.060.87 ± 0.060.82 ± 0.05^0.001*Waist-height ratio, Mean ± SD0.46 ± 0.070.48 ± 0.070.43 ± 0.07^0.001*Sum of SFT (mm), Mean ± SD46.3 ± 17.753.8 ± 16.434.0 ± 12.2^ < 0.001*SBP systolic blood pressure, DBP diastolic blood pressure, BMI body mass index, SFT skinfold thickness^Independent t-test#Chi square test§Fishers Exact test*Significant ($p \leq 0.05$)
## The Laboratory findings were as follows
Considering the CBC findings; cases with fatty liver significantly had lower Hemoglobin, RBC count, Hematocrit value compared to cases with normal liver ($$p \leq 0.003$$, 0.037 and 0.005) respectively, while no significant difference between the two groups in the other blood parameters was detected ($p \leq 0.05$).
Regarding the lipid profile, the mean serum concentrations of total cholesterol, triglycerides, and LDL were statistically significantly higher in the NAFLD group compared to the other group ($p \leq 0.001$, $$p \leq 0.019$$, $$p \leq 0.001$$, respectively). Nevertheless, HDL serum levels were significantly lower in NAFLD cases ($$p \leq 0.001$$).
Comparing the glycemic control; FBG levels were significantly higher among the NAFLD group of patients ($$p \leq 0.007$$). While HbA1c % levels measured at the time of the study and the mean HbA1c % levels during the last year revealed a statistically significant difference among the two groups ($$p \leq 0.001$$ and 0.003 respectively) being higher in the NAFLD group. The number of cases with HbA1c % levels and means > $10\%$ measured at the time of the study and during the last year was significantly higher in the NAFLD group ($$p \leq 0.014$$ and 0.002) respectively.
With respect to the liver enzymes, serum levels of AST and ALT were significantly higher in NAFLD cases compared to normal liver group ($$p \leq 0.019$$ and 0.015). We considered the level of AST > 35 IU/l in males and > 31 IU/l in females abnormal, while ALT > 45 IU/l in males and > 34 IU/l in females was considered abnormal; it was detected that the number of cases with elevated AST serum levels was significantly increased in NAFLD patients compared to cases with normal liver ($19.6\%$ vs. $0.0\%$) with ($$p \leq 0.011$$) while no significant difference was found regarding the number of patients with elevated ALT ($$p \leq 0.285$$). These findings are shown in (Table 4).Table 4Laboratory findings among the studied cases and comparison according to the presence of NAFLDVariablesTotal cases ($$n = 74$$)NAFLDP-valuePresent ($$n = 46$$)Absent ($$n = 28$$)Hemoglobin13.1 ± 1.412.8 ± 1.313.7 ± 1.2^0.003*RBCs4.8 ± 0.54.7 ± 0.54.9 ± 0.4^0.037*Hematocrit38.8 ± 4.137.7 ± 4.140.5 ± 3.6^0.005*MCHC33.9 ± 0.933.9 ± 1.134.0 ± 0.5^0.627TLC12.3 ± 5.612.0 ± 6.012.9 ± 5.0^0.533Platelets354.4 ± 84.9364.7 ± 86.5337.5 ± 81.1^0.184Cholesterol148.3 ± 43.2161.7 ± 45.2126.4 ± 28.8^ < 0.001*Triglycerides117.1 ± 74.8132.9 ± 88.891.1 ± 29.6^0.019*HDL52.6 ± 8.650.2 ± 7.156.7 ± 9.4^0.001*LDL71.1 ± 31.380.7 ± 33.455.5 ± 19.5^0.001*FBG216.2 ± 75.6234.5 ± 76.8186.1 ± 64.1^0.007*HbA1c10.3 ± 2.010.9 ± 1.79.3 ± 2.0^0.001*No. of cases with HbA1c ≥ 10.040 ($54.1\%$)30 ($65.2\%$)10 ($35.7\%$)#0.014*Mean HbA1c/year10.7 ± 2.011.2 ± 1.99.9 ± 1.8^0.003*No. of cases with mean HbA1c ≥ 10.0/year50 ($67.6\%$)37 ($80.4\%$)13 ($46.4\%$)#0.002*AST25.5 ± 18.829.5 ± 22.619.0 ± 6.1^0.019*No. of cases with elevated AST9 ($12.2\%$)9 ($19.6\%$)0 ($0.0\%$)§0.011*ALT14.4 ± 8.516.2 ± 10.111.3 ± 3.2^0.015*No. of cases with elevated ALT3 ($4.1\%$)3 ($6.5\%$)0 ($0.0\%$)§0.285^Independent t-test#Chi square test§Fishers exact test*Significant ($p \leq 0.05$) In the context of evaluation of liver fibrosis; (Table 5) shows different stages of fibrosis detected by ARFI elastography in comparison to NAFLD diagnosed by hepatic U/S. The majority of type1 diabetic patients had stage 1 ($45.9\%$), a few had stage 2 ($7\%$) or 3 ($5.4\%$), and one case ($1.4\%$) had stage 4 fibrosis. The proportions of cases with liver Fibrosis detected by ARFI ranging from F2-F4 were significantly more in fatty liver cases than normal liver cases detected by U/S, while the proportion of cases with normal liver (F0) detected by ARFI was significantly higher in cases with U/S normal liver ($53.6\%$ vs. $28.3\%$).Table 5Different stages of fibrosis diagnosed by (ARFI) compared to NAFLD diagnosed by abdominal ultrasonographyFibrosis by ARFIAll cases ($$n = 74$$) (%)NAFLD by US Present ($$n = 46$$) (%)NAFLD by US Absent ($$n = 28$$) (%)P-valueF028 (37.8)13 (28.3)15 (53.6)§0.021*F134 (45.9)21 (45.7)13 (46.4)F27 (9.5)7 (15.2)0 (0.0)F34 (5.4)4 (8.7)0 (0.0)F41 (1.4)1 (2.1)0 (0.0)§Fisher’s exact test*Significant ($p \leq 0.05$) Comparison of cases regarding the presence of liver fibrosis delineated that cases with liver fibrosis had significantly higher Cholesterol, Triglycerides and ALT levels compared to cases without liver fibrosis ($p \leq 0.05$). As well as significantly had more frequent Abnormal AST ($p \leq 0.05$) as shown in (Table 6).Table 6Laboratory findings among the studied cases and comparison according to liver fibrosisVariablesFibrosisP-valuePresent ($$n = 46$$)Absent ($$n = 28$$)Hemoglobin13.0 ± 1.313.4 ± 1.4^0.193RBC4.7 ± 0.54.9 ± 0.5^0.219Hematocrit38.2 ± 4.139.7 ± 4.0^0.118MCHC34.0 ± 0.833.7 ± 1.0^0.219TLC12.4 ± 5.612.2 ± 5.6^0.840Platelets359.8 ± 83.8346.1 ± 87.5^0.504Cholesterol157.5 ± 45.8134.1 ± 35.0^0.022*Triglycerides130.8 ± 90.895.9 ± 29.3^0.020*HDL53.5 ± 9.551.3 ± 7.0^0.296LDL75.2 ± 34.964.8 ± 23.9^0.164FBG214.0 ± 79.3219.6 ± 70.7^0.762HbA1c10.2 ± 1.710.3 ± 2.4^0.881No. of cases with HbA1c ≥ 10.024 ($52.2\%$)16 ($57.1\%$)#0.877Mean HbA1c/y10.7 ± 1.810.8 ± 2.2^0.702No. of cases with mean HbA1c ≥ 10.0/y30 ($65.2\%$)20 ($71.4\%$)#0.837AST28.7 ± 23.120.6 ± 6.1^0.069No. of cases with abnormal AST9 ($19.6\%$)0 ($0.0\%$)§0.010*ALT16.1 ± 10.311.7 ± 3.2^0.028*No. of cases with abnormal ALT3 ($6.5\%$)0 ($0.0\%$)§0.275^Independent t-test#Chi square test§Fishers exact test*Significant ($p \leq 0.05$) Our results demonstrated that liver fibrosis stage was significantly positively correlated with BMI Z-score, Waist-Hip ratio, Waist-Height ratio, SFT, Cholesterol and Triglycerides levels ($p \leq 0.05$) as presented in (Table. 7).Table 7Correlations of liver fibrosis stages detected by ARFI among the studied cases with other characteristicsVariablesrpAge0.0750.527Age of onset0.1820.120Duration− 0.0290.806Insulin dose0.1280.276DKA Frequency0.0210.856SBP− 0.1510.198DBP0.0950.421BMI Z-score0.3390.003*Waist-Hip ratio0.3200.005*Waist-Height ratio0.411 < 0.001*SFT0.411 < 0.001*Hemoglobin− 0.1340.254RBC− 0.1380.240Hematocrit− 0.1770.131MCHC0.0780.510TLC− 0.0420.722Platelets0.0600.611Cholesterol0.2510.031*Triglycerides0.2440.036*HDL0.1270.280LDL0.1450.218FBG− 0.0230.846HbA1c0.0270.818Average HbA1c− 0.0140.905AST0.2220.05ALT0.2090.07Spearman correlation test*Significant ($p \leq 0.05$)
We investigated the factors associated with occurrence of Fatty liver in type 1 diabetic children and adolescents using Logistic regression models and the BMI-Z score ≥ + 1.0 and Age ≥ 15 years were significant factors that increased the risk of fatty liver occurrence ($$p \leq 0.034$$ and 0.002) with CI (1.140–26.204 and 1.657−9.598) respectively. While being a male was a significant protective factor ($$p \leq 0.020$$) as shown in (Table 8).Table 8Logistic regression models for factors associated with Fatty liver in T1DM children and adolescentsFactorsΒSEPOR ($95\%$ CI)BMI—Z score ≥ + 1.01.6990.8000.034*5.466 (1.140 − 26.204)Age ≥ 15 years1.3830.4480.002*3.988 (1.657 − 9.598)Male sex− 0.9840.4220.020*0.375 (0.164 − 0.854)β regression coefficient, SE Standard error, OR Odds ratio, CI Confidence interval, P is significant at < 0.05
## Discussion
The definition of NAFLD necessitates the confirmation of hepatic steatosis, whether by imaging or by histology, with absence of other reasons for secondary fat infiltration of the liver namely; excessive alcohol consumption, administration of drugs inducing steatosis or genetic diseases [22].
The ‘‘gold standard’’ for diagnosis of NAFLD is Liver biopsy. Nonetheless, it is invasive and has the possibility of complications [23]. Ultrasound is the preferred first-choice imaging method in clinical management [8]. Thus in the present study diagnosis of NAFLD was based on abdominal ultrasonographic findings.
The prevalence of NAFLD among apparently healthy young Egyptian adults aged 19–21 year was studied by Tomah et al. who concluded that fatty liver was present in $31.6\%$ of them [24].
The major finding of our study was the increased prevalence of NAFLD in children with type 1 diabetes as among the 74 patients; 46 ($62.2\%$) cases had NAFLD as diagnosed by abdominal ultra sound (US).
Whereas, Al-Hussaini et al. detected hepatic affection in $10\%$ of 106 children with type 1 diabetes in an Indian study [25], and El-Karaksy et al. ( a larger study of 692 Egyptian children with type 1 diabetes) declared a prevalence of $4.5\%$ of liver affection [26]. On the other hand, Farhan et al., reported abnormal hepatic findings in ($26\%$) of children with type 1 diabetes [7] and ElBaki et al. in their study detected $37.3\%$ of cases with NAFLD [27]. The high prevalence of hepatic affection in our study could be attributed to poorer glycemic control of the included patients.
Regarding NAFLD patients, they were significantly older in age than patients with normal liver ($$p \leq 0.015$$). This disagrees with the study of Barros et al. [ 28].
The number of females ($63\%$) in the NAFLD group was significantly more than those ($28.6\%$) in the group with normal hepatic US. On the other hand the percent of females ($63\%$) with NAFLD was more than males ($37\%$) in the same group. While in Farhan et al., study; patients with fatty liver ($69.2\%$) of them were females and in El-Karaksy et al. the female to male ratio was equal. Whereas in Barros et al., study, the female gender represented $75\%$ of the NAFLD group vs. $55.4\%$ in the normal liver group but this was not statistically significant [7, 26, 28].
Moreover, coherent with our results, Samuelsson et al., in their large population study; exihibited a sex difference; girls had poorer metabolic control, i.e., elevated HbA1c concentrations [29]. This could be because girls have poorer metabolic control over the period of adolescence than boys. The variations in hormones might be the influencers among the two genders during this period. Various researches have confirmed that both insulin doses and HbA1c levels were significantly elevated in girls [30, 31].
In examination of children, body mass index (BMI) is one of the most frequently used indicators in evaluation of obesity and undernourishment [32]. The waist-hip ratio (WHR) permits defining the kind of body contour and site of fat accumulation. Waist to height ratio (WHtR) is used to evaluate the dissemination of abdominal adiposity. In pediatric population with abdominal obesity and an amplified risk of metabolic syndrome, the index value is > 0.5 irrespective of sex [33].
Evaluation of anthropometric measures in diabetic children and adolescents must be done regularly. Approaches for nutritional evaluation are harmless and non-invasive, and the study outcomes may be used by clinicians in individuals with diabetes, assisting in monitoring their metabolic control, that affects the appropriate physical growth of children. Therefore, adjusting the nutritional state is of great importance as a whole, and not only stature and weight individually [31].
In this aspect, the anthropometric assessment in the present study revealed that, cases with NAFLD had higher BMI Z scores, waist/hip, waist/ height ratios and sum of skinfold thicknesses than those with normal liver and the difference was statistically significant ($p \leq 0.05$). This is in contrast to the findings of Barros et al. [ 28].
Moreover, our results revealed that cases with NAFLD had more frequent BMI Z-score grade = + 1.0 ($32.6\%$ vs. $7.1\%$). While, $63\%$ of NAFLD cases had normal mean BMI (Z-score = 0) vs. $64.3\%$ of cases with normal liver. None of our patients had obesity (BMI Z score ≥ + 2.0). On the other hand, in the NAFLD group no cases with malnutrition were detected, whereas, 2 cases ($7.1\%$) with normal liver had malnutrition (BMI Z score ≤ − 2.0).
In contrast to our results, Farhan et al. in their study reported; ($61.5\%$) of children with fatty liver had malnutrition while ($38.5\%$) of those children were normal in BMI, in children having normal hepatic findings, ($67.6\%$) were undernourished though ($32.4\%$) had normal BMI [7].
Regarding the lipid profile, our study revealed that patients with NAFLD showed significantly more serum lipid levels (cholesterol, triglyceride and LDL) ($$p \leq 0.001$$, 0.019 and 0.001) respectively, while there was significant decrease in HDL ($$p \leq 0.001$$). This agrees with the findings of previous studies [7, 34, 35].
In acceptance with our results, the findings of Barros et al. showed that cases with altered hepatic US findings had significant elevated triglycerides values and lesser HDL than normal liver patients ($$p \leq 0.028$$ and 0.034) respectively. However, Barros et al. in their study; found that there was no significant difference regarding total cholesterol and LDL levels between cases with Abnormal hepatic US findings and normal liver group [28]. This is opposite to our results.
Oscillations in blood glucose and insulin levels are significant factors in hepatic steatosis associated with type1 diabetes. HbA1c is a good indicator of metabolic control [36]. Proper metabolic control is necessary not only for adequate growth and development in diabetic children and adolescents, but also for reduction and delay of advancement of present complications [37].
Considering mean HbA1c % levels, the mean level of HbA1C of NAFLD patients enrolled in this study was ($11.2\%$ ± 1.9) while for those with normal ultrasound findings it was ($9.9\%$ ± 1.8) with significant difference ($$P \leq 0.003$$) being higher in the NAFLD group. The number of patients with poor glycemic control as indicated by HbA1c% > 10 was also significantly higher in the NAFLD group.
Parallel to this is the study of Farhan et al. who reported mean value of HbA1C of diabetic patients with fatty liver (10.69 ± 1.41) whereas in normal U/S cases it was (8.24 ± 2.04) with significant difference ($$P \leq 0.021$$) [7]. However, in El-Karaksy et al. study, the mean value of HbA1C in T1D children having liver affection was (8.1 ± 1.2) while for patients with normal liver was (7.6 ± 1.7) with insignificant difference ($$P \leq 0.05$$) [26].
Additionally, in Ismail et al. study, a statistically significant difference in HbA1c % mean levels, among the studied groups ($p \leq 0.001$) being higher in the NAFLD group with mean level = $8.41\%$ ± 0.8 [35]. This shows that NAFLD children in our study had more improper glycemic control than the previous studies.
Even though assessment of ALT serum level is usually done as a measure of liver functions, its significance is debatable [38]. The dependence on normal liver enzymes is one of the main causes for missing the diagnosis of NAFLD by general practitioners and diabetologists [39].
In this regards, the current study revealed that AST and ALT were statistically significantly higher in patients with NAFLD ($$P \leq 0.019$$ and 0.015). AST > 35 IU/l was considered abnormal in boys and > 31 IU/l in girls; 9 patients only had high level ($12.2\%$ of the total cases and they represented only $19.6\%$ of the NAFLD cases). On the other hand ALT > 45 IU/l was considered abnormal in boys and > 34 IU/l in girls; only 3 patients had increased level ($4.1\%$ of the total number of cases and they represented only $6.5\%$ of the NAFLD cases) this agrees with the opinion that normal liver enzymes do not exclude fatty liver [39], and further suggests that serum liver enzymes are good indicators for NAFLD diagnosis, however, ‘‘normal’’ standard levels used for exclusion of NAFLD are needed to be reviewed.
Our results agrees with Farhan et al. regarding AST levels, meanwhile, they disagree with them concerning ALT levels [7]. However, Ismail et al. in their study found that the mean serum ALT level of the NAFLD cases was at a high normal level and only three female patients had mildly elevated ALT level [35]. Moreover, our results are contradicting to the results of Barros et al. study, [28] and Singh et al. [ 10]. In addition, several researches demonstrated that the complete histological picture of NAFLD may be detected in patients with normal ALT levels [41, 42].
Diabetic individuals with NAFLD are highly prone to progress into more severe stage of NAFLD which can lead to liver cirrhosis and finally liver failure [8]. Diabetes had been detected as an independent risk factor for hepatic fibrosis [42].
Ultrasound can assess increase in liver size or diffuse increase in hepatic parenchyma echogenicity however it can’t detect fibrosis. ARFI elastography has the privilege that it is not an invasive technique to evaluate liver fibrosis [35].
ElBaky et al. concluded that it is essential to do abdominal ultrasound in type I diabetic pediatric patients as an early non-invasive evaluation of liver affection, whereas ARFI is required in more progressive stages [27]. Moreover, Farhan et al., concluded that NAFLD is significant as an early alarming sign of future result of diabetes mellitus in the form of progression to hepatic fibrosis, cirrhosis and failure [7] Another major finding in our study is different stages of liver fibrosis diagnosed by ARFI Fibro-Scan compared to abdominal ultrasonography findings. Most of our studied type1 diabetic children and adolescents were having grade one fibrosis ($$n = 34$$), a few were with grade 2 ($$n = 7$$) or 3 ($$n = 4$$), and only 1 patient was having grade 4 fibrosis while 28 patients had F0 stage i.e. no fibrosis.
There was a significant difference between proportions in liver affection using ARFI and those diagnosed by US. From the NAFLD free cases diagnosed by US there was liver stiffness (fibrosis) stage F1 in 13 cases ($46.4\%$) indicating that ARFI can detect liver affection more accurately than US.
Moreover, the number of cases with liver Fibrosis detected by ARFI ranging from F2-F4 was significantly higher among cases with fatty liver compared to cases without fatty liver detected by US ($$p \leq 0.021$$). There was a significant difference between proportions in liver affection using ARFI and those diagnosed by US. $53.6\%$ of the NAFLD free cases diagnosed by US had no fibrosis (F0), while no cases with F2, 3 and 4 were detected in this group. Meanwhile, in NAFLD cases, $28.3\%$ had no fibrosis (F0) and only $26\%$ had intermediate and severe form of fibrosis (F2, F3 and F4).
Ismail et al. in their study; stated that, ARFI elastography classification of fibrosis exhibited that 4 children ($8.0\%$) were having liver fibrosis stage 3 and 4 [35]. While in our study; $10.8\%$ of NAFLD cases had stage 3 and 4 liver fibrosis. Moreover, ElBaki et al. reported ARFI results in $7.7\%$ of patients with different stages of fibrosis [27].
Our results disagree with the results of Farhan et al. [ 7]. However their results were obtained by calculating the NAFLD fibrosis score and not by fibroscan. Furthermore, a research done by Singh et al. examining 4899 T1D patients aged 18–80 year with suspected NAFLD; showed a prevalence of advanced fibrosis of $22.1\%$ using AST/ALT > 1.4, demonstrating increased risk to develop progressive hepatic affection and its associated complications [10].
On the other hand, previous paediatric studies have shown ARFI to be an accepted non-invasive method. Hanquinet et al. compared ARFI values in children with biopsy-proven chronic liver disease and normal subjects and its value to differentiate between mild and severe (F > 2) fibrosis [43]. In contrast, Kummur et al. did not show any significantly increased prevalence of NAFLD in a paediatric cohort using ALT, ultrasound, and liver stiffness measures [44].
In addition, the findings of Barros et al. showed that of the total participants ($8.4\%$) had significant fibrosis (> F2). Whereas, F2 cases were $3.1\%$, F3 was present in $3.1\%$ and F4 was present in $2.1\%$ as detected by transient elastography (TE) [28].
Tuong and Duc reported $100\%$ successful rate of ARFI in their study and SWV had significant correlation with degree of liver fibrosis ($p \leq 0.05$). They concluded that ARFI was significantly better than APRI in evaluating the degree of liver fibrosis [45].
The current study revealed that Cases with liver fibrosis had significantly higher Cholesterol, Triglycerides and ALT serum levels compared to cases without liver fibrosis ($p \leq 0.05$). Cases with Abnormal AST serum level were significantly more frequent among the liver fibrosis group ($p \leq 0.05$).
Compatible with our study, Carter-Kent et al. mentioned that, the serum AST level correlated with the stage of fibrosis and might be used to differentiate significant from no or mild fibrosis [46]. In agreement with our findings also Farhan et al. reported that, AST serum level was found a good predictor of fibrosis in pediatric patients (P-value = 0.001) [7].
We analyzed factors correlated with the stage of liver fibrosis in the present study and we detected significant positive correlations of *Liver fibrosis* stages with BMI Z-score, Waist-Hip, Waist-Height, SFT, Cholesterol, Triglycerides ($p \leq 0.05$).
Multivariate logistic regression was applied to assess risk factors associated with occurrence of NAFLD; BMI-Z score ≥ + 1.0 and Age ≥ 15 years were detected as significant factors associated with increased risk of fatty liver development. While being a male was a significant protective factor.
In Barros et al. study; multivariable logistic regression assessing associated factors with fatty liver using both imaging technique; Gender, age and HbA1c were not associated to steatosis. This is contradicting to our results where we found gender and age as associated risk factors for NAFLD occurrence. However, Barros et al. study revealed that triglycerides were the only risk factor for fatty liver [28]. This could not be detected in our study.
However, Sae-wong et al. found that high BMI-SDS were the only risk factor associated with NAFLD (OR, 5.79) [47]. This coincides partly with our results.
## Limitations of the study
Our study has some limitations. First, we diagnosed NAFLD based on ultrasound which is operator-dependent and has a limited sensitivity. Second, ARFI fibroscan; the second technique we used, is not commonly used as first- choice investigation for diagnosis of liver fibrosis. Fibro—test consists of a panel of markers for diagnosis of liver fibrosis but unfortunately we couldn’t do it as we didn’t have enough fund for these tests which would cost too much so we didn’t perform them. We have used the ARFI fibroscan to diagnose *Liver fibrosis* according to previous studies which evaluated the diagnostic performance of ARFI fibroscan in diagnosing liver fibrosis [45, 48–51]. Thirdly, we did not have histological confirmation of our findings as the gold-standard method; liver biopsy, is invasive and prone to sample mistakes. Another limitation was the small sample size of the study.
As strengths in our study; NAFLD was diagnosed using two methods in this sample of type1 diabetic patients while most of NAFLD studies in type1diabetes used ultrasonography only. In addition, to the best of our knowledge, little previous studies were performed to diagnose steatosis and assess hepatic fibrosis in type1 diabetes children thus this data are represented for Egyptian children.
## Conclusions
The primary outcome of this study demonstrated a very high prevalence of NAFLD in T1D children and adolescents using US which was ($62.2\%$) and the percent of significant and advanced liver fibrosis (F2-F4) in NAFLD cases using ARFI elastography was $26\%$. The study clarified that, $46.4\%$ of the NAFLD free cases detected by US had mild liver fibrosis (F1) as detected by ARFI.
NAFLD represents a serious consequence of type 1 Diabetes in children and adolescents. It is an early warning sign of future consequence of diabetes mellitus in the form of progression to liver fibrosis, cirrhosis and failure; especially in those with poor glycemic control. NAFLD has the potential to evolve to fibrosis. Ultrasound can detect hepatomegaly and diffuse increase in hepatic parenchyma echogenicity but not fibrosis. ARFI fibro-scan assessment of liver fibrosis in T1DM patients with NAFLD detected the presence of significant fibrosis. BMI, age of patients and female gender were among the risk factors associated with NAFLD. Further prospective studies with larger number of patients are recommended.
## References
1. 1.International Diabetes Federation. IDF Diabetes Atlas—9th Edition 2019. https://diabetesatlas.org. Accessed 26 July 2019.
2. Wang XJ, Malhi H. **Nonalcoholic fatty liver disease**. *Ann Intern Med* (2018.0) **169** 60. PMID: 29507939
3. Chalasani N, Younossi Z, Lavine JE. **The diagnosis and management of nonalcoholic fatty liver disease: practice guidance from the American association for the study of liver diseases**. *Hepatology* (2018.0) **67** 328-357. DOI: 10.1002/hep.29367
4. Estes C, Razavi H, Loomba R, Younossi Z, Sanyal AJ. **Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease**. *Hepatology* (2018.0) **67** 123-133. DOI: 10.1002/hep.29466
5. Younossi Z, Tacke F, Arrese M. **Global perspectives on nonalcoholic fatty liver disease and nonalcoholic steatohepatitis**. *Hepatology* (2019.0) **69** 2672-2682. DOI: 10.1002/hep.30251
6. De Vries M, Westerink J, Kaasjager KHAH, de Valk HW. **prevalence of nonalcoholic fatty liver disease (NAFLD) in patients with type 1 diabetes mellitus: a systematic review and meta-analysis**. *J Clin Endocrinol Metab* (2020.0) **105** 3842-3853. DOI: 10.1210/clinem/dgaa575.PMID:328274323&27
7. Farhan R, Alzubaidi MA, Ghayyib SM. **Fatty liver disease in children and adolescents with type 1 diabetes mellitus (clinical and diagnostic aspects)**. *J Clin Gastroenterol Hepatol* (2018.0) **2** 14. DOI: 10.21767/2575-7733.1000043
8. Anstee QM, Targher G, Day CP. **Progression of NAFLD to diabetes mellitus, cardiovascular disease or cirrhosis**. *Nat Rev Gastroenterol Hepatol* (2013.0) **10** 330-344. DOI: 10.1038/nrgastro.2013.41
9. Marchesini G, Day CP, Dufour JF, Canbay A, Nobili V, Ratziu V, Tilg H, Roden M, Galstadelli A, Yki-Jarvinen H. **EASL-EASD-EASO Clinical practice guidelines for the management of non-alcoholic fatty liver disease**. *J Hepatol* (2016.0) **64** 1388-1402. DOI: 10.1016/j.jhep.2015.11.004
10. Singh A, Le P, Peerzada MM, Lopez R, Alkhouri N. **The Utility of noninvasive scores in assessing the prevalence of nonalcoholic fatty liver disease and advanced fibrosis in type 2 diabetic patients**. *J Clin Gastroenterol* (2018.0) **52** 268-272. DOI: 10.1097/MCG.0000000000000905
11. Awai HI, Newton KP, Sirlin CB, Behling C, Schwimmer JB. **Evidence and recommendations for imaging liver fat in children, based on systematic review**. *Clin Gastroenterol Hepatol* (2014.0) **12** 765-773. DOI: 10.1016/j.cgh.2013.09.050
12. Takahashi Y. **Histopathology of nonalcoholic fatty liver disease/nonalcoholic steatohepatitis**. *World J Gastroenterol* (2014.0) **20** 15539. DOI: 10.3748/wjg.v20.i42.15539
13. Clemente MG, Mandato C, Poeta M, Vajro P. **Pediatric non-alcoholic fatty liver disease: recent solutions, unresolved issues, and future research directions**. *World J Gastroenterol* (2016.0) **22** 8078. DOI: 10.3748/wjg.v22.i36.8078
14. Papachristodoulou A, Kavvadas D, Karamitsos A, Papamitsou T, Chatzidimitriou M, Sioga A. **Diagnosis and staging of pediatric non-alcoholic fatty liver disease: is classical ultrasound the answer?**. *Pediatr Rep* (2021.0) **13** 312-321. DOI: 10.3390/pediatric13020039
15. Kamble R, Sodhi KS, Thapa BR, Saxena AK, Bhatia A, Dayal D, Khandelwal N. **Liver acoustic radiation force impulse (ARFI) in childhood obesity: comparison and correlation with biochemical markers**. *J Ultrasound* (2017.0) **20** 33-42. DOI: 10.1007/s40477-016-0229-y
16. Berná-Serna JD, Sánchez-Jiménez R, Velázquez-Marín F, Sainz de Baranda P, Guzmán-Aroca F, Fernández-Hernández C, Domenech-Abellan E, Abellan-Rivero D, Ruiz-Merino G, Madrid-Conesa J. **Acoustic radiation force impulse imaging for detection of liver fibrosis in overweight and obese children**. *Acta Radiol* (2018.0) **59** 247-253. DOI: 10.1177/0284185117707359
17. Craig ME, Jefferies C, Dabelea D, Balde N, Seth A, Donaghue KC. **Definition, epidemiology, and classification of diabetes in children and adolescents**. *Pediatr Diabetes* (2014.0) **15** 4-17. DOI: 10.1111/pedi.12186
18. 18.WHO AnthroPlus for personal computers manual software for assessing growth of the world's children and adolescents. Geneva. WHO. 2009. (https://www.who.int/tools/growth-reference-data-for-5to19-years).
19. Hiernaux J, Tanner JM, Weiner JS, Lourie JA. **Growth and physique: anthropometry**. *Human biology, a guide to field methods* (1969.0) 2-42
20. Hirschler V, Oestreicher K, Maccallini G, Aranda C. **Relationship between obesity and metabolic syndrome among argentinean elementary school children**. *Clin Biochem* (2010.0) **43** 435-441. DOI: 10.1016/j.clinbiochem.2009.11.003
21. Palmeri ML, Wang MH, Dahl JJ, Frinkley KD, Nightingale KR. **Quantifying hepatic shear modulus in vivo using acoustic radiation force**. *Ultrasound Med Biol* (2008.0) **34** 546-558. DOI: 10.1016/j.ultrasmedbio.2007.10.009
22. Chalasani N. **The diagnosis and management of non-alcoholic fatty liver disease: practice guideline by the American gastroenterological association, American association for the study of liver diseases, and American college of gastroenterology**. *Gastroenterology* (2012.0) **142** 1592-1609. DOI: 10.1053/j.gastro.2012.04.001
23. Al K, Shiffman B. **Percutaneous liver biopsy in clinical practice**. *Liver Int* (2007.0) **27** 1166-1173. DOI: 10.1111/j.1478-3231.2007.01592.x
24. Tomah S, Hamdy O, Abuelmagd MM, Hassan AH, Alkhouri N, Al-Badri MR, Gardner H, Eldib AH, Eid EA. **Prevalence of and risk factors for non-alcoholic fatty liver disease (NAFLD and fibrosis among young adults in Egypt**. *BMJ Open Gastroenterol* (2021.0) **8** e000780. DOI: 10.1136/bmjgast-2021-000780
25. Al-Hussaini AA, Sulaiman NM, Alzahrani MD, Alenizi AS, Khan M. **Prevalence of hepatopathy in type 1 diabetic children**. *BMC Pediatr* (2012.0). DOI: 10.1186/1471-2431-12-160]
26. El-Karaksy HM, Anwar G, Esmat G, Mansour S, Sabry M. **Prevalence of hepatic abnormalities in a cohort of Egyptian children with type 1 diabetes mellitus**. *Pediatr Diabetes* (2010.0) **11** 467. DOI: 10.1111/j.1399-5448.2009.00627.x
27. ElBaky AM, Ismail NA, Abo-Hashesh MM, ElShaer TF, Abd El Aziz SH, Rasheed IA. **Non-alcoholic fatty liver disease (NAFLD) spectrum in children with type 1 diabetes mellitus evaluated with non-invasive fibrosis score and instrument: A cross-sectional study**. *Pediatr Pol* (2021.0) **96** 223-230. DOI: 10.5114/polp.2021.112395
28. Barros BSV, Monteiro FC, Terra C, Gomes MB. **Prevalence of non-alcoholic fatty liver disease and its associated factors in individuals with type 1 diabetes: a cross-sectional study in a tertiary care center in Brazil**. *Diabetol Metab Syndr* (2021.0) **13** 33. DOI: 10.1186/s13098-021-00649-0.PMID:33741024;PMCID:PMC7977593
29. Samuelsson U, Anderze´ NJ, Gudbjo¨rnsdottir S, Steineck I, kesson K, Hanberger L. **Teenage girls with type 1 diabetes have poorer metabolic control than boys and face more complications in early adulthood**. *J Diabetes Complicat* (2016.0) **30** 917-922. DOI: 10.1016/j.jdiacomp.2016.02.007
30. Setoodeh A, Mostafavi F, Hedayat T. **Glycaemic control in Iranian children with type 1 diabetes mellitus: effect of gender**. *Indian J Pediatr* (2012.0) **79** 896-900. DOI: 10.1007/s12098-011-0613-8
31. Grabia M, Markiewicz-Żukowska R. **Nutritional status of pediatric patients with type 1 diabetes mellitus from northeast Poland: a case-control study**. *Diabetes Ther* (2021.0) **12** 329-343. DOI: 10.1007/s13300-020-00972-1
32. Kułaga Z, zdzyn R, skawia S, tkowska A, Grajda A. **Percentile charts for growth and nutritional status assessment in polish children and adolescents from birth to 18 year of age**. *Stand Med Pediatr.* (2015.0) **12** 119-135
33. czyk-Puzon´ B, Koszowska A, Bieniek AJ. **Basic anthropometric measurements and derived ratios in dietary counseling: part one**. *Piel Zdr Publ* (2018.0) **8** 217-222. DOI: 10.17219/pzp/92366
34. Targher G, Bertolini L, Padovani R. **Prevalence of non-alcoholic fatty liver disease and its association with cardiovascular disease in patients with type 1 diabetes**. *J Hepatol* (2010.0) **53** 713-718. DOI: 10.1016/j.jhep.2010.04.030
35. Ismail NA, Abd El Baky AMNE, Ibrahim MH, Ezzat WM, Elhosary YA. **The role of serum cytokeratin 18 and platelet count as non-invasive markers in the diagnosis of nonalcoholic fatty liver disease in children with type 1 diabetes mellitus**. *Pediatr Polska Polish J Paediatr* (2020.0) **95** 141. DOI: 10.5114/polp.2020.100990
36. Mertens J, De Block C, Spinhoven M, Driessen A, Francque SM, Kwanten WJ. **Hepatopathy associated with type 1 diabetes: distinguishing non-alcoholic fatty liver disease from glycogenic hepatopathy**. *Front Pharmacol* (2021.0). DOI: 10.3389/fphar.2021.768576
37. Donaghue KC, Chiarelli F, Trotta D. **Microvascular and macrovascular complications associated with diabetes in children and adolescents**. *Pediatr Diabetes* (2009.0) **10** 195-203. DOI: 10.1111/j.1399-5448.2009.00576.x
38. Schwimmer JB, Dunn W, Norman GJ. **SAFETY study: alanine aminotransferase cutoff values are set too high for reliable detection of pediatric chronic liver disease**. *Gastroenterology* (2010.0) **138** 1357-1364. DOI: 10.1053/j.gastro.2009.12.052
39. Fracanzani AL. **Risk of severe liver disease in nonalcoholic fatty liver disease with normal aminotransferase levels: a role for insulin resistance and diabetes**. *Hepatology* (2008.0) **48** 792-798. DOI: 10.1002/hep.22429
40. Mofrad P, Contos MJ, Haque M. **Clinical and histologic spectrum of nonalcoholic fatty liver disease associated with normal ALT values**. *Hepatology* (2003.0) **37** 1286-1292. DOI: 10.1053/jhep.2003.50229
41. Flisiak-Jackiewicz M, Lebensztejn DM. **Update on pathogenesis, diagnostics and therapy of nonalcoholic fatty liver disease in children**. *Clin Exp Hepatol* (2019.0) **5** 11-21. DOI: 10.5114/ceh.2019.83152
42. Hossain N. **Independent predictors of fibrosis in patients with nonalcoholic fatty liver disease**. *Clin Gastroenterol Hepatol* (2009.0) **7** 1224-1229. DOI: 10.1016/j.cgh.2009.06.007
43. Hanquinet S, Courvoisier D, Kanavaki A. **Acoustic radiation force impulse imaging-normal values of liver stiffness in healthy children**. *Pediatr Radiol* (2013.0) **43** 539-544. DOI: 10.1007/s00247-012-2553-5
44. Kummer S, Klee D, Kircheis G. **Screening for non-alcoholic fatty liver disease in children and adolescents with type 1 diabetes mellitus: a cross-sectional analysis**. *Eur J Pediatr* (2017.0) **176** 529-536. DOI: 10.1007/s00431-017-2876-1
45. Tuong TTK, Duc NM. **The role of ARFI and APRI In diagnosis of liver fibrosis on patients with common chronic liver diseases**. *Imaging Med* (2018.0) **10** 133-138. DOI: 10.14303/Imaging-Medicine.100014
46. Carter-Kent C, Yerian LM, Brunt EM, Angulo P, Kohli R. **Nonalcoholic steatohepatitis in children: a multicenter clinicopathological study**. *Hepatology* (2009.0) **50** 1113-1120. DOI: 10.1002/hep.23133
47. Sae-wong J, Chaopathomkul B, Phewplung T, Chaijitraruch N, Sahakitrungruang T. **The prevalence of nonalcoholic fatty liver disease and its risk factors in children and young adults with type 1 diabetes mellitus**. *J Pediatr* (2021.0) **230** 37.e1. DOI: 10.1016/j.jpeds.2020.10.043
48. Musso G, Gambino R, Cassader M. **Meta-analysis: natural history of non-alcoholic fatty liver disease (NAFLD) and diagnostic accuracy of non-invasive tests for liver disease severity**. *Ann Med* (2011.0) **43** 617-649. DOI: 10.3109/07853890.2010.518623
49. Singh D, Das CJ, Baruah MP. **Review article imaging of non - alcoholic fatty liver disease: a road less travelled**. *Indian J Endocrinol Metab* (2013.0) **17** 990-995. DOI: 10.4103/2230-8210.122606
50. Dietrich CF, Ferraioli G, Sirli R, Popescu A, Sporea I, Pienar C. **General advice in ultrasound based elastography of pediatric patients**. *Med Ultrason* (2019.0) **21** 315-326. DOI: 10.11152/mu-2063
51. Lee S, Choi YH, Cho YJ, Lee SB, Cheon J-E, Kim WS. **The usefulness of noninvasive liver stiffness assessment using shear-wave elastography for predicting liver fibrosis in children**. *BMC Med Imaging* (2021.0) **21** 68. DOI: 10.1186/s12880-021-00601-8
|
---
title: Exosomal Lnc NEAT1 from endothelial cells promote bone regeneration by regulating
macrophage polarization via DDX3X/NLRP3 axis
authors:
- Yuxuan Chen
- Yuanhao Wu
- Linlin Guo
- Shijie Yuan
- Jiaming Sun
- Kangcheng Zhao
- Jiecong Wang
- Ran An
journal: Journal of Nanobiotechnology
year: 2023
pmcid: PMC10029245
doi: 10.1186/s12951-023-01855-w
license: CC BY 4.0
---
# Exosomal Lnc NEAT1 from endothelial cells promote bone regeneration by regulating macrophage polarization via DDX3X/NLRP3 axis
## Abstract
### Background
Bone regeneration is a complex procedure that involves an interaction between osteogenesis and inflammation. Macrophages in the microenvironment are instrumental in bone metabolism. Amount evidence have revealed that exosomes transmitting lncRNA is crucial nanocarriers for cellular interactions in various biotic procedures, especially, osteogenesis. However, the underlying mechanisms of the regulatory relationship between the exosomes and macrophages are awaiting clarification. In the present time study, we aimed to explore the roles of human umbilical vein endothelial cells (HUVECs)-derived exosomes carrying nuclear enrichment enriched transcript 1 (NEAT1) in the osteogenesis mediated by M2 polarized macrophages and elucidate the underlying mechanisms.
### Results
We demonstrated HUVECs-derived exosomes expressing NEAT1 significantly enhanced M2 polarization and attenuated LPS-induced inflammation in vitro. Besides, the conditioned medium from macrophages induced by the exosomes indirectly facilitated the migration and osteogenic differentiation of bone marrow-derived mesenchymal stem cells (BMSCs). Mechanically, Exos carrying NEAT1 decreased remarkably both expression of dead-box helicase 3X-linked (DDX3X) and nod-like receptor protein 3 (NLRP3). The level of NLRP3 protein increased significantly after RAW264.7 cells transfected with DDX3X overexpression plasmid. Additionally, the knockdown of NEAT1 in exosomes partially counteracted the aforementioned effect of Exos. The results of air pouch rat model demonstrated that HUVECs-derived exosomes increased anti-inflammatory cytokines (IL-10) and decreased pro-inflammatory cytokines (IL-1β and IL-6) significantly in vivo, contributing to amelioration of LPS-induced inflammation. Afterwards, we further confirmed that the HUVECs-derived exosomes encapsulated in alginate/gelatin methacrylate (GelMA) interpenetrating polymer network (IPN) hydrogels could promote the bone regeneration, facilitate the angiogenesis, increase the infiltration of M2 polarized macrophages as well as decrease NLRP3 expression in the rat calvarial defect model.
### Conclusions
HUVECs-derived exosomes enable transmitting NEAT1 to alleviate inflammation by inducing M2 polarization of macrophages through DDX3X/NLRP3 regulatory axis, which finally contributes to osteogenesis with the aid of alginate/GelMA IPN hydrogels in vivo. Thus, our study provides insights in bone healing with the aid of HUVECs-derived exosomes-encapsulated composite hydrogels, which exhibited potential towards the use of bone tissue engineering in the foreseeable future.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12951-023-01855-w.
## Introduction
Cranial reconstruction is as such among the most demanding issues in craniomaxillofacial surgery for skull defect, mainly in brain tumor operations, traumatic injuries, craniotomies, and congenital cranial anomalies [1]. Current therapies for repairing skull defects normally use a cranial implant to accurately replace the absent cranial bone, include autografts, allografts and xenografts, metal biomaterials and macromolecular biomaterials [2, 3]. In order to best match the apertures on the cranium, the customized material and plastic bone replacements need to correspond accurately to the morphology of defect, as well as facilitate the bone regeneration.
Recent studies have indicated that the polarization state of macrophage was instrumental in bone healing [4]. Macrophages play critical roles in removing the tissue debris and secreting different signaling macromolecules to enlist the progenitors. Depletion of macrophages has been reported to result in an impaired and delayed osseous repair in fracture models of murine, indicating that the vital contributions of macrophages to bone regeneration [5, 6]. Macrophages are mainly classified into M1 and M2 subgroups, decreasing the M1 phenotype and increasing the M2 phenotype could remarkably enhanced osteogenesis and vascularization, furthermore, leading to bone healing [7]. Pioneer works have illustrated the ability of mesenchymal stem cells (MSCs)-derived exosomes to promote differentiation of macrophages into the M2 subtype, thereby reducing inflammatory responses and promoting tissue repair [8]. Compelling evidence demonstrated that M2 macrophages have been gradually recognized as a positive regulator of bone formation [9]. Nevertheless, the fundamental mechanisms associated with macrophages that manipulate the fate of BMSCs remain unclear.
Exosomes are extracellular vesicles rich in DNA, RNA, lipids, and proteins [10]. With the development of exosome-based therapeutics, exosomes act as mediators in cell–cell communications and are able to re-program recipient cells according to the parent cell's acculturation environment [11, 12]. In addition, exosome-based cell-free therapy is gaining attention due to effectively avoiding the risks of low survival rate, strong immune rejection, and high tumorigenicity of mutations caused by the direct use of cells [13, 14]. Human umbilical vein endothelial cells (HUVECs) derived exosomes are effective for promoting diabetic wound healing, reducing reperfusion damage, and protecting nerve cells against ischemia/reperfusion injury [17–19]. Inspired by these studies, our previous study has confirmed that the exosomes derived from HUVECs significantly improved the angiogenesis ability of endothelial progenitor cells and increased the survival area of the flap in vivo through nuclear enrichment enriched transcript 1 (NEAT1)/Wnt/β-catenin signal pathway [15]. Such suggestions indicate that HUVECs-derived exosomes may serve as a novel therapeutic tool for tissue repair, such as bone regeneration. Given the potential benefits of HUVECs-derived exosomes, therefore, it is urgent to find a way to investigate the role of HUVECs-derived exosomes on bone regeneration in this study.
To address this, long noncoding RNAs (lncRNAs) could be loaded within exosomes to regulate gene expression in host cell via cell communication [16]. LncRNA NEAT1 is a classic lncRNA which resides in paraspeckles [17] and can act as an inflammatory meditor [18]. Of note, NEAT1 could also activate NLR family CARD domain containing 4 (NLRC4) and nod-like receptor protein 3 (NLRP3) inflammasome, and stabilize the caspase-1 to promote IL-1β production and pyroptosis [19–21]. However, several studies have also illustrated that NEAT1 could not only ameliorate LPS-induced inflammation, but inhibit the activation of NLRP3 inflammasome by interacting with specific pathways and miRNAs, finally leading to M2 polarization [20, 22]. Dead-box helicase 3X-linked (DDX3X) is essential for assembly of NLRP3 inflammasomes and stress granules due to its binding action to NLRP3 [23]. DDX3X deletion could alleviate cardiomyocyte pyroptosis induced by LPS via inhibiting NLRP3 inflammasome activation [24]. Abundant evidence has reported that the axis of DDX3X-NLRP3-mediated pyroptosis could be regulated by several factors, such as AKT, aluminum, TLR4 and so on [25–27]. Considering the undefined role of NEAT1 in immunity, works need to be done to decipher whether NEAT1 could regulate NLRP3 inflammasome activation mediated through DDX3X for macrophage polarization, suppression of chronic inflammation and bone regeneration.
In vivo, unloaded exosome-based therapeutics still face challenges, due to the short half-life and fast removal rate, especially applied in bone repair, which is a long-term and complex multiple process [28]. Therefore, it is critical to select a suitable carrier to carry exosomes to maintain the function of exosomes and achieve sustained release at the target site. Here, we report on an alginate/GelMA IPN hydrogel to deliver bioactive exosomes and prefabricated in a mold to match the defective region to promote osseous restoration.
In our work, we aimed to investigate the effect of exosomes derived from HUVECs associated with DDX3X/NLRP3 axis on osteogenesis. Our study revealed that NEAT1 in HUVECs-derived exosomes has dramatic effects on the regulation of macrophage plasticity to resolve chronic inflammation, promote the osteogenic function of BMSCs and enhance in vivo osteogenesis, via suppressing NLRP3 inflammasome activation regulated by DDX3X. We propose this study may provide potential insight for the therapeutic application of exosomal lncRNAs in osteogenesis.
## Characterization of exosomes
Regarding the identification of exosomes, we performed nanoparticle tracking analysis (NTA), western blotting and TEM. The exosomes appeared as typical cup-shaped morphology by TEM (Fig. 1A). Western blotting analysis revealed the positive expression of exosomal specific markers CD63, CD81 and TSG101 in the isolated particles with no significant difference between the two groups (Fig. 1B). In addition, NTA results showed that the size distribution of exosomes from HUVECs (Exos) and HUVECs transfected with NEAT1 siRNA (si-Exos) displayed similar peak diameters of around 105 nm (Exos: 103.6 ± 44.4 nm, si-Exos: 108.5 ± 37.5 nm), and both the diameters of exosomes in the range of 50–150 nm was > $99\%$ (Fig. 1C). As shown in Fig. 1D, the red fluorescence of Dil-labeled exosomes were clearly observed in the cytoplasm of RAW264.7 around the nucleus indicating the two kinds of exosomes could display a cellular transmission activity similarly which was not affected by parental NEAT1 depletion. Quantitative Real-time Polymerase Chain Reaction (qRT-PCR) data showed a significantly decrease in NEAT1 expression in the secreted exosomes following the transfection of the NEAT1 inhibitor, compared with the expression level in non-treated exosomes (Additional file 1: Fig. S1).Fig. 1Characterization and internalization of HUVECs-derived Exos by RAW264.7. A Morphology of Exos and si-Exos identified by TEM. Scale bar = 200 nm. B Western blot analysis of the specific markers of exosomes, including CD63, CD81 and TSG101. C The particle size distribution and particle concentration of Exos and si-Exos detected by NTA. D The uptake of Exos and si-Exos by RAW264.7 cells. FITC-phalloidin (green) and DAPI (blue) were used to stain the cytoskeleton and nucleus of RAW264.7, respectively. Exosomes were labeled with Dil (red). Scale bar = 50 μm
## Characterization of exosomes embedded with hydrogel
To explore the effects of the exosomes in vivo, the Exos or si-Exos were embedded with alginate/GelMA IPN hydrogels to establish composites. According to the 3D reconstruction images of exosomes encapsulated in hydrogels (Fig. 2A–D), amount of red fluorescence Dil-labeled exosomes were homogeneously distributed in the hydrogels. Figure 2E indicated the composite hydrogels had a continuously slow and controlled release effect on exosomes during the monitoring span. The release curves of two different kinds of exosomes in hydrogel showed similar trends and no significant differences were observed at all time points between the two groups. Notably, approximately $50\%$ of the exosomes were still remained inside the hydrogels after 15 days. Fig. 2Exosome retention ability of alginate/GelMA IPN hydrogel. A/C 3D image of Dil-labeled Exos/si-Exos incorporated in alginate/GelMA IPN hydrogel. B/D overlapping image of A/C. E Release curves of Exos and si-Exo from composite hydrogels. Scale bar = 100 μm
## HUVECs derived exosomes attenuated inflammation by enhancing M2 polarization in vitro
To determine the macrophage polarization state under stimulation with Exos and si-Exos after treated with LPS, we tested M1 and M2 polarization markers by flow cytometry, enzyme-linked immunosorbent assay (ELISA) and qRT-PCR. CD86, IL-1β and IL-6 represent the inflammatory state of M1 polarization, while CD206, IL-10 and Arg1 are the markers of anti-inflammatory state of M2 polarization.
By flow cytometry, CD86 expression was significantly increased in LPS group, but markedly decreased after treated with the two kinds of exosomes. In contrast, the expression of M2 macrophage marker (CD206) significantly elevated with the treatment of the exosomes to varying degrees. Meanwhile, the knockdown of NEAT1 resulted in lower proportions of M2 macrophages than did normal HUVECs-derived exosomes (Fig. 3A/B, $P \leq 0.01$). Next, by ELISA, IL-1β as well as IL-6 secretion were significantly decreased, while IL-10 secretion was markedly increased in the groups treated with Exos and si-Exos, especially treated with Exos, compared to the control group (Fig. 3C, $P \leq 0.05$).Fig. 3HUVECs-derived Exos attenuated inflammation by enhancing M2 polarization in vitro. A Representative images of the percentage of CD86 and CD206 positive cells detected by flow cytometry analysis. B Quantification of flow cytometry analysis of the percentage of CD86 and CD206 positive cells. C The concentrations of IL-1β, IL-6 and IL-10 of the supernatants detected by ELISA. D The relative gene expression of IL-1β, IL-6, IL-10 and Arg1 detected by qRT-PCR. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$ Furthermore, qRT-PCR results showed that the expression levels of IL-1β and IL-6 was significantly decreased, while the expression of IL-10 and Arg-1 was significantly increased in Exos and si-Exos groups, with the highest enhancement in Exos group. ( Fig. 3D, $P \leq 0.05$).
## HUVECs-derived exosomes promoted osteogenic differentiation and migration of BMSCs
To explore whether the exosomes could promote the osteointegration by enhancing M2 polarization, the supernatant from RAW264.7 with different treatments (PBS, LPS, LPS + Exos, LPS + si-Exos) as conditioned medium (CM) were collected.
By RT-PCR assay, the mRNA levels of osteogenic genes, alkaline phosphatase (ALP), OCN and RUNX2 of BMSCs significantly elevated in either LPS + Exos CM or LPS + si-Exos CM from days 7 -14 days compared to the other groups (Fig. 4A, $P \leq 0.05$). Likewise, RUNX2, OCN and ALP were upregulated dramatically through a western blotting assay (Fig. 4B/C, $P \leq 0.05$). Accordantly, after knockdown of NEAT1, the positive effect of HUVECs-derived exosomes on osteogenesis of BMSCs was partially attenuated. Fig. 4Conditional medium from macrophages treated by the exosomes promoted migration and osteogenic differentiation of BMSCs. A qRT-PCR analysis for mRNA expressions of ALP, OCN and RUNX2 on day 7 and day 14. B Western blot analysis and quantification C of protein levels of ALP, OCN and RUNX2 on day 7 and day 14. D Representative images of transwell assay and quantification (F) of cell migration. Representative images (E) and quantification (G) of ALP staining after 7 days of osteogenic induction. Scale bar = 200 μm. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$ To further elucidate the effect of NEAT1 on the migration capacity of BMSCs, transwell assay was performed. As shown in Fig. 4D/F, LPS + Exos CM and LPS + si-Exos CM attracted more MSCs to the lower chamber than did LPS CM group ($P \leq 0.01$). However, conditioned medium derived from unstimulated macrophages had no apparent influence on the migratory ability of BMSCs. Meanwhile, there was a significance difference from the LPS + Exos CM versus LPS + si-Exos CM ($P \leq 0.01$), indicating a stronger transmigrated capability of the HUVECs-derived exosomes without NEAT1 inhibition.
Similarly, ALP activity was significantly increased in the LPS + Exos CM and the LPS + si-Exos CM group compared to the other two groups, which was highest in the LPS + Exos CM group (Fig. 4E/G, $P \leq 0.01$).
## Exosomal NEAT1 inhibited LPS-induced inflammation via the DDX3X/NLRP3 pathway
To investigate the underlying mechanisms of NEAT1 mediated the HUVEC derived exosomes regulation of osteogenesis, RAW264.7 cells were successfully transfected with DDX3X overexpression plasmid. The transfection efficiency was verified by qRT-PCR (Fig. 5A).Fig. 5Lnc NEAT1 inhibited NLRP3 inflammasome activation by targeting DDX3X. A qRT-PCR analysis for mRNA level of DDX3X. B–F Western blot analysis and quantification of protein levels of ASC, Caspase-1, NLRP3 and DDX3X. G ELISA analysis for the concentrations of IL-1β of the supernatants. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$ Immunoblotting demonstrated that the expression levels of ASC, Caspase-1 and NLRP3 were found to be significantly increased following LPS stimulation and DDX3X overexpression. Besides, the expression of NLRP3 increased following the application of NEAT1 inhibited exosomes. NEAT1-shRNA transfection increased the LPS-induced increase in protein expression of ASC, Caspase-1, NLRP3 and DDX3X. The changes in the expression of NLRP3 in each group were in accordance with those of DDX3X indicating the inhibitory effect of NEAT1 on NLRP3 was dependent on the suppression of DDX3X. Similar to NEAT1 depletion, protein levels of ASC, Caspase-1 and NLRP3 were significantly increased by overexpression of DDX3X. Of note, overexpression of DDX3X exerted the increasing effect on expression of NLRP3 inflammasome-associated proteins which was similar to LPS. However, increased levels of ASC, Caspase-1, NLRP3 and DDX3X were partially ameliorated by Exos/si-Exos in both the LPS and pcDDX3X groups, with the strongest inhibition in the Exos group (Fig. 5B–F, $P \leq 0.05$). The ELISA results showed that similar to the groups treated with LPS, the cytokine concentration of IL-1β was significantly elevated in the groups overexpressed DDX3X compared to control group. In addition, the increasing level of IL-1β in the pc-DDX3X groups was inhibited by Exos and si-Exos, with the Exos group having a more prominent inhibitory activity (Fig. 5G, $P \leq 0.05$).
## HUVECs-derived exosomes inhibited the inflammation by enhancing M2 polarization in vivo
To assess the effect of the exosomes on polarization of macrophages in vivo, an air pouch model was established (Fig. 6A). The level of inflammation associated cytokines were evaluated by ELISA. IL-1β and IL-6, both for M1 phenotype, were highest, in LPS group, but decreased significantly after the application of Exos and si-Exos. Exos and si-Exos remarkably elevated IL-10 secretion, a marker of M2 polarization, which was significantly reduced subjected to LPS treatment. In particular, the inhibition of NEAT1 reversed the effect of M2 polarization in si-Exos group (Fig. 6B, $P \leq 0.$ 01). The results of immunofluorescence staining showed that both M2 macrophages (CD206 labeled, red) and M1 macrophages (CD86 labeled, green) were visible in all groups. The ratio of M2/M1 was significant increased in the group treated with exosomes, especially the exosomes without NEAT1 knockdown (Fig. 6C/D, $P \leq 0.$ 05).Fig. 6Air pouch model in vivo. A Schematic illustration of air pouch model establishment. B ELISA analysis of IL-1β, IL-6 and IL-10 concentration of lavage fluid. C Representative confocal images for CD86 (green) and CD206 (red). The nucleus were counterstained with DAPI (blue). D Quantification of the ratio of CD206/CD86 positive cells per field. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$
## Micro-CT and histological analysis of bone regeneration
The 3D reconstruction of micro-CT images of the rat cranial defects at 4 and 12 weeks were shown in Fig. 7A. The images at 4 weeks showed that sporadic newly formed bone filled the defects of the Exos + Gel and si-Exos + Gel groups, but no obvious bone structure was generated in the control group. With the increase of time after implantation, more new bone was formed. In particular, at 12 weeks, the largest amount of new bone, which nearly filled the calvarial defects, was found in the Exos + Gel group than that the other three groups. Interestingly, little new bone was generated in the control group, while more newly formed bone could be observed in Gel group. Fig. 7Micro-CT analysis and histological staining of rat bone reconstruction in vivo. A 3D reconstruction images of calvarial bone at 4 and 12 weeks postoperatively. B *Quantitative analysis* of BV/TV, Tb. N and Tb. Sp in 3D micro-CT images at 4 and 12 weeks. H&E staining (C) and Masson’s trichrome staining (D). Scale bar = 200 μm. HB host bone; NB new bone. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$ Based on micro-CT images, quantitative analysis including BV/TV, Tb. N and Tb. Sp demonstrated that BV/TV ratio and Tb. N in Exos + Gel and si-Exos + Gel groups were all significantly higher and Tb. Sp was significantly lower than those in the other two groups, with the Exos + Gel group showing a stronger modulation (Fig. 7B, $P \leq 0.05$).
Histological analysis was performed to observe the tissue in the cranial defect area. H&E staining showed an increased deposition of new bone that were generated both along the margin and into the center of calvarial defects after application of the two kinds of exosomes, especially HUVECs-derived exosomes (Fig. 7C). Whereas, in the control group, the most regions of the defects were filled with fibrotic connective tissue with few visible bone regenerations. Masson’s trichrome staining exhibited that in the Gel group, differs from the Blank group, a small amount of new bone can be observable and both Exos + Gel and si-Exos + Gel groups induced a significantly large amount of the osteoid matrix formation and among the four treatments, composite hydrogels with Exos facilitated the most bone regeneration (Fig. 7D).
## Immunohistochemical and immunofluorescence staining of bone regeneration
To further assess the insights into the effect of the exosomes on osteogenesis, immunohistochemical staining of CD31, ALP, OCN, RUNX2 and NLRP3 were performed. The number of CD31-positive vessels was significantly increased in the Exos + Gel group than the other three groups (Fig. 8A/B; $P \leq 0.01$). More neovascularization was observed in si-Exos + Gel group compared to Gel and blank groups. Moreover, osteogenic marker ALP, OCN and RUNX2, were more abundantly expressed in both exosomes groups compared to the other two groups, especially in Exos group (Fig. 8A/C–E; $P \leq 0.05$). In addition, NLRP3 inflammasome components were significantly decreased in the Exos + Gel and si-Exos + Gel groups, with a much sharper decline in Exos + Gel group, demonstrating that the knockdown of NEAT1 increased the expression of NLRP3 (Fig. 8A/F; $P \leq 0.05$).Fig. 8Immunohistochemistry and immunofluorescence staining of bone defect. Representative images of A CD31, ALP, OCN, RUNX2 and NLRP3. Quantitative analysis of B CD31, C ALP, D OCN, E RUNX2 and F NLRP3. G Representative confocal images of calvarial sections for CD86 (green) and CD206 (red). In all images the nucleus were counterstained with DAPI (blue). H Quantification of the ratio of CD206/CD86 positive cells per field. Scale bar = 100 μm. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001$ Moreover, immunofluorescence staining in Fig. 8G revealed, M2 macrophages, labeled by CD206 (red) and M1 macrophages, labeled by CD86 (green), were distributed in the fibrous tissues in all groups. Quantitative analysis indicated that a significant increase of M2/M1 ratio in the Exos + Gel group, indicating a dominant phenotype of M2 polarized macrophages (Fig. 8H; $P \leq 0.05$). Although ratio of M2 macrophage was slightly increased in the Gel group, they were still lower than that in the hydrogels composed with exosomes. These effects of Exos were partly compromised by NEAT1 inhibition.
## Discussion
Restoration of bone deficiency is still a clinically challenging problem to treat [29]. Generally, bone regeneration typically undergoes three sequential phases: inflammation, regeneration, and remodeling [30]. The local inflammatory response is crucial for bone tissue regeneration, and an appropriate grade of inflammation can promote bone healing [31, 32]. Recent studies have found that the proportion of M2-polarized macrophages increases during fracture repair, which induce osteogenic differentiation by secreting various growth factors [33, 34]. Among the approaches to modulate M1 to M2 macrophage conversion, strategies including drugs, exosomes, and hydrogels have been exploited to regulate macrophage polarization and promote bone recovery [35, 36]. Exosomes, as emerging cell-free therapies, have received growing attention for their immunomodulatory capabilities [8]. To present, however, very little researches have concerned the effect of exosomes from HUVECs towards osseous restoration, and few studies are available on the mechanisms through which macrophages and specific exosomes stimulate osseous restoration. Therefore, our research sought to unravel the exosomes produced by HUVECs stimulated osteogenesis via involving macrophages through the NEAT1/DDX3X/NLRP3 signaling pathway, through which we found the close association between macrophage polarization and local immune microenviroment, contributing to the bone repair.
Numerous studies in the past have found that exosomes can effectively participate in the immune response to facilitate tissue regeneration [37, 38]. Exosomes from HUVECs were also reported as offering a positive effect in protecting nerve cells from ischemia/reperfusion injury, improving fibroblast photoaging, inducing endothelial progenitor cell homing and inhibiting osteoclast formation to reduce bone resorption [15, 39–41]. Particularly, we demonstrated that HUVEC-derived exosomes could directly promote osteogenic differentiation and increase the migratory capacity of BMSCs, and the aforementioned phenomenon was slightly inhibited with the knockdown of NEAT1 (Additional file 3: Fig. S3). However, the modulatory ability of HUVECs-derived exosomes on immune countermeasure remains largely unknown. In the present study, compared to the conspicuous inflammatory reaction induced by LPS, HUVECs-derived exosomes initiated an obvious alleviated inflammatory reaction characterized by significantly enhanced M2-phenotype polarization of RAW264.7 cells. In accordance, with the advent of polarization, the polarized macrophages further secreted anti-inflammatory cytokines, such as IL-10 and Arg-1. And the level of pro-inflammatory-related cytokines (IL-1β, IL-6) was reduced. Nevertheless, the application of NEAT1 silencer could partially counteracted the trends but not entirely abolish, which further verified the critical role of NEAT1 on M2 polarization. Consistently, Zhang et al. found M2 polarization can be promoted by NEAT1 and promotes choroidal neovascularization by sponging miRNA-148a-3p [42]. Besides, NEAT1 accelerates multiple myeloma progression by regulating B7-H3 to promote M2 macrophage polarization [43]. Since there are various bioactivators in exosomes which may also be responsible for M2 phenotype polarization by diverse mechanisms, therefore, NEAT1 inhibitor could only suppress the signaling pathway associated with NEAT1, rendering it incapable of fully eliminating the effect of Exos on promotion of M2 polarization [8, 44, 45].
Emerging evidence has demonstrated that the transition of macrophages to M2 polarization is of great importance on tissue remodeling and able to enhance the osteogenic differentiation and migration of MSCs [46, 47]. It has been widely recognized that M2 polarized macrophages elicit a variety of cytokine including bone morphogenetic protein 2 (BMP-2), IL-10, platelet-derived growth factor (PDGF), transforming growth factor-β (TGF-β) and arginine, which contribute to vascular sprouting and bone healing [48]. In line with these literature studies, our observations suggested that exosomal NEAT1 indirectly enhanced the osteogenic differentiation and migration ability of BMSCs mediated by increased M2 polarized macrophages in a conditioned culture system, which was weakened by NEAT1 silencing. Furthermore, in air pouch model, the involvement of HUVECs-derived exosomes remarkably ameliorated the LPS-induced inflammatory response. With the knockdown of NEAT1, the phenomenon of increase of anti-inflammatory cytokines (IL-10), M2 macrophages (CD206 labeled) and decrease of pro-inflammatory cytokines (IL-6 and IL-1β), M1 macrophages (CD86 labeled) in the lavage fluid was partially reversed. Combining the aforementioned findings, we confirmed that exosomal NEAT1 plays an indispensable role in increasing the proportion of M2 macrophages in vitro and in vivo, and further promoted BMSC migration and osteogenic differentiation.
To achieve optimal in vivo application of exosomes in various tissues repair, a suitable carrier to maintain effective local concentration and function of exosomes during the repair process should to be taken into consideration [49, 50]. Hydrogels has generally been presumed a favorable carrier for exosomes because of its similarities to the extracellular matrix (ECM), appropriate physical strength and good biocompatibility [51, 52]. In our release profile, the alginate/GelMA IPN hydrogel represented stable release capacity for HUVECs-derived exosomes with or without NEAT1 inhibition and retained approximately $50\%$ exosomes in the alginate hydrogel during 15 days. Moreover, the composite hydrogels keep releasing exosomes which still maintain the typical cup-like structure at day 15 (Additional file 2: Fig. S2). These results suggest that our synthetic alginate/GelMA IPN hydrogels are reliable vehicles of exosomes to maintain a uniform distribution, sustained release and stable function of exosomes, ultimately achieving the therapeutic goal.
To deeply evaluate the bone repair ability of HUVECs-derived exosomes loaded hydrogels in vivo, a rat cranial defect model was employed. Based on the images and quantification of micro-CT analysis, the in vivo application of the exosomes and hydrogel composites markedly enhanced bone regeneration compared to the other groups, especially the naive exosomes without NEAT1 knockdown. Furthermore, the histological and IHC evaluation revealed significant higher levels of angiogenic and osteogenic markers, increased local infiltration of M2 polarized macrophages in the Exos group, leading to a better repair of rat cranial defect, which corresponded to their strongest ability in vitro. Our results are in parallel with the former conclusion that exosomal NEAT1 could promote angiogenesis [15]. Moreover, bone regeneration highly depends on angiogenesis, which is a vital step to restore blood flow providing nutrients, further exerting a positive feedback to bone healing [53]. These encouraging findings supported that exosomal NEAT1 could facilitate bone regeneration in vivo.
In regard to the potential effect of NEAT1 on regulation of macrophages, our study further focused on investigating the downstream molecular mechanism of NEAT1. The noteworthy targets, particularly the NLRP3 inflammasomes, assembled from ASC, caspase-1 and NLRP3, are vital members of the innate immune system [54]. NLRP3 inflammasome plays a significant role in bone inflammation, because of the causal caspase-1 activation and its correlation to inhibition osteogenic adipose accumulation and differentiation in bone tissues [55]. Occupation of NLRP3 inflammasome can accelerate bone resorption, promote osteoclast differentiation and aggravate inflammation, which increase the risk of osteoporosis [56]. However, the regulatory relationship of NEAT1 and NLRP3 still remains to be controversial. Herein we hypothesized that DDX3X worked as a downstream target of NEAT1 to regulate NLRP3 inflammasome activation. Recently, DDX3X has been shown to drive the assembly of NLRP3 inflammasome and determine the fate of cells by interacting with NLRP3 [23]. Knockdown of DDX3X significantly suppressed NLRP3 inflammasome activation caused by LPS and attenuated pyroptosis and cell injury in H9c2 cells [24]. The application of HUVECs-derived exosomes decreased the expressions of ACS, caspase-1, NLRP3 and DDX3X dramatically, which is in consistence with the in vivo findings. Notably, with the use of DDX3X mimic, the level of NLRP3 inflammasome increased highly, and then decreased to some extent when combining the use of the specific exosomes. Thus, these exciting findings supported that HUVECs derived exosomal NEAT1 inhibited NLRP3 inflammasome activation mediated by DDX3X, thereby alleviating LPS-induced inflammation. Coincidentally, it has been suggested that NEAT1 alleviate ischemic stroke inhibiting NLRP3 mediated via miR-10b-5p/BCL6 axis [57]. Likewise, Nong et al. indicated NEAT1 could sponge miR-193a-3p via NF-κB signal pathway to alleviate inflammation of normal human fibroblast cells induced by LPS [58]. Conversely, others have demonstrated contrary evidence to those in the present study. The inhibition of NEAT1 could protect endothelial cells from hypoxia-induced NLRP3 inflammasome activation by targeting miR-204/BRCC axis [59]. The reason for these differences is likely to be attributed to the different experimental details and the mechanism network involved, which finally exert diverse effects.
As far as we know, we have taken the first step in the direction that exosomal NEAT1 inhibited LPS-induced inflammatory responses contributing to bone regeneration via the DDX3X/NLRP3 signaling pathway. Regarding to the mutual influence between macrophages and BMSCs, it would be of great importance to gain an in-depth understanding of NEAT1 function involved in the bone healing process. Notably, our study not only proposed a novel cell therapy replacement but also provides a prospective therapeutic strategy to broaden the translational use of HUVECs-derived exosomes for the treatment of clinical bone defects.
## Conclusion
Taken together, the role of HUVECs derived exosomal NEAT1 in bone regeneration was unraveled in our current work and the underlying mechanisms were portrayed. Our study clearly indicated NEAT1 can promote M2 polarization and suppress inflammatory responses by modulating the DDX3X/NLRP3 axis, and further promote the osteogenic differentiation and migration potential of BMSC to facilitate bone repair in vivo and in vitro (Fig. 9). The positive impact was suppressed in part by NEAT1 knockdown. Meanwhile, alginate/GelMA IPN hydrogels were verified to provide a reliable delivery platform for exosomes. Therefore, our findings confirmed a promising role of HUVECs derived exosomal NEAT1 as a therapeutic alternative for bone defect. Fig. 9Schematic of the HUVECs derived exosomal NEAT1 mediated bone regeneration mediated by macrophage polarization via DDX3X/NLRP3 axis
## Cell isolation and culture
HUVECs and murine-derived RAW264.7 cells, macrophage cell line, were both purchased from Cell Bank of Chinese Academy of Sciences (Shanghai, China) and cultured in DMEM medium (HyClone, Logan, UT, USA) and RPMI 1640 medium (HyClone), respectively, containing $1\%$ penicillin–streptomycin solution (Biosharp, Hefei, Anhui, China) and $10\%$ fetal bovine serum (FBS) (Evergreen, Zhejiang, China) at 37 °C with $5\%$ CO2.
BMSCs were isolated from neonatal male Sprague–Dawley (SD) rats (5-days-old) according a previous medthod [60]. Briefly, after execution of the rats using the spinal dislocation method, they were then immersed in $75\%$ alcohol for 30 min. The femur and tibia were isolated. Then, the cartilage at both ends of the bone was cut and the bone marrow cavity was repeatedly flushed with culture medium until the bone appeared white. Finally, the rinsed bone marrow tissue was inoculated in a culture dish with 6 mL of culture medium (L-DMEM containing $10\%$ FBS, $1\%$ penicillin and streptomycin). The culture was incubated at 37 °C and the culture medium was changed every 3 days. The passages of BMSC used in studies were 1 to 3.
## Transfection assay
LncRNA NEAT1 expression in HUVECs was knocked down using NEAT1 siRNA lentivirus, synthesized by Wuhan Biofavor Biothech Service Co., with Lipofectamine 3000 Transfection Reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The sequences of NEAT1 siRNA were as follow: forward: GCC TTG TAG ATG GAG CTT GC; reverse: GCA CAA CAC AAT GAC ACC CT. RAW264.7 were inoculated on 6-well plates and cultured overnight. The sequence of DDX3X (Additional file 4: Table S1) was provided by Huayan Biotechnology Co., LTD (Wuhan, China) and the plasmids pcDNA3.1-DDX3X were constructed. The plasmids were introduced into RAW264.7 using Lipofectamine 3000 Transfection Reagent (Invitrogen) according to the manufacturer's instructions. Transfection efficiency was verified by qRT-PCR, and cells were collected for subsequent experiments.
## Extraction and identification of exosomes
Exosomes, including Exos and si-Exos, were both isolated. Specifically, HUVECs were pretreated by transfecting with NEAT1 siRNA (si-NEAT1-HUVECs). Afterwards, the mediums were changed to exosome free mediums and cultured for 72 h when the confluence reached 80–$90\%$. The supernatant of HUVECs and si-NEAT1-HUVECs group was collected for extracting exosomes. Cells and cell debris were first removed by centrifugation at 300 × g for 10 min and then at 2,000 × g for 10 min. Then, centrifugation at 10,000 × g for 30 min and 0.22 μm sterilized filters (MilliporeExpress® PES membrane, Millex, Bedford, MA, USA) were used to remove the larger diameter extracellular vesicles. Finally, the exosomes were obtained by centrifugation at 110,000 × g for 70 min and resuspended with pre-cooled PBS. The exosomes were lysed with radio-immunoprecipitation assay (RIPA) lysis buffer, and the protein concentration in the exosomes was detected with the bicinchoninic acid (BCA) protein assay kit (Servicebio, Wuhan, Hubei, China) according to the manufacturer's instructions. qRT-PCR assay was applied to detect the expression level of NEAT1 in Exos and si-Exos.
The exosomes were identified by the marker proteins CD81, CD63, and TSG101 by western blotting. Morphological identification was observed by Transmission electron microscopy microscope (TEM, Tecnai G2 20, Thermo Fisher Scientific, Cleveland, OH, USA). The particle size distribution and nanoparticle concentration were evaluated by NTA using ZetaView (Particle Metri, Bavaria, Germany). Three 60 s video recordings of the Brownian motion of the exosomes were taken while the blue laser (488 nm) irradiated the exosomes, and finally the NTA software analysis was performed using ZetaView software (version 8.02.31, Particle Metri, Bavaria, Germany).
## Internalization of exosomes
To trace the internalization of exsomes in macrophages, exosomes were labeled with Dil dye (Solarbio, Beijing, China). Then, the Dil-labeled exosomes were co-cultured with RAW264.7 cells for 3 h. The cells were stained with phalloidin-FITC (C1033, Beyotime Biotechnology, Shanghai, China) and DAPI (C1002, Beyotime Biotechnology) to visualize the cytoskeleton and nucleus, respectively. The uptake was photographed with a confocal microscope (Nikon, Tokyo, Japan).
## Effects of conditioned media on BMSC responses
A conditioned culture system was designed to investigate the effects of exosomes-induced macrophages on the osteogenic differentiation of BMSCs. To obtain the CM from macrophages, RAW264.7 macrophages were seeded in 6-well plate at a density of 2 × 105/well and achieved a confluence of 70–$80\%$. Then, the cells were cultured in normal complete medium containing 100 ng/mL LPS (Biosharp) for 24 h. Thereafter, Exos and si-Exos (100 μg/mL) were added to study the effect of exosomes on the polarization of macrophages, respectively. At day 3, the supernatant of the macrophages in four groups (PBS, LPS (100 ng/mL), LPS + Exos (100 μg/mL) and LPS + si-Exos (100 μg/mL)) was collected, centrifuged at 1200 rpm for 10 min to remove precipitates, then mixed with osteogenic medium (HyCyte, Jiangsu, Suzhou, China) at a ratio of 1:1 (v/v) to obtain the mixed RAW264.7 CM. Specifically, BMSCs were seeded at a density of 2 × 105 cells/well in 6-well plate. After 24 h, the culture medium of BMSCs was replaced with different RAW264.7 CM. At day 7 and 14, the total RNA of BMSCs in conditioned culture system were extracted. The expression of osteogenic related genes and protein was detected by qRT-PCR and western blotting. When the culture reaches day 7, ALP staining was performed using the BCIP/NBT alkaline phosphatase color development kit (C3206, Beyotime Biotechnology) according to the instructions provided by the manufacturer.
## Transwell migration assay
BMSCs were seeded in the upper chamber at a density of 5 × 104 in a transwell plate (24-well plate, JETBIOFIL, Guangzhou, Guangdong, China). 600 μL conditioned medium derived from the macrophages (PBS; LPS (100 ng/mL); LPS + Exos (100 μg/mL) and LPS + si-Exos (100 μg/mL)) were added to the lower chamber. After incubation for 24 h, the cells remained on the upper side of the chamber were removed. The cells which migrated to the lower chamber were stained with DAPI (Beyotime Biotechnology). Images were observed by a confocal microscope (Nikon) and analyzed with ImageJ software (National Institutes of Health, USA). The three fields of view per well were analyzed by counting the number of DAPI-labeled positive staining cells.
## Preparation of alginate/GelMA IPN Hydrogel and exosome-hydrogel composite
To prepare the alginate/GelMA IPN hydrogel, 200 mg alginic acid sodium salt (SA) was dissolved in 10 mL calcium-free dulbecco phosphate-buffered saline (DPBS) solution (Procell, Wuhan, Hubei, China) to obtain SA solution. 500 μL SA solution was mixed with photo-initiator lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP, StemEasy, Jiangsu, China) to the target concentration of $0.5\%$ (m/v) LAP and $1\%$ (m/v) SA. Afterwards, 100 mg gelatin methacryloyl (GelMa) was added into the mixture and then put in 70 °C for 30 min until completely dissolved to obtain the hybrid $1\%$ (m/v) alginate/$10\%$ (m/v) GelMA IPN-hybrid hydrogel. 100 μg exosomes (including Exos and si-Exos) were resuspended in 100 μL hybrid hydrogel in the dark at 37℃ to produce the exosomes + Gel mixed solutions. Then, the mixed solutions were photopolymerized after exposed to ultraviolet (UV) light for 30 s and subsequently immersed in $2\%$ (m/v) CaCl2 solution for crosslinking to finally obtain exosomes + Gel composites. In order to measure the distribution of exosomes in hydrogel, exosomes were labeled with Dil and then observed by the confocal microscope (Nikon).
## The exosome retaining ability of hydrogels
The release profile of exosomes from alginate/GelMA IPN hydrogel was evaluated using BCA protein assay kit [61] and TEM. Briefly, 100 μL of the above prepared exosome/ hydrogels composites containing 100 μg exosomes were placed in a 96-well plate supplemented with 100 μL PBS. Subsequently, the supernatant in the well was collected and the well was refilled with another 100 μL PBS at specific time points. The protein concentration in PBS collected was measured to draw a release curve. Morphological characterization of exosomes released from the hydrogel composites was observed by TEM.
## Flow cytometry
In order to identify the polarization of RAW264.7, LPS-preconditioned cells were incubated in medium containing Exos or si-Exos (100 μg/mL). Subsequently, treated cells with TruStain Fc X™ PLUS (156603, Biolegend, USA) and Triton X-100 (P0096-100 ml, Beyotime Biotechnology), then they were incubated with the antibodies. Thereafter, the cells were analyzed by flow cytometry (BD LSR FortessaTM X-20, San Jose, USA) utilizing the flowJo software (version 10.7.1, Stanford University, USA). Flow cytometry analysis was performed on each sample in triplicate. The antibodies were used as follows: Phycoerythrin (PE) conjugated-CD86 (12-0862-82, eBioscience, USA), PE/Cyanine7 conjugated-CD206 (141719, Biolegend).
## Quantitative real-time polymerase chain reaction
TRIzol reagent (Invitrogen) was utilized to extract total RNA. cDNA was synthesized by RevertAid First Strand cDNA Synthesis Kit (K1622, Thermo Fisher Scientific) following the manufacturer’s instructions. The levels of mRNA were calculated and normalized to GAPDH using the 2 −ΔΔCt method. The primer sequences were listed in Additional file 5: Table S2.
## Western blot analysis
Total protein was measured by the BCA Protein Detection Kit (Thermo Fisher Scientific). The proteins were extracted by gradient sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred to nitrocellulose membranes. After being blocked by $5\%$ skimmed milk, the membrane was incubated at 4 °C along with the primary antibodies overnight, followed by incubation with horseradish peroxidase (HRP)-conjugated secondary antibodies at room temperature for 1 h. Band intensities was analyzed using Image-Pro Plus (Media Cybernetics, USA) by densitometry. β-actin was used as an internal control. The following primary antibodies were used: NLRP3 (1:1000, ab214185; Abcam, Cambridge, UK), IL-1β (1:1000, A16288, ABclonal, Woburn, MA, USA), Caspase-1 (1:1000, ab207802, Abcam), the adaptor protein apoptosis-associated speck-like protein containing a CARD (ASC; 1:1000, ab180799, Abcam), DDX3X (1:1000, ab235940, Abcam), GAPDH (1:5000, ab8245, Abcam) and β-actin (1:5000, ab8227, Abcam).
## Enzyme-linked immunosorbent assay
Cell supernatants of RAW264.7 cells incubated on 24-well plates were treated with PBS, LPS, LPS + Exos and LPS + si-Exos for 24 h and 4 groups (PBS, LPS (100 ng/mL), LPS + Exos (100 μg/mL) and LPS + si-Exos (100 μg/mL)) lavage fluid from air pouch were collected for detecting IL-6, IL-1β and IL-10 by ELISA kits (Sigma-Aldrich, St. Louis, MO, USA), respectively, according to the manufacturer’s specifications. Besides, the supernatants of RAW264.7 cells treated with PBS, LPS, LPS + Exos, LPS + si-Exos, pc-DDX3X, pc-DDX3X + Exos, and pc-DDX3X + si-Exos were collected respectively for detection of IL-1β by ELISA kits (Sigma-Aldrich).
## Animal experiments
In our study, All the procedures were approved by the guidelines of the Animal Research Committee of the Huazhong University of Science and Technology.
## Air pouch assay in vivo
12 C57BL/6 mice (40 g; 10 weeks old; male) were used and randomly assigned to four groups ($$n = 4$$): PBS, LPS, LPS + Exos and LPS + si-Exos. The mouse were anesthetized by intraperitoneal injection of $5\%$ pentobarbital sodium (Sigma-Aldrich) at 0.5 mL/kg before surgery, and injectied 5 mL sterilized air into loose dorsal tissue and supplemented with 2 mL of sterile air on days 3 and 5 to establish a stable air pouch model (Fig. 0.6A). At day 7, 2 mL PBS, LPS (100 ng/mL), LPS + Exos and LPS + si-Exos (100 μg/mL) were injected subcutaneously. After 24 h, 2 mL PBS was applied for washing the subcutaneous pouch to collect the lavage fluid. The levels of IL-6, IL-10 and IL-1β were detected by ELISA. Afterwards, the mice were sacrificed. All airpouches were harvested and processed for immunofluorescence staing to determine the inflammatory response by the number of infiltrated macrophages.
## Cranial defect rat model establishment in vivo
16 Sprague–Dawley rats (400 g; 8 weeks old; male) were used for this operation and randomly assigned to four groups ($$n = 4$$): [1] alginate/GelMA IPN hydrogel groups (Gel): 40 μL hydrogel; [2] HUVECs-Exos + hydrogel (Exos + Gel): 40 μg HUVECs-Exos dissolved in 40 μL hydrogel; [3] si-NEAT1-Exos/hydrogel (si-Exos + Gel): 40 μg si-Exos dissolved in 40 μL hydrogel; [4] control group: 40 μL PBS. All rats were anaesthetized by an intraperitoneal injection of $5\%$ pentobarbital sodium (Sigma–Aldrich) at a dose of 0.5 mL/kg prior to surgery. Afterwards, two full-thickness critical size of cranial defects (5 mm in diameter) were created symmetrically on each side of the rat’s cranium with a trephine bur. The different components of implant materials (5 mm in diamter and 1 mm in depth) were prepared in advance and gently implanted into the cranial defect according to the grouping, respectively. Afterwards, the soft tissues were closed and the skin was sutured. Animals were euthanized at 4 and 12 weeks postoperatively to harvest specimens from the defect sites for the further study.
## Micro-computed tomography (micro-CT)
The cranium defect sites were evaluated via Micro-CT (SkyScan 1176). The 3D bone reconstruction images were obtained by using CTvox software (version 3.1.1, Bruker). As the density of bone tissue and hydrogel differs, the MicroView software (Bruker version 1.15.4) was used to distinguish one from the other. The percentage of new bone volume/total tissue volume (BV/TV), bone trabecular separation (Tb. Sp) and bone trabecular number (Tb. N) were analyzed.
## Histology, immunohistochemistry and immunofluorescence analysis
The harvested specimens were fixed, decalcified, gradually dehydrated, and embedded in paraffin. Samples were sectioned into 3 μm. Haematoxylin and Eosin (H&E) and Masson’s trichrome staining was performed for the observation of osteogenesis in calvarial bone defects. For immunohistochemistry staining (IHC), the sections were incubated with anti-rat CD31 antibody (1:1000, 28083-1-AP, PTG, Chicago, USA) to label neovascularization and anti-rat antibodies against ALP (1:100, 11400-1-AP, PTG), OCN (1:200, 23418-1-AP, PTG) and RUNX2 (1:1000, 20700-1-AP, PTG) to characterize the new bone in the tissue, followed with 3,3-diaminobenzidine substrate (DAB, Vector Laboratories, Burlingame, CA 4 min), and counterstained with Mayer’s hematoxylin (Sigma-Aldrich). Slides were imaged using a Leica microscope (Leica, Wetzlar, Germany). ImageJ software (National Institutes of Health) was used to analyze CD31 positive vessels and calculated the area of ALP, OCN and RUNX2-positive region in each field of view. In each section of each sample, at least three sections and three areas were randomly selected for observation and analysis.
For immunofluorescence staining, to detect the macrophages, the sections were incubated with primary antibodies against CD86 (1:100, A2353, Abclonal) and CD206 (1:2000, 60143-1-IG, PTG) overnight at 4 °C, followed with secondary antibodies (Goat Anti-Rabbit IgG H&L (HRP), ab205718, Abcam, Goat Anti-Mouse IgG H&L (HRP), ab6789, Abcam) at room temperature for 1 h in the dark. Cell nucleus were stained with DAPI (Beyotime Biotechnology) for 20 min. Images were acquired by a confocal microscope (Nikon). The ratio of CD206-positive cells to CD86-positive cells was analyzed using ImageJ software (National Institutes of Health).
## Statistical analysis
All Results are expressed as mean ± SD derived from at least three independent experiments. One way analysis of variance (ANOVA) and unpaired t-test were used to assess the differences between groups with GraphPad Prism 8.0 software (San Diego, CA, USA). Statistical significance was shown as follows: * $p \leq 0.05$, ** $p \leq 0.01$, *** $p \leq 0.001$, **** $p \leq 0.0001$; ns, no statistically significant difference ($p \leq 0.05$).
## Supplementary Information
Additional file 1: Figure S1. Quantification of NEAT1 level in Exos or si-Exos. *** $P \leq 0.001.$Additional file 2: Figure S2. Observation of the morphology of Exos and si-Exos released from the hydrogel at day 15 under TEM. Scale bar = 200nm. Additional file 3: Figure S3. Exos/si-Exos promoted osteogenic differentiation and migration of BMSCs. ( A) qRT-PCR analysis for mRNA expressions of ALP, OCN and RUNX2 on day 7. ( B) Western blot analysis and quantification. ( C) of protein levels of ALP, OCN and RUNX2 on day 7. ( D) Representative images of transwell assay and quantification (F) of cell migration. Images (E) and quantification (G) of ALP staining after 7 days of osteogenic induction. Scale bar = 200 μm. * $P \leq 0.05$, ** $P \leq 0.01$, *** $P \leq 0.001.$Additional file 4: Table S1. Construction of Lentiviral Overexpression Vector. Additional file 5: Table S2. Primer sequences employed for reverse transcription-quantitative polymerase chain reaction.
## References
1. Fearon JA, Griner D, Ditthakasem K, Herbert M. **Autogenous bone reconstruction of large secondary skull defects**. *Plast Reconstr Surg* (2017) **139** 427-438. DOI: 10.1097/PRS.0000000000002941
2. Szpalski C, Barr J, Wetterau M, Saadeh PB, Warren SM. **Cranial bone defects: current and future strategies**. *Neurosurg Focus* (2010) **29** E8. DOI: 10.3171/2010.9.FOCUS10201
3. Grayson WL, Bunnell BA, Martin E, Frazier T, Hung BP, Gimble JM. **Stromal cells and stem cells in clinical bone regeneration**. *Nat Rev Endocrinol* (2015) **11** 140-150. DOI: 10.1038/nrendo.2014.234
4. Pajarinen J, Lin T, Gibon E. **Mesenchymal stem cell-macrophage crosstalk and bone healing**. *Biomaterials* (2019) **196** 80-89. DOI: 10.1016/j.biomaterials.2017.12.025
5. Raggatt LJ, Wullschleger ME, Alexander KA. **Fracture healing via periosteal callus formation requires macrophages for both initiation and progression of early endochondral ossification**. *Am J Pathol* (2014) **184** 3192-3204. DOI: 10.1016/j.ajpath.2014.08.017
6. Miron RJ, Bosshardt DD. **OsteoMacs: key players around bone biomaterials**. *Biomaterials* (2016) **82** 1-19. DOI: 10.1016/j.biomaterials.2015.12.017
7. Ferrante CJ, Leibovich SJ. **Regulation of macrophage polarization and wound healing**. *Adv Wound Care* (2012) **1** 10-16. DOI: 10.1089/wound.2011.0307
8. Arabpour M, Saghazadeh A, Rezaei N. **Anti-inflammatory and M2 macrophage polarization-promoting effect of mesenchymal stem cell-derived exosomes**. *Int Immunopharmacol* (2021) **97** 107823. DOI: 10.1016/j.intimp.2021.107823
9. Jiang G, Li S, Yu K. **A 3D-printed PRP-GelMA hydrogel promotes osteochondral regeneration through M2 macrophage polarization in a rabbit model**. *Acta Biomater* (2021) **128** 150-162. DOI: 10.1016/j.actbio.2021.04.010
10. Kalluri R, LeBleu VS. **The biology, function, and biomedical applications of exosomes**. *Science (New York N.Y.).* (2020). DOI: 10.1126/science.aau6977
11. Pegtel DM, Gould SJ. **Exosomes**. *Annu Rev Biochem* (2019) **88** 487-514. DOI: 10.1146/annurev-biochem-013118-111902
12. Liang X, Ding Y, Zhang Y, Tse HF, Lian Q. **Paracrine mechanisms of mesenchymal stem cell-based therapy: current status and perspectives**. *Cell Transplant* (2014) **23** 1045-1059. DOI: 10.3727/096368913X667709
13. Verweij FJ, Revenu C, Arras G. **Live tracking of inter-organ communication by endogenous exosomes in vivo**. *Dev Cell* (2019) **48** 573-589.e574. DOI: 10.1016/j.devcel.2019.01.004
14. El Andaloussi S, Mäger I, Breakefield XO, Wood MJ. **Extracellular vesicles: biology and emerging therapeutic opportunities**. *Nat Rev Drug Discov* (2013) **12** 347-357. DOI: 10.1038/nrd3978
15. Guo L, Chen Y, Feng X. **Oxidative stress-induced endothelial cells-derived exosomes accelerate skin flap survival through Lnc NEAT1-mediated promotion of endothelial progenitor cell function**. *Stem Cell Res Ther* (2022). DOI: 10.1186/s13287-022-03013-9
16. Kopp F, Mendell JT. **Functional classification and experimental dissection of long noncoding RNAs**. *Cell* (2018) **172** 393-407. DOI: 10.1016/j.cell.2018.01.011
17. Hirose T, Yamazaki T, Nakagawa S. **Molecular anatomy of the architectural NEAT1 noncoding RNA: The domains, interactors, and biogenesis pathway required to build phase-separated nuclear paraspeckles**. *Wiley Interdiscip Rev RNA.* (2019) **10** e1545. DOI: 10.1002/wrna.1545
18. Pan Y, Wang T, Zhao Z. **Novel insights into the emerging role of NEAT1 and its effects downstream in the regulation of inflammation**. *J Inflamm Res* (2022) **15** 557-571. DOI: 10.2147/JIR.S338162
19. Zhang P, Cao L, Zhou R, Yang X, Wu M. **The lncRNA Neat1 promotes activation of inflammasomes in macrophages**. *Nat Commun* (2019) **10** 1495. DOI: 10.1038/s41467-019-09482-6
20. Lin S, Wen Z, Li S. **LncRNA Neat1 promotes the macrophage inflammatory response and acts as a therapeutic target in titanium particle-induced osteolysis**. *Acta Biomater* (2022) **142** 345-360. DOI: 10.1016/j.actbio.2022.02.007
21. Zhang M, Zheng Y, Sun Y. **Knockdown of NEAT1 induces tolerogenic phenotype in dendritic cells by inhibiting activation of NLRP3 inflammasome**. *Theranostics* (2019) **9** 3425-3442. DOI: 10.7150/thno.33178
22. Yang Y, Ma S, Ye Z. **NEAT1 in bone marrow mesenchymal stem cell-derived extracellular vesicles promotes melanoma by inducing M2 macrophage polarization**. *Cancer Gene Ther* (2022) **29** 1228-1239. DOI: 10.1038/s41417-021-00392-8
23. Samir P, Kesavardhana S, Patmore DM. **DDX3X acts as a live-or-die checkpoint in stressed cells by regulating NLRP3 inflammasome**. *Nature* (2019) **573** 590-594. DOI: 10.1038/s41586-019-1551-2
24. Feng D, Guo L, Liu J. **DDX3X deficiency alleviates LPS-induced H9c2 cardiomyocytes pyroptosis by suppressing activation of NLRP3 inflammasome**. *Exp Ther Med* (2021) **22** 1389. DOI: 10.3892/etm.2021.10825
25. Guo X, Chen S, Yu W. **AKT controls NLRP3 inflammasome activation by inducing DDX3X phosphorylation**. *FEBS Lett* (2021) **595** 2447-2462. DOI: 10.1002/1873-3468.14175
26. Hao W, Hao C, Wu C. **Aluminum impairs cognitive function by activating DDX3X-NLRP3-mediated pyroptosis signaling pathway**. *Food Chem Toxicol Int J Publ Br Ind Biol Res Assoc* (2021) **157** 112591. DOI: 10.1016/j.fct.2021.112591
27. Wang J, Zhang F, Xu H. **TLR4 aggravates microglial pyroptosis by promoting DDX3X-mediated NLRP3 inflammasome activation via JAK2/STAT1 pathway after spinal cord injury**. *Clin Transl Med* (2022) **12** e894. DOI: 10.1002/ctm2.894
28. Zhang Y, Xie Y, Hao Z. **Umbilical mesenchymal stem cell-derived exosome-encapsulated hydrogels accelerate bone repair by enhancing angiogenesis**. *ACS Appl Mater Interfaces* (2021) **13** 18472-18487. DOI: 10.1021/acsami.0c22671
29. Dimitriou R, Jones E, McGonagle D, Giannoudis PV. **Bone regeneration: current concepts and future directions**. *BMC Med* (2011) **9** 66. DOI: 10.1186/1741-7015-9-66
30. Marsell R, Einhorn TA. **The biology of fracture healing**. *Injury* (2011) **42** 551-555. DOI: 10.1016/j.injury.2011.03.031
31. Jiao H, Xiao E, Graves DT. **Diabetes and its effect on bone and fracture healing**. *Curr Osteoporos Rep* (2015) **13** 327-335. DOI: 10.1007/s11914-015-0286-8
32. Wheatley BM, Nappo KE, Christensen DL, Holman AM, Brooks DI, Potter BK. **Effect of NSAIDs on bone healing rates: a meta-analysis**. *J Am Acad Orthop Surg* (2019) **27** e330-e336. DOI: 10.5435/JAAOS-D-17-00727
33. Muñoz J, Akhavan NS, Mullins AP, Arjmandi BH. **Macrophage polarization and osteoporosis: a review**. *Nutrients* (2020) **12** 2999. DOI: 10.3390/nu12102999
34. Xiong Y, Chen L, Yan C. **M2 Macrophagy-derived exosomal miRNA-5106 induces bone mesenchymal stem cells towards osteoblastic fate by targeting salt-inducible kinase 2 and 3**. *J Nanobiotechnol* (2020) **18** 66. DOI: 10.1186/s12951-020-00622-5
35. Annamalai RT, Turner PA, Carson WFT, Levi B, Kunkel S, Stegemann JP. **Harnessing macrophage-mediated degradation of gelatin microspheres for spatiotemporal control of BMP2 release**. *Biomaterials* (2018) **161** 216-227. DOI: 10.1016/j.biomaterials.2018.01.040
36. Mi B, Chen L, Xiong Y. **Osteoblast/osteoclast and immune cocktail therapy of an exosome/drug delivery multifunctional hydrogel accelerates fracture repair**. *ACS Nano* (2022) **16** 771-782. DOI: 10.1021/acsnano.1c08284
37. Wei F, Li Z, Crawford R, Xiao Y, Zhou Y. **Immunoregulatory role of exosomes derived from differentiating mesenchymal stromal cells on inflammation and osteogenesis**. *J Tissue Eng Regen Med* (2019) **13** 1978-1991. DOI: 10.1002/term.2947
38. Huang Y, He B, Wang L. **Bone marrow mesenchymal stem cell-derived exosomes promote rotator cuff tendon-bone healing by promoting angiogenesis and regulating M1 macrophages in rats**. *Stem Cell Res Ther* (2020) **11** 496. DOI: 10.1186/s13287-020-02005-x
39. Xiao B, Chai Y, Lv S. **Endothelial cell-derived exosomes protect SH-SY5Y nerve cells against ischemia/reperfusion injury**. *Int J Mol Med* (2017) **40** 1201-1209. DOI: 10.3892/ijmm.2017.3106
40. Ellistasari EY, Kariosentono H, Purwanto B. **Exosomes derived from secretome human umbilical vein endothelial cells (Exo-HUVEC) ameliorate the photo-aging of skin fibroblast**. *Clin Cosmet Investig Dermatol* (2022) **15** 1583-1591. DOI: 10.2147/CCID.S371330
41. Song H, Li X, Zhao Z. **Reversal of osteoporotic activity by endothelial cell-secreted bone targeting and biocompatible exosomes**. *Nano Lett* (2019) **19** 3040-3048. DOI: 10.1021/acs.nanolett.9b00287
42. Zhang P, Lu B, Zhang Q. **LncRNA NEAT1 sponges MiRNA-148a-3p to suppress choroidal neovascularization and M2 macrophage polarization**. *Mol Immunol* (2020) **127** 212-222. DOI: 10.1016/j.molimm.2020.08.008
43. Gao Y, Fang P, Li WJ. **LncRNA NEAT1 sponges miR-214 to regulate M2 macrophage polarization by regulation of B7–H3 in multiple myeloma**. *Mol Immunol* (2020) **117** 20-28. DOI: 10.1016/j.molimm.2019.10.026
44. Xian P, Hei Y, Wang R. **Mesenchymal stem cell-derived exosomes as a nanotherapeutic agent for amelioration of inflammation-induced astrocyte alterations in mice**. *Theranostics* (2019) **9** 5956-5975. DOI: 10.7150/thno.33872
45. Liu W, Yu M, Xie D. **Melatonin-stimulated MSC-derived exosomes improve diabetic wound healing through regulating macrophage M1 and M2 polarization by targeting the PTEN/AKT pathway**. *Stem Cell Res Ther* (2020) **11** 259. DOI: 10.1186/s13287-020-01756-x
46. Zhang Y, Böse T, Unger RE, Jansen JA, Kirkpatrick CJ, van den Beucken J. **Macrophage type modulates osteogenic differentiation of adipose tissue MSCs**. *Cell Tissue Res* (2017) **369** 273-286. DOI: 10.1007/s00441-017-2598-8
47. Gong L, Zhao Y, Zhang Y, Ruan Z. **The macrophage polarization regulates MSC osteoblast differentiation in vitro**. *Ann Clin Lab Sci* (2016) **46** 65-71. PMID: 26927345
48. Schlundt C, Fischer H, Bucher CH, Rendenbach C, Duda GN, Schmidt-Bleek K. **The multifaceted roles of macrophages in bone regeneration: a story of polarization, activation and time**. *Acta Biomater* (2021) **133** 46-57. DOI: 10.1016/j.actbio.2021.04.052
49. Sun J, Yin Z, Wang X, Su J. **Exosome-laden hydrogels: a novel cell-free strategy for in-situ bone tissue regeneration**. *Front Bioeng Biotechnol* (2022) **10** 866208. DOI: 10.3389/fbioe.2022.866208
50. Xie Y, Guan Q, Guo J, Chen Y, Yin Y, Han X. **Hydrogels for exosome delivery in biomedical applications**. *Gels Basel, Switzerland.* (2022) **8** 328. PMID: 35735672
51. Pishavar E, Luo H, Naserifar M. **Advanced hydrogels as exosome delivery systems for osteogenic differentiation of MSCs: application in bone regeneration**. *Int J Mol Sci* (2021) **22** 6203. DOI: 10.3390/ijms22126203
52. Safari B, Aghazadeh M, Davaran S, Roshangar L. **Exosome-loaded hydrogels: a new cell-free therapeutic approach for skin regeneration**. *Eur J Pharm Biopharm* (2022) **171** 50-59. DOI: 10.1016/j.ejpb.2021.11.002
53. Kusumbe AP, Ramasamy SK, Adams RH. **Coupling of angiogenesis and osteogenesis by a specific vessel subtype in bone**. *Nature* (2014) **507** 323-328. DOI: 10.1038/nature13145
54. Chi W, Chen H, Li F, Zhu Y, Yin W, Zhuo Y. **HMGB1 promotes the activation of NLRP3 and caspase-8 inflammasomes via NF-κB pathway in acute glaucoma**. *J Neuroinflamm* (2015) **12** 137. DOI: 10.1186/s12974-015-0360-2
55. Wang L, Chen K, Wan X, Wang F, Guo Z, Mo Z. **NLRP3 inflammasome activation in mesenchymal stem cells inhibits osteogenic differentiation and enhances adipogenic differentiation**. *Biochem Biophys Res Commun* (2017) **484** 871-877. DOI: 10.1016/j.bbrc.2017.02.007
56. Jiang N, An J, Yang K. **NLRP3 inflammasome: a new target for prevention and control of osteoporosis?**. *Front Endocrinol* (2021) **12** 752546. DOI: 10.3389/fendo.2021.752546
57. Zhou ZW, Ren X, Zhou WS, Li AP, Zheng LJ. **LncRNA NEAT1 alleviates ischemic stroke via transcriptional inhibition of NLRP3 mediated by the miR-10b-5p/BCL6 axis**. *Acta Neurobiol Exp* (2022) **82** 12-21
58. Nong W. **Long non-coding RNA NEAT1/miR-193a-3p regulates LPS-induced apoptosis and inflammatory injury in WI-38 cells through TLR4/NF-κB signaling**. *Am J Transl Res* (2019) **11** 5944-5955. PMID: 31632562
59. Yao T, Song Y, Li S, Gu J, Yan X. **Inhibition of lncRNA NEAT1 protects endothelial cells against hypoxia/reoxygenation-induced NLRP3 inflammasome activation by targeting the miR-204/BRCC3 axis**. *Mol Med Rep* (2022). DOI: 10.3892/mmr.2021.12548
60. Zhou C, Liu S, Li J. **Collagen functionalized with graphene oxide enhanced biomimetic mineralization and in situ bone defect repair**. *ACS Appl Mater Interfaces* (2018) **10** 44080-44091. DOI: 10.1021/acsami.8b17636
61. Wang M, Wang C, Chen M. **Efficient Angiogenesis-Based Diabetic Wound Healing/Skin Reconstruction Through Bioactive Antibacterial Adhesive Ultraviolet Shielding Nanodressing With Exosome Release**. *ACS Nano* (2019) **13** 10279-10293. DOI: 10.1021/acsnano.9b03656
|
---
title: 'Gut microbiota and host cytochrome P450 characteristics in the pseudo germ-free
model: co-contributors to a diverse metabolic landscape'
authors:
- Shanshan Wang
- Qiuyu Wen
- Yan Qin
- Quan Xia
- Chenlin Shen
- Shuai Song
journal: Gut Pathogens
year: 2023
pmcid: PMC10029254
doi: 10.1186/s13099-023-00540-5
license: CC BY 4.0
---
# Gut microbiota and host cytochrome P450 characteristics in the pseudo germ-free model: co-contributors to a diverse metabolic landscape
## Abstract
### Background
The pseudo germ-free (PGF) model has been widely used to research the role of intestinal microbiota in drug metabolism and efficacy, while the modelling methods and the utilization of the PGF model are still not standardized and unified. A comprehensive and systematic research of the PGF model on the composition and function of the intestinal microbiota, changes in host cytochrome P450 (CYP450) enzymes expression and intestinal mucosal permeability in four different modelling cycles of the PGF groups are provided in this paper.
### Results
16S rRNA gene amplicon sequencing was employed to compare and analyze the alpha and beta diversity, taxonomic composition, taxonomic indicators and predicted function of gut microbiota in the control and PGF groups. Bacterial richness and diversity decreased significantly in the PGF group beginning after the first week of establishment of the PGF model with antibiotic exposure. The PGF group exposed to antibiotics for 4-week-modelling possessed the fewest indicator genera. Moreover, increased intestinal mucosal permeability occurred in the second week of PGF model establishment, indicating that one week of antibiotic exposure is an appropriate time to establish the PGF model. The results of immunoblots revealed that CYP1A2, CYP2C19 and CYP2E1 expression was significantly upregulated in the PGF group compared with the control group, implying that the metabolic clearance of related drugs would change accordingly. The abundance of functional pathways predicted in the gut microbiota changed dramatically between the control and PGF groups.
### Conclusions
This study provides information concerning the microbial and CYP450 enzyme expression profiles as a reference for evaluating drug metabolism differences co-affected by gut microbiota and host CYP450 enzymes in the PGF model.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13099-023-00540-5.
## Introduction
Cytochrome P450 (CYP450) enzymes, important objects of preclinical drug metabolism research, are a major factor of variability in drug response and pharmacokinetic characteristics. Six isoenzymes—CYP1A2, CYP2C9, CYP2C19, CYP2D6, CYP2E1 and CYP3A4—are responsible for more than $90\%$ of clinical drugs metabolised by the liver [1]. Many researchers have demonstrated that the large inter-individual variability of CYP450 enzymes, influenced by genetic polymorphisms, underlies unpredictable and varied clinical drug responses [2]. However, the genetic variation of metabolic enzymes can only explain part of the individual variation in drug responses [2, 3].
The human gut microbiome which has emerged as an active metabolic ‘organ’ is tightly interconnected with drug metabolism and plays an increasingly important role in the field of individualised drug treatment. Moreover, emerging data have revealed that the compositional characteristics of the gut microbiota and bacterial enzymatic activity might vary with the influence of external factors such as treatment with antibiotics [3]. Therefore, it is crucial to investigate the effect of microbial changes on individual variability in drug responses in addition to changes in CYP450 enzymatic activity. Insights into gut microbiome–directed metabolism of drugs and individual differences in drug metabolic outcomes have led to general research on varying pharmacokinetic profiles by utilization of pseudo germ-free (PGF) and germ-free (GF) models [4–8]. For example, Wu et al. [ 9] revealed that gut microbiota occupied a prominent position in altering metformin pharmacokinetics by using the PGF model. Guo et al. [ 10] discovered that changes in gut microbiota–mediated biotransformation alter the blood exposure of Panax notoginseng saponin metabolites in PGF rats. Recently, however, the PGF model rather than the GF model has been frequently used to explore the relationship between the gut microbiota and xenobiotic metabolism because the PGF model has healthier intestinal mucosa and more robust immune function than the GF model [11].
PGF models, treated with antibiotics, have been extensively employed to study the potential roles of the gut microbiome in physiological and pathological processes of host life [6, 12–14]. Subsequently, metabolomic and proteomic research methods have been used to study the physiological characteristics of PGF models, such as urine metabolic profiles and liver and kidney proteomic expression profiles [15, 16]. Nevertheless, there are few systematic studies involving a comparative characterisation of the gut microbiome between the PGF model and the normal intestinal state. Interestingly, researchers have unveiled that the gut microbiome could modify host gene expression, including CYP450, through a comparative study of GF and conventional mice [17, 18], which might alter the pharmacokinetic features of oral drugs. It is not clear whether such changes in CYP450 enzyme expression would occur in the PGF model established by antibiotic exposure and reconstruct the metabolic phenotype of the PGF model.
Of note, ampicillin sodium, neomycin sulphate, metronidazole and vancomycin hydrochloride are common antibiotic combinations used to establish PGF model, but the modelling methods and the modelling time of PGF models described in the literature lack uniformity and standardization [6, 12, 19, 20]. In addition, an increasing number of researchers have probed into the exact role of microbial biotransformation on individual drug metabolism differences using the PGF model, with a tendency to focus on the gut microbiota itself while ignoring host CYP450. The existing studies have little systematic awareness about the PGF model in terms of the gut microbiota and the hepatic CYP450 protein expression.
Given the fact that the gut microbiota and host hepatic CYP450 expression affect on drug metabolism outcomes in vivo and the extensive application of the PGF model in the field of drug metabolism differences, a comparative study of the gut microbiota and host hepatic CYP450 expression in normal and PGF rats was carried out in the present study. In addition, the function of gut microbiota was predicted in the PGF model and intestinal mucosal permeability and inflammation levels of PGF rats were determined to evaluate the possible contributing factors on the intestinal absorption in the PGF model after oral administration of drugs. The results of this study provide a reference in an effort to standardise how a PGF model is established and exhibits more characteristic evidence of the PGF model for differential metabolic study.
## Materials and reagents
Vancomycin hydrochloride was purchased from Macklin Co., Ltd (Shanghai, China). Ampicillin sodium, neomycin sulphate and metronidazole were provided by Aladdin Co., Ltd (Shanghai, China). Radio Immunoprecipitation Assay (RIPA) lysis buffer and phenylmethanesulfonyl fluoride (PMSF) were obtained from Beyotime Biotechnology Co., Ltd (Shanghai, China). The following primary antibodies were used: anti-CYP1A2, anti-CYP2D6, anti-CYP2E1, anti-CYP3A4 (Affinity Cincinnati, OH, USA), anti-CYP2C9 (Biosynthesis Biotechnology, Beijing, China), anti-CYP2C19 (Abcam, Cambridge, UK). The secondary antibodies were peroxidase-conjugated anti-rabbit IgG (Affinity Cincinnati, OH, USA). ECL-chemiluminescent kit and Polyvinylidene fluoride membrane were purchased from Millipore (MA, USA). Prestained Protein Marker was obtained from Biosynthesis Biotechnology (Beijing, China). BCA kit was purchased from Beyotime Biotechnology Co., Ltd (Shanghai, China). Rapid transfer solution and blocking buffer were purchased from NCM Biotech (Suzhou, China). Other reagents used in preparation of $10\%$ SDS-page gel and separation solution were purchased from Sigma-Aldrich Co., Ltd (St. Louis, MO, USA). ELISA kits of rat diamine oxidase (DAO), rat endotoxin (ET) and rat leptin (LEP) were purchased from Jiangsu Feiya Biological Technology Co., Ltd (Yancheng, China). The HiPure Stool DNA Extraction Kit was purchased from Magen (Guangzhou, China). Polymerase chain reaction (PCR) related reagents were obtained from New England Biolabs (Ipswich, MA, USA). AxyPrep DNA Gel Extraction Kit was purchased from Axygen Biosciences (Union City, CA, USA).
## Animals
Six- to eight-week-old male Sprague–Dawley rats (200 ± 20 g) were provided by the Laboratory Animal Center of Anhui Medical University; they were all specific pathogen-free grade (Certificate no. SCXK 2017–001). The rats were all housed in a specific pathogen-free class experimental animal room. The animal experimental protocol was in compliance with the specific pathogen free class animal laboratory operating procedures. The animal experiments were conducted in accordance with the Experimental Animal Ethics Committee of Anhui Medical University (LLSC20170348).
## Construction of PGF rat model
Four PGF groups (each group $$n = 6$$) were exposed to an antibiotic cocktail for different times: 1 week (PGF-1W), 2 weeks (PGF-2W), 3 weeks (PGF-3W) and 4 weeks (PGF-4W), while the control group ($$n = 6$$) received autoclaved water without antibiotics. The four PGF groups were administered with an antibiotic cocktail including 1 g/L ampicillin sodium, 1 g/L metronidazole, 1 g/L neomycin sulfate and 0.5 g/L vancomycin hydrochloride in autoclaved water. The drinking water was replaced once every 2 days. During the construction of the PGF model, faeces of all PGF groups were collected by massage stimulation on an ultra-clean bench. The collected faeces (each group $$n = 6$$) was placed in tubes, immediately flash-frozen in liquid nitrogen for 5 min and then stored in − 80 °C freezer for 16S rRNA sequencing.
## Assessment of body weight in rats
Six male rats were randomly divided equally into the control group and PGF groups (each group $$n = 3$$). Body weight gain of the control and PGF groups were recorded during antibiotic treatment.
## DAO/ET/LEP assay
Blood samples from the control and PGF groups (each group $$n = 3$$) were taken from the fundus venous plexus of rats at the various modelling cycles. Serum was isolated from the blood and the levels of diamine oxidase (DAO), endotoxin (ET) and leptin (LEP)—indicators of the intestinal mucosal barrier—were detected using the DAO Kit, ET Kit and LEP Kit according to the manufacturer's instructions.
## Western blot analysis
Total proteins of hepatic tissues from the control and PGF-1W groups (each group $$n = 3$$) were extracted with RIPA buffer containing PMSF and protein concentration was determined using BCA protein assay kit (Beyotime Biotechnology, Shanghai, China). The lysates were then subjected to western blot to explore the expression levels of CYP450 enzymes [21]. In brief, a prestained protein marker and the 5 μL protein samples were separated by $10\%$ SDS-page gel under electrophoretic conditions. The separated protein was transferred to a polyvinylidene fluoride membrane using a rapid transfer solution. The membrane was then incubated with blocking buffer (to block nonspecific protein binding) and then incubated with primary followed by secondary antibodies (see “Materials and reagents” Section). Finally, signals were visualised using the ECL-chemiluminescent detection kit. ImageJ software was used to analyse the blots.
## PCR amplification of the V3–V4 region of the 16S rRNA gene
Faecal microbial DNA from all rats was extracted using the HiPure Stool DNA Extraction Kit (Magen, Guangzhou, China) according to the manufacturer’s instructions. Then, the V3-V4 region of 16S rRNA gene was amplified by PCR. The primer sequences used in this assay were as follows: 341-forward: 5′-CCTACGGGNGGCWGCAG-3′; 806- reversed: 5′-GGACTACHVGGGTATCTAAT-3′. Each 50 μL reaction mixture contained 10 μL of 5 × Reaction Buffer, 10 μL of 5 × High GC Enhancer, 1.5 μL of 2.5 mM dNTPs, 1.5 μL of upstream and downstream primers (10 μM final concentration), 0.2 μL of High-Fidelity DNA Polymerase, and 50 ng of template DNA. The thermal cycling conditions were: 95 °C for 5 min; 30 cycles of 95 °C for 1 min, 60 °C for 1 min and 72 °C for 1 min; and 72 °C for 7 min.
## Illumina sequencing
DNA library sequencing was performed on NovaSeq 6000 platform using paired-end 250 bases mode with single indexing. Amplicons were extracted from $2\%$ agarose gels, purified using the AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, Union City, CA, USA) according to the manufacturer's instructions, and quantified using the ABI StepOnePlus Real-Time PCR System (Life Technologies, Foster City, CA, USA). The purified amplicons were submitted to paired-end sequencing (PE250) on the Illumina platform according to standard procedures.
## Bioinformatics
The sequenced data was submitted to the DADA2 method [22] for de-duplication, correction, noise reduction, and chimera removal to obtain amplicon sequence variant (ASV) sequences and abundances. Taxonomic annotation was performed using SILVA database [23] and bacterial abundance information at each level was obtained based on ASV abundance. Using ASVs and bacterial abundance table, the taxa abundance and taxa classification of the gut microbiota were analysed by compositional analysis. Then the alpha diversity indexes containing Chao1, Shannon and Phylogenetic diversity (PD) tree and beta diversity of faecal samples were compared to analyse the similarities and differences of bacterial diversity among the diverse groups. Linear discriminant analysis effect size (LEfse) [24] and indicator analysis of the microbial composition were used to screen the dominant taxa groups. Taxonomic indicators with statistical difference were filtered by the P-value of less than 0.01. The sequencing results were annotated with the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to predict the metabolic pathways involved in the gut microbiota of the samples using functional prediction analysis tool such as TAX4FUN [25].
## Statistical analysis
Independent sample t tests were used for the comparative analysis of the statistical significance of differences between two groups, while one-way analysis of variance was used for statistical analysis of differences among three or more groups. The comparison of relative values of body weight were carried out by a two-way analysis of variance. Analysis for significant differences between two groups was performed by Welch’s t test, and Kruskal–Wallis test was used for analysis of three or more groups in bioinformatics analysis such as KEGG functional prediction analysis in the control and PGF groups.
## Evaluation of the PGF rat model
To determine whether the gut microbiota of PGF rats was depleted during the 4 weeks-modeling periods, the relative abundance of intestinal bacteria was quantified using quantitative polymerase chain reaction (qPCR) in our previous work [2]. The result showed that the relative abundance of intestinal bacteria in the intestine of rats was continuously depleted [4] during the antibiotic intervention. In the present study, PGF rats were evaluated from the perspective of gut microbial diversity. The alterations in body weight during antibiotic treatment are shown in Fig. 1 and Additional file 1: Fig. S1. There was no difference in relative weight between two groups (each group $$n = 3$$) during antibiotic treatment. Raw data of the body weight was attached in the Additional file 2.Fig. 1Body weight in relative values of the control and PGF groups ($$n = 3$$ rats per group) during antibiotic treatment. There was no difference in relative weight during antibiotic treatment. Each bar in the graph represents the mean ± SD of values 16S rRNA sequencing was performed on the faeces of the control and PGF groups to analyse the alpha diversity of the gut microbiota. As shown in Fig. 2A, the rarefaction curve of observed species (Sob) tends to flatten with the increase of tags, indicating that the sequencing depth of this study can cover much of the bacterial taxa in the samples, and the amount of data is reasonable. The Chao1 and Shannon indexes were used as indicators to reflect microbial richness and richness and evenness, respectively. As an alpha diversity index based on phylogenetic trees, PD-tree was used to reflect the microbial community diversity. The Chao1, Shannon and PD-tree indexes in the PGF groups were significantly lower ($P \leq 0.01$) than those in the control group (Fig. 2B–D), indicating that the alpha diversity of gut microbiota composition in the PGF groups was significantly reduced ($P \leq 0.01$). However, alpha diversity was not significantly different among PGF groups (Additional file 1: Tables S1, S2) except that the PD-tree index was significantly reduced in the PGF-4W group compared with the PGF-3W group (Additional file 1: Table S3).Fig. 2The rarefaction curve of Sob A and the Chao1 B, Shannon C and PD-tree D alpha diversity indexes were different between the control group and PGF groups in different modelling periods (each group $$n = 6$$). * $P \leq 0.05$, ***$P \leq 0.001$ versus the control group Unweighted Pair Group Method with Arithmetic Mean (UPGMA) clustering tree (Fig. 3A, D) and PCoA principal coordinate analysis (Fig. 3B, E) based on Bray–Curtis and weighted UniFrac distances were used to evaluate beta diversity using the Vegan package (version 2.5.3) in R. The results showd that there was significant separation between the control and the PGF groups. Welch’s t test of beta diversity (Fig. 3C, F) based on Bray–Curtis and weighted UniFrac distances was performed to calculate the distance index (degree of dissimilarity) of microbial composition to reflect the beta diversity of the control and PGF groups. The value range of the Bray–Curtis distance was from 0 to 1, that is, the types and abundances of the two groups were from completely consistent to totally distinct. The beta diversity of the PGF groups was significantly reduced compared with the control group. The information of reads, ASVs and tags were attached in the Additional files 3, 4, 5.Fig. 3UPGMA cluster diagram A and D, PCoA principal coordinate analysis B and E and the statistic analysis of beta diversity C and F based on the Bray–Curtis A–C and weighted UniFrac D–F distances in the control and PGF groups (each group $$n = 6$$). ** $P \leq 0.01$ versus the control group Overall, the 16S rRNA gene analysis revealed that the bacterial abundance and diversity of the gut microbiota decreased significantly after 1 week of antibiotic exposure in PGF rats.
## Gut microbiota composition in the PGF model
The faeces of the control and PGF groups were subjected to 16S rRNA sequencing and analysis. The relative abundance for each taxa were displayed using Krona (version 2.6), and taxa Venn diagrams were plotted against the abundance of ASVs (Fig. 4A). Based on the taxa annotation information, the number of tag sequences in each group at each taxonomic level (kingdom, phylum, class, order, family and genus) was counted in this study. The top 10 taxa in abundance at each taxonomic level (phylum, class, order and family) and top 12 taxa in abundance at the genus level were selected in present study. In all samples, Bacteroidetes was mainly represented by the class Bacteroidia; the order Bacteroidales; the families Bacteroidaceae, Prevotellaceae and Muribaculaceae; and the genus Bacteroides, Sediminibacterium, Prevotellaceae_NK3B31_group and Prevotellaceae_UCG-001. In parallel, the composition of Firmicutes in samples included diverse genera, such as Roseburia, Fusicatenibacter and Lachnospiraceae_NK4A136_group (Fig. 4). Compared with the control group, the proportions of taxa at all taxonomic levels changed in the PGF groups (Fig. 4B–F), while the proportion of taxa at all taxonomic levels among the PGF groups was similar, implying that the gut microbial richness and diversity of all PGF model rats decreased to varying degrees. Fig. 4Venn diagram A and the most abundant taxa at the phylum B, class C, order D, family E, and genus F levels between the control and PGF groups (each group $$n = 6$$). PGF-1W, PGF-2W, PGF-3W and PGF-4W: the first, second, third and fourth weeks of antibiotic exposure
## Indicator taxa of the PGF model
The dominant taxa at the genus level in the control and PGF groups (determined by the vegan version 2.5.3 package in R) are shown in Fig. 5A–D. The control group contained more indicator taxa than the PGF groups ($P \leq 0.05$). The LefSe [24] software (version 1.0) was used to screen the indicator taxa at each level (phylum, class, order, family, and genus). Figure 6A–B shows the difference taxa based on a linear discriminant analysis (LDA) value > 3.5, and the histogram shows the influence of significant distinct taxa at each level. Most bacterial genera were depleted in the PGF groups compared with the control group ($P \leq 0.01$). However, the relative abundance of Escherichia–Shigella, as an indicator genus in the PGF groups, was significantly higher than that in the control group. In addition, the relative abundance of Ralstonia and Bradyrhizobium in the PGF-1W, PGF-2W and PGF-3W groups (Fig. 5A–C), but not in the PGF-4W group (Fig. 5D), was significantly higher than that in the control group. The gut microbiota of the PGF-4W group was depleted at a wider scale. Fig. 5Comparative analysis of indicator genera in the control and PGF groups (each group $$n = 6$$). *Indicator* genera in two groups were identified by the P-value less than 0.01. PGF-1W (A), PGF-2W (B), PGF-3W (C) and PGF-4W (D): the first, second, third and fourth weeks of antibiotic exposureFig. 6LEfSe analysis of the control and PGF groups (each group $$n = 6$$) with the linear discriminant analysis (LDA) value > 3.5. A Cladogramof intestinal microbiota from phylum to genus. B Histogram of LDA value distribution
## Changes in CYP450 enzymes expression in the PGF model
As previously described, the gut microbiota and CYP450 co-contribute to the diverse metabolic landscape [26]. Many researchers have explored the impact of the gut microbiota on the metabolism of foreign substances using PGF model, often ignoring changes in host hepatic metabolic enzymes while establishing the model. In the present study, the relative expression was calculated as follows: the gray value of target protein band divided by the gray value of actin protein band from the same sample. There were obvious changes in the relative expression of hepatic metabolic enzymes in PGF group after 1 week of antibiotic exposure. Compared with the control group, the relative expression levels of CYP1A2, CYP2C19 and CYP2E1 were significantly upregulated in the PGF groups (Fig. 7A–G). Meanwhile, CYP2C9, CYP2D6 and CYP3A4 expression was not significantly different between the control and the PGF group. Overall, the hepatic metabolic enzyme expression of CYP1A2, CYP2C19, and CYP2E1 in PGF model rats were higher than those in the control group. Fig. 7Comparison of the relative protein expression of hepatic CYP450 enzymes between the control group and PGF group at week 1 (each group $$n = 3$$). A *Densitometric analysis* of protein band intensity from western blots B–G. Data are mean ± SD; *$P \leq 0.05$, **$P \leq 0.01$ versus the control group
## Intestinal mucosal barrier function of the PGF model
As indicators of intestinal mucosal permeability and inflammatory level, the DAO, ET and LEP levels were compared between the control and PGF groups. Specifically, DAO level indicates intestinal mucosal permeability [27], while the levels of ET and LEP indicates mucosal inflammation [28, 29]. As shown in Fig. 8 and Additional file 1: Tables S4–S6, there were no significant differences in serum ET and LEP among the groups, while DAO level increased significantly in the PGF-2W, PGF-3W and PEG-4W groups compared with the control group. These findings indicate that the level of inflammation did not change among the groups while intestinal mucosal permeability was increased in the PGF groups compared with the control group beginning at week 2. Holota et al. [ 30] found that long-term antibiotic exposure increased intestinal mucosal permeability, which was associated with reduced levels of short-chain fatty acids. Dysbiosis in gut microbiota due to antibiotic treatment led to an increase of gut mucosal permeability [30, 31], which may augment systemic exposure to xenobiotics. Hence, increased intestinal mucosal permeability may promote the absorption of oral drugs, thereby increasing drug exposure in the body. These results provide a new perspective on the differences in blood drug concentrations in the PGF rat model, that is, the effect of increased absorption of oral drug on its blood exposure. Fig. 8Serum DAO, ET and LEP levels in the control and PGF groups (each group $$n = 3$$). The data are presented as the mean ± SD; *$P \leq 0.05$ versus the control group
## Gut microbial function of the PGF model
To describe the overall metabolic landscape of gut microbiota metabolism, the bacteria in the samples were annotated with KEGG functional pathways using Tax4Fun (version 1.0). The river plot (Fig. 9A) shows the functional pathway abundance characteristics of the control and PGF groups. Each PGF group exhibited similar functional pathway abundance profiles. The statistical differences of functional pathways of the groups were analysed by vegan package (version 2.5.3) in R. The P-value threshold of 0.001 was used as a filter to display the functional prediction results. The predicted results of the gut microbiota function in the control and PGF groups are shown in Fig. 9B–C, and the functional pathways listed in Fig. 9C–D are all statistically different metabolic pathways. What is striking is that the pathway abundance of nucleotide metabolism, lipid metabolism, metabolism of other amino acids, and metabolism of terpenoids and polyketides showed significant differences between the control and PGF groups, implying that the metabolic enzymes involved in the metabolic pathways from the gut microbiota of the PGF groups were altered. Moreover, the abundance of metabolism of cofactors and vitamins and metabolism of terpenoids and polyketides decreased significantly in the PGF-4W group compared to the PGF-3W group (Fig. 9D).Fig. 9Functional prediction results of faecal microbiota in the control and PGF groups. A: river chart of functional abundance; B and C: heat maps of functional abundance of the control and PGF groups. D: statistical analysis of functional abundance of all groups
## Discussion
The gut microbiota and host hepatic CYP450s enzymes together have profound effects on the field of drug metabolism research [5, 6], where PGF models are widely used [4, 6–9]. However, the PGF model research is not sufficient, and the application of this model is not standardized and unified [6, 12, 19, 20]. Ampicillin sodium, neomycin sulfate, metronidazole and vancomycin hydrochloride have been used in many studies to deplete most intestinal bacteria and thus to establish PGF model. In our previous study [2], antibiotic cocktails could deplete the relative abundance of gut microbiota within one week. The present study discovered that richness and diversity of the gut microbiota decreased significantly beginning 1 week after antibiotic exposure (Fig. 2). PGF-4W model, submitted to antibiotics for 4 weeks, had the lowest dominant taxa, and the gut microbiota was depleted to a greater extent than the other PGF groups (Fig. 5). It seems that a modelling time of 4 weeks is the best choice for establishing the PGF model. However, beginning at week 2 of antibiotic exposure, DAO was significantly higher compared with the control group. Given that DAO is an indicator of intestinal permeability, the results suggest that intestinal permeability had altered (Fig. 8). Considering that long-term use of antibiotics can change intestinal permeability, to save time and cost, and to achieve the goal of reducing gut microbial abundance and diversity, 1 week of antibiotic treatment is best to establish the PGF model.
A study found that antibiotic treatment led to the perturbation of the gut microbiota in rats, which reduced the antitumor efficacy of 5-Fluorouracil on colorectal cancer [32]. Zhang et al. [ 33] confirmed that the hypoxic environment of the plateau could lead to changes in the number and composition of faecal microbiota, which demonstrates that the gut environment could change the intestinal microbiota and lead to individual differences in intestinal microbiota. These changes in the intestinal microbiota could alter drug metabolic activity in the body, resulting in increased systemic blood drug exposure and therapeutic efficacy of drug. A recent report showed that the abundance of *Faecalibacterium prausnitzii* in the feces of renal transplant patients was positively correlated with the oral tacrolimus dose. Incubation of *Faecalibacterium prausnitzii* with tacrolimus produced two compounds; however, these compounds were not observed when tacrolimus was incubated with liver microsomes [34]. Therefore, a comparative study of the gut microbiota composition between the control and PGF groups is crucial when using the PGF model to explore the effect of gut microbiota on differences in drug metabolism. Based on the presented analysis of taxonomic composition, there were significant differences in the distribution of intestinal bacteria at different taxonomic levels between the control and PGF groups and the control group (Fig. 4). The control group was mainly composed of Bacteroidetes and Firmicutes, while the PGF group mainly consisted of Proteobacteria. As indicator genera in the PGF groups, Escherichia–Shigella in the PGF groups had a higher relative abundance than that in the control group. In addition, at the family and genus level, the control group had greater richness and diversity (Figs. 2–5). However, in the process of studying the effect of gut microbial metabolism on differences in plasma drug concentration exposure, the effect of host hepatic CYP450s metabolic enzyme variations on differences in drug metabolism is often overlooked.
Together, the gut microbiota and host CYP450 enzymes establish the metabolic system of body, and these factors are important factors affecting pharmacokinetics. The expression of CYP450 varies among individuals due to genetic polymorphism, which has been reported to be associated with individual differences in the efficacy of many drugs [35, 36]. For example, the gut microbiota and expression of CYP450 enzymes were altered in non-alcoholic steatohepatitis model mice, thus jointly affecting the pharmacokinetic profiles [26]. Previous studies have demonstrated that CYP3A11 expression varied between in conventional and GF mice [17], which may affect the metabolism of related drugs. There may be a crosstalk between gut microbiota and hepatic drug metabolism. It is worth mentioning that the present study comprehensively considered changes in six host CYP450 that metabolism > $90\%$ of clinical drugs after establishing a PGF model [1]. The results revealed that the expression levels of CYP1A2, CYP2C19 and CYP2E1 were significantly increased in the PGF model (Fig. 7A–G), meaning that metabolic clearance of relevant drugs may be enhanced, possibly leading to changes in pharmacokinetic characteristics.
The gut microbiota has been increasingly recognized as an important but often neglected component of drug metabolism. For example, the importance of CYP3A4 and other host metabolic enzymes in irinotecan metabolism is highlighted by the significant correlation between irinotecan and the clearance of the CYP3A probe drug midazolam [37]. Nonetheless, recent studies have discovered that the gut microbiota plays an increasingly significant role in irinotecan metabolism and the toxic side effects of irinotecan diarrhea [38, 39]. To understand the role of the gut microbiota in drug metabolism and to compare and analyse the functions of the gut microbiota in the established PGF model, the function of the gut microbiota in all groups were predicted based on KEGG metabolic pathway annotation. The differential functional pathways are listed in Fig. 9. Of note, the abundance of nucleotide metabolism pathway was significantly enriched in the control group compared with the PGF groups, implying the metabolic enzymes involved in this pathway, such as hypoxanthine phosphoribosyl transferase, are more abundant than in the PGF model, which is verified by a recent study [4]. The result of this study showed that the blood level of the active metabolite 6-thioguanine nucleotide was significantly decreased in the PGF group compared with the control group [4], which may be related to the decrease in the abundance of related metabolic enzymes in the gut microbiota of the PGF groups. Furthermore, other metabolic pathways such as the terpenoids and polyketides metabolic pathways showed significant changes in the PGF groups, suggesting the dramatic changes occurred in the gut microbiota function in PGF rats compared with the control group.
Unfortunately, this study did not explore how the gut microbiota and host CYP450 affect individual differences in drug metabolism and their contribution to drug metabolism. More advanced experimental methods and more samples are needed to explore the xenobiotic metabolic ability of PGF model and more characteristic evidence of the PGF model.
*In* general, this study represents a comprehensive and systematic investigation on the characteristics of the gut microbiota and the host CYP450 enzymes in the PGF model. Given that the PGF model needs to reduce the relative abundance and diversity of intestinal bacteria, and should not affect intestinal mucosal permeability, 1 week treatment with an antibiotic cocktail is the best choice. Moreover, researchers need to take into account changes in host CYP450 and intestinal permeability when using the PGF model to study the influence of intestinal microbiota on the blood concentrations of drugs and their metabolites.
## Conclusion
The richness and diversity of the gut microbiota decreased significantly in the PGF groups beginning after 1 week of antibiotic exposure, and the PGF group submitted to antibiotics for 4 weeks had the least dominant taxa. In view of the changes in intestinal permeability beginning at week 2 of antibiotic exposure, 1 week of antibiotic administration is the best choice for establishing a PGF model. The composition and metabolic function of the gut microbiota, and the expression of CYP450 enzymes were drastically altered in the PGF model. These findings provide deep insight into the understanding of how oral drug acts in the control and PGF groups and may serve as a model reference for future studies.
## Supplementary Information
Additional file 1: Figure S1. Body weight in absolute values of control and PGF groups (3 rats in each group) during antibiotic treatment. Each bar in the graph represents the mean ± SD of values. Table S1. Statistical results of one-way ANOVA analysis of Chao1 index of alpha diversity among the four PGF groups. Table S2. Statistical results of one-way ANOVA analysis of Shannon index of alpha diversity among the four PGF groups. Table S3. Statistical results of one-way ANOVA analysis of Phylogenetic diversity (PD) tree index of alpha diversity among the four PGF groups. Table S4. Statistical results of one-way ANOVA analysis of serum DAO level among the control and PGF groups. Table S5. Statistical results of one-way ANOVA analysis of serum ET level among the control and PGF groups. Table S6. Statistical results of one-way ANOVA analysis of serum LEP level among the control and PGF groups. Additional file 2. Raw data of the body weight. Additional file 3. The number of reads per sample. Additional file 4. Statistical table of the number of ASVs and Tags in different samples. Additional file 5. Tags length.
## References
1. Tang L, Ye L, Lv C, Zheng Z, Gong Y, Liu Z. **Involvement of CYP3A4/5 and CYP2D6 in the metabolism of aconitine using human liver microsomes and recombinant CYP450 enzymes**. *Toxicol Lett* (2011) **202** 47-54. DOI: 10.1016/j.toxlet.2011.01.019
2. Gao B, Tan T, Cao X, Pan M, Yang C, Wang J. **Relationship of cytochrome P450 gene polymorphisms with blood concentrations of hydroxychloroquine and its metabolites and adverse drug reactions**. *BMC Med Genomics* (2022) **15** 23. DOI: 10.1186/s12920-022-01171-6
3. Doestzada M, Vila AV, Zhernakova A, Koonen DPY, Weersma RK, Touw DJ. **Pharmacomicrobiomics: a novel route towards personalized medicine?**. *Protein Cell* (2018) **9** 432-445. DOI: 10.1007/s13238-018-0547-2
4. Wang S, Qin Y, Wen Q, Xia Q, Gu R, Wang S. **Intestinal microbiota-mediated biotransformations alter the pharmacokinetics of the major metabolites of azathioprine in rats after oral administration**. *Drug Metab Pharmacokinet* (2022) **45** 100458. DOI: 10.1016/j.dmpk.2022.100458
5. Cussotto S, Walsh J, Golubeva AV, Zhdanov AV, Strain CR, Fouhy F. **The gut microbiome influences the bioavailability of olanzapine in rats**. *EBioMedicine* (2021) **66** 103307. DOI: 10.1016/j.ebiom.2021.103307
6. Liu M, Yuan J, Hu WJ, Ke CQ, Zhang YF, Ye Y. **Pretreatment with broad-spectrum antibiotics alters the pharmacokinetics of major constituents of Shaoyao-Gancao decoction in rats after oral administration**. *Acta Pharmacol Sin* (2019) **40** 288-296. DOI: 10.1038/s41401-018-0011-0
7. Kida H, Akao T, Meselhy MR, Hattori M. **Metabolism and pharmacokinetics of orally administered saikosaponin b1 in conventional, germ-free and Eubacterium sp. A-44-infected gnotobiote rats**. *Biol Pharm Bull* (1998) **21** 588-93. DOI: 10.1248/bpb.21.588
8. Newmark J, Brady RO, Grimley PM, Gal AE, Waller SG, Thistlethwaite JR. **Amygdalin (Laetrile) and prunasin beta-glucosidases: distribution in germ-free rat and in human tumor tissue**. *Proc Natl Acad Sci U S A* (1981) **78** 6513-6. DOI: 10.1073/pnas.78.10.6513
9. Wu B, Chen M, Gao Y, Hu J, Liu M, Zhang W. **In vivo pharmacodynamic and pharmacokinetic effects of metformin mediated by the gut microbiota in rats**. *Life Sci* (2019) **226** 185-192. DOI: 10.1016/j.lfs.2019.04.009
10. Guo YP, Chen MY, Shao L, Zhang W, Rao T, Zhou HH. **Quantification of Panax notoginseng saponins metabolites in rat plasma with in vivo gut microbiota-mediated biotransformation by HPLC-MS/MS**. *Chin J Nat Med* (2019) **17** 231-240. PMID: 30910060
11. Hernandez-Chirlaque C, Aranda CJ, Ocon B, Capitan-Canadas F, Ortega-Gonzalez M, Carrero JJ. **Germ-free and Antibiotic-treated Mice are Highly Susceptible to Epithelial Injury in DSS Colitis**. *J Crohns Colitis* (2016) **10** 1324-1335. DOI: 10.1093/ecco-jcc/jjw096
12. Wang Z, Klipfell E, Bennett BJ, Koeth R, Levison BS, Dugar B. **Gut flora metabolism of phosphatidylcholine promotes cardiovascular disease**. *Nature* (2011) **472** 57-63. DOI: 10.1038/nature09922
13. Seki E, De Minicis S, Osterreicher CH, Kluwe J, Osawa Y, Brenner DA. **TLR4 enhances TGF-beta signaling and hepatic fibrosis**. *Nat Med* (2007) **13** 1324-32. DOI: 10.1038/nm1663
14. Rakoff-Nahoum S, Paglino J, Eslami-Varzaneh F, Edberg S, Medzhitov R. **Recognition of commensal microflora by toll-like receptors is required for intestinal homeostasis**. *Cell* (2004) **118** 229-241. DOI: 10.1016/j.cell.2004.07.002
15. Kuno T, Hirayama-Kurogi M, Ito S, Ohtsuki S. **Effect of intestinal flora on protein expression of drug-metabolizing enzymes and transporters in the liver and kidney of germ-free and antibiotics-treated mice**. *Mol Pharm* (2016) **13** 2691-2701. DOI: 10.1021/acs.molpharmaceut.6b00259
16. Lee SH, An JH, Park HM, Jung BH. **Investigation of endogenous metabolic changes in the urine of pseudo germ-free rats using a metabolomic approach**. *J Chromatogr B Analyt Technol Biomed Life Sci* (2012) **887–888** 8-18. DOI: 10.1016/j.jchromb.2011.12.030
17. Han LW, Wang L, Shi Y, Dempsey JL, Pershutkina OV, Dutta M. **Impact of microbiome on hepatic metabolizing enzymes and transporters in mice during pregnancy**. *Drug Metab Dispos* (2020) **48** 708-722. DOI: 10.1124/dmd.120.000039
18. Kindt A, Liebisch G, Clavel T, Haller D, Hormannsperger G, Yoon H. **The gut microbiota promotes hepatic fatty acid desaturation and elongation in mice**. *Nat Commun* (2018) **9** 3760. DOI: 10.1038/s41467-018-05767-4
19. Liang W, Zhao L, Zhang J, Fang X, Zhong Q, Liao Z. **Colonization potential to reconstitute a microbe community in pseudo germ-free mice after fecal microbe transplant from equol producer**. *Front Microbiol* (2020) **11** 1221. DOI: 10.3389/fmicb.2020.01221
20. Hintze KJ, Cox JE, Rompato G, Benninghoff AD, Ward RE, Broadbent J. **Broad scope method for creating humanized animal models for animal health and disease research through antibiotic treatment and human fecal transfer**. *Gut Microbes* (2014) **5** 183-91. DOI: 10.4161/gmic.28403
21. Liu G, Wei C, Yuan S, Zhang Z, Li J, Zhang L. **Wogonoside attenuates liver fibrosis by triggering hepatic stellate cell ferroptosis through SOCS1/P53/SLC7A11 pathway**. *Phytother Res* (2022). DOI: 10.1002/ptr.7558
22. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. **DADA2: high-resolution sample inference from illumina amplicon data**. *Nat Methods* (2016) **13** 581-3. DOI: 10.1038/nmeth.3869
23. Pruesse E, Quast C, Knittel K, Fuchs BM, Ludwig W, Peplies J. **SILVA: a comprehensive online resource for quality checked and aligned ribosomal RNA sequence data compatible with ARB**. *Nucleic Acids Res* (2007) **35** 7188-96. DOI: 10.1093/nar/gkm864
24. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS. **Metagenomic biomarker discovery and explanation**. *Genome Biol* (2011) **12** R60. DOI: 10.1186/gb-2011-12-6-r60
25. Asshauer KP, Wemheuer B, Daniel R, Meinicke P. **Tax4Fun: predicting functional profiles from metagenomic 16S rRNA data**. *Bioinformatics* (2015) **31** 2882-2884. DOI: 10.1093/bioinformatics/btv287
26. Guo J, Xu Y, Chen LJ, Zhang SX, Liou YL, Chen XP. **Gut microbiota and host Cyp450s co-contribute to pharmacokinetic variability in mice with non-alcoholic steatohepatitis: effects vary from drug to drug**. *J Adv Res* (2022) **39** 319-332. DOI: 10.1016/j.jare.2021.10.004
27. Zhang Z, Xiang Y, Wang B, Chen H, Cai X, Wang X. **Intestinal mucosal permeability of children with cefaclor-associated serum sickness-like reactions**. *Eur J Pediatr* (2013) **172** 537-43. DOI: 10.1007/s00431-012-1926-y
28. Bolte LA, Vich VA, Imhann F, Collij V, Gacesa R, Peters V. **Long-term dietary patterns are associated with pro-inflammatory and anti-inflammatory features of the gut microbiome**. *Gut* (2021) **70** 1287-1298. DOI: 10.1136/gutjnl-2020-322670
29. Clemente-Postigo M, Oliva-Olivera W, Coin-Araguez L, Ramos-Molina B, Giraldez-Perez RM, Lhamyani S. **Metabolic endotoxemia promotes adipose dysfunction and inflammation in human obesity**. *Am J Physiol Endocrinol Metab* (2019) **316** E319-E332. DOI: 10.1152/ajpendo.00277.2018
30. Holota Y, Dovbynchuk T, Kaji I, Vareniuk I, Dzyubenko N, Chervinska T. **The long-term consequences of antibiotic therapy: role of colonic short-chain fatty acids (SCFA) system and intestinal barrier integrity**. *PLoS ONE* (2019) **14** e0220642. DOI: 10.1371/journal.pone.0220642
31. de la Cuesta-Zuluaga J, Mueller NT, Alvarez-Quintero R, Velasquez-Mejia EP, Sierra JA, Corrales-Agudelo V. **Higher fecal short-chain fatty acid levels are associated with gut microbiome dysbiosis, obesity, hypertension and cardiometabolic disease risk factors**. *Nutrients* (2018). DOI: 10.3390/nu11010051
32. Yuan L, Zhang S, Li H, Yang F, Mushtaq N, Ullah S. **The influence of gut microbiota dysbiosis to the efficacy of 5-fluorouracil treatment on colorectal cancer**. *Biomed Pharmacother* (2018) **108** 184-193. DOI: 10.1016/j.biopha.2018.08.165
33. Zhang J, Chen Y, Sun Y, Wang R, Zhang J, Jia Z. **Plateau hypoxia attenuates the metabolic activity of intestinal flora to enhance the bioavailability of nifedipine**. *Drug Deliv* (2018) **25** 1175-1181. DOI: 10.1080/10717544.2018.1469687
34. Guo Y, Crnkovic CM, Won KJ, Yang X, Lee JR, Orjala J. **Commensal gut bacteria convert the immunosuppressant tacrolimus to less potent metabolites**. *Drug Metab Dispos* (2019) **47** 194-202. DOI: 10.1124/dmd.118.084772
35. Nehlig A. **Interindividual differences in caffeine metabolism and factors driving caffeine consumption**. *Pharmacol Rev* (2018) **70** 384-411. DOI: 10.1124/pr.117.014407
36. Jiang XL, Samant S, Lesko LJ, Schmidt S. **Clinical pharmacokinetics and pharmacodynamics of clopidogrel**. *Clin Pharmacokinet* (2015) **54** 147-66. DOI: 10.1007/s40262-014-0230-6
37. Mathijssen RH, de Jong FA, van Schaik RH, Lepper ER, Friberg LE, Rietveld T. **Prediction of irinotecan pharmacokinetics by use of cytochrome P450 3A4 phenotyping probes**. *J Natl Cancer Inst* (2004) **96** 1585-92. DOI: 10.1093/jnci/djh298
38. Parvez MM, Basit A, Jariwala PB, Gaborik Z, Kis E, Heyward S. **Quantitative investigation of irinotecan metabolism, transport, and gut microbiome activation**. *Drug Metab Dispos* (2021) **49** 683-693. DOI: 10.1124/dmd.121.000476
39. Guthrie L, Gupta S, Daily J, Kelly L. **Human microbiome signatures of differential colorectal cancer drug metabolism**. *NPJ Biofilms Microbiomes* (2017) **3** 27. DOI: 10.1038/s41522-017-0034-1
|
---
title: 'Plantar pressure and falling risk in older individuals: a cross-sectional
study'
authors:
- Yifeng Yan
- Jianlin Ou
- Hanxue Shi
- Chenming Sun
- Longbin Shen
- Zhen Song
- Lin Shu
- Zhuoming Chen
journal: Journal of Foot and Ankle Research
year: 2023
pmcid: PMC10029259
doi: 10.1186/s13047-023-00612-4
license: CC BY 4.0
---
# Plantar pressure and falling risk in older individuals: a cross-sectional study
## Abstract
### Background
Falls are commonplace among elderly people. It is urgent to prevent falls. Previous studies have confirmed that there is a difference in plantar pressure between falls and non-falls in elderly people, but the relationship between fall risk and foot pressure has not been studied. In this study, the differences in dynamic plantar pressure between elderly people with high and low fall risk were preliminarily discussed, and the characteristic parameters of plantar pressure were determined.
### Methods
Twenty four high-fall-risk elderly individuals (HR) and 24 low-fall-risk elderly individuals (LR) were selected using the Berg Balance Scale 40 score. They wore wearable foot pressure devices to walk along a 20-m-long corridor. The peak pressure (PP), pressure time integral (PTI), pressure gradient (maximum pressure gradient (MaxPG), minimum pressure gradient (MinPG), full width at half maximum (FWHM)) and average pressure (AP) of their feet were measured for inter-group and intra-group analysis.
### Results
The foot pressure difference comparing the high fall risk with low fall risk groups was manifested in PP and MaxPG, concentrated in the midfoot and heel ($p \leq 0.05$), while the only time parameter, FWHM, was manifested in the whole foot ($p \leq 0.05$). The differences between the left and right foot were reflected in all parameters. The differences between the left and right foot in LR were mainly reflected in the heel ($p \leq 0.05$), while it in the HR was mainly reflected in the forefoot ($p \leq 0.05$).
### Conclusions
The differences comparing the high fall risk with low fall risk groups were mostly reflected in the midfoot and heel. The HR may have been more cautious when landing. In the intra-group comparison, the difference between the right and left foot of the LR was mainly reflected during heel striking, while it was mainly reflected during pedalling in the HR. The sensitivity of PP, PTI and AP was lower and the newly introduced pressure gradient could better reflect the difference in foot pressure between the two groups. The pressure gradient can be used as a new foot pressure parameter in scientific research.
## Background
The aging of the global population is becoming more and more serious, and with it the related problems [1]. Studies showed approximately $28\%$ of elderly people aged 65 and above fall every year [2, 3] and its incidence is higher among the older group. Falls occur with high frequency, great harmfulness and many sequelae, which bring a heavy burden to society [4, 5]. And fall fatalities have increased over the past decade [6]. As such, fall prevention among elderly people and reduction of fall rate is of great urgency.
The increased risk of falls is associated with an increment of gait variability [7]. The ability to walk is one of the most natural and basic forms of human movement and a prerequisite for independent human activity and self-care. With age growing, gait changes because of the alteration of balance control, degeneration of the musculoskeletal system, and diminished sensorimotor function. As gait changes, plantar pressure will change correspondingly. Therefore, the plantar pressure in walking is often used to study normal and abnormal gait characteristics. Plantar pressure has now been applied widely in studies related to falls in elderly people. The plantar center of pressure (COP) trajectory is the most widely used. Estevez-Pedraza presents a statistical model for estimating fall risk from the COP data [8]. Also, studies by Pizzigalli concluded that certain swaying characteristics of silent posture, particularly in the medial-lateral direction, are significantly different from those of non-falling and falling people [9]. Muir's study indicated that COP displacement was significantly worse in those who fell [10]. Although a recent systematic review concluded that certain COP measures may be linked to a fall in certain circumstances rather than others [11], the force platform parameters may indeed be useful for fall risk prediction [12, 13]. However, the aforementioned studies have focused on static balance posture control and lacked studies on dynamic gait parameters, with only Mickle showing that compared to non-fallers, fallers featured significantly higher peak pressures and pressure time integrals [14]. Pol found the fallers had greater medial midfoot, medial forefoot, and bunion loading [15]. Mickle and Pol have used the "two-step method" to test plantar pressure, with equipment limited to the laboratory.
In addition, we found that previous studies have focused on falls between fallers and non-fallers when a primary injury from a fall has already developed. Whereas the most important thing against falls is the prevention of falls, there are few studies on the risk and foot pressure before falls. Plantar pressure is an essential feature during walking, and it has a good potential to improve our awareness before a fall, and future wearable wireless plantar pressure devices will be more convenient [16]. Therefore, based on the wearable intelligent footwear system [17–20], a preliminary study was made on plantar pressure in elderly people with a high or low risk of falling.
The two aims of this study were to investigate whether there are differences in dynamic plantar pressure among elderly people at high and low risk of falls; if there are differences, we tried to search for the plantar pressure characteristic parameters.
## Methods
A cross-sectional study was applied.
## Participants
Participants were recruited from January 2021 to May 2022 at the Department of Rehabilitation Medicine of the First Affiliated Hospital of Jinan University, Guangzhou, China. Participants were primarily outpatient follow-up patients and hospital attendants. Recruitment announcements were posted on the department bulletin boards and outpatient department. Participants of interest were screened according to inclusion and exclusion criteria, and those who met the requirements were invited to participate in the study.
Inclusion criteria: (a) age 65 or older; (b) capable of independent walking for 3 min without assistance; (c) clear consciousness, able to cooperate with the assessment, Mini-mental State Examination score > 24; (d) with available informed consent. Exclusion criteria: (a) Those with foot injuries, deformities and other conditions that temporarily affect their daily walking; (b) People receiving trunk or lower limb therapy that affect lower extremity biomechanics; (c) patients with serious or unstable cardiac, pulmonary, renal and other medical diseases who can not tolerate the study; (d) patients with a history of mania, delirium and other psychiatric disorders who cannot cooperate to complete the test.
According to previous study [15] and pre-experiment results, it is expected that the combined sample standard deviation σ is 1.36, and the difference between the two groups' means δ is 1.1. Bilateral α = 0.05 is set, and power (1-β) is $80\%$. The sample size was calculated according to the formula [1], and $$n = 24$$ were obtained. So each group requires 24 samples.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=\frac{2{\left({z}_{\alpha }+{z}_{\beta }\right)}^{2}*{\sigma }^{2}}{{\delta }^{2}}$$\end{document}$$n = 2$$zα+zβ2∗σ2δ2 Finally, forty-eight elderly people were selected for general data collection and Berg Balance Scale (BBS) assessment, of whom 24 BBS scored ≤ 40 for the high fall risk group (HR) and the other 24 BBS scored > 40 for the low fall risk group (LR). *The* general data includes sex, age, height, body weight and body mass index.
## Apparatus and equipment
The plantar pressure was detected using a wearable intelligent footwear system developed in cooperation with the Human Data Science Engineering Center of South China University of Technology and Zhongshan Super Sense Technology Co [19, 20]. As shown in Fig. 1, there are eight pressure sensors at different points in the insole of each foot. The pressure sensor has the characteristics of a short response time, large range, high sensitivity, and high wear resistance [17]. Wirelessly connected mobile phones can receive real-time datums collected by the footwear system. And the phone APP sets different colours according to the pressure grading, indicating the dynamic changes of plantar pressure, as shown in Fig. 2. It was confirmed that satisfactory accuracy, repeatability and wearing comfort was showed by this intelligent footwear system [18].Fig. 1Position of eight pressure sensors. The right foot is shown as an exampleFig. 2Different colours according to the pressure grading. During the test, the color will change dynamically with the pressure
## Procedure
During the experiment, the researcher provided uniform cotton socks and the participants were asked to choose the right size of socks and intelligent shoes to ensure that their feet would not slide in the shoes while walking. Before the experiment began, the participants wore cotton socks and intelligent shoes for 1-2 min to adapt. During the formal experiment, participants need to walk for more than two minutes along a 20-m corridor at their normal gait and usual walking speed. The experiments were supervised by one investigator, with no physical contact or verbal induction. Each experiment was supervised by the same investigator. And participants went through the whole process in one day.
## Observation criteria and data extraction
Based on previous studies combined with plantar mechanics, the plantar area was divided into 8 regions for analysis: 1st toe (T1), 1st metatarsal head (M1), 2nd-3rd metatarsal head (M2-3), 4th-5th metatarsal head (M4-5), medial midfoot (MMF), lateral midfoot (LMF), medial heel (MH), and lateral heel (LH), as shown in Fig. 3.Fig. 3Eight regions for analysis. These 8 regions correspond to the position of the pressure sensor one by one The raw plantar pressure data is exported from the smart terminal mobile APP background. After weight normalization and identification of valid gait cycles, for each foot region, we calculate the following parameters: peak pressure (PP), pressure-time integral (PTI), pressure gradient (maximum pressure gradient (MaxPG), minimum pressure gradient (MinPG), full width at half maximum (FWHM)), and average pressure (AP). In this study, we used the characteristic calculation method of Botros [21] and Dongran Wang [17] et al.
For each plantar region i, where $i = 1$, 2,…, 16 and i ∈ [1-8] for the left foot and i ∈ [9-16] for the right foot, Pi(m) is the pressure datums for each sample suiting to the region, where $m = 1$,2,…, M. M is the length of the sample data for one valid gait cycle. Then for each Pi(m), calculate 12 feature Fri and perform the average of individual effective gait cycles, where $r = 1$, 2,…, 12.
Taking the left foot as an example, with r ∈ [1-6], the 1st parameter PP can be calculated by Eq. [ 2].2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{F}}_{1\mathrm{i}}{= }_{\mathrm{m}\in \left[1,\mathrm{M}\right]}^{\mathrm{ max}}{\mathrm{P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L}}$$\end{document}F1i=m∈1,MmaxPim|L The 2nd parameter PTI can be calculated by Eqs. [ 3].3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{F}}_{2\mathrm{i}}= {\sum }_{\mathrm{m}=1}^{\mathrm{M}-1}{(\mathrm{P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L}}+{\mathrm{P}}_{\mathrm{i}}(\mathrm{m}+1){|}_{\mathrm{L}}).\Delta \mathrm{m}/2$$\end{document}F2i=∑$m = 1$M-1(Pim|L+Pi(m+1)|L).Δm/2 The 3rd parameter MaxPG and the 4th parameter MinPG are calculated by Eqs. [ 4] and [5], respectively.4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{F}}_{3\mathrm{i}}{= }_{\mathrm{m}\in \left[1,\mathrm{M}\right]}^{\mathrm{ max}}{\nabla \mathrm{P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L}}{= }_{\mathrm{m}\in \left[1,\mathrm{M}\right]}^{\mathrm{ max}}{ [\mathrm{ P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L }}{ -\mathrm{ P}}_{\mathrm{i}}\left(\mathrm{m}-1\right){|}_{\mathrm{L}}]/\Delta \mathrm{m}$$\end{document}F3i=m∈1,Mmax∇Pim|L=m∈1,Mmax[Pim|L-Pim-1|L]/Δm5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{F}}_{4\mathrm{i}}{= }_{\mathrm{m}\in \left[1,\mathrm{M}\right]}^{\mathrm{ min}}{\nabla \mathrm{P}}_{\mathrm{i}}(\mathrm{m}){|}_{\mathrm{L}}{= }_{\mathrm{m}\in \left[1,\mathrm{M}\right]}^{\mathrm{ min}}{ [\mathrm{ P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L }}{ -\mathrm{ P}}_{\mathrm{i}}(\mathrm{m}-1){|}_{\mathrm{L}}]/\Delta \mathrm{m}$$\end{document}F4i=m∈1,Mmin∇Pi(m)|L=m∈1,Mmin[Pim|L-Pi(m-1)|L]/Δm The 5th parameter FWHM can be calculated by Eqs. [ 6] and [7], where mi1 and mi2 denote the longness of the sample datums when the pressure value is half of the PP.6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{F}}_{5\mathrm{i}}={\mathrm{m}}_{\mathrm{i}2}{|}_{\mathrm{L}}{ -\mathrm{ m}}_{\mathrm{i}1}{|}_{\mathrm{L}}$$\end{document}F5i=mi2|L-mi1|L7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{P}}_{\mathrm{i}}{(\mathrm{m}}_{\mathrm{i}2}{)|}_{\mathrm{L}}={\mathrm{P}}_{\mathrm{i}}{(\mathrm{m}}_{\mathrm{i}1}{)|}_{\mathrm{L}}=0.5\times {\mathrm{F}}_{1\mathrm{i}}$$\end{document}Pi(mi2)|L=Pi(mi1)|$L = 0.5$×F1i
The sixth parameter AP can be calculated by Eq. [ 8].8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{{\mathrm{F} }_{6\mathrm{i}}}=\frac{1}{\mathrm{m}}\sum_{\mathrm{r}=1}^{\mathrm{m}}{\mathrm{P}}_{\mathrm{i}}\left(\mathrm{m}\right){|}_{\mathrm{L}}$$\end{document}F6i¯=1m∑$r = 1$mPim|L
## Statistical analysis
Except for parameter calculations, all data were processed by IBM SPSS Statistics for Windows, Version 27.0 (IBM Corp, Armonk, NY, USA). Data were expressed as mean ± SD or median (Q1, Q3), as appropriate. Also, group comparison was made using two independent sample t-test or Mann-Whitney U test. Besides, paired-sample t-test and Wilcoxon test were made for within-group comparisons. The significance level was set at α = 0.05.
## Ethical considerations
Approval was granted by the Medical Ethics Committee of the First Affiliated Hospital of Jinan University on December 23, 2020 (KY-2020-087). All participants were informed of the study purpose, procedure, anonymity and confidentiality of participation, and received written informed consent.
## Baseline data
The baseline data, including sex, age, height, body weight and body mass index of the two groups were compared, and the differences were not statistically significant ($p \leq 0.05$). As shown in Table 1.Table 1Comparison of baseline information of high and low fall risk groupsGroupNumber of casesSex (cases)AgeHeightBody weightBody mass indexmenwomen(age, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overset-{\mathrm x}$$\end{document}x- ± s)(cm, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overset{-}{\mathrm x}$$\end{document}x- ± s)(kg, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overset-{\mathrm x}$$\end{document}x- ± s)(kg/m2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overset-{\mathrm x}$$\end{document}x- ± s)LR24101472.63 ± 5.97159.88 ± 7.7760.17 ± 8.4323.55 ± 2.91HR2415976.33 ± 6.96163.38 ± 8.6862.46 ± 8.5023.41 ± 2.88P value0.1490.0540.1480.3530.875LR Low fall risk group, HR High fall risk group
## Peak pressure
Between-group comparison: Compared with the LR, the PP was reduced in the left-MMF, left-LH and right-MH in the HR ($p \leq 0.05$). Within-group comparison: In the LR, differences comparing the left foot with the right foot were shown in the LH and MMF ($p \leq 0.05$). In the HR, it was shown in the MMF and the T1 ($p \leq 0.05$), as shown in Fig. 4.Fig. 4Between-group and within-group effects of PP in elderly people with high and low fall risk. Plot by the median (Q1,Q3). 1where significant at $p \leq .05$ for pared-samples T-test or Wilcoxon paired test. # where significant at $p \leq .05$ for independent-samples T-test or Mann-Whitney U test
## Pressure–time integral
Within-group comparison: In the LR, differences comparing the left foot with the right foot were shown in the LH and MMF ($p \leq 0.05$). In the HR, it was shown in the T1, the M2-3 and the MMF ($p \leq 0.05$) (Fig. 5). Comparison between groups: none of them was statistically different. Fig. 5Between-group and within-group effects of PTI in elderly people with high and low fall risk. Plot by the median (Q1,Q3). 1where significant at $p \leq .05$ for pared-samples T-test or Wilcoxon paired test
## Full width at half maximum
Comparison between groups: Compared with the LR, the FWHM increased in elderly people in the HR in the following regions ($p \leq 0.05$): the left-T1, the left-M1, the left-M2-3, the left-M4-5, the left-LMF, the right-M1, the right-M2-3, the right-M4-5, the right-LMF, the right-MH, the right-LH.
Within-group comparison: in the LR, the difference comparing the left foot with the right foot was shown on the LH ($p \leq 0.05$). In the HR, it was shown at the M1 ($p \leq 0.05$), as shown in Fig. 6.Fig. 6Between-group and within-group effects of FWHM in elderly people with high and low fall risk. Plot by the median (Q1,Q3). 1where significant at $p \leq .05$ for pared-samples T-test or Wilcoxon paired test. # where significant at $p \leq .05$ for independent-samples T-test or Mann-Whitney U test
## Maximum pressure gradient
Between-group comparison: Compared with the LR, the MaxPG in elderly people in the HR decreased in the following regions ($p \leq 0.05$): left-LMF, left-MH, left-LH, right-LMF, right-MH, and right-LH.
Within-group comparison: In the LR, differences comparing the left foot with the right foot were shown in the T1 and LMF ($p \leq 0.05$). In the HR, it was shown in the T1 and the M2-3 ($p \leq 0.05$), as shown in Fig. 7.Fig. 7Between-group and within-group effects of MaxPG in elderly people with high and low fall risk. Plot by the median (Q1,Q3). 1where significant at $p \leq .05$ for pared-samples T-test or Wilcoxon paired test. # where significant at $p \leq .05$ for independent-samples T-test or Mann-Whitney U test
## Minimum pressure gradient
Within-group comparison: only in the LR, the left and right feet of the M4-5 showed differences ($p \leq 0.05$). None of the between-group comparisons was statistically different.
## Average pressure
Within-group comparison: in the LR, differences comparing the left foot with the right foot were shown in the LH and MMF ($p \leq 0.05$). As shown in Fig. 8, in the HR, it was shown in the T1, M2-3 and MMF ($p \leq 0.05$). The difference between groups was not statistical. Fig. 8Between-group and within-group effects of AP in elderly people with high and low fall risk. Plot by the median (Q1,Q3). 1where significant at $p \leq .05$ for pared-samples T-test or Wilcoxon paired test
## Discussion
In this paper, pressure gradients are first introduced for foot pressure analysis, and it is clear from our results that PP, PTI and AP are less sensitive in high and low fall risk studies, while the spatial variation in pressure expressed by pressure gradients can rather better reflect the difference in plantar pressure on intergroup comparison. This paper prospectively investigates the relationship between fall risk and plantar pressure. The results showed that the differences between group comparison were mainly focused on midfoot and heel, so we assume that HR may have been more cautious when landing. As for the comparison within the group, the differences between the left and right foot in LR occurred mainly when the foot heel contacted the ground, while it in HR occurred mainly when the foot stroked the ground. In the within-group comparisons between the two groups, the differences in foot pressure were in different areas.
Walking is a fundamental function of human activity, and the walking ability reflects the physiological, structural, and functional state of the lower limbs and even the whole body. Plantar pressure refers to the amount of pressure exerted on the sole during standing or walking. It can be used to reflect the walking condition. Currently, the technique of foot pressure measurement has found wide application in clinical evaluation and scientific experiment, among which the most commonly used is the plantar pressure measurement platform, but the plantar pressure measurement platform is limited to the laboratory, which has a high demand for the measurement environment. In this study, we adopted a wearable intelligent footwear system developed by a research group in cooperation, which was designed with a footwear system that can be monitored in real-time during daily life, which is convenient, applicable to various living environments, conducive to further promotion and utilization and does not bring extra burden to elderly people.
## Peak pressure and pressure–time integral
The PP is one of the most commonly used variables to indicate plantar load, representing the maximum pressure value during the contact phase. In our study, the PP was smaller in the HR compared to the LR. One study showed that the PP was smaller in elderly people with falls [22]. It may be due to the slower walking velocity of elderly people [23–26] and possibly their activation of physical protection mechanisms. The differences between the two groups were concentrated on the heel both the medial and lateral, most likely because when elderly people contact the ground with the heel close to the body’s gravity center, the foot tends to be vertical [24, 27–29], thus reducing the impact force on the foot when landing [29]. These effects may be more pronounced in HR.
The PTI is also a variable frequently used to assess plantar load. It represents the accumulated effect of plantar pressure over time and can be simply interpreted as the product of pressure and contact time, which reflects the total plantar load contact value during the walking cycle. In our study, no significant differences were reflected between HR and LR, probably because PTI is a measure of the area size of the foot pressure curve over a walking period. To eliminate the influence of individual unnatural data, we adopted the calculation method of taking the average of multiple gait cycles, which may also filter out some extremely minimal differences. Even if the effective sample size is greater than 20 gait cycles, it may not accurately reflect the differences between the two groups [30]. This may also be the reason why the AP did not differ significantly between the HR and LR. Some studies have pointed out that there is a difference in the PTI between fallers and non-fallers [14, 15], but in our study no difference was shown in the history of falls between the two groups, suggesting that the difference comparing the HR with the LR may be smaller than the difference between falls and non-falls. High fall risk is not equivalent to fall history, which is the reason why we introduced a new foot pressure parameter, pressure gradient.
## Pressure gradient
The pressure gradient is a new index introduced in this study for foot pressure studies in elderly people, which was previously widely used in clinical studies of plantar pressure [31, 32]. Mueller pioneered the concept of pressure gradient, which suggests that higher pressure gradients, namely greater pressure changes in adjacent areas of the foot surface, are more damaging to plantar soft tissues [33]. Pressure gradients can be useful indicators of skin vulnus because spacial variations in high plantar pressure can recognize high concentrations of stress within the soft tissues. Theoretically, the authors believe that the introduced pressure gradient parameter is more specific and detailed, and better reflects the subtle differences comparing the HR with the LR. Our study calculated three parameters related to pressure gradients: MaxPG, MinPG and FWHM [21].
There was a difference in the MaxPG with the HR being significantly lower than the LR. This contradicts the conventional understanding that HR has a greater chance of soft tissue damage and a larger MaxPG. However, our results showed that the MaxPG was smaller in the HR, while the FWHM was found to be significantly higher. Because the FWHM refers to the difference between the two-time points at which half of the PP is reached, it indicates the time it takes for the participants to rise from the half-peak pressure to the PP and then to fall to the half-peak pressure. The significant increase in the HR indicates that the HR walking process is slower and the entire process is kept for a longer time, while the MaxPG of the HR is significantly lower than that of the LR, which further proves that elderly people with high fall risk will be more cautious during the walking process and their exposure to spatial variation of pressure will be smaller. This may also suggest that elderly people at high risk of falls do not necessarily have a higher likelihood of plantar soft tissue injury, which can be studied further and deeper later.
## Within-group comparison
As for the comparison of left and right feet in each index group, it can be found regardless of the high or low fall risk group, there was a difference between their left and right feet, and the asymmetry of both feet led to a decrease in their balance ability, which is in line with previous studies [24, 34]. Our study found that the differences comparing the left foot with the right foot in the LR were concentrated at the midfoot and heel, while those in the HR were concentrated at the anterior metatarsal head. It can be simply interpreted that the difference between the feet in the LR occurred when the foot contacted the ground with the heel, and it in the HR occurred when the foot pedalled. This may be related to changes in the toe grip force of the dominant foot in the older [35]. It also may be related to the enhanced toe flexion of the long and short toe flexors. Their enhanced toe flexion could compensate for reduced plantar fascia function in the 2nd and 3rd metatarsals, enhance forefoot stirrups, and increase proprioceptive feedback to the plantar side of the foot, improving postural stability in elderly people [36]. We also have the right to assume that the change in the walking style of elderly people occurs during the heel landing in the early stages of fall risk, and that the forefoot pedalling style changes when the fall risk progresses to a certain level.
## Between-group comparison
The difference in foot pressure between HR and LR is mainly the reduction of HR reflected in the midfoot and heel by PP and MaxPG (FWHM belongs to a time parameter), which is different from similar studies conducted by previous researchers. Mickle believed that the PP of non-fallers was significantly lower than that of fallers [14], but it is aimed at the total plantar pressure value and is related to foot pain, In our study, all the subjects had no foot pain, and there was no difference in the fall history between our two groups of subjects. Secondly, we needed to wear socks and shoes during the experiment, while Mickle's experiment was barefoot walking. And walking barefoot exhibited different plantar pressure than walking in shoes [25]. Whereas Pol found that fallers had higher PTI in the medial foot compared to non-fallers [15]. Brenton-Rule also found higher plantar pressures in fallers among adults with rheumatoid arthritis [37]. A recent study also showed that an increase in PTI is associated with falling fears [38]. However, previous studies focused on dividing experimental groups based on fall history as fall risk, and most of the experimental methods adopted the "one-step method" or "two-step method" on the force measurement platform, which is different from the experimental design of this study. In this study, the fall risk scale was adopted, and the walking state in daily life was restored as much as possible in the experiment, which is also the innovation of this study.
## Limitation
This study also has certain limitations. Firstly, when including subjects, we simply excluded subjects with obvious foot diseases that would affect daily walking, without considering the subjects' foot structure in detail, such as flat feet, etc., which may affect the results of plantar pressure. Secondly, our limited sample sizes fail to ensure a fully representative and widespread conclusion. Yet, it’s underpinned by medical theoretical knowledge, which could offer some clues to the research. Thirdly, Plantar pressure was not standardized by shoe size. Although most of our subjects adopted shoe size 37 for women and 41 for men, and our analysis was a region-specific analysis, our research group believes that the relationship between shoe size and foot pressure needs further investigation. This may also be another research direction for us in the future.
## Conclusion
These results preliminary suggest that there were indeed differences in plantar pressure between high and low fall risk in older adults and that plantar pressure may be used to determine fall risk, especially pressure gradient. In the subsequent fall prevention studies, we believe that focusing on the prior study, that is, starting from the fall risk, rather than distinguishing whether a people falls from the history of falls can fundamentally solve a series of problems caused by falls in elderly people. Meanwhile, the successful use of pressure gradient also prompts us to pay attention to the analysis and application of parameters reflecting small changes in plantar pressure, especially in real studies with small differences.
## References
1. 1.WHO. WHO global report on falls prevention in older age. Geneva,Switzerland. 2007. https://apps.who.int/iris/bitstream/handle/10665/43811/9789241563536_kor.pdf;jsessionid=6273F0199914A9F9EF6E7A0EF9428FD7?sequence=2.
2. Moreland B, Kakara R, Henry A. **Trends in nonfatal falls and fall-related injuries among adults aged >= 65 years - united states, 2012–2018**. *Mmwr-Morbidity and Mortality Weekly Report* (2020.0) **69** 875-881. DOI: 10.15585/mmwr.mm6927a5
3. Bergen G, Stevens MR, Burns ER. **Falls and fall injuries among adults aged >= 65 years - united states, 2014**. *Mmwr-Morbidity and Mortality Weekly Report* (2016.0) **65** 993-998. DOI: 10.15585/mmwr.mm6537a2
4. Burns ER, Stevens JA, Lee R. **The direct costs of fatal and non-fatal falls among older adults - United States**. *J Safety Res* (2016.0) **58** 99-103. DOI: 10.1016/j.jsr.2016.05.001
5. Florence CS, Bergen G, Atherly A, Burns E, Stevens J, Drake C. **Medical Costs of Fatal and Nonfatal Falls in Older Adults**. *J Am Geriatr Soc* (2018.0) **66** 693-698. DOI: 10.1111/jgs.15304
6. Global MM. **regional, and national burden of diseases and injuries for adults 70 years and older: systematic analysis for the Global Burden of Disease 2019 Study**. *Bmj-British Med J* (2019.0) **2022** 376. DOI: 10.1136/bmj-2021-068208
7. Toebes MJ, Hoozemans MJ, Furrer R, Dekker J, van Dieen JH. **Local dynamic stability and variability of gait are associated with fall history in elderly subjects**. *Gait Posture* (2012.0) **36** 527-531. DOI: 10.1016/j.gaitpost.2012.05.016
8. Estevez-Pedraza AG, Parra-Rodriguez L, Martinez-Mendez R, Portillo-Rodriguez O, Ronzon-Hernandez Z. **A novel model to quantify balance alterations in older adults based on the center of pressure (CoP) measurements with a cross-sectional study**. *PLoS ONE* (2021.0) **16** e0256129. DOI: 10.1371/journal.pone.0256129
9. Pizzigalli L, Micheletti Cremasco M, Mulasso A, Rainoldi A. **The contribution of postural balance analysis in older adult fallers: A narrative review**. *J Bodyw Mov Ther* (2016.0) **20** 409-417. DOI: 10.1016/j.jbmt.2015.12.008
10. Muir JW, Kiel DP, Hannan M, Magaziner J, Rubin CT. **Dynamic Parameters of Balance Which Correlate to Elderly Persons with a History of Falls**. *PLoS ONE* (2013.0) **8** e70566. DOI: 10.1371/journal.pone.0070566
11. Quijoux F, Vienne-Jumeau A, Bertin-Hugault F, Lefevre M, Zawieja P, Vidal PP. **Center of pressure characteristics from quiet standing measures to predict the risk of falling in older adults: A protocol for a systematic review and meta-analysis**. *Syst Rev* (2019.0) **8** 232. DOI: 10.1186/s13643-019-1147-9
12. Oliveira MR, Vieira ER, Gil AWO, Fernandes KBP, Teixeira DC, Amorim CF. **One-legged stance sway of older adults with and without falls**. *PLoS ONE* (2018.0) **13** e0203887. DOI: 10.1371/journal.pone.0203887
13. Kozinc Z, Loefler S, Hofer C, Carraro U, Sarabon N. **Diagnostic Balance Tests for Assessing Risk of Falls and Distinguishing Older Adult Fallers and Non-Fallers: A Systematic Review with Meta-Analysis**. *Diagnostics* (2020.0) **10** 667. DOI: 10.3390/diagnostics10090667
14. Mickle KJ, Munro BJ, Lord SR, Menz HB, Steele JR. **Foot pain, plantar pressures, and falls in older people: A prospective study**. *J Am Geriatr Soc* (2010.0) **58** 1936-1940. DOI: 10.1111/j.1532-5415.2010.03061.x
15. Pol F, Forghany S, Hosseini SM, Taheri A, Menz HB. **Structural and functional foot and ankle characteristics associated with falls in older people**. *Gait Posture* (2021.0) **88** 78-83. DOI: 10.1016/j.gaitpost.2021.05.012
16. Niu J, Zheng Y, Liu H, Chen X, Ran L. **Stumbling prediction based on plantar pressure distribution**. *Work (Reading, Mass)* (2019.0) **64** 705-712. DOI: 10.3233/wor-193032
17. Dongran W, Ji O, Peiru Z, Junlan Y, Lin S, Xiangmin X. **A novel low-cost wireless footwear system for monitoring diabetic foot patients**. *IEEE Trans Biomed Circuits Syst* (2021.0) **15** 43-54. DOI: 10.1109/TBCAS.2020.3043538
18. 18.Shu L, Mai KY, Tao XM, Li Y, Wong WC, Lee KF, et al. Monitoring Diabetic Patients by Novel Intelligent Footwear System. International Conference on Computerized Healthcare (ICCH). Hong Kong, PEOPLES R CHINA2012. p. 95-+. https://www.webofscience.com/wos/alldb/full-record/WOS:000350201500019.
19. 19.Shibin W, Jianlin O, Lin S, Guohua H, Zhen S, Xiangmin X, et al. Mhnet: Multi-scale spatio-temporal hierarchical network for real-time wearable fall risk assessment of the elderly. Computers in biology and medicine. 2022;144:105355 10.1016/j.compbiomed.2022.105355.
20. Zhen S, Jianlin O, Lin S, Guohua H, Shibin W, Xiangmin X. **Fall Risk Assessment for the Elderly Based on Weak Foot Features of Wearable Plantar Pressure**. *IEEE Trans Neural Syst Rehabil Eng* (2022.0) **30** 1060-1070. DOI: 10.1109/tnsre.2022.3167473
21. 21.Botros FS, Taher MF, ElSayed NM, Fahmy AS, Ieee. Prediction of diabetic foot ulceration using spatial and temporal dynamic plantar pressure. in 8th Cairo International Biomedical Engineering Conference (CIBEC),Cairo, EGYPT,. 2016;pp:43–7. https://www.webofscience.com/wos/alldb/full-record/WOS:000400192500011.
22. Yan C, Wei S, Fan Y, Xin-fa G. **Plantar pressure characteristics of elderly people with fall histories in different walking conditions**. *J Shandong Inst Phys Educ Sports* (2014.0) **30** 57-60. DOI: 10.14104/j.cnki.1006-2076.2014.03.011
23. Kernozek TW, Lamott EE. **Comparisons of plantar pressures between the elderly and young adults**. *Gait Posture* (1995.0) **3** 143-148. DOI: 10.1016/0966-6362(95)99064-R
24. Michel-Pellegrino V, Hewson D, Hogrel J-Y, Duchene J. **Biomechanical analysis of older adults stepping up: A method of evaluating balance**. *J Aging Phys Act* (2008.0) **16** 171-187. DOI: 10.1123/japa.16.2.171
25. Burnfield JM, Few CD, Mohamed OS, Perry J. **The influence of walking speed and footwear on plantar pressures in older adults**. *Clin Biomech (Bristol, Avon)* (2004.0) **19** 78-84. DOI: 10.1016/j.clinbiomech.2003.09.007
26. Fernandes WC, Machado A, Borella C, Carpes FP. **Influence of gait speed on plantar pressure in subjects with unilateral knee osteoarthritis**. *Rev Bras Reumatol* (2014.0) **54** 441-445. DOI: 10.1016/j.rbr.2014.06.004
27. Bennetts CJ, Owings TM, Erdemir A, Botek G, Cavanagh PR. **Clustering and classification of regional peak plantar pressures of diabetic feet**. *J Biomech* (2013.0) **46** 19-25. DOI: 10.1016/j.jbiomech.2012.09.007
28. 28.Mille ML, Creath RA, Prettyman MG, Johnson Hilliard M, Martinez KM, Mackinnon CD, et al. Posture and locomotion coupling: A target for rehabilitation interventions in persons with parkinson’s disease. Parkinsons Dis. 2012;2012:754186. 10.1155/2012/754186.
29. Simpson KM, Munro BJ, Steele JR. **Effects of prolonged load carriage on ground reaction forces, lower limb kinematics and spatio-temporal parameters in female recreational hikers**. *Ergonomics* (2012.0) **55** 316-326. DOI: 10.1080/00140139.2011.642004
30. McClymont J, Savage R, Pataky TC, Crompton R, Charles J, Bates KT. **Intra-subject sample size effects in plantar pressure analyses**. *PeerJ* (2021.0) **9** e11660. DOI: 10.7717/peerj.11660
31. Jan YK, Lung CW, Cuaderes E, Rong D, Boyce K. **Effect of viscoelastic properties of plantar soft tissues on plantar pressures at the first metatarsal head in diabetics with peripheral neuropathy**. *Physiol Meas* (2013.0) **34** 53-66. DOI: 10.1088/0967-3334/34/1/53
32. Lung CW, Wu FL, Liao F, Pu F, Fan Y, Jan YK. **Emerging technologies for the prevention and management of diabetic foot ulcers**. *J Tissue Viability* (2020.0) **29** 61-68. DOI: 10.1016/j.jtv.2020.03.003
33. 33.Mueller MJ, Zou DQ, Lott DJ. “Pressure gradient” as an indicator of plantar skin injury. Diabetes Care. 2005;28(12):2908–12. 10.2337/diacare.28.12.2908.
34. 34.Blaszczyk JW, Orawiec R. Assessment of postural control in patients with parkinson’s disease: Sway ratio analysis. Hum Mov Sci. 2011;30(2):396–404. 10.1016/j.humov.2010.07.017.
35. 35.Matsuno S, Yoshimura A, Yoshiike T, Morita S, Fujii Y, Honma M, et al. Toe grip force of the dominant foot is associated with fall risk in community-dwelling older adults: a cross-sectional study. Journal of Foot and Ankle Research. 2022;15(1). doi: 10.1186/s13047-022-00548-1.
36. Shaojun W, Lina S, Jingxian L, Dongqing X. **Comparison of the characteristics of plantar kinetics in older and young male adults during walking**. *J Med Biomech* (2021.0) **36** 304-308. DOI: 10.16156/j.1004-7220.2021.02.021
37. Brenton-Rule A, Dalbeth N, Menz HB, Bassett S, Rome K. **Foot and ankle characteristics associated with falls in adults with established rheumatoid arthritis: A cross-sectional study**. *BMC Musculoskelet Disord* (2016.0) **17** 22. DOI: 10.1186/s12891-016-0888-z
38. Pol F, Khajooei Z, Hosseini SM, Taheri A, Forghany S, Menz HB. **Foot and ankle characteristics associated with fear of falling and mobility in community-dwelling older people: a cross-sectional study**. *J Foot Ankle Res* (2022.0) **15** 86. DOI: 10.1186/s13047-022-00593-w
|
---
title: 'The association of gestational age and birthweight with blood pressure, cardiac
structure, and function in 4 years old: a prospective birth cohort study'
authors:
- Bowen Du
- Hualin Wang
- Yujian Wu
- Zhuoyan Li
- Yiwei Niu
- Qianchuo Wang
- Lin Zhang
- Sun Chen
- Yurong Wu
- Jihong Huang
- Kun Sun
- Jian Wang
journal: BMC Medicine
year: 2023
pmcid: PMC10029264
doi: 10.1186/s12916-023-02812-y
license: CC BY 4.0
---
# The association of gestational age and birthweight with blood pressure, cardiac structure, and function in 4 years old: a prospective birth cohort study
## Abstract
### Background
Current evidence relating birthweight and gestational age to cardiovascular risk is conflicting. Whether these factors have independent or interactive impacts on cardiovascular parameters during early childhood remains unclear. The goal of this study was to explore whether there were any independent and interactive effects of gestational age and birthweight on blood pressure, left ventricle (LV) structure, and function in 4 years old.
### Methods
This study included 1194 children in the Shanghai Birth Cohort from 2013 to 2016. Information about the mothers and children was recorded at time of birth using a questionnaire. Follow-up measurements, including anthropometric, blood pressure, and echocardiography, were taken between 2018 and 2021, when the children were 4 years old. Multiple linear or logistic regressions and restricted cubic spline were used to explore the association of birthweight and gestational age with cardiovascular measurements.
### Results
Gestational age had a significant negative correlation with both systolic blood pressure [β = − 0.41, $95\%$ CI: (− 0.76, − 0.07)] and mean arterial pressure [β = − 0.36, $95\%$CI: (− 0.66, − 0.07)]. The risk of prehypertension decreased with increased gestational age [OR = 0.54, $95\%$ CI: (0.32, 0.93)]. The relationship between birthweight with blood pressure was U-shape (P for non-linear < 0.001). The wall thickness, volume, mass, and cardiac output of LV increased with birthweight, though the ejection fraction [β = − 1.02, $95\%$ CI: (− 1.76, − 0.27)] and shorten fraction [β = 0.72, $95\%$ CI: (− 1.31, − 0.14)] decreased with birthweight. The risk of LV hypertrophy was not associated with birthweight [OR = 1.59, $95\%$ CI: (0.68, 3.73)].
### Conclusions
In this study, we found different associations of birthweight and gestational age with cardiovascular measurements in the offspring at 4 years old. Gestational age influenced blood pressure independent of birthweight. Heart size and function at 4 years old was influenced mostly by birthweight and not by gestational age.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12916-023-02812-y.
## Background
The risk of cardiovascular disease (CVD) in adulthood may be determined from early life [1] and is potentially influenced by both the intrauterine environment and early childhood development [2], which may be reflected in birthweight (BW) and gestational age (GA). The potential influence of BW and GA on the cardiovascular system is complicated and controversial. Recent studies have found that children with low BW or GA are linked to increase risk of CVD, such as hypertension [3, 4], heart failure [5], hypertrophy [6], ischemia disease [7, 8], or arrhythmias [9] in adolescence or in adulthood [10, 11]. Other studies have found that children with large BW have an increased future risk of CVD or hypertension [12–14]. Conversely, a British study suggested that large BW was associated with a lower future blood pressure (BP) [15]. Studies focused on preterm birth have found that it might increase the risk of hypertension or other CVD from childhood to adulthood [5, 7, 16, 17]. Any potential non-linear relationship between GA or BW and cardiovascular measurements remains unknown.
BW is strongly, but not only, influenced by GA. Children with low BW are usually preterm birth. Several studies have focused only on the simple linear correlation between BW and GA with cardiovascular indexes [16, 18, 19]. Other studies have focused primarily on participants per se, such as only prematurity or low BW, often without accounting for BW and GA independently [20]. Juonala et al. [ 21] observed increased BP levels only among preterm low BW participants and not in full term participants. These results suggested that GA might influence BP independently from BW. Other studies have focused on CVD risk in adolescence or adulthood [5, 7, 22], though the influence of BW and GA on the cardiovascular system may start in childhood. Therefore, studies investigating these risks during the early childhood period are still valuable. Additionally, evidence on the independent or interactive regulation of BW and GA on cardiovascular risk remains limited.
This study aimed to explore the independent and interactive associations of BW and GA with BP, left ventricle (LV) structure, and function in early childhood, based on detailed BP and echocardiography evaluation from a birth cohort in China.
## Participants
The Shanghai Birth Cohort (SBC) is an ongoing prospective cohort study conducted in six collaborating hospitals in Shanghai, China. A detailed description of the cohort has been provided elsewhere [23]. In this study, 1194 mother–child pairs joined our cardiovascular cohort between 2013 and 2016. The maternal self-reported questionnaires were conducted after enrollment and during pregnancy with the help of trained staff. Information on maternal demographic and sociodemographic characteristics (e.g., maternal income, race, education level) and lifestyle factors (e.g., passive smoking or alcohol drinking status during pregnancy) were collected using structured questionnaires. The history of hypertensive disorders in pregnancy (HDP) and gestational diabetes mellitus (GDM) was collected using structured questionnaires or extraction of inpatient history from medical records. Follow-up measurements (including weight, height, blood pressure and echocardiography) of children were carried-out between 2018 and 2021 when they were 4 years old. Children with congenital heart disease or other birth defects and those lost to follow-up, who were uncooperative, or without all available records were excluded. A total of 943 children were finally included in all analyses. Ethical approval was granted by the Ethical Committee of Xinhua Hospital affiliated to Shanghai Jiao Tong University School of Medicine (No. XHEC-C-2013–001-2). All parents or guardians of participants signed written informed consent forms prior to enrollment.
## Anthropometric measurement
The BW and GA were recorded after birth and extracted from the inpatient medical history of the pregnant women or from pediatric medical records. Between 2018 and 2021, when the children were 4 years old, their gender, height, and weight were measured and recorded. BMI was calculated by weight (kg)/height (cm)2. Preterm was defined as newborns with GA < 37 weeks. Term was defined as children with GA between 37 and 42 weeks. The BW groups were classified as follows: low birth weight (LBW) group < 2500 g, normal birthweight (NBW) group 2500–4000 g, and macrosomia group with birthweight > 4000 g.
## Blood pressure (BP)
Systolic BP (SBP) and diastolic BP (DBP) of children were assessed by a single, trained staff member on the left arm at heart level and with the appropriate cuff size for arm circumference while they were in the supine position. The OMRAN HBP-1300 automatic BP device (Omron Healthcare, Guangzhou, China) was used. Three measurements were taken at 5 min intervals once each child had calmed down. The mean of the three measurements was used in all analyses. Mean arterial pressure (MAP) was calculated as MAP = DBP + (SBP-DBP)/3. The 90th and 95th percentile of SBP and DBP for the sex and height were defined according to the Chinese standard [24]. Prehypertension was defined as P95th > SBP and/or DBP ≥ P90th. Hypertension was defined as SBP and/or DBP ≥ P95th.
## Echocardiography
Transthoracic echocardiography examinations were performed on the children by trained operators according to the American Society of Echocardiography recommendations [25] and using the Philip EPIQ7C (Philips Healthcare, Andover, USA) ultrasound that uses the X5-1 (1-5 MHz) or S8-3 (3-8 MHz) matrix-array transducers (Philips Healthcare, Andover, USA). LV interventricular or posterior wall thickness, internal diameter, volume in systole and diastole (IVSs, IVSd, LVPWs, LVPWd, LVIDs, LVIDd, ESV and EDV), ejection fraction (LVEF), and shorten fraction (LVFS) were measured. Relative wall thickness (RWT), LV mass (LVM), LVM index (LVMI), E/A ratio, Tei index, and global peak longitudinal strain (GLS) were measured as previously described [26].
Aortic annular diameter was measured. LV velocity time integral (VTI) was assessed by identifying the apical view in which peak flow velocity was maximal and, after calibration, tracing the black-white interface outlining the Doppler flow envelope. Aortic annular cross-sectional area was calculated as follows: π × (diameter/2)2. Doppler stroke volume (SV) was calculated as annular cross-sectional area multiplied by the VTI. Cardiac output (CO) was calculated as SV × HR. Total peripheral vascular resistance (TPVR) was calculated as MAP divided by CO [27]. The carotid intima-media thickness (cIMT), which was defined as the thickness of the intima and media of carotid artery, was averaged by measurements of 6 times at common carotid artery 1 cm below the carotid bulb on each side in 2-D images captured by C8-3 (3–8 MHz) or L15-7io (7–15 MHz) transducers [28, 29].
The sex-specific 95th percentiles (P95th) of LVMI (male: 33.76 g/m2.7 and female: 33.24 g/m2.7) were derived from our own cohort. LV hypertrophy (LVH) was defined as LVMI ≥ the sex-specific P95th of LVMI [26, 30]. All examinations were performed by experienced operators. Both the sonographers and the observers were blinded to the details of the participants.
## Statistical analysis
Comparisons of continuous variables were performed using the one-way analysis of variance (one-way ANOVA), followed by the Bonferroni post hoc test to adjust for multiple comparisons in different groups, when normality and homogeneity of variance assumptions are satisfied. Otherwise, the equivalent nonparametric tests were used. Both the Kolmogorov–Smirnov and Levene’s test were used to evaluate the normal distribution and homogeneity of variances. Categorical variables were compared using chi-square tests or Fisher’s exact tests.
Analysis of the associations of BW and GA with BP, LV structure, and function was conducted by constructing multi-factor linear regression models. To explore the independent effects of BW and GA on cardiovascular parameters, linear regression models were established in different BW and GA subgroups. Measurements for BW and GA were also used in the same model to analyze their interrelationship or mediated effect with each other. Odds ratios (OR) were calculated using a logistic regression model adjusted for the maternal and children’s factors. The interaction of BW and GA was analyzed in linear and logistic models. The additive interaction in logistic regression was assessed using the R package “epiR.” Restricted cubic spline models with 3 knots at the 5th, 50th, and 95th percentiles were constructed to determine the nonlinear correlation of BW or GA with cardiovascular measurements with “rms,” “splines,” and “ggplot2” R packages. Tests for nonlinearity were conducted by likelihood ratio tests.
Postnatal weight gain has been proven to play an important role in cardiovascular outcomes in preterm or LBW [31, 32]. BP may influence the LV structure and function (Additional file 1: Table S1). Weight gain from birth to 4 years old and the SBP of children were inputted into the regression models to examine their effect on cardiovascular parameters. For sensitivity analysis, multiple imputation was used to input the missing values in maternal information to analyze the relationship of BW and GA with blood pressure, LV structure, and function.
All statistical analysis were carried out using the SPSS 25.0 software program (IBM Corp., Armonk, NY, USA), Stata 15.0 (Stata Corporation, College Station, TX, USA), and R 4.0.4 (R Foundation for Statistical Computing). All tests were two-sided with a significance level of 0.05.
## The baseline characteristics
A total of 1194 mother–child pairs were included in the cardiovascular cohort in SBC. After excluding children due to missing information ($$n = 240$$) and birth defects ($$n = 11$$), a total of 943 ($78.9\%$) children were included for final analysis (Additional file 1: Figure S1). In total, the study population consisted of $8.6\%$ preterm ($$n = 81$$), $4.8\%$ LBW ($$n = 45$$), and $7.6\%$ macrosomia ($$n = 72$$) newborns. At 4 years of age, the difference in anthropometric indexes disappeared from GA groups but was maintained in the BW groups. LBW and preterm children gained more weight from birth to 4 years old than NBW and term (Table 1).Table 1Basic characteristics of participantsLBW ($$n = 45$$)NBW ($$n = 826$$)Macrosomia ($$n = 72$$)P valuePreterm ($$n = 81$$)Term ($$n = 862$$)P valueTotal ($$n = 943$$)Maternal characteristics Pre-pregnancy BMI, mean (SD), kg/m222.34 (3.08)21.54 (3.19)22.65 (3.23)0.0121.88 (3.20)21.61 (3.18)0.5121.52 (3.03) Maternal age, mean (SD), years31.33 (3.32)30.88 (3.51)30.38 (3.56)0.3531.10 (4.12)30.84 (3.47)0.5130.83 (3.52) HDP, N (%)5 ($11.1\%$)75 ($9.1\%$)4 ($5.6\%$)0.437 ($8.6\%$)77 ($8.9\%$)0.4884 ($8.9\%$) Passive smoke, N (%)12 ($26.7\%$)391 ($47.3\%$)32 ($44.4\%$)0.1327 ($33.3\%$)406 ($47.1\%$)0.63435 ($46.1\%$) Drink during pregnancy, N (%)5 ($11.1\%$)127 ($15.4\%$)7 ($9.7\%$)0.8210 ($12.3\%$)129 ($15.0\%$)0.86139 ($14.7\%$) University background, N (%)29 ($64.4\%$)690 ($83.5\%$)46 ($63.9\%$)0.5151 ($62.9\%$)711 ($82.5\%$)0.41765 ($81.1\%$) Ethic Han, N (%)41 ($91.1\%$)826 ($100\%$)70 ($97.2\%$)0.5275 ($92.6\%$)862 ($100.0\%$)0.28937 ($99.4\%$) Income ≥ 100,000 yuan/year, N (%)20 ($44.4\%$)459 ($55.6\%$)36 ($50.0\%$)0.2935 ($43.2\%$)478 ($55.4\%$)0.66515 ($54.6\%$)Children characteristics Birth height, mean (SD), cm46.84 (1.89)49.75 (1.23)51.60 (1.81) < 0.00147.95 (1.87)49.89 (1.39) < 0.00149.8 (1.50) BW, mean (SD), kg2.19 (0.33)3.31 (0.35)4.23 (0.22) < 0.0012.52 (0.48)3.39 (0.42) < 0.0013.33 (0.48)BW groups, N (%) LBW////27 ($33.3\%$)18 ($2.9\%$) < 0.00145 ($4.8\%$) NBW54 ($66.7\%$)772 ($89.6\%$) < 0.001826 ($87.6\%$) MBW0 ($0.0\%$)72 ($8.4\%$) < 0.00172 ($7.6\%$)GA, mean (SD), weeks35.55 (2.79)39.21 ((1.41)39.85 (0.79) < 0.00134.92 (2.15)39.43 (1.05) < 0.00139.13 (1.51)GA groups, N (%) Preterm27 ($60.0\%$)54 ($6.5\%$)0 ($0\%$) < 0.001///81 ($8.6\%$) Term18 ($40.0\%$)772 ($93.5\%$)72 ($100\%$) < 0.001862 ($91.4\%$)Gender, N (%) Boys23 ($51.1\%$)479 ($58.0\%$)50 ($69.4\%$)0.0132 ($39.5\%$)520 ($54.1\%$)0.13552 ($58.5\%$) Girls22 ($48.9\%$)347 ($42.0\%$)22 ($30.6\%$)0.0149 ($60.5\%$)342 ($39.7\%$)0.13391 ($41.5\%$)Height at 4, mean (SD),cm106.62 (4.33)108.07 (4.81)110.25 (4.62) < 0.001107.91 (4.73)107.24 (4.53)0.71107.26 (4.53)Weight at 4, mean (SD), kg16.99 (2.08)17.75 (2.83)19.30 (3.08) < 0.00117.50 (2.66)17.27 (2.60)0.7017.28 (2.56)BMI at 4, mean (SD) kg/m214.93 (1.43)15.13 (1.65)15.83 (1.98)0.00214.98 (1.66)14.97 (1.59)0.2514.97 (1.59)Postnatal weight gain, mean (SD), kg14.77 (2.15)13.82 (2.42)15.06 (3.01) < 0.00115.18 (2.61)13.86 (2.45) < 0.00113.94 (2.48)The bold values were $P \leq 0.05$GA Gestational age, BW Birthweight, LBW Low birthweight, NBW Normal birthweight, BMI Body mass index, HDP Hypertensive disorders in pregnancy, GDM Gestational diabetes mellitus
## BP, LV structure, and function differed between the GA and BW groups
In different BW groups, LVIDd, IVSs, LVIDs, LVM, SV, EDV, and ESV were smaller in the LBW group and larger in the macrosomia group than NBW. The SBP and LV function were not significantly different among BW groups. In different GA groups, the cardiovascular measurements were not significantly different in full term and preterm birth children (Table 2). After adjusting for maternal or children’s factors, the results were consistent (Additional file 1: Tables S2 and S3).Table 2The BP, LV structure, and function in different BW and GA groupsLBW ($$n = 45$$)NBW ($$n = 826$$)Macrosomia ($$n = 72$$)P valuePreterm ($$n = 81$$)Term ($$n = 862$$)P valueTotal($$n = 943$$)Blood pressure SBP, mean (SD), mmHg98.93 (7.93)98.16 (7.67)100.22 (7.24)0.0998.24 (8.17)97.54 (7.72)0.8097.57 (7.72) DBP, mean (SD), mmHg57.63 (7.35)57.46 (6.27)57.20 (6.45)0.9457.72 (6.37)57.04 (6.32)0.8957.04 (6.19) MAP, mean (SD), mmHg72.24 (6.65)71.45 (5.98)71.59 (5.61)0.6971.88 (6.55)70.89 (5.99)0.1770.97 (5.95)LV structure IVSd, mean (SD), mm3.88 (0.63)3.84 (0.53)3.82 (0.31)0.883.79 (0.33)3.81 (0.59)0.803.79 (0.56) LVIDd, mean (SD), mm34.92 (2.44)35.75 (2.55)36.78 (2.20)a,b0.00135.52 (2.72)35.48 (2.53)0.8935.47 (2.51) LVPWd, mean (SD), mm4.02 (0.51)4.14 (0.61)4.22 (0.47)0.284.15 (0.63)4.15 (0.59)10.004.14 (0.56) IVSs, mean (SD), mm6.46 (0.86)6.53 (0.98)6.93 (0.93)a0.0066.53 (0.88)6.57 (0.97)0.786.57 (0.95) LVIDs, mean (SD), mm22.20 (1.78)22.81 (1.97)23.43 (2.10)a,b0.0122.69 (1.94)22.57 (1.96)0.6322.57 (1.94) LVPWs, mean (SD), mm7.82 (0.93)7.80 (0.99)8.03 (0.95)0.187.85 (0.90)7.87 (0.95)0.847.84 (0.91) RWT0.23 (0.04)0.23 (0.04)0.22 (0.02)0.350.23 (0.03)0.23 (0.04)0.920.20 (0.10) LVM, mean (SD), g30.44 (4.66)32.34 (6.38)34.47 (5.55)a,b0.00631.62 (5.09)31.83 (6.33)0.7931.66 (6.04) LVMI, mean (SD), g/m2.725.58 (3.57)26.19 (4.83)26.30 (3.74)0.7525.73 (4.03)26.34 (4.97)0.3226.20 (4.74) cIMT, mean (SD), *10−2mm41.35 (4.84)40.45 (4.58)40.82 (4.59)0.5840.88 (4.76)40.76 (4.81)0.8740.78 (4.81)LV function LVEF, mean (SD), %66.99 (4.27)69.21 (4.08)66.93 (4.42)0.9766.54 (3.98)66.93 (4.12)0.4666.90 (4.14) LVFS, mean (SD), %36.37 (3.37)36.17 (3.21)36.50 (3.49)0.703 6.06 (3.12)36.36 (3.23)0.4736.34 (3.25) Tei index, mean (SD), %44.82 (6.44)43.77 (7.03)43.01 (6.17)0.4544.20 (7.85)43.22 (6.27)0.2143.26 (6.20) E/A1.78 (0.35)1.81 (0.36)1.79 (0.39)0.711.81 (0.30)1.80 (0.34)0.791.81 (0.40) CO, mean (SD), L/min4.09 (1.01)4.22 (1.20)4.55 (1.39)0.104.37 (1.27)4.19 (1.13)0.224.19 (1.16) SV, mean (SD), mL43.37 (10.32)46.36 (12.98)51.13 (14.94)a,b0.00847.89 (12.87)45.70 (12.41)0.1645.71 (12.56) TPVR, mean (SD), dyne*s/cm51540.19 (525.32)1465.55 (447.13)1386.20 (491.63)0.281427.89 (455.43)1463.51 (444.36)0.541469.40 (450.46) EDV, mean (SD), %50.97 (8.80)53.94 (9.23)57.65 (8.26)a,b0.00153.22 (9.85)53.00 (8.98)0.8552.95 (8.89) ESV, mean (SD), %16.77 (3.50)17.99 (3.85)19.22 (4.11)a,b0.00817.76 (3.84)17.52 (3.77)0.6117.50 (3.75) VTI, mean (SD), cm21.72 (3.63)21.50 (3.24)21.74 (3.93)0.8022.31 (3.51)21.52 (3.25)0.0521.57 (3.28) GLS, mean (SD), %23.49 (2.27)23.57 (2.30)23.36 (2.33)0.8723.39 (2.62)23.58 (2.28)0.6523.54 (2.26) Prehypertension, N (%)4 ($8.9\%$)131 ($15.8\%$)6 ($8.3\%$)0.1311 ($13.6\%$)130 ($15.1\%$)0.09141 ($14.9\%$) Hypertension, N (%)6 ($13.3\%$)85 ($10.3\%$)10 ($13.9\%$)0.7111 ($13.6\%$)89 ($10.3\%$)0.85101 ($10.7\%$) LVH, N (%)0 ($0.0\%$)53 ($6.4\%$)2 ($2.8\%$)0.272 ($2.5\%$)53 ($6.1\%$)0.5755 ($5.8\%$)The bold values were $P \leq 0.05$LBW Low birthweight, NBW Normal birthweight, SBP Systolic blood pressure, DBP Diastolic blood pressure, LV Left ventricle, IVSd Ventricle interventricular septal thickness in diastole, LVIDd LV internal diameter in diastole, LVPWd LV posterior wall thickness in diastole, IVSs Ventricle interventricular septal thickness in systole, LVIDs LV internal diameter in systole, LVPWs LV posterior wall thickness in systole, RWT Relative wall thickness, LVM Left ventricle mass, LVMI LV mass index, cIMT carotid artery intima-media thickness, LVEF LV ejection fraction, LVFS LV fractional shorting, CO Cardiac output, SV Stroke volume, TPVR Total peripheral vascular resistance, MAP Mean arterial pressure, EDV End diastolic volume, ESV End systolic volume, VTI Velocity time integral, GLS Global longitudinal strain, LVH LV hypertrophyaCompared with NBW were significant using Bonferroni testbCompared with LBW were significant using Bonferroni test
## The different effect of GA and BW on BP
GA had a significant negative correlation with SBP [β = − 0.41, $95\%$ CI: (− 0.76, − 0.07)] and MAP [β = − 0.36, $95\%$ CI: (− 0.66, − 0.07)], but not with DBP after adjusting for maternal and children’s factors (Table 3). The risk of prehypertension decreased with increasing GA [OR = 0.54, $95\%$ CI: (0.32, 0.93)] after adjusting for BMI (Table 4). BMI is a key negative confounder in GA and BP. After adjusting for postnatal weight gain, the significant relationship between BP and GA disappeared, suggesting postnatal weight gain was a completed mediator in GA and SBP (Additional file 1: Table S4).Table 3The independent and interaction effect of BW and GA on BP, LV structure, and functionBWGACrude ModelModel 1Model 2Crude ModelModel 1Model 2P for interaction#Blood pressure SBP (mmHg)0.79 (− 0.24, 1.81) − 0.03 (− 1.31, 1.24) − 0.77 (− 2.02, 0.48) − 0.30 (− 0.58, − 0.02) − 0.47 (− 0.84, − 0.11) − 0.41 (− 0.76, − 0.07)0.63 DBP (mmHg) − 0.47 (− 1.29, 0.36) − 0.29 (− 1.39, 0.81) − 0.59 (− 1.71, 0.52) − 0.15 (− 0.38, 0.08) − 0.17 (− 0.49, 0.15) − 0.16 (− 0.48, 0.16)0.97 MAP (mmHg) − 0.22 (− 1.00, 0.56) − 0.51 (− 1.54, 0.52) − 0.95 (− 1.99, 0.08) − 0.21 (− 0.43, 0.01) − 0.38 (− 0.67, − 0.08) − 0.36 (− 0.66, − 0.07)0.98LV structure IVSd (mm)0.04 (− 0.03, 0.12)0.03 (− 0.06, 0.12)0.02 (− 0.07, 0.12) − 0.01 (− 0.03, 0.01) − 0.02 (− 0.05, 0.01) − 0.41 (− 0.92, 0.11)0.51 LVIDd (mm)1.04 (0.71, 1.38)1.02 (0.61, 1.43)0.78 (0.37, 1.19)0.11 (0.01, 0.20)0.11 (− 0.01, 0.23) − 0.02 (− 0.05, 0.01)0.01 LVPWd (mm)0.13 (0.54, 0.21)0.11 (0.01, 0.21)0.08 (− 0.02, 0.18) − 0.01 (− 0.03, 0.01) − 0.16 (− 0.05, 0.01) − 0.02 (− 0.04, 0.01)0.97 IVSs (mm)0.31 (0.18, 0.44)0.24 (0.09, 0.40)0.18 (0.02, 0.34)0.02 (− 0.02, 0.05)0.01 (− 0.03, 0.06)0.01 (− 0.03, 0.06)0.90 LVIDs (mm)0.71 (0.44, 0.97)0.93 (0.61, 1.26)0.79 (0.46, 1.11)0.05 (− 0.02, 0.13)0.10 (0.00, 0.19)0.10 (0.00, 0.19)0.08 LVPWs (mm)0.24 (0.12, 0.36)0.12 (− 0.04, 0.28)0.03 (− 0.13, 0.19)0.00 (− 0.03, 0.04) − 0.01 (− 0.06, 0.04) − 0.01 (− 0.05, 0.04)0.18 RWT0.00 (− 0.01, 0.01)0.00 (− 0.01, 0.01)0.00 (− 0.01, 0.01)0.00 (0.00, 0.00)0.00 (0.00, 0.00)0.00 (0.00, 0.00)0.31 LVM (g)2.58 (1.78, 3.39)2.44 (1.42, 3.45)1.88 (0.87, 2.90)0.09 (− 0.15, 0.33)0.00 (− 0.30, 0.31)0.01 (− 0.28, 0.30)0.05 LVMI (g/cm2.7)0.66 (0.01, 1.30)0.59 (− 0.26, 1.25)0.15 (− 0.69, 1.00)0.09 (− 0.10, 0.28)0.09 (− 0.15, 0.33)0.09 (− 0.15, 0.33)0.67 cIMT (*10−2 mm)0.28 (− 0.53, 1.09)0.04 (− 1.00, 1.09)0.08 (− 0.99, 1.15)0.11 (− 0.11, 0.32)0.16 (− 0.12, 0.45)0.16 (− 0.12, 0.45)0.51LV function LVEF (%) − 0.15 (− 0.71, 0.42) − 1.00 (− 1.73, − 0.28) − 1.02 (− 1.76, − 0.27)0.05 (− 0.11, 0.20) − 0.13 (− 0.34, 0.08) − 0.13 (− 0.34, 0.08)0.74 LVFS (%) − 0.03 (− 0.47, 0.42) − 0.71 (− 1.28, − 0.14) − 0.72 (− 1.31, − 0.14)0.04 (− 0.08, 0.17) − 0.10 (− 0.26, 0.07) − 0.10 (− 0.26, 0.07)0.61 Tei index (%) − 0.23 (− 1.06, 0.60)0.03 (− 1.05, 1.12)0.16 (− 0.94, 1.26) − 0.12 (− 0.36, 0.11)0.03 (− 0.28, 0.34)0.03 (− 0.28, 0.34)0.63 E/A0.00 (− 0.04, 0.04)0.01 (− 0.05, 0.06)0.01 (− 0.04, 0.07)0.00 (− 0.01, 0.02)0.01 (− 0.01, 0.26)0.01 (− 0.01, 0.03)0.82 CO (L/min)0.24 (0.08, 0.40)0.24 (0.04, 0.44)0.15 (− 0.06, 0.35) − 0.01 (− 0.06, 0.03) − 0.02 (− 0.08, 0.04) − 0.01 (− 0.07, 0.04)0.53 SV (ml)3.91 (2.21, 5.61)4.35 (2.19, 6.51)3.40 (1.23, 5.56)0.07 (− 0.40, 0.54)0.10 (− 0.52, 0.71)0.11 (− 0.49, 0.72)0.80 TPVR (dynes*s/cm5) − 77.19 (− 141.11, − 13.26) − 93.15 (− 177.34, − 8.96) − 75.53 (− 160.79, 9.74)6.17 (− 11.09, 23.48)3.85 (− 20.19, 27.88)2.87 (− 21.03, 26.77)0.53 EDV (ml)3.74 (2.56, 4.93)3.72(2.24, 5.19)2.86 (1.39, 4.32)0.36 (0.01, 0.70)0.39 (− 0.04, 0.83)0.40 (− 0.02, 0.81)0.009 ESV (ml)1.38 (0.87, 1.88)1.80 (1.17, 2.44)1.51 (0.87, 2.15)0.09 (− 0.05, 0.23)0.19 (0.00, 0.38)0.19 (0.01, 0.37)0.07 VTI (cm)0.02 (− 0.43, 0.46)0.11 (− 0.48, 0.69) − 0.08 (− 0.67, 0.51) − 0.05 (− 0.17, 0.08) − 0.04 (− 0.21, 0.12)0.04 (− 0.21, 0.12)1.00 GLS (%) − 0.15 (− 0.56, 0.26) − 0.40 (− 0.94, 0.15) − 0.32 (− 0.87, 0.23)0.07 (− 0.05, 0.18) − 0.01 (− 0.16, 0.15) − 0.01 (− 0.17, 0.15)0.09The data were presented as β ($95\%$ CI) in linear regression models. The missing values were not inputted (crude model: $$n = 943$$, model 1: $$n = 786$$, model 2: $$n = 583$$)The bold values were $P \leq 0.05$Model 1: adjusted for maternal nationality, scholarship, income, HDP, GDM, drink history, passive smoke history, and gender of childrenModel 2: model 1 + BMI at 4 years oldBW Birthweight, GA Gestational age, SBP Systolic blood pressure, DBP Diastolic blood pressure, LV Left ventricle, IVSd Ventricle interventricular septal thickness in diastole, LVIDd LV internal diameter in diastole, LVPWd LV posterior wall thickness in diastole, IVSs Ventricle interventricular septal thickness in systole, LVIDs LV internal diameter in systole, LVPWs LV posterior wall thickness in systole, RWT Relative wall thickness, LVM Left ventricle mass, LVMI LV mass index, cIMT carotid artery intima-media thickness, LVEF LV ejection fraction, LVFS LV fractional shorting, CO Cardiac output, SV Stroke volume, TPVR Total peripheral vascular resistance, MAP Mean arterial pressure, EDV End diastolic volume, ESV End systolic volume, VTI Velocity time integral, GLS Global longitudinal strain, BMI Body mass index, HDP Hypertensive disorders in pregnancy,GDM Gestational diabetes mellitus#Adjusted for maternal nationality, scholarship, income, HDP, GDM, drink history, passive smoking, gender of children, and BMI at 4Table 4The cardiovascular risk with GA and BWPrehypertension ($$n = 141$$)Hypertension ($$n = 101$$)LVH ($$n = 55$$)CrudeModel 1Model 2CrudeModel 1Model 2CrudeModel 1Model 2BW0.99 (0.89, 1.10)0.90 (0.79, 1.03)0.92 (0.80, 1.05)0.93 (0.83, 1.05)0.92 (0.79, 1.07)0.94 (0.80, 1.10)1.60 (0.84, 3.04)1.00 (1.00, 1.00)1.59 (0.68, 3.73)GA0.76 (0.50, 1.13)0.59 (0.35, 1.00)0.54 (0.32 0.93)1.16 (0.70, 1.76)1.02 (0.57, 1.84)0.80 (0.44, 1.45)1.10 (0.91, 1.33)1.05 (0.82, 1.33)1.05 (0.82, 1.34)Data were presented as OR ($95\%$ CI). P for multiplicative and addictive interaction of BW and GA were not significant. The missing values were not inputted (crude model: $$n = 943$$, model 1: $$n = 786$$, model 2: $$n = 583$$)The bold values were $P \leq 0.05$Model 1: adjusted for maternal nationality, scholarship, income, HDP, GDM, drink history, passive smoke history, and gender of childrenModel 2: model 1 + BMI at 4 years oldThe 90th and 95th percentile of SBP and DBP for the sex and height were defined according to the Chinese standard [24]. Prehypertension was defined as P95th > SBP and/or DBP ≥ P90th. Hypertension was defined as SBP and/or DBP ≥ P95th. LV hypertrophy (LVH) was defined as LVMI ≥ the sex-specific P95th of LVMIBW Birthweight, GA Gestational age, LVH Left ventricle hypertrophy, HDP Hypertensive disorders in pregnancy, GDM Gestational diabetes mellitus The BW was found not to have a significant association with BP or hypertension in either the linear or the logistic regression models (Tables 3 and 4). However, a non-linear association between BP and BW was found. The smoothing line of BW and SBP was in a U-shape (P for non-linear < 0.001) (Fig. 1).Fig. 1The restricted cubic spline of BW and SBP ($$n = 943$$). The red lines indicate the predicted cardiovascular parameters derived from the restricted cubic spline regression model with 3 knots at the 5th, 50th, and 95th percentiles of BW. The shadow indicates the $95\%$CIs. Tests for non-linearity were conducted by using likelihood ratio tests. BW, birthweight; SBP, systolic blood pressure The positive linear correlation between BW and GA was strong [β = 0.18, $95\%$ CI: (0.16, 0.19)], though GA still influenced the BP independently from BW. When putting BW and GA in the same model, the SBP remained negatively correlated with GA but not with BW (Additional file 1: Table S5). In different BW groups, SBP was negatively associated with GA in participants with preterm birth children, including LBW [β = − 0.58, $95\%$ CI: (− 1.13, − 0.04)] and NBW groups [β = − 0.55, $95\%$ CI: (− 1.04, − 0.07)]). In different GA subgroups, the BW had a negative correlation with SBP only in preterm children not in term [β = − 4.22, $95\%$ CI: (− 8.02, − 0.42)] (Additional file 1: Table S6). Preterm birth played a more important role in the occurrence of high BP than LBW.
## The different effect of GA and BW on LV structure and function
BW was found to be associated with higher LV wall thickness [IVSs: β = 0.18, $95\%$ CI:(0.02,0.34)], interventricular diameters [LVIDd: β = 0.78, $95\%$ CI:(0.37,1.19), LVIDs: β = 0.79, $95\%$ CI:(0.46, 1.11)], volume [EDV: β = 2.86, $95\%$ CI:(1.39,4.32), ESV: β = 1.51, $95\%$ CI:(0.87, 2.15)], and LVM [β = 1.88, $95\%$ CI:(0.87, 2.90)]. LV function indexes of LVEF [β = − 1.02, $95\%$ CI: (− 1.76, − 0.27)] and LVFS [β = − 0.72, $95\%$ CI: (− 1.31, − 0.14)] decreased with increasing BW following adjustment for confounders (Table 3). These results were stable in non-linear regression models (Additional file 1: Figure S2), though the risk of LVH was not significant (Table 4). In the linear regression or non-linear models, the correlation of LV structure and function with GA was not significant (Table 3 and Additional file 1: Figure S2).
BW influenced the LV structure and function independently from GA. When inputting BW and GA into the same models (Additional file 1: Table S5), the effect of BW on the LV structure and function were not changed, even after adjusting for the postnatal weight gain and SBP (Additional file 1: Table S4). However, the GA was found to have a negative correlation with LVPWd, CO, SV, and LVM (Additional file 1: Table S5). After adjusting for postnatal weight gain and SBP, the significance of the relationship between GA and these indexes disappeared (Additional file 1: Table S4). In different GA groups, the effect of BW on the LV structure and function were stable in full term children. GA had no significant association with LV structure and function in any of the BW subgroups (Additional file 1: Table S6). In sensitivity analysis, the results were stable (Additional file 1: Table S7-S8).
## Discussion
To the best of our knowledge, this study is the first to suggest that GA and BW have different effects on the cardiovascular system, though with a strong association with each other. GA mainly influenced the BP while BW had an independent impact on cardiac structure and function. The relationship between BW and BP was in a U-shape, and the interaction effect of GA and BW on cardiovascular parameters was not significant.
This study suggested that GA had a negative association with BP and low GA increased the risk of prehypertension, even in early childhood. The risk of hypertension was not significant associated with GA, potentially because of the small sample size of preterm children with hypertension in our population. Recent studies have also found a significant inverse correlation between SBP and GA [3, 16, 33], and individuals of preterm birth or those with fetal growth restriction have a tendency for elevated BP levels in childhood and adulthood [34–36]. Some studies found that LBW might increase the risk of hypertension in the future [37, 38], but LBW individuals are usually preterm. However, no study has explored the independent association of GA and BW on BP. Our results suggest that for LBW children, the increase in BP maybe arise independently from preterm birth. Increased BW for preterm birth infants could potentially reduce BP in childhood.
BP is mainly influenced by BMI [39]. Obese children usually have higher BP which could explain the high BW children have higher BP than normal BW in the U-shape line of BW and BP. For preterm or LBW children, the BMI was lower than normal; however, they may still have higher BP and risk of hypertension during adolescence and adulthood [40–42]. This phenomenon is particularly interesting. It can be hypothesized that the mechanism is as follows: GA mainly influences children by the duration of the intra or extra uterine environmental exposure and affects the degree of fetal organ development through perinatal reprogramming [43]. It is determined by the maternal and/or fetal health condition. High SBP and normal or low DBP indicates high pulse pressure and increased blood vessel stiffness. Preterm and fetal growth restriction babies have defects in renal development with less nephron [44]. Vascular or cardiac development defection or early exposure to extra uterine environment can cause early hypoxic or inflammation damage, which then leads to accelerated vascular aging, vascular elastance reduction, and cardiac remodeling [40–42]. Additionally, the sympathetic system is overactivated, and parasympathetic nervous system tone is deficient in preterm infants [33]. Early stress response stimulates elevation of BP raising hormones that finally contribute to a greater SBP [43]. Furthermore, higher BP in preterm birth children has been proposed to be influenced by other intrauterine or postnatal factors, aside from their own BMI state, such as maternal disease, like hypertension or diabetes, and children’s early fast weight gain or catch-up growth pattern [45]. In this study, after adjusting for BMI, the risk of low GA with prehypertension was significant. Postnatal weight gain played a mediating effect on GA and BP, suggesting postnatal growth might play an important role in the relationship between GA and BP. For preterm or LBW, early overnutrition can improve malnutrition or growth retardation, but it increases the risk of CVD in the future [46–48]. Early growth management in preterm infants remains a challenge and is worthy of further exploration.
Recent studies have found similar results of a relationship between BW and LV structure and function. The studies by Toemen et al. [ 49] and Zhang et al. [ 50] found that children who are larger at birth, and have a longtime burden of excessive growth, are at increased risk of LV hypertrophy in adolescence and adulthood. A study by Harris et al. [ 51] similarly found that very low BW adults had smaller LVs, higher LV elastance, and lower arterial elastance. However, existing studies usually focus on specific disease participants and have only addressed either GA or BW with cardiovascular measurements. Our study is the first to report an independent and interactive association of GA and BW on cardiovascular measurements in continuous data.
BW influences LV structure and function potentially via BMI burden from birth. BW reflects the size and growth condition of body and organs during the fetal stage. It is influenced by GA, maternal nutrition state, intrauterine environment, fetal organ weight, and so on. However, both BP and LV structure are determined to some degree by the growth and BMI of children. The number of fat cells has been determined previously in neonates [52]. The body or organ size in the future is determined by the size in early life [53]. It is known that high BW may contribute to maintenance of BMI burden throughout life [22, 54]. BP and LV wall thickness are known to have a positive association with BMI [39]. Excessive BMI can be detrimental to LV function and increase the risk of LV dysfunction, hypertrophy, and hypertension in both adolescents and adults [19, 55–58]. However, in early childhood, changes in LV structure may arise from physiological factors rather than pathological ones. Function damage would be relatively small while the heart would get bigger and heavier as the body grows. Therefore, the growth of the body is typically faster than the growth of the heart. LVMI therefore might not change significantly with BW while only increases in LVM are observed.
In other studies, GA has been found to influence both cardiac structure and function. Preterm children might have defects in both cardiac development and growth. Goss et al. [ 10] and Mohlkert et al. [ 17] found that children, adolescents, and young adults born prematurely had significantly smaller biventricular cardiac chamber size and decreased cardiac mass with altered systolic and diastolic functions. But, in our study, the cardiac structure and function were not influenced by GA. The reason might be like this. GA had a negative association with BP, which means that preterm babies might have a higher BP in the future. It has been proposed that LV hypertrophy is caused by high SBP [59]. According to current studies, preterm children have smaller heart sizes and masses than term babies. For preterm children, BP and the influence of postnatal catch-up growth on cardiac growth could balance the change of cardiac structure in the early childhood. However, if the high BP condition is persistent, cardiac hypertrophy might occur in the future [6].
It should be noted that this study had some limitations. First, the sample size of extreme preterm and post-term birth and extremely low or macrosomia children were relatively small in our study, which may have contributed to bias to the results. Further studies with larger sample sizes are needed to draw stronger conclusions. Second, echocardiography for younger children requires sedatives to be administered during the examination, and so it was only performed at 4 years old, presenting a major study limitation as these parameters may have changed during early development. As this is an ongoing cohort, longitudinal assessment of cardiovascular parameters will be performed again in a further follow-up. Third, we did not collect blood metabolism biomarkers, other cardiovascular measurements, and information of other confounders, such as daily exercise, diet, and sleep. Further studies with more detailed information are recommended.
## Conclusions
GA and BW have different effects on cardiovascular measurements in 4 years old. BP was negatively correlated with GA. The relationship of BW and BP was non-linear. BW had a greater impact on LV structure and function than GA. The LV wall thickness, interventricular diameter, volume, and mass increased with BW. The LV function decreased with BW. Small GA played a more important role in the occurrence of high BP than low or high BW. Heart size and function at 4 years old was influenced mostly by BW and not by GA. BW and GA are important factors in early childhood cardiovascular health. Therefore, BW control, prevention of preterm birth, and early cardiovascular risk screening in children are recommended to help early prevention on children from getting cardiovascular diseases.
## Supplementary Information
Additional file 1: Figure S1. The flowchart of the study. Figure S2. The restricted cubic spline of BW and GA with cardiovascular parameters. Table S1. The influence of SBP and DBP on LV structure and function. Table S2-S3. The BP, LV structure and function in different BW and GA groups in linear regression models. Table S4. The association of BW and GA with BP, LV structure and function after adjusted for children weight gain from birth and BP. Table S5. The effect of BW and GA on BP, LV structure and function by putting them in the same models. Table S6. The association of BW and GA with BP, LV structure and function in fixed GA or BW groups. Table S7-S8. The effect of BW and GA on BP, LV structure and function after multiple imputation.
## References
1. Barker DJ, Winter PD, Osmond C, Margetts B, Simmonds SJ. **Weight in infancy and death from ischaemic heart disease**. *Lancet* (1989.0) **2** 577-580. DOI: 10.1016/S0140-6736(89)90710-1
2. Lynch J, Smith GD. **A life course approach to chronic disease epidemiology**. *Annu Rev Public Health* (2005.0) **26** 1-35. DOI: 10.1146/annurev.publhealth.26.021304.144505
3. de Jong F, Monuteaux MC, van Elburg RM, Gillman MW, Belfort MB. **Systematic review and meta-analysis of preterm birth and later systolic blood pressure**. *Hypertension* (2012.0) **59** 226-234. DOI: 10.1161/HYPERTENSIONAHA.111.181784
4. 4.Wang X, Dong Y, Zou Z, Ma J, Yang Z, Gao D, et al. Low birthweight is associated with higher risk of high blood pressure in Chinese girls: results from a national cross-sectional study in China. Int J Environ Res Public Health. 2019;16(16):2898.
5. Carr H, Cnattingius S, Granath F, Ludvigsson JF, Edstedt Bonamy AK. **Preterm birth and risk of heart failure up to early adulthood**. *J Am Coll Cardiol* (2017.0) **69** 2634-2642. DOI: 10.1016/j.jacc.2017.03.572
6. Lewandowski AJ, Augustine D, Lamata P, Davis EF, Lazdam M, Francis J. **Preterm heart in adult life: cardiovascular magnetic resonance reveals distinct differences in left ventricular mass, geometry, and function**. *Circulation* (2013.0) **127** 197-206. DOI: 10.1161/CIRCULATIONAHA.112.126920
7. Crump C, Howell EA, Stroustrup A, McLaughlin MA, Sundquist J, Sundquist K. **Association of preterm birth with risk of ischemic heart disease in adulthood**. *JAMA Pediatr* (2019.0) **173** 736-743. DOI: 10.1001/jamapediatrics.2019.1327
8. Lawlor DA, Ronalds G, Clark H, Smith GD, Leon DA. **Birth weight is inversely associated with incident coronary heart disease and stroke among individuals born in the 1950s: findings from the Aberdeen Children of the 1950s prospective cohort study**. *Circulation* (2005.0) **112** 1414-1418. DOI: 10.1161/CIRCULATIONAHA.104.528356
9. Larsson SC, Drca N, Jensen-Urstad M, Wolk A. **Incidence of atrial fibrillation in relation to birth weight and preterm birth**. *Int J Cardiol* (2015.0) **178** 149-152. DOI: 10.1016/j.ijcard.2014.10.138
10. Goss KN, Haraldsdottir K, Beshish AG, Barton GP, Watson AM, Palta M. **Association between preterm birth and arrested cardiac growth in adolescents and young adults**. *JAMA Cardiol* (2020.0) **5** 910-919. DOI: 10.1001/jamacardio.2020.1511
11. Fattal-Valevski A, Bassan H, Bernheim J, Redianu B, Leitner Y, Harel S. **Blood pressure values in 8–12 year old children with a history of intrauterine growth retardation**. *Isr Med Assoc J* (2011.0) **13** 480-484. PMID: 21910372
12. 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. DOI: 10.1056/NEJMoa1803527
13. Renom Espineira A, Fernandes-Rosa FL, Bueno AC, de Souza RM, Moreira AC, de Castro M. **Postnatal growth and cardiometabolic profile in young adults born large for gestational age**. *Clin Endocrinol (Oxf)* (2011.0) **75** 335-341. DOI: 10.1111/j.1365-2265.2011.04054.x
14. Wang H, Du B, Wu Y, Li Z, Niu Y, Ouyang F. **Sex-disparity in the association between birthweight and cardiovascular parameters in 4-year-old children: a Chinese cohort study**. *Front Nutr* (2021.0) **8** 756512. DOI: 10.3389/fnut.2021.756512
15. Hardy R, Kuh D, Langenberg C, Wadsworth ME. **Birthweight, childhood social class, and change in adult blood pressure in the 1946 British birth cohort**. *Lancet* (2003.0) **362** 1178-1183. DOI: 10.1016/S0140-6736(03)14539-4
16. Lawlor DA, Hübinette A, Tynelius P, Leon DA, Smith GD, Rasmussen F. **Associations of gestational age and intrauterine growth with systolic blood pressure in a family-based study of 386,485 men in 331,089 families**. *Circulation* (2007.0) **115** 562-568. DOI: 10.1161/CIRCULATIONAHA.106.646661
17. 17.Mohlkert LA, Hallberg J, Broberg O, Rydberg A, Halvorsen CP, Liuba P, et al. The preterm heart in childhood: left ventricular structure, geometry, and function assessed by echocardiography in 6-year-old survivors of periviable births. J Am Heart Assoc. 2018;7(2):e007742.
18. Edvardsson VO, Steinthorsdottir SD, Eliasdottir SB, Indridason OS, Palsson R. **Birth weight and childhood blood pressure**. *Curr Hypertens Rep* (2012.0) **14** 596-602. DOI: 10.1007/s11906-012-0311-6
19. Hardy R, Ghosh AK, Deanfield J, Kuh D, Hughes AD. **Birthweight, childhood growth and left ventricular structure at age 60–64 years in a British birth cohort study**. *Int J Epidemiol* (2016.0) **45** 1091-1102. PMID: 27413103
20. Lurbe E, Ingelfinger J. **Developmental and early life origins of cardiometabolic risk factors: novel findings and implications**. *Hypertension* (2021.0) **77** 308-318. DOI: 10.1161/HYPERTENSIONAHA.120.14592
21. Juonala M, Cheung MM, Sabin MA, Burgner D, Skilton MR, Kähönen M. **Effect of birth weight on life-course blood pressure levels among children born premature: the Cardiovascular Risk in Young Finns Study**. *J Hypertens* (2015.0) **33** 1542-1548. DOI: 10.1097/HJH.0000000000000612
22. Zhao Y, Wang SF, Mu M, Sheng J. **Birth weight and overweight/obesity in adults: a meta-analysis**. *Eur J Pediatr* (2012.0) **171** 1737-1746. DOI: 10.1007/s00431-012-1701-0
23. Zhang J, Tian Y, Wang W, Ouyang F, Xu J, Yu X. **Cohort profile: the Shanghai Birth Cohort**. *Int J Epidemiol.* (2019.0) **48** 21-g. DOI: 10.1093/ije/dyy277
24. Hui F, Yinkun Y, Jie Mi. **Blood pressure reference standard for Chinese children aged 3–17 by sex, age and height**. *Zhonghua Gao Xue Ya Za Zhi* (2017.0) **25** 428-435
25. Lai WW, Geva T, Shirali GS, Frommelt PC, Humes RA, Brook MM. **Guidelines and standards for performance of a pediatric echocardiogram: a report from the Task Force of the Pediatric Council of the American Society of Echocardiography**. *J Am Soc Echocardiogr* (2006.0) **19** 1413-1430. DOI: 10.1016/j.echo.2006.09.001
26. Wang J, Du B, Wu Y, Li Z, Chen Q, Zhang X. **Association of maternal gestational weight gain with left ventricle geometry and function in offspring at 4 years of age: a prospective birth cohort study**. *Front Pediatr* (2021.0) **9** 722385. DOI: 10.3389/fped.2021.722385
27. Drukteinis JS, Roman MJ, Fabsitz RR, Lee ET, Best LG, Russell M. **Cardiac and systemic hemodynamic characteristics of hypertension and prehypertension in adolescents and young adults: the Strong Heart Study**. *Circulation* (2007.0) **115** 221-227. DOI: 10.1161/CIRCULATIONAHA.106.668921
28. Cobble M, Bale B. **Carotid intima-media thickness: knowledge and application to everyday practice**. *Postgrad Med* (2010.0) **122** 10-18. DOI: 10.3810/pgm.2010.01.2091
29. Naqvi TZ, Lee MS. **Carotid intima-media thickness and plaque in cardiovascular risk assessment**. *JACC Cardiovasc Imaging* (2014.0) **7** 1025-1038. DOI: 10.1016/j.jcmg.2013.11.014
30. Ganau A, Devereux RB, Roman MJ, de Simone G, Pickering TG, Saba PS. **Patterns of left ventricular hypertrophy and geometric remodeling in essential hypertension**. *J Am Coll Cardiol* (1992.0) **19** 1550-1558. DOI: 10.1016/0735-1097(92)90617-V
31. Bowers K, Liu G, Wang P, Ye T, Tian Z, Liu E. **Birth weight, postnatal weight change, and risk for high blood pressure among Chinese children**. *Pediatrics* (2011.0) **127** e1272-e1279. DOI: 10.1542/peds.2010-2213
32. Lule SA, Namara B, Akurut H, Muhangi L, Lubyayi L, Nampijja M. **Are birthweight and postnatal weight gain in childhood associated with blood pressure in early adolescence? Results from a Ugandan birth cohort**. *Int J Epidemiol* (2019.0) **48** 148-156. DOI: 10.1093/ije/dyy118
33. Sutherland MR, Bertagnolli M, Lukaszewski MA, Huyard F, Yzydorczyk C, Luu TM. **Preterm birth and hypertension risk: the oxidative stress paradigm**. *Hypertension* (2014.0) **63** 12-18. DOI: 10.1161/HYPERTENSIONAHA.113.01276
34. Lewington S, Clarke R, Qizilbash N, Peto R, Collins R. **Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective studies**. *Lancet* (2002.0) **360** 1903-1913. DOI: 10.1016/S0140-6736(02)11911-8
35. Rotteveel J, Twisk JW. **Infant and childhood growth patterns, insulin sensitivity, and blood pressure in prematurely born young adults**. *Pediatrics.* (2008.0) **122** 313-21. DOI: 10.1542/peds.2007-2012
36. Willemsen RH, de Kort SW, van der Kaay DC, Hokken-Koelega AC. **Independent effects of prematurity on metabolic and cardiovascular risk factors in short small-for-gestational-age children**. *J Clin Endocrinol Metab* (2008.0) **93** 452-458. DOI: 10.1210/jc.2007-1913
37. Kanda T, Murai-Takeda A, Kawabe H, Itoh H. **Low birth weight trends: possible impacts on the prevalences of hypertension and chronic kidney disease**. *Hypertens Res* (2020.0) **43** 859-868. DOI: 10.1038/s41440-020-0451-z
38. Huxley R, Neil A, Collins R. **Unravelling the fetal origins hypothesis: is there really an inverse association between birthweight and subsequent blood pressure?**. *Lancet* (2002.0) **360** 659-665. DOI: 10.1016/S0140-6736(02)09834-3
39. Liu Y, Yan Y, Jiang T, Li S, Guo Y, Fernandez C. **Impact of long-term burden of body mass index and blood pressure from childhood on adult left ventricular structure and function**. *J Am Heart Assoc* (2020.0) **9** e016405. DOI: 10.1161/JAHA.120.016405
40. Gluckman PD, Seng CY, Fukuoka H, Beedle AS, Hanson MA. **Low birthweight and subsequent obesity in Japan**. *Lancet* (2007.0) **369** 1081-1082. DOI: 10.1016/S0140-6736(07)60524-8
41. Harder T, Schellong K, Stupin J, Dudenhausen JW, Plagemann A. **Where is the evidence that low birthweight leads to obesity?**. *Lancet* (2007.0) **369** 1859. DOI: 10.1016/S0140-6736(07)60847-2
42. Saenger P, Czernichow P, Hughes I, Reiter EO. **Small for gestational age: short stature and beyond**. *Endocr Rev* (2007.0) **28** 219-251. DOI: 10.1210/er.2006-0039
43. Dumeige L, Nehlich M, Viengchareun S, Perrot J, Pussard E, Lombès M. **Preterm birth is associated with epigenetic programming of transgenerational hypertension in mice**. *Exp Mol Med* (2020.0) **52** 152-165. DOI: 10.1038/s12276-020-0373-5
44. Sutherland MR, Gubhaju L, Black MJ. **Stereological assessment of renal development in a baboon model of preterm birth**. *Am J Nephrol* (2011.0) **33** 25-33. DOI: 10.1159/000327073
45. Wang J, Wu Y, Du B, Li Z, Ye Y, Wang H. **Growth patterns in early childhood and cardiovascular structure and function at 4 years old: A prospective cohort study**. *Nutr Metab Cardiovasc Dis* (2021.0) **31** 3492-3501. DOI: 10.1016/j.numecd.2021.08.028
46. Tang A, Slopen N, Nelson CA, Zeanah CH, Georgieff MK, Fox NA. **Catch-up growth, metabolic, and cardiovascular risk in post-institutionalized Romanian adolescents**. *Pediatr Res* (2018.0) **84** 842-848. DOI: 10.1038/s41390-018-0196-4
47. Jain V, Singhal A. **Catch up growth in low birth weight infants: striking a healthy balance**. *Rev Endocr Metab Disord* (2012.0) **13** 141-147. DOI: 10.1007/s11154-012-9216-6
48. Ong KK. **Catch-up growth in small for gestational age babies: good or bad?**. *Curr Opin Endocrinol Diabetes Obes* (2007.0) **14** 30-34. DOI: 10.1097/MED.0b013e328013da6c
49. Toemen L, Gaillard R, Roest AA, van der Geest RJ, Steegers EA, van der Lugt A. **Fetal and infant growth patterns and left and right ventricular measures in childhood assessed by cardiac MRI**. *Eur J Prev Cardiol* (2020.0) **27** 63-74. DOI: 10.1177/2047487319866022
50. Zhang T, Li S, Bazzano L, He J, Whelton P, Chen W. **Trajectories of childhood blood pressure and adult left ventricular hypertrophy: the Bogalusa Heart Study**. *Hypertension* (2018.0) **72** 93-101. DOI: 10.1161/HYPERTENSIONAHA.118.10975
51. Harris SL, Bray H, Troughton R, Elliott J, Frampton C, Horwood J. **Cardiovascular outcomes in young adulthood in a population-based very low birth weight cohort**. *J Pediatr* (2020.0) **225** 74-9.e3. DOI: 10.1016/j.jpeds.2020.06.023
52. Spalding KL, Arner E, Westermark PO, Bernard S, Buchholz BA, Bergmann O. **Dynamics of fat cell turnover in humans**. *Nature* (2008.0) **453** 783-787. DOI: 10.1038/nature06902
53. 53.Vollmer J, Casares F, Iber D. Growth and size control during development. Open Biol. 2017;7(11):170190.
54. Kapral N, Miller SE, Scharf RJ, Gurka MJ, DeBoer MD. **Associations between birthweight and overweight and obesity in school-age children**. *Pediatr Obes* (2018.0) **13** 333-341. DOI: 10.1111/ijpo.12227
55. Dhuper S, Abdullah RA, Weichbrod L, Mahdi E, Cohen HW. **Association of obesity and hypertension with left ventricular geometry and function in children and adolescents**. *Obesity (Silver Spring)* (2011.0) **19** 128-133. DOI: 10.1038/oby.2010.134
56. Cuspidi C, Rescaldani M, Sala C, Grassi G. **Left-ventricular hypertrophy and obesity: a systematic review and meta-analysis of echocardiographic studies**. *J Hypertens* (2014.0) **32** 16-25. DOI: 10.1097/HJH.0b013e328364fb58
57. Mangner N, Scheuermann K, Winzer E, Wagner I, Hoellriegel R, Sandri M. **Childhood obesity: impact on cardiac geometry and function**. *JACC Cardiovasc Imaging* (2014.0) **7** 1198-1205. DOI: 10.1016/j.jcmg.2014.08.006
58. Erbs S, Broniecki H, Scheuermann K, Winzer E, Adam J, Spielau U. **Impact of weight reduction during adolescence on parameters of cardiac geometry and function in obese children**. *JACC Cardiovasc Imaging* (2018.0) **11** 1915-1917. DOI: 10.1016/j.jcmg.2018.05.023
59. Wu Y, Li Z, Du B, Ye Y, Wang H, Niu Y. **Different associations of systolic blood pressure and body mass index with cardiac structure and function in young children**. *Hypertension* (2022.0) **79** 2583-2592. DOI: 10.1161/HYPERTENSIONAHA.122.19396
|
---
title: Loss of Müller cell glutamine synthetase immunoreactivity is associated with
neuronal changes in late-stage retinal degeneration
authors:
- Hallur Reynisson
- Michael Kalloniatis
- Erica L. Fletcher
- Mohit N. Shivdasani
- Lisa Nivison-Smith
journal: Frontiers in Neuroanatomy
year: 2023
pmcid: PMC10029270
doi: 10.3389/fnana.2023.997722
license: CC BY 4.0
---
# Loss of Müller cell glutamine synthetase immunoreactivity is associated with neuronal changes in late-stage retinal degeneration
## Abstract
### Introduction
A hallmark of photoreceptor degenerations is progressive, aberrant remodeling of the surviving retinal neurons and glia following photoreceptor loss. The exact relationship between neurons and glia remodeling in this late stage of retinal degeneration, however, is unclear. This study assessed this by examining Müller cell dysfunction via glutamine synthetase immunoreactivity and its spatial association with retinal neuron subpopulations through various cell markers.
### Methods
Aged Rd1 mice retinae (P150 – P536, n = minimum 5 per age) and control heterozygous rd1 mice retinae (P536, $$n = 5$$) were isolated, fixed and cryosectioned. Fluorescent immunolabeling of glutamine synthetase was performed and retinal areas quantified as having low glutamine synthetase immunoreactivity if proportion of labeled pixels in an area was less than two standard deviations of the mean of the total retina. Other Müller cell markers such as Sox9 and Glial fibrillary acidic protein along with neuronal cell markers Calbindin, Calretinin, recoverin, Protein kinase C-α, Glutamic acid decarboxylase 67, and Islet-1 were then quantified within areas of low and normal synthetase immunoreactivity.
### Results
Glutamine synthetase immunoreactivity was lost as a function of age in the rd1 mouse retina (P150 – P536). Immunoreactivity of other Müller cell markers, however, were unaffected suggesting Müller cells were still present in these low glutamine synthetase immunoreactive regions. Glutamine synthetase immunoreactivity loss affected specific neuronal populations: Type 2, Type 8 cone, and rod bipolar cells, as well as AII amacrine cells based on reduced recoverin, protein kinase Ca and parvalbumin immunoreactivity, respectively. The number of cell nuclei within regions of low glutamine synthetase immunoreactivity was also reduced suggesting possible neuronal loss rather than reduced cell marker immunoreactivity.
### Conclusion
These findings further support a strong interplay between glia-neuronal alterations in late-stage degeneration and highlight a need for future studies and consideration in intervention development.
## 1. Introduction
Retinitis Pigmentosa (RP) is the most common inherited retinal degeneration affecting one in 4,000 individuals (Hartong et al., 2006). It results from the death of photoreceptors leading to significant visual impairment and ultimately, blindness. A large body of evidence shows that the impact of RP continues to the inner retina which undergoes change secondary to photoreceptor loss (Jones et al., 2003; Marc et al., 2003; Strettoi et al., 2003; Jones and Marc, 2005; Marc et al., 2007; Chua et al., 2009, 2013; Kalloniatis et al., 2013; Greferath et al., 2015; Pfeiffer et al., 2020b; Strettoi et al., 2022). These changes are complex, and include anatomical, metabolic and functional alterations of both neurons and glia in the inner retina (Jones et al., 2003; Marc et al., 2003; Strettoi et al., 2003; Jones and Marc, 2005; Marc et al., 2007; Chua et al., 2009, 2013; Kalloniatis et al., 2013; Greferath et al., 2015).
An added complexity is that inner retinal changes are dependent on disease stage. For example, Chua et al. [ 2009] found that neurochemical remodeling of ionotropic glutamate receptors on bipolar cells in the rd1 mouse was only evident during active cone degeneration and was lost following total photoreceptor death. Marc et al. [ 2003] noted that specific, significant anatomical changes to the inner retina such as neuronal migration, cell death and glial seal completion only occurred late in the course of disease, well beyond photoreceptor death (Jones et al., 2003). Considering that many vision restoration strategies are targeted to individuals with well-established vision loss, furthering understanding of late-stage degeneration is most relevant to the successful development and deployment of such interventions.
Glial remodeling also exhibits separate, time dependent phases of change. For example, early degeneration is associated with loss of Müller cell processes and hyperexpression of glial fibrillary acidic protein (GFAP) while late degeneration is associated with the converse, Müller cell hypertrophy and glial seal formation (Bringmann et al., 2006; Chua et al., 2013). Metabolic profile and protein expression of Müller cells also becomes distinctly “chaotic” in late-stage degeneration. Specifically, metabolic amino acid signatures between neighboring Müller cells are significantly heterogenous with no obvious explanation for variation (Jones et al., 2016; Pfeiffer et al., 2020a). Major metabolic enzymes, cellular retinaldehyde-binding protein CRALBP and glutamine synthetase also undergo varying levels of loss with the latter demonstrated within human RP retinae (Jones et al., 2016), the P347L rabbit retinae (Pfeiffer et al., 2016, Pfeiffer et al., 2020b) and the rd1-Fos-Tau-LacZ (rd1-FTL) mouse retinae (Greferath et al., 2015). Greferath et al. [ 2015] further postulated that inner retinal neurons within regions of abnormal glutamine synthetase immunolabeling were abnormal based on FTL expression suggesting a potential cause-and-effect relationship to explain neuronal and glial changes in late-stage retinal degeneration. Such a relationship could help predict the course of inner retinal change in late-stage degeneration and be of significant benefit in guiding intervention approaches.
A quantitative time course of Müller cell dysfunction (based on loss of glutamine synthetase immunoreactivity) is currently unknown. However, qualitatively data from Pfeiffer et al. ( 2020b) presents a potential degeneration dependent loss in the P347L rabbit. Similarly, investigation of the specific neuronal populations which are present within regions of Müller cell dysfunction is also limited. Thus, the aim of this study was to determine the time course of glutamine synthetase loss in Müller cells in late-stage retinal degeneration and the identity the neuronal cell types remaining within these regions of altered Müller cell function.
## 2.1. Animals
Rd1 mice (on a C57Bl/6 background) (Farber et al., 1994) were studied at post-natal day P150 - P536 (n = minimum 5 per age). The control mice (C57Bl/6) were examined at the oldest age only (P536; $$n = 5$$). Animals were maintained on a 12 h light/dark cycle and had access to standard mouse chow and water ad libitum. The experimental protocols in this study were approved by the University of Melbourne and UNSW Sydney Animal Ethics committees.
## 2.2. Tissue fixation and immunolabeling
Mouse retinae were processed and immunostained as described previously (Nivison-Smith et al., 2013, 2014, 2015, 2017). Briefly, mice were killed by cervical dislocation. Eyes were enucleated immediately, and the anterior structures removed, under constant fluorescent room lighting, creating an eyecup preparation. Eyecups were then fixed for 30 min in $4\%$ (w/v) paraformaldehyde and $0.01\%$ (w/v) glutaraldehyde in 0.10 M phosphate buffer. Tissues were then washed in 0.10 M phosphate buffer before cryo-protection in graded $30\%$ (w/v) sucrose and cryo-sectioned in the vertical plane at a thickness of 120 μm.
For immunostaining, retinal sections were blocked for 60 min with $6\%$ (v/v) goat serum, $1\%$ (w/v) bovine serum albumin, $0.1\%$ (v/v) Triton-X then incubated overnight at 4°C with primary antibodies at the dilutions specified in Table 1. Primary antibodies were detected with anti-chicken AlexaFluor 488, anti-rabbit AlexaFluor 488 or 594, or anti-mouse AlexaFluor 405 or 488 (Thermo Fisher Scientific, Waltham, MA, USA). Sections were incubated with secondary antibodies at a 1:500 dilution for 2 h at room temperature. All antibody dilutions were made in $3\%$ (v/v) goat serum, $1\%$ (w/v) bovine serum albumin, $0.1\%$ (v/v) Triton-X. In a subset of samples, counterstaining was performed with 2-(4-amidinophenyl)-1H -indole-6-carboxamidine (DAPI) diluted 1:1000 in MilliQ water. All samples were mounted in Citifluor mounting media (ProSciTech, QLD, Australia). Sections were imaged using an FV1200 Scanning Laser Microscope (Olympus Australia, Notting Hill, VIC, Australia).
**TABLE 1**
| Antigen | Immunogen | Specificity* | Manufacturer, cat no. | Host | Dilution | Retinal cell types labeled | Length analysed (μm) |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Calbindin | Purified bovine kidney calbindin-D-28K | – | Sigma-Aldrich; C9848 | Ms; monoclonal | 1:1000 | Subpopulations of amacrine cells and horizontal cells | 7165 |
| Calretinin (CR) | Rat calretinin, amino acids 38–151 | The antibody binds to the Ca2+ binding protein Calretinin | BD Transduction; 610908 | Ms; monoclonal | 1:1000 | Populations of amacrine and ganglion cells | 24555 |
| Glial fibrillary acidic protein (GFAP) | GFAP from pig spinal cord | The antibody reacts specifically with GFAP in immunoblotting assays and labels astrocytes. | Sigma Aldrich; G3893 | Ms; monoclonal | 1:1000 | Müller cell presence | 6698 |
| Glutamic acid decarboxylase GAD67 | Synthetic peptide from mouse GAD67 (amino acids 87–106 | Reacts specifically with GAD67 | Sigma-Aldrich; G5419 | Ms; monoclonal | 1:500 | GABAergic amacrine cells | 7630 |
| Glutamine synthetase (GS) | Glutathione conjugated to glutaraldehyde | – | Abcam; ab93439 | Rb; polyclonal | 1:500 | Müller glia in terms of glutamate metabolism | 191030 |
| Islet-1 | Truncated rat islet protein corresponding to amino acids 178–349 | – | Developmental Studies Hybridoma Bank; 39.4D5 | Ms; monoclonal | 1:200 | ON bipolar cells | 14270 |
| Parvalbumin (PV) | Frog muscle parvalbumin | Recognizes parvalbumin in a Ca2+ ion-dependent manner. | Sigma-Aldrich; P3088 | Ms; monoclonal | 1:500 | Amacrine cells, specifically AII amacrine cells | 11860 |
| Protein kinase C-α (PKCα) | Purified bovine brain PKC | Reacts with the 80 kDa polypeptide of PKC. | Sigma-Aldrich; P5704 | Ms; monoclonal | 1:400 | Rod bipolar cells | 35620 |
| Recoverin | Recombinant human recoverin | Recognizes recoverin. | Chemicon (Millipore); AB5585 | Rb; polyclonal | 1:1000 | Cone bipolar cell type 2 and type 8 | 13590 |
| SOX9 | C-terminal sequence of human Sox9 | Recognizes Sox9 | Chemicon (Millipore); AB5535 | Rb; polyclonal | 1:2000 | Astrocytes and Müller cells | – |
All samples that were stained for the same antibody were processed at the same time with no difference in antibody batch number to address possible issues from histological preparation. To decrease effect of eccentricity all samples were taken within 500 μm of the central retina from temporal to nasal. Images were taken within a week of staining with fixed laser settings for DAPI, AlexaFluor488, and AlexaFluor594 labeling to decrease errors due to laser variability.
## 2.3.1. Sliding window and pixel threshold criteria
A systematic approach was developed to assess quantitatively the area specific differences in immunoreactivity within and between areas of the retina in the control mice at P536 and rd1 at P150, P300, and P536, in confocal images. Immunolabeling of cell markers was assessed from microscope images in a using a sliding window analysis in ImageJ (version 1.53n; provided in the public domain by the National Institutes of Health, Bethesda, MD, USA)1 and a custom MATLAB® (R2020b, v9.9.0, Mathworks, Natrick, MA, USA) script that utilized the Image Processing Toolbox.
The window width was set to the pixel equivalent of 25 μm, translated from the scale bar on each image, and only pixels within the region of the neural retina were counted in the window. The neural retina in each window was defined as the pixels contained between the inner limiting membrane to the outermost part of the neural retina. Due to degeneration in rd1 mice, this outer border was manually delineated by user tracing of the retina within the image. As window width was constant, total pixel count within a given window was a function of retinal thickness.
To quantify an immunolabeled cell marker within a given window, images were separated into RGB channels and then individual pixels (P) were thresholded based on their respective pixel values, PPV by the criterion where PT is the binary thresholded pixel value and MIPV is the mean pixel value of the image for their respective channel and immunolabel. In short, individual pixels were considered above threshold if their value was greater than twice that of the relevant mean pixel value. The positive pixel ratio, PPratio of each window for each channel or immunolabel was then determined by where n is the total number of pixels in the window counted as neural retina. In short, the positive pixel ratio revealed the proportion of pixels above threshold to the total number of pixels of the retina within the window, thus normalizing the positive pixel count to the thickness of the neural retina. The analysis was repeated across 25 μm windows across the whole length of the retina for each immunolabel in each microscope image using a window step size of 5 μm. In total, sliding window analysis was conducted on 37 images spanning 24,005 μm retinal length for control mice ($$n = 5$$ at P536), 90 images spanning 57,540 μm retinal length for rd1 mice at P150 ($$n = 5$$), 80 images spanning 51,050 μm retinal length for rd1 mice at P300 ($$n = 4$$) and 102 images spanning 58,435 μm of retina for rd1 mice at P536 ($$n = 6$$).
## 2.3.2. Defining areas of normal and low glutamine synthetase immunolabeling
After analyzing the GS immunolabeling for all 24,005 μm of control retinae, the GS labeling of an Area (A) of rd1 retina was defined as either normal (GSnorm) or low (GSlow) utilizing the mean and standard deviations obtained from the control using the formula where MPPratio is the mean positive pixel ratio of the GS immunolabel for the control retina and SPPratio is the standard deviation of the positive pixel ratio of the GS immunolabel for the control. In short, an area of rd1 retina was considered to have low GS immunolabeling if the proportion of GS pixels above threshold relative to total retina was less than the mean proportion of GS pixels relative to total retina by approximately two standard deviations or more.
## 2.3.3. Defining colocalization of cell markers
Pixel colocalization between any two immunolabels (for example labels A and B; Figures 1A, B and colocalisation in Figure 1C) in a retinal image was defined as a binary zero or one for each pixel within a retinal confocal image depending on whether its corresponding pixel values were over a set threshold (Figures 1D, E) such that
**FIGURE 1:** *Method for defining colocalization of immunoreactivity. (A) Retinal tissue was double labeled with glutamine synthetase (GS; magenta) and (B) glial fibrillary acidic protein (GFAP; green). (C) Merged channels by merging GS and GFAP from images (A,B). (D,E) Images (A,B) were thresholded with respect to the images’ mean pixel values to get binary maps of positive pixels. (F) The binary maps were then joined into a binary map where if and only if a pixel was positive for both GS and GFAP it was considered positive for colocalization of GS and GFAP. Note the similarities between the faint white in (C) and the binary map in (F). Scale bar is 50 μm.*
Where PC is the thresholded co-localization pixel, PAPV and PBPV are the individually thresholded pixel values for immunolabels A and B, respectively, and MIAPV and MIBPV are the mean pixel values in the image for immunolabels A and B, respectively (Figure 1F). In short, if a pixel in a two-channel image matrix was positive for both A and B it was a site of colocalization. Colocalization ratio was then quantified as pixel values over threshold for both immunolabels divided by the total number of pixels of the respective area. All colocalization results were normalized to normal GS areas within the same retinal image for their relative assessment between images. For the assessment of colocalization between GS and GFAP, we assessed 6,698 μm of GFAP labeled rd1 (P536; $$n = 5$$) retina (see Table 1).
## 2.3.4. Assessment of areas of normal and low glutamine synthetase immunolabeling
A total of 24,675 μm of rd1 retina (n = minimum 5 per age) was stained with DAPI for nuclear labeling. All positive pixel ratios, except GS, were normalized to the mean positive pixel ratios of normal GS areas within each retinal image to allow consistent, relative assessment of differences in other immunolabels between areas of normal vs. low GS areas within a retinal image. For the majority of cell markers, only retina at P536 were assessed as this was the only age that demonstrated numerous and long areas of low glutamine synthetase immunoreactivity that could be reliably quantified.
## 2.4. Statistical analysis
All variables are expressed as mean ± standard error. Data was analysed using the two-sample t-test and one way analysis of variance (ANOVA), with an α of 0.05. For t-tests with multiple numbers of tests the α was adjusted to the Bonferroni adjusted α (αB) of 0.0125 and 0.0167, when the number of tests were 4 and 3, respectively. For multiple groups a Tukey-Kramer post-hoc test was performed. Statistical analyses were performed using MATLAB® (Mathworks, Natrick, MA, USA) v9.9.0.1570001 (R2020b) and the Statistics and Machine Learning Toolbox v12.0.
## 3.1. Glutamine synthetase immunoreactivity is reduced in the rd1 mouse as a function of age
The rd1 retinae was assessed at post-natal days P150, P300, and P536 and compared to control tissue at post-natal day P536 (Figures 2A–D). Total thickness of the rd1 retina was significantly reduced at all time points compared to control (t-test, all $p \leq 0.0001$, αB = 0.0125; Figure 2E). However, there was no significant difference in thickness between each of the rd1 age groups (t-test, $p \leq 0.0125$, αB = 0.0125; Tukey-Kramer, $p \leq 0.05$), though some regional thickening was qualitatively observed in low GS regions (Figure 2D). Thickness of individual retinal layers was not explored due to nuclear migration which prohibited accurate manual layer segmentation.
**FIGURE 2:** *Glutamine synthetase immunoreactivity as a function of age in the rd1 retina. Representative images from a (A) healthy control retina at P536, and rd1 retina at (B) P150, (C) P300 and (D) P536 labeled for glutamine synthetase (GS; red) and 4’,6-diamidino-2-phenylindole (DAPI; blue). Scale bar is 50 μm. (E) Quantification of total retinal thickness confirming loss in the rd1 retina at all time points compared to control (n = 5, 5, 4, and 6 mice; mean ± SEM = 170.0 ± 7.4, 95.1 ± 4.7, 98.0 ± 1.4, and 84.3 ± 4.1; for C57Bl/6, rd1 P150, P300, P536, respectively). (F) Quantification of positive pixel ratio for GS demonstrating gradual loss in the rd1 retina relative to control (n = 5, 5, 4, and 6 mice; mean ± SEM = 31.6 ± 2.5, 24.2 ± 3.0, 22.7 ± 2.7, and 12.2 ± 1.6%; for C57Bl/6, rd1 P150, P300, P536, respectively). (G) Quantification of total amount of GSlow areas in rd1 retina defined as area where GS positive pixel count was below the mean-1.96 × SD of the control retina (n = 5, 4, and 6 mice; mean ± SEM = 10.4 ± 11.1, 17.7 ± 8.3, and 51.9 ± 8.2%; for rd1 P150, P300, P536, respectively). (H) Quantification of number of GSlow areas in the rd1 retina per cm (n = 4, 2, and 5 mice; mean ± SEM = 10.6 ± 6.5, 48.2 ± 4.5, and 139.4 ± 43.7 patches/cm; for rd1 P150, P300, P536, respectively). All data is presented as mean ± SEM. Statistical comparisons were performed via t-test with αB = 0.0125 for (E,F) and αB = 0.0167 for (G,H) and. All significant p-values are annotated on graphs. INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer; *p ≤ 0.05; **p ≤ 0.01; ****p ≤ 0.0001.*
Glutamine Synthetase immunoreactivity was not significantly different between the rd1 P150 and control retinae (t-test, $$p \leq 0.73$$; Figures 2A, B, F). At P300, there was evidence of areas of the rd1 retina where GS immunoreactivity was absent or low compared to control retinae (Figure 2C). Quantification confirmed this with a $28\%$ loss in GS positive pixel ratio in the rd1 retina at P300 relative to control, however, this was not significant following Bonferroni adjusted α (t-test, $p \leq 0.05$, αB = 0.0125; Figure 2F). At P536, GS immunoreactivity further decreased (Figure 2D) with a significant $61\%$ loss in GS positive pixel ratio relative to control (t-test, $p \leq 0.0001$, αB = 0.0125; Tukey-Kramer, $p \leq 0.0001$). The standard deviation for the Positive Pixel ratio of GS in the control was 0.565. For the rd1 at P150, P300, and P536 it was 0.0666, 0.0537, and 0.0400, respectively, such that raw variability was similar for all groups.
Areas of low or absent GS immunoreactivity in the rd1 retina increased in length and number as a function of age. Specifically, areas of low GS immunoreactivity grew from 10.4 ± $5.0\%$ of the total retina at P150 to 51.9 ± $8.2\%$ by P536 (t-test, $p \leq 0.0167$, αB = 0.0167; Turkey-Kramer, $p \leq 0.01$; Figure 2G). Similarly, the number of individual areas of low GS immunoreactivity per unit length of retina significantly increased 13-fold from P150 to P536 (t-test, $p \leq 0.0167$, αB = 0.0167; Tukey-Kramer, $p \leq 0.05$; Figure 2H).
## 3.2. Müller cells are conserved in areas of low glutamine synthetase immunoreactivity
To determine if loss of GS immunoreactivity reflected loss of Müller cells, we assessed immunoreactivity of two alternative Müller cell markers, glial fibrillary acidic protein (GFAP) and Sox9. Despite the observation of numerous areas of low GS immunoreactivity in rd1 retina at P536 (Figure 3A), GFAP immunoreactivity was present throughout the retina and followed the Müller cell trunk, spanning the entire retinal thickness. In areas of normal GS immunoreactivity, GFAP was co-localized with GS suggesting both markers were likely reflective of Müller cell presence (Figure 3A). However, in areas of low GS immunoreactivity, the colocalization was reduced by over $50\%$ (t-test, $p \leq 0.0001$; Figure 3C). When quantified, there was no significant difference in GFAP immunolabeling in areas of low GS immunoreactivity compared to those with normal GS immunoreactivity (t-test, $$p \leq 0.49$$; Figure 3B) which suggests that Müller cells were still present in areas of low GS immunoreactivity.
**FIGURE 3:** *Representative images of the rd1 retina at P536 labeled for (A) glutamine synthetase (GS; magenta), glial fibrillary acidic protein (GFAP; green), GFAP colocalized with GS (white), and 4’,6-diamidino-2-phenylindole (DAPI) counterstaining (blue). Dotted boxes indicated a representative area of normal and low GS immunoreactivity, respectively. (B) Graph depicting normalized GFAP immunoreactivity in normal versus low GS immunoreactivity areas (n = 7 eyes; GSnorm = 1 ± 0.10, GSlow = 0.91 ± 0.12; p = 0.49). (C) Graph depicting the normalized colocalization of pixels for GS and GFAP (n = 7 eyes; GSnorm = 1 ± 0.07, GSlow = 0.35 ± 0.10). (D) Representative image of the rd1 retina at P536 labeled for glutamine synthetase (GS; magenta), Sox9 (green), and DAPI (blue). (E) Graph depicting normalized Sox9 immunoreactivity in normal versus low GS immunoreactivity areas (n = 5 eyes; GSnorm = 1 ± 0.13, GSlow = 1.27 ± 0.19; p = 0.20). (F) Graph depicting the cell count of Sox9 positive cells per 100 μm of retina for areas of normal and low GS immunoreactivity (n = 5 eyes; GSnorm = 8.92 ± 0.81, GSlow = 8.11 ± 0.92 Sox9 positive cells/100 μm; p = 0.52). All GSlow pixel ratio data is presented as mean ± SEM. Statistical comparisons were performed via t-test with α = 0.05. Only significant p-values are annotated on graphs, all other p-values are noted in this legend. Scale bar is 50 μm; GSnormal, area of normal GS expression; GSlow, area of low GS expression; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer; ****p ≤ 0.0001.*
Similarly, Sox9 immunoreactivity was present throughout the rd1 retina at P536 and in areas where normal GS immunoreactivity was present, Sox9 was assessed in normal and low GS areas (Figures 3D–F). When quantified, there was no significant difference in Sox9 immunolabeling in areas of low GS immunoreactivity compared to normal GS immunoreactivity in terms of positive pixel ratio (t-test, $$p \leq 0.20$$; Figure 3E), nor cell count (t-test, $$p \leq 0.52$$; Figure 3F). This further supports the notion that Müller cells were still present in areas of low GS immunoreactivity.
## 3.3. Loss of GS immunoreactivity disproportionately affects neuronal populations
Glutamine synthetase contributes to a major function of Müller cells: the clearance of extracellular glutamate to maintain retinal neurons and their microenvironment (Matsui et al., 1999; Bringmann et al., 2006; Kalloniatis et al., 2013). Thus, localized loss of GS in the rd1 retina could significantly affect surrounding neural tissue through elevated extracellular glutamate levels (Robin and Kalloniatis, 1992; Lieth et al., 1998; Dkhissi et al., 1999; Delyfer et al., 2005). To assess this, we analyzed major retinal neural populations within areas of low GS immunoreactivity using established cell markers (Table 1) and compared them to normal areas of GS expression in the rd1 retina at P536.
## 3.3.1. Bipolar cell marker immunoreactivity is reduced in areas of low GS immunoreactivity
Bipolar cells were assessed using cell markers against Recoverin (labels cone bipolar cell type 2 and type 8), protein kinase C-alpha (labels rod bipolar cells; PKC-α) and Islet-1 (labels all ON bipolar cells). Recoverin immunoreactivity was significantly decreased by $30\%$ in areas of low GS immunoreactivity compared to normal GS immunoreactivity (t-test, $p \leq 0.01$; Figures 4A, B). A manual count of Recoverin labeled cells supported those findings (GSnormal, 2.73 ± 0.31 Recoverin positive cells/100 μm; GSlow, 1.57 ± 0.30 Recoverin positive cells/100 μm, t-test, $p \leq 0.05$). Similarly, PKC-α immunoreactivity was significantly decreased by $13\%$ in areas of low versus normal GS immunoreactivity in the rd1 retina (t-test, $p \leq 0.001$; Figures 4C, D). There was a $27\%$ decrease in Islet-1 immunoreactivity, however, this difference was not significant (t-test, $$p \leq 0.099$$; Supplementary Figures 1A, B).
**FIGURE 4:** *Representative images of the rd1 retina at P536 labeled for glutamine synthetase (GS; magenta) and (A) Recoverin (green), (C) PKCα (green), (E) Parvalbumin (green), or (G) 4’,6-diamidino-2-phenylindole (DAPI) (blue). Scale bar is 50 μm. Graphs on the right show normalized (B) Recoverin (n = 11 eyes; GSnorm = 1 ± 0.09; GSlow = 0.71 ± 0.08), (D) PKCα (n = 11 eyes; GSnorm = 1 ± 0.04; GSlow = 0.87 ± 0.03), (F) Parvalbumin (n = 8 eyes; GSnorm = 1 ± 0.09; GSlow = 0.62 ± 0.13), and (H) DAPI (n = 7 eyes; GSnorm = 1 ± 0.03; GSlow = 0.88 ± 0.03) positive pixels within areas of normal versus low GS immunoreactivity. All GSlow data is presented as mean ± SEM. Statistical comparisons were performed via t-test with α = 0.05. All significant p-values are annotated on graphs. GSnormal, area of normal GS expression; GSlow, area of low GS expression; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer; *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001.*
## 3.3.2. Select amacrine cell markers are reduced in areas of low GS immunoreactivity
Changes in GS immunoreactivity also impacted certain populations of amacrine cells, specifically AII amacrine cells based on a significant $40\%$ reduction in Parvalbumin immunoreactivity in areas of low GS immunoreactivity, relative to normal GS immunoreactive areas (t-test, $p \leq 0.05$; Figures 4A, B). Assessment of horizontal cells and a subpopulation of amacrine cells via the cell marker Calbindin revealed no significant difference between areas of normal and low GS immunoreactivity (t-test, $$p \leq 0.082$$; Supplementary Figures 1C, D). Calretinin, which stains amacrine cells in the inner nuclear layer as well as wide-field amacrine cells and some ganglion cells in the ganglion cell layer, was also not significantly different between the two areas (t-test, $$p \leq 0.058$$; Supplementary Figures 1E, F). Finally, there was no significant difference in Glutamic acid decarboxylase 67 (GAD67) immunoreactivity in areas of low versus normal GS immunoreactivity, suggesting alterations in GS immunoreactivity did not affect GABAergic amacrine cells (t-test, $$p \leq 0.955$$; Supplementary Figures 1G, H).
## 3.3.3. Nuclei based immunoreactivity is reduced in areas of low GS immunoreactivity
Finally, we explored if the loss of Recoverin, PKCα and PV cell marker immunoreactivity in areas of low GS immunoreactivity reflected the loss of the specific neuron population or simply loss of the cell marker. By assessing DAPI we found that areas of low GS immunoreactivity had a significant $12\%$ decrease in total DAPI immunoreactivity relative to areas of normal GS immunoreactivity ($p \leq 0.01$; Figures 4G, H). This suggests that the loss of neurochemical markers noted above was due to loss of somata rather than a reduction in immunoreactivity alone.
## 4. Discussion
Diseases that result in photoreceptor death such as RP lead to aberrant functional and anatomical changes that follow a stage wise progression. The progression of these stages differs between the neural and glial population in the retina. While previous work indicates that GS expression remains stable in wildtype mice central nervous system for up to 18 months, (Olabarria et al., 2011), our results show that GS immunoreactivity is progressively diminished in the rd1 retina as areas of low GS both increase in number and size as the disease progresses.
Unaltered immunoreactivity of GFAP and Sox9 between regions of normal and low GS immunoreactivity in the rd1 retina suggest Müller cells were still however, present in these regions. GS immunoreactivity only affected some subpopulations of retinal neurons with reduction in bipolar cell marker immunoreactivity and AII amacrine cell marker immunoreactivity. Reduced DAPI labeling within low GS regions further suggested loss of cell marker immunoreactivity was at least in part attributed cell loss rather than a reduction in immunoreactivity alone.
## 4.1. Reduced GS immunoreactivity likely reflects Müller cell dysfunction, not absence in the rd1 retina
This study found GS immunoreactivity was reduced in the rd1 retina as a function of disease progress. Reduction in GS expression has been previously noted following photoreceptor degeneration (Härtig et al., 1995; Greferath et al., 2015; Jones et al., 2016; Pfeiffer et al., 2016, 2020b), retinal injury (Grosche et al., 1995), and retinal detachment (Lewis et al., 1989) and is possibly a consequence of the glutamate-release from dying photoreceptors. We qualitatively observed regional thickening of the retina in areas of low GS immunoreactivity however, quantitatively, no significant change in the overall thickness of the rd1 retina was found from P150 to P536, which suggested that strategies to minimize cell death in the INL may be of little value for RP treatments aimed at these stages of degeneration. Loss of GS immunoreactivity in the rd1 mouse appeared to reflect Müller cell dysfunction rather than Müller cell loss as immunoreactivity of other Müller cell markers, notably GFAP and Sox9, were not lost across the rd1 retina. Greferath et al. [ 2015] reported similar findings in the rd1-FTL retina. The exact process of dysfunction in non GS immunoreactive Müller cells is unclear however, it has been postulated that absence of this well-established metabolic pathway leads to “unmasking” of alternative, less energetically favorable metabolic pathways that attempt to continue the glutamate-glutamine cycle (Pfeiffer et al., 2020a).
Loss of GS immunoreactivity was highly variable across the retina with areas of low GS immunoreactivity immediately flanked by areas of GS immunoreactivity that were comparable to age-matched control retinae. While similar observations were made in the human RP, the P34TL rabbit and the rd1-FTL mouse (Jones et al., 2016; Pfeiffer et al., 2016, Pfeiffer et al., 2020b; Greferath et al., 2015), to our knowledge, this is the first attempt to specifically quantify this variable GS immunoreactivity loss in late-stage retinal degeneration of the rd1 mouse. Our analysis found that both the number and size of low GS immunoreactive areas increased significantly as a function of age and that by P536, $50\%$ of the total retinal area demonstrated low GS immunoreactivity.
A possible reason for “patchy” loss of GS immunoreactivity in the rd1 retina could be the nature of photoreceptor loss which is also variable. This is supported by Greferath et al. [ 2015] who demonstrated that normal glutamine synthetase immunoreactivity was only maintained in the rd1-FTL retina in areas with remnant cone photoreceptor terminals. While we did not investigate GS immunoreactivity relative to any synaptic markers in this study, previous work indicates up to $5\%$ of cone photoreceptors remain in the rd1 retina by P536 and therefore a spatial association between GS immunoreactivity and photoreceptor degeneration could potentially exist up to this stage.
Variable photoreceptor degeneration may have induced GS loss through uneven glutamate release and subsequent response from Müller cells (Härtig et al., 1995). However, Jones et al. [ 2003] noted that, in some instances, glutamate levels are elevated in Müller cells of degenerate retinae which would suggest greater demand for GS. Alternatively, release of basic fibroblast growth factor (bFGF) in response to neuronal damage (Gao and Hollyfield, 1995, 1996; Cao et al., 1997) could decrease GS expression in Müller cells through activation of the c-Jun signaling pathway (Kruchkova et al., 2001). Disruption of GS expression has been shown to lead to a breakdown of the blood-retinal barrier (Shen et al., 2010) which could lead to further cell death and release of bFGF, thus creating a feedback loop that contributes to the spread of low GS areas with disease progression.
## 4.2. Areas of low GS immunoreactivity display advanced features of degeneration
Greferath et al. [ 2015] found that in the rd1 FTL mouse, c-fos was increased within areas of low GS immunoreactivity which suggested greater neural activity, cell death, and/or plasticity (Rich et al., 1997). However, no individual neuronal populations have been previously analyzed in areas of low GS immunoreactivity. In this study, we found the immunoreactivity of bipolar cell markers Recoverin and PKCα and the AII amacrine cell marker Parvalbumin were significantly reduced in areas of low GS immunoreactivity. Nuclei labeling was also significantly reduced in areas of low GS immunoreactivity suggesting that loss of these markers at least partly reflected cell loss rather than solely a reduction in cell marker immunoreactivity. Alternatively, reduction in marker immunoreactivity and cell nuclei number in low GS immunoreactive areas could be a consequence of neuronal migration described in advanced retinal remodeling (Jones et al., 2003). The direction of cell migration in retinal degeneration has been previously described to follow along glial surfaces to ectopic sites, which is inconsistent with the decreased DAPI observed in low GS immunoreactive areas in this study. However, retinal migration is still poorly understood with a number of different migration patterns observed including evidence of neurons migrating out of the degenerating retina via the choroid (Sullivan et al., 2003).
The loss of PKCα and Parvalbumin immunoreactivity which labels rod bipolar cells and AII amacrine cells, respectively, suggests that inner retinal neurons involved in the rod-mediated pathway are significantly affected by low GS immunoreactivity. Reduction in Recoverin immunoreactivity may reflect this as well, as type 2 bipolar cells have a high degree of chemical synaptic contacts with AII amacrine cells (Tsukamoto and Omi, 2017). Greater impairment of the rod-mediated pathway also aligns with the model of degeneration as rod loss precedes cone death in the rd1 retina (Jiménez et al., 1996).
## 4.3. Retinal degeneration as a feedback loop
Based on the results of this study we postulate that following photoreceptor death, areas of low GS develop. These areas become more numerous and expand, which likely leads to excess extracellular glutamate. As a result, local populations of glutamatergic neurons to become hyperactive and possibly excitotoxic. This could explain the increased c-fos labeling in the rd1-FTL mouse in regions of GS loss (Greferath et al., 2015) and the reduction in amacrine and bipolar cell marker immunoreactivity seen in this study. Altered glutamate homeostasis could also shed light on the canonical glutamate receptor function and class switching seen in late-stage retinal degeneration and remodeling (Puthussery et al., 2009; Jones et al., 2016). Finally, death or dysfunction of select amacrine and bipolar cell populations within areas of low GS could lead to the release of exogenous bFGF, which would feedback to further decreasing GS expression and exacerbating remodeling. Future work is needed to characterize Müller cells and retinal neuron subtypes within these regions as it will help build our understanding of these glia-neuronal alterations reported in late-stage degeneration.
## 4.4. Limitations
Due to the late stage of degeneration, we could not easily make a distinction between the inner nuclear layer, inner plexiform layer, and ganglion cell layer in the rd1 retina and thus assessed all retinal layers as one. As a result, we are limited in the conclusions we can make with regard to retinal migration. We were also unable to account for any lateral displacement meaning neurons which originate from low GS immunoreactive areas but migrate to normal GS immunoreactive areas were counted as the later. We believe this effect however, was minimal based on regions of low GS immunoreactivity being large in the P536 retina and very few quantifying markers being at the borders of low and normal GS areas.
Another by-product is that small changes in immunoreactivity in selected neuronal populations may have been masked by cell markers that label multiple cell subtypes. For example, we found no significant difference in Islet-1 immunoreactivity in areas of normal vs. low GS immunoreactivity despite evidence of rod bipolar cell loss through PKCα labeling. This may have been due to Islet-1 labeling of surviving amacrine cells which were found to be unaffected by GS immunoreactivity loss. Similarly, Calretinin immunoreactivity was not significantly altered between normal and low GS immunoreactive areas. However, this could have been due to labeling displaced amacrine cells and ganglion cells in the ganglion cell layer. Despite this our sliding window quantitative analysis ensured our whole retina analysis was a systematic and unbiased evaluation of immunoreactivity across the retina. Our analysis did not assess variability as a function of eccentricity as all samples were within 500 μm of the central retina. Future work with more specific cell markers could determine if other neuronal populations are affected by GS immunoreactivity changes.
## 5. Conclusion
Glutamine synthetase (GS) is lost in the rd1 retina as discrete regions that increased in both size and number as a function of age. This loss is not likely due to Müller cell loss as GFAP immunoreactivity was unaltered. Specific loss of neural macromolecular markers pertaining to specific amacrine and bipolar cell populations occurred in areas of low GS immunoreactivity and this was likely, in part, due to neuronal death based on decreased DAPI labeling. These data shed light on glia-neuronal alterations in late-stage degeneration and could provide insight for interventions to combat them.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
This animal study was reviewed and approved by University of Melbourne and University of New South Wales Animal Ethics Committees.
## Author contributions
LN-S performed the tissue preparation, immunostaining, and imaging. HR performed the quantification, programming, and statistical analysis. EF supplied the mice. MS provided study funding in part. 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.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnana.2023.997722/full#supplementary-material
## References
1. Bringmann A., Pannicke T., Grosche J., Francke M., Wiedemann P., Skatchkov S. N.. **Müller cells in the healthy and diseased retina.**. (2006) **25** 397-424
2. Cao W., Wen R., Li F., Lavail M. M., Steinberg R. H.. **Mechanical injury increases bFGF and CNTF mRNA expression in the mouse retina.**. (1997) **65** 241-248. DOI: 10.1006/exer.1997.0328
3. Chua J., Fletcher E. L., Kalloniatis M.. **Functional remodeling of glutamate receptors by inner retinal neurons occurs from an early stage of retinal degeneration.**. (2009) **514** 473-491. DOI: 10.1002/cne.22029
4. Chua J., Nivison-Smith L., Fletcher E. L., Trenholm S., Awatramani G. B., Kalloniatis M.. **Early remodeling of Müller cells in the rd/rd mouse model of retinal dystrophy.**. (2013) **521** 2439-2453. DOI: 10.1002/cne.23307
5. Delyfer M. N., Forster V., Neveux N., Picaud S., Léveillard T., Sahel J. A.. **Evidence for glutamate-mediated excitotoxic mechanisms during photoreceptor degeneration in the rd1 mouse retina.**. (2005) **11** 688-696. PMID: 16163266
6. Dkhissi O., Chanut E., Wasowicz M., Savoldelli M., Nguyen-Legros J., Minvielle F.. **Retinal TUNEL-positive cells and high glutamate levels in vitreous humor of mutant quail with a glaucoma-like disorder.**. (1999) **40** 990-995. PMID: 10102297
7. Farber D. B., Flannery J. G., Bowes-Rickman C.. **The rd mouse story: Seventy years of research on an animal model of inherited retinal degeneration**. (1994) **13** 31-64
8. Gao H., Hollyfield J. G.. **Basic fibroblast growth factor in retinal development: differential levels of bFGF expression and content in normal and retinal degeneration (rd) mutant mice.**. (1995) **169** 168-184. DOI: 10.1006/dbio.1995.1135
9. Gao H., Hollyfield J. G.. **Basic fibroblast growth factor: increased gene expression in inherited and light-induced photoreceptor degeneration.**. (1996) **62** 181-189. PMID: 8698078
10. Greferath U., Anderson E., Jobling A., Vessey K., Martinez G., de Iongh R.. **Inner retinal change in a novel rd1-FTL mouse model of retinal degeneration.**. (2015) **9**. DOI: 10.3389/fncel.2015.00293
11. Grosche J., Härtig W., Reichenbach A.. **Expression of glial fibrillary acidic protein (GFAP), glutamine synthetase (GS), and Bcl-2 protooncogene protein by Müller (glial) cells in retinal light damage of rats.**. (1995) **185** 119-122. DOI: 10.1016/0304-3940(94)11239-f
12. Härtig W., Grosche J., Distler C., Grimm D., el-Hifnawi E., Reichenbach A.. **Alterations of Müller (glial) cells in dystrophic retinae of RCS rats.**. (1995) **24** 507-517. DOI: 10.1007/BF01179976
13. Hartong D. T., Berson E. L., Thaddeus P. D.. **Retinitis pigmentosa.**. (2006) **368** 1795-1809. PMID: 17113430
14. Jiménez A. J., García-Fernández J. M., González B., Foster R. G.. **The spatio-temporal pattern of photoreceptor degeneration in the aged rd/rd mouse retina.**. (1996) **284** 193-202. DOI: 10.1007/s004410050579
15. Jones B. W., Marc R. E.. **Retinal remodeling during retinal degeneration.**. (2005) **81** 123-137. PMID: 15916760
16. Jones B. W., Pfeiffer R. L., Ferrell W. D., Watt C. B., Marmor M., Marc R. E.. **Retinal remodeling in human retinitis pigmentosa.**. (2016) **150** 149-165. PMID: 27020758
17. Jones B. W., Watt C. B., Frederick J. M., Baehr W., Chen C. K., Levine E. M.. **Retinal remodeling triggered by photoreceptor degenerations.**. (2003) **464** 1-16. PMID: 12866125
18. Kalloniatis M., Loh C. S., Acosta M. L., Tomisich G., Zhu Y., Nivison-Smith L.. **Retinal amino acid neurochemistry in health and disease.**. (2013) **96** 310-332
19. Kruchkova Y., Ben-Dror I., Herschkovitz A., David M., Yayon A., Vardimon L.. **Basic fibroblast growth factor: a potential inhibitor of glutamine synthetase expression in injured neural tissue.**. (2001) **77** 1641-1649. DOI: 10.1046/j.1471-4159.2001.00390.x
20. Lewis G. P., Erickson P. A., Guérin C. J., Anderson D. H., Fisher S. K.. **Changes in the expression of specific Müller cell proteins during long-term retinal detachment.**. (1989) **49** 93-111. DOI: 10.1016/0014-4835(89)90079-1
21. Lieth E., Barber A. J., Xu B., Dice C., Ratz M. J., Tanase D.. **Glial reactivity and impaired glutamate metabolism in short-term experimental diabetic retinopathy. Penn State Retina Research Group.**. (1998) **47** 815-820. DOI: 10.2337/diabetes.47.5.815
22. Marc R. E., Jones B. W., Anderson J. R., Kinard K., Marshak D. W., Wilson J. H.. **Neural reprogramming in retinal degeneration.**. (2007) **48** 3364-3371. PMID: 17591910
23. Marc R. E., Jones B. W., Watt C. B., Strettoi E.. **Neural remodeling in retinal degeneration.**. (2003) **22** 607-655. PMID: 12892644
24. Matsui K., Hosoi N., Tachibana M.. **Active role of glutamate uptake in the synaptic transmission from retinal nonspiking neurons.**. (1999) **19** 6755-6766. DOI: 10.1523/JNEUROSCI.19-16-06755.1999
25. Nivison-Smith L., Acosta M. L., Misra S., O’Brien B. J., Kalloniatis M.. **Vinpocetine regulates cation channel permeability of inner retinal neurons in the ischaemic retina.**. (2014) **66** 1-14. PMID: 24412512
26. Nivison-Smith L., Khoo P., Acosta M. L., Kalloniatis M.. **Pre-treatment with vinpocetine protects against retinal ischemia.**. (2017) **154** 126-138. DOI: 10.1016/j.exer.2016.11.018
27. Nivison-Smith L., O’Brien B. J., Truong M., Guo C. X., Kalloniatis M., Acosta M. L.. **Vinpocetine modulates metabolic activity and function during retinal ischemia.**. (2015) **308** C737-C749. PMID: 25696811
28. Nivison-Smith L., Sun D., Fletcher E. L., Marc R. E., Kalloniatis M.. **Mapping kainate activation of inner neurons in the rat retina.**. (2013) **521** 2416-2438. DOI: 10.1002/cne.23305
29. Olabarria M., Noristani H. N., Verkhratsky A., Rodríguez J. J.. **Age-dependent decrease in glutamine synthetase expression in the hippocampal astroglia of the triple transgenic Alzheimer’s disease mouse model: mechanism for deficient glutamatergic transmission.**. (2011) **6**. DOI: 10.1186/1750-1326-6-55
30. Pfeiffer R. L., Marc R. E., Jones B. W.. **Persistent remodeling and neurodegeneration in late-stage retinal degeneration.**. (2020b) **74**. DOI: 10.1016/j.preteyeres.2019.07.004
31. Pfeiffer R. L., Marc R. E., Jones B. W.. **Müller cell metabolic signatures: evolutionary conservation and disruption in disease.**. (2020a) **31** 320-329. DOI: 10.1016/j.tem.2020.01.005
32. Pfeiffer R. L., Marc R. E., Kondo M., Terasaki H., Jones B. W.. **Müller cell metabolic chaos during retinal degeneration.**. (2016) **150** 62-70. DOI: 10.1016/j.exer.2016.04.022
33. Puthussery T., Gayet-Primo J., Pandey S., Duvoisin R. M., Taylor W. R.. **Differential loss and preservation of glutamate receptor function in bipolar cells in the rd10 mouse model of retinitis pigmentosa.**. (2009) **29** 1533-1542. DOI: 10.1111/j.1460-9568.2009.06728.x
34. Rich K. A., Zhan Y., Blanks J. C.. **Aberrant expression of c-Fos accompanies photoreceptor cell death in the rd mouse.**. (1997) **32** 593-612. DOI: 10.1002/(sici)1097-4695(19970605)32:6<593::aid-neu5>3.0.co;2-v
35. Robin L. N., Kalloniatis M.. **Interrelationship between retinal ischaemic damage and turnover and metabolism of putative amino acid neurotransmitters, glutamate and GABA.**. (1992) **80** 273-300. DOI: 10.1007/BF00154376
36. Shen W., Li S., Chung S. H., Gillies M. C.. **Retinal vascular changes after glial disruption in rats.**. (2010) **88** 1485-1499. PMID: 20029988
37. Strettoi E., Di Marco B., Orsini N., Napoli D.. **Retinal plasticity.**. (2022) **23**
38. Strettoi E., Pignatelli V., Rossi C., Porciatti V., Falsini B.. **Remodeling of second-order neurons in the retina of rd/rd mutant mice.**. (2003) **43** 867-877. DOI: 10.1016/s0042-6989(02)00594-1
39. Sullivan R., Penfold P., Pow D. V.. **Neuronal migration and glial remodeling in degenerating retinas of aged rats and in nonneovascular AMD.**. (2003) **44** 856-865. DOI: 10.1167/iovs.02-0416
40. Tsukamoto Y., Omi N.. **Classification of mouse retinal bipolar cells: type-specific connectivity with special reference to rod-driven AII amacrine pathways.**. (2017) **11**. DOI: 10.3389/fnana.2017.00092
|
---
title: The prevalence of chronic kidney disease in South Africa - limitations of studies
comparing prevalence with sub-Saharan Africa, Africa, and globally
authors:
- Sudesh Hariparshad
- Rajendra Bhimma
- Louansha Nandlal
- Edgar Jembere
- Saraladevi Naicker
- Alain Assounga
journal: BMC Nephrology
year: 2023
pmcid: PMC10029276
doi: 10.1186/s12882-023-03109-1
license: CC BY 4.0
---
# The prevalence of chronic kidney disease in South Africa - limitations of studies comparing prevalence with sub-Saharan Africa, Africa, and globally
## Abstract
### Background
Chronic kidney disease (CKD) is a globally significant non-communicable disorder. CKD prevalence varies between countries and within a country. We compared the prevalence rates of CKD in South Africa with sub-Saharan Africa, Africa, and globally.
### Methods
We registered a systematic review with the International Prospective Register of Systematic Reviews for prevalence studies reporting CKD stages III-V from 2013 to 2021. The analysis sought to explain any significant differences in prevalence rates. The R statistical package was used for data analysis. Comparisons included measures of effect size due to the large sample sizes analysed. We also compared sex differences in prevalence rates, common aetiologies, and type of study methodologies employed.
### Results
Eight studies were analysed, with two from each region. The matched prevalence rates of CKD between the various regions and South Africa showed significant differences, except for one comparison between South Africa and an African study [$$p \leq 0.09$$ ($95\%$ CI − 0.04–0.01)]. Both sub-Saharan African studies had a higher prevalence than South Africa. One study in Africa had a higher prevalence, while the other had a lower prevalence, whilst one Global study had a higher prevalence, and the other had a lower prevalence compared to South Africa. The statistical differences analysed using the Cramer’s V test were substantially less than 0.1. Thus, differences in comparisons were largely due to differences in sample sizes rather than actual differences.
### Conclusion
Variable prevalence rates between regions included disparities in sample size, definitions of CKD, lack of chronicity testing and heterogeneous laboratory estimations of eGFR. Improved consistency and enhanced methods for diagnosing and comparing CKD prevalence are essential.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12882-023-03109-1.
## Background
The estimated number of people with chronic kidney disease (CKD) globally is approximately 844 million [1]. Patients with CKD are estimated to be twice the number of people with diabetes worldwide and more than twenty times the number of people affected by human immunodeficiency virus/acquired immune deficiency syndrome (HIV/AIDS) [1]. Kidney diseases are among the most common global non-communicable diseases (NCDs) [1]. The worldwide all-age prevalence of CKD has increased by $29.3\%$ over the past three decades [2]. CKD has therefore become a universal public health priority [3]. Even though CKD prevalence has been researched more widely in economically developed countries, the disease burden is even more significant in developing countries [4]. The systematic review by Mills et al. estimated the global prevalence of CKD to be $11.1\%$ [4]. However, the numbers affected by CKD rest on data of various qualities, approximations, and assumptions [1, 5]. It is acknowledged that CKD is common, but the challenge is to define its true prevalence [5].
Non-communicable diseases (NCDs) are increasingly contributing to morbidity and mortality over the last three decades [6]. The factors contributing to NCDs rise are increasing longevity, urbanization, and cultural changes [6]. Metabolic disorders such as diabetes mellitus have contributed heavily to NCD deaths [7]. There is a projected increase of $156\%$ in diabetes mellitus, with about 25 million more cases estimated from 2017 to 2045 [7]. The high estimated prevalence of CKD will cause a significant disruption of healthcare provision obliging fundamental infrastructural changes with increasing expenditure [8]. Comparative CKD prevalence studies involving different countries or within a continent have revealed statistically significant differences in prevalence rates [9, 10]. The variances proposed were due to actual differences and disparities in study methods [9, 10].
Due to the dearth of epidemiological data from the majority of the continent, the prevalence of CKD in Africa continues to be underestimated [11] The majority of CKD prevalence studies conducted in Africa are not optimal [11, 12] Sub-Saharan Africa comprises $85\%$ of the African population with a higher prevalence of CKD compared to the continent’s north [11]. The most frequent causes of CKD in Africa are hypertension and diabetes mellitus followed by chronic glomerulonephritis and tubulointerstitial disorders [11]. Poverty and a lower socioeconomic status are two independent risk factors for developing CKD in Africa and hasten the course of the disease [13].
The International Society of Nephrology (ISN) Global Health Atlas survey for Africa estimated the prevalence of CKD in South Africa to be $10.7\%$ ($95\%$ CI 9.94–11.57) [14]. The distribution of NCDs in South Africa displays socioeconomic disparities, with the most onerous burden falling on poor communities in urban areas [15]. The World Health Organization (WHO) estimates that the burden of NCDs in South *Africa is* two to three times higher than in other developing countries [15].
The lack of comprehensive CKD registries in South Africa and the rest of Africa has resulted in limited knowledge of CKD prevalence. The ISN has underscored that the current and future burden of CKD will be concentrated in lower socioeconomic countries, which often lack systematized and coordinated policies to manage the problem [16]. Accurate CKD prevalence rates allow for efficient preparation and execution of intervention and prevention programs [17]. The purpose of this review is to compare the CKD prevalence rates in South Africa with prevalence rates in sub-Saharan Africa, Africa as a whole, and globally. The study sought to explain the causes of any substantial differences in prevalence rates if this was present.
## Method
The study was registered with the International Prospective Register of Systematic Reviews (PROSPERO). The reference number for the review was CRD42022330121. Two reviewers applied the eligibility criteria independently. Decisions were checked by a third reviewer. Disagreements were resolved through discussion and reaching a consensus. The study searched for publications on CKD using Google Scholar, Scopus, Embase, and PubMed/Medline.
The search terms included “prevalence,“ “epidemiology,” “chronic kidney disease,“ “renal insufficiency,” “renal impairment,” “nephropathy,” “stage III-V CKD,” “proteinuria, “albuminuria,” “meta-analysis,” “systematic reviews,” “cohort,” “cross-sectional,” “population-based,” “South Africa,“ “sub-Saharan Africa,“ “Africa,” “global.
The current Kidney Disease Improving Global Outcomes (KDIGO) staging criteria for CKD were included (stage III-V); hence the period for the studies was limited predominantly to the last decade (2013–2021) [18]. The search included only those reporting CKD (stage III-V) prevalence as not all studies included stages I-II CKD. Inclusion criteria included adult studies and English language articles. Studies that were translated into English were also included. The search included meta-analyses, systematic reviews, cohort, and cross-sectional studies. The studies were expected to use the prevailing definition of CKD. Criteria were also limited to those directly reporting studies of CKD in South Africa, sub-Saharan Africa, and globally.
Exclusion criteria were studies with patients under 12 years of age, those with inaccessible full texts, non-English studies that were not translated into English. Studies involving specific populations such as pregnant women, acute kidney injury or transplantation were excluded. ( Fig. 1). The first reviewer developed a data extraction tool. The data extracted included author, year of study, region, the prevalence of CKD, study population, and study design. Information acquired was tabulated on an Excel Spreadsheet (Microsoft Office for Windows, version 10; Microsoft Corporation, Redmond, WA®) for analysis.
The prevalence of CKD in South African studies was compared with the prevalence in sub-Saharan Africa, Africa, and globally to determine if there were statistically significant differences. Some of the papers reviewed had studied large numbers of patients. It was hence necessary to use the effect size to assess the strength of correlations where the chi-square test of independence would have shown dependence.
The null hypothesis proposed that there was no statistically significant difference in the prevalence of CKD between South Africa and sub-Saharan Africa, Africa, and globally. The R package was used for data analysis. In R, the test for difference between two proportions and the chi-square test for independence provided the same chi-square and p values. Rejection of the null hypothesis could be interpreted as evidence that the variables being considered are statistically dependent. An alternative interpretation for the rejection was that the sample proportions being compared were significantly different.
The probability of finding a significant difference between proportions is increased with large sample sizes. The increased chi-square statistic may not represent a strong pattern of dependence between variables but reflects an increase in sample size. It was necessary to review the test of independence between two variables and use the effect size to assess whether significant differences were not due to large sample sizes. The Cramer’s V test was used as an effect size measurement for the chi-square test of independence. The test measured how strongly categorical fields, regions, and CKD are associated.
Fig. 1Selection of papers for analysis of the prevalence of CKD from South Africa, sub-Saharan Africa, Africa and Globally
## Results
The analysis incorporated eight studies. ( Table 1).
Table 1Studies in South Africa, sub-Saharan Africa, Africa, and globally on prevalence of CKDAuthorRegionYearStudy type$95\%$ Confidence intervalReported CKD III-V prevalence rateCKD number of patientsNon-CKD number of patientsTotal number of patientsMatsha et al. [ 19]South Africa2013CohortPopulation based5.0-$8.58.7\%$1041 1111202Adeniyi et al. [ 20]South Africa2016CohortPopulation based3.2–$9.76.4\%$31458489George et al. [ 8]Sub-Saharan Africa2019Cross sectionalPopulation based9.9–$11.710.7\%$86872428110Stanifer et al [22]Sub-Saharan Africa2014Systematic reviewPopulation based12.2–$15.713.9\%$893955,36864,307Kaze et al. [ 12]Africa2018Systematic reviewPopulation and hospital based3.3–$6.14.6\%$45289 390498 432Abd ElHafeez et al. [ 11]Africa2018Systematic reviewPopulation based9.8–$10.510.1\%$15 150134 850150 000Hill et al. [ 4]Global2016Systematic reviewPopulation based9.2–$12.210.6\%$742 0006 258 0007 000 000Bikbov et al. [ 2]Global2017Systematic reviewPopulation based3.5–$4.34.1\%$314 262 5097 350 676 7347 664 939 243 There were two studies each from South Africa, sub-Saharan Africa, Africa, and globally. A total of 7 665 961 783 participants were included: 315 034 128 ($4.1\%$) having CKD stages III-V. The sample size of the studies ranged from 489 in a South African study [19, 20] to 7 664 939 243 in a global study [2]. The prevalence rates for CKD ranged from 6.4 to $8.7\%$ in South Africa [19, 20], 10.7–$13.9\%$ in sub-Saharan Africa [21, 22], 4.6–$10.1\%$ in Africa [16, 17] and 4.1–$10.6\%$ globally [2, 4].
Matsha et al. published a regional cohort study on CKD in the Western Cape, South Africa, in one of the two South African studies [19]. The age-standardized prevalence using the Chronic Kidney Disease Epidemiology Collaboration equation (CKD-EPI) of CKD was $8.7\%$ ($95\%$ CI 7.5–9.9) [19]. The overall mean age of participants was 52.9 ± 14.8 years; females constituted $75.3\%$ of the study group. The risk factors involved included hypertension ($33.0\%$) which also doubled the risk of developing CKD [19]. The prevalence of diabetes was $26.0\%$, with obesity being a significant risk factor for developing diabetes mellitus [19]. The prevalence of HIV was not reported.
The second South African study was a cross-sectional survey of CKD prevalence from the Western Cape by Adeniyi et al. [ 20]. The age-standardized prevalence using the CKD-EPI equation for CKD was $6.4\%$ ($95\%$ CI 3.2–$9.7\%$) [20]. Patients had a mean age of 46.3 ±8.5 years with the majority ($70.3\%$) being female [20]. Risk factors included hypertension and diabetes, with a prevalence of $55.2\%$ and $20.7\%$, respectively, while the prevalence of HIV was not reported [20].
In the sub-Saharan African study by George et al. in 2019, using a population-based study, the authors investigated the CKD prevalence in four sub-Saharan countries, viz. Burkina Faso, Ghana, Kenya, and South Africa [21]. The overall prevalence of CKD was $10.7\%$ ($95\%$ CI 9.9–11.7) [10]. South Africa had the highest prevalence of $12.9\%$ ($95\%$ CI 10.6–11.5) compared to the East and West African countries [21]. The mean age of participants was 49.9 ±5.8 years [10]. Females accounted for $49.2\%$ of the study participants [21]. Women had a higher prevalence of CKD of $12.0\%$ ($95\%$ CI 10.8–13.2) compared to men, with a prevalence of $9.5\%$ ($95\%$ CI 8.3–10.8) [10]. Prevalence of the risk factors of hypertension, diabetes, and HIV were $32.6\%$ ($95\%$ CI 31.3–34), $5.6\%$ ($95\%$ CI 5-6.2), and $15.9\%$ ($95\%$ CI 14.9–17.1), respectively [21].
A systematic review by Stanifer et al. in 2014 of 22 medium and high-quality studies in sub-Saharan Africa reported the prevalence of CKD to be $13.9\%$ ($95\%$ CI 13.8–19.6) [22]. The mean age in the different quality studies was 41.5± 4.1 years, with females constituting $57.5\%$ of participants [22]. Risk factors included hypertension and diabetes, and HIV, with a median prevalence of $16.8\%$ and $17.1\%$, and $11.9\%$, respectively [22].
In a meta-analysis of 98 CKD studies in Africa by Kaze et al. in 2018, the overall prevalence of CKD was $4.6\%$ ($95\%$ CI 3.3–6.1) [12]. The mean age of participants was 43.0 ± 6.2 years. [ 12] The proportion of female participants was not reported. The main risk factors for CKD were hypertension, diabetes, and HIV [12]. The prevalence rates for the risk factors were $35.6\%$ ($95\%$ CI 27.9–43.7), $13.3\%$ ($95\%$ CI 10.7–16), and $17.9\%$ ($95\%$ CI 10.9–26.1), respectively [12].
In another systematic review of 152 CKD stage III-V prevalence studies in Africa in 2018 by Abd El Hafeez et al. [ 11], the CKD prevalence rate was $10.1\%$ ($95\%$ CI 9.8–10.5) [11]. The median age was 52.8 ± 11.7 years [11]. The overall proportion of female participants was $64.3\%$ [11]. The pooled risk factor prevalence of hypertension was $34.5\%$ ($95\%$ CI 34.0–36.0), diabetes $24.7\%$ (95 CI 23.6–25.7), and HIV $5.6\%$ ($95\%$ CI 5.4–5.8) [11].
The global study by Hill et al. in 2016 was a systematic review and meta-analysis of 100 observational studies involving seven million patients [4]. The estimated prevalence of CKD was $10.6\%$ ($95\%$ CI 9.2–$12.2\%$) [4].The mean age of all participants was 49.0 ± 8.5 years [4]. The proportion of female participants studied was $55.0\%$ [4]. The prevalence of CKD in males was $8.1\%$ ($95\%$ CI 6.3–10.2) [4]. The CKD prevalence in females was $12.1\%$ ($95\%$ CI 10.6–13.8) [4]. The median prevalence of the two major risk factors was hypertension ($40.1\%$) and diabetes mellitus ($15.1\%$) [4]. HIV was not reported as a risk factor.
Bikbov et al. in 2020 reported a systemic analysis of the Global Burden of Disease (GBD) study based on published literature, registration systems, chronic kidney failure registries, and household surveys [2]. The estimated prevalence in a study population for CKD stage III was $3.9\%$ ($95\%$ CI 3.5–$4.3\%$), $0.16\%$ ($95\%$ CI 0.0.13–$0.19\%$) for CKD stage IV, and $0.07\%$ ($95\%$ CI 0.06–$0.08\%$) for CKD stage V [2]. The mean age and proportion of female participants were not reported, but the prevalence of CKD in females was 1.29-fold ($95\%$ CI 1.28–1.3) more than in males [2]. The age-standardized prevalence of CKD in females was $9.5\%$ ($95\%$ CI 8.8–10.2] and $7.3\%$ ($95\%$CI 6.8–7.9) in males [2]. Major risk factors for CKD in the study were hypertension, with a prevalence of $43.2\%$ ($95\%$ CI 42.3–54.1), and diabetes, with a prevalence of $57.6\%$ ($95\%$ CI 50.5–63.8) [2]. There was no reporting of HIV as a risk factor.
The first comparison was between South Africa and sub-Saharan Africa. Both sub-Saharan studies had a higher prevalence of CKD compared to Matsha et al. [ 19]. Once more, when compared to Adeniyi et al. [ 20], both sub-Saharan studies revealed a higher prevalence of CKD..
When comparing South Africa with Africa, only one study comparing Adeniyi et al. [ 20](South Africa) versus Kaze et al. [ 12] (Africa) displayed no significant difference. The African study by Kaze et al. [ 12] displayed a lower CKD prevalence. Abd El Hafeez et al. [ 11](Africa) had a higher prevalence of CKD than both South African studies. ( Table 2) Table 2shows the results of statistical tests of differences in CKD prevalence in South *Africa versus* sub-Saharan Africa, African and global studiesComparative RegionAuthorsProportions$95\%$ CIChi-Squaredp-valueCramer’s VSub-Saharan AfricaMatsha et al. [ 19] (SA)0.07570715-0.08: − 0.0539.249< 0.0010.025Stanifer et al. [ 22]0.13900508Matsha et al. [ 19] (SA)0.07570715-0.05: -0.0110.78< 0.0010.035George et al. [ 8]0.10702836Adeniyi et al. [ 20] (SA)0.06339468-0.10: -0.0522.633< 0.0010.019Stanifer et al. [ 22]0.13900508Adeniyi et al. [ 20] (SA)0.06339468-0.07: -0.028.919< 0.0010.033George et al. [ 8]0.10702836AfricaMatsha et al. [ 19] (SA)0.075707150.01: 0.0423.035< 0.0010.015Kaze et al. [ 12]0.04600130Matsha et al. [ 19] (SA)0.07570715-0.04: -0.018.13960.0040.007Abd ElHafeez et al. [ 11]0.10100000Adeniyi et al. [ 20] (SA)0.06339468-0.01: 0.042.96440.0900.006Kaze et al. [ 12]0.04600130Adeniyi et al. [ 20] (SA)0.06339468-0.06: -0.017.19050.0070.007Abd ElHafeez et al. [ 11]0.10100000GlobalMatsha et al. [ 19] (SA)0.07570715-0.04: -0.0111.321< 0.0010.001Hill et al. [ 4]0.106Adeniyi et al. [ 20](SA)0.06339468-0.06: -0.028.9222< 0.0010.001Hill et al. [ 4]0.106Matsha et al. [ 19](SA)0.075707150.02: 0.0535.947< 0.0010.00Bikbov et al. [ 2]0.0410000Adeniyi et al. [ 20] (SA)0.06339468-0.0002: 0.045.68070.0170.00Bikbov et al. [ 2]0.0410000 The final CKD prevalence comparison was between South Africa and global studies. The global study by Bikbov et al. [ 2] had a lower prevalence of CKD, while the global study by Hill et al. [ 4] had a higher prevalence of CKD when both were compared to the South African studies.
Overall, there were statistically significant differences in comparisons between all studies, except for one study comparing South Africa against Africa. The prevalence of CKD in both sub-Saharan studies was higher than in South African studies. One African study had a lower prevalence of CKD than the South African studies, while the other had a higher prevalence. Similarly, one global study had a lower prevalence of CKD than the South African studies, while the global study had a higher prevalence. However, the maximum Cramer’s V value for all comparisons was 0.035, all considerably less than 0.1, which suggested that these statistical differences were an effect of sample size rather than actual differences.
Table 3 compares the studies analysed in each geographical region, incorporating the mean age range of participants, number of female participants, and prevalence of risk factors. In addition, the table compared whether the Kidney Disease Improving Global Outcomes (KDIGO) guidelines were used to define CKD, including whether testing for chronicity of more than three months duration was used for the diagnosis of CKD. Further comparisons included the type of serum creatinine assay used, estimating equations to calculate the estimated glomerular filtration rate (eGFR), and if ethnicity co-efficient were employed.
Table 3Comparison of eGFR equations, the mean age of participants, number of female participants, and prevalence of risk factors and laboratory methodsAuthorRegionMean age of participants (Years)FemaleParticipants(Percentage)Prevalence of HypertensionPrevalence of DiabetesPrevalence of HIVKDIGOCKDcriteriaTest for chronicity > 3 monthsSerum creatinine measurement: enzyme or JaffeeGFR equations studiedEthnicity co-efficient usedMatsha et al. [ 19]South Africa52.9 ± 14.875.333.026.0NotmentionedNoNoEnzymeCKD-EPIMDRDCockcroft-GaultMeasured with and without ethnic co-efficientAdeniyi et al. [ 20]South Africa46.3 ±8.570.355.220.7NotmentionedNoNoEnzymeCKD-EPIMDRDNoGeorge et al. [ 8]Sub-Saharan Africa49.9 ±5.849.232.65.615.9YesNoEnzymeCKD-EPIMDRDMeasured with and without ethnic co-efficientStanifer et al. [ 22]Sub-Saharan Africa41.5 ± 4.157.516.817.111.9NoNoNotmentionedMDRDCockcroft-GaultNoKaze et al. [ 12]Africa43.0 ± 6.2Notmentioned35.613.317.9NoNoNotmentionedCKD-EPIMDRDCockcroft-GaultCystatin CNoAbd ElHafeez et al. [ 11]Africa52.8 ±11.764.334.424.75.6NoNoEnzyme and JaffeCKD-EPIMDRDCockcroft-GaultNoHill et al. [ 4]Global49.0 ± 8.555.040.115.1NotmentionedNoNoEnzyme and JaffeCKD-EPIMDRDNoBikbov et al. [ 2]GlobalNot studiedNotmentioned43.257.6NotmentionedModelledNoNotmentionedCKD-EPIMDRDNo Most study participants in all studies assessed were in the fourth to fifth decade of life. There was a predominance of female participants in the prevalence studies. Hypertension and diabetes mellitus were the most common risk factors in all studies, with HIV identified as a common risk factor in sub-Saharan Africa and Africa.
Only George et al. [ 21] (sub-Saharan African study) used the KDIGO definition of CKD. None of the selected studies considered chronicity of more than three months for CKD. Matsha et al. [ 19], Adeniyi et al. [ 20] (South Africa), and George et al. [ 21] (sub-Saharan Africa) calculated serum creatinine with the enzyme-linked assay. Abd ElHafeez et al. [ 11] (Africa) and Hill et al. [ 4] (Global) analysed serum creatinine that was calculated using the enzyme-linked and Jaffe assays. The CKD-EPI, Modification of Diet in Renal Disease (MDRD), and Cockcroft-Gault equations were the most widely used for the estimated glomerular filtration rate (eGFR). Matsha et al. [ 19] (South Africa) and George et al. [ 21] (sub-Saharan Africa) calculated the eGFR with and without ethnicity co-efficient.
## Discussion
On prima facie analysis, there were statistically significant differences in CKD prevalence rates between South Africa and sub-Saharan Africa, Africa, and globally in all except for one comparison. The single comparison that did not show a statistically significant difference in CKD prevalence was between the South African study by Adeniyi et al. [ 20] compared to the African study by Kaze et al. [ 12]. The prevalence of CKD in sub-Saharan studies was higher than those in South African studies. However, it could not be determined whether the prevalence of CKD in South Africa was higher or lower than the African and global prevalence.
The wide variations in sample sizes between comparative groups limited the interpretation of statistical tests such as the p-values and confidence intervals [23]. The significant differences in prevalence may be due to large sample sizes. The analysis of the Cramer’s V effect size indicated a weak association between CKD prevalence rates and the regions. The statistically significant differences in prevalence rates across the regions may be due to differences in sample sizes rather than dependence of CKD prevalence between each geographical region. Our analysis shows a similarity to previous comparative studies between geographical regions [9, 10, 15]. Differences in prevalence rates of CKD between countries and regions have been documented, with variations being due to true differences or limitations caused by the heterogeneity of studies [15]. True variations result from high protein diets, smoking, physical activity, socioeconomic status, ethnicity, genetics, and birth weight [15]. International comparisons of CKD prevalence have been hindered by differences in age, sex distribution, sampling, and definitions of CKD [15]. Regional variations of CKD prevalence within a country are also frequent, and the degree of variations may fluctuate [15]. A rapid epidemiological transition could also explain the different prevalence of environmental changes, adoption of western lifestyles, and rapid urbanisation in Africa [12]. The clinical, demographic, and laboratory causes of variations in CKD prevalence will be discussed.
The median age of developing CKD in lower-middle-income countries was 43.7 years [12]. Observational and cohort studies in Africa have consistently shown an increased risk of cardiovascular disease mortality in the early stages, with nearly $40\%$ of deaths from CKD occurring before 65 years [12]. The mean age of patients diagnosed with CKD in South Africa, sub-Saharan Africa, and Africa was younger, between the fourth and fifth decade, compared to the global CKD study by Hill et al. [ 4], where the highest prevalence was between the fifth and sixth decade. Lower kidney function was associated with a significant and progressive reduction of life expectancy in middle age for both men and women [24]. An earlier age at diagnosis heralds a worse prognosis.
The KDIGO criteria, if used to define stages of CKD, result in a considerable increase in prevalence with age and the method used to estimate GFR [25]. The threshold of 60ml/min/1.73m2 for the diagnosis of CKD could contribute to an overestimation of CKD in older patients [26]. Elderly populations exhibit a normal “physiological” decline in GFR with aging (renal senescence) [27]. Epidemiologic studies using a “once-off” testing of eGFR, especially with elderly participants, may also overestimate the burden of CKD in older patients [25]. The controversy of whether the decline in GFR is due to aging, as opposed to disease, has not been directly resolved [25]. A suggested method to overcome false positives would be to use the third percentile of eGFR creatinine levels and age-calibrated thresholds [26]. Alternatively, the Berlin Initiative Study 1 equation would be more suitable for subjects older than 70 years [28].
Most patients diagnosed with CKD were female, in keeping with the majority of worldwide CKD prevalence studies [29]. The prevalence of CKD in the United States of *America is* higher in females than males, but males have a higher prevalence of newly treated chronic kidney failure (CKF) [30]. The cause of this was indeterminate but may be multifactorial [3]. There is a possibility of overdiagnosis of CKD in older women than in men [31]. Women, on average, have lower estimated GFR and measured GFR (uncorrected for body surface area) and tend to progress to a GFR value of < 60ml/min/1.73m2 before men, although men progress more rapidly to CKF [31]. This physiological sex difference could contribute to an overdiagnosis of CKD in women than men as they age, especially in the absence of albuminuria [31].
The role of the social environment and economic conditions is an emerging component in the pathway from CKD risk to the development and complications of CKD and chronic kidney failure [32]. Socioeconomically underprivileged inhabitants worldwide show an unevenly high burden of CKD [32]. The burden is compounded by the inability to receive evidence-based care leading to poor clinical outcomes [32]. Lower socioeconomic status was related to a greater risk of prevalence of CKD [32].The poor with a higher kidney disease burden often have fewer resources to meet treatment costs [32]. The consequence is “catastrophic spending” (defined as out-of-pocket payments above $40\%$ of non-food expenditure) [32]. Thus, advanced CKD could be considered a risk factor for poverty along with low education level, employment status, and ethnicity [33]. The entire family becomes affected by the reduction in resources [33]. Poverty can also directly affect adherence to medical treatment as the affected patient may be unable to access follow-up care or afford kidney replacement therapy when required [33].
Countries with a higher CKD prevalence have a higher risk factor profile [10]. Sub-Saharan *Africa is* estimated to have 18.65 million people with diabetes mellitus [22]. A similar number is estimated to develop hypertension by the end of this decade [22]. There would also be an estimated 22 million people living with HIV/AIDS during this time, posing a further substantial burden of CKD in this region [22]. In Africa, the dominant risk factors for developing CKD are hypertension, diabetes mellitus, and HIV [11, 12]. Africa also has the highest prevalence of HIV-1 infection [34]. There is a robust association between Apolipoprotein L1 gene variants found only on African chromosomes resulting in an increased probability of developing focal segmental glomerulosclerosis and HIV-associated nephropathy [34, 35]. Resource limitations lead to the late initiation of ARVs (antiretroviral agents), which predisposes to HIV-associated nephropathy [36]. The combination of genetic susceptibility with delayed treatment of HIV contributes to the increase in CKD prevalence and disease burden. Africa is, therefore, subject to a dual burden of non-communicable and endemic infectious diseases such as HIV leading to CKD [37].
Global studies identified hypertension, diabetes mellitus, female sex, and increasing age as the major risk factors for the development of CKD [2, 4]. International differences in the prevalence of risk factors for CKD could be affected by sample selection [10]. CKD prevalence fluctuates with time, as some international differences in CKD prevalence may be explained by differences in the study periods and the associated transition of risk factor profiles [10]. Increased prevalence within some regions compared to neighbouring areas with similar demographics may also indicate increased recognition and recording of CKD [29]. The epidemiological transition from communicable to non-communicable diseases, with significant increases in hypertension and diabetes mellitus with aging, may also account for the increased prevalence of CKD [11, 38, 39].
The estimated requirements for kidney replacement therapy will double from 2.62 million in 2010 to 5.4 million people by 2030 [40]. Global deaths due to kidney disease are projected at between 5 and 10 million people annually due to a shortage of kidney replacement therapy services [40]. Higher-income countries spend 2–$3\%$ of their annual health budget on CKF treatment for approximately $0.03\%$ of the total population [40]. Lower-income countries are not able to provide similar resources for chronic kidney failure (CKF). They will most likely experience the societal, health, and economic burden of mostly untreated CKF.
Over the past decade, there have been substantial developments in standardising assays for serum creatinine [41]. The re-calibration of serum creatinine assays to an isotope dilution mass spectrometry reference method has resulted in more specific assays traceable to the International System of units [42]. The introduction of isotope dilution mass spectrometry calibration for serum creatinine assays has addressed the variability of serum creatinine data [42]. However, difficulties persist concerning using eGFR to assess CKD prevalence in epidemiological studies [42]. A continuing complication is that the effect of assay calibration differs between eGFR equations [43]. Variations in calibration have a more significant effect on the MDRD equation than on the CKD-EPI equation for eGFR [43]. The variation is due to the mathematical exponent applied to serum creatinine in elevated eGFR ranges and is lower in CKD-EPI than the MDRD equation [43]. The CKD-EPI equation gives a lower prevalence of CKD due to a higher eGFR in general or specific population participants than other equations [43]. In contrast, the systematic underestimation of eGFR with the MDRD equation is associated with an overestimation of CKD prevalence in epidemiological studies [43].
The lack of standardized equations to calculate eGFR was highlighted in the studies by several authors in this paper [2, 4, 17, 20–22]. Most studies reviewed displayed an analytical heterogeneity used to measure creatinine. Evaluation of eGFR is fundamental to medical practice, research, and public health [44]. Serum creatinine is the most commonly utilized biomarker to assess eGFR [45]. However, individual values may vary due to factors that include mass, age, sex, ethnicity, and diet unrelated to CKD [45]. Measured GFR (mGFR) and gold-standard measurements using inulin clearance are, unfortunately, too cumbersome to perform in extensive epidemiologic studies [31]. In a collaborative study from Malawi, Uganda, and South Africa that prospectively measured kidney function, it was established that creatinine-based GFR-estimating equations overestimate kidney function [46]. The implication is that the burden of kidney disease may be significantly underestimated in Africa [46].
A common limitation in CKD prevalence studies is the “once-off testing” of serum creatinine (and hence eGFR). Other limitations included quantifying albuminuria; the different formulae used to calculate eGFR, the absence of proteinuria and haematuria testing, and heterogeneity in sample data used to calculate the prevalence of CKD. Once off, eGFR testing or confirming chronicity was reported here as a limitation in numerous studies [2, 4, 7, 8, 16, 20]. Glassock et al. contend that although CKD is widespread, the contention that the prevalence is increasing in many countries may be incorrect [31]. The authors maintain that using “once-off testing” of eGFR and albuminuria to define prevalence in epidemiological studies is controversial, as these “single test” studies do not adhere to the KDIGO CKD definition of three-month duration [18]. The “once-off” testing produces a false positive diagnostic rate of about $30\%$ for eGFR and even higher for albuminuria [47]. Conversely, false-negative results, which primarily involve the younger population, arise when they have an eGFR above 60 ml/min/1.73m2 [48]. This subset does not meet the criteria for the definition of CKD and is without proteinuria, but they have a low eGFR for their age, below the 3rd percentile for age and sex category [48].
Using ancestry coefficients, sex, and age of patients can further contribute to the limitations of prevalence studies. The ancestry coefficient is a significant constituent of the MDRD and CKD-EPI equations [31]. It was recommended to improve the understanding of the prevalence of CKD in ethnically diverse populations [31]. However, the African American coefficient results in the MDRD and CKD-EPI equations for eGFR being $21\%$ and $15\%$ more elevated, respectively, than the same equations without coefficients [31]. It can be contended that the use of race in eGFR equations is a social and not a biological concept [46]. The inclusion of race ignores diversity within and among racial groups [46]. Alterations in estimating equations can affect the calculation of the burden of CKD and potentially disrupt patient care [46]. It can also be debated that keeping a race term in GFR equations adversely affects access to kidney replacement therapy [49].
Alternatives to calculating eGFR without using race are currently being evaluated [50]. The estimation of GFR with the usage of cystatin C was similar to estimations using serum creatinine [50]. Cystatin C-based estimations did not use race or ancestry and were not enhanced or changed by their inclusion [50]. Most recent eGFR equations use creatinine and cystatin C without race [51]. They are more accurate in estimating GFR than either equation using creatinine or cystatin C alone [51]. This has resulted in reduced differences from measured GFR between race groups [51]. A systematic review of epidemiological studies from sub-Saharan Africa highlighted the source’s potential for bias [52]. These include variability in the requirements for serum creatinine assays, appropriate choice of estimating equations to calculate eGFR, and appropriate diagnostic criteria for CKD [52]. The results were consistent with other worldwide studies [52]. The ongoing evolution of data from eGFR equations will further inform clinical practice, research, and public health considerations [52].
An essential requirement for the management of CKD is for efficient and sustainable solutions to capture high-quality population-based health data and extrapolate it into health information systems [53]. This will allow a better understanding of CKD epidemiology and variations in CKD prevalence [53]. The CKD in Africa (CKD-Africa) project is a continental collaboration network that aims to provide uniformly reliable estimates for CKD prevalence [53]. The collaboration has currently networked 12 African countries in sub-Saharan Africa, totalling 39 studies and 35 747 participants [53]. This collective health system would be able to effectively advise future health services planning and policy for CKD management in Africa [53].
The study limitations include analysing two studies from South Africa from the same region. These studies may not represent the country’s prevalence of CKD because regional variations in CKD prevalence can occur within a country [15]. The South African studies had relatively small numbers of participants compared to those in sub-Saharan Africa, Africa, and globally. HIV, a significant risk factor for CKD in sub-Saharan Africa, was not investigated amongst participants in the South African studies. The population sampling was also not representative of the South African population demographics. A further limitation was the low number of studies that were eligible for inclusion in the analysis.
## Conclusion
There was a statistically significant variation in the prevalence of CKD between South Africa and sub-Saharan Africa, Africa, and globally in all except one comparison. However, there was a poor correlation due to the effect size, which suggests that these differences may be due to comparing studies with large sample sizes than to actual differences in the prevalence. This review echoed the marked heterogeneity when comparing CKD prevalence from different regions. These included varying sample sizes, differences in the study methodology, the criteria for the definition of CKD, the lack of chronicity reporting, and variances in serum creatinine measurements leading to variable eGFRs. Enhanced uniformity and novel approaches are crucial for performing and reporting CKD prevalence studies to advance the accuracy of comparing the burden of the disease.
## Electronic supplementary material
Below is the link to the electronic supplementary material.
Supplementary Table 1: Search Strategy
## References
1. 1.Jager KJ, Kovesdy C, Langham R, Rosenberg M, Jha V, Zoccali C. A single number for advocacy and communication-worldwide more than 850 million individuals have kidney diseases. Oxford University Press; 2019. pp. 1803–5. 10.1093/ndt/gfz174
2. Bikbov B, Purcell CA, Levey AS, Smith M, Abdoli A, Abebe M. **Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the global burden of Disease Study 2017**. *The Lancet* (2020.0) **395** 709-33. DOI: 10.1016/S0140-6736(20)30045-3
3. Levin A, Tonelli M, Bonventre J, Coresh J, Donner J-A, Fogo AB. **Global kidney health 2017 and beyond: a roadmap for closing gaps in care, research, and policy**. *The Lancet* (2017.0) **390** 1888-917. DOI: 10.1016/S0140-6736(17)30788-2
4. Hill NR, Fatoba ST, Oke JL, Hirst JA, O’Callaghan CA, Lasserson DS. **Global prevalence of chronic kidney disease–a systematic review and meta-analysis**. *PLoS ONE* (2016.0) **11** e0158765. DOI: 10.1371/journal.pone.0158765
5. 5.Black C, van der Veer SN. Unlocking the value of variation in CKD prevalence. Am Soc Nephrol. 2016;1874–7. 10.1681/asn.2015111280
6. Matsha TE, Erasmus RT. **Chronic kidney disease in sub-saharan Africa**. *The Lancet Global Health* (2019.0) **7** e1587-e8. DOI: 10.1016/S2214-109X(19)30467-X
7. Saeedi P, Petersohn I, Salpea P, Malanda B, Karuranga S, Unwin N. (2019.0). DOI: 10.1016/j.diabres.2019.107843
8. George JA, Brandenburg J-T, Fabian J, Crowther NJ, Agongo G, Alberts M. **Kidney damage and associated risk factors in rural and urban sub-saharan Africa (AWI-Gen): a cross-sectional population study**. *The Lancet Global Health* (2019.0) **7** e1632e43. DOI: 10.1016/S2214-109X(19)30443-7
9. Wang F, He K, Wang J, Zhao M-H, Li Y, Zhang L. **Prevalence and risk factors for CKD: a comparison between the adult populations in China and the United States**. *Kidney Int Rep* (2018.0) **3** 1135-43. DOI: 10.1016/j.ekir.2018.05.011
10. Stel VS, Brück K, Fraser S, Zoccali C, Massy ZA, Jager KJ. **International differences in chronic kidney disease prevalence: a key public health and epidemiologic research issue**. *Nephrol Dialysis Transplantation* (2017.0) **32** ii129-ii35. DOI: 10.1093/ndt/gfw420
11. Abd ElHafeez S, Bolignano D, D’Arrigo G, Dounousi E, Tripepi G, Zoccali C. **Prevalence and burden of chronic kidney disease among the general population and high-risk groups in Africa: a systematic review**. *BMJ open* (2018.0) **8** e015069. DOI: 10.1136/bmjopen-2016-015069
12. Kaze AD, Ilori T, Jaar BG, Echouffo-Tcheugui JB. **Burden of chronic kidney disease on the african continent: a systematic review and meta-analysis**. *BMC Nephrol* (2018.0) **19** 1-11. DOI: 10.1186/s12882-018-0930-5
13. Nyirenda MJ. **Non-communicable diseases in sub-saharan Africa: understanding the drivers of the epidemic to inform intervention strategies**. *Int health* (2016.0) **8** 157-8. DOI: 10.1093/inthealth/ihw021
14. Bello AK, Levin A, Lunney M, Osman MA, Ye F, Ashuntantang G, Bellorin-Font E, Benghanem GM, Ghnaimat M, Harden P. *Global kidney Health Atlas: a report by the International Society of Nephrology on the global burden of end-stage kidney disease and capacity for kidney replacement therapy and conservative care across world countries and regions* (2019.0)
15. Mayosi BM, Flisher AJ, Lalloo UG, Sitas F, Tollman SM, Bradshaw D. **The burden of non-communicable diseases in South Africa**. *The lancet* (2009.0) **374** 934-47. DOI: 10.1016/S0140-6736(09)61087-4
16. 16.Working Group International Society of Nephrology ISN Framework for DevelopingDialysis Programs in low-resource settings2021Brussels, Belgium. International society of NephrologyInternational Society of Nephrology. *Dialysis Programs in low-resource settings* (2021.0)
17. Brück K, Stel V, Gambaro G, Hallan S, Völzke H, Ärnlöv J. **European CKD Burden Consortium: CKD prevalence varies across the european general population**. *J Am Soc Nephrol* (2015.0) **27** 2135. DOI: 10.1681/ASN.2015050542
18. **KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease**. *Kidney Int Suppl* (2013.0) **3** 1-50. DOI: 10.1016/j.kint.2020.06.019
19. Matsha TE, Yako YY, Rensburg MA, Hassan MS, Kengne AP, Erasmus RT. **Chronic kidney diseases in mixed ancestry south african populations: prevalence, determinants and concordance between kidney function estimators**. *BMC Nephrol* (2013.0) **14** 1-10. DOI: 10.1186/1471-2369-14-75
20. Adeniyi AB, Laurence CE, Volmink JA, Davids MR. **Prevalence of chronic kidney disease and association with cardiovascular risk factors among teachers in Cape Town, South Africa**. *Clin kidney J* (2017.0) **10** 363-9. DOI: 10.1093/ckj/sfw138
21. 21.George JA, Brandenburg JT, Fabian J, Crowther NJ, Agongo G, Alberts M, Ali S, Asiki G, Boua PR, Gómez-Olivé FX, Mashinya F. Kidney damage and associated risk factors in rural and urban sub-Saharan Africa (AWI-Gen): a cross-sectional population study. The Lancet Global Health. 2019 Dec1; 7(12):e1632-4310.1016/s2214-109x(19)30443-7
22. Stanifer JW, Jing B, Tolan S, Helmke N, Mukerjee R, Naicker S. **The epidemiology of chronic kidney disease in sub-saharan Africa: a systematic review and meta-analysis**. *The Lancet Global Health* (2014.0) **2** e174-e81. DOI: 10.1016/s2214-109x(14)70002-6
23. 23.Shreffler J, Huecker MR. Hypothesis testing, P values, confidence intervals, and significance. StatPearls [Internet] Treasure Island (FL): StatPearls Publishing; 2022 Jan–. PMID: 32491353.
24. Turin TC, Tonelli M, Manns BJ, Ravani P, Ahmed SB, Hemmelgarn BR. **Chronic kidney disease and life expectancy**. *Nephrol Dialysis Transplantation* (2012.0) **27** 3182-6. DOI: 10.1093/ndt/gfs052
25. Glassock R, Delanaye P, El Nahas M. **An age-calibrated classification of chronic kidney disease**. *JAMA* (2015.0) **314** 559-60. DOI: 10.1001/jama.2015.6731
26. Delanaye P, Glassock RJ, De Broe ME. **Epidemiology of chronic kidney disease: think (at least) twice!**. *Clin Kidney J* (2017.0) **10** 370-4. DOI: 10.1093/ckj/sfw154
27. Glassock RJ, Winearls C. **The global burden of chronic kidney disease: how valid are the estimates?**. *Nephron Clin Pract* (2008.0) **110** c39-c47. DOI: 10.1159/000151244
28. Matašin M, Domislović V, Fuček M, Gellineo L, Jelaković A, Dika Ž. **Berlin Initiative Study 1 equation and HUGE formula for more accurate estimation of kidney function in elderly**. *Rad Hrvatske akademije znanosti i umjetnosti Medicinske znanosti* (2021.0) **547** 24-31. DOI: 10.21857/ydkx2cw8k9
29. Carrero JJ, Hecking M, Chesnaye NC, Jager KJ. **Sex and gender disparities in the epidemiology and outcomes of chronic kidney disease**. *Nat Rev Nephrol* (2018.0) **14** 151. DOI: 10.1038/nrneph.2017.181
30. Coresh J, Selvin E, Stevens LA, Manzi J, Kusek JW, Eggers P. **Prevalence of chronic kidney disease in the United States**. *JAMA* (2007.0) **298** 2038-47. DOI: 10.1001/jama.298.17.2038
31. Glassock RJ, Warnock DG, Delanaye P. **The global burden of chronic kidney disease: estimates, variability and pitfalls**. *Nat Rev Nephrol* (2017.0) **13** 104. DOI: 10.1038/nrneph.2016.163
32. 32.Nicholas SB, Kalantar-Zadeh K, Norris KC. Socioeconomic disparities in chronic kidney disease. Advances in chronic kidney disease. 2015;22(1):6–15. 10.1053/j.ackd.2014.07.002
33. Morton RL, Schlackow I, Gray A, Emberson J, Herrington W, Staplin N. **Impact of CKD on household income**. *Kidney Int Rep* (2018.0) **3** 610-8. DOI: 10.1016/j.ekir.2017.12.008
34. Kasembeli AN, Duarte R, Ramsay M, Mosiane P, Dickens C, Dix-Peek T. **APOL1 risk variants are strongly associated with HIV-associated nephropathy in black South Africans**. *J Am Soc Nephrol* (2015.0) **26** 2882-90. DOI: 10.1681/asn.2014050469
35. Kopp JB, Nelson GW, Sampath K, Johnson RC, Genovese G, An P. **APOL1 genetic variants in focal segmental glomerulosclerosis and HIV-associated nephropathy**. *J Am Soc Nephrol* (2011.0) **22** 2129-37. DOI: 10.1681/asn.2011040388
36. 36.Szczech LA. Renal disease: the effects of HIV and antiretroviral therapy and the implications for early antiretroviral therapy initiation. Current Opinion in HIV and AIDS. 10.1097/coh.0b013e328329c6462009;4(3):167-70.
37. Hodel NC, Hamad A, Praehauser C, Mwangoka G, Kasella IM, Reither K. **The epidemiology of chronic kidney disease and the association with non-communicable and communicable disorders in a population of sub-saharan Africa**. *PLoS ONE* (2018.0) **13** e0205326. DOI: 10.1371/journal.pone.0205326
38. Cockwell P, Fisher L-A. **The global burden of chronic kidney disease**. *The Lancet* (2020.0) **395** 662-4. DOI: 10.1016/s0140-6736(19)32977-0
39. Madala ND, Thusi GP, Assounga AG, Naicker S. **Characteristics of south african patients presenting with kidney disease in rural KwaZulu-Natal: a cross sectional study**. *BMC Nephrol* (2014.0) **15** 1-9. DOI: 10.1186/1471-2369-15-61
40. Luyckx VA, Tonelli M, Stanifer JW. **The global burden of kidney disease and sustainable development goals**. *Bull World Health Organ* (2018.0) **96** 414. DOI: 10.2471/blt.17.206441
41. Piéroni L, Delanaye P, Boutten A, Bargnoux A-S, Rozet E, Delatour V. **A multicentric evaluation of IDMS-traceable creatinine enzymatic assays**. *Clin Chim Acta* (2011.0) **412** 2070-5. DOI: 10.1016/j.cca.2011.07.012
42. Delanaye P, Cavalier E, Cristol J-P, Delanghe JR. **Calibration and precision of serum creatinine and plasma cystatin C measurement: impact on the estimation of glomerular filtration rate**. *J Nephrol* (2014.0) **27** 467-75. DOI: 10.1007/s40620-014-0087-7
43. White SL, Polkinghorne KR, Atkins RC, Chadban SJ. **Comparison of the prevalence and mortality risk of CKD in Australia using the CKD epidemiology collaboration (CKD-EPI) and modification of Diet in Renal Disease (MDRD) Study GFR estimating equations: the AusDiab (australian diabetes, obesity and lifestyle) study**. *Am J Kidney Dis* (2010.0) **55** 660-70. DOI: 10.1053/j.ajkd.2009.12.011
44. Levey AS, Coresh J, Tighiouart H, Greene T, Inker LA. **Strengths and limitations of estimated and measured GFR**. *Nat Rev Nephrol* (2019.0) **15** 784. DOI: 10.1038/s41581-019-0213-9
45. 45.Centres for disease control and Prevention. Chronic kidney disease in the United States., 2019. Atlanta, GA: US Department of Health and Human Services, Centers for Disease Control and Prevention. 2019. March,3
46. Fabian J, Kalyesubula R, Mkandawire J, Hansen CH, Nitsch D, Musenge E. **Measurement of kidney function in Malawi, South Africa, and Uganda: a multicentre cohort study**. *The Lancet Global Health* (2022.0) **10** e1159-e69. DOI: 10.1016/s2214-109x(22)00239-x
47. Eriksen B, Ingebretsen O. **The progression of chronic kidney disease: a 10-year population-based study of the effects of gender and age**. *Kidney Int* (2006.0) **69** 375-82. DOI: 10.1038/sj.ki.5000058
48. Gharbi MB, Elseviers M, Zamd M, Alaoui AB, Benahadi N, Trabelssi EH. **Chronic kidney disease, hypertension, diabetes, and obesity in the adult population of Morocco: how to avoid “over”-and “under”-diagnosis of CKD**. *Kidney Int* (2016.0) **89** 1363-71. DOI: 10.1016/j.kint.2016.02.019
49. 49.Delgado C, Baweja M, Burrows NR, Crews DC, Eneanya ND, Gadegbeku CA et al. Reassessing the inclusion of race in diagnosing kidney diseases: An interim report from the NKF-ASN task force. Am J Kidney Dis. 2021 Jul; 78(1):103–115. 10.1053/j.ajkd.2021.03.008
50. Inker LA, Eneanya ND, Coresh J, Tighiouart H, Wang D, Sang Y. **New creatinine-and cystatin C–based equations to estimate GFR without race**. *N Engl J Med* (2021.0) **385** 1737-49. DOI: 10.1056/nejmoa2102953
51. Hsu C-y, Yang W, Parikh RV, Anderson AH, Chen TK, Cohen DL. **Race, genetic ancestry, and estimating kidney function in CKD**. *N Engl J Med* (2021.0) **385** 1750-60. DOI: 10.1056/nejmoa2103753
52. Fabian J, George JA, Etheredge HR, van Deventer M, Kalyesubula R, Wade AN. **Methods and reporting of kidney function: a systematic review of studies from sub-saharan Africa**. *Clin kidney J* (2019.0) **12** 778-87. DOI: 10.1093/ckj/sfz089
53. 53.George C, Stoker S, Okpechi I, Woodward M, Kengne A. The Chronic Kidney Disease in Africa (CKD-Africa) collaboration: lessons from a new pan-African network. BMJ Global Health. 2021 Aug1; 6(8):e006454. 10.1136/bmjgh-2021-006454
|
---
title: The association of center volume with transplant outcomes in selected high-risk
groups in kidney transplantation
authors:
- Massini Merzkani
- Su-Hsin Chang
- Haris Murad
- Krista L. Lentine
- Munis Mattu
- Mei Wang
- Vangie Hu
- Bolin Wang
- Yazen Al-Hosni
- Obadah Alzahabi
- Omar Alomar
- Jason Wellen
- Tarek Alhamad
journal: BMC Nephrology
year: 2023
pmcid: PMC10029277
doi: 10.1186/s12882-023-03099-0
license: CC BY 4.0
---
# The association of center volume with transplant outcomes in selected high-risk groups in kidney transplantation
## Abstract
### Background
In context of increasing complexity and risk of deceased kidney donors and transplant recipients, the impact of center volume (CV) on the outcomes of high-risk kidney transplants(KT) has not been well determined.
### Methods
We examined the association of CV and outcomes among 285 U.S. transplant centers from 2000–2016. High-risk KT were defined as recipient age ≥ 70 years, body mass index (BMI) ≥ 35 kg/m2, receiving kidneys from donors with kidney donor profile index(KDPI) ≥ $85\%$, acute kidney injury(AKI), hepatitisC +. Average annual CV for the specific-high-risk KT categorized in tertiles. Death-Censored-Graft-Loss(DCGL) and death at 3 months, 1, 5, and 10 years were compared between CV tertiles using Cox-regression models.
### Results
Two hundred fifty thousand five hundred seventy-four KT were analyzed. Compared to high CV, recipients with BMI ≥ 35 kg/m2 had higher risk of DCGL in low CV(aHR = 1.11,$95\%$CI = 1.03–1.19) at 10 years; recipients with age ≥ 70 years had higher risk of death in low CV(aHR = 1.07,$95\%$CI = 1.01–14) at 10 years. There was no difference of DCGL or death in low CV for donors with KDPI ≥ $85\%$, hepatitisC +, or AKI.
### Conclusions
Recipients of high-risk KT with BMI ≥ 35 kg/m2 have higher risk of DCGL and recipients age ≥ 70 years have higher risk of death in low CV, compared to high CV. Future studies should identify care practices associated with CV that support optimal outcomes after KT.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12882-023-03099-0.
## Introduction
Kidney transplantation (KT) is the treatment of choice for end stage renal disease (ESRD), as it improves quality of life and reduces the mortality rate of patients with ESRD, compared to dialysis, at lowest costs to the healthcare system [1]. The growth in the number of patients on the waiting list far exceeds the rate at which kidney transplantation is performed [2, 3]. To narrow this gap, several strategies have been used to expand the pool of deceased donors. Importantly, even high-risk KT is cost-effective compared to dialysis [1].
Strategies to increase deceased donor pool include the use of high-risk deceased donors, e.g., kidneys from donors with high kidney donor profile index (KDPI), acute kidney injury (AKI), or hepatitis C positivity. For donors with AKI, it has been shown that recipients had similar graft survival at both short- and long-term compared to recipients of kidneys from donors without AKI [4, 5]. Recent data have shown that patients receiving kidneys from donors with Hepatitis C viremic did not experience increased risk for graft loss compared to those receiving kidneys with no viremic donors [6]. KDPI ≥ $85\%$ is associated with lower graft survival [7], but is imprecise and carries chance of misclassifying risk, which may increase the likelihood of organ discard.
In this context, the complexity of recipients is increased with the growing number of elderly and obese candidates on the waiting list [8]. Older recipient age has been associated with increased comorbidities, frailty, and risk of infection and death with functional allograft, when compared to younger population [9, 10]. Obesity has been also associated with an increased risk for delayed graft function, proteinuria, rejection, and graft failure in transplant recipients [11, 12].
Prior studies in solid organ transplants have shown an association between transplant center volume (CV) and patient outcomes. Centers performing relatively fewer solid organ transplants may have inferior allograft outcomes, whereas high-volume centers are associated with improved survival outcomes [13–21]. However, there are no large-scale studies that examined this relationship on high-risk KT, defined as recipient age ≥ 70 years, body mass index (BMI) ≥ 35 kg/m2, or receiving kidneys from donors with KDPI ≥ $85\%$, AKI, or hepatitis C antibody positivity. This is particularly important in the modern era with the growing number of high-risk donors in the donor pool and high-risk patients in the waitlist candidates.
## Study population
The study cohort was composed of all adult patients age ≥ 18 years, who received a solitary KT between 2000 and 2016. Data from the Organ Procurement and Transplantation Network (OPTN) were used. This study was exempt of IRB and no informed consent was needed, as this is a data registry. Exclusion criteria were combined kidney-liver and kidney-pancreas ($$n = 6$$,615), transplants with missing values for kidney graft failure time or status, recipient's BMI, donor’s age and donor's BMI ($$n = 8$$,351), resulting in 250,574 kidney transplants (Fig. 1).Fig. 1Flowchart This study cohort was then divided into the following high-risk kidney transplant groups; a) recipient age at the time of transplant ≥ 70 years old ($$n = 15$$,775); b) BMI at transplant ≥ 35 kg/m2 ($$n = 25$$,976); patients receiving kidneys from donors with: c) KDPI ≥ $85\%$ ($$n = 17$$,485); d) AKI (serum creatinine ≥ 2 mg/dl) ($$n = 12$$,662); and e) or hepatitis C antibody positive (HCV +) ($$n = 4$$,223).
## Transplant center volume
For the overall population and for each of the high-risk KT groups, transplant center volume was categorized into tertiles: low-, medium-, and high-volume based on their average annual volume of that KT. Therefore, the tertiles cutoffs were different across these high-risk KT groups. For each of the high-risk group were also divided in to tertiles to determine the volumes for this specific characteristic as shown in Tables 1 and 2.Table 1Baseline characteristics for the cohort and individual high-risk groupVariablesEntire group ($$n = 250$$,574)Age older than 70 years old ($$n = 15$$,775)Recipient BMI > 35 kg/m2 ($$n = 25$$,976)Low ($$n = 84$$,903)Medium ($$n = 83$$,883)High ($$n = 81$$,788)p-valueLow ($$n = 5254$$)Medium ($$n = 5192$$)High ($$n = 5329$$)p-valueLow ($$n = 8674$$)Medium ($$n = 8630$$)High ($$n = 8672$$)p-valueRecipient and Transplant Factors Age (%) <.0001-0.4027 18–35 years$16.9\%$$16.7\%$$16.1\%$---$13.3\%$$12.4\%$$12.9\%$ 36–50 years$30.4\%$$31.1\%$$30.5\%$$34.1\%$$34.7\%$$34.4\%$ 51–65 years$39.0\%$$39.3\%$$39.3\%$---$42.9\%$$42.9\%$$42.3\%$ > 65 years$13.7\%$$12.9\%$$14.1\%$$100\%$$100\%$$100\%$$9.7\%$$10\%$$10.4\%$ Gender (%) <.0010.97440.0078 Male$61.5\%$$60.1\%$$60.0\%$$66.2\%$$66.4\%$$66.2\%$$55.3\%$$52.9\%$$54.3\%$ Female$38.5\%$$39.9\%$$40.0\%$$33.8\%$$33.6\%$$33.8\%$$44.7\%$$47.1\%$$45.7\%$ Race (%) <.0001 <.0001 <.0001 White$55.5\%$$48.5\%$$54.8\%$$66.1\%$$67.4\%$$68.1\%$$54.2\%$$49.1\%$$54.7\%$ Black$25.1\%$$28.1\%$$23.3\%$$18.2\%$$16.6\%$$14.9\%$$28.8\%$$38\%$$31.4\%$ Hispanic$12.9\%$$16.0\%$$13.8\%$$9.8\%$$9.7\%$$8.8\%$$12.7\%$$9.5\%$$11.1\%$ Other$6.4\%$$7.4\%$$8.1\%$$6\%$$6.3\%$$8.3\%$$4.3\%$$3.4\%$$2.8\%$BMI (kg/m2)27.827.527.6 <.000127.527.327.40.109837.7137.9138.22 <.0001 Time on dialysis (months) <.0001 <.0001 <.0001 $020.0\%$$21.1\%$$26.8\%$$20.6\%$$22.5\%$$28.1\%$$16.5\%$$17.1\%$$22.6\%$ 1–$129.6\%$$8.7\%$$10.2\%$$7\%$$7.9\%$$7.3\%$$7.9\%$$7\%$$9\%$ > 12 to $2414.7\%$$13.9\%$$13.3\%$$15.2\%$$15.1\%$$14.7\%$$14.2\%$$12.5\%$$14.3\%$ > $2455.6\%$$56.4\%$$49.7\%$$57.3\%$$54.5\%$$49.9\%$$61.4\%$$63.4\%$$54.1\%$ Time on waiting list (years) <.00010.0048 <.0001 0–$141.7\%$$41.0\%$$43.3\%$$37.8\%$$40\%$$40.3\%$$40.4\%$$36\%$$43.4\%$ 1 to $335.2\%$$35.0\%$$32.2\%$$36.9\%$$37.9\%$$37.3\%$$36.2\%$$35.9\%$$32.5\%$ > $323.1\%$$24.0\%$$24.5\%$$25.3\%$$22.1\%$$22.4\%$$23.4\%$$28.1\%$$24.1\%$ HLA Mismatch Level0.00060.2117 <.0001 $09.2\%$$9.1\%$$9.2\%$$6.4\%$$7\%$$6.2\%$$8.7\%$$8.9\%$$8.8\%$ 1 to $212.2\%$$12.0\%$$12.5\%$$12.1\%$$11.4\%$$11.1\%$$10.9\%$$10\%$$12.4\%$ 3 to $678.5\%$$78.9\%$$78.3\%$$81.6\%$$81.6\%$$82.8\%$$80.4\%$$81.1\%$$78.8\%$ PRA (%) <.0001 <.0001 <.0001 $057.8\%$$56.8\%$$59.4\%$$63.1\%$$67.8\%$$66.2\%$$58.7\%$$56.9\%$$60.9\%$ 0–$2018.1\%$$16.0\%$$15.5\%$$17\%$$14\%$$15.7\%$$16.8\%$$15.9\%$$15\%$ 20–$8013.8\%$$15.2\%$$14.3\%$$13.4\%$$11.4\%$$11.9\%$$14.5\%$$15.4\%$$14.2\%$ > $8010.2\%$$12.0\%$$10.8\%$$6.5\%$$6.8\%$$6.2\%$$9.9\%$$11.8\%$$9.8\%$ Induction <.0001 <.0001 <.0001 Basiliximab$29.2\%$$22.3\%$$20.2\%$$33.2\%$$34.3\%$$25.9\%$$23.5\%$$23.7\%$$14.1\%$ Alemtuzumab$7.7\%$$11.0\%$$13.8\%$$7.9\%$$8.3\%$$9.5\%$$10.2\%$$8.9\%$$21.7\%$ Thymoglobulin$41.1\%$$44.0\%$$47.4\%$$37.1\%$$38\%$$47.1\%$$47.1\%$$45.7\%$$45.8\%$ Other$0.7\%$$0.5\%$$0.9\%$$0.5\%$$0.2\%$$0.4\%$$0.3\%$$0.6\%$$0.8\%$ Missing$21.3\%$$22.2\%$$17.8\%$$21.3\%$$19.1\%$$17.1\%$$18.8\%$$21.1\%$$17.6\%$ Year of transplantation <.0001 <.0001 <.0001 2000 ~ $200534.1\%$$30.8\%$$28.2\%$$22.8\%$$23.6\%$$19.8\%$$28.1\%$$25.3\%$$23.5\%$ 2006 ~ $201136.0\%$$37.0\%$$37.1\%$$38\%$$40.4\%$$40.2\%$$37\%$$39.7\%$$38.8\%$ 2012 ~ $201629.9\%$$32.2\%$$34.7\%$$39.2\%$$36\%$$40\%$$34.9\%$$34.9\%$$37.7\%$Donor FactorsAge (years)38.938.239.9 <.000145.64747.5 <.000139.3839.6440.190.0024Gender <.00010.02 <.0001 Male$53.1\%$$53.3\%$$51.2\%$$52\%$$51.5\%$$49.4\%$$54.1\%$$55.2\%$$51.5\%$ Female$46.9\%$$46.7\%$$48.8\%$$48\%$$48.5\%$$50.6\%$$45.9\%$$44.8\%$$48.5\%$Race <.0001 <.0001 <.0001 White$72.6\%$$67.2\%$$69.8\%$$75.4\%$$76.5\%$$73.1\%$$73.7\%$$70.3\%$$71.7\%$ Black$11.9\%$$13.8\%$$12.2\%$$10.9\%$$10.5\%$$10.6\%$$11.6\%$$16.8\%$$14.9\%$ Hispanic$12.0\%$$14.4\%$$13.5\%$$9.8\%$$9.7\%$$11.4\%$$11.2\%$$9.9\%$$11.1\%$ Other$3.6\%$$4.5\%$$4.4\%$$3.9\%$$3.3\%$$4.9\%$$3.6\%$$3\%$$2.3\%$BMI26.927.026.990.00127.427.827.50.0227.6427.7828.11 <.0001Hypertension$17.4\%$$18.3\%$$16.7\%$ <$.000130.8\%$$32.4\%$$32.7\%$$0.0920.8\%$$21.6\%$$18.9\%$ <.0001Cause of death <.0001 <.00010.35 Anoxia$22.4\%$$23.6\%$$25.1\%$$22.4\%$$23.7\%$$24.5\%$$24\%$$25.5\%$$25.2\%$ CVA$34.4\%$$35.0\%$$35.2\%$$43.8\%$$46.9\%$$47.6\%$$34.2\%$$34.2\%$$34.1\%$ Other$43.2\%$$41.4\%$$39.6\%$$33.8\%$$29.4\%$$28\%$$41.7\%$$40.3\%$$40.7\%$*Cold ischemia* time (hours) <.0001 <.0001 <.0001 < $1254.6\%$$52.6\%$$58.4\%$$48\%$$46\%$$47.2\%$$53.1\%$$50.5\%$$58.2\%$ 12 to $2434.6\%$$34.8\%$$26.7\%$$40.1\%$$41.2\%$$28.8\%$$36\%$$36.1\%$$27.6\%$ > $2410.9\%$$12.6\%$$15.0\%$$12\%$$12.9\%$$24\%$$10.9\%$$13.4\%$$14.2\%$Table 2Baseline characteristics for the cohort and individual high-risk groupVariablesKDPI ≥ $85\%$ ($$n = 17$$,485)Donor with AKI ($$n = 12$$,662)Donor with Hepatitis C + ($$n = 4$$,223)Low ($$n = 5873$$)Medium ($$n = 5917$$)High ($$n = 5695$$)p-valueLow ($$n = 4227$$)Medium ($$n = 4132$$)High ($$n = 4303$$)p-valueLow ($$n = 1424$$)Medium ($$n = 1394$$)High ($$n = 1405$$)p-valueRecipient and Transplant Factors Age (%) <.0001 <.00010.16 18–$354.03.03.511.8\%$$11.9\%$$10.2\%$$2.8\%$$1.8\%$$1.9\%$ 36–$5014.5\%$$15.614.5\%$$29.2\%$$30.1\%$$27.1\%$$25.1\%$$24\%$$24.5\%$ 51–$6548.4\%$$51.5\%$$48.7\%$$43.2\%$$41.4\%$$42.5\%$$60.5\%$$61.5\%$$63.4\%$ > $6533\%$$30\%$$33.3\%$$15.8\%$$16.7\%$$20.1\%$$11.5\%$$12.7\%$$10.2\%$ Gender (%)0.670.190.11 Male$62.2\%$$62.9\%$$62.2\%$$62.4\%$$60.9\%$$62.7\%$$80.8\%$$82.5\%$$79.4\%$ Female$37.8\%$$37.1\%$$37.8\%$$37.6\%$$39.1\%$$37.3\%$$19.2\%$$17.5\%$$20.6\%$ Race (%) <.0001 <.0001 <.0001 White$50.8\%$$38.3\%$$37.3\%$$45.2\%$$44.5\%$$30.7\%$$28.9\%$$21.7\%$$18.4\%$ Black$28\%$$44.3\%$$33.5\%$$31.7\%$$41.5\%$$28.4\%$$55.8\%$$68.7\%$$71\%$ Hispanic$12.5\%$$10.7\%$$16.5\%$$15.5\%$$8.8\%$$27.7\%$$11.8\%$$7.1\%$$7.5\%$ Other$8.7\%$$6.6\%$$12.7\%$$7.6\%$$5.2\%$$13.2\%$$3.6\%$$2.4\%$$3.1\%$BMI (kg/m2)282827.2 <.000128.228.327.7 <.000127.2327.2426.920.053 Time on kidney waiting list <.0001 <.00010.003 0–1 year$31.1\%$$26.8\%$$30.7\%$$30.3\%$$27.7\%$$26.3\%$$57.9\%$$59\%$$64\%$ 1 to 3 years$40.5\%$$41\%$$39.4\%$$38.2\%$$36.7\%$$38.3\%$$32\%$$30.8\%$$28.7\%$ > 3 years$28.4\%$$32.2\%$$29.9\%$$31.5\%$$35.6\%$$35.4\%$$10.2\%$$10.2\%$$7.3\%$ HLA Mismatch Level (%)0.002 <.00010.1068 $04.3\%$$3.7\%$$3.2\%$$6.5\%$$6.1\%$$3.7\%$$1\%$$0.7\%$$0.5\%$ 1 to $24.9\%$$4.1\%$$4\%$$6.2\%$$6.5\%$$5.3\%$$3.6\%$$3.4\%$$2.2\%$ 3 to $690.8\%$$92.2\%$$92.8\%$$87.2\%$$87.5\%$$91\%$$95.4\%$$95.9\%$$97.3\%$ PRA (%) <.0001 <.0001 <.0001 $060.1\%$$61.9\%$$65.9\%$$56.2\%$$55.2\%$$63.5\%$$69.1\%$$60.4\%$$72.7\%$ 0–$2021.9\%$$18.2\%$$17.5\%$$18.7\%$$16.9\%$$14.1\%$$15.2\%$$21.2\%$$12.3\%$ 20–$8013.1\%$$14.6\%$$11.9\%$$14.1\%$$14.9\%$$13.1\%$$11.5\%$$14.3\%$$10.9\%$ > $805\%$$5.3\%$$4.7\%$$11\%$$12.9\%$$9.3\%$$4.2\%$$4.1\%$$4\%$ Induction (%) <.0001 <.0001 <.0001 Basiliximab$23.9\%$$21.2\%$$16\%$$19.1\%$$18.7\%$$13.2\%$$28.4\%$$26.2\%$$16.9\%$ Alemtuzumab$8.3\%$$13.3\%$$14.5\%$$10.8\%$$13.8\%$$10.6\%$$5.3\%$$10.8\%$$2.9\%$ Thymoglobulin$49.3\%$$42.4\%$$47.7\%$$51.6\%$$50.1\%$$59.5\%$$44.6\%$$43.2\%$$41.9\%$ Other$0.5\%$$0.5\%$$0.4\%$$0.4\%$$0.6\%$$0.4\%$$0.8\%$$0.5\%$$0.4\%$ Missing$18\%$$22.6\%$$21.3\%$$18.1\%$$16.9\%$$16.3\%$$20.9\%$$19.3\%$$37.9\%$ Year of Transplantation (%) <.0001 <.00010.13 2000 ~ $200530.6\%$$28.2\%$$27.1\%$ <$.000123.6\%$$17\%$$17.7\%$ <$.000131\%$$29.6\%$$30.7\%$0.13 2006 ~ $201141.2\%$$42.4\%$$40.9\%$$35\%$$43.1\%$$33\%$$31.5\%$$34.8\%$$35.5\%$ 2012 ~ $201628.2\%$$29.4\%$$32\%$$41.4\%$$40\%$$49.3\%$$37.4\%$$35.7\%$$33.7\%$Donor Categories Age59.358.757.4 <.000135.335.536.9 <.000138.9739.2340.040.02 Gender <.00010.030.14 Male$42.7\%$$43.5\%$$47.4\%$ <.0001 <$.000173.4\%$$71.3\%$$71\%$0.03 <$.000166.3\%$$65.4\%$$62.8\%$0.140.005 Female$57.3\%$$56.5\%$$52.6\%$$26.6\%$$28.7\%$$29\%$$33.7\%$$34.6\%$$37.2\%$ Race White$63.7\%$$53.3\%$$54.3\%$ <$.00010.2465\%$$65.7\%$$55.6\%$ <$.00010.00278.2\%$$77\%$$75.4\%$0.0050.97 Black$24.3\%$$33.1\%$$27.5\%$$18.5\%$$22.9\%$$17.5\%$$9.2\%$$13.3\%$$12.7\%$ Hispanic$8.3\%$$9.5\%$$11.7\%$$13.2\%$$8.7\%$$21.7\%$$11.4\%$$8.5\%$$10.8\%$ Other$3.7\%$$4\%$$6.5\%$$3.3\%$$2.6\%$$5.2\%$$1.3\%$$1.2\%$$1.1\%$BMI28.128.32828.628.929.226.1526.1326.32 Hypertension <.00010.010.59 No$27\%$$24.8\%$$29.8\%$ <.0001 <$.000174.3\%$$73.4\%$$71.3\%$0.01 <$.000178\%$$78.8\%$$77.2\%$0.590.22 Yes$73\%$$75.2\%$$70.2\%$$25.7\%$$26.6\%$$28.7\%$$22\%$$21.2\%$$22.8\%$ Cause of death Anoxia$10.8\%$$11.3\%$$15.3\%$ <.0001 <$.000136.5\%$$41.6\%$$46.5\%$ <.0001 <$.000129.8\%$$30.2\%$$33\%$0.22 <.0001 CVA$77.2\%$$78.7\%$$72.4\%$$24.8\%$$23.4\%$$26.3\%$$33.4\%$$31.7\%$$32.5\%$ Other$12\%$$10\%$$12.3\%$$38.7\%$$35\%$$27.2\%$$36.8\%$$38.1\%$$34.5\%$ *Cold ischemia* time (hours) < $1230.2\%$$26\%$$18.4\%$ <.0001 <$.000125.9\%$$21.3\%$$14.9\%$ <.0001 <$.000128.9\%$$22.5\%$$21.5\%$ <.0001 <.0001 12 to $2453.8\%$$53.1\%$$42\%$$56.1\%$$52.3\%$$40.3\%$$52.4\%$$49.9\%$$45.6\%$ > $2416\%$$20.9\%$$39.5\%$$18\%$$26.4\%$$44.8\%$$18.7\%$$27.5\%$$32.9\%$ Total months from diagnosis to transplantation $09.3\%$$10.5\%$$12.3\%$ <$.000110.3\%$$12.2\%$$10.5\%$ <$.00018.8\%$$8.5\%$$13.1\%$ <.0001 0–$125.9\%$$4.2\%$$5.7\%$$4.7\%$$4.4\%$$3.3\%$$10\%$$7.9\%$$11.6\%$ 12 to $2414\%$$12\%$$13.5\%$$12.3\%$$11.2\%$$10.5\%$$20.9\%$$20.7\%$$21\%$ > $2470.8\%$$73.4\%$$68.5\%$$72.7\%$$72.2\%$$75.8\%$$60.4\%$$62.9\%$$54.3\%$
## Outcomes and covariates
Death Censored Graft Loss (DCGL) was defined as returning to dialysis or receiving another renal transplant. Death was defined as recipient demise. Recipient characteristics included age, gender, race, BMI at time of KT, pre-transplant dialysis, and time on dialysis. Donor characteristics include KDPI, calculated using 10 donor factors including age, height, weight, ethnicity, history of hypertension, history of diabetes, cause of death, serum creatinine, HCV serological status, and Donation after Cardiac Death (DCD) Status, as well as each component separately.
## Statistical methods
Patient characteristics were summarized using proportions for categorical variables and means and standard deviations for continuous variables. Differences between center volume categories (high, medium, low) were compared using χ2 test for categorical variables and analysis of variance test or Kruskal Wallis tests for continuous variables, depending on the distribution of the variable. Kaplan–Meier analyses was performed on DCGF and death for the three categories of transplant center volume were compared using Log Rank tests.
Multivariable Cox regression analyses were used to assess the independent association of center volume with the two outcomes (DCGL and death), controlling for all aforementioned recipient and donor characteristics as well as transplant factors e.g., cold ischemic time greater than 24 h, except for the variable used to define high-risk. For each group of the high-risk group were also analyzed with Cox regression for our two outcomes (DCGL and death). DCGL was evaluated at 3 months, 1, 5 and 10 years of follow-up following KT. The results for 10 years are reported in the main text, and the other results are reported in the Supplemental Materials. All tests are two-sided. A p-value less than 0.05 was considered statistically significant for all tests. All analyses were performed using SAS 9.4 software (Cary, NC).
## Results
The cohort included 250,574 KT performed in 285 transplant centers between 2000 and 2016 (Fig. 1). Overall, patients transplanted at high volume centers were more likely to be older (age > 65 years), have longer waiting time and cold ischemia time, and to receive T cell depletion (thymoglobulin or alemtuzumab) for induction (Tables 1 and 2). The baseline characteristics of each high-risk group stratified by individual transplant center volume characteristics are described also in Tables 1 and 2.
## Death censored graft loss and death for the entire group
Compared with high CV, patients undergoing KT at low CV (adjusted hazard ratio, [aHR] = 1.04; $95\%$ confidence interval [CI], 1.02–1.07) and at medium CV (aHR = 1.03; $95\%$ CI, 1.00–1.05) had higher risk of DCGL at 10 years. Furthermore, low (but not medium) CV volume was associated with higher risk for death at 10 years (aHR = 1.07; $95\%$ CI, 1.05–1.09), when compared to medium center volume (Supplemental Table 1, Fig. 2A.1 and A.2).Fig. 2Subgroup *Multivariate analysis* for transplant center volume. A.1 Overall kidney transplants center volume associated DCGL. A.2 Overall kidney transplants center volume associated death. B.1 Center Volume for Recipient age ≥ 70 years associated DCGL. B.2 Center Volume for Recipient age ≥ 70 years associated death. C.1 Center Volume for Recipient BMI ≥ 35 kg/m2 associated DCGL. C.2 Center Volume for *Recipient a* BMI ≥ 35 kg/m2 associated death. D.1 Center Volume for Transplants with KDPI ≥ $85\%$ associated DCGL. D.2 Center Volume for Transplants with KDPI ≥ $85\%$ associated death. E.1 Center Volume for Transplants with Donor AKI with Serum Creatinine ≥ 2 mg/dl associated DCGL. E.2 Center Volume for Transplants with Donor AKI with Serum Creatinine ≥ 2 mg/dl associated death. F.1 Center Volume for Transplants with Donor with hepatitis C associated DCGL. F.2 Center Volume for Transplants with Donor with hepatitis C associated death
## Recipient Age greater or equal to 70 years old
The risk for death at 10 years for recipient age ≥ 70 years, when compared with high CV, was higher in low (aHR = 1.07; $95\%$ CI, 1.01–1.14) and medium CV (aHR = 1.09; $95\%$ CI, 1.03–1.15) (Fig. 2B.2). However, no statistically significant difference was observed in risk for DCGL for low, medium CV when compared with high CV (Fig. 2B.1) (Supplemental Table 2).
## Kidney recipients with Body Mass Index greater or equal than 35 kg/m2
The risk for DCGL in patients with BMI ≥ 35 kg/m2 was higher in low volume centers (aHR = 1.11; $95\%$ CI, 1.03–1.19) and for medium CV (aHR = 1.07, $95\%$ CI, 1.00–1.15) when compared with the high CV at 10 years (Fig. 2C.1). There was no difference for death when comparing high CV with low and medium CV (Fig. 2C.2) (Supplemental Table 3).
## Kidney Donor Profile Index greater or equal than 85%
For low CV and medium CV there was no difference in DCGL or death when comparing with high CV (Supplemental Table 4) (Fig. 2D.1 and D.2).
## Donors with acute kidney injury
Compared to high CV, there was no evidence that low or medium CV was associated with different risk for DCGL at any time points (Fig. 2E.1). However, we observed an increased risk of death for low CV at 3 month (aHR = 1.48; $95\%$ CI, 1.04–2.10), 5 year (aHR = 1.13; $95\%$ CI, 1.00–1.28) but not at 1 year or 10 years when compared with high CV (Fig. 2E.2) (Supplemental Table 5).
## Donors with Hepatitis C
For low and medium CV, no statistically significant difference in DCGL or death was observed, when compared with high CV. However, medium CV was associated with lower risk of death at 10 years when compared with high CV (aHR = 0.87; $95\%$ CI, 0.76–0.99), but not statistically significant different at 3 months, 1 year, or 5 years (Supplemental Table 6) (Fig. 2F.1 and F.2).
## Kaplan–meier analysis
A statistically significant difference in DCGL by center volume was observed ($p \leq 0.001$) and overall death ($p \leq 0.001$) (Fig. 3A and B).Fig. 3A Kaplan Meier Overall kidney transplants center volume associated DCGL. B Kaplan Meier Overall kidney transplants center volume associated death
## Discussion
Accepting kidneys from high-risk donors and performing kidney transplantation in high-risk recipients requires adequate staff support to monitor their post-transplant outcomes. In this large, national cohort analysis, when examining high-risk recipient subgroups, we found that low CV (compared to high CV) was associated with higher risk of death in elderly (age ≥ 70 years) and higher risk of graft failure in obese patients (BMI ≥ 35 kg/m2) at short- and long-term follow-up.
Elderly recipients require more intense medical care and closer follow up after transplantation. The number of elderly recipients continue to increase over the last 10 years [8]. A recent report showed an increase of elderly recipients from $17.6\%$ in 2009 to $24.2\%$ in 2019 [22]. Older recipients are associated with an increased number of comorbidities and at higher risk of infection, cardiovascular diseases, and malignancies, and they are more prevalent to be frail [9, 23–28]. In terms of obesity, it has been associated with increased risk for peritransplant complications, including delayed graft function, wound infections, and graft loss [29–32]. Resources and care practices to manage these complications might be better at high CV, resulting in better outcomes as reported in our paper.
The cause of higher risk of DCGL in obese recipients and death in elderly recipients in low volume centers is most likely multifactorial. Higher volume centers presumably have a complex multidisciplinary team, and broader resources for management and follow up [33, 34]. Higher volume centers might have a larger number of transplant nephrologists and surgeons with different expertise and interests. The impact of a large specialized network of transplant coordinators likely also helps manage and follow up patients with tailored protocols that are needed to improve patient and graft survival. In addition, high volume centers are more likely to have increased availability of other advanced specialties such as transplant cardiology, transplant infectious disease and oncology that may influence outcomes. High Volume Center for each specific high -risk group for elderly and obese might have lower threshold to accept these population that they have practice care specific for them and prepare for potential for the complications.
When compared with high center volume, our study did not find any difference of DCGL or death for low center volume in patients receiving kidneys from donors with AKI, HCV +, or high KDPI. Given the increasing prevalence of KT using kidneys from donors with KDPI ≥ $85\%$, AKI, and HCV +, it is important to evaluate what factors contribute to the improved outcomes [2, 35, 36].
Our findings for high risk donors are consistent with prior publications that selected deceased kidney donors with AKI were not associated with higher risk of graft failure or death compared to those receiving non-AKI kidneys [4, 5, 37–40]. In the new era of effective direct acting antivirals (DAA), patients receiving kidneys from donors with HCV + do not seem to have an increased risk when compared with those receiving kidneys from donors with seronegative hepatitis C [6, 41, 42]. Our study did not asses the effect of DAA at long term as this were started in 2014 and our study population included patients transplanted up to 2016. High KDPI is a well-known risk factor to have decreased survival of the allograft [43]. Our study showed center volume by KDPI did not seem to play a role in DCGL and death.
Our study has several important strengths. To our knowledge, this is one of the first studies to highlight the outcome implications of KT CV for specific risk factors in the current era with the increasing prevalence of having higher risk donors with high KDPI, donor with AKI and donor with HCV +. Second, we used national data that allowed us to include a large number of kidney transplants with long-term follow-up for several time points of interest. There are alsolimitations. First, it is a retrospective study based on registry data, which is limited by available variables and existing data quality. Second, the analyses did not account for patient socioeconomic status, which might impact transplant outcomes.
In conclusion, elderly patients who received KT in low-volume centers had increased risk of death compared to those who received KT in high volume centers. Moderate obese recipients who received KT in low-volume centers had increased risk of graft loss compared to those who received KT in high volume centers. Future studies should seek to identify care processes that support optimal outcomes after kidney transplantation irrespective of center volume.
## Supplementary Information
Additional file 1: Supplemental Table 1. Overall kidney transplants center volume associated Death Censored Graft Failure and Death. Supplemental Table 2. Kidney transplants center volume for recipient age> 70 years associated Death Censored Graft Failure and Death. Supplemental Table 3. Kidney transplants center volume for Recipient BMI> 35 Kg/m2 associated Death Censored Graft Failure and Death. Supplemental Table 4. Kidney transplants center volume for Donors with KDPI>$85\%$ associated Death Censored Graft Failure and Death. Supplemental Table 5. Kidney transplant center volume for Transplants with Donor AKI with Serum Creatinine > 2 mg/dl associated Death Censored Graft Failure and Death. Supplemental Table 6. Kidney transplant center volume for Transplants with Donors with Hepatitis C associated Death Censored Graft Failure and Death.
## Public access open
The database was obtain UNOS/OPTN which is public access open for reasonable request to UNOS/OPTN.
## References
1. Axelrod DA, Schnitzler MA, Xiao H. **An economic assessment of contemporary kidney transplant practice**. *Am J Transplant* (2018.0) **18** 1168-1176. DOI: 10.1111/ajt.14702
2. 2.Collins AJ, Foley RN, Chavers B, et al. 'United States Renal Data System 2011 Annual Data Report: Atlas of chronic kidney disease & end-stage renal disease in the United States. Am J Kidney Dis. 2012;59(1 Suppl 1):A7, e1–420.
3. Port FK, Merion RM, Goodrich NP, Wolfe RA. **Recent trends and results for organ donation and transplantation in the United States, 2005**. *Am J Transplant Off J Am Soc Transplant Am Soc Transplant Surg* (2006.0) **6** 1095-1100. DOI: 10.1111/j.1600-6143.2006.01268.x
4. Heilman RL, Smith ML, Kurian SM. **Transplanting Kidneys from Deceased Donors With Severe Acute Kidney Injury**. *Am J Transplant Off J Am Soc Transplant Am Soc Transplant Surg* (2015.0) **15** 2143-2151. DOI: 10.1111/ajt.13260
5. Heilman RL, Smith ML, Smith BH. **Long-term Outcomes Following Kidney Transplantation From Donors With Acute Kidney Injury**. *Transplantation* (2019.0) **103** e263-e272. DOI: 10.1097/TP.0000000000002792
6. Potluri VS, Goldberg DS, Mohan S. **National Trends in Utilization and 1-Year Outcomes with Transplantation of HCV-Viremic Kidneys**. *J Am Soc Nephrol* (2019.0) **30** 1939-1951. DOI: 10.1681/ASN.2019050462
7. Dahmen M, Becker F, Pavenstädt H, Suwelack B, Schütte-Nütgen K, Reuter S. **Validation of the Kidney Donor Profile Index (KDPI) to assess a deceased donor’s kidneys’ outcome in a European cohort**. *Sci Rep* (2019.0) **9** 11234. DOI: 10.1038/s41598-019-47772-7
8. Lentine KL, Pastan S, Mohan S. **A roadmap for innovation to advance transplant access and outcomes: a position statement from the national kidney foundation**. *Am J Kidney Dis* (2021.0) **78** 319-332. DOI: 10.1053/j.ajkd.2021.05.007
9. Knoll GA. **Kidney transplantation in the older adult**. *Am J Kidney Dis* (2013.0) **61** 790-797. DOI: 10.1053/j.ajkd.2012.08.049
10. Gaston RS, Fieberg A, Helgeson ES. **Late Graft Loss After Kidney Transplantation: Is "Death With Function" Really Death With a Functioning Allograft?**. *Transplantation* (2020.0) **104** 1483-1490. DOI: 10.1097/TP.0000000000002961
11. Kwan JM, Hajjiri Z, Metwally A, Finn PW, Perkins DL. **Effect of the Obesity Epidemic on Kidney Transplantation: Obesity Is Independent of Diabetes as a Risk Factor for Adverse Renal Transplant Outcomes**. *PLoS ONE* (2016.0) **11** e0165712-e0165712. DOI: 10.1371/journal.pone.0165712
12. Aziz F, Ramadorai A, Parajuli S. **Obesity: An Independent Predictor of Morbidity and Graft Loss after Kidney Transplantation**. *Am J Nephrol* (2020.0) **51** 615-623. DOI: 10.1159/000509105
13. Asrani SK, Kim WR, Edwards EB. **Impact of the center on graft failure after liver transplantation**. *Liver Transplantation* (2013.0) **19** 957-964. DOI: 10.1002/lt.23685
14. Arnaoutakis GJ, George TJ, Allen JG. **Institutional volume and the effect of recipient risk on short-term mortality after orthotopic heart transplant**. *J Thorac Cardiovasc Surg* (2012.0) **143** 157-167, 167.e151. DOI: 10.1016/j.jtcvs.2011.09.040
15. Kilic A, George TJ, Beaty CA, Merlo CA, Conte JV, Shah AS. **The effect of center volume on the incidence of postoperative complications and their impact on survival after lung transplantation**. *J Thorac Cardiovasc Surg* (2012.0) **144** 1502-1508. DOI: 10.1016/j.jtcvs.2012.08.047
16. Shuhaiber JH, Moore J, Dyke DB. **The effect of transplant center volume on survival after heart transplantation: a multicenter study**. *J Thorac Cardiovasc Surg* (2010.0) **139** 1064-1069. DOI: 10.1016/j.jtcvs.2009.11.040
17. Thabut G, Christie JD, Kremers WK, Fournier M, Halpern SD. **Survival differences following lung transplantation among US transplant centers**. *JAMA* (2010.0) **304** 53-60. DOI: 10.1001/jama.2010.885
18. Weiss ES, Allen JG, Meguid RA. **The impact of center volume on survival in lung transplantation: an analysis of more than 10,000 cases**. *Ann Thorac Surg* (2009.0) **88** 1062-1070. DOI: 10.1016/j.athoracsur.2009.06.005
19. Axelrod DA, Guidinger MK, McCullough KP, Leichtman AB, Punch JD, Merion RM. **Association of center volume with outcome after liver and kidney transplantation**. *Am J Transplant Off J Am Soc Transplant Am Soc Transplant Surg* (2004.0) **4** 920-927. DOI: 10.1111/j.1600-6143.2004.00462.x
20. Schurman SJ, Stablein DM, Perlman SA, Warady BA. **Center volume effects in pediatric renal transplantation. A report of the North American Pediatric Renal Transplant Cooperative Study**. *Pediatric Nephrology (Berlin, Germany)* (1999.0) **13** 373-378. DOI: 10.1007/s004670050626
21. Alhamad T, Malone AF, Brennan DC. **Transplant Center Volume and the Risk of Pancreas Allograft Failure**. *Transplantation* (2017.0) **101** 2757-2764. DOI: 10.1097/TP.0000000000001628
22. 22.Hart A, Lentine KL, Smith JM, et al. OPTN/SRTR 2019 Annual Data Report: Kidney. 2021;21(S2): 21–137.
23. Trouillhet I, Benito N, Cervera C. **Influence of age in renal transplant infections: cases and controls study**. *Transplantation* (2005.0) **80** 989-992. DOI: 10.1097/01.TP.0000173822.05877.D7
24. Martins PN, Pratschke J, Pascher A. **Age and immune response in organ transplantation**. *Transplantation* (2005.0) **79** 127-132. DOI: 10.1097/01.TP.0000146258.79425.04
25. Meier-Kriesche H-U, Ojo AO, Hanson JA, Kaplan B. **Exponentially increased risk of infectious death in older renal transplant recipients**. *Kidney Int* (2001.0) **59** 1539-1543. DOI: 10.1046/j.1523-1755.2001.0590041539.x
26. Farrugia D, Mahboob S, Cheshire J. **Malignancy-related mortality following kidney transplantation is common**. *Kidney Int* (2014.0) **85** 1395-1403. DOI: 10.1038/ki.2013.458
27. Kappes U, Schanz G, Gerhardt U, Matzkies F, Suwelack B, Hohage H. **Influence of age on the prognosis of renal transplant recipients**. *Am J Nephrol* (2001.0) **21** 259-263. DOI: 10.1159/000046259
28. Legeai C, Andrianasolo RM, Moranne O. **Benefits of kidney transplantation for a national cohort of patients aged 70 years and older starting renal replacement therapy**. *Am J Transplant Off J Am Soc Transplant Am Soc Transplant Surg* (2018.0) **18** 2695-2707. DOI: 10.1111/ajt.15110
29. Lynch RJ, Ranney DN, Shijie C, Lee DS, Samala N, Englesbe MJ. **Obesity, surgical site infection, and outcome following renal transplantation**. *Ann Surg* (2009.0) **250** 1014-1020. DOI: 10.1097/SLA.0b013e3181b4ee9a
30. Meier-Kriesche HU, Kaplan B. **Waiting time on dialysis as the strongest modifiable risk factor for renal transplant outcomes: a paired donor kidney analysis**. *Transplantation* (2002.0) **74** 1377-1381. DOI: 10.1097/00007890-200211270-00005
31. Aalten J, Christiaans MH, de Fijter H. **The influence of obesity on short- and long-term graft and patient survival after renal transplantation**. *Transplant International* (2006.0) **19** 901-907. DOI: 10.1111/j.1432-2277.2006.00367.x
32. Molnar MZ, Kovesdy CP, Mucsi I. **Higher recipient body mass index is associated with post-transplant delayed kidney graft function**. *Kidney Int* (2011.0) **80** 218-224. DOI: 10.1038/ki.2011.114
33. Finks JF, Osborne NH, Birkmeyer JD. **Trends in hospital volume and operative mortality for high-risk surgery**. *N Engl J Med* (2011.0) **364** 2128-2137. DOI: 10.1056/NEJMsa1010705
34. Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE, Lucas FL. **Surgeon volume and operative mortality in the United States**. *N Engl J Med* (2003.0) **349** 2117-2127. DOI: 10.1056/NEJMsa035205
35. Moist LM, Fenton S, Kim JS. **Canadian Organ Replacement Register (CORR): reflecting the past and embracing the future**. *Can J Kidney Health Dis* (2014.0) **1** 26-26. DOI: 10.1186/s40697-014-0026-5
36. Snoeijs MG, Schaubel DE, Hene R. **Kidneys from donors after cardiac death provide survival benefit**. *J Am Soc Nephrol* (2010.0) **21** 1015-1021. DOI: 10.1681/ASN.2009121203
37. Liu C, Hall IE, Mansour S, Thiessen Philbrook HR, Jia Y, Parikh CR. **Association of Deceased Donor Acute Kidney Injury With Recipient Graft Survival**. *JAMA Netw Open* (2020.0) **3** e1918634-e1918634. DOI: 10.1001/jamanetworkopen.2019.18634
38. Oppong YD, Farber JL, Chervoneva I, Martinez Cantarin MP. **Correlation of acute tubular injury in reperfusion biopsy with renal transplant outcomes**. *Clin Transplant* (2016.0) **30** 836-844. DOI: 10.1111/ctr.12757
39. Yu CC, Ho HC, Yu TM. **Kidneys from standard-criteria donors with different severities of terminal acute kidney injury**. *Transpl Proc* (2014.0) **46** 3335-3338. DOI: 10.1016/j.transproceed.2014.11.002
40. Anil Kumar MS, Khan SM, Jaglan S. **Successful transplantation of kidneys from deceased donors with acute renal failure: Three-year results**. *Transplantation* (2006.0) **82** 1640-1645. DOI: 10.1097/01.tp.0000250908.62948.8f
41. Goldberg DS, Abt PL, Blumberg EA. **Trial of Transplantation of HCV-Infected Kidneys into Uninfected Recipients**. *N Engl J Med* (2017.0) **376** 2394-2395. DOI: 10.1056/NEJMc1705221
42. Molnar MZ, Nair S, Cseprekal O. **Transplantation of kidneys from hepatitis C–infected donors to hepatitis C–negative recipients: Single center experience**. *Am J Transplant* (2019.0) **19** 3046-3057. DOI: 10.1111/ajt.15530
43. Gandolfini I, Buzio C, Zanelli P. **The Kidney Donor Profile Index (KDPI) of marginal donors allocated by standardized pretransplant donor biopsy assessment: distribution and association with graft outcomes**. *Am J Transplant Off J Am Soc Transplant Am Soc Transplant Surg* (2014.0) **14** 2515-2525. DOI: 10.1111/ajt.12928
|
---
title: Are threat perceptions associated with patient adherence to antibiotics? Insights
from a survey regarding antibiotics and antimicrobial resistance among the Singapore
public
authors:
- Si Yu Lee
- Yang Shanshan
- May O. Lwin
journal: BMC Public Health
year: 2023
pmcid: PMC10029286
doi: 10.1186/s12889-023-15184-y
license: CC BY 4.0
---
# Are threat perceptions associated with patient adherence to antibiotics? Insights from a survey regarding antibiotics and antimicrobial resistance among the Singapore public
## Abstract
### Background
Public health strategies to improve patient adherence to antibiotics rely mostly on raising awareness of the threat of antimicrobial resistance (AMR) and improving knowledge about antibiotics. We aimed to evaluate how adherence to antibiotics relates to knowledge and the threat perceptions proposed by the Protection Motivation Theory (PMT).
### Method
A cross-sectional online survey was conducted in September-December 2020 with 1002 participants aged 21-70 years in Singapore. Two items, which were reverse coded, evaluated adherence to antibiotics: ‘how often do you obtain antibiotics that were left over from the previous prescription’ and ‘how often did you treat yourself with antibiotics in the past year’. Questions about the PMT-related constructs, and knowledge regarding antibiotics and AMR knowledge were also included. Hierarchical regression models were performed at a $5\%$ significance level.
### Results
Adherence to antibiotics was associated with knowledge level (β = 0.073, $p \leq 0.05$), education level (β = − 0.076, $p \leq 0.01$), and four of the five PMT constructs: “perceived response cost” (β = 0.61, $p \leq 0.01$), “perceived response efficacy of adherence to antibiotic” (β = 0.096, $p \leq 0.01$), “perceived susceptibility to AMR” (β = 0.097, $p \leq 0.01$), and “perceived severity of AMR” (β = − 0.069, $p \leq 0.01$). Knowledge about AMR, perceived self-efficacy in adhering to antibiotics, age, and sex were not associated with adherence.
### Conclusions
In Singapore, patient adherence to antibiotics appear to be driven by the perceived costs of visiting a doctor to obtain antibiotics, followed by perceptions of AMR as a threat and to a lesser extent, knowledge about antibiotics. Public health strategies to mitigate antibiotic misuse should consider these patient barriers to medical care.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12889-023-15184-y.
## Introduction
Antimicrobial resistance (AMR) is one of the top ten global public health threats [1]. While it is a naturally occurring process, the misuse and overuse of antimicrobials have accelerated AMR. In fact, a main cause of AMR is patient non-adherence to antibiotics [2], with a meta-analysis of studies conducted across four continents observing that nearly half of the respondents stopped their antibiotic course prematurely once they begin to feel better [3]. Adherence to antibiotics is, therefore, an important gap in the global efforts against AMR.
Patient adherence is a complex and multifactorial behavior that refers to the extent to which a patient’s behavior corresponds with recommendations from a healthcare provider [4, 5]. In the context of antibiotic use, adherence involves the patient completing the antibiotic course as prescribed and not self-medicating without a doctor’s prescription [6]. Patient-related determinants of suboptimal adherence include poor attitudes and perceptions toward antibiotics, poor understanding of the received regimen, and a rapid improvement in clinical outcomes [7–9]. On the clinician level, determinants include poor recognition of non-adherence, complex and multidrug prescriptions, and insufficient patient-physician communication [10]. On the health system level, determinants include poor inpatient and outpatient care coordination, and antibiotic accessibility [11]. This multiplicity of determinants makes the identification of modifiable determinants a critical research priority.
One major psychosocial determinant of adherence to antibiotics is how the public perceives the threat of AMR. Threat perceptions refer to how individuals appraise their susceptibility to and the severity of the adverse consequence of an event or action. As a core component of health behavioral theories such as the Protection Motivation Theory (PMT), threat perceptions have been identified as a driver of medication adherence in the management of chronic diseases such as diabetes and cardiovascular disease [12, 13]. However, no study to date has investigated how the public’s threat perceptions of AMR relate to their adherence to antibiotics.
This study aims to address this gap by applying the PMT to investigate the relationship between the public’s threat perceptions of AMR and their adherence to antibiotics. The PMT provides a theoretical framework to understand why and how people respond to health threats. It posits that people’s response to health threats depends on how they appraise the threats and the preventive action [14]. Then, according to PMT, the patient adherence to antibiotics will depend on their perceived severity of and vulnerability to AMR, as well as their perceived response efficacy, self-efficacy, and response cost of non-adherence.
This research is based in Singapore, a city-state in Southeast Asia with a population size of 5.5 million [15]. Antibiotics are largely classified as prescription-only drugs in Singapore under the regulation of the National Health Science Authority and it is usually dispensed to the public through primary care providers in government-funded polyclinics or private clinics operated by general practitioners [16]. The public’s knowledge of antibiotics and AMR are comparable to those of other countries - a recent study has found the public in Singapore to hold similar beliefs and misconceptions commonly identified in population-level antibiotic surveys of other countries [17].
## Study design
The present study is performed as a part of a larger study to understand the knowledge, attitudes, and practices about antimicrobial use and AMR among the Singapore public. It was conducted from September to December 2020 using an online survey hosted on the survey platform Qualtrics. The survey respondents were recruited by Qualtrics following a set of criteria on age, gender, and ethnicity to ensure that the participant makeup is nationally representative. A total of 1002 citizens or permanent residents in Singapore between the ages of 21 and 70 were recruited for the study.
## Survey design
The survey was conducted in English which is commonly spoken in Singapore. To ensure validity, it was pilot tested to identify and rectify any issues. It has four main sections: (i) socio-demographic information, (ii) antibiotics-related behaviors, (iii) knowledge about antibiotics and AMR, as well as (iv) PMT-related perceptions about antibiotics and AMR. Relevant to the present study, socio-demographic information was measured using five questions on sex, residency status, ethnicity, education level, and monthly household income.
Patient Adherence to antibiotics was operationalized by measuring the dimensions related to use of leftover antibiotics and self-medication with antibiotics in the previous year.
The items are ‘how often do you obtain antibiotics that were leftover from the previous prescription’ and ‘how often did you treat yourself (self-medicate without a doctor’s prescription) with antibiotics in the past one year’. Participants answered these two questions on a five-point Likert scale (1 = ‘never’ and 5 = ‘always’). Both items were then reverse coded and averaged into an adherence score that ranges from 1 (low adherence) to 5 (high adherence). The inter-item correlation of the 2 items is $r = 0.67.$
Knowledge about antibiotics was assessed using 28 true or false questions adapted from past studies [18, 19] (see TS 1). It comprises 11 questions on general facts about antibiotics, 10 questions on the diseases that can be treated with antibiotics, and 7 questions on the side effects of antibiotics. A score of one point was given for each correct answer. For each respondent, points were summated into an overall antibiotic knowledge score, which ranges from 0 to 28. Knowledge about AMR was assessed using 7 true or false questions adapted from the WHO’s multi-country public awareness survey on AMR [20] (see TS 2). One point was given for each correct answer and the points were then summated into an overall AMR knowledge score that ranges from 0 to 7 for each respondent.
Lastly, PMT-related perceptions of antibiotics and AMR were assessed with 17 questions using a 5-point Likert scale (See TS 3). An index for each of the PMT constructs was then created by summating the score for its constituting items.
## Data analysis
Descriptive analysis was run to assess the distribution of the participant’s socio-demographics, antibiotics knowledge, and AMR knowledge. Cronbach’s Alpha was also used to assess the internal consistency of adherence to antibiotics and the five constructs under PMT (see TS 4). All scales have a moderate to high level of reliability with a Cronbach’s alpha score of.66 to.81.
As guided by past applications of PMT in health [21, 22], we employed hierarchical regression to examine how adherence to antibiotic is associated with the PMT constructs, knowledge about antibiotics and AMR, as well as socio-demographical variables. This method allows us to compare the effect sizes across models comprising different predictors, and hence determine the best determinants of adherence to antibiotics. The hierarchical regression was conducted in a few steps. First, only age, sex, and education level were included to form regression Model 1. Then, knowledge about antibiotics and AMR were added to the socio-demographic variables to form regression Model 2. In the last step, we further added the five PMT constructs in the model to form regression Model 3. Additionally, we wanted to compare whether the composite constructs of threat appraisal (sum of perceived severity and perceived susceptibility) and coping appraisal (sum of perceived self-efficacy and perceived response efficacy minus perceived response cost) would be more effective than its five individual PMT constructs. We hence performed a separate regression model by replacing the five PMT constructs with threat appraisal and coping appraisal to form Model 4.
## Participants
The questionnaire was completed by 1002 participants. Thirty-five responses were removed as they were either completed in under 330 seconds or had duplicate IP addresses. The remaining 967 valid responses were used for analysis. Table 1 details the sociodemographic characteristics of the respondents. The average age of the respondents was 44.41 (SD = 12.14). The distribution of sex was roughly balanced with $50.8\%$ female respondents and $49.1\%$ male respondents. The ethnic distribution of the sample is also approximately representative of the ethnic makeup of the Singapore population in 2021 [23], with $75.0\%$ of the respondents being Chinese, $14.5\%$ being Malay, and $7.4\%$ Indian. Table 1Sociodemographic characteristics of participantsN = 967Sociodemographic Characteristicsn (%)Sex Female491 (50.8) Male475 (49.1)Residency Status Singapore Citizen846 (87.5) Singapore Permanent Resident121 (12.5)Age 21–29130 (13.4) 30–39234 (24.2) 40–49257 (26.6) 50–59223 (23.1) 60 and above123 (12.7)Ethnicity Chinese725 (75.0) Malay140 (14.5) Indian72 (7.4) Others30 (3.1)Education No formal schooling4 (0.4) PSLE or equivalent9 (0.9) GCE O/N Level155 (16.0) GCE A Level/ diploma263 (27.2) Degree/higher education536 (55.4)Monthly household income (In SGD) Below $100037 (3.8) 1000–4999257 (26.6) 5000–9999329 (34.0) 10,000–14,999179 (18.5) 15,000–19,99978 (8.1) 20,000 and over61 (6.3) Do not wish to answer26 (2.7)
## Antibiotics adherence
Table 2 presents the descriptive statistics of respondents’ adherence to antibiotic recommendations. Adherence to antibiotics recommendations is generally high among the respondents with $60.8\%$ of them “always adhering to antibiotics recommendations” ($$n = 578$$) and $26.8\%$ adhering to them most of the time ($$n = 255$$).Table 2Descriptive statistics of adherence to antibioticsNMean (SD)Frequency, N(%)NeverSometimesOftenMost of the timeAlwaysAdherence to antibiotics9514.57 (0.73)10 (1.1)23 (2.4)85 (8.9)255 (26.8)578 (60.8) Item 1: Self-medicate without doctor’s prescription9531.45 (0.83)670 (69.3)183 (18.9)64 (6.6)23 (2.4)13 (1.3) Item 2: Use antibiotics from previous leftover prescription9651.41 (0.84)714 (73.8)162 (16.8)48 (5.0)24 (2.5)17 (1.8)
## Knowledge about antibiotics and AMR
Respondents correctly answered an average of 12.5 (SD = 5.01) of the 28 questions on antibiotics. On the general statements asked about antibiotics, −respondents correctly answered an average of 3.61 (SD = 2.81) of the 11 questions. On the side effects of antibiotics, respondents correctly identified an average of 2.00 (SD = 1.33) of the 7 side effects of antibiotics, with fever, bloating, and vomiting being the three least known side effects. As for AMR knowledge, respondents correctly answered an average of 3.15 (SD = 1.71) of the 7 questions.
## Predicting adherence to antibiotics with PMT
Table 3 presents the descriptive statistics of the PMT constructs and their correlations. Of the five PMT constructs, perceived response cost has the lowest response rating with an average of 1.48 (SD =.82) on a 5-point Likert scale. This means that respondents, on average, tend to perceive adherence to antibiotic as an action with low cost. Perceived susceptibility to AMR has the second-lowest response rating with an average of 3.11 (SD = 0.77) on a 5-point Likert scale. This means that most respondents tended to be neutral about whether AMR will affect themselves or the people around them. Perceived response efficacy, on the other hand, has the highest response rating of 3.92 (SD = 0.74), meaning that respondents tended to believe that adherence to antibiotics is effective in coping with the threat of AMR.Table 3Descriptive statistics and intercorrelations of PMT constructs and adherence to antibioticsVariablesPearson correlation coefficientMean (SD)Range123456781. Adherence to antibiotics–4.57 (0.73)1–52. Susceptibility.27**–3.11 (0.77)1–53. Severity.08*.005 ($$p \leq .12$$)–3.72 (0.73)1–54. Self-Efficacy.11**−.07* ($$p \leq .03$$).49**–3.82 (0.67)1–55. Response-Efficacy.23**0.04 ($$p \leq .21$$).53**.48**–3.92 (0.74)1–56. Response Cost−.68**−.28**−.11**−.10**−.20**–1.49 (0.82)1–57. Threat appraisal.25**.74**.71**.28**.38**−.28**–3.41 (0.54)1–56. Coping appraisal.51**.13**.53**.70**.78**−.65**.44**–2.08 (0.52)1–5*Denotes correlation is significant at the 0.05 level**Denotes correlation is significant at the 0.01 level The results also show that adherence to antibiotics is significantly correlated with all five PMT constructs. Coping and threat appraisals are significantly correlated with antibiotics adherence ($r = .$ 51and.25 respectively, $p \leq .01$). Among the five constructs, the perceived response cost of adherence to antibiotics has the strongest correlation with adherence to antibiotics (r = −.69, $p \leq .001$). It is also the only construct negatively correlated with adherence to antibiotics. On the other hand, the perceived severity of AMR has the weakest correlation with adherence to antibiotics ($r = .08$, $$p \leq .01$$).
Table 4 shows the results of the hierarchical regression analysis to adherence to antibiotics based on the PMT model. Model 1 served as the baseline that informs us how participants’ demographic traits could influence their adherence to antibiotics. It accounted for only $6.3\%$ of the variance in adherence to antibiotics, F [3, 842] = 19.82, $p \leq 0.001$ with R2 =.066. For Model 2, the inclusion of the knowledge about antibiotics and AMR improved the effect size of the model, from R2 =.066 to R2 =.14, F [2, 840] = 34.62, $p \leq 0.001$, specifically. Finally, Model 3, with five PMT constructs added to the analysis, presented the best predictive model with the biggest R2 change, F [5, 835] = 126, $p \leq 0.001$, which increased the R2 by.37. As shown in the results in Model 3, perceived response cost emerged as the most salient predictor of adherence to antibiotics, t = − 22.57, $p \leq .001.$ One unit increase in perceived response cost decreases the average adherence score, which ranges from 1 to 5, by.61 ((β =.61, $p \leq 0.001$). Perceived response efficacy of adherence to antibiotics ($t = 3.1$, $$p \leq .002$$; β =.096), perceived susceptibility to AMR ($t = 3.6$, $p \leq .001$; β =.097), education level (t = − 3.02, $p \leq .001$, β = −.076), antibiotic knowledge ($t = 2.55$, $$p \leq .01$$; β = 0.073), and perceived severity of AMR (t = − 2.17, $$p \leq .03$$, β = −.069) also emerged as significant predictors of adherence to antibiotics. Table 4Summary of hierarchical regression analysis for variables predicting adherence to antibioticsPredictorsModel 1Model 2Model 3Model 4BSE BβBSE BβBSE BβBSE BβAge0.0130.0020.222**0.0110.0020.183**0.0030.0020.049 ($$p \leq .06$$)0.0050.0020.077* ($$p \leq .01$$)Sex0.0430.0490.029 ($$p \leq .38$$)0.0120.0480.008 ($$p \leq .80$$)−0.0310.036−0.021 ($$p \leq .40$$)0.0030.0430.002 ($$p \leq .94$$)Education level−0.0790.032−0.084 ($$p \leq .01$$)− 0.090.031−0.095** ($$p \leq .004$$)−0.0720.024−0.076** ($$p \leq .003$$)−0.0730.028−0.078** ($$p \leq .009$$)Antibiotics knowledge0.0440.0050.287**0.0110.0040.073* ($$p \leq .01$$)0.0160.0050.104** ($$p \leq .002$$)AMR knowledge−0.0330.015−0.077* ($$p \leq .03$$)−0.0180.012−0.041 ($$p \leq .15$$)−0.0640.014−0.148**Susceptibility0.0910.0250.097**Severity−0.0700.032−0.069** ($$p \leq .03$$)Self-Efficacy0.0410.0330.036 ($$p \leq .22$$)Response-Efficacy0.0950.0310.096** ($$p \leq .002$$)Response Cost−0.5440.024−0.61**Threat appraisal0.0550.0230.082* ($$p \leq .02$$)Coping appraisal0.2120.0160.455**R20.0660.1370.5080.316Adjusted R20.0630.1320.5020.310F for change in R219.82**34.62**126.00**109.22***Denotes correlation is $p \leq 0.05$ level while ** denotes $p \leq 0.01$ In the last regression (Model 4), the five PMT dimensions were replaced with their overarching constructs – coping appraisal and threat appraisal. The three sociodemographic variables, two knowledge variables, together with the coping appraisal, and threat appraisal were entered together in the regression. The final model is also significant, F [7, 838] = 55.18, $p \leq 0.01$, However, it yields a smaller R2 change =.18 (F [2, 838] = 109.22, $p \leq .001$) as compared to that of model 3 (R2 change =.37) with the five dimensions of PMT included. Therefore, model 3 is the model with better predictive power.
## Discussion
This study aimed to investigate the utility of AMR threat perceptions in predicting and explaining adherence to antibiotics among the public. By applying the PMT to quantitatively explore this relationship, this research provides empirical evidence on AMR threat perceptions as modifiable psycho-social determinants of patient adherence to antibiotics. Our findings show that patient adherence to antibiotics can be significantly explained by PMT, with the five constructs accounting for $37.1\%$ of the variance in this behavior. This amount of predictive variance exceeds those reported in past applications of PMT, such as in the study of breast cancer therapy adherence [24], preventative asthma treatment adherence [25], and eye patching adherence [26]. The PMT is henceforth a promising theoretical framework that can be used to understand why people fail to adhere to antibiotic recommendations and provide a blueprint for the development of future interventions.
Coping appraisal was found to exert a moderately-large effect on adherence to antibiotics with perceived response cost accounting for the largest effect among the five PMT constructs. This finding corroborates a meta-analysis of PMT studies which found response cost to be the most salient predictor of health behaviors [27]. The large negative effect size of perceived response cost suggests that one major determinant of non-adherence to antibiotics is the perception that people do not have the time and/or money to obtain antibiotics from healthcare providers and complete their prescribed antibiotic course. The smaller positive effect size of perceived response efficacy suggests that adherence is also, albeit to a notably smaller extent, driven by the belief that adhering to antibiotic recommendations can address the threat of AMR. Perceived self-efficacy, on the other hand, was not associated with adherence to antibiotics. This means that beliefs about whether an individual can enact behaviors to address AMR were not found to influence adherence to antibiotics. Collectively, these findings point to the utility of coping appraisal in predicting and explaining adherence to antibiotics with perceived response cost being the most promising modifiable psycho-social determinant.
We observed that the way individuals appraise the threat of AMR, as measured by perceived susceptibility and severity, has limited utility in explaining adherence to antibiotics, as very weak correlations were observed. This finding diverges from past literature which has found risk perceptions to be salient determinants of medication adherence for preventive or long-term treatment of chronic diseases [12, 13, 25]. One plausible reason for this contradictory finding is the psychological distance in which individuals view the risks of non-adherence to different medications. Psychological distance refers to how removed an event is perceived to be from direct experience [28]. Heightened psychological distance has been shown to reduce people’s intentions to adopt behaviors in a wide variety of contexts such as climate change [29] and the consumption of sugar-sweetened beverages [30]. While psychological distance has yet to be examined in the context of medication adherence, it is tenable that people are more likely to perceive the risks of antibiotics non-adherence, which are often less direct, immediate, and salient than those associated with other medication, as more psychologically distant. According to the construal level theory, this increased psychological distance can, in turn, reduce the effect of threat perceptions in motivating individuals to adhere to antibiotic recommendations [31]. Future studies could hence explore the saliency of psychological distance as a possible psycho-social barrier to adherence to antibiotics as well as examine the interactions between psychological distance and AMR threats perceptions on adherence. Future public health interventions can also explore ways to reduce people’s perceived psychological distance to AMR as a way to improve patient adherence to antibiotics.
On the other hand, while knowledge about antibiotics and knowledge regarding AMR were indeed found to be statistically significant determinants of adherence, they only accounted for $6.9\%$ of its variability. The magnitude of the effect of knowledge on adherence was even more reduced when the PMT constructs were considered. This finding offers some viable explanations for the mounting body of conflicting evidence between knowledge and adherence to antibiotics [32]. While some studies have found higher levels of antibiotic knowledge to be associated with lower levels of adherence, others have found support for the reverse, and some even found the lack of relationship between antibiotic knowledge and adherence [33].
## Study limitations
The results from our study should be considered in light of several potential limitations that future research could address. First, this was a single-country study conducted in Singapore, a developed nation in Southeast Asia, and our findings may be less generalizable to nations with different economic, cultural, and development profiles. Second, the self-report of adherence is prone to desirability and recall biases that can underestimate the prevalence of adherence. Besides, as the general public may not be able to fully distinguish antibiotics from other medications, it cannot be ruled out that our measure of adherence to antibiotics captures how well respondents adhere to medication in general instead of antibiotics specifically. However, as we have emphasized in the survey that this was a study on antibiotics, we believe that the two items we have chosen are with sufficient face validity. Since non-adherence to antibiotics includes various types of behaviors, future studies should include other dimensions such as not being compliant with the intake schedule. Third, participants may have used online search tools when completing the self-administered survey – a limitation that can overestimate the knowledge level regarding antibiotics and AMR.
## Insights for future antibiotic stewardship programs
Our findings may translate into practical implications for future antibiotic stewardship programs, suggesting other targets besides the lack of understanding of responsible antibiotics practices or adverse consequences of AMR. In fact, the large effect of coping appraisal and the comparatively weaker effect of threat appraisals suggest that persuasion strategies aimed at heightening concerns about AMR, such as the use of fear appeals, might not be as effective, and that may be better served with constructive communication strategies which include health education. In addition, future interventions should consider ways to assuage the public’s concerns about costs and time needed to adhere to antibiotic recommendations.
## Conclusion
As a pioneering study to examine how AMR threat perceptions relate to patient adherence to antibiotics, we found that while adherence was indeed driven by people’s perceptions of the severity of and their susceptibility to AMR, the effect of AMR threat perceptions was very weak. Adherence was instead found to be more influenced by how people perceived antibiotic-related costs. Knowledge about AMR and antibiotics was found to exert very little effect on adherence. Taken together, these findings demonstrate the complex and multifactorial nature of patient adherence to antibiotics and the relevancy of psychosocial constructs to the design of public health interventions.
## Supplementary Information
Additional file 1.
## References
1. 1.World Health Organization (WHO)Antimicrobial resistance2021GenevaWHO. *Antimicrobial resistance* (2021.0)
2. 2.Centre for Disease Control and PreventionInfographic: antibiotic resistance the global threat2019. *Infographic: antibiotic resistance the global threat* (2019.0)
3. Gualano MR, Gili R, Scaioli G, Bert F, Siliquini R. **General population’s knowledge and attitudes about antibiotics: a systematic review and meta-analysis**. *Pharmacoepidemiol Drug Saf* (2014.0) **24** 2-10. DOI: 10.1002/pds.3716
4. 4.WHOWorld antimicrobial awareness week2021GenevaWHO. *World antimicrobial awareness week* (2021.0)
5. Franklin BD, Abel G, Shojania KG. **Medication non-adherence: an overlooked target for quality improvement interventions**. *BMJ Qual Saf* (2019.0) **29** 271-273. DOI: 10.1136/bmjqs-2019-009984
6. Vrijens B, Urquhart J. **Patient adherence to prescribed antimicrobial drug dosing regimens**. *J Antimicrob Chemother* (2005.0) **55** 616-627. DOI: 10.1093/jac/dki066
7. 7.Zanichelli V, Tebano G, Gyssens IC, Vlahović-Palčevski V, Monnier AA, Stanic Benic M, et al. Patient-related determinants of antibiotic use: a systematic review. Clin Microbiol Infect. 2019;25(1). 10.1016/j.cmi.2018.04.031.
8. McQuaid EL, Landier W. **Cultural issues in medication adherence: disparities and directions**. *J Gen Intern Med* (2018.0) **33** 200-206. DOI: 10.1007/S11606-017-4199-3/TABLES/2
9. Axelsson M. **Report on personality and adherence to antibiotic therapy: a population-based study**. *BMC Psychol* (2013.0) **1** 1-6. DOI: 10.1186/2050-7283-1-24
10. 10.Haag M, Hersberger KE, Arnet I. Assessing medication adherence barriers to short-term oral antibiotic treatment in primary care—development and validation of a self-report questionnaire (Biotica). Int J Environ Res Public Health. 2021;18(15). 10.3390/ijerph18157768.
11. Gast A, Mathes T. **Medication adherence influencing factors - an (updated) overview of systematic reviews**. *Syst Rev* (2019.0) **8** 112, 1-17. DOI: 10.1186/s13643-019-1014-8
12. Shreck E, Gonzalez JS, Cohen HW, Walker EA. **Risk perception and self-management in urban, diverse adults with type 2 diabetes: the improving diabetes outcomes study**. *Int J Behav Med* (2014.0) **21** 88. DOI: 10.1007/S12529-013-9291-4
13. Claassen L, Henneman L, Kindt I, Marteau TM, Timmermans DRM. **Perceived risk and representations of cardiovascular disease and preventive behaviour in people diagnosed with familial hypercholesterolemia: a cross-sectional questionnaire study**. *J Health Psychol* (2010.0) **15** 33-43. DOI: 10.1177/1359105309345170
14. Stainback RD, Rogers RW. **Identifying effective components of alcohol abuse prevention programs: effects of fear appeals, message style and source expertise**. *Int J Addict* (1983.0) **18** 393-405. DOI: 10.3109/10826088309039356
15. 15.Singapore department of statistics. Population trends. 2021. https://www.singstat.gov.sg/-/media/files/publications/population/population2021.pdf.
16. Guo H, Hildon ZJ-L, Loh VW, Sundram M, Ibrahim MA, Tang WE. **Exploring antibiotic prescribing in public and private primary care settings in Singapore: a qualitative analysis informing theory and evidence-based planning for value-driven intervention design**. *BMC Fam Pract* (2021.0) **22** 205. DOI: 10.1186/s12875-021-01556-z
17. Lim JM, Duong MC, Cook AR, Hsu LY, Tam CC. **Public knowledge, attitudes and practices related to antibiotic use and resistance in Singapore: a cross-sectional population survey**. *BMJ Open* (2021.0) **11** e048157. DOI: 10.1136/BMJOPEN-2020-048157
18. Wang J, Sheng Y, Ni J, Zhu J, Zhou Z, Liu T. **Shanghai parents’ perception and attitude towards the use of antibiotics on children: a cross-sectional study**. *Infect Drug Resist* (2019.0) **12** 3259. DOI: 10.2147/IDR.S219287
19. Lim KK, Teh CC. **A cross sectional study of public knowledge and attitude towards antibiotics in Putrajaya, Malaysia**. *South Med Rev* (2012.0) **5** 26. PMID: 23532680
20. 20.World Health OrganizationAntibiotic resistance: multi-country public awareness survey2015. *Antibiotic resistance: multi-country public awareness survey* (2015.0)
21. Wu F, Yuan Y, Deng Z, Yin D, Shen Q, Zeng J. **Acceptance of Covid-19 booster vaccination based on the protection motivation theory: a cross-sectional study in China**. *J Med Virol* (2022.0) **94** 4115-4124. DOI: 10.1002/jmv.27825
22. Lwin MO, Stanaland AJ, Chan D. **Using protection motivation theory to predict condom usage and assess HIV health communication efficacy in Singapore**. *Health Commun* (2010.0) **25** 69-79. DOI: 10.1080/10410230903473540
23. 23.National population and talent division. Population in brief 2021. 2021. https://www.population.gov.sg/files/mediacentre/publications/population-in-brief-2021.pdf.
24. Karmakar M, Pinto SL, Jordan TR, Mohamed I, Holiday-Goodman M. **Predicting adherence to aromatase inhibitor therapy among breast cancer survivors: an application of the protection motivation theory**. *Breast Cancer* (2017.0) **11** 117822341769452. DOI: 10.1177/1178223417694520
25. Bennett P, Rowe A, Katz D. **Reported adherence with preventive asthma medication: a test of protection motivation theory**. *Psychol Health Med* (1998.0) **3** 347-354. DOI: 10.1080/13548509808400609
26. Norman P, Searle A, Harrad R, Vedhara K. **Predicting adherence to eye patching in children with amblyopia: an application of protection motivation theory**. *Br J Health Psychol* (2003.0) **8** 67-82. DOI: 10.1348/135910703762879219
27. Milne S, Sheeran P, Orbell S. **Prediction and intervention in health-related behavior: a meta-analytic review of protection motivation theory**. *J Appl Soc Psychol* (2000.0) **30** 106-143. DOI: 10.1111/j.1559-1816.2000.tb02308.x
28. Liberman N, Trope Y, McCrea SM, Sherman SJ. **The effect of level of construal on the temporal distance of activity enactment**. *J Exp Soc Psychol* (2007.0) **43** 143-149. DOI: 10.1016/j.jesp.2005.12.009
29. Jones C, Hine DW, Marks AD. **The future is now: reducing psychological distance to increase public engagement with climate change**. *Risk Anal* (2016.0) **37** 331-341. DOI: 10.1111/risa.12601
30. Ahn SJ. **Incorporating immersive virtual environments in health promotion campaigns: a construal level theory approach**. *Health Commun* (2014.0) **30** 545-556. DOI: 10.1080/10410236.2013.869650
31. Trope Y, Liberman N. **Construal-level theory of psychological distance**. *Psychol Rev* (2010.0) **117** 440-463. DOI: 10.1037/a0018963
32. Norris P, Chamberlain K, Dew K, Gabe J, Hodgetts D, Madden H. **Public beliefs about antibiotics, infection and resistance: a qualitative study**. *Antibiotics* (2013.0) **2** 465-476. DOI: 10.3390/antibiotics2040465
33. Byrne MK, Miellet S, McGlinn A, Fish J, Meedya S, Reynolds N. **The drivers of antibiotic use and misuse: the development and investigation of a theory driven community measure**. *BMC Public Health* (2019.0) **19** 1425. DOI: 10.1186/s12889-019-7796-8
|
---
title: 'Leptin and leptin receptor expression as biomarkers for breast cancer: a retrospective
study'
authors:
- Yan Wang
- Lili Du
- Jiexian Jing
- Xianwen Zhao
- Xing Wang
- Shenghuai Hou
journal: BMC Cancer
year: 2023
pmcid: PMC10029294
doi: 10.1186/s12885-023-10617-8
license: CC BY 4.0
---
# Leptin and leptin receptor expression as biomarkers for breast cancer: a retrospective study
## Abstract
### Background
Effective screening and treatment have reduced the number of women dying from breast cancer (BC). However, the long-term sequelae of BC treatment and psychosocial factors seriously affect the life quality of BC patients and survivors. Therefore, the discovery and application of targeted biomarkers to improve the functional outcome and life quality of BC patients is necessary.
### Aims
To explore the impact of leptin (LEP)/ leptin receptor (LEPR) expression on occurrence and survival of BC.
### Methods
Totally 132 primary BC and 66 non-BC patients who underwent surgery in department of breast surgery in Shanxi Cancer Hospital from January to October in 2009 were enrolled in this retrospective study. LEP and LEPR were examined in BC tissues, benign breast tissues, para-carcinoma tissues using immunohistochemical staining. Kaplan–Meier curve was generated to test survival time.
### Results
The high level expression of LEP and LEPR in BC tissues were significantly higher than that in benign breast tissues and in para-carcinoma tissues (all $P \leq 0.05$). The LEP expression in patients with lymph node metastases was significantly higher than that in patients without lymph nodes metastases ($$P \leq 0.002$$). LEPR expression was correlated with higher Ki-67 rate ($$P \leq 0.002$$). LEP and LEPR both had no impact on survival (all $P \leq 0.05$).
### Conclusions
High LEP/LEPR expression were risk factors for occurrence of BC, but without impact on survival.
## Introduction
Approximately more than 2 million women worldwide are diagnosed with breast cancer (BC) each year. The number of women dying from BC has fallen thanks to screening and effective treatment. However, radical mastectomy, chemoradiotherapy, and targeted therapy all have different degrees of harm to BC patients, leading to long-term sequelae of BC treatment. For BC patients and survivors, long-term sequelae as well as many psychosocial determinants, such as health care system factors, work constraints, spirit, and coping, seriously affect the life quality of BC patients and survivors [1, 2]. BC is arguably responsible for global burden and disability. Immunometabolism is found to be a new approach for tumor immunotherapy. Drugs to control dyslipidemia can well control lipid metabolism, thereby improving the tumor immune microenvironment of BC patients and inhibiting the recurrence and metastasis of BC [3]. It can be argued that health care today is not only about prolonging the life of patients, but also about maintaining and improving their health. Therefore, the discovery and application of targeted biomarkers to improve the functional outcome and quality of life of breast cancer patients is necessary. Obesity has been proved to be an independent risk factor for development of (BC, which was induced by adipose tissue secreted peptides [4–6]. As a neuroendocrine hormone exclusively generated by adipose tissue, Leptin (LEP) plays multiple biological functions, such as regulating insulin secretion and promoting cell proliferation and angiogenesis, through receptor-mediated pathways [4]. The relationship between LEP and cancer progression, has been widely explored [7–11].
It has been confirmed that LEP can significantly promote proliferation of BC cells compared with that of normal breast cells [12], and only BC cells will respond to LEP rather than normal breast cells [13]. Our recent study also found that BC patients had higher LEP levels than healthy controls [14, 15]. LEP plays biological functions by binding to leptin receptor (LEPR). LEPR is a single-pass transmembrane protein which plays a regulatory role in metabolism. Metabolic disorder is a direct cause for obesity and obesity-related diseases [16–18]. Previous studies showed that LEPR and LEP genes were two important genes related to obesity, and also closely related to BC [19, 20].
Present study aims to detect the expression of LEP and LEPR in BC patients, and further explore the impact of LEP/LEPR on occurrence and survival of BC.
## Specimen
Totally 132 primary BC and 66 non-BC patients who underwent surgery in department of breast surgery in Shanxi Cancer Hospital from January to October in 2009 were enrolled in this retrospective study. Patients meeting these criteria were included: (I) Female patients with breast cancer or benign breast disease; (II) Age between 30 to 70 years; (III) No Chemotherapy, radiotherapy and endocrinotherapy before surgery; (IV) pathologically confirmed after surgery. Patients with diabetes or malignancies other than breast cancer were excluded from analysis. The flow chart was shown in Fig. 1. All specimens collected during surgery were fixed with $10\%$ neutral formalin within 4 h after surgery and embedded in paraffin. Age, menopausal status, height, weight and circumference of waist and hip were collected and used to calculate body mass index (BMI) and ratio of waist to hip (WHR).Fig. 1The flowchart of patient enrollment. A total of 198 BC patients and 66 non-BC were primarily enrolled. Thirteen cases were excluded for metastatic BC, 19 cases were excluded for being performed chemotherapy, radiotherapy, or endocrinotherapy before surgery, and 5 cases were excluded for being complicated with diabetes. Totally 132 BC patients and 66 non-BC patients were finally enrolled All procedures performed in this study involving human participants were performed in accordance with the ethical standards of Shanxi Medical University Committee and with the 1964 Helsinki declaration and its later amendments (approve number: 2011134).
## Reagents and instruments
Rabbit LEP (A-20) and Rabbit LEPR (M-18) antibody to human were purchased from Santa Cruz, USA. Rapid immunohistochemistry EnVision™ Kit 5004, and diaminobenzidine (DAB) were purchased from Fuzhou Maixin Biotech, China.
## Immunohistochemistry
BC and breast fibroma tissues were prepared as tissue microarray samples using a tissue microarray preparation machine (Hengtai Co, LTD, Liaoning), and then sliced into 4 um serial sections. After incubation at 60°C for 4 h, the sections were immediately dewaxed, hydrated, and hot repaired under high pressure. After blocking endogenous peroxidase activity using H2O2, the sections were incubated with a drop of LEP/LEPR antibody at 4°C for overnight (LEP: 1:100, LEPR: 1:90), immersed with EnVision™ solution, stained with DAB and counter-stained with hematoxylin. After dehydration, transparent sections were mounted and examined with microscope. Human adipocyte tissues were used as positive controls, and samples without incubation of LEP or LEPR antibodies were used as negative controls.
## Result evaluation
LEP is expressed on cytosol and LEPR is expressed on both cytosol and membrane. Yellow or yellowish brown staining was considered as positive. The percentage of positive cells under optical microscope was calculated. Samples with less than $10\%$ positive cell were given 1 point, 2 points when 10–$50\%$ cells were positive, and 3 points when ≥ $50\%$ cells were positive. Signal intensity was scored as weak (1 point), moderate (2 points), or strong (3 points). Both scores were multiplied [21], and the resulting score was used to categorize LEP/LEPR expression as low (< 4) or high (≥ 4). Estrogen receptor (ER), progesterone receptor (PR) and Ki-67 are expressed in the nuclei of cancer cells. Yellowish brown particles were considered as positive.
## Statistical analyses
All the data collected in this study were analyzed using SPSS 22.0 software. Normally distributed measurement data were expressed as mean ± standard deviation (SD), and the comparisons were examined by Student-t test and Mann–Whitney test (non parametric distribution). Categorical data were expressed as n (%), and the differences between the two groups were examined by chi-square analysis or Fisher's exact test. The correlation between LEP/LEPR expression and clinical characteristics were analyzed by chi-square analysis. Survival time was calculated from surgery to death of patient or the last follow-up time. Kaplan–Meier methods was used for survival analysis. $P \leq 0.05$ was considered statistically significant.
## General information
Of these 132 patients, 10 patients were with well-differentiated, 60 patients were with moderately differentiated, and 62 patients were with poorly differentiated BC. Median age of these patients was 46 (30–70) years old. Sixty-six patients with benign breast diseases and 30 para-carcinoma tissues, which were normal breast tissues and 4 cm away from the tumor tissues from the 132 cancer patients, were included in controls. Among the patients with benign breast diseases, 37 had fibroadenoma, 15 had multiple intraductal papilloma and 14 had lobular proliferative diseases. They were aged between 34 and 68 years old with median age of 43 years old.
There was no difference of BMI (24.04 vs 23.28 kg/m2, $$P \leq 0.142$$) and WHR (0.86 vs 0.82, $$P \leq 0.241$$) between BC patients and benign ones. Baseline characteristics of patients with breast cancer were shown in Table 1.Table 1Baseline characteristics of patientsCharacteristicsNumber of patientsPatients with breast cancer132Age (years) ≤ 55110 > 5522BMI [kg/m2,(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{\upchi } \pm s$$\end{document}χ¯±s)]24.04 ± 3.47WHR (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{\upchi } \pm s$$\end{document}χ¯±s)0.86 ± 0.32Menstrual status Premenopausal98 Postmenopausal34Histological type Invasive ductal carcinoma58 In situ ductal carcinoma55 Invasive lobular carcinoma9 Lobular neoplasia10TNM stage 02 I37 II71 III20 IV2Status at last follow-up Alive`114 Dead7BMI body mass index, WHR waist-to-hip ratio
## LEP and LEPR expression
LEP and LEPR expression was evaluated in 132 tumors and 66 non-BC patients, which were classified as low expression and high expression. Both LEP and LEPR showed high expression in breast cancer tissues and low expression in benign breast tissues. There was no LEP and LEPR expressed in normal para-cancerous breast tissues (Fig. 2). The proportion of patients with LEP and LEPR high expression in different tissues was shown in Table 2.Fig. 2Expression of LEP and LEPR in breast cancer tissues, benign breast tissue and normal para-cancerous breast tissue; Note: High expression of LEP (A) and LEPR (D) in breast cancer tissues; Low expression of LEP (B) and LEPR (E) in benign breast tissues; No expression of LEP (C) and LEPR (F) in normal para-cancerous breast tissues. Original magnification × 400Table 2LEP and LEPR expressionVariablesTumorBenign lesionP-valuePara-carcinoma tissueP-valueTotal number1326630LEP High101370.001*140.003*LEPR High93370.005*170.044*p-value were all compared with tumor LEP leptin, LEPR leptin receptor The expression of LEP and LEPR were $86.66\%$ ($\frac{13}{15}$) and $73.33\%$ ($\frac{11}{15}$) of multiple intraductal papillomas disease, respectively, which resulting high rate of LEP and LEPR in all patients with benign disease. However, the expression decreased to $45.1\%$ and $51.\%$ in patients with benign disease regardless of multiple intraductal papilloma.
The expression of LEP and LEPR did not showed association with menopausal status, histological type, tumor size, tumor grade and metastasis status ($P \leq 0.05$), as well as the expression of ER, PR ($P \leq 0.05$). However, the positive rate of LEP expression in patients with lymph node metastasis was $91.8\%$, which was significantly higher than $67.5\%$ in patients without lymph node metastasis ($$P \leq 0.002$$). We also found that LEPR expression was correlated with Ki-67 expression ($$P \leq 0.002$$), as shown in Table 3.Table 3Correlation of clinicopathological features with the LEP and LEPR expressionClinicopathological featuresLEPR expression level ($$n = 132$$)LEP expression level ($$n = 132$$)High (%)Low (%)P-valueHigh (%)Low (%)P-valueMenstrual status0.8200.161 Premenopausal71 (72.4)27 (27.6)72 (73.5)26 (26.5) Postmenopausal24 (70.6)10 (29.4)29 (85.3)5 (14.7)Tumor size0.7090.832 ≤ 2 cm35 (66.0)18 (34.0)42 (79.2)11 (20.8) 2 ~ 5 cm52 (73.2)19 (26.8)53 (74.6)18 (25.4) > 5 cm6 (75.0)2 (25.0)6 [75]2 [25]Tumor grade0.8740.397 I7 (70.0)3 (30.0)6 (60.0)4 (40.0) II43 (72.9)16 (27.1)47 (79.7)12 (20.3) III43 (68.3)20 (31.7)48 (76.2)15 (24.2)TNM stage0.8490.428 01 (50.0)1 (50.0)1 (50.0)1 (50.0) I26 (68.4)12 (31.6)28 (73.7)10 (26.3) II49 (70.0)21 (30.0)52 (74.3)18 (25.7) III14 (73.7)5 (26.3)17 (89.5)2 (10.5) IV3 (100.0)0 [0]3 (100.0)0 [0]Histological type0.8310.508 IDC42 (72.4)16 (27.6)45 (77.6)13 (22.4) DCIS39 (70.9)16 (29.1)42 (76.4)13 (23.6) ILC6 (66.7)3 (33.3)8 (88.9)1 (11.1) LIN6 (60.0)4 (40.0)6 (60.0)4 (40.0)Lymph node metastasis0.6600.002 Yes36 (73.5)13 (26.5)45 (91.8)4 (8.2) No58 (69.9)25 (30.1)56 (67.5)27 (32.5)Distant metastases0.8840.684 Yes2 (66.7)1 (33.3)2 (66.7)1 (33.3) No91 (70.5)38 (29.5)99 (76.7)30 (23.3)Estrogen receptor0.3940.564 Positive71 (72.4)27 (27.6)73 (74.5)25 (25.5) Negative22 (64.7)12 (35.3)27 (79.4)7 (20.6)Progesterone receptor0.6480.298 Positive68 (69.4)30 (30.6)72 (73.5)26 (26.5) Negative25 (73.5)9 (26.5)28 (82.4)6 (17.6)Ki-670.0020.487 Positive75(68.8)34(31.2)70(64.2)39(35.8) Negative8(34.8)15(65.2)13(56.5)10(43.5)LEP leptin, LEPR leptin receptor, IDC Invasive Ductal Carcinoma, DCIS Ductal Carcinoma In Situ, ILC Invasive lobular Carcinoma, LIN Lobular Intraepithelial Neoplasia
## LEP and LEPR expression with prognosis
The follow-up time ranged from 22 to 89 months and all patients alive were followed-up for at least 48 months. Eleven patients were lost to follow-up. Two patients refused further treatment after surgery, five were with cancer progression during follow-up, and 7 patients died at last follow-up time. The 5-year overall survival rate of entire cohort was $94.9\%$.
LEP and LEPR expression did not significantly correlated with overall survival, nor with disease-free survival (Fig. 3).Fig. 3Survival curve of patients with different LEP and LEPR expression. A: Overall survival curve comparing patients with high and low expression of LEP; B: Overall survival curve comparing patients with high and low expression of LEPR; C: Disease-free survival curve comparing patients with high and low expression of LEP; D: Disease-free survival curve comparing patients with high and low expression of LEPR
## Discussions
Previous studies have indicated the association between BC and overweight, obesity, excessive nutrition, and metabolic disorders. Incidence of BC was increased with weight gain in middle-aged women [22]. Elevated BMI and WHR were also important risk factors for BC in women [23]. Our previous study also found that BC patients had significantly higher BMI and WHR than patients with benign breast diseases and healthy controls [24]. Kuriyama et al. reported that the increased LEP level and higher BMI would contribute to the increased incidence of BC [25]. Present study partly confirmed previous findings which indicated that LEP/LEPR expression was correlated with progression of BC.
LEPR is a multifunctional single-pass transmembrane protein and widely distributed in many organs [26]. It is the receptor of LEP, which can balance the energy consumption and glucose metabolism by activating JAK2-STAT3 and ERK pathways [27]. Disorder of the processes mentioned above would cause obesity and obesity-related diseases [28]. LEP and LEPR are widely expressed in various tissues including hypothalamus, adipose tissue, nerves, heart, kidney, breast, lung, liver and islet cell surface [29]. Ishikaw et al. detected the expression of LEP and LEPR in 76 cases of invasive ductal carcinoma and 32 cases of normal para-carcinoma breast tissues, and found that the expression of LEPR was significantly higher in tumor epithelium than normal breast epithelium. In addition, LEP was also over-expressed in tumor epithelium than in normal epithelium [30]. Garofalo et al. analyzed 148 BC tissue and benign breast lesions indicating that LEPR was expressed in $41.2\%$ of BC tissues but not in benign breast lesions [31]. Previous study analyzed the expression of LEP, long isoform of LEPR and short isoform of LEPR in 322 primary BC tissues and found that the long isoform of LEPR and the short isoform of LEPR were expressed in all tumor tissues and LEP was expressed in 318 samples [32]. In addition, the expression of these three proteins was positively associated with expression of estradiol and progesterone receptors, but not with tumor diameter and malignant degree. Present study showed that LEPR was expressed in $70.5\%$ of tumor tissues, which was significantly higher than $56.3\%$ of benign breast tissues and $44.0\%$ of normal para-carcinoma tissues. A previous study of our team showed that LEP was over-expressed in BC tissues and significantly associated with LEPR expression [33]. The co-expression of LEP and LEPR in primary BC showed that LEP expressed on mammary tumor cells via an autocrine pathway [18]. Highly expressed LEPR in BC make them more sensitive to LEP stimulation. LEP is associated with tumor cell migration and invasion, as well as angiogenesis in some tumors. It is also involved in several signaling pathways, as JAK/STAT, protein kinase B, phosphatidylinositol 3-kinase and mitogen activated protein kinase. These pathways, controlled by LEP-LEPR, are strongly related to cell survival and differentiation [34, 35].
LEP expression might be correlated with lymph nodes metastases of BC. Garofalo et al. found that LEPR expression rate was $51.5\%$ in BC tissues of patients with lymph node metastasis, which was significantly higher than $41.2\%$ in BC tissues of patients without lymph node metastasis [33]. In contrast, another study showed that low LEPR expression increased the risk of lymph node metastasis by fourfold [36]. Present study found that LEPR expression did not significantly correlated with lymph node metastases ($$p \leq 0.66$$), but the expression rate of LEP in patients with lymph node metastasis was significantly higher than that of patients without lymph node metastasis ($$p \leq 0.002$$). It indicated that LEP is involved in the proliferation process of BC and plays a very important role in the development of BC. Moreover, this study also found that the expression of LEP and LEPR were balanced when stratified by patients’ age, menopausal status, tumor size, tumor pathological classification, distant metastasis and the expression of ER and PR, which was consistent with previous study [30, 37].
Another finding of present study was that LEPR expression was correlated with Ki-67 ($$P \leq 0.002$$). Ki67 was related to cell proliferation cycle and its expression has been correlated with the development of a variety of malignant tumors. The results suggested that LEPR could promote BC cell proliferation after activation.
Previous study has shown an association between high LEP expression levels and poor prognosis in several cancers [38]. Other reported low LEP expression and high expression of Ob-R mRNA in breast cancer tissue correlated with shorter OS and RFS [39]. LEP and LEPR expression did not correlated with overall survival, nor disease-free survival in our analysis. Similar results have been reported in oral and oropharyngeal cancer [36]. However, another study found that the expression of LEP was significantly associated with overall survival. Further more, for the postmenopausal patients or triple negative patients and lymph node metastasis patients, LEP-positive group had worse prognosis [40].
There were also several limitations of present study. First, there was unavoidable biases due to the retrospective nature of present study. Second, this was a study with small sample size which resulting in limited reliability. Thus, all conclusions should be interpreted cautiously.
## Conclusion
In conclusion, the positive rate of LEP and LEPR expression in BC tissues was significantly higher than that in benign breast tissues and normal para-carcinoma tissues. The LEP and LEPR expression were significantly correlated with lymph node metastasis and Ki-67 expression, respectively. High LEP/LEPR expression were risk factors for occurrence of BC, but without impact on survival.
## References
1. Culbertson MG. **The psychosocial determinants of quality of life in breast cancer survivors: a scoping review**. *BMC Cancer* (2020) **20** 948. DOI: 10.1186/s12885-020-07389-w
2. Invernizzi M. **Editorial: quality of life in breast cancer patients and survivors**. *Front Oncol* (2020) **10** 620574. DOI: 10.3389/fonc.2020.620574
3. Goto W. **Relationship between serum lipid levels and the immune microenvironment in breast cancer patients: a retrospective study**. *BMC Cancer* (2022) **22** 167. DOI: 10.1186/s12885-022-09234-8
4. Lorincz AM, Sukumar S. **Molecular links between obesity and breast cancer**. *Endocr Relat Cancer* (2006) **13** 279-292. DOI: 10.1677/erc.1.00729
5. Soheila Sayad, Dastgheib SeyedAlireza, Meraj Farbod, Fatemeh Asadian, Karimi-Zarchi Mojgan, Seyedali S. **Association of PON1, LEP and LEPR polymorphisms with susceptibility to breast cancer: a meta-analysis**. *APJCP* (2021) **22** 2323-34. DOI: 10.31557/APJCP.2021.22.8.2323
6. Crovesy Louise, Rosado ElianeL. **Interaction between genes involved in energy intake rgulation and diet in obesity**. *Nutrition.* (2019) **110547** 67-68. DOI: 10.1016/j.nut.2019.06.027
7. Chen DC, Chung YF, Yeh YT, Chaung HC, Kuo FC, Fu OY. **Serum adiponectin and leptin levels in Taiwanese breast cancer patients**. *Cancer Lett* (2006) **237** 109-114. DOI: 10.1016/j.canlet.2005.05.047
8. Ribeiro R, Lopes C, Medeiros R. **The link between obesity and prostate cancer: the leptin pathway and therapeutic perspectives**. *Prostate Cancer Prostatic Dis* (2006) **9** 19-24. DOI: 10.1038/sj.pcan.4500844
9. Tutino V, Notarnicola M, Guerra V, Lorusso D, Caruso MG. **Increased soluble leptin receptor levels are associated with advanced tumor stage in colorectal cancer patients**. *Anticancer Res* (2011) **31** 3381-3383. PMID: 21965750
10. White PB, True EM, Ziegler KM, Wang SS, Swartz-Basile DA, Pitt HA. **Insulin, leptin, and tumoral adipocytes promote murine pancreatic cancer growth**. *J Gastrointest Surg* (2010) **14** 1888-93. DOI: 10.1007/s11605-010-1349-x
11. Wu MH, Chou YC, Chou WY, Hsu GC, Chu CH, Yu CP. **Circulating levels of leptin, adiposity and breast cancer risk**. *Br J Cancer* (2009) **100** 578-582. DOI: 10.1038/sj.bjc.6604913
12. Hu X, Juneja SC, Maihle NJ, Cleary MP. **Leptin–a growth factor in normal and malignant breast cells and for normal mammary gland development**. *J Natl Cancer Inst* (2002) **94** 1704-1711. DOI: 10.1093/jnci/94.22.1704
13. Cleary MP, Juneja SC, Phillips FC, Hu X, Grande JP, Maihle NJ. **Leptin receptor-deficient MMTV-TGF-alpha/Lepr(db)Lepr(db) female mice do not develop oncogene-induced mammary tumors**. *Exp Biol Med (Maywood)* (2004) **229** 182-193. DOI: 10.1177/153537020422900207
14. Han C, Zhang HT, Du L, Liu X, Jing J, Zhao X. **Serum levels of leptin, insulin, and lipids in relation to breast cancer in china**. *Endocrine* (2005) **26** 19-24. DOI: 10.1385/ENDO:26:1:019
15. Du LL, Han CZ, Liu XY, Jing JX, Zhao XW, Tian BG. **Relationships of serum leptin level and body mass index to breast cancer**. *Cancer Research and Clinic* (2006) **18** 307-309
16. Chen H, Charlat O, Tartaglia LA, Woolf EA, Weng X, Ellis SJ. **Evidence that the diabetes gene encodes the leptin receptor: identification of a mutation in the leptin receptor gene in db/db mice**. *Cell* (1996) **84** 491-495. DOI: 10.1016/s0092-8674(00)81294-5
17. Cleveland RJ, Gammon MD, Long CM, Gaudet MM, Eng SM, Teitelbaum SL. **Common genetic variations in the LEP and LEPR genes, obesity and breast cancer incidence and survival**. *Breast Cancer Res Treat* (2010) **120** 745-752. DOI: 10.1007/s10549-009-0503-1
18. Jarde T, Caldefie-Chezet F, Damez M, Mishellany F, Penault-Llorca F, Guillot J. **Leptin and leptin receptor involvement in cancer development: a study on human primary breast carcinoma**. *Oncol Rep* (2008) **19** 905-911. PMID: 18357374
19. Okobia MN, Bunker CH, Garte SJ, Zmuda JM, Ezeome ER, Anyanwu SN. **Leptin receptor Gln223Arg polymorphism and breast cancer risk in Nigerian women: a case control study**. *BMC Cancer* (2008) **8** 338. DOI: 10.1186/1471-2407-8-338
20. Nyante SJ, Gammon MD, Kaufman JS, Bensen JT, Lin DY, Barnholtz-Sloan JS. **Common genetic variation in adiponectin, leptin, and leptin receptor and association with breast cancer subtypes**. *Breast Cancer Res Treat* (2011) **129** 593-606. DOI: 10.1007/s10549-011-1517-z
21. Campos AH, Aldred VL, Ribeiro KC, Vassallo J, Soares FA. **Role of immunoexpression of nitric oxide synthases by Hodgkin and Reed-Sternberg cells on apoptosis deregulation and on clinical outcome of classical Hodgkin lymphoma**. *Mol Cell Biochem* (2009) **321** 95-102. DOI: 10.1007/s11010-008-9923-1
22. O'Brien SN, Welter BH, Price TM. **Presence of leptin in breast cell lines and breast tumors**. *Biochem Biophys Res Commun* (1999) **259** 695-698. DOI: 10.1006/bbrc.1999.0843
23. Borugian MJ, Sheps SB, Kim-Sing C, Olivotto IA, Van Patten C, Dunn BP. **Waist-to-hip ratio and breast cancer mortality**. *Am J Epidemiol* (2003) **158** 963-968. DOI: 10.1093/aje/kwg236
24. Han CZ, Shi J, Du LL, Jin JX, Zhao XW, Tian BG. **Association among lipids, leptin and leptin receptor polymorphism with risk of breast cancer**. *Chin J Epidemiol* (2007) **28** 136-140
25. Gui Y, Pan QW, Chen XC, Xu SM, Luo XD, Chen L. **The association between obesity related adipokines and risk of breast cancer: a meta-analysis**. *Oncotarget* (2017) **8** 75389-75399. DOI: 10.18632/oncotarget.17853
26. Margetic S, Gazzola C, Pegg GG. **Leptin: a review of its peripheral actions and interactions**. *Int J Obes Relat Metab Disold* (2002) **26** 1407-33. DOI: 10.1038/sj.ijo.0802142
27. Morton GJ, Schwartz MW. **Leptin and the central nervous system control of glucose metabolism**. *Physiol Rev* (2011) **91** 389-411. DOI: 10.1152/physrev.00007.2010
28. Enns JE, Taylor CG, Zahradka P. **Variations in adipokine genes AdipoQ, Lep, and Lepr are associated with risk for obesity-related metabolic disease: the modulatory role of gene-nutrient interactions**. *J Obes* (2011) **2011** 168659. DOI: 10.1155/2011/168659
29. Margetic S, Gazzola C, Pegg GG, Hill RA. **Leptin: a review of its peripheral actions and interactions**. *Int J Obes Relat Metab Disord* (2002) **26** 1407-1433. DOI: 10.1038/sj.ijo.0802142
30. Khabaz MN, Abdelrahman A, Butt N, Damnhory L, Elshal M, Aldahlawi AM. **Immunohistochemical staining of leptin is associated with grade, stage, lymph node involvement, recurrence, and hormone receptor phenotypes in breast cancer**. *BMC Womens Health* (2017) **17** 105. DOI: 10.1186/s12905-017-0459-y
31. Garofalo C, Koda M, Cascio S, Sulkowska M, Kanczuga-Koda L, Golaszewska J. **Increased expression of leptin and the leptin receptor as a marker of breast cancer progression: possible role of obesity-related stimuli**. *Clin Cancer Res* (2006) **12** 1447-1453. DOI: 10.1158/1078-0432.CCR-05-1913
32. Revillion F, Charlier M, Lhotellier V, Hornez L, Giard S, Baranzelli MC. **Messenger RNA expression of leptin and leptin receptors and their prognostic value in 322 human primary breast cancers**. *Clin Cancer Res* (2006) **12** 2088-2094. DOI: 10.1158/1078-0432.CCR-05-1904
33. Wang Y, Han C, Jing J. **Association between expressions of leptin and leptin receptor genes in breast cancer**. *Cancer Res Clin* (2012) **24** 16-19
34. Rodrigues PRS, Maia LL, Santos M. **Leptin receptor expression and Gln223Arg polymorphism as prognostic markers in oral and oropharyngeal cancer**. *Genet Mol Res* (2015) **14** 14979-14988. DOI: 10.4238/2015.November.24.5
35. Bracho-Riquelme RL, Loera-Castañeda V, Torres-Valenzuela A, Loera-Castañeda GA, Sánchez-Ramírez JP. **Leptin and leptin receptor polymorphisms are associated with poor outcome (death) in patients with non-appendicular secondary peritonitis**. *Crit Care* (2011) **15** R227. DOI: 10.1186/cc10467
36. Rodrigues PR, Maia LL, Santos M, Peterle GT, Alves LU, Takamori JT. **Leptin receptor expression and Gln223Arg polymorphism as prognostic markers in oral and oropharyngeal cancer**. *Genet Mol Res* (2015) **14** 14979-14988. DOI: 10.4238/2015.November.24.5
37. Kim HS. **Leptin and leptin receptor expression in breast cancer**. *Cancer Res Treatment* (2009) **41** 155-163. DOI: 10.4143/crt.2009.41.3.155
38. Lipsey CC, Harbuzariu A, Daley-Brown D, Gonzalez-Perez RR. **Oncogenic role of leptin and Notch interleukin-1 leptin crosstalk outcome in cancer**. *World J Methodol* (2016) **6** 43-55. DOI: 10.5662/wjm.v6.i1.43
39. Jin TY, Saindane M, Park KS, Kim SH, Nam SE, Yoo YB. **LEP as a potential biomarker in prognosis of breast cancer: systemic review and meta analyses (PRISMA)**. *Medicine* (2021) **100** e26896. DOI: 10.1097/MD.0000000000026896
40. Chen W, Zha XM, Liu XA, Zhang WM, Yin YM. **Clinical significance and prognosis of expressions of leptin/ leptin receptor in breast cancer**. *Chin Clin Oncol* (2010) **06** 493-497
|
---
title: Isoginkgetin-loaded reactive oxygen species scavenging nanoparticles ameliorate
intervertebral disc degeneration via enhancing autophagy in nucleus pulposus cells
authors:
- Hao Yu
- Yun Teng
- Jun Ge
- Ming Yang
- Haifeng Xie
- Tianyi Wu
- Qi Yan
- Mengting Jia
- Qing Zhu
- Yanping Shen
- Lianxue Zhang
- Jun Zou
journal: Journal of Nanobiotechnology
year: 2023
pmcid: PMC10029295
doi: 10.1186/s12951-023-01856-9
license: CC BY 4.0
---
# Isoginkgetin-loaded reactive oxygen species scavenging nanoparticles ameliorate intervertebral disc degeneration via enhancing autophagy in nucleus pulposus cells
## Abstract
Excessive reactive oxygen species (ROS) in nucleus pulposus cells (NPCs) promote extracellular matrix (ECM) degradation and cellular inflammatory responses by activating a variety of cellular pathways, ultimately inducing cell apoptosis and leading to the development of low back pain. Here, we designed and fabricated an isoginkgetin-loaded ROS-responsive delivery system (IGK@SeNP) based on diselenide block copolymers. Successfully encapsulated IGK was released intelligently and rapidly in a microenvironment with high ROS levels in degenerative disc. Controlled-release IGK not only efficiently scavenged ROS from the intervertebral disc together with diselenide block copolymers but also effectively enhanced autophagy in NPCs to inhibit ECM degradation and cell apoptosis, and showed significant therapeutic effects in the rat intervertebral disc degeneration (IDD) model. Overall, the synergistic effects of IGK@SeNP in ROS scavenging and autophagy enhancement endowed it with an attractive therapeutic strategy for IDD treatment.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12951-023-01856-9.
## Introduction
Low back pain caused by intervertebral disc degeneration (IDD) has become an important public health concern [1, 2]. In degenerated discs, excessive ROS promote organelle damage, extracellular matrix (ECM) degradation, and cellular inflammatory responses by activating various pathways, ultimately inducing functional cell death in the disc, causing spinal motor unit instability and radicular pain [3–7]. As there are no blood vessels in the intervertebral disc, the concentration of drugs in the intervertebral disc is low, and the efficacy is poor when administered systemically [8, 9]. Currently, the commonly used conservative treatment for IDD is pharmacological analgesia and dehydration [10]. However, these methods can only temporarily relieve pain and eliminate nerve swelling, but cannot remove excessive ROS to delay the progress of IDD, let alone achieve the purpose of curing IDD.
The intervertebral disc consists of a gelatinous inner nucleus pulposus and tough outer annulus fibrosus. The nucleus pulposus is the site of initial degeneration caused by oxidative stress [11]. Early recognition and removal of excessive ROS is a viable therapeutic strategy [12, 13]. For instance, Yang et al. [ 14] designed an injectable composite hydrogel, which sustained the release of Prussian blue nanoparticles to clear excessive ROS and showed good therapeutic effects in a rat IDD model. Although this strategy can effectively remove excessive ROS, nucleus pulposus cells (NPCs) cannot completely repair oxidative damaged biomolecules and organelles, which seriously affects the metabolic balance and cell viability, and may eventually lead to apoptosis [15, 16]. Therefore, it is necessary to develop a new therapeutic strategy that can not only remove too much ROS in NPCs but also promote the rapid repair of oxidative damage.
When NPCs are subjected to oxidative stress, cells can repair themselves by regulating the level of autophagy to breakdown damaged cellular components [17]. However, when ROS are cleared, cellular autophagy is downregulated, and cells cannot effectively clear damaged organelles, which may cause the cells to undergo apoptosis. Several studies have shown that upregulation of autophagy in NPCs can promote the repair of oxidative damage and reduce apoptosis [18–20]. Isoginkgetin (IGK), a natural flavonoid extracted from ginkgo biloba leaves, can enhance autophagy and scavenge ROS [21, 22]. We believe that it may be a potential drug for the treatment of IDD.
In this study, we designed and prepared an ROS-responsive IGK-loaded nanodelivery system (IGK@SeNP) that contained Se-Se bonds and showed good biocompatibility. The encapsulated IGK was intelligently released under the stimulation of high levels of ROS in the degenerative intervertebral disc. Under the combined treatment of IGK and Se-Se bonds, intracellular ROS was effectively cleared. Moreover, controlled-release IGK increased the autophagy level in NPCs, which balanced the anabolism and catabolism of ECM, inhibited apoptosis, and showed a significant therapeutic effect on IDD in vivo (Scheme 1). The strategy of removing intracellular ROS and enhancing autophagy to repair oxidative damage provides a safe and effective novel treatment for IDD.Scheme 1Diagram of IGK@SeNP for the treatment of IDD. A ginkgo biloba extract, IGK, is loaded in Se-Se bond-containing nanoparticles (SeNP). IGK@SeNP is easily taken up by NPCs, and then SeNP scavenges ROS and releases IGK to enhance autophagy, which synergistically protects the ECM of NPCs and reduces cell apoptosis under oxidative stress
## Construction and characterization of IGK@SeNP
The diselenide-containing polymer (Se polymer) was synthesized according to our previous study (Fig. 1A). Briefly, di(1-hydroxylundecyl) diselenide was first synthesized, polymerized with isophorone diisocyanate (IPDI), and finally reacted with methoxypolyethylene glycol (mPEG). The 1H NMR spectrum demonstrated the successful synthesis of di(1-hydroxylundecyl) diselenide (Additional file 1: Fig. S1) and Se-polymer (Fig. 1B). GPC analysis determined the molecular weight of the Se-polymer to be 11950 g/mol (Additional file 1: Fig. S2).Fig. 1Construction and characterization of SeNP. A Synthetic route of diselenide-containing polymer. B 1H NMR characterization of diselenide-containing polymer. C TEM image and size distribution of SeNP. Scale bar = 200 nm. D Cumulative drug release curves of Cy5@SeNP at different concentrations of H2O2. $$n = 3$$, mean ± SD Owing to its amphiphilicity, the Se-polymer can self-assemble into nanoparticles (SeNP) in aqueous solution. The amphiphilic Se-polymer and hydrophobic IGK self-assembled into micelle nanoparticles in aqueous solution (hereafter referred to as IGK@SeNP). Transmission electron microscopy (TEM) revealed that the IGK@SeNP possessed a spherical morphology with an average diameter of approximately 100 nm (Fig. 1C). Dynamic light scattering (DLS) indicated that the average hydrodynamic diameter was approximately 108 nm, with a polydispersity index of 0.25.
## In vitro drug release
The in vitro ROS-responsive drug release behavior of the nanoparticles was investigated. Insoluble Cy5 was loaded as the model cargo given its easy detectability (named Cy5@SeNP hereafter). Different concentrations of hydrogen peroxide (H2O2) divided this experiment into four groups: (a) PBS control group, (b) 50 μM H2O2 co-treatment group, (c) 100 μM H2O2 co-treatment group, and (d) 200 μM H2O2 co-treatment group.
As shown in Fig. 1D, in the control group, Cy5@SeNP achieved a drug release rate of only ~ $9\%$ at 36 h, which indicated that Cy5@SeNP had favorable stability under physiological conditions. In contrast, in the three H2O2 treatment groups, the release rate significantly increased with increasing concentrations of H2O2. For example, under the stimulation of 200 μM H2O2, the drug was released rapidly, with a cumulative release percentage of ~ $63\%$ at 36 h. The above results suggest that SeNP exhibited an excellent property of controlled release in the presence of ROS. Therefore, we expect SeNP to release drugs to treat IDD under oxidative conditions.
## Biocompatibility and ROS-scavenging properties in vitro
To determine the biocompatibility of the nanoparticles, we first treated NPCs with various concentrations of SeNP. The CCK-8 assay results showed that SeNP at 0–60 μg/ml had no significant toxicity on NPCs within 72 h (Fig. 2A). Therefore, we chose the concentration of SeNP to be 60 μg/ml in the subsequent study. IGK@SeNP at the above concentrations were co-incubated with NPCs for 72 h to detect their biocompatibility. Live/dead staining indicated that NPCs maintained good growth ability after co-culture with IGK@SeNP for 72 h, and the percentage of viable cells was not statistically different from that in the 24 h group (Fig. 2B, D). These results showed that the IGK@SeNP had excellent compatibility with NPCs. Fig. 2Biocompatibility and ROS scavenging properties in vitro. A CCK-8 assay was used to evaluate the cell cytotoxicity of SeNP and IGK@SeNP against NPCs after incubation for 24 h, 48 h, and 72 h. $$n = 6$$, mean ± SD. B, D Live/dead staining and cell viability of NPCs co-cultured with IGK@SeNP (60 μg/ml) for 24 h, 48 h, and 72 h. Scale bar = 200 μm. $$n = 3$$, mean ± SD. ns, not significant. C Fluorograph of the cellular uptake of Cy5@SeNP by NPCs. Scale bar = 100 μm. E, F Flow cytometry analysis and relative geometric mean fluorescence intensity of intracellular ROS levels in NPCs treated with H2O2 (200 μM) and different treatments. $$n = 3$$, mean ± SD. * $p \leq 0.05$ compared to the H2O2 group Cy5 was used as a drug model to assess cellular internalization of the nanocarriers. As shown in Fig. 2C, when NPCs were co-cultured with Cy5@SeNP (red) for 2 h, red fluorescence was observed in cells, indicating that Cy5@SeNP was internalized into NPCs. After 6 h of culture, the red fluorescence intensity was significantly enhanced, indicating that more Cy5@SeNP were internalized by the NPCs. These results indicate that SeNP can be effectively internalized by NPCs.
To verify the ROS scavenging ability of IGK, SeNP, and IGK@SeNP, we established a cell model of oxidative stress in NPCs by co-incubating the NPCs with 200 μM H2O2 for 24 h and calculated the intracellular ROS content using geometric mean fluorescence intensity (gMFI). We found that intracellular ROS levels in the H2O2 group (gMFI = 1891 ± 53) increased 21-fold compared with the control group (gMFI = 77 ± 3). Subsequently, different NPC treatments were administered. As shown in Fig. 2E, F, IGK decreased intracellular ROS levels (gMFI = 683 ± 66), but its effect was weaker than that of SeNP (gMFI = 506 ± 44). In contrast, IGK@SeNP exhibited the strongest antioxidant capacity (158 ± 52). These results indicate that IGK@SeNP has an excellent synergistic effect in eliminating excessive ROS in NPC.
## Anti‑ECM degradation effects of IGK@SeNP on NPCs
Western blotting was performed to evaluate the protective effect of IGK@SeNP on ECM (Fig. 3A, B). H2O2 treatment decreased the expression of the ECM anabolic proteins COL2, SOX9, and ACAN and increased the expression of the ECM catabolic enzymes MMP3, MMP13, and ADAMTS5. These changes indicate that oxidative stress tilts the metabolic balance of the ECM toward catabolism. Supplementation with IGK, SeNP, or IGK@SeNP reversed the disordered ECM metabolism induced by H2O2. However, the correction of protein expression in the IGK group was slightly weaker than that in the SeNP group, which may be due to the poor water solubility of IGK and its difficulty in entering cells to exert biological effects. Notably, in the IGK@SeNP group, the expression of ECM metabolism-related proteins was significantly restored, with a better effect than that in the IGK and SeNP groups. Furthermore, immunofluorescence (IF) staining was conducted for the representative component (ACAN) and stromelysin (MMP3) in the nucleus pulposus (Fig. 3C–F). Consistent with previous results, the IGK@SeNP group most significantly increased the fluorescence intensity of ACAN and decreased the fluorescence intensity of MMP3. Collectively, the above data showed that H2O2-induced oxidative stress can disrupt the metabolic balance of ECM, while the combination of SeNP and IGK can better regulate the anabolism and catabolism of ECM than individual treatments. Fig. 3IGK@SeNP attenuates H2O2-induced ECM degradation. A Western blotting analysis of the expression of ECM anabolic proteins (COL2, SOX9, and ACAN) and ECM catabolic enzymes (MMP3, MMP13, and ADAMTS5) in 200 μM H2O2-induced NPCs treated with IGK (20 μM), SeNP (60 μg/ml), and IGK@SeNP (60 μg/ml). B *Densitometric analysis* of ECM metabolism-related protein levels. C–F IF staining of ACAN and MMP3 in NPCs. Scale bar = 100 μm. $$n = 3$$, mean ± SD. * $p \leq 0.05$ compared to the H2O2 group, †$p \leq 0.05$ compared to the IGK group
## Anti‑apoptosis effects of IGK@SeNP on NPCs
To determine the protective effect of the nanodrug delivery system on NPCs under oxidative stress, western blotting was used to detect changes in the expression of key proteins in the apoptotic pathway (Fig. 4A, B). After treatment with H2O2 for 24 h, the expression of BCL2 was downregulated, while BAX and cleaved caspase3 (CASP3) were upregulated, indicating that apoptosis was activated by oxidative stress. However, after treatment with IGK, SeNP, and especially IGK@SeNP, the changes in the expression of the above proteins became weaker, indicating that the apoptosis pathway was blocked. Fig. 4IGK@SeNP inhibits H2O2-induced NPC apoptosis. A Western blotting analysis of the expression of an antiapoptotic protein (BCL2) and pro-apoptosis proteins (BAX and cleaved CASP3). B *Densitometric analysis* of apoptosis-related protein levels. C, D IF staining of cleaved CASP3 in NPCs. Scale bar = 100 μm. E, F TUNEL staining of NPCs after different treatments. Scale bar = 100 μm. G, H Flow cytometry analysis with FITC-annexin V/PI assay. * $p \leq 0.05$ compared to the H2O2 group, †$p \leq 0.05$ compared to the IGK group Accumulation of ROS has been demonstrated to increase the activity of CASP3, a critical protein in the apoptosis pathway, leading to amplification of the proteolytic cleavage cascade, ultimately causing cell apoptosis [23]. IF staining showed that the high level of cleaved CASP3 induced by H2O2 decreased in the IGK, SeNP, and IGK@SeNP groups, in which IGK@SeNP achieved the optimal effect (Fig. 4C, D). TUNEL staining showed increased DNA fragmentation in the H2O2 group, indicating that the NPCs underwent apoptosis under H2O2 stimulation (Fig. 4E, F). Fortunately, the number of TUNEL-positive cells decreased in the three experimental groups, and again, the IGK@SeNP group showed the strongest cytoprotection. These results were also confirmed with flow cytometry, showing that the apoptosis rate was 4.93 ± $0.28\%$ in the control groups and 17.49 ± $0.75\%$ in the H2O2 groups, which was reduced to 11.18 ± $0.14\%$, 9.71 ± $0.44\%$, and 7.02 ± $0.98\%$ in the IGK, SeNP, and IGK@SeNP groups, respectively (Fig. 4G, H). The above results indicated that IGK@SeNP could effectively inhibit NPC apoptosis induced by H2O2.
## Pro-autophagy effects of IGK@SeNP on NPCs
Moderate autophagy helps cells survive in harsh environments, whereas excessive autophagy also induces cell death [24–26]. Therefore, we explored the role of autophagy in the process of IGK treatment in IDD. As autophagy is regulated by multiple proteins at different stages, several key indicators are used to reflect changes in autophagic flux. For example, Beclin1 (BECN1) binds to pre-autophagosomes to initiate autophagosome formation. ATG7 participates in the assembly of autophagosomes by promoting the conjugation of ATG12 to ATG5 and the conversion of LC3I to LC3II. Then, LC3II binds to autophagosome membranes and is widely used as a marker for autophagy. Eventually, SQSTM1/p62 delivers autophagic substrates to the autophagosome and is degraded with these substrates in autolysosomes. As shown in Fig. 5A, B, H2O2 activated autophagy, marked by the upregulated expression of BECN1 and ATG7 and the ratio of LC3B II/I, while downregulating the expression of SQSTM1/p62. Treatment with IGK further amplified these changes in the protein levels. IF staining of LC3B also confirmed that there were more punctate structures in the IGK and IGK@SeNP groups, suggesting that the number of autophagosomes combined with LC3B II might be increased (Fig. 5C, D).Fig. 5IGK@SeNP promotes autophagy in H2O2-treated NPCs. A Western blotting analysis of the expression of autophagy-related proteins (BECN1, ATG7, LC3B, and SQSTM1/p62). B *Densitometric analysis* of autophagy-related protein levels. C, D IF staining of LC3B in NPCs incubated with IGK, SeNP, and IGK@SeNP in the presence of H2O2 stimulation. Scale bar = 20 μm. E, F TEM analysis of morphology and the number of autophagosomes and autolysosomes in NPCs. Red arrows indicate autolysosomes with single membranes, while black arrows indicate autophagosomes with double membranes. M indicates mitochondria. Scale bar = 1 μm (upper row) and 500 nm (lower row). $$n = 3$$, mean ± SD. * $p \leq 0.05$ compared to the H2O2 group, †$p \leq 0.05$ compared to the IGK group *Autophagy is* a complex process because an increase in LC3II levels may be caused by either enhanced autophagic flux or insufficient autolysosomes [27]. Therefore, we detected autophagy-associated vesicles by TEM. As shown in Fig. 5E, in the H2O2 group, autolysosomes wrapped the damaged organelles, and mitochondria dissolved into vacuoles, suggesting that oxidative stress damaged mitochondria. However, the dysfunctional mitochondria were not completely cleared, which may have led to their continuous production of ROS and eventually induced apoptosis of NPCs. In addition, a higher number of autophagy-associated vesicles was observed after IGK treatment, especially in the IGK@SeNP group, indicating that the SeNP drug delivery system favored the autophagy-enhancing properties of IGK (Fig. 5F).
We also noticed that the SeNP group showed a slight inhibition of autophagy compared with the H2O2 group, but interestingly, the IGK-loaded SeNP still showed a more obvious autophagy enhancement effect. These results indicate that the autophagy stimulator IGK was steadily transferred and released from the ROS-responsive SeNP and more effectively enhanced autophagy, thus regulating ECM metabolism and restraining apoptosis of NPCs.
## Effects of autophagy blockers on the therapeutic efficacy of IGK@SeNP
To confirm that IGK protects NPCs by enhancing autophagy, we treated NPCs with 10 μM chloroquine (CQ), an autophagic blocker that inhibits the fusion of autophagosomes with lysosomes. The protein levels of LC3B and SQSTM1/p62 increased after treatment with CQ, indicating an interrupted autophagic flux (Fig. 6A, B). Moreover, in the presence of CQ, the effect of IGK@SeNP in regulating the expression of ECM metabolism-related proteins (COL2, SOX9, ACAN, MMP3, MMP13, and ADAMTS5) and apoptosis-related proteins (BCL2, BAX, and cleaved CASP3) decreased (Fig. 6C–F). These results further verified that IGK@SeNP protects NPCs by enhancing autophagy under oxidative stress. Fig. 6Autophagy blocker reduces the anti-ECM degradation effect and anti-apoptotic effect of IGK@SeNP. A Western blotting analysis of the expression of autophagy-related proteins in NPCs induced by H2O2 with or without CQ (10 μM) for 24 h. B *Densitometric analysis* of autophagy-related protein levels. C Western blotting analysis of the expression of ECM metabolism-related proteins in NPCs induced by H2O2 with or without CQ (10 μM) for 24 h. D *Densitometric analysis* of ECM metabolism-related protein levels. E Western blotting analysis of the expression of apoptosis-related proteins in NPCs induced by H2O2 with or without CQ (10 μM) for 24 h. F *Densitometric analysis* of apoptosis-related protein levels. $$n = 3$$, mean ± SD. * $p \leq 0.05$ compared to the H2O2 group, †$p \leq 0.05$ compared to the IGK group, #$p \leq 0.05$ compared to the same treatment groups without CQ
## Therapeutic efficacy of IGK@SeNP in the rat IDD model
Given the favorable results of in vitro anti-ECM degradation and anti‑apoptosis potential, the efficacy of IGK@SeNP in treating IDD in vivo was further evaluated in a rat model (Fig. 7A). The IDD rat model was established using needle puncture. Degenerative discs were locally injected with PBS, IGK, SeNP, or IGK@SeNP. Imaging examinations, including X-ray and magnetic resonance imaging (MRI), were used to evaluate the degree of disc degeneration 4 and 8 weeks after surgery. As shown in Fig. 7B, C, intervertebral height was significantly preserved in the IGK@SeNP group compared to that in the IDD group. However, the intervertebral height in the IGK group gradually decreased in 8 weeks, indicating that in vivo treatment with IGK alone is not very effective, while combining it with the SeNP drug delivery system enhanced the therapeutic effect. MRI revealed that the T2-weighted signal in the IDD group had declined, indicating a loss of water in these discs (Fig. 7D). After 6 weeks of treatment, the water content was preserved in the IGK@SeNP group, with significantly low signals in the IGK groups. The Pfirrmann grade results further confirmed that IGK@SeNP had superior efficacy for treating IDD in vivo (Fig. 7E).Fig. 7IGK@SeNP protects the nucleus pulposus in needle puncture-induced IDD rats. A Treatment schedule of animal experiments. Two weeks after surgery, the IDD model was validated by MRI. PBS, IGK, SeNP, and IGK@SeNP were injected weekly thereafter. Imaging examinations were performed at weeks 4 and 8. B X-ray images of rat coccygeal spines at the 4th and 8th-week post-surgery. C Intervertebral heights of rat coccygeal spines measured on the X-ray. D T2-weighted MRI of coccygeal discs at the 4th and 8th-week post-surgery. E Pfirrmann grade scores of coccygeal discs measured on MRI. F, G HE and SF staining of rat coccygeal discs. Glycosaminoglycans are shown with safranin O. Scale bar = 1 mm. H Histological grade scores of coccygeal discs at the 4th and 8th-week post-surgery. $$n = 5$$, mean ± SD. * $p \leq 0.05$ compared to the IDD group, †$p \leq 0.05$ compared to the IGK group Histological sections were collected at 4 and 8 weeks after surgery. Hematoxylin–eosin (HE) staining was used to observe the morphology of the nucleus pulposus and annulus fibrosus (Fig. 7F). The nucleus pulposus of the IDD group prominently decreased in volume and was replaced by fibrous tissue at 8 weeks, and the arrangement of the annulus fibrosus was disordered. In addition, a substantial loss of glycosaminoglycans was observed at 4 and 8 weeks in the IDD group by safranin O-fast green (SF) staining (Fig. 7G). Histological changes induced by needle puncture were weakened in the IGK@SeNP group. However, in the IGK and SeNP groups, the loss of the nucleus pulposus was evident with destruction of the disc structure. Based on the histological grade, the IDD group recorded the highest score, whereas the IGK@SeNP group recorded the lowest score, except for the control group (Fig. 7H).
We performed immunochemical (IHC) staining to analyze the expression levels of ACAN and MMP3, which are representative indices of matrix metabolism in the nucleus pulposus. After needle puncture surgery, the expression of ACAN in all model groups was degraded, while the IGK@SeNP group maintained the highest ACAN expression among the model groups (Fig. 8A). Conversely, the expression of MMP3 was substantially increased in the IDD group and was relatively restrained in the IGK, SeNP, and particularly in the IGK@SeNP group, indicating that the ECM of the nucleus pulposus was rescued (Fig. 8B).Fig. 8IGK@SeNP attenuates matrix degradation and cell apoptosis in vivo. A, B IHC staining of ACAN and MMP3 in rat coccygeal discs. C *Quantitative analysis* of ACAN- and MMP3-positive rates in different groups. D IF staining of cleaved CASP3 in rat coccygeal discs. E TUNEL staining of rat coccygeal discs. F *Quantitative analysis* of cleaved CASP3- and TUNEL-positive cells in different groups. Scale bar = 100 μm. $$n = 5$$, mean ± SD. * $p \leq 0.05$ compared to the IDD group, †$p \leq 0.05$ compared to the IGK group Subsequently, we used TUNEL assay and IF staining of cleaved CASP3 to analyze the apoptosis of NPCs in degenerative discs (Fig. 8D, E). In the IDD group, the apoptosis pathway was upregulated, and the apoptosis rate increased remarkably with time. However, treatment with IGK@SeNP inhibited this pathological progression in vivo. The above results indicate that IGK@SeNP exerts a prominent therapeutic effect on IDD in vivo.
## Discussion
The incidence of IDD increases with age and can lead to spinal instability, radicular pain, and even disability, making it a critical health issue [2, 28]. Many studies have found that ROS accumulation and high levels of oxidative stress are important causes of IDD development because they can cause ECM degradation and NPC death [29, 30]. Therefore, ROS have become a key target for the treatment of IDD. However, traditional therapeutic agents have difficulty accessing the interior of avascular intervertebral discs, so they can only reduce inflammation and temporarily relieve pain outside the intervertebral disc but cannot remove accumulated ROS, which makes them unable to fundamentally prevent the progression of IDD [31, 32]. Although some studies have used novel biomaterials to deliver drugs into the intervertebral disc and effectively reduce ROS levels, oxidatively damaged organelles are not removed simultaneously, which may cause blockage in the repair of cell viability and metabolic function, ultimately affecting the therapeutic effect [14, 33, 34]. Therefore, there is an urgent need to develop new strategies to effectively treat IDD by removing damaged organelles and reducing ROS. In this study, we fabricated an ROS-responsive nanodelivery system based on diselenide block copolymers that can load and intelligently release IGK, a natural autophagy stimulator. Our experiments confirmed that IGK@SeNP could effectively scavenge ROS and delay the progression of IDD in rats by enhancing autophagy to degrade oxidatively damaged mitochondria. Moreover, blocking autophagy attenuated the protective function of IGK@SeNP in regulating ECM metabolism and inhibiting apoptosis, indicating that IGK@SeNP has the dual efficacy of ROS clearance and autophagy enhancement, which makes it advantageous for IDD treatment.
Oxidative stress plays an important role in many physiological processes, such as cell signaling and immune responses. An important mechanism is that ROS can promote the expression of BAX and release cytochrome C by increasing the permeability of the outer mitochondrial membrane, which activates caspase-mediated cascade amplification and promotes apoptosis, resulting in tissue entrapment into a pathological state. Cheng et al. [ 33] revealed that the combination of antioxidants and thermosensitive hydrogels could decrease oxidative stress induced by H2O2 and restore the content of GAGs in the early stages of IDD. Bai et al. [ 35] developed a rapamycin-loaded hydrogel and successfully induced macrophage differentiation into the M2 phenotype, thereby alleviating the inflammatory microenvironment of the intervertebral disc. Zhu et al. [ 34] used MnO2 nanoparticles to control the release of TGF-β3, and this approach suppressed H2O2-induced oxidative damage by increasing the expression of antioxidative genes. Our study showed that SeNP was able to scavenge ROS owing to the reducibility of Se-Se bonds, which prevented H2O2-induced ECM degradation and NPC apoptosis. However, this effect was insufficient. We found that although SeNP removed intracellular ROS, the autophagy level decreased, marked by the downregulation of initiation-related proteins (BECN1, ATG7, and LC3B). This deficiency resulted in the inability of NPCs to repair oxidatively damaged organelles, as vacuolization in mitochondria was observed in the SeNP group (Fig. 5E), and eventually induced NPC apoptosis.
Chen et al. [ 36] revealed that the PERK/eIF2α pathway could promote autophagy; however, blocking the PERK/eIF2α pathway or inhibiting autophagy reduced NPC viability. Further experiments that supported the cellular protective effect of autophagy showed that H2O2 induced DNA damage in NPCs and caused AIM2-associated inflammatory cascades. After enhancing autophagy, autophagy-dependent secretion promotes AIM2 inflammasome secretion from NPCs, which inhibits cellular DNA damage and apoptosis, thus delaying the progression of IDD [18]. In our experiments, IGK significantly activated autophagy and protected NPCs by removing damaged organelles. A recent study confirmed that IGK can inhibit proteasome degradation to stabilize the Nrf2 protein, thus promoting nuclear translocation of Nrf2, which in turn triggers the activation of the antioxidant system and plays a role in protecting mitochondria [37]. In the dysfunctional mitochondria, more electrons leak from mitochondrial complexes, which subsequently generate H2O2 under the catalysis of superoxide dismutase [38]. Endogenous ROS can destroy intracellular biomacromolecules and cause cellular dysfunction or even death. As an autophagy stimulator, IGK can both resist mitochondrial collapse and induce impaired mitochondrial clearance; thus, it has powerful therapeutic potential in ROS-induced IDD.
Many studies have attempted to treat IDD using multidrug administration. This strategy allows the pharmacological effects of several drugs to be superimposed on each other and achieves greater therapeutic effects at lower drug concentrations [39–41]. A recent study revealed that dysfunction of either autophagy or antioxidant systems can lead to aggravated detrimental effects of H2O2 on intervertebral disc cells [42]. Therefore, we combined an autophagy stimulator (IGK) with ROS scavenger (SeNP) and found that the assembled nanodelivery system exerted superior efficacy. Moreover, an autophagy blocker attenuated the therapeutic effects of IGK@SeNP on ECM metabolism regulation and inhibition of apoptosis, indicating that IGK-induced autophagy is necessary for NPC self-repair. In addition, SeNP allows IGK to be more stably present within the disc, which is beneficial in reducing the frequency of medication and the complications of intradiscal injection. These characteristics make its clinical application promising.
However, we also considered the limitations of this study. Although we employed H2O2 to induce NPC degeneration, there may be more complex pathophysiological features in the actual degenerated discs [43]. Therefore, we will use NPCs obtained from patients with IDD for further validation. Moreover, the intervertebral disc is an avascular closed structure, and nanoparticles injected in situ are rarely distributed to other organs [31]. Thus, our study did not investigate blood compatibility and toxicity in vivo. Although HE staining of intervertebral discs showed no obvious inflammatory cell infiltration, we will verify the in vivo biocompatibility if IGK@SeNP is used in other disease models. Overall, we propose a novel ROS-responsive nanodrug delivery system that can delay the progression of IDD by synergistically enhancing autophagy and scavenging ROS.
## Conclusion
In this study, we synthesized a novel ROS-responsive nanodelivery system for SeNP loaded with IGK for IDD treatment. Our findings demonstrated that the nanoparticles had an ROS-stimulated responsive drug release effect. SeNP combined with IGK synergistically eliminated ROS and enhanced autophagy. IGK@SeNP protected NPCs and delayed the pathological process of IDD, both in vitro and in vivo. Collectively, this study suggests that IGK@SeNP may be an attractive therapeutic agent for treating IDD.
## Materials and reagents
Selenium powder, sodium borohydride, isophorone diisocyanate (IPDI), methoxypolyethylene glycol (mPEG), dibutyltin dilaurate (DBTDL), 11-bromoundecanol, methoxypolyethylene glycols, Tetrahydrofuran (THF), hydrogen peroxide (H2O2), Cy 5 (Aladdin, Shanghai, China). Isoginkgetin (IGK), chloroquine (CQ), (MedChemExpress, Monmouth Junction, USA). Primary antibodies against aggrecan, collagen II, SOX9, MMP3, MMP13, ADAMTS5, ATG7, Beclin1, LC3B, SQSTM1/p62 (Abcam, Cambridge, UK), primary antibodies against BCL2, BAX and cleaved caspase-3 (Cell Signalling Technology, Danvers, USA), primary antibodies against β-actin (Proteintech, Wuhan, China). Goat anti-rabbit IgG H&L (Alexa Fluor® 488), goat anti-rabbit IgG H&L (Alexa Fluor® 594), (Abcam, Cambridge, UK), HRP-labeled goat anti-rabbit IgG (Proteintech, Wuhan, China). Penicillin–streptomycin (Sigma-Aldrich, St. Louis, USA). Cell counting kit-8 (CCK-8), Calcein/PI assay kit, ROS assay kit, and TUNEL apoptosis assay kit (Beyotime, Shanghai, China). Phosphate-buffered saline (PBS), FITC-annexin V/PI apoptosis detection kit, fetal bovine serum (FBS), DMEM/F12 (Thermo Fisher Scientific, Waltham, USA). Nitrocellulose membranes (Pall, New York, USA). Electrochemiluminescence substrate (Meilunbio, Dalian, China). Zoletile (Virvac, Carros, France).
## Synthesis of IGK@SeNP
The synthesis of SeNP was consistent with our previous work description [44]. In brief, selenium powder (1 g, 12.7 mmol) was dissolved in deionized water (15 mL) with stirring under N2 flow. Sodium borohydride solution (0.1 g/mL, 10 mL) was dropwise added to the reaction for 15 min. Secondly, another selenium powder (1 g, 12.7 mmol) was added and reacted for 15 min. Afterward, 11-bromoundecanol (6.33 g, 25.2 mmol) was dissolved in THF (25 mL) and added to the reaction system. The mixture was heated at 50 °C and maintained for 24 h, and then filtered while hot. The obtained solution was extracted three times with 20 mL of dichloromethane and dried with anhydrous sodium sulfate. The product was purified by column chromatography with a 1:4 mixture of ethyl acetate and methylene chloride as eluent. The di(1-hydroxylundecyl) diselenide was obtained after removing the solvent by rotary evaporation. Di(1-hydroxylundecyl) diselenide (2 g, 4 mmol) and IPDI (0.98 g, 4.40 mmol) were dissolved in THF (60 mL) under N2 flow. Then DBTDL (0.05 mg, 0.1 mmol) was added as a catalyst. After a reaction for 2 h at 50 °C, mPEG2000 (13 g, 6.50 mmol) was added and the reaction was continued at 50 °C for 24 h. THF was removed by rotary evaporation. The obtained Se-polymer was purified by cold methanol precipitated and collected after vacuum drying.
IGK (1 mg), triethylamine (50 μL), and Se-polymer (10 mg) were dissolved in N,N-dimethylformamide (DMF, 2 mL) and stirred for 2 h. Deionized water (10 mL) was slowly added with stirring. Subsequently, the solution was dialyzed for 24 h (MWCO 3500 Da) against deionized water. IGK-loaded nanoparticles (IGK@SeNP) were obtained by ultrafiltration centrifugation (MWCO 10000 Da). The SeNP and Cy5@SeNP were constructed using the same method. IGK@SeNP were freeze-dried and dissolved in DMSO, then detected using a UV−vis spectrometer. The IGK loading content was $8.69\%$ and the drug entrapment efficiency was $41.78\%$.
## Characterization
1H NMR spectroscopy was used to detect the Se-polymer modified with IPDI and mPEG. The number-average molecular weight (M̅n) and weight-average molecular weights (M̅w) of the Se-polymers were determined using gel permeation chromatography (GPC). The morphology of SeNP was observed and reported using a transmission electron microscope (TEM) at an acceleration voltage of 120 kV. The particle diameters and dispersion of particle size were reported by dynamic light scattering (DLS).
## ROS-responsive drug release
5 mL of Cy5@SeNP solution was treated with gradient concentrations of H2O2 (0, 50, 100, 200 μM). The Cy5 content was measured using a fluorescence spectrophotometer at 650 nm excitation wavelength.
## Cell culture and treatment
Human nucleus pulposus cells (NPCs, ScienCell, Carlsbad, USA) were cultured in DMEM/F12 medium supplemented with $10\%$ FBS, $1\%$ penicillin–streptomycin at 37 °C and $5\%$ CO2. To simulate an oxidative microenvironment of degenerative disc, NPCs were pretreated with PBS, IGK, SeNP, or IGK@SeNP for 4 h, then stimulated with 200 μM H2O2 for 24 h.
## Cellular uptake
Due to the lack of fluorescent properties of IGK, we used Cy5 as a drug model to investigate the cellular uptake of Cy5@SeNP in NPCs [45]. NPCs were seeded in 24-well plates. After washes with PBS, Hoechst 33342 was incubated with the cells for 30 min at 37 °C. Subsequently, the culture medium contained Cy5@SeNP replaced the original medium. The cellular uptake was observed by fluorescence microscope.
## Live/dead staining assay
NPCs were seeded into 24-well plates and co-cultured with IGK@SeNP (60 μg/mL). After washing with PBS, the Calcein/PI staining dye was added and incubated for 30 min at 37 °C for 24, 48, and 72 h. After incubation, the cells were washed again and observed under a fluorescence microscope.
## ROS-scavenging evaluation
Intracellular ROS level was detected by DCFH-DA. NPCs were cultured in 6-well plates and were treated as per the aforementioned protocol. Afterward, the cells were stained with DCFH-DA (10 μM) for 20 min at 37 °C and washed 3 times with serum-free DMEM/F12. The flow cytometer analysis was executed by FACSVerse.
## Rat IDD model and treatment
All animal experiments were performed following the Ethics Committee for Animal Experiments. A total of sixty 8-week-old male SD rats were randomly allocated to 5 groups: Control, IDD, IGK, SeNP, and IGK@SeNP. The IDD model was established by needle puncture [46]. Rats were intraperitoneally anesthetized with 50 mg/kg Zoletile and 3 mg/kg xylazine hydrochloride. After successful anesthesia, the rats were placed prone on a clean operating table. The rat tails were disinfected and draped, and a 20G needle was used to penetrate 5 mm percutaneously into the dorsal center on Co $\frac{7}{8}$ and Co $\frac{8}{9.}$ The needle was rotated 360° and removed after staying for 30 s. The surgical area was disinfected again to avoid infection.
To determine whether the modeling was successful, a magnetic resonance imaging (MRI) examination was conducted after 2 weeks. The degenerative discs were orthotopically injected with PBS (2 μL), IGK (2 μL, 20 μM), SeNP (2 μL, 60 μg/mL), or IGK@SeNP (2 μL, 60 μg/mL) using a 33G needle.
## Radiology evaluation
X-rays and MRI were performed at 4 and 8 weeks after IDD construction. The anesthetized rats were placed supine and the tails were straightened. The intervertebral heights were measured by image J on X-ray images. A 1.5 T MRI scanner was used to acquire the intervertebral disc signals on T2 weighted phase. Further, the severity of IDD was assessed according to the modified Pfirrmann grading system [47]
## Histological analysis
The rats were sacrificed by injecting over a dose of pentobarbital sodium at 4 weeks or 8 weeks. Intervertebral discs were harvested under aseptic conditions and fixed in $4\%$ paraformaldehyde for 24 h. After decalcification by $10\%$ EDTA for 2 weeks, the discs were sliced into 4 μm thick sections. Hematoxylin–eosin (HE), safranin O-fast green (SF), immunochemical (IHC), and IF staining were performed according to the manufacturer’s instructions [48, 49]. The morphology of intervertebral discs was observed under a light microscope and determined the histological grade [50].
## Statistical analysis
All statistical analyses were conducted with SPS software version 26.0 (IBM SPSS Corp. Chicago, USA). Data were presented as mean ± standard deviations (SD). Student’s t-test or one-way analysis of variance (ANOVA) was used to evaluate the differences among various treatment groups. $P \leq 0.05$ was considered statistically significant. (*, †, and # denoted $P \leq 0.05$).
## Supplementary Information
Additional file 1: Fig S1. 1H NMR characterization of di(1-hydroxylundecyl) diselenide in CDCl3. Fig S2. GPC plot of diselenide-containing polymer in DMF.
## References
1. Knezevic NN, Candido KD, Vlaeyen JWS, Van Zundert J, Cohen SP. **Low back pain**. *Lancet* (2021) **398** 78-92. DOI: 10.1016/S0140-6736(21)00733-9
2. Francisco V, Pino J, Gonzalez-Gay MA, Lago F, Karppinen J, Tervonen O, Mobasheri A, Gualillo O. **A new immunometabolic perspective of intervertebral disc degeneration**. *Nat Rev Rheumatol* (2022) **18** 47-60. DOI: 10.1038/s41584-021-00713-z
3. Alvarez-Garcia O, Matsuzaki T, Olmer M, Miyata K, Mokuda S, Sakai D, Masuda K, Asahara H, Lotz MK. **FOXO are required for intervertebral disk homeostasis during aging and their deficiency promotes disk degeneration**. *Aging Cell* (2018) **17** e12800. DOI: 10.1111/acel.12800
4. Che H, Li J, Li Y, Ma C, Liu H, Qin J, Dong J, Zhang Z, Xian CJ, Miao D. **p16 Deficiency attenuates intervertebral disc degeneration by adjusting oxidative stress and nucleus pulposus cell cycle**. *Elife* (2020). DOI: 10.7554/eLife.52570
5. Zhang W, Li G, Luo R, Lei J, Song Y, Wang B, Ma L, Liao Z, Ke W, Liu H. **Cytosolic escape of mitochondrial DNA triggers cGAS-STING-NLRP3 axis-dependent nucleus pulposus cell pyroptosis**. *Exp Mol Med* (2022) **54** 129-142. DOI: 10.1038/s12276-022-00729-9
6. Yu BP. **Cellular defenses against damage from reactive oxygen species**. *Physiol Rev* (1994) **74** 139-162. DOI: 10.1152/physrev.1994.74.1.139
7. Kang L, Liu S, Li J, Tian Y, Xue Y, Liu X. **The mitochondria-targeted anti-oxidant MitoQ protects against intervertebral disc degeneration by ameliorating mitochondrial dysfunction and redox imbalance**. *Cell Prolif* (2020) **53** e12779. DOI: 10.1111/cpr.12779
8. Grunhagen T, Shirazi-Adl A, Fairbank JC, Urban JP. **Intervertebral disk nutrition: a review of factors influencing concentrations of nutrients and metabolites**. *Orthop Clin North Am* (2011) **42** vii. PMID: 21944584
9. Colella F, Garcia JP, Sorbona M, Lolli A, Antunes B, D'Atri D, Barre FPY, Oieni J, Vainieri ML, Zerrillo L. **Drug delivery in intervertebral disc degeneration and osteoarthritis: selecting the optimal platform for the delivery of disease-modifying agents**. *J Control Release* (2020) **328** 985-999. DOI: 10.1016/j.jconrel.2020.08.041
10. Haro H, Ebata S, Inoue G, Kaito T, Komori H, Ohba T, Sakai D, Sakai T, Seki S, Shiga Y. **Japanese Orthopaedic Association (JOA) clinical practice guidelines on the management of lumbar disc herniation, third edition—secondary publication**. *J Orthop Sci* (2022) **27** 31-78. DOI: 10.1016/j.jos.2021.07.028
11. Pope MH. **Biomechanics of the lumbar spine**. *Ann Med* (1989) **21** 347-351. DOI: 10.3109/07853898909149219
12. Liu T, Xiao B, Xiang F, Tan J, Chen Z, Zhang X, Wu C, Mao Z, Luo G, Chen X, Deng J. **Ultrasmall copper-based nanoparticles for reactive oxygen species scavenging and alleviation of inflammation related diseases**. *Nat Commun* (2020) **11** 2788. DOI: 10.1038/s41467-020-16544-7
13. Li X, Liu Y, Qi X, Xiao S, Xu Z, Yuan Z, Liu Q, Li H, Ma S, Liu T. **Sensitive activatable nanoprobes for real-time ratiometric magnetic resonance imaging of reactive oxygen species and ameliorating inflammation in vivo**. *Adv Mater* (2022) **34** e2109004. DOI: 10.1002/adma.202109004
14. Yang L, Yu C, Fan X, Zeng T, Yang W, Xia J, Wang J, Yao L, Hu C, Jin Y. **Dual-dynamic-bond cross-linked injectable hydrogel of multifunction for intervertebral disc degeneration therapy**. *J Nanobiotechnol* (2022) **20** 433. DOI: 10.1186/s12951-022-01633-0
15. Zhang Z, Xu T, Chen J, Shao Z, Wang K, Yan Y, Wu C, Lin J, Wang H, Gao W. **Parkin-mediated mitophagy as a potential therapeutic target for intervertebral disc degeneration**. *Cell Death Dis* (2018) **9** 980. DOI: 10.1038/s41419-018-1024-9
16. Hu S, Zhang C, Qian T, Bai Y, Chen L, Chen J, Huang C, Xie C, Wang X, Jin H. **Promoting Nrf2/Sirt3-dependent mitophagy suppresses apoptosis in nucleus pulposus cells and protects against intervertebral disc degeneration**. *Oxid Med Cell Longev* (2021) **2021** 6694964. DOI: 10.1155/2021/6694964
17. Levine B, Mizushima N, Virgin HW. **Autophagy in immunity and inflammation**. *Nature* (2011) **469** 323-335. DOI: 10.1038/nature09782
18. Li S, Liao Z, Luo R, Song Y, Wang K, Feng X, Ou Y, Wu X, Zhang Y, Gao Y. **Autophagy-based unconventional secretory for AIM2 inflammasome drives DNA damage resistance during intervertebral disc degeneration**. *Front Cell Dev Biol* (2021) **9** 672847. DOI: 10.3389/fcell.2021.672847
19. He R, Wang Z, Cui M, Liu S, Wu W, Chen M, Wu Y, Qu Y, Lin H, Chen S. **HIF1A alleviates compression-induced apoptosis of nucleus pulposus derived stem cells via upregulating autophagy**. *Autophagy* (2021) **17** 3338-3360. DOI: 10.1080/15548627.2021.1872227
20. Shao Z, Ni L, Hu S, Xu T, Meftah Z, Yu Z, Tian N, Wu Y, Sun L, Wu A. **RNA-binding protein HuR suppresses senescence through Atg7 mediated autophagy activation in diabetic intervertebral disc degeneration**. *Cell Prolif* (2021) **54** e12975. DOI: 10.1111/cpr.12975
21. Liu L, Wang Y, Zhang J, Wang S. **Advances in the chemical constituents and chemical analysis of ginkgo biloba leaf, extract, and phytopharmaceuticals**. *J Pharm Biomed Anal* (2021) **193** 113704. DOI: 10.1016/j.jpba.2020.113704
22. Yao J, Tang S, Shi C, Lin Y, Ge L, Chen Q, Ou B, Liu D, Miao Y, Xie Q. **Isoginkgetin, a potential CDK6 inhibitor, suppresses SLC2A1/GLUT1 enhancer activity to induce AMPK-ULK1-mediated cytotoxic autophagy in hepatocellular carcinoma**. *Autophagy* (2022). DOI: 10.1080/15548627.2022.2119353
23. He Y, Li W, Zheng Z, Zhao L, Li W, Wang Y, Li H. **Inhibition of protein arginine methyltransferase 6 reduces reactive oxygen species production and attenuates aminoglycoside- and cisplatin-induced hair cell death**. *Theranostics* (2020) **10** 133-150. DOI: 10.7150/thno.37362
24. Dong J, Zhang KJ, Li GC, Chen XR, Lin JJ, Li JW, Lv ZY, Deng ZZ, Dai J, Cao W, Jiang Q. **CDDO-im ameliorates osteoarthritis and inhibits chondrocyte apoptosis in mice via enhancing Nrf2-dependent autophagy**. *Acta Pharmacol Sin* (2022) **43** 1793-1802. DOI: 10.1038/s41401-021-00782-6
25. Yuan L, Zhang F, Qi X, Yang Y, Yan C, Jiang J, Deng J. **Chiral polymer modified nanoparticles selectively induce autophagy of cancer cells for tumor ablation**. *J Nanobiotechnol* (2018) **16** 55. DOI: 10.1186/s12951-018-0383-9
26. Peng Z, Yuan L, XuHong J, Tian H, Zhang Y, Deng J, Qi X. **Chiral nanomaterials for tumor therapy: autophagy, apoptosis, and photothermal ablation**. *J Nanobiotechnol* (2021) **19** 220. DOI: 10.1186/s12951-021-00965-7
27. Mizushima N, Yoshimori T, Levine B. **Methods in mammalian autophagy research**. *Cell* (2010) **140** 313-326. DOI: 10.1016/j.cell.2010.01.028
28. Kim HS, Wu PH, Jang IT. **Lumbar degenerative disease part 1: anatomy and pathophysiology of intervertebral discogenic pain and radiofrequency ablation of basivertebral and sinuvertebral nerve treatment for chronic discogenic back pain: a prospective case series and review of literature**. *Int J Mol Sci* (2020). DOI: 10.3390/ijms21041483
29. Zorov DB, Juhaszova M, Sollott SJ. **Mitochondrial reactive oxygen species (ROS) and ROS-induced ROS release**. *Physiol Rev* (2014) **94** 909-950. DOI: 10.1152/physrev.00026.2013
30. Yin H, Wang K, Das A, Li G, Song Y, Luo R, Cheung JPY, Zhang T, Li S, Yang C. **The REDD1/TXNIP complex accelerates oxidative stress-induced apoptosis of nucleus pulposus cells through the mitochondrial pathway**. *Oxid Med Cell Longev* (2021) **2021** 7397516. DOI: 10.1155/2021/7397516
31. Hickman TT, Rathan-Kumar S, Peck SH. **Development, pathogenesis, and regeneration of the intervertebral disc: current and future insights spanning traditional to omics methods**. *Front Cell Dev Biol* (2022) **10** 841831. DOI: 10.3389/fcell.2022.841831
32. Lim S, An SB, Jung M, Joshi HP, Kumar H, Kim C, Song SY, Lee JR, Kang M, Han I, Kim BS. **Local delivery of senolytic drug inhibits intervertebral disc degeneration and restores intervertebral disc structure**. *Adv Healthc Mater* (2022) **11** e2101483. DOI: 10.1002/adhm.202101483
33. Cheng YH, Yang SH, Lin FH. **Thermosensitive chitosan-gelatin-glycerol phosphate hydrogel as a controlled release system of ferulic acid for nucleus pulposus regeneration**. *Biomaterials* (2011) **32** 6953-6961. DOI: 10.1016/j.biomaterials.2011.03.065
34. Zhu L, Yang Y, Yan Z, Zeng J, Weng F, Shi Y, Shen P, Liu L, Yang H. **Controlled release of TGF-β3 for effective local endogenous repair in IDD using rat model**. *Int J Nanomed* (2022) **17** 2079-2096. DOI: 10.2147/IJN.S358396
35. Bai J, Zhang Y, Fan Q, Xu J, Shan H, Gao X, Ma Q, Sheng L, Zheng X, Cheng W. **Reactive oxygen species-scavenging scaffold with rapamycin for treatment of intervertebral disk degeneration**. *Adv Healthc Mater* (2020) **9** e1901186. DOI: 10.1002/adhm.201901186
36. Chen L, Liu L, Xie ZY, Wang F, Zhu L, Zhang C, Fan P, Sinkemani A, Hong X, Wu XT. **Protein kinase RNA-like ER kinase/eukaryotic translation initiation factor 2α pathway attenuates tumor necrosis factor alpha-induced apoptosis in nucleus pulposus cells by activating autophagy**. *J Cell Physiol* (2019) **234** 11631-11645. DOI: 10.1002/jcp.27820
37. Wu X, Huang J, Tang J, Sun Y, Zhao G, Yan C, Liu Z, Yi W, Xu S, Yu X. **Isoginkgetin, a bioactive constituent from ginkgo biloba, protects against obesity-induced cardiomyopathy via enhancing Nrf2/ARE signaling**. *Redox Biol* (2022). DOI: 10.1016/j.redox.2022.102485
38. Filomeni G, De Zio D, Cecconi F. **Oxidative stress and autophagy: the clash between damage and metabolic needs**. *Cell Death Differ* (2015) **22** 377-388. DOI: 10.1038/cdd.2014.150
39. Cho H, Lee S, Park SH, Huang J, Hasty KA, Kim SJ. **Synergistic effect of combined growth factors in porcine intervertebral disc degeneration**. *Connect Tissue Res* (2013) **54** 181-186. DOI: 10.3109/03008207.2013.775258
40. Colombier P, Clouet J, Boyer C, Ruel M, Bonin G, Lesoeur J, Moreau A, Fellah BH, Weiss P, Lescaudron L. **TGF-β1 and GDF5 act synergistically to drive the differentiation of human adipose stromal cells toward nucleus pulposus-like cells**. *Stem Cells* (2016) **34** 653-667. DOI: 10.1002/stem.2249
41. Yu C, Li D, Wang C, Xia K, Wang J, Zhou X, Ying L, Shu J, Huang X, Xu H. **Injectable kartogenin and apocynin loaded micelle enhances the alleviation of intervertebral disc degeneration by adipose-derived stem cell**. *Bioact Mater* (2021) **6** 3568-3579. DOI: 10.1016/j.bioactmat.2021.03.018
42. Kang L, Liu S, Li J, Tian Y, Xue Y, Liu X. **Parkin and Nrf2 prevent oxidative stress-induced apoptosis in intervertebral endplate chondrocytes via inducing mitophagy and anti-oxidant defenses**. *Life Sci* (2020) **243** 117244. DOI: 10.1016/j.lfs.2019.117244
43. Huang YC, Leung VY, Lu WW, Luk KD. **The effects of microenvironment in mesenchymal stem cell-based regeneration of intervertebral disc**. *Spine J* (2013) **13** 352-362. DOI: 10.1016/j.spinee.2012.12.005
44. Zhang L, Zhang S, Xu J, Li Y, He J, Yang Y, Huynh T, Ni P, Duan G, Yang Z, Zhou R. **Low-dose X-ray-responsive diselenide nanocarriers for effective delivery of anticancer agents**. *ACS Appl Mater Interfaces* (2020) **12** 43398-43407. DOI: 10.1021/acsami.0c11627
45. Yang G, Fan M, Zhu J, Ling C, Wu L, Zhang X, Zhang M, Li J, Yao Q, Gu Z, Cai X. **A multifunctional anti-inflammatory drug that can specifically target activated macrophages, massively deplete intracellular H(2)O(2), and produce large amounts CO for a highly efficient treatment of osteoarthritis**. *Biomaterials* (2020) **255** 120155. DOI: 10.1016/j.biomaterials.2020.120155
46. Ge J, Yan Q, Wang Y, Cheng X, Song D, Wu C, Yu H, Yang H, Zou J. **IL-10 delays the degeneration of intervertebral discs by suppressing the p38 MAPK signaling pathway**. *Free Radic Biol Med* (2020) **147** 262-270. DOI: 10.1016/j.freeradbiomed.2019.12.040
47. Qian J, Ge J, Yan Q, Wu C, Yang H, Zou J. **Selection of the optimal puncture needle for induction of a rat intervertebral disc degeneration model**. *Pain Physician* (2019) **22** 353-360. PMID: 31337166
48. Ge J, Cheng X, Yan Q, Wu C, Wang Y, Yu H, Yang H, Zhou F, Zou J. **Calcitonin inhibits intervertebral disc degeneration by regulating protein kinase C**. *J Cell Mol Med* (2020) **24** 8650-8661. DOI: 10.1111/jcmm.15496
49. Xu X, Wang D, Zheng C, Gao B, Fan J, Cheng P, Liu B, Yang L, Luo Z. **Progerin accumulation in nucleus pulposus cells impairs mitochondrial function and induces intervertebral disc degeneration and therapeutic effects of sulforaphane**. *Theranostics* (2019) **9** 2252-2267. DOI: 10.7150/thno.30658
50. Han B, Zhu K, Li FC, Xiao YX, Feng J, Shi ZL, Lin M, Wang J, Chen QX. **A simple disc degeneration model induced by percutaneous needle puncture in the rat tail**. *Spine* (2008) **33** 1925-1934. DOI: 10.1097/BRS.0b013e31817c64a9
|
---
title: Residual periodontal ligament in the extraction socket promotes the dentin
regeneration potential of DPSCs in the rabbit jaw
authors:
- Bin Luo
- Yu Luo
- Lin He
- Yangyang Cao
- Qingsong Jiang
journal: Stem Cell Research & Therapy
year: 2023
pmcid: PMC10029302
doi: 10.1186/s13287-023-03283-x
license: CC BY 4.0
---
# Residual periodontal ligament in the extraction socket promotes the dentin regeneration potential of DPSCs in the rabbit jaw
## Abstract
### Background
Because of the low regeneration efficiency and unclear underlying molecular mechanism, tooth regeneration applications are limited. In this study, we explored the influence of residual periodontal ligament on the dentin regeneration potential of dental pulp stem cells (DPSCs) in the jaw.
### Methods
To establish a tooth regeneration model, the incisors of New Zealand white rabbits were extracted while preserving residual periodontal ligament, followed by the implantation of DPSCs. After 3 months, micro-computed tomography (micro-CT), stereomicroscopy and scanning electron microscopy (SEM) were used to observe the volume, morphology and microstructure of regenerated tissue. Histological staining and immunostaining analyses were used to observe the morphological characteristics and expression of the dentin-specific proteins DMP1 and DSPP. To explore the mechanism, DPSCs and periodontal ligament stem cells (PDLSCs) were cocultured in vitro, and RNA was collected from the DPSCs for RNA-seq and bioinformatic analysis.
### Results
The results of micro-CT and stereomicroscopy showed that the number of sites with regeneration and the volume of regenerated tissue in the DPSCs/PDL group ($\frac{6}{8}$, 1.07 ± 0.93 cm3) were larger than those in the DPSCs group ($\frac{3}{8}$, 0.23 ± 0.41 cm3). The results of SEM showed that the regenerated dentin-like tissue in the DPSCs and DPSCs/PDL groups contained dentin tubules. Haematoxylin and eosin staining and immunohistochemical staining indicated that compared with the DPSCs group, the DPSCs/PDL group showed more regular regenerated tissue and higher expression levels of the dentin-specific proteins DMP1 and DSPP (DMP1: $$P \leq 0.02$$, DSPP: $$P \leq 0.01$$). RNA-seq showed that the coculture of DPSCs with PDLSCs resulted in the DPSCs differentially expressing 427 mRNAs (285 upregulated and 142 downregulated), 41 lncRNAs (26 upregulated and 15 downregulated), 411 circRNAs (224 upregulated and 187 downregulated), and 19 miRNAs (13 upregulated and 5 downregulated). Bioinformatic analysis revealed related Gene Ontology function and signalling pathways, including extracellular matrix (ECM), tumour necrosis factor (TNF) signalling and chemokine signalling pathways.
### Conclusions
Residual periodontal ligament in the extraction socket promotes the dentin regeneration potential of DPSCs in the jaw. RNA-seq and bioinformatic analysis revealed that ECM, TNF signalling and chemokine signalling pathways may represent the key factors and signalling pathways.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13287-023-03283-x.
## Background
Tooth loss caused by caries, periodontal disease, and trauma is a common oral condition. Some progress has been made in the prevention and treatment of oral diseases, and the structure and function of missing teeth can be restored in many ways; however, tooth loss remains an outstanding public health issue [1, 2]. Methods for replacing missing teeth include the use of traditional fixed bridges, removable partial dentures and implant restoration [3]. However, traditional restorations lack the periodontal ligament structure and cannot provide the same perception or stress buffering capability as natural teeth [4, 5]. With the development of stem cell biology and tissue engineering technology, researchers have attempted to regenerate tooth structures and even whole teeth [6, 7]. Regenerated teeth have biomechanical properties and periodontal ligament and dentin matrix structures similar to those of natural teeth [8–10]. Tooth regeneration strategies can be roughly divided into two categories. One type is based on an epithelial-mesenchymal bioengineered tooth germ, but this type of method is difficult to apply due to the limited cell resource and uncontrollable morphology of the regenerated teeth [11]. In contrast, another type of biological root regeneration strategy is based on the combination of mesenchymal stem cells (MSCs) and scaffold materials; this type of method is more feasible, and the combination of a preformed root scaffold and MSCs results in the formation of a functional tooth root in the alveolar bone [8, 9].
The ultimate goal of tooth regeneration research is to regenerate tooth structures similar to those of natural teeth and thus restore the missing teeth. Generally, the method for tooth regeneration in vivo is to prepare a cavity in the jawbone and then implant a composite consisting of a scaffold and MSCs. In a recent study, functional tooth roots were regenerated by implanting a composite comprising a root scaffold material and dental pulp stem cells (DPSCs) in the swine jawbone; however, compared with the $100\%$ success rate of implant-supported dentures, the $21.7\%$ success rate of regenerated biological tooth roots indicates low regeneration efficiency [10]. In these studies, an artificially prepared cavity was used as the implant bed for MSCs, and the bone-derived microenvironment in the cavity is more likely to induce the osteogenic differentiation of MSCs, resulting in low tooth regeneration efficiency in vivo. However, some studies have found that adding bone morphogenetic protein 2 (BMP2), secreted frizzled-related protein 2 (SFRP2) and other factors to the MSC microenvironment can improve the efficiency of dentin regeneration [12, 13], which indicates that regulation of the microenvironment may have a positive impact on the dentinogenic differentiation of MSCs.
A large number of studies have confirmed that the microenvironment has a very large impact on the physiological functions, pathological changes and therapeutic effects of stem cells [14, 15]. Physiologically, the microenvironment of MSCs is composed of various tissue components, cell populations and soluble factors, which strictly regulate the behaviour of MSCs [16, 17]. Under pathological conditions such as osteoporosis and periodontitis, the viability and differentiation of MSCs are severely impaired, leading to aggravation of the disease and impaired tissue healing [18–20]. In addition, in cell therapy and tissue engineering, the donor and recipient microenvironments play key roles in determining the regenerative efficacy of the transplanted MSCs [21, 22]. The extracellular matrix (ECM) is an important component of the cellular microenvironment. ECM obtained by decellularization contains a large number of growth factors and can significantly promote the proliferation and differentiation of MSCs [23]. In summary, these studies further demonstrate the key role of cell–microenvironment interactions in MSC-mediated tooth regeneration, and it is necessary to explore the influence of different microenvironments on MSC-mediated tooth regeneration.
Periodontal ligament (PDL) is a natural connective tissue between teeth and alveolar bone. Periodontal ligament is mainly composed of periodontal ligament fibres, cells and ECM [24, 25]. In the clinic, a large number of teeth are removed for different reasons. After a tooth is extracted, some periodontal ligament will remain in the extraction socket and maintain an odontogenic microenvironment in the extraction socket, different from the bone-derived microenvironment created by artificial cavity preparation. To clarify whether the odontogenic microenvironment maintained by residual periodontal ligament in the extraction socket can promote the dentin regeneration capability of MSCs, in this study, we extracted rabbit incisors while retaining the periodontal ligament structure in the fresh extraction socket to establish an odontogenic microenvironment and then implanted DPSCs. We explored the effect of the residual periodontal ligament microenvironment on the dentin regeneration potential of DPSCs in the rabbit jaw. The null hypothesis was that residual periodontal ligament in the extraction socket promotes the dentin regeneration potential of DPSCs in the rabbit jaw.
## Methods
New Zealand white rabbits were selected as experimental animals. The right upper and lower incisors were extracted, the residual periodontal ligament in the extraction socket was retained, and DPSCs were implanted to establish a tooth regeneration model. The dentin regeneration potential of the DPSCs in the residual periodontal ligament microenvironment was detected by micro-computed tomography (micro-CT), stereomicroscopy, scanning electron microscopy (SEM), histological staining and immunostaining. To explore the underlying mechanism, we cocultured rabbit DPSCs and periodontal ligament stem cells (PDLSCs) in vitro and collected DPSCs for RNA-seq and bioinformatic analysis.
## Cell culture and identification
The animal research involved in this work was approved by the Animal Ethics and Walfare Committee of Beijing Stomatological Hospital Affiliated to Capital Medical University (Reference number: KQYY-202101-003 and KQYY-202111-005). The care and use of animals were performed strictly following the regulations on the management of experimental animals. Rabbit DPSCs and PDLSCs were obtained as previously described [26, 27]. Incisors were extracted from 3-month-old New Zealand white rabbits after oral disinfection. Phosphate-buffered saline (PBS) was used to rinse the tooth tissue and then collected the pulp and periodontal ligament. The pulp and periodontal ligament were cut with scissors and digested in 3 mg/ml type I collagenase and 4 mg/ml dispase for 1 h. After centrifugation, the rabbit DPSCs and PDLSCs were resuspended in Dulbecco's modified Eagle's medium containing foetal bovine serum and cultured in a cell incubator. The rabbit DPSCs and PDLSCs from passage 3–5 were used for subsequent research.
Flow cytometry was used to identify the surface markers of rabbit DPSCs before application. When the cell cultures reached 80–$90\%$ confluence, a cell suspension was obtained by trypsin digestion and aliquoted into a few sterile tubes, with each tube containing 1 × 106 cells. Then, 5 µl (1:200) of anti- rabbit CD90 (cat no. ab225; Abcam, Cambridge, UK), CD105 (cat no. ab11414; Abcam), CD34 (cat no. ab81289; Abcam), CD45 (cat no. ab10558; Abcam), CD44 (cat no. MA5-28376, Invitrogen), and vimentin (cat no. GTX79851, GeneTex) antibodies were added to the samples and incubated at 4 °C for 60 min in the dark. After washing 3 times with PBS, the samples were incubated with the secondary antibody at 4 °C for 60 min, and then flow cytometry was used to identify the surface markers.
## Cell transplantation into the rabbit jaw
Twelve 3-month-old New Zealand white rabbits were randomly divided into three groups, i.e., the blank control group, the DPSCs group and the DPSCs/PDL group, with 4 rabbits in each group. All surgical procedures were completed under anaesthesia established with an intramuscular injection of 0.25 ml/kg Zoletil 50. After disinfecting the oral cavity, the right upper and lower incisors of the rabbit were removed. In the blank control group and the DPSCs group, each extraction socket was thoroughly cleaned to remove the residual periodontal ligament. In the DPSCs/PDL group, the residual periodontal ligament was retained in the extraction socket. In the blank control group, 150 µl of hydrogel was implanted, while in the DPSCs and DPSCs/PDL groups, a mixture of 100 µl of DPSCs suspension (1 × 106 cells) and 50 µl of hydrogel was implanted. Absorbable sutures were then used to close the wound. Penicillin was injected intramuscularly for 3 days to avoid infection.
## Micro-CT, stereomicroscopy and SEM observation
Three months after the model was established, the rabbits were sacrificed, and the upper and lower jawbones were obtained for examination by micro-CT (80 kV, 2 s, Siemens Inveon, Munich, Germany). CTAn software was used to reconstruct and calculate the volume of regenerated tissue in the blank control, DPSCs and DPSCs/PDL groups.
After rabbit jawbones from the blank control, DPSCs and DPSCs/PDL groups were cut on the coronal plane, a stereomicroscope was used for general observation of the regenerated tissue, and SEM (Phenom-World Co., Ltd., Netherlands) was used to observe the microstructure.
## Histological staining and immunostaining
Rabbit jawbones were decalcified with $10\%$ acetic acid buffer (pH 8.0) for 4 months, embedded in paraffin and then sectioned at 5 µm. The tissue sections were routinely deparaffinized and hydrated, incubated in sodium citrate solution for tissue antigen retrieval, and then incubated with the blocking solution at room temperature for 20 min to block endogenous peroxidase activity. The samples were subsequently blocked with normal goat serum for 40 min at room temperature, followed by incubation with primary antibodies overnight at 4 °C. Finally, a 3′-diaminobenzidine (DAB) kit was used to detect antigen expression in the DPSCs and DPSCs/PDL groups. The primary antibodies used were as follows: rabbit anti-dentin sialophosphoprotein (DSPP; bs-10316R; Bioss) and rabbit anti-dentin matrix acidic phosphoprotein 1 (DMP1; bs-12359R; Bioss). Conventional haematoxylin and eosin (H&E) staining and microscopy (OLYMPUS BX53) were used for histomorphological analysis in the blank control, DPSCs and DPSCs/PDL groups.
## RNA-seq and bioinformatic analysis in vitro
We next aimed to explore the mechanism by which residual periodontal ligament promotes the dentin regeneration potential of DPSCs in the jaw. We cocultured DPSCs and PDLSCs in vitro and compared the gene expression profiles of DPSCs cocultured with and cultured without PDLSCs to identify differentially expressed genes. DPSCs and PDLSCs were cocultured in six-well Transwell plates, separated by a 0.4-µm pore-size filter membrane. DPSCs were collected after a total of 3 days of coculture, and TRIzol reagent (Invitrogen) was used to extract total RNA. The RNA quantity and quality were determined by a multiImager and spectrophotometer (Meriton, China). Then, according to the manufacturer’s instructions, a library was constructed using TruSeq Stranded Total RNA with Ribo-Zero Gold (Illumina). Transcriptome sequencing and analysis were performed by OE Biotech Co., Ltd. DESeq was used to screen for differentially expressed genes according to the conditions of q < 0.05 and fold change > 2 or fold change < 0.5. After identifying the differentially expressed genes, Gene Ontology (GO) enrichment analysis was performed to describe their functions, and the Kyoto Encyclopedia of Genes and Genomes (KEGG) database was used to perform a pathway analysis.
## Statistical analysis
All statistical analyses were performed with SPSS 22 statistical software. The statistical significance of the differences was determined by Student’s t-test, including for comparison of the volume of dentin-like tissue and expression of DMP1 and DSPP in the DPSCs and DPSCs/PDL groups and the expression of mRNAs in the DPSCs and DPSCs/PDLSCs groups. $P \leq 0.05$ was considered significant.
## Expression of surface markers on rabbit DPSCs
Flow cytometry was used to identify the surface markers of rabbit DPSCs. The results showed positive expression of CD44 and vimentin and negative expression of CD34, CD45, CD90, and CD105, indicating rabbit DPSCs (Fig. 1).Fig. 1Rabbit DPSCs surface marker expression. Rabbit DPSCs negatively expressed A CD105, B CD90, C CD45 and D CD34 and positively expressed E CD44 and F vimentin
## Residual periodontal ligament promotes the dentin regeneration potential of DPSCs in the rabbit jaw
Rabbit jawbones were obtained 3 months after the model was established and examined by micro-CT. The results showed complete bone-like healing without any high-density shadows in the blank control group. In the DPSCs group, high-density dentin-like tissue was observed at a small number of sites ($\frac{3}{8}$), while in the DPSCs/PDL group, more sites ($\frac{6}{8}$) and larger volumes of high-density dentin-like tissue were observed. Subsequently, we used CTan software to perform three-dimensional reconstruction of the micro-CT images and volume calculation of the high-density regions of regeneration. The volume of the high-density regions of regeneration in the DPSCs/PDL group (1.07 ± 0.93 cm3) was larger than that in the DPSCs group (0.23 ± 0.41 cm3), and the difference was statistically significant ($$P \leq 0.04$$) (Fig. 2).Fig. 2Micro-CT findings. A Micro-CT showed complete bone-like healing in the blank control group, a small number of sites of high-density dentin-like tissue in the DPSCs group, and larger-volume sites of high-density dentin-like tissue in the DPSCs/PDL group. B The volume of the regions of regenerated high-density tissue was larger in the DPSCs/PDL group than in the DPSCs group. Statistical significance was determined by Student’s t-test. SD is represented by bars. * $P \leq 0.05$ The results of stereomicroscope were similar to those of micro-CT. The blank control group showed complete bone healing without any dentin-like tissue; small areas of dentin-like tissue were observed in the DPSCs group; and larger areas of dentin-like tissue were observed in the DPSCs/PDL group (Fig. 3A). Finally, the microstructure of the regenerated dentin-like tissue was observed by SEM. These results showed that the regenerated dentin-like tissue in the DPSCs and DPSCs/PDL groups contained dentin tubules, confirming that the regenerated tissue was dentin. We also noticed that compared with the regenerated dentin tubules in the DPSCs group, those in the DPSCs/PDL group were more regular and had clearer structures (Fig. 3B).Fig. 3Stereomicroscopy and SEM findings. A *The* general morphology of the tissue was observed by stereomicroscopy. Complete bone healing without any dentin-like tissue was observed in the blank control group; small areas of dentin-like tissue were observed in the DPSCs group; and larger areas of dentin-like tissue were observed in the DPSCs/PDL group. B SEM revealed regenerated dentin-like tissue in the DPSCs group and dentin tubules in addition to regenerated dentin-like tissue in the DPSCs/PDL group H&E staining showed that the regenerated tissue in the blank control group was bone tissue, while the dentin-like tissue observed in the DPSCs and DPSCs/PDL groups were surrounded by odontoblasts. In the DPSCs group, the regenerated dentin structure was chaotic, the dentin tubules were irregular, and there were fewer surrounding odontoblasts. However, the regenerated dentin structure in the DPSCs/PDL group was regular, and the dentin tubules were clear in shape, arranged neatly, and surrounded by a large number of odontoblasts (Fig. 4). To confirm that the regenerated high-density tissue was dentin, we used immunohistochemical staining to detect dentin-specific proteins in the regenerated tissue. The results showed significantly more DMP1- and DSPP-positive cells in the DPSCs/PDL group than in the DPSCs group (DMP1: $$P \leq 0.02$$, DSPP: $$P \leq 0.01$$) (Fig. 5).Fig. 4H&E staining results. The regenerated tissue in the blank control group was bone tissue. The regenerated tissue in the DPSCs group as similar to dentin, with irregular dentin tubules visible. The dentin tubules in the DPSCs/PDL group were arranged regularly and surrounded by odontoblast-like cellsFig. 5Immunohistochemical staining results. A, B DMP1- and C, D DSPP-positive cells were significantly more abundant in the DPSCs/PDL group than in the DPSCs group. Black arrow: DMP1- or DSPP-positive cells. Statistical significance was determined by Student’s t-test. SD is represented by bars. * $P \leq 0.05$
## Identification of differentially expressed genes in DPSCs after coculture with PDLSCs in vitro
Using $P \leq 0.05$ and fold change > 2.0 or fold change < 0.5 as screening criteria, 427 differentially expressed mRNAs were detected; 285 mRNAs were upregulated, and 142 mRNAs were downregulated (see Additional file 2). Forty-one differentially expressed lncRNAs were detected; 26 lncRNAs were upregulated, and 15 lncRNAs were downregulated (see Additional file 3). A total of 411 differentially expressed circRNAs were detected; 224 circRNAs were upregulated, and 187 circRNAs were downregulated (see Additional file 4). Nineteen differentially expressed miRNAs were detected, including 13 upregulated miRNAs and 6 downregulated miRNAs (see Additional file 5). Among the differentially expressed genes, we selected the 4 genes (CCL2, METTL24, CA9, CA12) with the largest fold changes and 4 genes (DKK1, FGF11, RPS6KA1, EDAR) related to dentinogenic differentiation pathways to verify the accuracy and credibility of the microarray results. The real-time RT–PCR results showed that the cocultivation of PDLSCs and DPSCs resulted in the downregulation of CCL2, METTL24 and RPS6KA1 and the upregulation of CA9, CA12, DKK1, FGF11 and EDAR in DPSCs, consistent with the RNA-seq results (Fig. 6 and Additional file 1).Fig. 6Cocultivation with PDLSCs altered DPSCs gene expression levels. A–D Cocultivation of PDLSCs and DPSCs resulted in downregulation of CCL2 and METTL24 and upregulation of CA9 and CA12 in DPSCs. E–H Cocultivation of PDLSCs and DPSCs resulted in upregulation of DKK1, FGF11 and EDAR and downregulation of RPS6KA1 in DPSCs. GAPDH was used as an internal reference. Statistical significance was determined by Student’s t test. SD is represented by bars. * $P \leq 0.05$ The GO functional enrichment analysis of differentially expressed mRNAs, lncRNAs, circRNAs and miRNA target mRNAs was used to investigate function in terms of three aspects, i.e., biological process, cellular component and molecular function. The GO enrichment analysis of differentially expressed mRNAs revealed 198 upregulated GO functions and 80 downregulated GO functions (see Additional file 6). The upregulated GO functions included glycolytic process, myosin filament and inward rectifier potassium channel activity (Fig. 7A). The downregulated GO functions included cholesterol biosynthetic process, extracellular matrix and chemokine activity (Fig. 7B). The GO enrichment analysis of differentially expressed lncRNAs revealed 109 upregulated GO functions and 130 downregulated GO functions (see Additional file 7). The upregulated GO functions included biological regulation, cell and binding (see Additional file 8). The downregulated GO functions included biological adhesion, cell and binding (see Additional file 9). The GO enrichment analysis of differentially expressed circRNAs revealed 111 upregulated GO functions and 161 downregulated GO functions (see Additional file 10). The upregulated GO functions included biological adhesion, cell and binding (see Additional file 11). The downregulated GO functions included biological adhesion, cell and binding (see Additional file 12). The GO enrichment analysis of differentially expressed miRNA target mRNAs revealed 3124 differentially regulated GO functions (see Additional file 13). The differentially regulated GO functions included biological adhesion, cell and antioxidant activity (see Additional file 14).Fig. 7GO functional enrichment analysis of differentially expressed mRNAs. A Top 30 upregulated GO functions of the differentially expressed mRNAs. B Top 30 downregulated GO functions of the differentially expressed mRNAs *Pathway analysis* of differentially expressed mRNAs, lncRNAs, circRNAs and miRNA target mRNAs was performed using the KEGG database. Differentially expressed mRNAs revealed a total of 118 upregulated pathways and 119 downregulated pathways (see Additional file 15). The upregulated pathways included glycolysis/gluconeogenesis, carbon fixation in photosynthetic organisms and carbon metabolism (Fig. 8A). The downregulated pathways included the tumour necrosis factor (TNF) signalling pathway, steroid biosynthesis and chemokine signalling pathways (Fig. 8B). Differentially expressed lncRNAs revealed a total of 46 upregulated pathways and 37 downregulated pathways (see Additional file 16). The upregulated pathways included galactose metabolism, starch and sucrose metabolism and inositol phosphate metabolism (see Additional file 17). The downregulated pathways included the AMPK signalling pathway, the TGF-β signalling pathway and fc gamma R-mediated phagocytosis (see Additional file 18). Differentially expressed circRNAs revealed a total of 102 upregulated pathways and 119 downregulated pathways (see Additional file 19). The upregulated pathways included the apoptosis—fly, HIF-1 signalling and p53 signalling pathways (see Additional file 20). The downregulated pathways included the MAPK signalling pathway—fly, alanine, aspartate and glutamate metabolism and cholinergic synapse (see Additional file 21). Differentially expressed miRNA target mRNAs revealed a total of 238 differentially regulated pathways (see Additional file 22). The differentially regulated pathways included axon guidance, the oestrogen signalling pathway and actin cytoskeleton regulation (see Additional file 23).Fig. 8Pathway analysis of differentially expressed mRNAs using the KEGG database. A Top 20 upregulated pathways of the differentially expressed mRNAs. B Top 20 downregulated pathways of the differentially expressed mRNAs
## Discussion
With the development of stem cell tissue engineering technology, significant breakthroughs in tooth regeneration mediated by odontogenic stem cells have been achieved in recent years, but there are still many problems that need to be resolved in the field of tooth regeneration [28, 29]. In this study, we established a tooth regeneration model in the rabbit jaw in which residual periodontal ligament is retained in the fresh extraction socket. The results show that the residual periodontal ligament microenvironment in the extraction socket can promote the dentin regeneration potential of DPSCs in the rabbit jaw.
The cellular microenvironment supports and maintains the proliferation, differentiation and regeneration potential of MSCs. A large number of studies have clarified the importance of the MSC microenvironment [30, 31]. The cellular microenvironment is also an important factor that determines cell behaviour and tooth morphogenesis. In tooth regeneration research, stem cells, ECM, growth factors and multiple interactions among them determine the formation, development and eruption of teeth [32, 33]. Our results show that the residual periodontal ligament microenvironment in the rabbit jaw is conducive to dentin regeneration by DPSCs in terms of improving the efficiency of DPSCs-mediated dentin regeneration, producing regenerated dentin with a more regular structure, yielding dentin tubules with a more orderly arrangement, and increasing the expression of the dentin-specific proteins DMP1 and DSPP.
Periodontal ligament is a natural connective tissue existing between the tooth root and alveolar bone and is mainly composed of periodontal fibrous ligament, cells and ECM [24, 25]. After a tooth is extracted from the jaw, some periodontal ligament tissue will remain on the inner wall of the extraction socket [34]. After DPSCs are implanted in the extraction socket, the periodontal ligament tissue becomes the odontogenic cellular microenvironment surrounding the implanted MSCs, which regulates the regeneration potential and differentiation of DPSCs. The ECM is a noncellular component of tissue and a highly organized and complex structure composed of structural and functional proteins [35]. Each tissue has unique ECM characteristics, which provide guiding cues for cell differentiation, cell migration, wound healing and immune responses. In short, the ECM determines cell and tissue function. A recent study found that the ECM contains a large number of growth factors and proteins related to MSC differentiation, including VEGF, RUNX2, and BMP2 [23]. In the RNA-seq results of this study, we found 427 differentially expressed protein-coding RNAs; 85 mRNAs were upregulated, and 142 mRNAs were downregulated (see Additional file 2). According to the bioinformatic analysis, we found some of the differentially expressed genes, including MMP13, TIMP4, FBN2, WISP1, TNXB, and ADAMTS3, are related to the ECM. These ECM regulators are widely involved in the composition and conversion of ECM [37, 38]. MMPs are a class of metal ion-dependent proteolytic enzymes, and TIMPs are their inhibitors. The main function of MMPs is to degrade the ECM, and they can also activate growth factors and adhesion molecule enzymes. The MMP/TIMP system is the most important enzyme system regulating the dynamic balance of the ECM, and it plays an important role in the development of inflammation in various tissues, the formation of new blood vessels and the regeneration of tissue. Studies have found that MMPs/TIMPs participate in tooth movement, tissue regeneration and tissue remodelling by maintaining ECM homeostasis [36–38]. Fibrillins (FBNs) are structural components of the ECM that serve to distribute, concentrate and regulate local TGF-β and BMP signals, which regulate numerous cellular activities, including ECM formation and remodelling and tissue regeneration [39, 40]. TGF-β and BMP have been proven to promote the dentinogenic differentiation of DPSCs [41]. A study combining TGF-β and BMP into bioscaffold materials demonstrated that the combination promoted the regeneration of biological tooth roots [42]. WISP1 is a connective tissue growth factor and a target of the wnt/frizzled pathway [43], and the wnt pathway is an important regulatory pathway for dentinogenic MSC differentiation. In addition, the identified ECM-related differentially expressed genes include a variety of genes for collagen, TNXB and ADAMTS-1, which may also be involved in the beneficial effect of residual periodontal ligament on DPSCs differentiation. In conclusion, residual periodontal ligament in the extraction socket may be used as a kind of odontogenic ECM to regulate the dentinogenic differentiation of DPSCs in the jaw through a variety of factors.
To explore the specific mechanism by which residual periodontal ligament promotes the dentin regeneration potential of DPSCs in the jaw, we cocultured PDLSCs isolated from the periodontal ligament with DPSCs. Then, transcriptome sequencing of the DPSCs was performed to determine the differentially expressed gene profile, and GO enrichment and KEGG pathway analyses of the differentially expressed genes were carried out. The GO functions related to the influence of PDLSCs on DPSCs included glycolytic process, ECM and chemokine activity. These findings are consistent with those of our previous analysis, indicating that the ECM in residual periodontal ligament may be used as a microenvironment to regulate the process of DPSCs differentiation.
The differentially regulated pathways are related to the influence of PDLSCs on DPSCs. The upregulated pathways included glycolysis/gluconeogenesis, carbon fixation in photosynthetic organisms and carbon metabolism, while the downregulated pathways included the TNF signalling pathway, steroid biosynthesis and chemokine signalling pathways. TNF-α is a cytokine with pleiotropic biological effects that can affect the differentiation of MSCs [44–46]. In this study, it was found that PDLSCs negatively regulate TNF signalling in DPSCs, potentially promoting the dentinogenic differentiation of DPSCs. Chemokines are small cytokines or signal proteins secreted by cells. MSCs from different sources have the ability to secrete different chemokines and are regulated by chemokines [47]. The differentiation of MSCs is the result of the interaction of multiple signalling pathways. These differentially regulated signalling pathways jointly regulate the dentinogenic differentiation of DPSCs in the jaw. However, much research is still needed to further explore the possible mechanisms.
The microenvironment is involved in determining the differentiation fate of cells. In this study, an odontogenic microenvironment was created by retaining residual periodontal ligament tissue in the extraction socket, and this odontogenic microenvironment was applied in an animal model of tooth regeneration. The results show that the periodontal ligament microenvironment promotes the dentin regeneration potential of DPSCs in the jaw, which provides a theoretical basis for the study of tooth regeneration. However, because the residual periodontal ligament in the extraction socket could not be completely separated for in vitro studies, the in vivo situation cannot be completely simulated in vitro. In the future, other methods are needed to simulate the periodontal ligament microenvironment in vitro to explore not only potential mechanisms in more detail but also ways to enhance their effects.
## Conclusions
In conclusion, our results indicate that residual periodontal ligament can promote the dentin regeneration potential of DPSCs in the jaw. RNA-seq and bioinformatic analysis revealed that ECM, TNF signalling and chemokine signalling pathways may represent key factors and signalling pathways through which residual periodontal ligament promotes the dentin regeneration potential of DPSCs. These discoveries provide new insights for further research on MSC-mediated tooth regeneration.
## Supplementary Information
Additional file 1: Primers for specific genes. Additional file 2: Differentially expressed mRNAs in DPSCs regulated by PDLSCs. Additional file 3: Differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 4: Differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 5: Differentially expressed miRNAs in DPSCs regulated by PDLSCs. Additional file 6: GO functional enrichment analysis of differentially expressed mRNAs in DPSCs regulated by PDLSCs. Additional file 7: GO functional enrichment analysis of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 8: Upregulated GO functions of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 9: Downregulated GO functions of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 10: GO functional enrichment analysis of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 11: Upregulated GO functions of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 12: Downregulated GO functions of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 13: GO functional enrichment analysis of differentially expressed miRNA target mRNAs in DPSCs regulated by PDLSCs. Additional file 14: GO functions of differentially expressed miRNA target mRNAs in DPSCs regulated by PDLSCs. Additional file 15: KEGG enrichment analysis of differentially expressed mRNAs in DPSCs regulated by PDLSCs. Additional file 16: KEGG enrichment analysis of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 17: Upregulated pathways of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 18: Downregulated pathways of differentially expressed lncRNAs in DPSCs regulated by PDLSCs. Additional file 19: KEGG enrichment analysis of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 20: Upregulated pathways of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 21: Downregulated pathways of differentially expressed circRNAs in DPSCs regulated by PDLSCs. Additional file 22: KEGG enrichment analysis of differentially expressed miRNA target mRNAs in DPSCs regulated by PDLSCs. Additional file 23: Pathways of differentially expressed miRNA target mRNAs in DPSCs regulated by PDLSCs.
## References
1. Cooper LF. **The current and future treatment of edentulism**. *J Prosthodont* (2009) **18** 116-122. DOI: 10.1111/j.1532-849X.2009.00441.x
2. Mark AM. **Preventing tooth loss**. *J Am Dent Assoc* (2020) **151** 712. DOI: 10.1016/j.adaj.2020.06.021
3. Donovan TE. **Longevity of the tooth/restoration complex: a review**. *J Calif Dent Assoc* (2006) **34** 122-128. DOI: 10.1080/19424396.2006.12222180
4. Mozaffari MS, Emami G, Khodadadi H, Baban B. **Stem cells and tooth regeneration: prospects for personalized dentistry**. *EPMA J* (2019) **10** 31-42. DOI: 10.1007/s13167-018-0156-4
5. Lee DJ, Lee JM, Kim EJ, Takata T, Abiko Y, Okano T. **Bio-implant as a novel restoration for tooth loss**. *Sci Rep* (2017) **7** 7414. DOI: 10.1038/s41598-017-07819-z
6. Li H, Sun J, Yang H, Han X, Luo X, Liao L. **Recruited CD68(+)CD206(+) macrophages orchestrate graft immune tolerance to prompt xenogeneic-dentin matrix-based tooth root regeneration**. *Bioact Mater* (2021) **6** 1051-1072. DOI: 10.1016/j.bioactmat.2020.09.029
7. Nakahara T. **Potential feasibility of dental stem cells for regenerative therapies: stem cell transplantation and whole-tooth engineering**. *Odontology* (2011) **99** 105-111. DOI: 10.1007/s10266-011-0037-y
8. Sonoyama W, Liu Y, Fang D, Yamaza T, Seo BM, Zhang C. **Mesenchymal stem cell-mediated functional tooth regeneration in swine**. *PLoS ONE* (2006) **1** e79. DOI: 10.1371/journal.pone.0000079
9. Wei F, Song T, Ding G, Xu J, Liu Y, Liu D. **Functional tooth restoration by allogeneic mesenchymal stem cell-based bio-root regeneration in swine**. *Stem Cells Dev* (2013) **22** 1752-1762. DOI: 10.1089/scd.2012.0688
10. Gao ZH, Hu L, Liu GL, Wei FL, Liu Y, Liu ZH. **Bio-root and implant-based restoration as a tooth replacement alternative**. *J Dent Res* (2016) **95** 642-649. DOI: 10.1177/0022034516639260
11. Hu L, Liu Y, Wang S. **Stem cell-based tooth and periodontal regeneration**. *Oral Dis* (2018) **24** 696-705. DOI: 10.1111/odi.12703
12. Wang C, Wang Y, Wang H, Yang H, Cao Y, Xia D. **SFRP2 enhances dental pulp stem cell-mediated dentin regeneration in rabbit jaw**. *Oral Dis* (2021) **27** 1738-1746. DOI: 10.1111/odi.13698
13. Hung CN, Mar K, Chang HC, Chiang YL, Hu HY, Lai CC. **A comparison between adipose tissue and dental pulp as sources of MSCs for tooth regeneration**. *Biomaterials* (2011) **32** 6995-7005. DOI: 10.1016/j.biomaterials.2011.05.086
14. Lane SW, Williams DA, Watt FM. **Modulating the stem cell niche for tissue regeneration**. *Nat Biotechnol* (2014) **32** 795-803. DOI: 10.1038/nbt.2978
15. Wagers AJ. **The stem cell niche in regenerative medicine**. *Cell Stem Cell* (2012) **10** 362-369. DOI: 10.1016/j.stem.2012.02.018
16. Crisan M, Yap S, Casteilla L, Chen CW, Corselli M, Park TS. **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
17. Voog J, Jones DL. **Stem cells and the niche: a dynamic duo**. *Cell Stem Cell* (2010) **6** 103-115. DOI: 10.1016/j.stem.2010.01.011
18. Sui BD, Hu CH, Liu AQ, Zheng CX, Xuan K, Jin Y. **Stem cell-based bone regeneration in diseased microenvironments: challenges and solutions**. *Biomaterials* (2019) **196** 18-30. DOI: 10.1016/j.biomaterials.2017.10.046
19. Sui BD, Hu CH, Zheng CX, Jin Y. **Microenvironmental views on mesenchymal stem cell differentiation in aging**. *J Dent Res* (2016) **95** 1333-1340. DOI: 10.1177/0022034516653589
20. Sui B, Hu C, Liao L, Chen Y, Zhang X, Fu X. **Mesenchymal progenitors in osteopenias of diverse pathologies: differential characteristics in the common shift from osteoblastogenesis to adipogenesis**. *Sci Rep* (2016) **6** 30186. DOI: 10.1038/srep30186
21. Sui BD, Hu CH, Zheng CX, Shuai Y, He XN, Gao PP. **Recipient glycemic micro-environments govern therapeutic effects of mesenchymal stem cell infusion on osteopenia**. *Theranostics* (2017) **7** 1225-1244. DOI: 10.7150/thno.18181
22. Ko KI, Coimbra LS, Tian C, Alblowi J, Kayal RA, Einhorn TA. **Diabetes reduces mesenchymal stem cells in fracture healing through a TNFalpha-mediated mechanism**. *Diabetologia* (2015) **58** 633-642. DOI: 10.1007/s00125-014-3470-y
23. Gao CY, Huang ZH, Jing W, Wei PF, Jin L, Zhang XH. **Directing osteogenic differentiation of BMSCs by cell-secreted decellularized extracellular matrixes from different cell types**. *J Mater Chem B* (2018) **6** 7471-7485. DOI: 10.1039/C8TB01785A
24. Berkovitz BK. **Periodontal ligament: structural and clinical correlates**. *Dent Update* (2004) **31** 46-50. DOI: 10.12968/denu.2004.31.1.46
25. Shuttleworth CA, Smalley JW. **Periodontal ligament**. *Int Rev Connect Tissue Res* (1983) **10** 211-247. DOI: 10.1016/B978-0-12-363710-9.50010-1
26. Hao J, Yang H, Cao Y, Zhang C, Fan Z. **IGFBP5 enhances the dentinogenesis potential of dental pulp stem cells via JNK and ErK signalling pathways**. *J Oral Rehabil* (2020) **47** 1557-1565. DOI: 10.1111/joor.13047
27. Qin Q, Yang H, Zhang C, Han X, Guo J, Fan Z. **lncRNA HHIP-AS1 promotes the osteogenic differentiation potential and inhibits the migration ability of periodontal ligament stem cells**. *Stem Cells Int* (2021) **2021** 5595580. DOI: 10.1155/2021/5595580
28. Cheng N, Wen J, Hitching R, Lei C, Xu C. *Tooth bioengineering and whole tooth regeneration: regenerative approaches in dentistry* (2021)
29. Baranova J, Buchner D, Gotz W, Schulze M, Tobiasch E. **Tooth formation: are the hardest tissues of human body hard to regenerate?**. *Int J Mol Sci* (2020) **21** 4031. DOI: 10.3390/ijms21114031
30. Donnelly H, Salmeron-Sanchez M, Dalby MJ. **Designing stem cell niches for differentiation and self-renewal**. *J R Soc Interface* (2018) **15** 20180388. DOI: 10.1098/rsif.2018.0388
31. Bloom AB, Zaman MH. **Influence of the microenvironment on cell fate determination and migration**. *Physiol Genom* (2014) **46** 309-314. DOI: 10.1152/physiolgenomics.00170.2013
32. Zheng C, Chen J, Liu S, Jin Y. **Stem cell-based bone and dental regeneration: a view of microenvironmental modulation**. *Int J Oral Sci* (2019) **11** 23. DOI: 10.1038/s41368-019-0060-3
33. Zhai Q, Dong Z, Wang W, Li B, Jin Y. **Dental stem cell and dental tissue regeneration**. *Front Med* (2019) **13** 152-159. DOI: 10.1007/s11684-018-0628-x
34. Wang L, Shen H, Zheng W, Tang L, Yang Z, Gao Y. **Characterization of stem cells from alveolar periodontal ligament**. *Tissue Eng Part A* (2011) **17** 1015-1026. DOI: 10.1089/ten.tea.2010.0140
35. Bronckers ALJJ, Lyaruu DM, Wöltgens JHM. **Immunohistochemistry of extracellular matrix proteins during various stages of dentinogenesis**. *Connect Tissue Res* (1989) **22** 691-696. DOI: 10.3109/03008208909114121
36. Leonardi R, Talic NF, Loreto C. **MMP-13 (collagenase 3) immunolocalisation during initial orthodontic tooth movement in rats**. *Acta Histochem* (2007) **109** 215-220. DOI: 10.1016/j.acthis.2007.01.002
37. Lausch E, Keppler R, Hilbert K, Cormier-Daire V, Nikkel S, Nishimura G. **Mutations in MMP9 and MMP13 determine the mode of inheritance and the clinical spectrum of metaphyseal anadysplasia**. *Am J Hum Genet* (2009) **85** 168-178. DOI: 10.1016/j.ajhg.2009.06.014
38. Austin JS, Gordon GM, Fini ME. **Global expression analysis of the MMP–TIMP gene families in corneal epithelial regeneration**. *Investig Ophthalmol Vis Sci* (2005) **46** 2128-2228
39. Olivieri J, Smaldone S, Ramirez F. **Fibrillin assemblies: extracellular determinants of tissue formation and fibrosis**. *Fibrogenesis Tissue Repair* (2010) **3** 24. DOI: 10.1186/1755-1536-3-24
40. Hubmacher D, Wang LW, Mecham RP, Reinhardt DP, Apte SS. **Adamtsl2 deletion results in bronchial fibrillin microfibril accumulation and bronchial epithelial dysplasia—a novel mouse model providing insights into geleophysic dysplasia**. *Dis Model Mech* (2015) **8** 487-499. DOI: 10.1242/dmm.017046
41. Zhou C, Yang G, Chen M, Wang C, He L, Xiang L. **Lhx8 mediated Wnt and TGFbeta pathways in tooth development and regeneration**. *Biomaterials* (2015) **63** 35-46. DOI: 10.1016/j.biomaterials.2015.06.004
42. Chen J, Liao L, Lan T, Zhang Z, Guo W. **Treated dentin matrix-based scaffolds carrying TGF-β1/BMP4 for functional bio-root regeneration**. *Appl Mater Today* (2020) **20** 100742. DOI: 10.1016/j.apmt.2020.100742
43. Mill C, Monk BA, Williams H, Simmonds SJ, Jeremy JY, Johnson JL. **Wnt5a-induced Wnt1-inducible secreted protein-1 suppresses vascular smooth muscle cell apoptosis induced by oxidative stress**. *Arterioscler Thromb Vasc Biol* (2014) **34** 2449-2456. DOI: 10.1161/ATVBAHA.114.303922
44. Yuan J, Wang X, Ma D, Gao H, Zheng D, Zhang J. **Resveratrol rescues TNFalphainduced inhibition of osteogenesis in human periodontal ligament stem cells via the ERK1/2 pathway**. *Mol Med Rep* (2020) **21** 2085-2094. PMID: 32186753
45. Zhang X, Chen Q, Liu J, Fan C, Wei Q, Chen Z. **Parthenolide promotes differentiation of osteoblasts through the Wnt/beta-catenin signaling pathway in inflammatory environments**. *J Interferon Cytokine Res* (2017) **37** 406-414. DOI: 10.1089/jir.2017.0023
46. Liu X, Tan GR, Yu M, Cai X, Zhou Y, Ding H. **The effect of tumour necrosis factor-alpha on periodontal ligament stem cell differentiation and the related signaling pathways**. *Curr Stem Cell Res Ther* (2016) **11** 593-602. DOI: 10.2174/1574888X11666160429122858
47. Kangari P, Talaei-Khozani T, Razeghian-Jahromi I, Razmkhah M. **Mesenchymal stem cells: amazing remedies for bone and cartilage defects**. *Stem Cell Res Ther* (2020) **11** 492. DOI: 10.1186/s13287-020-02001-1
|
---
title: Analysis of nutritional status and influencing factors in patients with thoracoabdominal
aortic dissection receiving 3D printing-assisted stent graft fenestration
authors:
- Yan Zhou
- Ying Xu
- Ying Cai
journal: Journal of Cardiothoracic Surgery
year: 2023
pmcid: PMC10029305
doi: 10.1186/s13019-023-02185-6
license: CC BY 4.0
---
# Analysis of nutritional status and influencing factors in patients with thoracoabdominal aortic dissection receiving 3D printing-assisted stent graft fenestration
## Abstract
### Background
To investigate the nutritional status of patients with aortic dissection (AD) treated with using 3D printing-assisted stent graft fenestration and explore the important factors affecting the nutrition status of patients with different numbers of fenestrations (holes).
### Methods
Ninety-nine hospitalized patients with AD in a grade A tertiary hospital in Nanjing from January 2020 to December 2020 were selected as the study subjects. According to the different number of fenestrations, the patients were divided into four groups: one fenestration (group A), two fenestrations (group B), three fenestrations (group C) and four fenestrations (group D); and the nutrition status of patients in the four groups was analyzed. Then, according to whether the calories provided via infusion reached the $80\%$ goal calories (25 kcal/kg/day) on postoperative day 5, the patients were assigned to the Reached group and Not reached group, and their inflammatory parameters, including white blood cell (WBC) and C-reactive protein (CRP), on postoperative days 1 and 5 were analyzed.
### Results
Compared with patients in group B ($18.8\%$), C ($19.4\%$) and D ($6.7\%$), patients in group A ($48.6\%$) had the highest rate of reaching the nutrition requirement ($80\%$ goal calories). Further, in the Reached group, WBC count and CRP concentration were significantly reduced on postoperative day 5 compared with postoperative day 1, and the proportion of patients with abnormal WBC count was significantly decreased. In contrast, although the CRP concentration on postoperative day 5 in the Not reached group was significantly lower than that on postoperative day 1, no significant changes in WBC count were observed.
### Conclusion
In 3D printing-assisted stent graft fenestration for AD, multiple fenestrations (holes) were associated with a low rate of reaching nutrition requirements, which might be related to imflammation. Therefore, effective nutritional support should be given to patients with multiple fenestrations after operation to improve their nutritional status and prognosis.
## Background
Aortic dissection (AD) is a pathological change that begins with aortic intima rupture, enabling blood to rush through the tear in the inner layer and move along the long axis of the aorta, separating the intima from the media, causing a septum between the true and false lumens of the aorta [1]. Although AD is a rare disease with an incidence of only one in 200,000, aortic rupture has a mortality rate as high as $34.5\%$ [2, 3]. 3D printing-assisted stent graft fenestration are an important approach for aortic lesions involving visceral, subclavian, or carotid arteries. Contrast-enhanced cardiac CT scans from patients were post-processed and obtained the precise spatial data, then transformed and reconstructed into 3D models, finally thoracic aorta models of aortic aneurysm and aortic dissection were precisely printed by 3D printing technology [4]. It enables the anatomical structure and positional relationship between each branch artery and aneurysm visible, increasing the landing zone, and achieving optimized endovascular repair, which is an effective means of treating complex aortic diseases [5, 6].
However, surgery significantly affects the whole body, causing disruptions in the normal nutritional and immune status, which increase the risk of postoperative inflammatory response and infection in patients. Also, hypercatabolic syndrome occurring at the early stage of aortic disease may disbalance the nutritional status of patients, affecting their prognosis [7, 8]. Therefore, proper postoperative nutritional support is necessary, which not only can improve the negative nitrogen balance caused by protein metabolism disorders but also promote the recovery of immune function, improve the inflammatory response, and reduce the incidence of infection [9].
Briefly, nutritional support contributes to improving postoperative recovery and prognosis of patients. Presently, accurate nutritional support therapy is mostly used for patients in ICU. Lack of nutritional evaluation and treatment of complex aortic diseases affects patient outcomes. Therefore, this study investigated the nutritional status of patients after aortic 3D printing-assisted stent graft fenestration and explored the important factors affecting the nutritional status of patients after surgery to provide a basis for more accurate nutritional support therapy and nursing intervention in clinical practice.
## Study subjects
We initially screened the records of 130 patients with AD who were hospitalized in Nanjing Drum Tower Hospital from January 2020 to December 2020. The study inclusion criteria were: [1] Patients diagnosed with type B AD and underwent 3D printing-assisted stent graft fenestration; [2] Hospital stay > 7 days; [3] Age ≤ 90 years, life expectancy > 1 year (based on the hospital annual screening database); [4] had nutritional status (based on the parameters investigated in this study) within normal ranges prior to surgery. The exclusion criteria were: [1] Hospital stay > 2 months or underwent urgent procedures; [2] History of stroke or hemorrhage in organs in the past 3 months; [3] Complicated with immune system diseases, tumors, digestive system disease; [4] Severe heart, liver, kidney and other vital organ lesions. Based on these criteria, a total of 99 patients were eligible for this study. The patient screening process is shown in Fig. 1.Fig. 1Study flow chart of the patient selection process According to different number of fenestrations, the patients were divided into four groups: one fenestration (group A, $$n = 37$$), two fenestrations (group B, $$n = 16$$), three fenestrations (group C, $$n = 31$$) and four fenestrations (group D, $$n = 15$$). Additionally, using the criteria of 25 kcal/kg/day based on the actual body weight of the patients as the goal calories, the patients were divided into a Reached group and a Not reached group depending on whether their calories reached $80\%$ of the goal calories via infusion on postoperative day 5 [10, 11]. This study was approved by the Ethics Committee of Nanjing Drum Tower Hospital, The Affiliated Hospital of Nanjing University Medical School (Approval Number: 2022-239-02) and informed consent was given by all patients.
## Collection of baseline data
The data collection form of this study was designed by the authors after consulting with relevant literature and experts. The form mainly included the general data of patients: age, sex, BMI, economic status, history of diabetes, history of smoking, and history of drinking.
## Collection of clinical variables
Clinical variables included operation duration, intraoperative blood loss, preoperative and postoperative NRS2002 score [12], as well as white blood cell (WBC) and C-reactive protein (CRP) levels on postoperative days 1 and 5.
## Statistical analysis
Statistical software SPSS 22.0 was used for data analysis. Measurement data that conformed to a normal distribution are presented as mean ± standard deviation (SD). The t test was performed for comparison between two groups, and one-way analysis of variance for comparison among multiple groups. Variables with skewed distribution are expressed as M (Q1, Q3), for which the Kruskal–Wallish test was used to compare multiple groups and the LSD test for pairwise comparison. Qualitative data are expressed as rate or constituent ratio, and means between groups were compared using the chi-square test. $P \leq 0.05$ was used to indicate a statistically significant difference.
## Baseline characteristics
Of the 99 patients who underwent 3D printing-assisted stent graft fenestration, there was no significant statistical difference in gender composition, mean age, BMI, underlying diseases (diabetes), economic status and living habits (smoking history, drinking history) between groups A, B, C, D ($P \leq 0.05$; Table 1).Table 1Baseline data of patients with aortic dissection in the four groupsVariableTotal($$n = 99$$)Group A($$n = 37$$)Group B($$n = 16$$)Group C($$n = 31$$)Group D($$n = 15$$)P valueSex, n (%)0.705 Male80 (80.8)31 (83.8)14 (87.5)24 (77.4)11 (73.3) Female19 (19.2)6 (16.2)2 (12.5)7 (22.6)4 (26.7)Age59.7 ± 12.762.2 ± 14.362.0 ± 10.658.7 ± 11.553.5 ± 11.70.127 Body mass index (BMI)24.2 ± 3.324.5 ± 3.525.5 ± 2.723.9 ± 3.122.8 ± 3.70.126Economic status, n (%)0.659 Medical insurance75 (75.8)27 (73.0)14 (87.5)22 (71.0)12 (80.0) Self-pay24 (24.2)10 (27.0)2 (12.5)9 (29.0)3 (20.0)Diabetes, n (%)0.536 No89 (89.9)34 (91.9)13 (81.2)29 (93.5)13 (86.7) Yes10 (10.1)3 (8.1)3 (18.8)2 (6.5)2 (13.3)History of smoking, n (%)0.812 No74 (74.7)26 (70.3)12 (75.0)25 (80.6)11 (73.3) Yes25 (25.3)11 (29.7)4 (25.0)6 (19.4)4 (26.7)History of drinking, n (%)0.118 No75 (75.8)25 (67.6)12 (75.0)28 (90.3)10 (66.7) Yes24 (24.2)12 (32.4)4 (25.0)3 (9.7)5 (33.3)*Measurement data* are expressed as mean ± SD, while qualitative data are expressed as n (%). group A, one fenestration; group B, two fenestrations; group C, three fenestrations; group D, four fenestrations
## Analysis of clinical variables in patients with different numbers of fenestrations
As shown in Table 2, there were significant differences in operation duration and intraoperative blood loss among the four groups ($P \leq 0.05$). Specifically, operation duration in group A (2.76 ± 1.45) was significantly shorter than that of group B (4.36 ± 2.25), group C (5.36 ± 2.39) and group D (5.10 ± 1.53) ($P \leq 0.05$). The intraoperative blood loss of group A (50 [50, 150]) was significantly lower than that of groups C (100 [100, 300]) and D (500 [100, 500]), but no marked difference in intraoperative blood loss was found between groups A and B ($P \leq 0.05$), groups C and D ($P \leq 0.05$). No significant differences were identified among the four groups in preoperative and postoperative NRS scores ($P \leq 0.05$). In terms of nutrition status, patients in group A had a significantly higher rate of reaching the nutritional requirements ($48.6\%$) compared with patients in groups B ($18.8\%$), C ($19.4\%$) and D ($6.7\%$) ($$P \leq 0.009$$), but little difference was found among groups B, C, and D ($P \leq 0.05$).Table 2Relationship between different number of fenestrations and clinical variables of patientsVariablesTotal($$n = 99$$)Group A($$n = 37$$)Group B($$n = 16$$)Group C($$n = 31$$)Group D($$n = 15$$)P valueOperation duration4.24 ± 2.242.76 ± 1.45bcd4.36 ± 2.255.36 ± 2.395.10 ± 1.53 < 0.001Preoperative NRS score, n (%)0.247 046 (46.5)19 (51.4)7 (43.8)13 (41.9)7 (46.7) 130 (30.3)13 (35.1)3 (18.8)9 (29.0)5 (33.3) 213 (13.1)5 (13.5)3 (18.8)5 (16.1)0 (0.0) 35 (5.1)0 (0.0)2 (12.5)1 (3.2)2 (13.3) 45 (5.1)0 (0.0)1 (6.2)3 (9.7)1 (6.7)Postoperative NRS score, n (%)0.159 27 (7.1)5 (13.5)0 (0.0)2 (6.5)0 (0.0) 342 (42.4)18 (48.6)9 (56.2)10 (32.3)5 (33.3) 437 (37.4)13 (35.1)5 (31.2)11 (35.5)8 (53.3) 56 (6.1)1 (2.7)1 (6.2)4 (12.9)0 (0.0) 67 (7.1)0 (0.0)1 (6.2)4 (12.9)2 (13.3)Intraoperative blood loss (ml)100 [50, 300]50 [50, 150]c,d300 [50, 525]100 [100, 300]500 [100, 500]0.002Nutritional status, n (%)0.009 Met nutrition requirement28 (28.3)18 (48.6)b,c,d3 (18.8)6 (19.4)1 (6.7) Failed to meet nutrition requirement71 (71.7)19 (51.4)13 (81.2)25 (80.6)14 (93.3)Operation duration is expressed as mean ± SD, preoperative and postoperative NRS score as n (%), and intraoperative blood loss as M (Q1, Q3). bcd represents a significant difference between group A and the other three groups. group A, one fenestration; group B, two fenestrations; group C, three fenestrations; group D, four fenestrations
## Changes of laboratory parameters after operation in patients with calorie provided reaching or not reaching the nutrition requirement
Of the 99 patients assessed, 28 ($28.3\%$) met the nutrition requirement on postoperative day 5 (Table 2). Additionally, WBC count and CRP concentration on postoperative day 5 in the Reached group were significantly lower than those on postoperative day 1 ($P \leq 0.05$), and the proportion of patients with abnormal WBC count was also significantly lower than that on postoperative day 1. CRP concentration on postoperative day 5 in the Not reached group was significantly lower than that on postoperative day 1 ($P \leq 0.05$), but WBC count was not markedly different from that on postoperative day 1 ($$P \leq 0.236$$). In addition, the Not-reached group showed no significant difference in the number of patients with abnormal WBC counts between postoperative days 1 and 5 ($$P \leq 0.253$$) (Table 3).Table 3Comparison of white blood cells and C-reactive protein on postoperative days 1 and 5VariablesTimeReached group ($$n = 29$$)Not reached group ($$n = 70$$)P valueWBCD111.70 (8.30,16.50)12.25 (8.20,16.15)0.479D510.20 (7.20,13.80)11.15 (8.78,16.15) < 0.001P < 0.0010.236CRPD183.90 (54.30,120.50)110.20 (72.53,165.03) < 0.001D563.70 (26.90,99.30)75.50 (30.90,128.80) < 0.001P < 0.0010.002Abnormal WBC countD1149 (59.60)132 (76.70)0.029D5121 (48.40)6123 (71.50) < 0.001P0.0060.253WBC count is expressed as n (%), and CRP concentration as M (Q1, Q3). WBC, white blood cell; CPR, C-reactive protein; D1, day 1; D5, day 5
## Discussion
3D printing-assisted stent graft fenestration technique is a new attempt developed by clinical workers in recent years to solve the problem of insufficient landing zone. With this technique, the anatomical structure and positional relationship between branch arteries and aneurysm can be visually displayed [12]. The 3D printing method can also reconstruct the affected visceral arteries, ensure the blood supply of target organs, and significantly reduce potential damage to target organs and patient mortality [13–15]. In terms of nutritional status, we found the rate of reaching the nutrition requirement was higher in patients with one fenestration ($48.6\%$), while the rate among the other three groups was not significantly different. Patients with single fenestration accounted for $64.29\%$ of all patients reaching nutrition requirements. Our findings further demonstrated that multiple fenestrations were associated with a low rate of reaching nutrition requirements. Therefore, more attention should be paid to the postoperative nutritional status of patients with multiple fenestrations, and a refined nutritional program is recommended as early as possible to promote the postoperative recovery of these patients. In terms of surgery-related parameters, group A (one fenestration) had the shortest operation duration and the least intraoperative blood loss. Group D (four fenestrations) had the most intraoperative blood loss. However, there was no significant statistical difference in the intraoperative blood loss between groups B, C and D, and the number of fenestrations did not increase positively with the intraoperative blood loss.
Patients undergoing surgery for AD have increased postoperative infection rates due to surgery trauma and decreased autoimmune function [16]. Long-term general anesthesia and gastrointestinal dysfunction can lead to the secretion of a large number of inflammatory factors, triggering systemic inflammatory response syndrome, which is an important cause of multiple organ function damage [17]. Inflammation can disrupt the metabolism of the patients, causing an increase in insulin resistance and reduction in appetite [18, 19], thus, inhibiting intestinal smooth muscle contraction and absorption of nutrients into cells and altering their gastrointestinal functions. WBC and CRP are direct reflections of inflammation and infection, and CRP is also a marker of acute inflammation. Gariballa et al. found that energy intake was significantly lower in patients with higher CRP concentrations [20]. In our study, the WBC count, CRP concentration and the number of patients with abnormal WBC count on postoperative day 5 in the Reached group were significantly lower than those on postoperative day 1, while WBC count and the number of patients with abnormal WBC count in the Not reached group on postoperative day 5 were not significantly different from those on postoperative day 1. These results suggest that in the first 5 days after surgery, the inflammatory response was alleviated in the Reached group but not in the Not reached group. Persistent high inflammatory levels in patients in the Not reached group may have a negative impact on gastrointestinal motility and function, affecting the digestion and absorption of nutrients and exacerbating nutritional intolerance. In summary, failure to reach the nutrition requirement may be related to the inflammatory status, especially persistent and serious inflammatory response, so it is speculated that the inflammatory status is an important factor affecting the nutrition status of patients.
Malnutrition after surgery for AD slows the rate of wound healing, increases the incidence of infection, prolongs hospital stay, increases medical costs, and impedes patient recovery [21]. Early recovery of immune function can reduce postoperative complications, which is of great significance for improving the prognosis of patients undergoing surgery for AD and is key to promoting patient rehabilitation [22, 23]. Additionally, timely and effective nutritional support for patients after aortic fenestration can improve their immune functions and reduce inflammatory stress responses, thus boosting the rate of reaching the nutrition requirement, reducing postoperative complications, and consequently improving treatment outcomes and patient recovery and quality of life.
The study limitations included the unavailability of data, such as dynamic changes in inflammatory factor levels and continuous gastrointestinal function scores, which forced us to select only WBC and CRP for analysis. Further, the single-center retrospective nature of this study and the limited number of patients analyzed, especially for group assessment, which might have led to a certain level of bias; therefore, large cohort and more in-depth studies using prospective settings are still required to confirm the significance of the 3D technique on patients recovery.
## Conclusion
In summary, among patients with AD treated with 3D printing-assisted stent graft fenestration, those with multiple fenestrations have a low rate of reaching nutrition requirements ($80\%$ of goal calories), and in which inflammation might be related to nutritional status. Therefore, we suggest that nutrition status be evaluated and graded for patients with multiple fenestrations after the surgery. Additionally, diversified nutrition regimens should be implemented early to promote the patients’ postoperative recovery.
## References
1. Tian JH, Jing L. **Nursing discussion of blood pressure control in patients with Stanford B aortic dissection**. *Clin Res Pract* (2018.0) **3** 175-176
2. Pichitchaipitak O, Ckumdee S, Apivanich S. **Predictive factors of catheter-related bloodstream infection in patients receiving home parenteral nutrition**. *Nutrition* (2018.0) **46** 1-6. DOI: 10.1016/j.nut.2017.08.002
3. Wanhainen A, Verzini F, Van Herzeele I. **Editor's Choice - European Society for Vascular Surgery (ESVS) 2019 clinical practice guidelines on the management of abdominal aorto-iliac artery aneurysms**. *Eur J Vasc Endovasc Surg* (2019.0) **57** 8-93. DOI: 10.1016/j.ejvs.2018.09.020
4. Sun Y, Wang YT, Dong DN. **Effectiveness of endovascular repair with fenestration stent and branch stent assisted by 3D printing in treating thoracoabdominal aortic aneurysm involving visceral artery area**. *Chin J Clin Thor Cardiovasc Surg* (2022.0) **01** 90-94
5. 5.Surgery EWGoV, Improvement ECoNCffCQ. Chinese experts’ consensus on the evaluation index system of endovascular thoracic aortic aneurysm repair. Chin Circul J. 2018;33:627–530.
6. Dai XC. **Application of preprocedural fenestrated technique during TEVAR for aortic arch disease to preserve supra-aortic branch**. *J Surg Concepts Pract* (2017.0) **22** 290-293
7. Banko D, Rosenthal N, Chung J. **Comparing the risk of bloodstream infections by type of parenteral nutrition preparation method: a large retrospective, observational study**. *Clin Nutr ESPEN* (2019.0) **30** 100-106. DOI: 10.1016/j.clnesp.2019.01.011
8. Lin QX. **Comparative analysis of early enteral nutrition and parenteral nutrition in the prevention of related infections after esophageal cancer surgery**. *China Health Stand Manag* (2018.0) **9** 11-13
9. Reitzel RA, Rosenblatt J, Chaftari AM. **Epidemiology of infectious and noninfectious catheter complications in patients receiving home parenteral nutrition: a systematic review and meta-analysis**. *JPEN J Parenter Enteral Nutr* (2019.0) **43** 832-851. DOI: 10.1002/jpen.1609
10. Singer P, Blaser AR, Berger MM. **ESPEN guideline on clinical nutrition in the intensive care unit**. *Clin Nutr* (2019.0) **38** 48-79. DOI: 10.1016/j.clnu.2018.08.037
11. Warren M, McCarthy MS, Roberts PR. **Practical application of the revised guidelines for the provision and assessment of nutrition support therapy in the adult critically ill patient: a case study approach**. *Nutr Clin Pract* (2016.0) **31** 334-341. DOI: 10.1177/0884533616640451
12. Kondrup J, Allison SP, Elia M. **ESPEN guidelines for nutrition screening 2002**. *Clin Nutr* (2003.0) **22** 415-421. DOI: 10.1016/S0261-5614(03)00098-0
13. Zhu JC, Hu FG, Zhang YW. **Preliminary study of endovascular aneurysm repair using fenestration on table for abdominal aortic aneurysm with short proximal neck**. *Zhonghua Yi Xue Za Zhi* (2016.0) **96** 3642-3646. PMID: 27978899
14. Cross J, Gurusamy K, Gadhvi V. **Fenestrated endovascular aneurysm repair**. *Br J Surg* (2012.0) **99** 152-159. DOI: 10.1002/bjs.7804
15. Oderich GS, Correa MP, Mendes BC. **Technical aspects of repair of juxtarenal abdominal aortic aneurysms using the Zenith fenestrated endovascular stent graft**. *J Vasc Surg* (2014.0) **59** 1456-1461. DOI: 10.1016/j.jvs.2013.10.060
16. Raphael BP, Hazekamp C, Samnaliev M. **Analysis of healthcare institutional costs of pediatric home parenteral nutrition central line infections**. *J Pediatr Gastroenterol Nutr* (2018.0) **67** e77-e81. DOI: 10.1097/MPG.0000000000002058
17. Yi F, Shang YQ, Li B. **Effect of endovascular repair of covered stent and drug conservative treatment on liver and renal function, inflammatory factors and prognosis in patients with stanford B aortic dissecting aneurysm**. *Prog Mod Biomed* (2019.0) **14** 16-18
18. Greenfield JR, Campbell LV. **Relationship between inflammation, insulin resistance and type 2 diabetes: ’cause or effect’?**. *Curr Diabetes Rev* (2006.0) **2** 195-211. DOI: 10.2174/157339906776818532
19. Shoelson SE, Lee J, Goldfine AB. **Inflammation and insulin resistance**. *J Clin Invest* (2006.0) **116** 1793-1801. DOI: 10.1172/JCI29069
20. Gariballa S, Forster S. **Effects of acute-phase response on nutritional status and clinical outcome of hospitalized patients**. *Nutrition* (2006.0) **22** 750-757. DOI: 10.1016/j.nut.2006.04.011
21. Phua AI, Hon KY, Holt A. **Candida catheter-related bloodstream infection in patients on home parenteral nutrition - Rates, risk factors, outcomes, and management**. *Clin Nutr ESPEN* (2019.0) **31** 1-9. DOI: 10.1016/j.clnesp.2019.03.007
22. Zhou SJ, Zhang P, Cai ML. **Postoperative nursing care of patients receiving debranching of abdominal artery and endovascular aortic aneurysm repair (EVAR)**. *J Nurs Sci* (2021.0) **23** 32-34
23. Liu J, Liu Y. **Effect of parenteral nutrition on acute inflammation and infection after aortic dissection operation**. *Chin J Nosocomiol* (2021.0) **10** 1552-1556
|
---
title: 'Screening for the primary prevention of fragility fractures among adults aged
40 years and older in primary care: systematic reviews of the effects and acceptability
of screening and treatment, and the accuracy of risk prediction tools'
authors:
- Michelle Gates
- Jennifer Pillay
- Megan Nuspl
- Aireen Wingert
- Ben Vandermeer
- Lisa Hartling
journal: Systematic Reviews
year: 2023
pmcid: PMC10029308
doi: 10.1186/s13643-023-02181-w
license: CC BY 4.0
---
# Screening for the primary prevention of fragility fractures among adults aged 40 years and older in primary care: systematic reviews of the effects and acceptability of screening and treatment, and the accuracy of risk prediction tools
## Abstract
### Background
To inform recommendations by the Canadian Task Force on Preventive Health Care, we reviewed evidence on the benefits, harms, and acceptability of screening and treatment, and on the accuracy of risk prediction tools for the primary prevention of fragility fractures among adults aged 40 years and older in primary care.
### Methods
For screening effectiveness, accuracy of risk prediction tools, and treatment benefits, our search methods involved integrating studies published up to 2016 from an existing systematic review. Then, to locate more recent studies and any evidence relating to acceptability and treatment harms, we searched online databases (2016 to April 4, 2022 [screening] or to June 1, 2021 [predictive accuracy]; 1995 to June 1, 2021, for acceptability; 2016 to March 2, 2020, for treatment benefits; 2015 to June 24, 2020, for treatment harms), trial registries and gray literature, and hand-searched reviews, guidelines, and the included studies. Two reviewers selected studies, extracted results, and appraised risk of bias, with disagreements resolved by consensus or a third reviewer. The overview of reviews on treatment harms relied on one reviewer, with verification of data by another reviewer to correct errors and omissions. When appropriate, study results were pooled using random effects meta-analysis; otherwise, findings were described narratively. Evidence certainty was rated according to the GRADE approach.
### Results
We included 4 randomized controlled trials (RCTs) and 1 controlled clinical trial (CCT) for the benefits and harms of screening, 1 RCT for comparative benefits and harms of different screening strategies, 32 validation cohort studies for the calibration of risk prediction tools (26 of these reporting on the Fracture Risk Assessment Tool without [i.e., clinical FRAX], or with the inclusion of bone mineral density (BMD) results [i.e., FRAX + BMD]), 27 RCTs for the benefits of treatment, 10 systematic reviews for the harms of treatment, and 12 studies for the acceptability of screening or initiating treatment.
In females aged 65 years and older who are willing to independently complete a mailed fracture risk questionnaire (referred to as “selected population”), 2-step screening using a risk assessment tool with or without measurement of BMD probably (moderate certainty) reduces the risk of hip fractures (3 RCTs and 1 CCT, $$n = 43$$,736, absolute risk reduction [ARD] = 6.2 fewer in 1000, $95\%$ CI 9.0–2.8 fewer, number needed to screen [NNS] = 161) and clinical fragility fractures (3 RCTs, $$n = 42$$,009, ARD = 5.9 fewer in 1000, $95\%$ CI 10.9–0.8 fewer, NNS = 169). It probably does not reduce all-cause mortality (2 RCTs and 1 CCT, $$n = 26$$,511, ARD = no difference in 1000, $95\%$ CI 7.1 fewer to 5.3 more) and may (low certainty) not affect health-related quality of life. Benefits for fracture outcomes were not replicated in an offer-to-screen population where the rate of response to mailed screening questionnaires was low. For females aged 68–80 years, population screening may not reduce the risk of hip fractures (1 RCT, $$n = 34$$,229, ARD = 0.3 fewer in 1000, $95\%$ CI 4.2 fewer to 3.9 more) or clinical fragility fractures (1 RCT, $$n = 34$$,229, ARD = 1.0 fewer in 1000, $95\%$ CI 8.0 fewer to 6.0 more) over 5 years of follow-up. The evidence for serious adverse events among all patients and for all outcomes among males and younger females (<65 years) is very uncertain. We defined overdiagnosis as the identification of high risk in individuals who, if not screened, would never have known that they were at risk and would never have experienced a fragility fracture. This was not directly reported in any of the trials. Estimates using data available in the trials suggest that among “selected” females offered screening, $12\%$ of those meeting age-specific treatment thresholds based on clinical FRAX 10-year hip fracture risk, and $19\%$ of those meeting thresholds based on clinical FRAX 10-year major osteoporotic fracture risk, may be overdiagnosed as being at high risk of fracture. Of those identified as being at high clinical FRAX 10-year hip fracture risk and who were referred for BMD assessment, $24\%$ may be overdiagnosed. One RCT ($$n = 9268$$) provided evidence comparing 1-step to 2-step screening among postmenopausal females, but the evidence from this trial was very uncertain.
For the calibration of risk prediction tools, evidence from three Canadian studies ($$n = 67$$,611) without serious risk of bias concerns indicates that clinical FRAX-Canada may be well calibrated for the 10-year prediction of hip fractures (observed-to-expected fracture ratio [O:E] = 1.13, $95\%$ CI 0.74–1.72, I2 = $89.2\%$), and is probably well calibrated for the 10-year prediction of clinical fragility fractures (O:$E = 1.10$, $95\%$ CI 1.01–1.20, I2 = $50.4\%$), both leading to some underestimation of the observed risk. Data from these same studies ($$n = 61$$,156) showed that FRAX-Canada with BMD may perform poorly to estimate 10-year hip fracture risk (O:$E = 1.31$, $95\%$ CI 0.91-2.13, I2 = $92.7\%$), but is probably well calibrated for the 10-year prediction of clinical fragility fractures, with some underestimation of the observed risk (O:E 1.16, $95\%$ CI 1.12–1.20, I2 = $0\%$). The Canadian Association of Radiologists and Osteoporosis Canada Risk Assessment (CAROC) tool may be well calibrated to predict a category of risk for 10-year clinical fractures (low, moderate, or high risk; 1 study, $$n = 34$$,060). The evidence for most other tools was limited, or in the case of FRAX tools calibrated for countries other than Canada, very uncertain due to serious risk of bias concerns and large inconsistency in findings across studies.
Postmenopausal females in a primary prevention population defined as <$50\%$ prevalence of prior fragility fracture (median $16.9\%$, range 0 to $48\%$ when reported in the trials) and at risk of fragility fracture, treatment with bisphosphonates as a class (median 2 years, range 1–6 years) probably reduces the risk of clinical fragility fractures (19 RCTs, $$n = 22$$,482, ARD = 11.1 fewer in 1000, $95\%$ CI 15.0–6.6 fewer, [number needed to treat for an additional beneficial outcome] NNT = 90), and may reduce the risk of hip fractures (14 RCTs, $$n = 21$$,038, ARD = 2.9 fewer in 1000, $95\%$ CI 4.6–0.9 fewer, NNT = 345) and clinical vertebral fractures (11 RCTs, $$n = 8921$$, ARD = 10.0 fewer in 1000, $95\%$ CI 14.0–3.9 fewer, NNT = 100); it may not reduce all-cause mortality. There is low certainty evidence of little-to-no reduction in hip fractures with any individual bisphosphonate, but all provided evidence of decreased risk of clinical fragility fractures (moderate certainty for alendronate [NNT=68] and zoledronic acid [NNT=50], low certainty for risedronate [NNT=128]) among postmenopausal females. Evidence for an impact on risk of clinical vertebral fractures is very uncertain for alendronate and risedronate; zoledronic acid may reduce the risk of this outcome (4 RCTs, $$n = 2367$$, ARD = 18.7 fewer in 1000, $95\%$ CI 25.6–6.6 fewer, NNT = 54) for postmenopausal females. Denosumab probably reduces the risk of clinical fragility fractures (6 RCTs, $$n = 9473$$, ARD = 9.1 fewer in 1000, $95\%$ CI 12.1–5.6 fewer, NNT = 110) and clinical vertebral fractures (4 RCTs, $$n = 8639$$, ARD = 16.0 fewer in 1000, $95\%$ CI 18.6–12.1 fewer, NNT=62), but may make little-to-no difference in the risk of hip fractures among postmenopausal females. Denosumab probably makes little-to-no difference in the risk of all-cause mortality or health-related quality of life among postmenopausal females. Evidence in males is limited to two trials (1 zoledronic acid, 1 denosumab); in this population, zoledronic acid may make little-to-no difference in the risk of hip or clinical fragility fractures, and evidence for all-cause mortality is very uncertain. The evidence for treatment with denosumab in males is very uncertain for all fracture outcomes (hip, clinical fragility, clinical vertebral) and all-cause mortality.
There is moderate certainty evidence that treatment causes a small number of patients to experience a non-serious adverse event, notably non-serious gastrointestinal events (e.g., abdominal pain, reflux) with alendronate (50 RCTs, $$n = 22$$,549, ARD = 16.3 more in 1000, $95\%$ CI 2.4–31.3 more, [number needed to treat for an additional harmful outcome] NNH = 61) but not with risedronate; influenza-like symptoms with zoledronic acid (5 RCTs, $$n = 10$$,695, ARD = 142.5 more in 1000, $95\%$ CI 105.5–188.5 more, NNH = 7); and non-serious gastrointestinal adverse events (3 RCTs, $$n = 8454$$, ARD = 64.5 more in 1000, $95\%$ CI 26.4–13.3 more, NNH = 16), dermatologic adverse events (3 RCTs, $$n = 8454$$, ARD = 15.6 more in 1000, $95\%$ CI 7.6–27.0 more, NNH = 64), and infections (any severity; 4 RCTs, $$n = 8691$$, ARD = 1.8 more in 1000, $95\%$ CI 0.1–4.0 more, NNH = 556) with denosumab. For serious adverse events overall and specific to stroke and myocardial infarction, treatment with bisphosphonates probably makes little-to-no difference; evidence for other specific serious harms was less certain or not available. There was low certainty evidence for an increased risk for the rare occurrence of atypical femoral fractures (0.06 to 0.08 more in 1000) and osteonecrosis of the jaw (0.22 more in 1000) with bisphosphonates (most evidence for alendronate). The evidence for these rare outcomes and for rebound fractures with denosumab was very uncertain.
Younger (lower risk) females have high willingness to be screened. A minority of postmenopausal females at increased risk for fracture may accept treatment. Further, there is large heterogeneity in the level of risk at which patients may be accepting of initiating treatment, and treatment effects appear to be overestimated.
### Conclusion
An offer of 2-step screening with risk assessment and BMD measurement to selected postmenopausal females with low prevalence of prior fracture probably results in a small reduction in the risk of clinical fragility fracture and hip fracture compared to no screening. These findings were most applicable to the use of clinical FRAX for risk assessment and were not replicated in the offer-to-screen population where the rate of response to mailed screening questionnaires was low. Limited direct evidence on harms of screening were available; using study data to provide estimates, there may be a moderate degree of overdiagnosis of high risk for fracture to consider. The evidence for younger females and males is very limited. The benefits of screening and treatment need to be weighed against the potential for harm; patient views on the acceptability of treatment are highly variable.
### Systematic review registration
International Prospective Register of Systematic Reviews (PROSPERO): CRD42019123767.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13643-023-02181-w.
## Rationale for the systematic reviews
There is no international consensus on the recommended approach to screening and subsequent treatment to prevent fragility fractures [1]. Screening has traditionally focused on measuring bone mineral density (BMD) with intervention in those with low bone mass, often referred to as osteoporosis [2]. More recent evidence suggests that fracture risk prediction may be improved by instead considering an array of clinical risk factors, alone or in addition to BMD, which may be incorporated into risk prediction tools to estimate the absolute short- to mid-term risk of fracture [2].
The 2010 Osteoporosis Canada screening strategy (presence of any of various clinical risk factors) has low sensitivity in identifying females aged 50 to 64 years for BMD testing who later experience a major osteoporotic fracture [3]. In addition, the screening strategy has not been evaluated in a randomized controlled trial (RCT), indicating that updated screening and treatment algorithms that incorporate the most recent evidence are needed. Since 2018, three RCTs have been published that integrate a 2-step approach to screening to prevent fragility fractures (i.e., risk assessment followed by BMD measurement in those exceeding a certain risk threshold, but without shared decision-making) [4–6]. A systematic review published in 2020 [7], after we began this review, reported on the effects of screening from these three trials on fractures and all-cause mortality. The review had slightly different eligibility criteria than ours (thus two studies included in our review are not included), did not address overdiagnosis (defined later), and did not review additional aspects such as alternative screening strategies or patient perspectives related to recommendations about screening in primary care.
Because randomized trials on screening were not anticipated to evaluate all possible screening tools and outcomes (e.g., harms from the treatment provided to those at high risk), we have included reviews on these topics to determine whether certain screening tools may be interchangeable, and whether treatment harms may impact the main screening recommendation.
## Description and burden of the condition
Fragility fractures are those that occur without stimulus during normal daily activities or secondary to minor incidents that in healthy adults would not normally result in a fracture [8]. Major independent risk factors for fragility fracture include low bone density, chronic use of certain medications (e.g., glucocorticoids), older age, female sex, low body weight, a personal or family history of fracture, a history of falls, smoking, higher levels of alcohol use, and living with type 2 diabetes and/or rheumatoid arthritis [9–14]. Advancing age, especially among postmenopausal females and older males [15], and menopausal status [16, 17] are strong predictors of fragility fracture, as is low bone density [18]. A reduction in bone mass and quality is a common consequence of the aging process.
Fragility fractures impose a substantial burden on societies worldwide [19]. By the year 2040, it has been projected that more than 319 million people globally will be considered to be at high risk of fragility fracture (based on the Fracture Risk Assessment Tool without incorporating BMD results [clinical FRAX]) [20]. In Canada in $\frac{2015}{16}$, the incidence of hip fractures among people aged 65 to 69 years was 87 per 100,000 and increased steeply with advancing age to a rate of 1156 per 100,000 in 85 to 89-year-olds [21]. Fragility fractures, particularly hip and clinical vertebral fractures, can result in significant morbidity (e.g., decreased mobility, pain, reduced quality of life) and increase the risk of mortality in the 5 years post-fracture [22–24]. Fragility fractures have been noted to result in more hospitalized days than either stroke or myocardial infarction [25].
## Screening for the primary prevention of fragility fractures
Screening in primary care aims to decrease the risk of future fragility fractures among those without a prior fracture, and to reduce fracture-related morbidity, mortality, and costs. Harms may be related to the screening test itself (e.g., minimal radiation exposure from dual X-ray absorptiometry [DXA]) [26] or the psychosocial or physical (if harmed from treatment) consequences of being labelled “at risk” [27, 28], which may be due to an inaccurate estimation of fracture risk (i.e., due to a risk prediction tool that is poorly calibrated), and/or detection of excess risk among people who, had they not been screened, would never have known their risk nor experienced a fracture. Though considered by the Task Force to be the ideal approach, shared decision-making for screening and subsequent treatment may not be the standard of care across Canada; many primary care providers may instead screen all people without a prior fracture for risk (e.g., using available risk prediction tools and/or offer of BMD assessment) and consider patients eligible for treatment when screening places them within pre-specified thresholds of BMD or fracture risk. It may instead be ideal to use shared decision-making during the clinical encounter, allowing patients to make informed decisions about screening and treatment after weighing the possible benefits against the potential harms. Information from screening can then be used, along with patient preferences, to consider preventive treatment among those who consider themselves to be at a high fracture risk.
There is large variation in the screening approaches suggested by international guidelines, which often consider the population burden of fragility fractures and mortality, competing societal priorities, and resource availability [1, 29]. A variety of approaches may be used within a single screening program, with recommendations often differing by population group based on age, sex, or menopausal status [1, 29]. Common approaches include (a) a one-step direct to BMD approach (e.g., in females >65 years old in Canada [30] and the USA [31]); and (b) a 2-step approach incorporating the assessment of absolute fracture risk followed by BMD assessment in individuals exceeding a pre-defined threshold [29]. The findings of BMD assessment may then be used independently or incorporated into revised clinical risk scores. Clinical risk factors alone may be used to estimate risk in circumstances where BMD is unavailable, but this is not recommended by current North American guidelines [30–35]. There are at least 12 published fracture risk prediction tools available [36, 37]; however, not all tools are easily accessible to clinicians nor have all tools been calibrated for Canada or validated in populations outside of their derivation cohort, limiting their use [38].
Treatment thresholds vary considerably across countries [1, 29, 39]. A common threshold for treatment used in Canada [30, 40], the USA [41], and several other countries is a fixed 10-year major osteoporosis-related fracture probability ≥$20\%$ [39]. In some countries (not Canada), a 10-year hip fracture probability ≥$3\%$ may also be used [39]. Other approaches include the use of variable thresholds based on age [39], and hybrid models that incorporate both age-based and fixed thresholds [42–44]. Few existing guidelines incorporate shared decision-making [45, 46], but ideally this could be applied to determine the point at which an individual patient, informed about the benefits and risks, would want to contemplate treatment. Bisphosphonates (i.e., alendronate, risedronate, or zoledronic acid) are the most commonly used first-line treatments for the prevention of fragility fractures [47, 48]. Denosumab may also sometimes be considered [47, 48], but this is less common due to its higher cost compared to bisphosphonates. Changing lifestyle factors (e.g., diet, exercise) and fall prevention are other approaches to preventing fragility fractures [30] but were not in the scope of these systematic reviews.
According to a systematic review commissioned by the United States Preventive Services Task Force (USPSTF) with a comprehensive search in 2016, compared to placebo, treatment with bisphosphonates probably reduces the risk of nonvertebral and vertebral fractures (moderate certainty), but may make little-to-no difference in the risk of hip fractures (low certainty) in females [37]. There was low certainty evidence for reduction in all fracture types with denosumab in females [37]. Evidence for males was limited across all pharmacologic treatments of interest [37]. The review authors did not rate the certainty for all clinical fractures, as is of interest for the current review, and updating the evidence may change findings for some outcomes. Various harms may be associated with treatment to various degrees, with some such as mild upper gastrointestinal distress being fairly benign. Others such as serious infections or cardiac events, osteonecrosis of the jaw, and atypical femoral fractures are potentially highly concerning [49].
The effectiveness of treatment relies on high uptake and adherence [50]. However, uptake of pharmacologic treatment is often low, and adherence tends to diminish over time [51]. Low uptake and adherence may be related to a variable assessment of the balance of benefits and harms by individual patients. Though shared decision-making is incorporated into few existing screening guidelines [45, 46], a large variation in treatment preferences across patients could support a shared decision-making approach in the place of recommended treatment thresholds based solely on fracture risk [52, 53].
## Objectives of systematic reviews
In these reviews, we have synthesized evidence relevant to screening for the primary prevention of fragility fractures and related mortality and morbidity among adults 40 years and older in primary care. The findings are among several considerations (including consultations with patients on outcome prioritization, information on issues of feasibility, acceptability, costs/resources, and equity) that will be used by the Canadian Task Force on Preventive Health Care (“Task Force”) to inform recommendations on screening for the prevention of fragility fractures among adults 40 years and older in Canada. Our key questions (KQs) were as follows: KQ1a: What are the benefits and harms of screening compared with no screening to prevent fragility fractures and related morbidity and mortality in primary care for adults ≥40 years?
KQ1b: Does the effectiveness of screening to prevent fragility fractures vary by screening program type (i.e., 1-step vs 2-step) or risk assessment tool?
KQ2: How accurate are screening tests at predicting fractures among adults ≥40 years?
KQ3a: What are the benefits of pharmacologic treatments to prevent fragility fractures among adults ≥40 years?
KQ3b: What are the harms of pharmacologic treatments to prevent fragility fractures among adults ≥40 years?
KQ4: For patients ≥40 years, what is the acceptability (i.e., positive attitudes, intentions, willingness, uptake) of screening and/or initiating treatment to prevent fragility fractures when considering the possible benefits and harms from screening and/or treatment?
Screening and treatment for risk factors related to fractures, such as fall risk, were not considered though the Task *Force is* currently developing separate recommendations about falls prevention interventions [54].
## Terminology
Throughout this report, we refer to “females” and “males”; these terms refer to biological sex (i.e., biological attributes, particularly the reproductive or sexual anatomy at birth) unless otherwise indicated.
## Review conduct
We followed a peer-reviewed protocol [55] for this review which was based on accepted systematic review methodology [56]. The review was registered prospectively in the International Prospective Register of Systematic Reviews (PROSPERO): CRD42019123767. The methods for the systematic review are reported in detail within the protocol [55]; we report on the methods here briefly, focusing on deviations from the original plans. We report the systematic review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 statement [57].
At the protocol stage, members of the Task Force rated outcomes on their importance for clinical decision-making using a 9-point scale according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [58]. In addition, the findings of surveys and focus groups with patients that were conducted by the Knowledge Translation team at St. Michael’s, Unity Health Toronto, were incorporated into the final outcome ratings. Outcomes rated as critical (7–$\frac{9}{9}$) were hip fracture, clinical fragility fractures, fracture-related mortality, quality of life or wellbeing, functionality and disability, serious adverse events, and prediction model calibration (KQ2 only). Outcomes rated as important (4–$\frac{6}{9}$) were all-cause mortality, non-serious adverse events, discontinuation due to adverse events, and overdiagnosis. The outcomes are defined in detail within our protocol [55]. As screening for risk of fracture does not result in a “diagnosis” but rather a risk for a future event, overdiagnosis has not been previously defined in the context of fracture risk assessment. However, as with conditions such as osteoporosis, overdiagnosis generally refers to identifying and labelling people with “problems,” or in this case “risks,” that would never have caused harm [59]. Thus, for the purpose of this review, we defined overdiagnosis as the identification of high risk in individuals who, if not screened, would never have known that they were at risk and would never have experienced a fragility fracture [59]. The systematic review protocol and this report were revised following review by external stakeholders ($$n = 7$$ and $$n = 4$$, respectively). The Task Force and their external clinical experts were involved with developing the scope of the review and the eligibility criteria ($$n = 4$$; see “Acknowledgments”), as well as with interpreting the findings ($$n = 2$$), but were not involved in the selection and risk of bias assessments of studies, data extraction, or analysis.
We reviewed the evidence following a staged approach, beginning by identifying direct evidence from trials (including all controlled trials but prioritizing evidence from RCTs) of primary screening versus no screening (KQ1a). Based on positive evidence from KQ1a, we proceeded to KQ1b, examining the comparative effectiveness of different screening approaches. We reviewed evidence related to the acceptability of screening and/or treatment (KQ4), as well as indirect evidence on the accuracy of screening tests (KQ2), concurrently with KQ1. The accuracy of screening tests was reviewed to better understand whether other well calibrated tools existed outside of those used in the screening trials, which could influence the tool ultimately recommended for screening. Because the Task Force believed that further information on the benefits and harms of pharmacologic treatment could be relevant to their recommendations, we proceeded with KQs 3a (benefits) and 3b (harms). After completing KQ3a on the benefits of treatment, discussions with the Task Force indicated that a rapid overview of reviews approach for KQ3b (harms of treatment) would be adequate to inform decision-making, while reducing the time and resources needed to review the evidence. We therefore amended our planned approach to KQ3b, as described herein.
## Eligibility criteria
Detailed PICOTs for each KQ are shown in Table 1. Here, we report changes from our original plans that occurred during the selection phase. For KQ1 (benefits and harms of screening), we had intended to exclude studies of patients already being treated with anti-fracture drugs and/or with prior fractures at baseline, but some relevant trials included unknown proportions of previously treated and/or fractured patients. The comparator of interest was no screening, but in reality the available trials included some degree of ad hoc screening in the comparison group. We considered these factors within the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) indirectness domain. Table 1Eligibility criteria for each key questionKey question 1 a & b(a) Benefits and harms of screening vs no screening(b) Comparative benefits and harms of different screening approaches/toolsKey question 2Predictive accuracy of screening testsKey question 3 a & b(a) Benefits of pharmacologic treatments(b) Harms of pharmacologic treatmentsKey question 4Acceptability of screening and/or treatmentPopulationIncludeAsymptomatic adults ≥40 years in the general population (≥$80\%$ of the sample or mean age -1 standard deviation is ≥40 years)Subgroups for decision-making: age, sex, menopausal statusMethods subgroups: diabetes, presence of prior fractures, baseline predicted fracture risk, length of follow-upAsymptomatic adults ≥40 years in the general populationSubgroups for decision-making: age, sex, menopausal statusMethods subgroups: treatment with anti-osteoporosis drugs, baseline predicted fracture risk, length of follow-upKQ3a: Adults ≥40 years in the general population who are at risk of fragility fractureKQ3b: Adults ≥40 years who are at risk of fragility fractureSubgroups for decision-making: age, sex, menopausal statusMethods subgroups (KQ3a): prior fracture, predicted fracture risk, length of follow-upAdults aged ≥40 yearsPopulation subgroups: absolute fracture risk (perceived or actual), prior screening, history of fracture, prior use of anti-osteoporotic medication, prior diagnosis of osteoporosis, level of concern or perceived severity of fracturesExclude- Adults <40 years- Treatment with anti-osteoporosis drugs- >$50\%$ with prior diagnosis of osteoporosis, prior fragility fracture, endocrine or other disorders related to metabolic bone disease, chronic use of glucocorticoid medications, cancer- Adults <40 years- >$50\%$ with prior diagnosis of osteoporosis, prior fragility fracture, endocrine or other disorders related to metabolic bone disease, chronic use of glucocorticoid medications, cancerKQ3a:- Adults <40 years- > $50\%$ with with prior fragility fracture, endocrine or other disorders related to metabolic bone disease, chronic use of glucocorticoid medications, cancerKQ3b:- Adults <40 years- Endocrine or other disorders related to metabolic bone disease, cancer- Adults <40 years- Current use of anti-osteoporosis drugs (>$10\%$ of population)- >$50\%$ with prior fragility fracture, endocrine or other disorders related to metabolic bone disease, chronic use of glucocorticoid medications, cancerIntervention/ ExposureIncludeScreeninga to prevent fragility fracture with any of the following:-Fracture risk assessment alone (validated or non-validated tools)-Bone mineral density (BMD) alone by dual x-ray absorptiometry ± vertebral fracture assessment (VFA)/spinal radiography - Fracture risk assessment followed by BMD if indicated ± vertebral fracture assessment/spinal radiographyTreatment is offered for participants meeting “high risk” thresholdScreening tool to prevent fragility fracture using any of the following:- Fracture risk assessment alone (validated or nonvalidated tools)- Bone mineral density (BMD) alone by dual X-ray absorptiometry ± vertebral fracture assessment/spinal radiography- Fracture risk assessment followed by/incorporating BMD ± vertebral fracture assessment (VFA)/spinal radiographyRisk assessment tools must be available to clinicians and have been externally validated to predict fragility fractures in a population within a very high human development index country with a fracture rate similar to Canada (i.e., moderate)Pharmacotherapy currently approved by Health Canada for the treatment of osteoporosis or prevention of fragility fractures that is commonly used in Canada as a first-line treatment:- Bisphosphonates (alendronate, risedronate, zoledronic acid); harms of bisphosphonates as a class will be included when no or very low certainty evidence is available for individual bisphosphonates- Denosumab (exposure is discontinuation of denosumab for rebound fracture outcome)*Adjunct calcium* and/or vitamin D (but not other drugs) will be included if it is used identically in both the intervention and comparison groupPopulation may or may not have knowledge of their own fracture risk but must have at least some general scenario or background information on the possible magnitude of benefits and/or harms from screening (same tools as KQ1) or treatment (bisphosphonates or denosumab) for fragility fractures or osteoporosis. ORInvestigators solicit the magnitude of benefits and/or harms where screening or treatment is acceptable. Exposure subgroups: different presentation of informationExclude- Other screening tests- VFA without BMD- Tools not externally validated- Tools not available to clinicians- Tools that do not provide absolute fracture risk (CAROC retained due to relevance to Canada)- Other countries (see inclusion)- Other BMD or osteoporosis-related screening tests- Pharmacotherapies not commonly used in Canada: hormone therapy, etidronate, raloxifene, teriparatide, calcitonin- Drugs used in combination- Off-label pharmaceuticals and dosages- Natural health products, dietary supplements (e.g., vitamins, minerals)- Complex interventions (e.g., pharmacotherapy + exercise)- Context of screening using other BMD or osteoporosis screening tests- Benefit and harm information about treatmentsComparatorIncludeKQ1a: no screeningKQ1b: another screening strategy or screening using a different risk assessment toolNot applicableKQ3a: PlaceboKQ3b: Placebo or no treatment; continuation of denosumab for rebound fracture outcome− None- Non-active exposure: intervention without information about the possible magnitude of benefits and/or harms of screening or treatment- Information on alternative screening or treatment strategyExclude- Other screening tests- Fracture liaison servicesNot applicableAll other comparatorsSee exposureOutcomeIncludeBenefits: hip fractures, clinical fragility fracturesb, fracture-related mortality, functionality and disability, quality of life or wellbeing, all-cause mortalityHarms: serious adverse eventsc (all serious cardiovascular events, serious cardiac rhythm disturbances, serious gastrointestinal adverse events, gastrointestinal cancer, atypical fractures, osteonecrosis of the jaw, overdiagnosis (defined in Additional file 3), discontinuation due to adverse events, non-serious adverse events (including “any” adverse events)Calibration (total/average and by differing estimated risks) for 5- and 10-year risk of hip and clinical fragility fracturesKQ3a: hip fractures, clinical fragility fracturesb, fracture-related mortality, functionality and disability, quality of life or wellbeing, all-cause mortalityKQ3b: serious adverse eventsc (all serious cardiovascular events, serious cardiac rhythm disturbances, serious gastrointestinal adverse events, gastrointestinal cancer, atypical fractures, osteonecrosis of the jaw, rebound fractures i.e. multiple vertebral fractures, discontinuation due to adverse events, non-serious adverse events (including “any” adverse events; non-serious gastrointestinal adverse events, musculoskeletal pain, dermatologic adverse events, infections)- Willingness or intentions to screen or initiate treatment- Acceptability of screening or initiating treatment- Uptake of screening or treatment- Absolute risk for fracture to make treatment acceptable- Others as suitable, as reported by authorsExcludeAll other outcomesDiscrimination (we will supplement with findings from 2018 USPSTF systematic review)All other outcomesAll other outcomesFollow-upInclude≥6 monthsAny length of follow-up; to make predictions for 5- or 10-year fracture≥6 monthsAnyExclude<6 monthsNot applicable<6 monthsNot applicableSettingIncludePrimary health carePrimary health careKQ3a: primary health careKQ3b: primary health care or long-term carePrimary health careExcludeLong-term careLong-term careKQ3a: long-term careKQ3b: all other settingsLong-term care or hospitalStudy designIncludeRandomized controlled trials, clinical controlled trials (if needed)d; manuscripts, reports, abstracts, dissertations, clinical trials registers if data are availableProspective or retrospective cohort studies; single arms of randomized trials; manuscripts, reports, abstracts, dissertations, clinical trials registers if data are availableKQ3a: randomized controlled trials; manuscripts, reports, abstracts, dissertations, clinical trials registers if data are availableKQ3b: systematic reviews of randomized trials or observational studiese; primary studies for rebound fractures after denosumab discontinuationAny quantitative primary study design, quantitative data from mixed methods studiesExcludeSystematic reviews, meta-analyses and pooled analyses; all other primary study designs; non-research; studies only available as gray literature if data are inadequate to assess study design and risk of biasSystematic reviews, meta-analyses and pooled analyses; all other primary study designs; non-research; studies only available as gray literature if data are inadequate to assess study design and risk of biasKQ3a: Systematic reviews, meta-analyses and pooled analyses; all other primary study designs; non-research; studies only available as gray literature if data are inadequate to assess study design and risk of biasKQ3b: primary research, overviews of reviewsSystematic reviews, meta-analyses and pooled analyses; qualitative studies; non-research; studies only available as gray literature if data are inadequate to assess study design and risk of biasLanguageIncludeEnglish or FrenchEnglish or FrenchEnglish or FrenchEnglish or FrenchExcludeAll other languagesAll other languagesAll other languagesAll other languagesDate of publicationIncludeAnyAnyKQ3a: anyKQ3b: 2015–present; 2020 for primary studies of rebound fractures1995–present (introduction of bisphosphonates)ExcludeNot applicableNot applicableKQ3a: not applicableKQ3b: pre-2015 (with exception of previously identified AHRQ review)Pre-1995AHRQ Agency for Healthcare Research and Quality, BMD Bone mineral density, USPSTF United States Preventive Services Task Force, VFA Vertebral fracture assessmenta Screening includes the intervention, follow-up, referral and/or treatment. Fracture risk assessment tools are considered to be any paper or electronic tool or set of questions using ≥2 demographic and/or clinical risk factors to assess risk of future fractureb Clinical fragility fractures include only symptomatic and radiologically confirmed fractures; sites per author definition, and may be defined as major osteoporotic fracturec A serious adverse event is any untoward medical occurrence that at any dose (a) results in death, (b) is life-threatening, (c) requires inpatient hospitalization or prolongation of existing hospitalization, (d) results in persistent or significant disability/incapacity, (e) is a congenital anomaly/birth defect, (f) is a medically important event or reactiond If certainty in the evidence is a barrier to the development of recommendations, and the CTFPHC believes that further evidence from CCTs may influence their recommendationse We will select one systematic review for each outcome comparison of interest For KQ2 (predictive accuracy of screening tests), based on clinical expert input, we decided to exclude tools that (a) are not freely available for use by clinicians or (b) do not provide an absolute risk prediction (e.g., provide only a risk categorization; Canadian Association of Radiologists and Osteoporosis Canada Risk Assessment [CAROC] tool retained due to relevance to Canada). We also considered external validations of FRAX-Canada to be most relevant, in comparison to FRAX tools calibrated for other countries. Though our original eligibility included studies from multiple countries, because of the applicability of Canadian studies (when tools are calibrated to this population) and those from Canada in our original search (in 2019) were among the highest quality, our search update in 2021 focused on finding new Canadian studies for which we limited our inclusion. Though not a deviation from our protocol, it is important to note that the discriminative ability of risk prediction tools was not rated as a critical or important outcome by the Task Force. For this reason, we did not review this information systematically within KQ2, but included data reported in a 2018 USPSTF review [60] within our GRADE Summary of Findings Tables for information purposes.
For KQ3a (benefits of treatment), we had planned to exclude the 5 mg/day dosage of alendronate but later included it as well as mixed doses (e.g., 5 mg followed by 10 mg) based on clinical expert input. This decision was supported by the apparent uncertainty about the superiority of the 10mg/day dose and the likelihood of some variability in the doses used in practice. For KQ3b (harms of treatment), we relied on systematic reviews published since 2015 rather than primary studies, as originally intended (see Review Conduct). We included the one most appropriate systematic review per outcome comparison by considering comprehensiveness (likelihood that the search captured all relevant studies, informed by domain 2 in the Risk Of Bias In Systematic reviews [ROBIS] tool [61]); recency (date of last search); and other relevant features (e.g., availability of subgroup and/or adjusted analyses; availability of absolute event rates for the pooled effect). We included systematic reviews of bisphosphonates as a class only for serious adverse events where findings were very uncertain for individual drugs (i.e., additional data may be useful). For rebound fractures (i.e., fractures resulting from increased bone turnover and reductions in BMD after stopping treatment) from denosumab, we compared discontinuation of denosumab to persistence of denosumab or discontinuation of placebo, based on Task Force input about this being the most relevant available comparison. We also added “multiple vertebral fractures” as the most valid potential outcome to capture the effects of rebound fractures. Further, because the reviews were limited on reporting rebound fractures, we added a search for recent (2020 onwards) primary studies for this outcome. For non-serious adverse events, we included: non-serious gastrointestinal adverse events, musculoskeletal pain, dermatologic adverse events, and infections. There were no changes to the original eligibility criteria for KQ4 (acceptability of screening/treatment).
## Literature search and selection of studies
The approach and dates used to search for and select studies for inclusion in the systematic reviews for each KQ are shown in Table 2. Briefly, for KQs 1 (benefits and harms of screening), 2 (predictive accuracy of screening tests), and 3a (benefits of treatment), we integrated eligible studies published up to 2016 from an existing systematic review by the USPSTF [60]. Due to differences in eligibility criteria, we also checked the USPSTF’s excluded studies list and the reference lists of other systematic reviews and major guidelines to identify studies published before 2016 that would have been excluded from the USPSTF review but met our inclusion criteria (e.g., studies that the USPSTF judged to have serious risk of bias concerns, and those examining the comparative effectiveness of screening approaches). We did not integrate studies from existing reviews for KQs 3b (harms of treatment) or 4 (acceptability) and instead relied solely on our search strategies. Table 2Approach to search and selection of studies for each key questionKey question 1 a & b(a) Benefits and harms of screening vs no screening(b) Comparative benefits and harms of screening approaches/ toolsKey question 2Predictive accuracy of screening testsKey question 3 a & b(a) Benefits of pharmacologic treatments(b) Harms of pharmacologic treatmentsKey question 4Acceptability of screening and/or treatmentSearchApproachIntegration of eligible studies from 2018 USPSTF review, with search from 2016 onwardIntegration of eligible studies from 2018 USPSTF review, with search from 2016 onwardKQ3a: integration of eligible studies from 2018 USPSTF review, with search update from 2016 onwardKQ3b: (i) search for reviews from 2015 onward; (ii) search for primary studies on rebound fractures from discontinuation of denosumab from 2020Search from 1995 (introduction of bisphosphonates)Search StrategySee protocolSee protocolKQ3a: see protocolKQ3b: Additional file 2See protocolConceptsBone health, fractures, osteoporosis, screeningBone health, fractures, osteoporosis, screening tests, risk assessment, prognosisOsteoporosis, drug treatment, harms (in general and specific harms; KQ3b only)Bone health, fractures, screening, risk assessment, osteoporosis - diagnosis, prevention, treatment, risk, decision-making, acceptabilityDatabasesOvid Medline, Ovid Embase, Wiley Cochrane LibraryOvid Medline, Ovid Embase, Wiley Cochrane LibraryOvid Medline, Ovid Embase (KQ3a only), Wiley Cochrane LibraryOvid Medline, Ovid Embase, Wiley Cochrane Library, PsycINFOAdditional sourcesClinicaltrials.gov; WHO ICTRP; scanned reference lists of systematic reviews, included studies, major guidelines; contacting authorsClinicaltrials.gov; WHO ICTRP; scanned reference lists of systematic reviews, included studies, major guidelines; contacting authorsKQ3a: Clinicaltrials.gov; WHO ICTRP; scanned reference lists of systematic reviews, included studies, major guidelines; contacting authorsKQ3b: PROSPERO, EpistemonikosScanned reference lists of systematic reviews, included studies; contacting authorsDate of search(es)2016 to 8 July 2019Update in databases 4 April 20222016 to 5 July 2019Update in databases 1 June 2021KQ3a: 2016 to 2 March 2020KQ3b: (i) 2015 to 24 June 2020; (ii) 2020 to 18 June 20211995 to 4 July 2019Update in databases 1 June 2021SelectionTitles and abstractsLiberal-accelerated (i.e., only requiring one reviewer to include to full text review)Liberal-acceleratedKQ3a: liberal-acceleratedKQ3b: single reviewerLiberal-acceleratedFull TextDuplicateDuplicateKQ3a: duplicateKQ3b: single reviewerDuplicateFinal inclusionBy consensus, with consultation of a third reviewer if neededBy consensus, with consultation of a third reviewer if neededBy consensus, with consultation of a third reviewer if neededSingle reviewer, with consultation of a second reviewer if neededKQ Key question, WHO ICTRP World Health Organization International Clinical Trials Registry Platform, USPSTF United States Preventive Services Task Force A research librarian developed and implemented comprehensive peer-reviewed [62] electronic search strategies for each KQ (see protocol [55]; Additional file 2 for KQ3b on harms of treatment). We also searched clinical trials registries and scanned the reference lists of relevant systematic reviews and the included studies. We exported the database results to EndNote (version X7 or X9, Clarivate Analytics, Philadelphia, PA) and removed duplicates before screening the records in DistillerSR (Evidence Partners Inc., Ottawa, Canada).
## Data extraction and risk of bias assessment
We had initially planned to rely (with verification) on data from the USPSTF systematic review [60] for older studies. However, during review conduct differences in outcome definitions, subgroups of interest, and methodology (e.g., updated version of the PROBAST tool became available) became apparent. Therefore, following a pilot round (with two reviewers), one reviewer independently extracted data from all included studies into a standardized form in Excel (Microsoft Corporation, Redmond, WA). Study characteristics were then verified by a second reviewer and outcome data were extracted in duplicate, with final data based on consensus. The full list of data extraction items is available in our protocol [55]. Since we altered our approach to rely on systematic reviews for KQ3b (harms of treatment), we additionally collected the following: databases searched and date of last search, scope of systematic review and selection criteria for the included studies, number and design of primary studies included, number of participants and summary characteristics, summary of interventions and comparators included, risk of bias/quality appraisal tool used to appraise included studies, analyses methods, and summary statistics for outcomes of interest.
Outcome-level risk of bias was appraised for each included study in duplicate (one reviewer with verification for KQ3b [harms of treatment]) using published design-specific tools as applicable (Cochrane risk of bias tool version 2011 for KQs 1 and 3a [63], PROBAST for KQ2 [64], AMSTAR 2 for KQ3b [65]), with final ratings determined by consensus. For KQ3b (harms of treatment), we also extracted information on the risk of bias of the systematic reviews’ included studies, but if these were missing, we did not perform these appraisals anew. Since there is no commonly used or accepted tool to assess risk of bias in studies of acceptability, we assessed risk of bias in the studies included for KQ4 (acceptability of screening/treatment) by considering the risk of bias subdomains within the GRADE guidance for assessing the certainty of evidence in studies of the importance of outcomes or patient values and preferences [66] (adapted to be suitable to acceptability). Assessments of risk of bias informed the study limitations domain of our assessments of the certainty of the body of evidence.
## Synthesis
We performed meta-analyses when appropriate based on clinical and methodological similarity across studies. For KQ1 (benefits and harms of screening), we pooled data for each outcome via pairwise meta-analysis using the DerSimonian and Laird random effects model [67] in Review Manager (version 5.3, The Cochrane Collaboration, Copenhagen, Denmark). Due to differences in the populations analyzed across studies, we pooled data from different population perspectives separately. The perspectives analyzed were (a) offer-to-screen, which included all those randomized and offered (by mail), but not necessarily completing any screening, and in the group they were originally assigned; (b) offer-to-screen in selected populations, which included those who independently completed a mailed clinical Fracture Risk Assessment (FRAX) questionnaire, in the group they were originally assigned (randomized before or after completion, depending on the trial); and (c) acceptors, which included those randomized who ultimately completed the entire screening process (i.e., clinical FRAX and BMD if meeting the risk threshold). In one study [68], hip fractures were presented only as counts (rather than number of participants with ≥1 fracture); we included this study among the others in meta-analysis based on clinical and statistical expert input indicating that the outcome was sufficiently rare that count and rate data would be similar. As described previously, we defined overdiagnosis as the identification of high risk in individuals who, if not screened, would never have known that they were at risk and would never have experienced a fragility fracture [59]. As this was not reported directly in any trial, we estimated this using available data from two trials, considering the proportion of participants exceeding the risk threshold in the study and the mean risk in these patients (see Additional file 3). For KQ3a (benefits of treatment), we pooled data by outcome as in KQ1; in several studies, there were zero events reported in one or both groups, and in these cases, we performed random effects meta-analysis using the reciprocal of the opposite treatment arm size correction for pooled odds ratio [69] in Stata (StataCorp, College Station, TX). We pooled data by sex and for each drug separately, but also performed an “all bisphosphonates” analysis including data from studies reporting on either of alendronate, risedronate, and zoledronic acid. For KQ3b, we report pooled effects directly as they were presented within the included systematic reviews and did not perform any re-analyses of data from primary studies.
We calculated absolute effects for each outcome comparison by applying the relative risk or odds ratio from the meta-analysis to the median control group event rates from the included studies [70]. For KQ1, we also incorporated a sensitivity analysis by calculating absolute effects using an assumed risk based on the general population in Canada (45 to 54 years and ≥65 years) [15, 71]. If statistically significant, we calculated the number needed to screen for an additional beneficial outcome (NNS), number needed to treat for an additional beneficial outcome (NNT), or number needed to treat for an additional harmful outcome (NNH) [72].
For KQ2 (predictive accuracy of screening tests), we chose not to pool the overall findings on calibration for most tools due to high levels of heterogeneity that could not be explained by a priori subgroups (age, sex, baseline risk within and across studies). We present the calibration findings by tool for both the population overall (average) and a summary of calibration within categories (e.g., quintiles, deciles) of baseline risk. We did pool data for the studies without high risk of bias reporting on the FRAX-Canada tool; we considered data from this subgroup to be most reliable and most directly applicable to Canada. These studies presented no major risk of bias concerns that would reduce our certainty in the findings, whereas all others generally had multiple major reasons to be seriously concerned about risk of bias. In all cases, we used the restricted maximum likelihood estimation approach and the Hartun-Knapp-Sidnick-Jonkman correction to derive $95\%$ CIs [73, 74]. We rescaled total observed versus expected fracture event ratios (O:E) and their variance (standard error) on the natural log scale prior to entering these into meta-analysis (or displaying on forest plots) to achieve approximate normality [75–77].
For KQ4 (acceptability of screening/treatment), we performed a narrative synthesis following the guidance of Popay et al. [ 78], recognizing that our question of acceptability differs to some extent from questions about interventions or implementation factors.
Across KQs, we considered several potential population and intervention/exposure subgroups of interest, for example in KQ1 analyses were stratified by age, while in KQ3a we analyzed data for postmenopausal females separately from males. In several cases, data on characteristics of interest were unavailable in the included study reports (e.g., baseline fracture risk). We also considered within-study subgroup analyses when these were available. We performed sensitivity analyses by risk of bias, applicability concerns (e.g., high-risk population in KQ1), and outcome ascertainment methods (e.g., clinical fragility fractures in KQ3a). When analyses for interventions contained at least eight trials of varying size, we assessed for small study bias using funnel plots and Harbord’s test (KQ3a) [79].
## Rating certainty of evidence and drawing conclusions
Two reviewers rated the certainty in the evidence for each outcome comparison of interest and agreed on the final rating and conclusion statements. Our certainty of evidence appraisals for effects of interventions were based on the absolute effects and considered only the direction of effect and not its magnitude. For KQ1 (benefits and harms of screening), KQ3 (benefits and harms of treatment), and KQ4 (acceptability of screening/treatment), we assessed the certainty of the evidence following the GRADE approach [80–86]. In the absence of published guidance on GRADE for reviews of risk prediction models, for KQ2 (predictive accuracy of screening tests) calibration outcomes, we considered input from an expert in GRADE to modify existing guidance [87] and assist in rating the evidence and developing conclusions. We decided a priori to consider tools to be well calibrated when the O:E ratio across the study populations consistently fell between 0.8 and 1.2 ($20\%$ over- or underestimation, respectively) [88]. We then rated certainty for one of four possible conclusions: well calibrated (O:E ratio consistently between 0.8 and 1.2), underestimation (O:E ratio >1.2 and adequately precise to draw clinically meaningful conclusions), overestimation (O:E ratio <0.8 and adequately precise to draw clinically meaningful conclusions), or poorly calibrated (wide variation across studies including over- and underestimation; unable to draw a clinically meaningful conclusion) (Additional file 4). For KQ3b, we relied preferentially on the certainty of evidence ratings presented by the included systematic reviews, with modifications if needed to align with our other appraisals. When these were not reported by the included systematic reviews, we performed our own GRADE appraisals, relying on the data available in the systematic reviews. When the data required to perform full evidence appraisals were missing from the included systematic reviews, we collected data from the included primary studies (if ≤5 studies) and/or made assumptions, as described in Additional file 4.
We developed informative statements based on our certainty in the evidence for each outcome comparison [89]. We adopted standard wording to describe our findings, using the word “may” together with the direction of effect to describe findings of low certainty and “probably” for those of moderate certainty. When our certainty in the evidence was very low, we describe the evidence only as “very uncertain” [89].
## KQ1a: What are the benefits and harms of screening compared with no screening to prevent fragility fractures and related morbidity and mortality in primary care for adults ≥40 years?
Of 7151 unique records retrieved by the searches for KQ1a and b, we assessed 163 for eligibility by full text and included five trials (4 randomized controlled trials [RCT] [4–6, 90], 1 controlled clinical trial [CCT] [68]) and two associated publications [91, 92] for KQ1a, and one RCT for KQ1b [93] (Fig. 1). Studies excluded after full text appraisal are listed with reasons in Additional file 5.Fig. 1Flow of records through the selection process. Legend: not applicable
## Study characteristics
Table 3 shows the characteristics of the included trials for KQ1a. The trials were conducted in countries with a moderate-to-high baseline fracture risk [94]: Denmark (ROSE [5]), the Netherlands (SALT [4]), the UK (SCOOP [6] and APOSS [90]), and the USA (Kern CCT [68]). Aside from the Kern CCT, which included a relatively equal proportion of males and females ≥65 years old [68], the trials included populations of exclusively peri-menopausal (aged 45 to 54 years) [90] or postmenopausal (mean ages 70 to 75.5 years; range 65 to 90 years) [4–6] females. When reported, between 10 and $44\%$ of the study population had a prior fracture [4–6]. The proportion of participants with a prior fracture was highest in the SALT trial ($44\%$), which enrolled females who reported at least one clinical risk factor on the clinical FRAX tool [4]. Participants were not treatment-naïve in all trials; in particular, the APOSS trial allowed enrollment by females with past use of hormone replacement therapy [90] and $11\%$ of participants in the ROSE trial were taking anti-osteoporosis medications at baseline [5].Table 3Characteristics of studies included for key questions 1a&b on the benefits and harms of screening versus no screening, and the comparative benefits and harms of different screening approachesStudy; design; country; funding; analysisPopulation characteristicsScreening approach (n randomized)Treatment thresholdRisk in those meeting thresholdAbove treatment threshold; Treatment uptakeOutcomes; follow-upKQ1a: benefits and harms of screening versus no screeningMerlijn 2019 (SALT) [4]RCTNetherlandsFoundation, industry, academicAnalysis: selected population (high risk)11,032 ($20.5\%$ of 53,794 in age-based sample) females aged 65 to 90 y with ≥1 clinical risk factor; $47\%$ of original sample completed FRAX, but $56\%$ of these were ineligible or did not have a risk factor. Mean (SD) 75.0 (6.7) y; $44\%$ prior fracture (location NR); $1\%$ type 1 diabetesScreening ($$n = 5575$$): 2-step- Completed FRAX-UK- BMD + VFA if ≥ 1 risk factor- $76\%$ of eligible for BMD participatedUsual care ($$n = 5457$$): completed FRAX-UK; advised to visit GP if ≥ 1 risk factor; $6\%$ underwent DXA and VFATreatment threshold: any of a) lumbar/thoracic fracture with vertebral height reduction, b) exceeding age-specific FRAX + BMD MOF risk threshold, or c) risk score ≥4 according to Dutch guidelinesaMean (SD) FRAX + BMD:10-y MOF risk: 23.9 ($9.6\%$)10-y hip fracture risk: 10.6 (10.1)%Above treatment threshold: $\frac{1417}{4228}$ ($34\%$) who underwent screening; $25\%$ for the screening groupSelf-report of any osteoporosis medication: $21\%$ in screened ($69\%$ with treatment indication); $5\%$ in usual care (mainly bisphosphonates)≥1 hip fractures: self-reported and verified≥1 MOF (hip, clinical vertebral, wrist, humerus): self-reported and verifiedAll-cause mortality: reported by relativesFollow-up: ≥36 monthsRubin 2018 (ROSE) [5]RCTDenmarkGovernment, academicAnalysis: offer-to-screen; selected population (completed FRAX)34,229 (18,605 with FRAX; $54.4\%$ of eligible) females aged 65 to 80 yMedian (IQR) 71 [8] y; $10\%$ prior fracture (location NR) in those with FRAX; diabetes NRScreening ($$n = 17$$,072; 9279 with FRAX): 2-step- Completed FRAX-Denmark- BMD + VFA when 10-y risk of MOF was ≥$15\%$- $71\%$ of eligible for BMD participatedUsual care ($$n = 17$$,157; 9326 with FRAX): completed FRAX, risk not calculated; $25\%$ had DXA scan after the index dateTreatment threshold: BMD T-score at any site ≤2.5; vertebral fracture on VFA.Median (IQR) FRAX + BMD: NR in those meeting threshold. Screened group ($$n = 5009$$) with DXA:10-y MOF risk: 22 [15, 29]%10-y hip fracture risk: 8.1 (5.6, 13)%Above treatment threshold:$\frac{1236}{9279}$ ($13\%$) who completed FRAX; $7\%$ for the screening groupAny osteoporosis medication (pharmacy records): $23\%$ in screened ($80\%$ with treatment indication); $18\%$ in controls≥1 hip fracture: records (ICD-10 codes)≥1 MOF (hip, clinical vertebral, wrist, humerus): records (ICD-10 codes)Follow-up: median (IQR) 5 (1.3) yShepstone 2018 (SCOOP) [6, 91]RCTGovernment, foundationUnited KingdomAnalysis: selected population (completed FRAX)12,483 ($32.3\%$ of eligible) females aged 70 to 85 yMean (SD) 75.5 (4.2) y; $24\%$ prior fracture (location NR); diabetes NRScreening ($$n = 6233$$): 2-step- Completed FRAX- BMD when 10-y risk of hip fracture met high risk threshold based on age- $92\%$ of eligible for BMD participatedUsual care ($$n = 6250$$): completed FRAX, fracture probability not calculated; GP received letter stating patient’s involvementTreatment threshold: exceeding age-specific 10-y hip fracture risk (FRAX-BMD) thresholdMean (SD) FRAX (no BMD):10-y MOF risk: 30.0 (10.7)%10-y hip fracture risk: 17.9 (10.9)%Above treatment threshold: $\frac{898}{2790}$ ($32\%$) who completed FRAX + BMD; $14\%$ for the screening groupAny osteoporosis prescription (GP records): $\frac{1486}{6233}$ ($24\%$) in screened ($78\%$ with treatment indication in first 6 months); $16\%$ in controls≥1 hip fracture: self-report, records≥1 osteoporosis-related fracture (not hands, feet, nose, skull, vertebrae): self-report; recordsAll-cause mortality: registry data, family members, GPsHealth-related quality of life: self-report via EuroQol-5D, Short-Form 12 Health SurveySerious AEs: GPs recorded serious AEsFollow-up: 5 yBarr 2010 (APOSS) [90, 92]RCTUnited KingdomFoundation, industryAnalysis: offer-to-screen; acceptors of screening (completed BMD)4800 (3128 attended / had complete follow-up; $65\%$ of eligible); peri-menopausal females aged 45 to 54 yMean (SD) 58.4 (3.7) y; prior fractures and diabetes NRScreening ($$n = 2400$$; 1764 attended): 1-step invitation to be screened by BMD via DXAUsual care ($$n = 2400$$; 1364 with complete follow-up): not invited to be screenedTreatment threshold: BMD at any site within the lowest quartile of first 1000 women screenedBaseline risk: NRAbove treatment threshold: NR; lowest quartileSelf-reported uptake of any osteoporosis medication >3 months (bisphosphonates, raloxifene, hormone replacement therapy): $69\%$ in screened (% with treatment indication NR); $59\%$ in controls≥1 hip fracture: self-reported and verified≥1 MOF (hip, wrist, vertebrae, humerus): self-report and verifiedGeneral health: self-reportedHealth status (2-y follow-up): self-reported on Short-Form 36 SurveyAll-cause mortality: NRFollow-up: median 9.1 y in screened, 8.8 y in controlsKern 2005 [68]CCT (non-random allocation based on availability of screening)United StatesGovernment, foundationAnalysis: selected population (enrolled in another study)3107 adults ≥65 years ($87\%$ of eligible study participants offered screening)Mean (SD) 76.2 (4.9) y; $56\%$ female; <$0.1\%$ with radius or ulna fracture in past 5 y, other fractures NR; $1\%$ diabetesScreening ($$n = 1422$$): 1-step offer to be screened by BMD via DXA; $97\%$ completed scansUsual care ($$n = 1685$$): not offered BMD scanRisk definition: BMD below age-matched mean of densitometer manufacturer’s reference groupRisk in those meeting threshold: NRAbove treatment threshold: $33\%$ of those completing a DXA scan (392 females, 69 males); $32\%$ in the screening groupAny bone-enhancing medication (includes calcium, multi-vitamins, estrogen, calcitonin, bisphosphonates): $27\%$ in screened ($31\%$ with treatment indication); NR in controlsTotal number of hip fractures: hospital records (ICD-9 codes); verified against Medicare claims dataAll-cause mortality: surveillance of hospital records and verified against Medicare claims dataFollow-up: mean 4.9 yKQ1b: comparative benefits and harms of different screening approachesLaCroix 2005 (OPRA) [93]RCT (3-arm)United StatesIndustryAnalysis: offer-to-screen9268 (3167 [$34\%$] participated) females aged 60 to 80 yMean (SD) 70.0 (5.6) y; $17\%$ prior fracture; diabetes NRUniversal screen ($$n = 1986$$; 415 participated): 1-step invitation to be screened by BMD via DXASCORE-based screen ($$n = 1940$$; 576 participated): 2-step- All completed SCORE- BMD offered if score ≥7- $74\%$ were eligible for BMDSOF-based screen ($$n = 5342$$; 2176 participated):- All completed SCORE- BMD offered if ≥5 clinical risk factors- $7\%$ were eligible for BMDRisk definition: ≥5 fracture risk factors and/or BMD T-score <−2.5 for 60–64 y or z-score <-0.43 for ≥65 y; prior fracture after age 50 (SOF-based group only)Risk in those meeting threshold: NRAbove treatment threshold: $28\%$ of those screened in the universal group ($6\%$ of allocated); $32\%$ of those completing the SCORE-based tool ($7\%$ of allocated); $18\%$ of completing the SOF-based tool ($7\%$ of allocated)Any dispensed prescription for osteoporosis medication (includes alendronate, hormone replacement therapy, calcitonin, raloxifene): $13\%$ in universal screening ($21\%$ of screened), $14\%$ in SCORE-based ($20\%$ of screened), $13\%$ in SOF-based ($17\%$ of screened) groupTotal number of hip fractures; all non-pathologic (osteoporotic) fractures: hospitalization and outpatient visit records (ICD-9 codes)Follow-up: mean (range) 28 (24–33) monthsAE Adverse events, BMD Bone mineral density, CCT Clinical controlled trial, DXA Dual-energy X-ray absorptiometry, FRAX Fracture Risk Assessment Tool, GP General practitioner, IQR Interquartile range, VFA Vertebral fracture assessment, MOF Major osteoporotic fracture, NR Not reported, RCT Randomized controlled trial, SD Standard deviation, y yearsaBone densitometry and VFA is indicated if the total risk score is ≥4 points (composite of vertebral fracture (4 points), recent fracture after age 50 years [4], age ≥60 years [1], age ≥70 years [1], non-recent fracture after age 50 years [1], additional non-recent fracture after age 50 years on a separate occasion [1], parental hip fracture [1], body weight <60 kg [1], severe immobility or 1 or more falls in the past year [1]). Bisphosphonate treatment is advised if BMD of either femoral neck or lumbar spine shows a T-score ≤ −2.5 or if a prevalent vertebral fracture (≥$25\%$ height reduction) is present The three more recent trials (published 2018–2019) [4–6] employed a 2-step approach to screening, whereby all participants completed a mailed questionnaire including data to assess risk with the clinical FRAX tool, and only those surpassing certain risk thresholds were offered BMD assessment. The threshold for BMD assessment varied across trials; in the SALT trial, the entire population had ≥1 risk factor and were offered BMD and vertebral fracture assessment [4], whereas ROSE offered BMD for those with a clinical FRAX-based 10-year major osteoporotic fracture risk ≥$15\%$ [6], and SCOOP used age-based thresholds of 10-year hip fracture risk [5]. The two older trials (APOSS [2010] and Kern CCT [2005]) used a one-step direct to BMD screening approach [68, 90]. No trials included a true “no screening” comparator; in all cases, the comparator was usual care, with evidence of varying levels of ad hoc screening and treatment (median $17\%$ treatment rate when this was reported, range 5 to $59\%$ [4–6, 90]) within the comparison groups.
Thresholds for treatment were also variable across the trials. In both the SALT and SCOOP trials, BMD assessment was used to recalculate the 10-year FRAX fracture risk with inclusion of BMD, and treatment was offered when participants exceeded age-specific thresholds [4, 6]; the SALT trial also allowed for several other treatment indications according to Dutch guidelines (e.g., vertebral fracture) [4]. Of note, in the SCOOP trial, only 898 females exceeded a treatment threshold despite 3064 being considered at elevated risk based on fairly similar thresholds but without incorporation of the BMD results into the risk prediction by clinical FRAX. In the ROSE trial, treatment was offered when the BMD T-score at any measured site was ≤2.5, and/or a fracture was detected on vertebral fracture assessment [5]. In the two 1-step screening trials, treatment was offered to those in the lowest quartile of BMD, based on the first 1000 participants screened (APOSS) [90], and to those below the age-matched mean of the reference group according to the densitometer’s manufacturer (Kern CCT) [68]. Across the trials, between 7 and $25\%$ of those assigned to screening had indications for treatment; the proportion was highest in the SALT trial, where higher-risk patients were enrolled [4]. The rate of treatment was lowest in the Kern CCT ($31\%$ of those with a treatment indication) [68]; among the remaining trials, more than two-thirds (69 to $80\%$) of those with a treatment indication reported using some form of anti-osteoporosis drugs during follow-up (variable treatments across studies, and sometimes including those such as calcitonin and hormone replacement therapy, which are no longer recommended; see Table 3). It was apparent that most of the treatment provided in the recent RCTs was pharmacologic, though at least one protocol (SALT) mentioned calcium and vitamin D supplementation, as well as notification of a high fall risk, that may have been acted upon by the primary care practitioner.
The trials provided data for hip fractures [4–6, 68, 90], clinical fragility fractures (described as major osteoporotic [4, 5, 90] or osteoporosis-related fractures [6]), serious adverse events [6], all-cause mortality [4, 6, 68, 90], and quality of life or wellbeing [6, 90, 92]; no trials reported on fracture-related mortality, functionality and disability, discontinuation due to adverse events, or non-serious adverse events. Though not directly reported, data were available in two trials to estimate the potential extent of overdiagnosis (see Additional file 3 for calculations) [4, 6]. Because of differences in design and reporting across the trials, we considered three possible population perspectives in our analyses. Two trials (APOSS and ROSE) provided data for an offer-to-screen population, whereby all eligible people invited for screening by mail, regardless of actual participation in any screening, were analyzed [5, 90]. The APOSS study also provided data for acceptors of screening, where the analysis included only those who attended for BMD measurement and thus completed screening. The SALT, ROSE, and SCOOP trials provided data for what we considered an offer-to-screen in selected population approach, because the analyses only included people who independently completed a mailed clinical FRAX questionnaire as part of 2-step screening [4–6]. The Kern CCT [68] also contributed data for this approach, as the sample population for screening was those already enrolled in the Cardiovascular Health Study (i.e., not the general population) [95]. We considered the “selected population” approach to be the one to be most applicable to primary care—where healthcare providers would complete risk assessment tools during the patient visit and then discuss the findings—although the participants in these trials are likely to be more accepting and compliant with screening, and possibly with treatment, than the general population presenting to primary care.
The risk of bias ratings for the included trials for KQ1a are in Table 4. The main risk of bias concerns were related to participant awareness of group assignments and contamination of the control groups in all trials (aforementioned ad hoc screening and treatment, likely to bias the findings toward the null) [4–6, 68, 90], and a high risk of attrition bias in the APOSS trial ($42\%$ lost to follow-up) in the offer-to-screen population [90]. The Kern CCT was not randomized, however patients were invited based on age- and sex-stratified random sampling and analyses were adjusted for baseline differences between groups [68]. We rated this trial, as well as the “acceptors” population for the APOSS and the “selected population” in the ROSE trial, to be at unclear risk of selection bias [5, 90], because in these analyses, the participants no longer represented the initially randomized population. Table 4Risk of bias assessments for trials included for KQ1a on the benefits and harms of screening vs. no screening, and KQ1b on the comparative benefits and harms of different screening approachesCCT Clinical controlled trial, PP Per protocol As indicated in the findings for KQ1a, one RCT (OPRA) [93] was included for the comparative effectiveness of different screening approaches. Characteristics of the OPRA trial are in Table 3. The trial included a mailed offer-to-screen population (acceptors of screening also available but less relevant to the primary care population). Eligible ($$n = 9268$$; $34\%$ participated) postmenopausal females were randomized to one of three screening approaches: 1-step screening using BMD via DXA; 2-step screening using the Simple Calculated Osteoporosis Risk Estimation (SCORE)-based tool, with BMD assessment offered when the score was ≥7 ($74\%$ eligible); and 2-step screening using the Study of Osteoporotic Fractures (SOF)-based tool, with BMD assessment offered to those with ≥5 clinical risk factors ($7\%$ eligible) [93]. Patients were eligible for potential treatment if they had ≥5 risk factors and/or BMD T-score below age-specific thresholds, or if they had a prior fracture after age 50 years (SOF-based group only) [93]. The proportion of patients dispensed a prescription (including alendronate, hormone replacement therapy, calcitonin, raloxifene) was similar across groups (13 to $14\%$ of those offered screening) [93]. The two outcomes reported by the trial were the total number of hip fractures, and clinical fragility fractures (reported as non-pathologic [osteoporotic] fractures) [93].
The risk of bias assessment for the OPRA trial is in Table 4. The trial was rated at unclear risk of bias due to the potential for selection bias (randomization and allocation concealment not clearly defined) and patient awareness of group assignment (those in the SCORE- and SOF-based groups not assigned to BMD testing would have increased awareness of risk and could seek further care) [93]. The trial was not powered to detect a difference in fracture outcomes across groups.
Additional file 6 shows the characteristics of the included studies and their associated publications. Half ($\frac{16}{32}$, $50\%$) of the included studies were composed of participants from the USA ($$n = 9$$) [104, 109, 110, 113, 136, 140, 143, 144, 153] and Canada ($$n = 7$$) [106, 111, 119, 128, 129, 134, 154]; the remaining studies took place in Spain ($$n = 4$$) [98, 99, 145, 150], Japan ($$n = 3$$) [117, 148, 149], France ($$n = 2$$) [146, 151], Israel ($$n = 2$$) [108, 112], Poland ($$n = 2$$) [107, 142], Australia ($$n = 1$$) [116], New Zealand ($$n = 1$$) [100], and Portugal ($$n = 1$$) [138]. The studies analyzed data from a total of 1,491,968 participants (median 3305, range 91 to 1,054,815), with mean age ranging from 51 to 74.2 years. In more than half of the studies, only females were included ($\frac{17}{32}$, $53\%$) [98–100, 104, 106, 107, 112, 134, 136, 142–144, 146, 148, 150, 151]; the remaining were equally split between including only males ($$n = 7$$, $22\%$; one cohort [129] included females but only the male population was used for analysis) [109, 110, 113, 116, 117, 153], and a mix of males and females ($$n = 8$$, $25\%$) [108, 111, 119, 128, 138, 140, 145, 154]. Participants were often recruited from patient, insurance, or resident (e.g., electoral rolls) registries ($$n = 16$$/32, $50\%$) [98, 108–113, 116, 119, 138, 140, 142, 143, 146, 148, 149]; ten ($31\%$) studies enrolled all those presenting for BMD assessment (potentially at higher risk depending on local practices) [99, 106, 107, 128, 129, 136, 144, 145, 150, 151], five ($16\%$) included patients already enrolled in other studies [100, 104, 117, 134, 154], and one ($3.2\%$) enrolled only veterans [153]. Studies most commonly provided findings for the calibration of clinical FRAX (i.e., without incorporation of BMD) or with incorporation of BMD results (i.e., FRAX + BMD; $$n = 26$$/32, $81\%$) [98–100, 104, 106–109, 111, 112, 116, 117, 129, 134, 138, 140, 142–144, 146, 148–151, 153, 154] and Garvan with or without BMD ($$n = 8$$, $25\%$) [100, 104, 108, 113, 119, 142, 145, 154]; there were few external validation studies reporting on QFracture ($$n = 3$$) [108, 113, 154], the Fracture Risk Calculator (FRC; $$n = 2$$) [110, 136], CAROC ($$n = 1$$) [128], and the Fracture and Immobilization Score (FRISC; $$n = 1$$) [149].
The risk of bias ratings for the included studies for KQ2 are in Additional file 7. Almost all of the studies were at high overall risk of bias; only four [106, 111, 128, 129] were lacking serious risk of bias concerns (rated at unclear risk of bias because proxy variables were used for some predictors, e.g., chronic obstructive pulmonary disease instead of smoking status). The primary risk of bias concerns across the included studies were related to predictor ascertainment (missing predictor data, predictors not handled as intended by the tool), outcome ascertainment (self-reported or including high trauma fractures), and the analysis (large losses to follow-up and/or competing risk of mortality not accounted for, inadequate number [<100] of fracture outcomes, follow-up duration not matching the prediction period [e.g., substantially shorter or longer than 10 years without adjustment]). Many studies did not account for the effect of treatment prior to risk assessment or during follow-up.
Detailed study characteristics are in Additional file 6. In total, there were 10 trials of alendronate (5 or 10 mg/day, or mixed doses, or 70 mg/week for 12 to 48 months) [172, 173, 176, 177, 183–185, 187, 193, 197], 7 trials of risedronate (2.5 or 5 mg/day for 12 to 36 months) [179, 182, 183, 186, 189, 190, 196], 6 trials of zoledronic acid (1 to 5 mg/year [5 mg/year most commonly] for 12 to 72 months) [175, 180, 181, 188, 194, 195], and 6 trials of denosumab (60 mg/6 months, or mixed doses for 12 to 36 months) [174, 178, 185, 191, 192, 198]. About half ($\frac{14}{27}$, $52\%$) of the trials were multi-country [172, 175, 178, 179, 183, 184, 187–191, 193, 194, 196], with the remaining taking place in the USA ($$n = 4$$) [173, 176, 177, 185], New Zealand ($$n = 3$$) [180, 181, 195], China ($$n = 3$$) [186, 197, 198], Australia ($$n = 1$$) [182], India ($$n = 1$$) [192], or the USA and Canada ($$n = 1$$) [174].
The trials included a total of 34,317 participants (median 398, range 50 to 9931), primarily postmenopausal females with low BMD (definition variable across trials). The prevalence of prior fracture was median $16.9\%$ (range 0 to $48\%$) when specified in the trials. There were only two trials of males with low BMD, one for zoledronic acid [175] and one for denosumab [191]. Most of the trials were small and probably underpowered to detect differences in fracture incidence, especially for hip fractures; analyses generally relied on one large trial per drug. Most ($\frac{23}{27}$, $82\%$) trials included adjunct calcium and/or vitamin D supplements in both groups (treatment and placebo). Length of follow-up for outcomes ranged from 0.5 to 6 years, which in almost all cases corresponded with the duration of treatment; rarely, the follow-up period extended 1 year beyond the end of treatment. The trials provided data for hip fractures [172, 175–178, 180, 181, 184, 186, 187, 189–193, 195–198], clinical fragility fractures [172–175, 177–196, 198], clinical vertebral fractures [172, 174, 176–178, 180, 181, 184, 186, 191, 192, 194–197], all-cause mortality [174–178, 180, 185, 188, 191, 192, 195–198], and health-related quality of life [178]; no trials reported on fracture-related mortality or functionality and disability. Discontinuation due to adverse events, serious and non-serious adverse events are addressed in KQ3b.
The risk of bias ratings for the trials included for KQ3a are in Additional file 7. One of the main risk of bias concerns was selective reporting, as many trials lacked protocols and did not pre-specify fractures as an outcome of interest (either in a protocol or in the “Methods” section); instead, these were often collected as potential harms. In these cases, it was often unclear whether the fracture outcomes were collected prospectively or systematically [172, 173, 176, 180, 181, 185, 190–192, 197, 198]. Several trials were at high risk of attrition bias, due to large or imbalanced losses to follow-up for various outcomes [172, 173, 179, 180, 186, 189–191]. One trial of alendronate was open-label [185] and therefore was at high risk of performance and detection biases. When applicable (“all bisphosphonates” analyses), we assessed for small study bias and this was not detected.
Characteristics of the systematic reviews and primary study are in Additional file 6. The systematic reviews were published between 2014 and 2020 and included either only RCTs [212, 216, 217] or a mix of RCTs and observational studies [60, 211, 213–215, 218]; occasionally, only observational studies were included when there existed no RCTs for rare harms [210]. The systematic reviews were generally focused on patients (males or females) with low BMD (often referred to as osteoporosis) or who had risk factors for fracture, though some included wider populations (e.g., patients with chronic use of glucocorticoids); in many cases, patients with other disorders of bone metabolism were excluded. Across the systematic reviews, risk of bias was usually not assessed specific to harm outcomes (assessed in 3 reviews [210, 216, 217]), and certainty of evidence was assessed for selected outcomes in only three of the systematic reviews [60, 211, 215]. Notably, no evidence (either no systematic reviews, or the systematic reviews located no primary studies) was located for the following outcome comparisons: serious stroke and thromboembolic events, atypical femoral fractures, osteonecrosis of the jaw, or myalgia, cramps, and limb pain with risedronate; serious gastrointestinal adverse events, gastrointestinal cancer, pulmonary embolism, and thromboembolic events with zoledronic acid; osteonecrosis of jaw with long-term bisphosphonates as a class; serious gastrointestinal adverse events, gastrointestinal cancer, thromboembolic events, cardiac death, and rebound hip fractures with denosumab. The primary study on rebound fractures (multiple vertebral fractures) after discontinuation versus persistence of denosumab was a retrospective cohort study of 3110 individuals ($91\%$ females; mean age 72 years; $42\%$ with prior fracture; denosumab as first-line therapy for $5.4\%$) conducted in Israel.
The appraisal of the quality of the systematic reviews and primary study included for KQ3b are shown in Additional file 7. Common methodological concerns across the reviews were potential errors in data extraction (because data were not collected in duplicate), limited description of the characteristics of the included studies, and lack of risk of bias appraisal (or risk of bias was assessed for benefits but not for harms). The primary study did not adjust findings for potential confounders, though there was demonstration of comparability across multiple characteristics between groups.
Detailed study characteristics are in Additional file 6. Half of the 12 studies were conducted in the USA ($\frac{6}{12}$, $50\%$) [221, 223, 225–227, 231]; the remaining were conducted in New Zealand ($$n = 3$$) [222, 229, 230], Canada ($$n = 1$$), the Netherlands ($$n = 1$$) [220], and China ($$n = 1$$) [224]. Across all studies, a total of 2188 participants (median 204, range 30 to 393) were included, primarily postmenopausal females. In three studies [222, 224, 230], both males and females were included. One study reported on the acceptability of screening among females who would be considered to be at low risk based on age (mean 57 years, range 50 to 65 years) [231]. The remaining 11 studies elicited patients’ views on the acceptability of initiating pharmacologic treatments. In four ($36\%$) of these studies, patients who were at risk for fracture based on BMD (T-score in osteoporosis or osteopenia range, definitions varied across studies) and were aware of their 10-year major osteoporotic and/or hip fracture probability were provided decision aids and were in the position to make real-life decisions about starting treatment. In the remaining studies, the decisions about starting treatment were based on hypothetical scenarios; patients in these studies were not always made aware of their fracture risk and would not necessarily have been eligible for treatment [220–224, 229–231].
The risk of bias assessments for studies included in KQ4 are in Additional file 7. Four studies were at high risk of bias due to low participation rates (<$40\%$ of those eligible) [222, 223, 229, 231]. Three studies were at high risk of bias because they provided participants no or inaccurate (based on our comparison to currently available evidence) information on the potential benefits or harms of treatment—we required information on at least one of benefits or harms for inclusion [222, 227, 230]. Two studies were considered to be at high risk of bias because they did not present findings for important subgroups of interest (e.g., baseline fracture risk) for whom results may be expected to differ [225, 227]. Other risk of bias concerns were infrequent.
## Findings
Table 5 summarizes the main findings for KQ1a; Additional file 3 contains the full GRADE Evidence Profiles and Summary of Findings Tables, with explanations for each rating as well as the forest plots, which include the results of statistical tests for subgroup differences. Among females aged 68–80 years, data from one trial showed that a mailed offer of screening in the general population may not reduce the risk of hip fractures, clinical fragility fractures, or all-cause mortality during 5 years of follow-up [5]. The evidence is very uncertain for all outcomes from a mailed offer of screening with BMD among females aged 45–54 years during 9 years of follow-up (1 trial) [90].Table 5Summary of findings for KQ1a on the benefits and harms of screening compared with no screeningStudy approachPopulation; studies; sample sizeFollow-up (y)Assumed population riskeAbsolute effectsCertaintyfWhat happens?Hip fracturesAll eligible / offer-to-screenFemales 45–54 y [90]The evidence from 1 RCT ($$n = 2979$$) is very uncertain. VERY LOWa-dVery uncertainFemales 68–80 y1 RCT; 34,229 [5]5Study data:25 per 10000.3 fewer in 1000(4.2 fewer to 3.9 more)LOWa-cMay not reduceGeneral:20 per 10000.2 fewer in 1000(2.4 fewer to 2.2 more)Acceptors of screeningFemales 45–54 y [90]The evidence from 1 RCT ($$n = 2604$$) is very uncertain. VERY LOWa-dVery uncertainOffer-to-screen in selected populationgFemales ≥65 y3 RCT+1 CCT; 43,736 [4–6, 68, 91]3 to 5Study data:31 per 10006.2 fewer in 1000(9.0 fewer to 2.8 fewer)MODERATEcProbably reducesGeneral:20 per 10004.0 fewer in 1000(5.8 fewer to 1.8 fewer)Males ≥70 y [68]The evidence from 1 CCT ($$n = 1380$$) is very uncertain. VERY LOWa-dVery uncertainClinical fragility fracturesAll eligible / offer-to-screenFemales 45–54 y [90]The evidence from 1 RCT ($$n = 2979$$) is very uncertain. VERY LOWa-dVery uncertainFemales 68–80 y1 RCT; 34,229 [5]5Study data:100 per 10001.0 fewer in 1000(8.0 fewer to 6.0 more)LOWa-cMay not reduceGeneral:168 per 10001.7 fewer in 1000(13.4 fewer to 10.1 more)Acceptors of screeningFemales 45–54 y [90]The evidence from 1 RCT ($$n = 2604$$) is very uncertain. VERY LOWa-dVery uncertainOffer-to-screen in selected populationgFemales ≥65 y3 RCT; 42,009 [4–6, 91]3 to 5Study data:84 per 10005.9 fewer in 1000(10.9 fewer to 0.8 fewer)MODERATEcProbably reducesGeneral:168 per 100011.8 fewer in 1000(21.8 fewer to 1.7 fewer)All-cause mortalityAll eligible / offer-to-screenFemales 45–54 y1 RCT; 4800 [90]9Study data: The evidence is very uncertain. VERY LOWb,dVery uncertainGeneral:3 per 1000No difference in 1000(0.8 fewer to 1.1 more)LOWb,dMay not reduceFemales 68–80 y1 RCT; 34,229 [5]5Study data:118 per 10003.5 fewer per 1000(9.4 fewer to 3.5 more)LOWb,dMay not reduceGeneral:57 per 10001.7 fewer per 1000(4.6 fewer to 1.7 more)Offer-to-screen in selected populationgFemales ≥65 yh2 RCT+1 CCT; 26,511 [4, 6, 68]3 to 5Study data:89 per 1000No difference in 1000(7.1 fewer to 5.3 more)MODERATEdProbably does not reduceGeneral:57 per 1000No difference in 1000(4.6 fewer to 5.1 more)Serious adverse eventsOffer-to-screen in selected populationgFemales 70–85 y [6]The evidence from 1 RCT ($$n = 12$$,483) is very uncertain. VERY LOWa,b,dVery uncertainHealth-related quality of life/WellbeingAll eligible / offer-to-screenFemales 45–54 y [90, 92]9 (self-rated health)2 (SF-36)NAThe evidence from 1 RCT ($$n = 2979$$) is very uncertain. VERY LOWa-cVery uncertainOffer-to-screen in selected populationgFemales 70–85 y1 RCT; 10,661 [6]5NASF-12 (range 0–100):Mental health: MD −0.30, $95\%$ CI −0.86 to 0.26Physical health: MD 0.30, $95\%$ CI −0.21 to 0.81EuroQol-5D (range 0–1):MD 0, $95\%$ CI −0.07 to 0.07LOWa,bMay be little to no differenceOverdiagnosisOffer-to-screen in selected populationgFemales 70–85 y (1 RCT; 6,233) [6] 14.4 × (100 − 17.9) /100 = $11.8\%$ overdiagnosedFemales 65–90 y (1 RCT; 5575) [4] 25.4 × (100 − 23.9) / 100 = $19.3\%$ overdiagnosed (selected higher-risk population)Among those considered at high riskFemales 70–85 y (1 RCT; 3064) [6] 29.3 × (100 − 17.9) / 100 = $24.1\%$ overdiagnosedCCT Clinical controlled trial, CI Confidence interval, RCT Randomized controlled trial, MD Mean difference, NA Not applicable, y yearsaRisk of bias; binconsistency; cindirectness; dimprecisione *Study data* refers to the median control events rates across trials, which is the main analysis. A sensitivity analysis used the effects without screening for the general risk population in Canada, estimated from PRIOR et al. ( Bone. 2015;71:237-43) based on 10-year follow-upf When our assessment of the certainty of evidence fell between levels, we assigned the level that best represented our actual certaintyg Selected population defined as those who completed the initial risk assessment tool (as part of 2-step screening). This population may be more accepting of screening and have higher compliance than the general (intention-to-screen) populationh *This analysis* included 1379 men from Kern 2005, representing $5.4\%$ of the total sample *Among a* selected population of females aged ≥65 years who are willing to independently complete a mailed fracture risk questionnaire, 2-step screening with risk assessment (clinical FRAX or FRAX-like tool) and BMD probably reduces the risk of hip fractures (3 RCTs + 1 CCT; $$n = 43$$,736; 6.2 fewer in 1000, $95\%$ confidence interval [CI] 9.0 fewer to 2.8 fewer; NNS=161) [4–6, 68] and clinical fragility fractures (3 RCTs; $$n = 42$$,009; 5.9 fewer in 1000, $95\%$ CI 10.9 fewer to 0.8 fewer; NNS=169) [4–6]. However, screening in this selected population probably does not reduce the risk of all-cause mortality (note: 1379 males were included in this analysis from the Kern CCT, representing $5.4\%$ of the total sample) [4, 6, 68]. Our sensitivity analyses using assumed/baseline risks from a general Canadian population (age roughly corresponding to that of the trials) suggest that the effects for clinical fragility fracture may be larger than found in the trial populations, but these analyses are considered exploratory (Table 5). Post hoc subgroup analyses from the SCOOP study showed that the effectiveness of screening on hip fracture risk was greater in females with higher baseline clinical FRAX 10-year hip fracture risk (HR [$95\%$ CI] 0.67 [0.53–0.84] in the 90th percentile of risk vs. 0.93 [0.71–1.23] in the 10th percentile, $$p \leq 0.021$$) and with prior fracture (HR [$95\%$ CI] 0.55 [0.38–0.79] vs. 0.87 [0.68–1.12], $$p \leq 0.040$$ without prior fracture) [91]. The evidence for the effect of an offer of screening in a selected population of males is very uncertain [68]. In females aged 70–85 years, screening may make little-to-no difference in health-related quality of life [6]. Between $11.8\%$ [6] and $19.3\%$ [4] of females in a selected population offered 2-step screening may be overdiagnosed, but the magnitude of these estimates is of low certainty due to serious concerns of indirectness from lack of data provided as required for the proposed equation (e.g., mean risk in the high-risk population in SCOOP was limited to results of clinical FRAX without incorporation of BMD as used for treatment decisions) and from use of data from the SALT trial where participants were all at increased risk. Among females aged 70–85 years who are considered at high-risk by FRAX 10-year hip fracture risk alone and are referred to BMD assessment, data from one trial indicate that $24.1\%$ may be overdiagnosed [6], but there is low certainty about this due to serious concerns about indirectness.
The evidence for hip and clinical fragility fractures among females aged 45–54 who accept 1-step screening with BMD measurement is very uncertain.
Additional file 3 contains the full analysis details for KQ1b, including the GRADE Summary of Findings Tables, with explanations for each rating and forest plots. The evidence from a single RCT showed that, among females aged 60–80 years, the evidence comparing 1-step (BMD) versus 2-step screening (risk assessment + BMD) and comparing different 2-step screening strategies (i.e., SCORE-based vs. SOF-based for the risk assessment) for risk of hip and clinical fragility fractures is very uncertain [93].
Additional file 4 contains the full analysis details for KQ2, including GRADE Summary of Findings Tables, with explanations for each rating, and forest plots. Within the Summary of Findings Tables, discrimination findings from the USPSTF’s review are shown. Due to a high degree of heterogeneity that could not be well explained by a priori subgroup analyses, we generally did not pool data on calibration, and instead present the findings descriptively. The exception was FRAX-Canada, where we pooled (and relied on primarily) data from the three Canadian studies without serious risk of bias concerns. This decision was based on recognition that FRAX is considered as a suite of tools (algorithm calibrated to various countries) rather than a single tool; therefore, these Canadian studies without serious risk of bias would provide the most directly applicable evidence.
Forest plots for the calibration of clinical FRAX and FRAX + BMD across studies with and without serious risk of bias concerns are in Figs. 2 and 3, respectively. For both the 10-year prediction of hip and clinical fragility fractures, there was a high degree of heterogeneity in O:E estimates across studies that was not well explained by subgroup analyses by age, sex, and baseline risk (Additional file 4). Most studies were at high risk of bias and did not use FRAX-Canada. We judged the performance of FRAX (with and without BMD) to be poor in these studies, but the evidence was rated as very uncertain due to concerns across all GRADE domains. Pooled data from three Canadian studies ($$n = 67$$,611) [106, 111, 129] without serious risk of bias indicate that clinical FRAX-Canada may be well calibrated for the 10-year prediction of hip fractures (O:$E = 1.13$, $95\%$ CI 0.74–1.72, I2 = $89.2\%$) and is probably well calibrated for the 10-year prediction of clinical fragility fractures (O:$E = 1.10$, $95\%$ CI 1.01–1.20, I2 = $50.4\%$), both with some underestimation of the observed risk. Data from these same studies ($$n = 61$$,156) [106, 111, 129] showed that FRAX-Canada with BMD may perform poorly to estimate 10-year hip fracture risk (O:$E = 1.31$, $95\%$ CI 0.91–2.13, I2 = $92.7\%$), but is probably well calibrated for the 10-year prediction of clinical fragility fractures, with some underestimation of the observed risk (O:E 1.16, $95\%$ CI 1.12–1.20, I2 = $0\%$). Within-study data from calibration plots (e.g., using deciles of baseline risk) were heterogeneous (7 studies for 10-year prediction of hip fractures [99, 100, 104, 109, 112, 143, 148] and 8 for clinical fragility fractures [99, 100, 104, 109, 112, 134, 143, 148] with clinical FRAX; 8 studies for the 10-year prediction of hip fractures [99, 100, 106, 109, 111, 125, 143, 148] and 10 for clinical fractures [99, 100, 106, 109, 111, 117, 140, 143, 148, 150] with FRAX + BMD), but two Canadian studies without serious concerns for risk of bias showed acceptable calibration of clinical FRAX-Canada in females at a baseline predicted risk above $5\%$ [106], and FRAX-Canada with BMD in females at a baseline predicted risk above 6 or $12\%$, depending on the study [106, 111].Fig. 2Calibration of clinical FRAX for the 10-year prediction of hip and clinical fragility fractures. Legend: Forest plots show the calibration ratios reported across the included studies; these were not pooled for the high risk of bias studies, and pooled for the studies without high risk of bias (reporting on FRAX-Canada)Fig. 3Calibration of FRAX with the incorporation of bone mineral density for the 10-year prediction of hip and clinical fragility fractures. Legend: Forest plots show the calibration ratios reported across the included studies; these were not pooled for the high risk of bias studies, and pooled for the studies without high risk of bias (reporting on FRAX-Canada) *There is* evidence to suggest acceptable calibration of FRAX to predict the 5-year risk of hip (FRAX + BMD only) and clinical fragility fractures (clinical FRAX and FRAX + BMD) (low certainty; most applicable to females) [129], but the prediction of 5-year risk is not a well-accepted or intended purpose of the tool. Findings on discrimination from Viswanathan 2018 [60] show an area under the curve (AUC) for the 10-year prediction of hip fractures in females of 0.76 ($95\%$ CI 0.72–0.81) for clinical FRAX and 0.79 ($95\%$ CI 0.76–0.81) for FRAX + BMD. The AUC for clinical fragility fractures in females was 0.67 ($95\%$ CI 0.65–0.68) for clinical FRAX and 0.70 (0.68–0.71) for FRAX + BMD [60]. Reported findings for males are in Additional file 4.
We are very uncertain about the ability of clinical Garvan (2 cohort; $$n = 67$$,923) [104, 113] and Garvan + BMD (5 cohort; $$n = 11$$,869) [100, 113, 119, 142, 145] to predict the 10-year risk of hip fractures and the 10-year risk of clinical fragility fractures [100, 113, 119, 142, 145]. Clinical Garvan (1 cohort; 1,054,815) [108] may underestimate the 5-year risk of hip fractures (O:E 2.17, $95\%$ CI 2.16 to 2.17; low certainty); evidence for calibration for 5-year risk of clinical fragility fractures is very uncertain [154]. The AUC for 10-year prediction of hip fractures reported by the USPSTF was 0.68 ($95\%$ CI not reported) for clinical Garvan and 0.73 for Garvan + BMD [60]. For clinical fragility fractures in females, the AUC was 0.66 ($95\%$ CI 0.61–0.72) for clinical Garvan and 0.68 for Garvan + BMD [60]. Data for males are in Additional file 4. There is evidence from one study ($$n = 34$$,060) to suggest that CAROC [128] (includes BMD) may be adequately calibrated to predict a category of 10-year risk of clinical fragility fracture; observed fracture risk ($95\%$ CI) was 6.4 (6.0–6.8)% in the low risk (<$10\%$) group, 13.8 (13.1–14.5)% in the moderate risk group (10–$20\%$), and 23.8 (22.5–25.0)% in the high-risk group (>$20\%$). The discrimination of this tool was not reported by the USPSTF [60]. There was very limited evidence for the remaining tools (QFracture [108, 113, 154], FRISC [149], FRC [110, 136] with and without BMD).
Additional file 4 contains the full analysis details for KQ3a, including GRADE Evidence Profiles and Summary of Findings Tables, with explanations for each rating and forest plots.
Additional file 4 contains the full analysis details for KQ3b, including GRADE Evidence Profiles and Summary of Findings Tables, with explanations for each rating.
Additional file 4 shows the full analysis details for KQ4, including GRADE Summary of Findings Tables, with explanations for each rating. One RCT ($$n = 258$$) [231] that included females aged 50–65 years (low risk based on age), revealed that this population had a strong intention to be screened over the next 5 years (mean [standard deviation] intention score 3.74 [0.96]/5). Participants were then provided a 1-page decision support sheet containing information on benefits in one of four formats (words, numbers, narrative, or framed narrative in terms of benefits of not screening). The sheet indicated that screening and treatment would be associated with a reduction in the risk of hip fractures by 2 per 1000 or “very few” females, and a reduction in other fractures in “few” females over 10 years. Risks were described as the potential for worry, minor stomach upset, and muscle or joint pain. Serious harms were described as rare—osteonecrosis of the jaw in 1 to 10 per 1000 or “very few” females and atypical fractures in 5 per 1000 or “very few” females over 10 years. Overdiagnosis was presented by showing that the incidence of low bone density (labelled as osteoporosis) exceeded important fracture outcomes. After reviewing the decision support sheet, participants’ intention to screen did not change substantially and also did not differ based on the format of information provided (1 study, $$n = 258$$; low certainty) [231].
Seven observational studies and two RCTs ($$n = 1930$$; sample size uncertain in one study) [220, 221, 224–230] reported on the acceptability of treatment. In five studies ($$n = 1010$$) [220, 221, 224, 229, 230], adults (primarily females) ≥50 years old were provided information on the benefits and harms of treatment in various formats; not all participants in these studies were considered to be at high fracture risk or eligible for treatment. In these studies, patients were asked to make hypothetical treatment decisions, with results of three studies showing that patients’ preference for treatment versus no treatment may be highly variable (3 studies, $$n = 317$$; low certainty) [220, 221, 224]. Two other studies showed that after receiving information on their personal fracture risk (median [IQR] 10-year hip fracture risk 2.2 [0.5–$2.7\%$] in one study, 5-year hip fracture risk 1.4 [0.8–$3.0\%$] in the other), relatively few (19 to $39\%$) patients may be willing to accept treatment (2 studies, $$n = 593$$; low certainty) [229, 230]. In the four remaining studies ($$n = 324$$; sample size uncertain in one study), postmenopausal females with low bone density (labelled as osteoporosis or osteopenia) who were in the position to make real-life decisions about treatment were provided decision aids outlining the potential benefits and harms of treatment. These studies showed that few (5–$20\%$) eligible patients who read decision aids and are aware of their fracture risk are willing to initiate treatment (2 studies, $$n = 240$$; sample size uncertain in one study [227, 228], but that somewhat more may be willing to start treatment when the decision aid is used during a clinical encounter (4–$44\%$ acceptance; 2 studies, $$n = 84$$ [225, 226] or when they have had a previous fracture or are at higher fracture risk (32–$45\%$; 1 study, $$n = 208$$) [53, 228]. Overall, a minority of postmenopausal females at increased risk for fracture may accept treatment (moderate certainty).
Three observational studies ($$n = 741$$) [220, 222, 224] reported on the minimum acceptable benefit of treatment among adults ≥50 years (mean 60 to 72 years) provided hypothetical scenarios about the benefits and harms of anti-osteoporosis treatment. These studies indicated that about two-thirds ($64\%$) of adults ≥50 years may have overly optimistic views of the benefits of treatment (1 study, $$n = 354$$) [222] and that these views may be highly variable (3 studies, $$n = 741$$; low certainty) [220, 222, 224]. For example, one study reported that patients may require a reduction of 20 to 200 fractures per 1000 to consider 10 years of bisphosphonate treatment with no major side effects to be acceptable (1 study, $$n = 354$$; low certainty) [222].
Six observational studies ($$n = 1091$$) [53, 220, 223, 226, 229, 230] reported on the level of risk at which treatment would be considered acceptable among adults ($97\%$ female) ≥45 years old who were aware of their personal fracture risk but not necessarily at high risk or making real-life treatment decisions. These studies reported that there is large heterogeneity in the level of risk at which treatment would be considered to be acceptable (6 studies, $$n = 1091$$; low certainty) [53, 220, 223, 226, 229, 230]. Many patients (19 to $51\%$) are willing to accept treatment even at low levels of fracture risk (5 to $20\%$); meanwhile, a large proportion (44 to $68\%$) of high-risk females (≥$3\%$ hip or ≥$20\%$ osteoporotic fracture risk; ≥$30\%$ in one study) would choose not to be treated (3 studies, $$n = 378$$; low certainty) [53, 226, 229].
## KQ2: How accurate are screening tests at predicting fractures among adults ≥40 years?
Of 6081 unique records retrieved by the searches for KQ2, we assessed 413 for eligibility by full text, and 59 external validation cohort studies [96–154] taking place in very high human development index countries with moderate fracture risk, met eligibility criteria for inclusion in the review (Fig. 1). From our search update in June 2021 when we changed our eligibility to Canadian reports of unique cohorts or that added data to that previously included, we included one study [154] and excluded 18 other reports [148, 155–171]. Studies excluded after full text appraisal are listed with reasons in Additional file 5. Among the initial set of included studies from our search in July 2019, there were several that analyzed cohorts with substantial overlap in participants. To prevent double-counting in the analysis, when cohorts were overlapping for a given tool-outcome comparison, we selected a single primary cohort study for analysis ($$n = 32$$) [98–100, 104, 106–113, 116, 117, 119, 128, 129, 134, 136, 138, 140, 142–146, 148–151, 153, 154]. We primarily considered recency in our choice of cohorts, but also considered the size of cohorts, quality of the methods (primarily more available data on predictors), and available outcomes. The remaining publications were then used for any reported supplementary data (e.g., calibration plots, subgroups of interest).
## KQ3a: What are the benefits of pharmacologic treatments to prevent fragility fractures among adults ≥40 years?
Of 11,693 unique records retrieved by the searches for KQ3a, we assessed 211 for eligibility by full text and included 27 RCTs [172–198] (one trial of alendronate was open-label [185]) and 11 associated publications [199–209] (Fig. 1). Studies excluded after full text appraisal are listed with reasons in Additional file 5.
## Bisphosphonates
In postmenopausal females at risk of fragility fractures, the risk of hip fractures may be reduced by median 2 (range 1 to 6) years of treatment with bisphosphonates as a class (alendronate, risedronate, or zoledronic acid; 14 RCTs; $$n = 21$$,038; 2.9 fewer in 1000, $95\%$ CI 4.6 fewer to 0.9 fewer; NNT=345; low certainty) compared to placebo [48, 172, 176, 177, 180, 181, 184, 186, 187, 189, 190, 193, 195–197, 201, 209]. Data for individual bisphosphonates showed that median 3 (range 1 to 3) years of treatment with risedronate may reduce the risk of hip fractures (4 RCTs; $$n = 9$$,672; 7.9 fewer in 1000, $95\%$ CI 13.0 fewer to 1.5 fewer; NNT=127; low certainty), but median 2 (range 1 to 4) years of treatment with alendronate and median 2 (range 2 to 6) years of treatment with zoledronic acid may not reduce the risk of hip fractures (low certainty). Within-study subgroup analyses were available for alendronate [177] and risedronate [189] (1 trial each) by age and baseline risk (BMD, prevalent fractures). These were not considered to be credible as they were available only in single trials (no evidence of consistency), may not have been adequately powered, and were not necessarily pre-specified (Additional file 4). One trial in males ($$n = 1199$$) showed that 2 years of treatment with zoledronic acid may not reduce the risk of hip fractures [175].
The risk of clinical fragility fractures in postmenopausal females is probably reduced by median 2 (range 1 to 6) years of treatment with bisphosphonates as a class (19 RCTs; $$n = 22$$,482; 11.1 fewer in 1000, $95\%$ CI 15.0 fewer to 6.6 fewer; NNT=90; moderate certainty) [172, 173, 177, 179–184, 186–190, 193–196, 200, 201, 203, 206, 209], median 2 (range 1 to 4) years of treatment with alendronate (8 RCTs; $$n = 8854$$; 14.7 fewer in 1000, $95\%$ CI 24.5 fewer to 2.6 fewer; NNT=68; moderate certainty) [172, 173, 177, 183, 184, 187, 193, 200, 203, 206, 209] and median 2 (range 1 to 6) years of treatment with zoledronic acid (5 RCTs; $$n = 3$$,218; 20.1 fewer in 1000, $95\%$ CI 27.6 fewer to 9.9 fewer; NNT=50; moderate certainty) compared to placebo [180, 181, 188, 194, 195, 201]. Median 2 (range 1 to 3) years of treatment with risedronate may reduce the risk of clinical fragility fractures (7 RCTs; $$n = 10$$,572; 7.8 fewer in 1000, $95\%$ CI 12.5 fewer to 2.3 fewer; NNT=128; low certainty) [179, 182, 183, 186, 189, 190, 196]. The analyses were robust to sensitivity analysis using only “nonvertebral fractures” in one trial of zoledronic acid where nonvertebral and vertebral fractures had been summed to determine the total number of people with fractures (could overestimate) [195]. One trial in males ($$n = 1199$$) showed that 2 years of treatment with zoledronic acid may not reduce the risk of clinical fragility fractures [175].
The risk of clinical vertebral fractures among postmenopausal females may be reduced by median 2 (range 1 to 6) years of treatment with bisphosphonates as a class (11 RCTs; $$n = 8921$$; 10.0 fewer in 1000, $95\%$ CI 14.0 fewer to 3.9 fewer; NNT=100; low certainty) [172, 176, 177, 179–181, 184, 194–197, 201, 203] and median 2 (range 1 to 6) years of treatment with zoledronic acid (4 RCTs; $$n = 2367$$; 18.7 fewer in 1000, $95\%$ CI 25.6 fewer to 6.6 fewer; NNT=53; low certainty) [180, 181, 194, 195]. The evidence for alendronate [172, 176, 177, 184, 197, 203] and risedronate [179, 196] is very uncertain. There were no studies in males that reported on clinical vertebral fractures.
Bisphosphonates as a class may not reduce the risk of all-cause mortality in postmenopausal females compared to placebo over 1 to 6 years of follow-up [176, 177, 180, 185, 188, 195–197, 202, 206]. Evidence for individual bisphosphonates is very uncertain (including for zoledronic acid in males).
The evidence was very uncertain for many adverse events, for example gastrointestinal cancers and several of the serious cardiovascular events. Compared to no treatment, alendronate may increase the risk of atypical subtrochanteric (0.08 more in 1000, $95\%$ CI 0.05 more to 0.14 more; systematic review of 1 cohort; $$n = 220$$,360; NNH=12,500; low certainty) [215] and femoral shaft fractures (0.06 more in 1000, $95\%$ CI 0.03 to 0.10; systematic review of 1 cohort; $$n = 220$$,360; NNH=16,667; low certainty) [215], and osteonecrosis of the jaw (systematic review of 1 cohort; $$n = 220$$,360; 0.22 more in 1000, $95\%$ CI 0.04 more to 0.59 more; NNH=4545; low certainty) [215]. The evidence for bisphosphonates as a class showed similar findings [48, 49, 211, 215]. The risk of “any serious adverse event” (composite outcome) is probably not increased with risedronate [37, 60] and zoledronic acid [37, 60] and may not be increased with alendronate [37, 60]. The risk of certain serious gastrointestinal adverse events (perforations, ulcers, and bleeds; serious esophageal) may not be increased with alendronate [48, 49, 211]. The risk of stroke and myocardial infarction probably does not increase with bisphosphonates as a class [216]; certainty was low for little-to-no difference in other serious cardiovascular events from individual drugs and from the drug class [48, 49, 211, 216].
The risk of non-serious gastrointestinal adverse events is probably increased by treatment with alendronate (systematic review of 50 RCTs; $$n = 22$$,549; 16.3 more in 1000, $95\%$ CI 2.4 more to 31.3 more; NNH=61; moderate certainty) [48, 49, 211], but probably not by treatment with risedronate [48, 49, 211]. Non-serious adverse events (composite outcome) are probably increased by treatment with zoledronic acid (systematic review of 6 RCTs; $$n = 9575$$; 51.8 more in 1000, $95\%$ CI no difference to 112.2 more; NNH=19; moderate certainty) [212], related to the potential increased risk of multiple influenza-like symptoms [48, 49, 211] including pyrexia [212], headache [212], chills [48, 49, 211], arthritis and arthralgia [48, 49, 211], and myalgia [48, 49, 211] (low-to-moderate certainty). With the exception of zoledronic acid, the risk of “any non-serious adverse event” (composite outcome) [212] and discontinuation due to adverse events [37, 60] do not appear to be increased by treatment with bisphosphonates (low-to-moderate certainty).
## Denosumab
In postmenopausal females the risk of hip fractures may not be reduced by median 1 (range 0.5 to 3) years of treatment with denosumab compared to placebo [178, 192, 198, 199, 207]. Within-study subgroup analyses were available by age, baseline BMD and FRAX score from one trial [178], but were not considered credible because there is no evidence that the effects are consistent as they have not been replicated in other trials (Additional file 4). The risk of clinical fragility fractures is probably reduced by median 1.5 (range 0.5 to 3) years of treatment with denosumab (6 RCTs; $$n = 9473$$; 9.1 fewer in 1000, $95\%$ CI 12.1 fewer to 5.6 fewer; NNT=110; moderate certainty) [174, 178, 185, 192, 198, 206, 207]. This analysis was robust to sensitivity analysis using only “nonvertebral” fractures for one trial [178] where vertebral and nonvertebral were summed to determine the total number of people with fractures. The risk of clinical vertebral fractures is probably reduced by median 1.5 (range 0.5 to 3) years of treatment with denosumab (4 RCTs; $$n = 8639$$; 16.0 fewer in 1000, $95\%$ CI 18.6 fewer to 12.1 fewer; NNT=62; moderate certainty) [174, 178, 192, 204, 205]. Denosumab probably does not reduce the risk of all-cause mortality over 0.5 to 3 years of follow-up [174, 178, 185, 192, 198, 205–207], and probably makes little-to-no difference in health-related quality of life over 3 years of follow-up [208]. The evidence for the effect of denosumab on the incidence of fractures (hip, clinical fragility, clinical vertebral) and all-cause mortality from one trial in males ($$n = 242$$) [191] is very uncertain.
The evidence was very uncertain for many adverse events, including serious infections [37, 60], venous thromboembolism [213], and rebound fractures after denosumab discontinuation [219]. Treatment with denosumab may not increase the risk of “any serious adverse event” (composite outcome) [37, 60] and does not appear to increase the risk of serious cardiovascular outcomes (stroke and various composite outcomes) [48, 49, 211, 213, 217] (low certainty).
The risks of non-serious gastrointestinal adverse events (systematic review of 3 RCTs; $$n = 8454$$; 64.5 more in 1000, $95\%$ CI 26.4 more to 113.3 more; NNH=16; moderate certainty) [48, 49, 211], rash or eczema (systematic review of 3 RCTs; $$n = 8454$$; 15.8 more in 1000, $95\%$ CI 7.6 more to 27.0 more; NNH=63; moderate certainty) [37, 60], and infections (any serious or non-serious; systematic review of 4 RCTs; $$n = 8691$$; 1.8 more per 1000, $95\%$ CI 0.1 more to 4.0 more; NNH=556; moderate certainty) [48, 49, 211] are probably increased by treatment with denosumab. Risks of any non-serious adverse event (composite outcome) [212, 213] and discontinuation due to adverse events [37, 60] do not appear to be increased by treatment with denosumab (moderate and low certainty, respectively).
## KQ3b: What are the harms of pharmacologic treatments to prevent fragility fractures among adults ≥40 years?
Of 721 unique records retrieved by the searches for KQ3b, we assessed 85 for eligibility by full text with 31 systematic reviews and one primary study meeting our eligibility criteria (Fig. 1). After reviewing these for key characteristics, we included 10 systematic reviews [60, 210–218], 3 associated publications [37, 48, 49], and one primary study on rebound fractures after discontinuation of denosumab [219]. Reviews excluded after full text appraisal, as well as systematic reviews that met inclusion criteria but were not selected for the overview, are listed with reasons in Additional file 5.
## KQ4: For adults ≥40 years, what is the acceptability of screening and/or initiating treatment to prevent fragility fractures when considering the possible benefits and harms from screening and/or treatment?
Of 8794 unique records retrieved by the searches for KQ4, we assessed 146 for eligibility by full text and included 12 studies (5 cross-sectional [220–224], 4 cohort [225–228], 3 RCTs [229–231]) and one associated publication of another study [53] (Fig. 1). Studies excluded after full text appraisal are listed with reasons in Additional file 5.
## Summary of principal findings for screening
In this review, we found that among a selected population of females aged 65 years and older who are willing to independently complete a mailed questionnaire about personal risk factors, an offer of 2-step screening using a fracture risk assessment tool (clinical FRAX) followed by assessment of BMD in those at increased risk (and treatment initiated based on various criteria) probably reduces the risk of hip (6.2 fewer in 1000, NNS=161) [4–6, 68, 91] and clinical fragility fractures (5.9 fewer in 1000, NNS=169) [4–6, 91] over 3 to 5 years of follow-up. The evidence is very uncertain for younger females [90] and for males [68]. A mailed offer of screening to females aged 68 to 80 years, where $54\%$ returned a completed questionnaire and were eligible, may not reduce the risk for hip or clinical fragility fractures over 5 years of follow-up [5]. Screening does not appear to make any difference in the risk of all-cause mortality nor wellbeing (very uncertain for younger females). The findings for the selected population (willing to independently complete clinical FRAX) are similar to those of a 2020 systematic review that pooled data only from the three most recent trials [7]. Minimal evidence related to the potential harms of screening is available; in one trial [6] no serious adverse events were reported but these did not appear to be collected systematically. Among selected females offered screening, $12\%$ of those meeting age-specific treatment thresholds based on FRAX 10-year hip fracture risk, and $19\%$ of those meeting thresholds based on FRAX 10-year major osteoporotic fracture risk, may be overdiagnosed according to our definition [4, 6, 59]. We did not locate convincing evidence to recommend one method of screening over another, although the evidence from the trials supports the use of clinical FRAX followed by BMD assessment in those at increased risk.
## Clinical considerations and implications
There appeared to be a considerable amount of ad hoc screening (and subsequent treatment; median $17\%$) in the control groups of the included trials; it is possible that the magnitude of effect would have been larger with a true “no screening” comparator. In all of the trials, the rate of completion of mailed risk assessment tools was low (generally less than two-thirds of those who were sent the tool), and 8 to $29\%$ of those eligible for BMD did not attend [4–6]. There appeared to be a healthy selection bias in several of the trials. For example, in the SALT trial $25\%$ of those who were offered DXA did not participate, and non-participants were among those at the highest fracture risk on clinical FRAX [4]. In the ROSE trial, the majority of fractures occurred in those who did not return the initial mailed risk assessment questionnaire [5]. In our review of the acceptability of screening, we similarly found that low risk (based on age) females have a high intention to be screened [231], but unfortunately we found no studies reporting on the intentions of higher-risk females. An analysis of non-participants in the ROSE trial showed that those who declined DXA scans were older, more likely to have comorbid conditions, had lower socioeconomic status, and were more likely to smoke and have high alcohol consumption [232]. Many of these factors may also place a person at increased risk for fracture. There are multiple reasons for which a person may choose not to be screened. For example, lack of interest may be related to a low perception (and perhaps underestimation) of personal fracture risk [232], the belief that low bone density is not a serious health issue [233], and fears of the potential serious harms of treatment despite their rare occurrence [234]. If screening for fracture risk is believed to be important, there may be a need to improve its accessibility for those at highest risk, and to attempt shared decision-making on the benefits and harms.
The mechanism by which the small reductions in fracture risk were achieved by screening is uncertain in light of other findings of this review. For example, among postmenopausal females, we found that treatment with bisphosphonates as a class may result in small reductions in the risk of hip (2.9 fewer in 1000; NNT=345) and clinical fragility fractures (11.1 fewer in 1000; NNT=90), of a magnitude similar to that seen in the screening trials, where only a small proportion of females were eligible for treatment and treated for a clinically meaningful length of time. In the screening context, we also observed an absolute risk reduction for hip fractures (6.2 per 1000) that was of similar magnitude to the reduction in clinical fragility fractures (5.9 per 1000) among females who independently completed the FRAX tool. The plausibility of this finding is difficult to ascertain. Notably, the one trial finding a statistically significant reduction in hip fracture risk with screening (SCOOP) did not find a similar reduction in the risk of clinical fractures [6], an outcome that occurs more frequently than hip fractures. It is possible that participants in this trial were better selected to benefit in terms of hip fracture reduction, because FRAX 10-year hip fracture risk was used in treatment thresholds, as opposed to 10-year major osteoporotic fracture risk used in the other trials. It is also possible that the treatments used in the trials were more effective at reducing hip rather than other clinical fractures, or simply that hip fractures were more reliably reported and ascertained than other fractures. Uncertainty remains because the trials do not provide information on which particular participants sustained fractures (i.e., those at increased risk or otherwise). Females in the screening trials may have been at higher risk overall than in the treatment trials due to older age (e.g., in SCOOP all were ≥70 years), though this is difficult to ascertain.
The effectiveness of screening may depend on uptake and persistence with anti-fracture treatments among those at high risk [50], but this tends to be suboptimal and declines with longer durations of treatment [51]. In the three more recent screening trials, uptake of anti-fracture drugs ranged from 69 to $80\%$ of those with a treatment indication [4–6]; however, these values could be overestimates as they were based on self-reports and prescription records. Longer-term follow-up from the SALT trial showed that by 36 months less than half ($43\%$) of those at high risk reported using anti-osteoporosis drugs [4]. In the larger treatment trials, full compliance with treatment was somewhat higher, ranging from about 50 to $80\%$ [177, 178, 189, 195]. One hypothesis is that the benefits seen from screening might be the result of unmeasured variables. For example, participation in screening may have provoked alterations in health behavior that helped participants to avoid fractures [235], like increasing weight-bearing exercise, stopping smoking, or taking preventive action to reduce the risk of falls. Post hoc analyses from the SCOOP trial showed, however, that screening had no significant impact on the risk of falls [236], and that the intervention was most beneficial in those at highest baseline hip fracture risk and those with prior fracture [91]. These findings suggest that the reduction in fracture risk seen with screening may be more related to treatment uptake and adherence (even if suboptimal) than other risk-reducing behaviors. It remains unclear from the trials whether the patients who sustained fractures were those who undertook treatment. It should be noted that decreased fracture risk in our review was only seen among highly motivated participants (those completing the clinical FRAX independently or accepting screening with BMD) who are probably more likely to adhere to treatment than the general population. The recent screening RCTs focused on treatment using first-line pharmacologic treatment and it is unclear what the impact may have been, if any, if they replaced this with or added therapies including vitamin D and calcium and/or interventions designed to prevent falls (e.g., exercise) or fractures from falls (e.g., hip protectors).
## Predictive value of screening strategies
If screening, overall, is believed to offer net benefit, there is limited certainty about which strategy to use. Two-step with risk assessment followed by BMD in those meeting a pre-determined risk threshold appears effective for reducing fractures, and the variable screening methods and treatment criteria in the trials suggest that some variation between strategies may be acceptable. The evidence from one comparative effectiveness trial suggests that BMD alone may be more effective than 2-step screening but we rated this evidence to be of very low certainty. The trials are most applicable to use of clinical FRAX for risk assessment and FRAX with BMD for treatment thresholds, and the evidence from KQ2 indicates that FRAX-Canada (with or without BMD) is probably well calibrated, with some underestimation, for the 10-year prediction of clinical fragility fractures [106, 111, 129]. Clinical FRAX-Canada may also be well calibrated, with some underestimation, to predict the 10-year risk of hip fracture, but the calibration of FRAX + BMD for this outcome may be poor [106, 111, 129]. One potential reason for the underestimation is lack of ability to incorporate a history of previous falls in FRAX; clinicians should be aware that those with previous falls may be at higher risk than estimated with FRAX [237]. The CAROC tool seemed to be adequately calibrated to predict a category of risk; however, it was not used in any of the included trials and requires the inclusion of BMD results. It was beyond the scope of this review to compare screening tools directly (e.g., with vs. without BMD); however, the evidence from this review showed clinical FRAX-Canada to be adequately calibrated without the addition of BMD. A review by Kanis et al. showed high concordance between risk categorization using either FRAX scores or BMD alone; people with higher scores are also generally those with a low BMD [238]. Also of interest is that in one of the trials (SCOOP) [6], only about one-third of those considered at high risk for 10-year hip fracture with clinical FRAX (using criteria suggested for treatment initiation in some cases [239]) were eligible for treatment (using only slightly different criteria) after their BMD results were incorporated into the predictions. Though not a focus of the current review, it is important to consider that the calibration of FRAX may vary by ethnicity. In a study using data from the Manitoba Bone Mineral Density Program registry, FRAX-Canada substantially overestimated 10-year risk of fracture in females who identified as Black or Asian as compared to White [240].
## Treatment effects
We found that treatment of postmenopausal females in a primary prevention population (<$50\%$ with prior fracture, but who are at risk of fragility fracture) with bisphosphonates as a class probably reduces the risk of clinical fragility fractures. Notably, our conclusion for the effect of bisphosphonates on the risk of hip fractures differs from the USPSTF who in 2018 reported low certainty evidence of no benefit [37]. We included additional trials in our analysis (including one large trial of zoledronic acid published after the USPSTF’s review was completed) and found a similar estimate of effect as the USPSTF but with improved precision, allowing for us to conclude that bisphosphonates may reduce the risk of hip fracture. Denosumab probably reduces the risk of clinical fragility fractures and clinical vertebral fractures, but may not reduce the risk of hip fractures. The limited evidence showed that zoledronic acid may not reduce the risk of hip or clinical fragility fractures in males with low BMD, and evidence for the use of denosumab in males was very uncertain. As reported in a recent review of risedronate for primary and secondary prevention of fractures [241], the trials for individual drugs are hampered by lack of power, as most studies focused on the impact of treatment on BMD as their main outcome of interest, rather than fractures which are then reported only as adverse events. Selection into treatment studies was often based on BMD, and no study used clinical risk scores to select patients. Similar to the screening trials, participants with prior fracture were often included, which differs somewhat from primary prevention where screening would be aimed at those without prior fracture. This review’s focus was determining estimates for the effects from the treatments used as first-line therapy in the RCTs on screening (mostly from anticipation of poor reporting on the harms), which largely employed pharmacologic treatment. Nevertheless, considering that most hip fractures occur as a direct result of a fall [242], preventing falls may be of value for people at high risk for fracture. The Task *Force is* currently developing recommendations about interventions for preventing falls [54].
## Patient perspectives
Though pharmacologic treatments appear to be beneficial, the magnitude of benefit may not be felt to be important enough to make treatment acceptable to patients. The most important findings of our acceptability review were that despite a high willingness to be screened among younger females, a minority of eligible older females may be willing to undergo treatment. Additionally, there was a large degree of variability in the level of risk at which individual patients would be willing to accept treatment (given information on benefits and/or harms). Many older adults have unrealistic views about the effectiveness of treatment and may require a reduction of 20 to 200 fractures per 1000 to consider 10 years of treatment with a bisphosphonate with no major side effects; this is at least double the magnitude of reduction in risk that was observed in our meta-analyses. Overall, though it was outside the scope of our review to determine the optimal length of treatment, a recent systematic review by Fink et al. found evidence of moderate certainty for no difference in the risk of clinical fragility fracture with 5 versus 10 years of treatment with alendronate and 3 versus 6 years of zoledronic acid [215]. There appeared to be some benefit of longer (10 vs. 5 years) treatment with alendronate on the risk of clinical vertebral fractures [215].
## Consideration of treatment harms and shared decision-making
Patients considering treatment should be able to weigh the proposed benefits with potential harms. We found increased risk for some non-serious adverse events; namely non-serious gastrointestinal events with alendronate; influenza-like symptoms with zoledronic acid; and non-serious gastrointestinal adverse events, dermatologic adverse events, and infections with denosumab. There was also low certainty evidence for an increased risk for the rare occurrence of atypical femoral fractures and osteonecrosis of the jaw with bisphosphonates (most evidence for alendronate). A concern about the risk of rebound fractures, and in particular multiple vertebral fractures, after cessation of treatment with denosumab has been raised by clinical experts [218, 243]. This requires more research focus as to date there is only minimal empiric evidence of very low certainty addressing these concerns; this finding was based on one available trial that compared discontinuation of denosumab with discontinuation of placebo (FREEDOM and its extension) [178, 244]. In this study, findings from patients initially randomized to denosumab or placebo who participated in the extension were analyzed for the occurrence of fractures after voluntary discontinuation (i.e., non-random sample). Ideally, trials would follow randomized participants from treatment initiation through an adequate time period after discontinuation to fully understand the net impact of denosumab treatment and subsequent discontinuation on the risk of fractures. The findings of our review also substantiate the large heterogeneity in the level of risk at which patients may accept treatment [52]. The finding that patients’ decisions about treatment may not correspond with guideline-recommended treatment thresholds [53, 225–227], and awareness of the complexity of decisions about treatment [245], supports the importance of shared decision-making about screening and subsequent treatment. A recent study of decision-making for osteoporosis treatment showed that allowing patients to make autonomous decisions after being provided information on the benefits and harms of treatment can result in better persistence with medication [246]. Most ($91\%$) of the females in this study who started pharmacotherapy continued to be treated after 1 year of follow-up [246].
## Strengths and limitations
We comprehensively reviewed evidence related to the benefits and harms of screening for the primary prevention of fragility fractures by first considering direct evidence from screening trials, and supplementing this by reviews on the accuracy of risk assessment, benefits and harms of treatment, and patient perceptions of the acceptability of screening and treatment. To our knowledge, this is the first systematic review to synthesize evidence on the calibration of fracture risk assessment tools. We implemented rigorous searches to locate all potentially relevant studies; though our searches were limited to English and French language studies, this has been shown not to bias the effect estimates from meta-analyses [247]. We limited our update search for the accuracy of risk assessment tools to Canadian studies because these were thought to be the most relevant; the studies included for other tools were all affected by serious risk of bias (among), such that conclusions were unlikely to be impacted by this limit. We did not update the evidence for KQ3a on the benefits of treatment because this data did not weigh heavily into the Task Force’s decision making for their guideline on screening, for which there were several RCTs. Since we took a rapid approach to KQ3b (harms of treatment), there is the small possibility that relevant systematic reviews were missed or that minor errors were overlooked; by using an experienced reviewer, we reduced the likelihood of major omissions that would impact the findings [248]. It is also possible that the evidence for this KQ was less up to date (versus using primary studies) or did not examine all outcomes of interest that could be available in primary studies; moderate certainty of evidence would suggest stable findings for several outcomes. For KQ2 (accuracy of risk prediction tools), we did not review discrimination as it was not rated as critical or important by the Task Force; reported findings from the USPSTF review [60] are therefore less up to date.
There was some indirectness in our findings due to populations, interventions, comparators, and outcomes differing from those of primary interest. Our findings focus mainly on a selected population of patients who completed a mailed clinical FRAX tool independently and who are likely to be more compliant with screening and potentially treatment than the general population. This differs to some extent from clinical practice, where ideally decisions about screening would be made in shared decision-making with between patients and providers, after which patients would have the opportunity to consider their level of risk, along with their perceived benefits and harms of treatment. In addition, some participants in the screening trials had previously used anti-osteoporosis drugs, and the comparator included ad hoc treatment. Across all KQs, the ascertainment of clinical fragility fractures was problematic; definitions differed across studies and in some cases could have included non-clinical vertebral fractures, or other fractures that were not related to fragility (e.g., due to trauma). Our findings were robust to sensitivity analyses removing studies with unclear ascertainment of outcomes, or including only a single type of fracture (e.g., if multiple were added to determine a total number, rather than number of patients with ≥1 fracture). There was concern for selective reporting across some outcomes. Minimal discussion of potential harms was included across the screening trials; in the treatment trials, it was often unclear whether fracture data were collected systematically, and many did not report on clinical vertebral fractures (though this information should have been available).
The evidence in this review is most applicable to postmenopausal females aged 65 and over. We located very limited evidence for males and younger females, and there were no screening trial data specific to females aged 55 to 65 years. In addition, though one trial provided evidence of increased effectiveness of screening among those at higher baseline risk, there is a need for analyses from other trials to substantiate these findings. There is a need for robust comparative effectiveness trials to inform the most effective screening strategy. Examining whether different treatment approaches may positively impact effects for those at high risk based on screening for fracture risk, especially for those individuals nonadherent or uninterested in anti-osteoporosis medications, may also be of value.
## Conclusion
Screening in primary care using clinical FRAX, followed by BMD assessment in those at increased risk, among selected females aged 65 years and older who are likely to be more compliant with screening (as ascertained by their willingness to independently complete a risk assessment questionnaire) probably results in a small reduction in the risk of clinical fragility fracture and hip fracture compared to no screening. This may differ to some extent from clinical practice, where healthcare providers would ideally engage in shared decision-making about screening and discuss the results of fracture risk estimation, as well as the risks and benefits of treatment, during the patient consultation. A mailed offer of screening in the general population, where uptake was relatively low, did not improve any patient-important outcomes. Minimal information on harms is available, although our calculated estimates of overdiagnosis were 12 and $19\%$ for hip and major osteoporotic fractures, respectively. The mechanism of the reduction in risk with screening is not fully clear, though there is some evidence to suggest it may be attributed to pharmacologic treatment rather than a reduction in falls or other risk behaviors. It is not clear which screening strategy would be most beneficial. The screening trials used diverse criteria when deciding for whom to offer treatment. There is some evidence for clinical FRAX and FRAX + BMD being adequately calibrated (particularly for clinical fragility fractures), with some underestimation, among Canadian studies; CAROC seems adequately calibrated to predict a category of risk and requires a BMD measurement. Treatment with bisphosphonates in primary prevention populations (at risk, but without prior fracture) probably reduces the risk of clinical fragility fractures and may reduce the risk of hip fractures and clinical vertebral fractures among postmenopausal females, to a similar magnitude as seen in the screening trials. Denosumab probably reduces the risk of clinical fragility fractures and clinical vertebral fractures but may not reduce the risk of hip fractures in postmenopausal females; evidence for males is very uncertain. Females at low risk seem to have a high willingness to be screened but there is large heterogeneity in the level of risk at which higher-risk patients would accept treatment, supporting a shared decision-making approach. The findings of this review will be used, among several other considerations (e.g., information on issues of feasibility, acceptability, costs/resources, and equity) by the Canadian Task Force on Preventive Health Care to inform recommendations on screening for the prevention of fragility fractures among adults 40 years and older in primary care in Canada.
## Supplementary Information
Additional file 1. PRISMA checklist. This file shows the completed reporting checklist for the systematic reviews. Additional file 2. Search strategy for KQ3b. This file shows the full search strategy for key question 3b on the harms of pharmacologic treatment. Additional file 3. Evidence sets for KQ1a-b. GRADE Evidence Profiles and Summary of Findings for KQ1. This file shows the analyses and GRADE Evidence Profiles, Summary of Findings, and contributing data for key question 1a on the benefits and harms of screening, and key question 1b on the comparative benefits and harms of different screening approaches. Additional file 4. Evidence sets for KQs 2-4. GRADE Summary of Findings for KQs 2-4. This file shows the analyses and GRADE Summary of Findings for key questions 2 (accuracy of screening tests), 3a (benefits of treatment), 3b (harms of treatment), and 4 (acceptability of screening/treatment).Additional file 5. Excluded studies list. This file lists all studies excluded after full text appraisal, with reasons. Additional file 6. KQ 2-4 study characteristics. This file shows the descriptive characteristics of the included studies for key questions 2 (accuracy of screening tests), 3a (benefits of treatment), 3b (harms of treatment), and 4 (acceptability of screening/treatment).Additional file 7. KQ 2-4 Risk of bias assessments. This file shows the risk of bias/quality appraisals for key questions 2 (accuracy of screening tests), 3a (benefits of treatment), 3b (harms of treatment), and 4 (acceptability of screening/treatment).Additional file 8. Responses to stakeholder comments. This file contains responses to the comments submitted by stakeholder during their review of a draft of this manuscript. Written informed consent to publish was obtained from the stakeholders who provided the stakeholder reviews. A copy of the written consent is available for review by the Editors-in-Chief of this journal. The stakeholder reviews have been anonymized.
## References
1. Leslie WD, Schousboe JT. **A review of osteoporosis diagnosis and treatment options in new and recently updated guidelines on case finding around the world**. *Curr Osteoporos Rep.* (2011.0) **9** 129-140. DOI: 10.1007/s11914-011-0060-5
2. Kanis JA. **Diagnosis of osteoporosis and assessment of fracture risk**. *Lancet.* (2002.0) **359** 1929-1936. DOI: 10.1016/S0140-6736(02)08761-5
3. Crandall CJ, Larson J, Manson JE, Cauley JA, LaCroix AZ, Wactawski-Wende J. **A comparison of US and Canadian osteoporosis screening and treatment strategies in postmenopausal women**. *J Bone Miner Res.* (2019.0) **34** 607-615. PMID: 30536628
4. Merlijn T, Swart KMA, van Schoor NM, Heymans MW, van der Zwaard BC, van der Heijden AA. **The effect of a screening and treatment program for the prevention of fractures in older women: a randomized pragmatic trial**. *J Bone Miner Res.* (2019.0) **20** 20
5. Rubin KH, Rothmann MJ, Holmberg T, Hoiberg M, Moller S, Barkmann R. **Effectiveness of a two-step population-based osteoporosis screening program using FRAX: the randomized Risk-stratified Osteoporosis Strategy Evaluation (ROSE) study**. *Osteoporos Int.* (2018.0) **29** 567-578. DOI: 10.1007/s00198-017-4326-3
6. Shepstone L, Lenaghan E, Cooper C, Clarke S, Fong-Soe-Khioe R, Fordham R. **Screening in the community to reduce fractures in older women (SCOOP): a randomised controlled trial**. *Lancet.* (2018.0) **391** 741-747. DOI: 10.1016/S0140-6736(17)32640-5
7. Merlijn T, Swart KMA, van der Horst HE, Netelenbos JC, Elders PJM. **Fracture prevention by screening for high fracture risk: a systematic review and meta-analysis**. *Osteoporos Int.* (2020.0) **31** 251-257. DOI: 10.1007/s00198-019-05226-w
8. Lentle B, Cheung AM, Hanley DA, Leslie WD, Lyons D, Papaioannou A. **Osteoporosis Canada 2010 guidelines for the assessment of fracture risk**. *Can Assoc Radiol J.* (2011.0) **62** 243-250. DOI: 10.1016/j.carj.2011.05.001
9. Friedman SM, Mendelson DA. **Epidemiology of fragility fractures**. *Clin Geriatr Med.* (2014.0) **30** 175-181. DOI: 10.1016/j.cger.2014.01.001
10. McCloskey EV, Vasireddy S, Threlkeld J, Eastaugh J, Parry A, Bonnet N. **Vertebral fracture assessment (VFA) with a densitometer predicts future fractures in elderly women unselected for osteoporosis**. *J Bone Miner Res.* (2008.0) **23** 1561-1568. DOI: 10.1359/jbmr.080515
11. Morin SN, Lix LM, Leslie WD. **The importance of previous fracture site on osteoporosis diagnosis and incident fractures in women**. *J Bone Miner Res.* (2014.0) **29** 1675-1680. DOI: 10.1002/jbmr.2204
12. Hodsman AB, Leslie WD, Tsang JF, Gamble GD. **10-year probability of recurrent fractures following wrist and other osteoporotic fractures in a large clinical cohort: an analysis from the manitoba bone density program**. *Arch Intern Med.* (2008.0) **168** 2261-2267. DOI: 10.1001/archinte.168.20.2261
13. Hippisley-Cox J, Coupland C. **Predicting risk of osteoporotic fracture in men and women in England and Wales: prospective derivation and validation of QFractureScores**. *BMJ.* (2009.0) **339** b4229. DOI: 10.1136/bmj.b4229
14. Robbins J, Aragaki AK, Kooperberg C. **Factors associated with 5-year risk of hip fracture in postmenopausal women**. *JAMA.* (2007.0) **298** 2389-2398. DOI: 10.1001/jama.298.20.2389
15. Prior JC, Langsetmo L, Lentle BC, Berger C, Goltzman D, Kovacs CS. **Ten-year incident osteoporosis-related fractures in the population-based Canadian Multicentre Osteoporosis Study—comparing site and age-specific risks in women and men**. *Bone.* (2015.0) **71** 237-243. DOI: 10.1016/j.bone.2014.10.026
16. Langdahl BL. **Osteoporosis in premenopausal women**. *Curr Opin Rheumatol.* (2017.0) **29** 410-415. DOI: 10.1097/BOR.0000000000000400
17. Cohen A. **Premenopausal osteoporosis**. *Endocrinol Metab Clin North Am.* (2017.0) **46** 117-133. DOI: 10.1016/j.ecl.2016.09.007
18. 18.World Health Organization. WHO Technical Report Series: assessment of fracture risk and its application to screening for postmenopausal osteoporosis. Geneva: WHO; 1994.
19. Curtis EM, Moon RJ, Harvey NC, Cooper C. **The impact of fragility fracture and approaches to osteoporosis risk assessment worldwide**. *Bone.* (2017.0) **104** 29-38. DOI: 10.1016/j.bone.2017.01.024
20. Odén A, McCloskey EV, Kanis JA, Harvey NC, Johansson H. **Burden of high fracture probability worldwide: secular increases 2010-2040**. *Osteoporos Int.* (2015.0) **26** 2243-2248. DOI: 10.1007/s00198-015-3154-6
21. 21.Public Health Agency of Canada. Public Health Infobase: Canadian Chronic Disease Indicators. Ottawa: PHAC; 2018.
22. Adachi JD, Adami S, Gehlbach S, Anderson FA, Boonen S, Chapurlat RD. **Impact of prevalent fractures on quality of life: baseline results from the global longitudinal study of osteoporosis in women**. *Mayo Clin Proc.* (2010.0) **85** 806-813. DOI: 10.4065/mcp.2010.0082
23. Ioannidis G, Papaioannou A, Hopman WM, Akhtar-Danesh N, Anastassiades T, Pickard L. **Relation between fractures and mortality: results from the Canadian Multicentre Osteoporosis Study**. *CMAJ.* (2009.0) **181** 265-271. DOI: 10.1503/cmaj.081720
24. Papaioannou A, Kennedy CC, Ioannidis G, Sawka A, Hopman WM, Pickard L. **The impact of incident fractures on health-related quality of life: 5 years of data from the Canadian Multicentre Osteoporosis Study**. *Osteoporos Int.* (2009.0) **20** 703-714. DOI: 10.1007/s00198-008-0743-7
25. Tarride JE, Hopkins RB, Leslie WD, Morin S, Adachi JD, Papaioannou A. **The burden of illness of osteoporosis in Canada**. *Osteoporos Int.* (2012.0) **23** 2591-2600. DOI: 10.1007/s00198-012-1931-z
26. Njeh CF, Fuerst T, Hans D, Blake GM, Genant HK. **Radiation exposure in bone mineral density assessment**. *Appl Radiat Isot.* (1999.0) **50** 215-236. DOI: 10.1016/S0969-8043(98)00026-8
27. Hansen CA, Abrahamsen B, Konradsen H, Pedersen BD. **Women’s lived experiences of learning to live with osteoporosis: a longitudinal qualitative study**. *BMC Womens Health.* (2017.0) **17** 17. DOI: 10.1186/s12905-017-0377-z
28. Barker KL, Toye F, Lowe CJM. **A qualitative systematic review of patients’ experience of osteoporosis using meta-ethnography**. *Arch Osteoporos.* (2016.0) **11** 33. DOI: 10.1007/s11657-016-0286-z
29. Hayawi LM, Graham ID, Tugwell P, Yousef AS. **Screening for osteoporosis: a systematic assessment of the quality and content of clinical practice guidelines, using the AGREE II instrument and the IOM Standards for Trustworthy Guidelines**. *PLoS one.* (2018.0) **13** e0208251. DOI: 10.1371/journal.pone.0208251
30. Papaioannou A, Morin S, Cheung AM, Atkinson S, Brown JP, Feldman S. **2010 clinical practice guidelines for the diagnosis and management of osteoporosis in Canada: summary**. *CMAJ.* (2010.0) **182** 1864-1873. DOI: 10.1503/cmaj.100771
31. **Screening for osteoporosis to prevent fractures: US Preventive Services Task Force recommendation statement**. *JAMA.* (2018.0) **319** 2521-2531. DOI: 10.1001/jama.2018.7498
32. **Management of osteoporosis in postmenopausal women: 2006 position statement of The North American Menopause Society**. *Menopause (New York, NY).* (2006.0) **13** 340. DOI: 10.1097/01.gme.0000222475.93345.b3
33. Siminoski K, O'Keeffe M, Brown JP, Burrell S, Coupland D, Dumont M. **Canadian Association of Radiologists technical standards for bone mineral densitometry reporting**. *Can Assoc Radiol J.* (2013.0) **64** 281-294. DOI: 10.1016/j.carj.2013.07.006
34. 34.The International Society for Clinical Densitometry. ISCD official position: Adults 2015 [Available from: https://www.iscd.org/official-positions/2015-iscd-official-positions-adult/].
35. Watts NB, Adler RA, Bilezikian JP, Drake MT, Eastell R, Orwoll ES. **Osteoporosis in men: an Endocrine Society clinical practice guideline**. *J Clin Endocrinol Metab.* (2012.0) **97** 1802-1822. DOI: 10.1210/jc.2011-3045
36. Rubin KH, Friis-Holmberg T, Hermann AP, Abrahamsen B, Brixen K. **Risk assessment tools to identify women with increased risk of osteoporotic fracture: complexity or simplicity? A systematic review**. *J Bone Miner Res.* (2013.0) **28** 1701-1717. DOI: 10.1002/jbmr.1956
37. Viswanathan M, Reddy S, Berkman N, Cullen K, Middleton JC, Nicholson WK. **Screening to prevent osteoporotic fractures: updated evidence report and systematic review for the US Preventive Services Task Force**. *JAMA.* (2018.0) **319** 2532-2551. DOI: 10.1001/jama.2018.6537
38. Moons KGM, de Groot JAH, Bouwmeester W, Vergouwe Y, Mallett S, Altman DG. **Critical appraisal and data extraction for systematic reviews of prediction modelling studies: The CHARMS checklist**. *PLOS Med.* (2014.0) **11** e1001744. DOI: 10.1371/journal.pmed.1001744
39. Kanis JA, Harvey NC, Cooper C, Johansson H, Oden A, McCloskey EV. **A systematic review of intervention thresholds based on FRAX : a report prepared for the National Osteoporosis Guideline Group and the International Osteoporosis Foundation**. *Arch Osteoporos.* (2016.0) **11** 25. DOI: 10.1007/s11657-016-0278-z
40. Siminoski K, Leslie WD, Frame H, Hodsman A, Josse RG, Khan A. **Recommendations for bone mineral density reporting in Canada**. *Can Assoc Radiol J.* (2005.0) **56** 178-188. PMID: 16144280
41. Cosman F, de Beur SJ, LeBoff MS, Lewiecki EM, Tanner B, Randall S. **Clinician's guide to prevention and treatment of osteoporosis**. *Osteoporos Int.* (2014.0) **25** 2359-2381. DOI: 10.1007/s00198-014-2794-2
42. Compston J, Cooper A, Cooper C, Gittoes N, Gregson C, Harvey N. **UK clinical guideline for the prevention and treatment of osteoporosis**. *Arch Osteoporos.* (2017.0) **12** 43. DOI: 10.1007/s11657-017-0324-5
43. Chakhtoura M, Baddoura R, El-Hajj Fuleihan G. *Lebanese FRAX-Based Osteoporosis Guidelines* (2013.0)
44. Kanis JA, Cooper C, Rizzoli R, Reginster JY. **European guidance for the diagnosis and management of osteoporosis in postmenopausal women**. *Osteoporos Int.* (2019.0) **30** 3-44. DOI: 10.1007/s00198-018-4704-5
45. Beithon J, Gallenberg M, Johnson K, Kildahl P, Krenik J, Liebow M. *Institute for Clinical Systems Improvement: diagnosis and treatment of osteoporosis* (2017.0)
46. Eastell R, Rosen CJ, Black DM, Cheung AM, Murad MH, Shoback D. **Pharmacological management of osteoporosis in postmenopausal women: An Endocrine Society Clinical Practice Guideline**. *J Clin Endocrinol Metab.* (2019.0) **104** 1595-1622. DOI: 10.1210/jc.2019-00221
47. Camacho PM, Petak SM, Binkley N, Clarke BL, Harris ST, Hurley DL. **American Association of Clinical Endocrinologists and American College of Endocrinology clinical practice guidelines for the diagnosis and treatment of postmenopausal osteoporosis—2016**. *Endocr Pract.* (2016.0) **22** 1-42. DOI: 10.4158/EP161435.GL
48. Qaseem A, Forciea M, McLean RM, Denberg TD. **Treatment of low bone density or osteoporosis to prevent fractures in men and women: a clinical practice guideline update from the American College of Physicians**. *Ann Intern Med.* (2017.0) **166** 818-839. DOI: 10.7326/M15-1361
49. Crandall CJ, Newberry SJ, Diamant A, Lim YW, Gellad WF, Suttorp MJ. *Treatment to prevent fractures in men and women with low bone density or osteoporosis: Update of a 2007 Report* (2012.0)
50. Keshishian A, Boytsov N, Burge R, Krohn K, Lombard L, Zhang X. **Examining the effect of medication adherence on risk of subsequent fracture among women with a fragility fracture in the U.S. Medicare Population**. *J Manag Care Spec Pharm* (2017.0) **23** 1178-1190. PMID: 29083977
51. Fatoye F, Smith P, Gebrye T, Yeowell G. **Real-world persistence and adherence with oral bisphosphonates for osteoporosis: a systematic review**. *BMJ Open.* (2019.0) **9** e027049. DOI: 10.1136/bmjopen-2018-027049
52. Hiligsmann M, Bours SPG, Boonen A. **A review of patient preferences for osteoporosis drug treatment**. *Curr Rheumatol Rep.* (2015.0) **17** 61. DOI: 10.1007/s11926-015-0533-0
53. 53.Billington EO, Feasel AL, Kline GA. At odds about the odds: women’s choices to accept osteoporosis medications do not closely agree with physician-set treatment thresholds. J Gen Intern Med. 2020;35(1):276-82.
54. Pillay J, Riva JJ, Tessier LA, Colquhoun H, Lang E, Moore AE. **Fall prevention interventions for older community-dwelling adults: systematic reviews on benefits, harms, and patient values and preferences**. *Syst Rev.* (2021.0) **10** 18. DOI: 10.1186/s13643-020-01572-7
55. Gates M, Pillay J, Thériault G, Limburg H, Grad R, Klarenbach S. **Screening to prevent fragility fractures among adults 40 years and older in primary care: protocol for a systematic review**. *Syst Rev.* (2019.0) **8** 216. DOI: 10.1186/s13643-019-1094-5
56. 56.Canadian Task Force on Preventive Health Care. Procedure Manual 2014 [Available from: https://canadiantaskforce.ca/methods/].
57. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD. **The PRISMA 2020 statement: an updated guideline for reporting systematic reviews**. *BMJ.* (2021.0) **372** n71. DOI: 10.1136/bmj.n71
58. Guyatt GH, Oxman AD, Kunz R, Atkins D, Brozek J, Vist G. **GRADE guidelines: 2. Framing the question and deciding on important outcomes**. *J Clin Epidemiol.* (2011.0) **64** 395-400. DOI: 10.1016/j.jclinepi.2010.09.012
59. 59.Theriault G, Reynolds D, Pillay J, Limburg H, Grad R, Gates M, et al. Expanding the measurement of overdiagnosis in the context of disease precursors and risk factors. BMJ Evid Based Med. 2023;bmjebm-2022-112117. 10.1136/bmjebm-2022-112117.
60. Viswanathan M, Reddy S, Berkman N, Cullen K, Middleton JC, Nicholson WK. *Screening to prevent osteoporotic fractures: an evidence review for the US Preventive Services Task Force* (2018.0)
61. Whiting P, Jelena S, Higgins LPT, Caldwell DM, Reeves BC, Shea B. **ROBIS: a new tool to assess risk of bias in systematic reviews was developed**. *J Clin Epidemiol.* (2016.0) **69** 225-234. DOI: 10.1016/j.jclinepi.2015.06.005
62. McGowan J, Sampson M, Salzwedel DM, Cogo E, Foerster V, Lefebvre C. **PRESS peer review of electronic search strategies: 2015 guideline statement**. *J Clin Epidemiol.* (2016.0) **75** 40-46. DOI: 10.1016/j.jclinepi.2016.01.021
63. Higgins JPT, Altman DG, Gøtzsche PC, Jüni P, Moher D, Oxman AD. **The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials**. *BMJ.* (2011.0) **343** d5928. DOI: 10.1136/bmj.d5928
64. Moons KM, Wolff RF, Riley RD. **Probast: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration**. *Ann Intern Med.* (2019.0) **170** W1-W33. DOI: 10.7326/M18-1377
65. Shea BJ, Reeves BC, Wells G, Thuku M, Hamel C, Moran J. **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
66. Zhang Y, Alonso-Coello P, Guyatt GH, Yepes-Nuñez JJ, Akl EA, Hazlewood G. **GRADE Guidelines: 19. Assessing the certainty of evidence in the importance of outcomes or values and preferences-Risk of bias and indirectness**. *J Clin Epidemiol.* (2019.0) **111** 94-104. DOI: 10.1016/j.jclinepi.2018.01.013
67. DerSimonian R, Laird N. **Meta-analysis in clinical trials**. *Control Clin Trials.* (1986.0) **7** 177-188. DOI: 10.1016/0197-2456(86)90046-2
68. Kern LM, Powe NR, Levine MA, Fitzpatrick AL, Harris TB, Robbins J. **Association between screening for osteoporosis and the incidence of hip fracture**. *Ann Intern Med.* (2005.0) **142** 173-181. DOI: 10.7326/0003-4819-142-3-200502010-00007
69. Sweeting M, Sutton A, Lambert P. **What to add to nothing? Use and avoidance of continuity corrections in meta-analysis of sparse data**. *Stat Med.* (2004.0) **23** 1351-1375. DOI: 10.1002/sim.1761
70. Schünemann HJ, Higgins J, Vist G, Glasziou P, Akl E, Skoetz N. **Chapter 14: Completing ‘Summary of findings’ tables and grading the certainty of the evidence**. *Cochrane Handbook for Systematic Reviews of Interventions version 61, Cochrane* (2020.0)
71. 71.Statistics CanadaTable 13-10-0710-01: Deaths and mortality rate, by age group2018. *Table 13-10-0710-01: Deaths and mortality rate, by age group* (2018.0)
72. Schünemann HJ, Vist G, Higgins J, Santesso N, Deeks JJ, Glasziou P. **Chapter 15: Interpreting results and drawing conclusions**. *Cochrane Handbook for Systematic Reviews of Interventions version 61: Cochrane* (2020.0)
73. Cornell JE, Mulrow CD, Localio R, Stack CB, Meibohm AR, Guallar E. **Random-effects meta-analysis of inconsistent effects: a time for change**. *Ann Intern Med.* (2014.0) **160** 267-270. DOI: 10.7326/M13-2886
74. IntHout J, Ioannidis JP, Borm GF. **The Hartung-Knapp-Sidik-Jonkman method for random effects meta-analysis is straightforward and considerably outperforms the standard DerSimonian-Laird method**. *BMC Med Res Methodol.* (2014.0) **14** 25. DOI: 10.1186/1471-2288-14-25
75. Snell KIE. *Development and application of statistical methods for prognosis research (Doctoral Thesis)* (2015.0)
76. van Klaveren D, Steyerberg EW, Perel P, Vergouwe Y. **Assessing discriminative ability of risk models in clustered data**. *BMC Med Res Methodol.* (2014.0) **14** 5. DOI: 10.1186/1471-2288-14-5
77. Qin G, Hotilovac L. **Comparison of non-parametric confidence intervals for the area under the ROC curve of a continuous-scale diagnostic test**. *Stat Methods Med Res.* (2008.0) **17** 207-221. DOI: 10.1177/0962280207087173
78. Popay J, Roberts H, Sowden A, Petticrew M, Arai L, Rodgers M. *Guidance on the conduct of narrative synthesis in systematic reviews: a product from the ESRC Methods Programme* (2006.0)
79. Harbord RM, Egger M, Sterne JA. **A modified test for small-study effects in meta-analyses of controlled trials with binary endpoints**. *Stat Med.* (2006.0) **25** 3443-3457. DOI: 10.1002/sim.2380
80. Guyatt GH, Oxman AD, Kunz R, Brozek J, Alonso-Coello P, Rind D. **GRADE guidelines 6. Rating the quality of evidence—imprecision**. *J Clin Epidemiol.* (2011.0) **64** 1283-1293. DOI: 10.1016/j.jclinepi.2011.01.012
81. Guyatt GH, Oxman AD, Kunz R, Woodcock J, Brozek J, Helfand M. **GRADE guidelines: 8. Rating the quality of evidence—indirectness**. *J Clin Epidemiol.* (2011.0) **64** 1303-1310. DOI: 10.1016/j.jclinepi.2011.04.014
82. Guyatt GH, Oxman AD, Kunz R, Woodcock J, Brozek J, Helfand M. **GRADE guidelines: 7. Rating the quality of evidence--inconsistency**. *J Clin Epidemiol.* (2011.0) **64** 1294-1302. DOI: 10.1016/j.jclinepi.2011.03.017
83. Guyatt GH, Oxman AD, Montori V, Vist G, Kunz R, Brozek J. **GRADE guidelines: 5. Rating the quality of evidence—publication bias**. *J Clin Epidemiol.* (2011.0) **64** 1277-1282. DOI: 10.1016/j.jclinepi.2011.01.011
84. Guyatt GH, Oxman AD, Vist G, Kunz R, Brozek J, Alonso-Coello P. **GRADE guidelines: 4. Rating the quality of evidence—study limitations (risk of bias)**. *J Clin Epidemiol.* (2011.0) **64** 407-415. DOI: 10.1016/j.jclinepi.2010.07.017
85. Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y, Alonso-Coello P. **GRADE: an emerging consensus on rating quality of evidence and strength of recommendations**. *BMJ.* (2008.0) **336** 924-926. DOI: 10.1136/bmj.39489.470347.AD
86. Hultcrantz M, Rind D, Akl EA, Treweek S, Mustafa RA, Iorio A. **The GRADE Working Group clarifies the construct of certainty of evidence**. *J Clin Epidemiol.* (2017.0) **87** 4-13. DOI: 10.1016/j.jclinepi.2017.05.006
87. Murad MH, Mustafa RA, Schünemann HJ, Sultan S, Santesso N. **Rating the certainty in evidence in the absence of a single estimate of effect**. *Evid Based Med.* (2017.0) **22** 85-87. DOI: 10.1136/ebmed-2017-110668
88. Debray TPA, Damen JAAG, Snell KIE, Ensor J, Hooft L, Reitsma JB. **A guide to systematic review and meta-analysis of prediction model performance**. *BMJ.* (2017.0) **356** i6460. DOI: 10.1136/bmj.i6460
89. Santesso N, Glenton C, Dahm P, Garner P, Akl EA, Alper B. **GRADE guidelines 26: informative statements to communicate the findings of systematic reviews of interventions**. *J Clin Epidemiol.* (2020.0) **119** 126-135. DOI: 10.1016/j.jclinepi.2019.10.014
90. Barr RJ, Stewart A, Torgerson DJ, Reid DM. **Population screening for osteoporosis risk: a randomised control trial of medication use and fracture risk**. *Osteoporos Int.* (2010.0) **21** 561-568. DOI: 10.1007/s00198-009-1007-x
91. McCloskey E, Johansson H, Harvey NC, Shepstone L, Lenaghan E, Fordham R. **Management of patients with high baseline hip fracture risk by FRAX reduces hip fractures-a post hoc analysis of the SCOOP study**. *J Bone Miner Res.* (2018.0) **33** 1020-1026. DOI: 10.1002/jbmr.3411
92. Torgerson DJ, Thomas RE, Campbell MK, Reid DM. **Randomized trial of osteoporosis screening. Use of hormone replacement therapy and quality-of-life results**. *Arch Intern Med.* (1997.0) **157** 2121-2125. DOI: 10.1001/archinte.1997.00440390121015
93. Lacroix AZ, Buist DS, Brenneman SK, Abbott TA. **Evaluation of three population-based strategies for fracture prevention: results of the osteoporosis population-based risk assessment (OPRA) trial**. *Med Care.* (2005.0) **43** 293-302. DOI: 10.1097/00005650-200503000-00012
94. Kanis JA, Odén A, McCloskey EV, Johansson H, Wahl DA, Cooper C. **A systematic review of hip fracture incidence and probability of fracture worldwide**. *Osteoporos Int.* (2012.0) **23** 2239-2256. DOI: 10.1007/s00198-012-1964-3
95. Fried LP, Borhani NO, Enright P, Furberg CD, Gardin JM, Kronmal RA. **The Cardiovascular Health Study: design and rationale**. *Ann Epidemiol.* (1991.0) **1** 263-276. DOI: 10.1016/1047-2797(91)90005-W
96. Azagra R, Roca G, Encabo G, Aguye A, Zwart M, Guell S. **FRAX(R) tool, the WHO algorithm to predict osteoporotic fractures: the first analysis of its discriminative and predictive ability in the Spanish FRIDEX cohort**. *BMC Musculoskelet Disord.* (2012.0) **13** 204. DOI: 10.1186/1471-2474-13-204
97. Azagra R, Roca G, Martin-Sanchez JC, Casado E, Encabo G, Zwart M. **FRAX(R) thresholds to identify people with high or low risk of osteoporotic fracture in Spanish female population**. *Med Clin.* (2015.0) **144** 1-8. DOI: 10.1016/j.medcli.2013.11.014
98. Azagra R, Zwart M, Aguye A, Martin-Sanchez JC, Casado E, Diaz-Herrera MA. **Fracture experience among participants from the FROCAT study: what thresholding is appropriate using the FRAX tool?**. *Maturitas.* (2016.0) **83** 65-71. DOI: 10.1016/j.maturitas.2015.10.002
99. Azagra R, Zwart M, Encabo G, Aguye A, Martin-Sanchez JC, Puchol-Ruiz N. **Rationale of the Spanish FRAX model in decision-making for predicting osteoporotic fractures: an update of FRIDEX cohort of Spanish women**. *BMC Musculoskelet Disord.* (2016.0) **17** 262. DOI: 10.1186/s12891-016-1096-6
100. Bolland MJ, Siu AT, Mason BH, Horne AM, Ames RW, Grey AB. **Evaluation of the FRAX and Garvan fracture risk calculators in older women**. *J Bone Miner Res.* (2011.0) **26** 420-427. DOI: 10.1002/jbmr.215
101. Bolton JM, Morin SN, Majumdar SR, Sareen J, Lix LM, Johansson H. **Association of mental disorders and related medication use with risk for major osteoporotic fractures**. *JAMA Psychiatry.* (2017.0) **74** 641-648. DOI: 10.1001/jamapsychiatry.2017.0449
102. Brennan SL, Leslie WD, Lix LM, Johansson H, Oden A, McCloskey E. **FRAX provides robust fracture prediction regardless of socioeconomic status**. *Osteoporos Int.* (2014.0) **25** 61-69. DOI: 10.1007/s00198-013-2525-0
103. Buehring B, Hansen KE, Lewis BL, Cummings SR, Lane NE, Binkley N. **Dysmobility syndrome independently increases fracture risk in the osteoporotic fractures in men (MrOS) prospective cohort study**. *J Bone Miner Res.* (2018.0) **33** 1622-1629. DOI: 10.1002/jbmr.3455
104. Crandall CJ, Larson J, LaCroix A, Cauley JA, LeBoff MS, Li W. **Predicting fracture risk in younger postmenopausal women: comparison of the Garvan and FRAX Risk Calculators in the Women's Health Initiative Study**. *J Gen Intern Med.* (2019.0) **34** 235-242. DOI: 10.1007/s11606-018-4696-z
105. Crandall CJ, Larson JC, Watts NB, Gourlay ML, Donaldson MG, LaCroix A. **Comparison of fracture risk prediction by the US Preventive Services Task Force strategy and two alternative strategies in women 50-64 years old in the Women's Health Initiative**. *J Clin Endocrinol Metab.* (2014.0) **99** 4514-4522. DOI: 10.1210/jc.2014-2332
106. Crandall CJ, Schousboe JT, Morin SN, Lix LM, Leslie W. **Performance of FRAX and FRAX-based treatment thresholds in women aged 40 years and older: the Manitoba BMD Registry**. *J Bone Miner Res.* (2019.0) **28** 28
107. Czerwinski E, Borowy P, Kumorek A, Amarowicz J, Gorkiewicz M, Milert A. **Fracture risk prediction in outpatients from Krakow Region using FRAX tool versus fracture risk in 11-year follow-up**. *Ortop Traumatol Rehabil.* (2013.0) **15** 617-628. DOI: 10.5604/15093492.1091517
108. Dagan N, Cohen-Stavi C, Leventer-Roberts M, Balicer RD. **External validation and comparison of three prediction tools for risk of osteoporotic fractures using data from population based electronic health records: retrospective cohort study**. *BMJ.* (2017.0) **356** i6755. DOI: 10.1136/bmj.i6755
109. Ettinger B, Ensrud KE, Blackwell T, Curtis JR, Lapidus JA, Orwoll ES. **Performance of FRAX in a cohort of community-dwelling, ambulatory older men: the Osteoporotic Fractures in Men (MrOS) study**. *Osteoporos Int.* (2013.0) **24** 1185-1193. DOI: 10.1007/s00198-012-2215-3
110. Ettinger B, Liu H, Blackwell T, Hoffman AR, Ensrud KE, Orwoll ES. **Validation of FRC, a fracture risk assessment tool, in a cohort of older men: the Osteoporotic Fractures in Men (MrOS) Study**. *J Clin Densitom.* (2012.0) **15** 334-342. DOI: 10.1016/j.jocd.2012.01.011
111. Fraser LA, Langsetmo L, Berger C, Ioannidis G, Goltzman D, Adachi JD. **Fracture prediction and calibration of a Canadian FRAX(R) tool: a population-based report from CaMos**. *Osteoporos Int.* (2011.0) **22** 829-837. DOI: 10.1007/s00198-010-1465-1
112. Goldshtein I, Gerber Y, Ish-Shalom S, Leshno M. **Fracture risk assessment with FRAX using real-world data in a population-based cohort from Israel**. *Am J Epidemiol.* (2018.0) **187** 94-102. DOI: 10.1093/aje/kwx128
113. Gourlay ML, Ritter VS, Fine JP, Overman RA, Schousboe JT, Cawthon PM. **Comparison of fracture risk assessment tools in older men without prior hip or spine fracture: the MrOS study**. *Arch Osteoporos.* (2017.0) **12** 91. DOI: 10.1007/s11657-017-0389-1
114. Harvey NC, Oden A, Orwoll E, Lapidus J, Kwok T, Karlsson MK. **Falls predict fractures independently of FRAX probability: a Meta-Analysis of the Osteoporotic Fractures in Men (MrOS) study**. *J Bone Miner Res.* (2018.0) **33** 510-516. DOI: 10.1002/jbmr.3331
115. Hillier TA, Cauley JA, Rizzo JH, Pedula KL, Ensrud KE, Bauer DC. **WHO absolute fracture risk models (FRAX): do clinical risk factors improve fracture prediction in older women without osteoporosis?**. *J Bone Miner Res.* (2011.0) **26** 1774-1782. DOI: 10.1002/jbmr.372
116. Holloway KL, Mohebbi M, Betson AG, Hans D, Hyde NK, Brennan-Olsen SL. **Prediction of major osteoporotic and hip fractures in Australian men using FRAX scores adjusted with trabecular bone score**. *Osteoporos Int.* (2018.0) **29** 101-108. DOI: 10.1007/s00198-017-4226-6
117. Iki M, Fujita Y, Tamaki J, Kouda K, Yura A, Sato Y. **Trabecular bone score may improve FRAX(R) prediction accuracy for major osteoporotic fractures in elderly Japanese men: the Fujiwara-kyo Osteoporosis Risk in Men (FORMEN) Cohort Study**. *Osteoporos Int.* (2015.0) **26** 1841-1848. DOI: 10.1007/s00198-015-3092-3
118. Kalvesten J, Lui LY, Brismar T, Cummings S. **Digital X-ray radiogrammetry in the study of osteoporotic fractures: Comparison to dual energy X-ray absorptiometry and FRAX**. *Bone.* (2016.0) **86** 30-35. DOI: 10.1016/j.bone.2016.02.011
119. Langsetmo L, Nguyen TV, Nguyen ND, Kovacs CS, Prior JC, Center JR. **Independent external validation of nomograms for predicting risk of low-trauma fracture and hip fracture**. *CMAJ.* (2011.0) **183** E107-E114. DOI: 10.1503/cmaj.100458
120. Langsetmo L, Peters KW, Burghardt AJ, Ensrud KE, Fink HA, Cawthon PM. **Volumetric bone mineral density and failure load of distal limbs predict incident clinical fracture independent HR-pQCT BMD and failure load predicts incident clinical fracture of FRAX and clinical risk factors among older men**. *J Bone Miner Res.* (2018.0) **33** 1302-1311. DOI: 10.1002/jbmr.3433
121. Leslie WD, Berger C, Langsetmo L, Lix LM, Adachi JD, Hanley DA. **Construction and validation of a simplified fracture risk assessment tool for Canadian women and men: results from the CaMos and Manitoba cohorts**. *Osteoporos Int.* (2011.0) **22** 1873-1883. DOI: 10.1007/s00198-010-1445-5
122. Leslie WD, Brennan SL, Lix LM, Johansson H, Oden A, McCloskey E. **Direct comparison of eight national FRAX(R) tools for fracture prediction and treatment qualification in Canadian women**. *Arch Osteoporos.* (2013.0) **8** 145. DOI: 10.1007/s11657-013-0145-0
123. Leslie WD, Johansson H, Kanis JA, Lamy O, Oden A, McCloskey EV. **Lumbar spine texture enhances 10-year fracture probability assessment**. *Osteoporos Int.* (2014.0) **25** 2271-2277. DOI: 10.1007/s00198-014-2761-y
124. Leslie WD, Lix LM. **Simplified 10-Year Absolute Fracture Risk Assessment: A Comparison of Men and Women**. *J Clin Densitom.* (2010.0) **13** 141-146. DOI: 10.1016/j.jocd.2010.02.002
125. Leslie WD, Lix LM, Johansson H, Oden A, McCloskey E, Kanis JA. **Independent clinical validation of a Canadian FRAX tool: fracture prediction and model calibration**. *J Bone Miner Res.* (2010.0) **25** 2350-2358. DOI: 10.1002/jbmr.123
126. Leslie WD, Lix LM, Johansson H, Oden A, McCloskey E, Kanis JA. **A comparative study of using non-hip bone density inputs with FRAX(R)**. *Osteoporos Int.* (2012.0) **23** 853-860. DOI: 10.1007/s00198-011-1814-8
127. Leslie WD, Lix LM, Majumdar SR, Morin SN, Johansson H, Oden A. **Total hip bone area affects fracture prediction with FRAX in Canadian white women**. *J Clin Endocrinol Metab.* (2017.0) **102** 4242-4249. DOI: 10.1210/jc.2017-01327
128. Leslie WD, Majumdar SR, Lix LM, Josse RG, Johansson H, Oden A. **Direct comparison of FRAX(R) and a simplified fracture risk assessment tool in routine clinical practice: a registry-based cohort study**. *Osteoporos Int.* (2016.0) **27** 2689-2695. DOI: 10.1007/s00198-016-3613-8
129. Leslie WD, Majumdar SR, Morin SN, Lix LM, Johansson H, Oden A. **FRAX for fracture prediction shorter and longer than 10 years: the Manitoba BMD registry**. *Osteoporos Int.* (2017.0) **28** 2557-2564. DOI: 10.1007/s00198-017-4091-3
130. Leslie WD, Majumdar SR, Morin SN, Lix LM, Schousboe JT, Ensrud KE. **Performance of FRAX in clinical practice according to sex and osteoporosis definitions: the Manitoba BMD registry**. *Osteoporos Int.* (2018.0) **29** 759-767. DOI: 10.1007/s00198-018-4415-y
131. Leslie WD, Morin S. **Fracture burden in relation to low bone mineral density and FRAX((R)) probability**. *J Clin Densitom.* (2011.0) **14** 279-285. DOI: 10.1016/j.jocd.2011.04.010
132. Leslie WD, Morin S, Lix LM, Johansson H, Oden A, McCloskey E. **Fracture risk assessment without bone density measurement in routine clinical practice**. *Osteoporos Int.* (2012.0) **23** 75-85. DOI: 10.1007/s00198-011-1747-2
133. Leslie WD, Tsang JF, Lix LM. **Simplified system for absolute fracture risk assessment: clinical validation in Canadian women**. *J Bone Miner Res.* (2009.0) **24** 353-360. DOI: 10.1359/jbmr.081012
134. Li G, Thabane L, Papaioannou A, Adachi JD. **Comparison between frailty index of deficit accumulation and fracture risk assessment tool (FRAX) in prediction of risk of fractures**. *Bone.* (2015.0) **77** 107-114. DOI: 10.1016/j.bone.2015.04.028
135. Lix LM, Leslie WD, Majumdar SR. **Measuring improvement in fracture risk prediction for a new risk factor: a simulation**. *BMC Res Notes.* (2018.0) **11** 62. DOI: 10.1186/s13104-018-3178-z
136. Lo JC, Pressman AR, Chandra M, Ettinger B. **Fracture risk tool validation in an integrated healthcare delivery system**. *Am J Manag Care.* (2011.0) **17** 188-194. PMID: 21504255
137. Majumdar SR, Leslie WD, Lix LM, Morin SN, Johansson H, Oden A. **Longer duration of diabetes strongly impacts fracture risk assessment: the Manitoba BMD Cohort**. *J Clin Endocrinol Metab.* (2016.0) **101** 4489-4496. DOI: 10.1210/jc.2016-2569
138. Marques A, Lucas R, Simoes E, Verstappen SMM, Jacobs JWG, da Silva JAP. **Do we need bone mineral density to estimate osteoporotic fracture risk? A 10-year prospective multicentre validation study**. *RMD Open.* (2017.0) **3** e000509. DOI: 10.1136/rmdopen-2017-000509
139. Martineau P, Leslie WD, Johansson H, Oden A, McCloskey EV, Hans D. **Clinical utility of using lumbar spine trabecular bone score to adjust fracture probability: the Manitoba BMD Cohort**. *J Bone Miner Res.* (2017.0) **32** 1568-1574. DOI: 10.1002/jbmr.3124
140. Melton LJ, Atkinson EJ, Achenbach SJ, Kanis JA, Therneau TM, Johansson H. **Potential Extensions of the US FRAX Algorithm**. *J Osteoporos.* (2012.0) **2012** 528790. DOI: 10.1155/2012/528790
141. Orwoll ES, Lapidus J, Wang PY, Vandenput L, Hoffman A, Fink HA. **The limited clinical utility of testosterone, estradiol, and sex hormone binding globulin measurements in the prediction of fracture risk and bone loss in older men**. *J Bone Miner Res.* (2017.0) **32** 633-640. DOI: 10.1002/jbmr.3021
142. Pluskiewicz W, Adamczyk P, Czekajlo A, Grzeszczak W, Drozdzowska B. **High fracture probability predicts fractures in a 4-year follow-up in women from the RAC-OST-POL study**. *Osteoporos Int.* (2015.0) **26** 2811-2820. DOI: 10.1007/s00198-015-3196-9
143. Premaor M, Parker RA, Cummings S, Ensrud K, Cauley JA, Lui LY. **Predictive value of FRAX for fracture in obese older women**. *J Bone Miner Res.* (2013.0) **28** 188-195. DOI: 10.1002/jbmr.1729
144. Pressman AR, Lo JC, Chandra M, Ettinger B. **Methods for assessing fracture risk prediction models: experience with FRAX in a large integrated health care delivery system**. *J Clin Densitom.* (2011.0) **14** 407-415. DOI: 10.1016/j.jocd.2011.06.006
145. Reyes Dominguez AI, Sosa Cabrera N, Saavedra Santana P, de Tejada Romero MJG, Jodar Gimeno E, Sosa HM. **Assessment of the predictive capacity of the garvan calculator of 10 year risk of fracture in a Spanish population**. *Rev de Osteoporos Metab Miner.* (2017.0) **9** 55-61
146. Sornay-Rendu E, Munoz F, Delmas PD, Chapurlat RD. **The FRAX tool in French women: how well does it describe the real incidence of fracture in the OFELY cohort?**. *J Bone Miner Res.* (2010.0) **25** 2101-2107. DOI: 10.1002/jbmr.106
147. Tamaki J, Iki M, Kadowaki E, Sato Y, Kajita E, Kagamimori S. **Fracture risk prediction using FRAX(R): a 10-year follow-up survey of the Japanese Population-Based Osteoporosis (JPOS) Cohort Study**. *Osteoporos Int.* (2011.0) **22** 3037-3045. DOI: 10.1007/s00198-011-1537-x
148. Tamaki J, Iki M, Sato Y, Winzenrieth R, Kajita E, Kagamimori S. **Does Trabecular Bone Score (TBS) improve the predictive ability of FRAX for major osteoporotic fractures according to the Japanese Population-Based Osteoporosis (JPOS) cohort study?**. *J Bone Miner Metab.* (2019.0) **37** 161-170. DOI: 10.1007/s00774-018-0910-7
149. Tanaka S, Yoshimura N, Kuroda T, Hosoi T, Saito M, Shiraki M. **The Fracture and Immobilization Score (FRISC) for risk assessment of osteoporotic fracture and immobilization in postmenopausal women--A joint analysis of the Nagano, Miyama, and Taiji Cohorts**. *Bone.* (2010.0) **47** 1064-1070. DOI: 10.1016/j.bone.2010.08.019
150. Tebe Cordomi C, Del Rio LM, Di Gregorio S, Casas L, Estrada MD, Kotzeva A. **Validation of the FRAX predictive model for major osteoporotic fracture in a historical cohort of Spanish women**. *J Clin Densitom.* (2013.0) **16** 231-237. DOI: 10.1016/j.jocd.2012.05.007
151. Tremollieres FA, Pouilles JM, Drewniak N, Laparra J, Ribot CA, Dargent-Molina P. **Fracture risk prediction using BMD and clinical risk factors in early postmenopausal women: sensitivity of the WHO FRAX tool**. *J Bone Miner Res.* (2010.0) **25** 1002-1009. DOI: 10.1002/jbmr.12
152. Yang S, Leslie WD, Morin SN, Lix LM. **Administrative healthcare data applied to fracture risk assessment**. *Osteoporos Int.* (2019.0) **30** 565-571. DOI: 10.1007/s00198-018-4780-6
153. Yin MT, Shiau S, Rimland D, Gibert CL, Bedimo RJ, Rodriguez-Barradas MC. **Fracture prediction with Modified-FRAX in older HIV-infected and uninfected men**. *J Acquir Immune Defic Syndr.* (2016.0) **72** 513-520. DOI: 10.1097/QAI.0000000000000998
154. Desbiens L-C, Mac-Way F, Sidibe A, Beaudoin C, Jean S. **Comparison of fracture prediction tools in individuals without and with early chronic kidney disease: a population-based analysis of CARTaGENE**. *J Bone Miner Res.* (2020.0) **35** 1048-1057. DOI: 10.1002/jbmr.3977
155. Chapurlat R, Bui M, Sornay-Rendu E, Zebaze R, Delmas PD, Liew D. **Deterioration of cortical and trabecular microstructure identifies women with osteopenia or normal bone mineral density at imminent and long-term risk for fragility fracture: a prospective study**. *J Bone Miner Res.* (2020.0) **35** 833-844. DOI: 10.1002/jbmr.3924
156. Crandall CJ, Larson J, Cauley JA, Schousboe JT, LaCroix AZ, Robbins JA. **Do additional clinical risk factors improve the performance of fracture risk assessment tool (FRAX) among postmenopausal women? Findings from the Women's Health Initiative Observational Study and Clinical Trials**. *JBMR Plus.* (2019.0) **3** E10239. DOI: 10.1002/jbm4.10239
157. Dagan N, Elnekave E, Barda N, Bregman-Amitai O, Bar A, Orlovsky M. **Automated opportunistic osteoporotic fracture risk assessment using computed tomography scans to aid in FRAX underutilization**. *Nat Med.* (2020.0) **26** 77-82. DOI: 10.1038/s41591-019-0720-z
158. de Abreu LLF, Holloway-Kew KL, Sajjad MA, Kotowicz MA, Pasco JA. **FRAX (Australia) scores in women with impaired fasting glucose and diabetes**. *Bone Rep.* (2019.0) **11** 100223. DOI: 10.1016/j.bonr.2019.100223
159. Harvey NC, Kanis JA, Liu E, Cooper C, Lorentzon M, Bea JW. **Predictive value of DXA appendicular lean mass for incident fractures, falls, and mortality, independent of prior falls, FRAX, and BMD: findings from the Women's Health Initiative (WHI)**. *J Bone Miner Res.* (2021.0) **36** 654-661. DOI: 10.1002/jbmr.4239
160. Holloway-Kew KL, Zhang Y, Betson AG, Anderson KB, Hans D, Hyde NK. **How well do the FRAX (Australia) and Garvan calculators predict incident fractures? Data from the Geelong Osteoporosis Study**. *Osteoporos Int.* (2019.0) **30** 2129-2139. DOI: 10.1007/s00198-019-05088-2
161. Kim H, Kim JH, Kim MJ, Hong AR, Choi H, Ku E. **Low predictive value of FRAX adjusted by trabecular bone score for osteoporotic fractures in Korean women: a community-based cohort study**. *Endocrinol Metab (Seoul, Korea).* (2020.0) **35** 359-366. DOI: 10.3803/EnM.2020.35.2.359
162. Lu T, Forgetta V, Keller-Baruch J, Nethander M, Bennett D, Forest M. **Improved prediction of fracture risk leveraging a genome-wide polygenic risk score**. *Genome Med.* (2021.0) **13** 16. DOI: 10.1186/s13073-021-00838-6
163. Pickhardt PJ, Graffy PM, Zea R, Lee SJ, Liu J, Sandfort V. **Automated abdominal CT imaging biomarkers for opportunistic prediction of future major osteoporotic fractures in asymptomatic adults**. *Radiology.* (2020.0) **297** 64-72. DOI: 10.1148/radiol.2020200466
164. Pluskiewicz W, Adamczyk P, Drozdzowska B. **Height loss in postmenopausal women-do we need more for fracture risk assessment? Results from the GO Study**. *Osteoporos Int.* (2021.0) **32** 2043-2049. DOI: 10.1007/s00198-021-05941-3
165. Xiao X, Wu Q. **The utility of genetic risk score to improve performance of FRAX for fracture prediction in US postmenopausal women**. *Calcif Tissue Int.* (2021.0) **108** 746-756. DOI: 10.1007/s00223-021-00809-4
166. Bisson EJ, Finlayson M, Ekuma O, Marrie RA, Leslie WD. **Accuracy of FRAX in people with multiple sclerosis: a Manitoba BMD registry-based cohort study**. *J Bone Miner Res.* (2018.0) **33** 284
167. Leslie WD, Lix LM, Binkley N. **Targeted vertebral fracture assessment for optimizing fracture prevention in Canada**. *Arch Osteoporos.* (2020.0) **15** 65. DOI: 10.1007/s11657-020-00735-2
168. Leslie WD, Morin SN, Lix LM, Binkley N. **Comparison of treatment strategies and thresholds for optimizing fracture prevention in Canada: a simulation analysis**. *Arch Osteoporos.* (2019.0) **15** 4. DOI: 10.1007/s11657-019-0660-8
169. Leslie WD, Morin SN, Lix LM, Binkley N. **Impact of spine-hip discordance on fracture risk assessment and treatment qualification in Canada: the Manitoba BMD registry**. *Arch Osteoporos.* (2020.0) **15** 85. DOI: 10.1007/s11657-020-00763-y
170. Leslie WD, Morin SN, Lix LM, Martineau P, Bryanton M, McCloskey EV. **Fracture prediction from self-reported falls in routine clinical practice: a registry-based cohort study**. *Osteoporos Int.* (2019.0) **30** 2195-2203. DOI: 10.1007/s00198-019-05106-3
171. Li G, Leslie WD, Kovacs CS, Prior J, Josse RG, Towheed T. **Combining frailty and trabecular bone score did not improve predictive accuracy in risk of major osteoporotic fractures**. *J Bone Miner Res.* (2020.0) **35** 1058-1064. DOI: 10.1002/jbmr.3971
172. Ascott-Evans BH, Guañabens N, Kivinen S, Stuckey BGA, Magaril CH, Vandormael K. **Alendronate prevents loss of bone density associated with discontinuation of hormone replacement therapy: a randomized controlled trial**. *Arch Intern Med.* (2003.0) **163** 789-794. DOI: 10.1001/archinte.163.7.789
173. Bell NH, Bilezikian JP, Bone Iii HG, Kaur A, Maragoto A, Santora AC. **Alendronate increases bone mass and reduces bone markers in postmenopausal African-American women**. *J Clin Endocrinol Metab.* (2002.0) **87** 2792-2797. DOI: 10.1210/jcem.87.6.8575
174. Bone HG, Bolognese MA, Yuen CK, Kendler DL, Wang H, Liu Y. **Effects of denosumab on bone mineral density and bone turnover in postmenopausal women**. *J Clin Endocrinol Metab.* (2008.0) **93** 2149-2157. DOI: 10.1210/jc.2007-2814
175. Boonen S, Reginster J-Y, Kaufman J-M, Lippuner K, Zanchetta J, Langdahl B. **Fracture risk and zoledronic acid therapy in men with osteoporosis**. *N Engl J Med.* (2012.0) **367** 1714-1723. DOI: 10.1056/NEJMoa1204061
176. Chesnut CH, McClung MR, Ensrud KE, Bell NH, Genant HK, Harris ST. **Alendronate treatment of the postmenopausal osteoporotic woman: effect of multiple dosages on bone mass and bone remodeling**. *Am J Med.* (1995.0) **99** 144-152. DOI: 10.1016/S0002-9343(99)80134-X
177. Cummings SR, Black DM, Thompson DE, Applegate WB, Barrett-Connor E, Musliner TA. **Effect of alendronate on risk of fracture in women with low bone density but without vertebral fractures results from the Fracture Intervention Trial**. *JAMA.* (1998.0) **280** 2077-2082. DOI: 10.1001/jama.280.24.2077
178. Cummings SR, Martin JS, McClung MR, Siris ES, Eastell R, Reid IR. **Denosumab for prevention of fractures in postmenopausal women with osteoporosis**. *N Engl J Med.* (2009.0) **361** 756-765. DOI: 10.1056/NEJMoa0809493
179. Fogelman I, Ribot C, Smith R, Ethgen D, Sod E, Reginster JY. **Risedronate reverses bone loss in postmenopausal women with low bone mass: results from a multinational, double-blind, placebo-controlled trial. BMD-MN Study Group**. *J Clin Endocrinol Metab.* (2000.0) **85** 1895-1900. PMID: 10843171
180. Grey A, Bolland M, Mihov B, Wong S, Horne A, Gamble G. **Duration of antiresorptive effects of low-dose zoledronate in osteopenic postmenopausal women: a randomized, placebo-controlled trial**. *J Bone Miner Res.* (2014.0) **29** 166-172. DOI: 10.1002/jbmr.2009
181. Grey A, Bolland MJ, Wattie D, Horne A, Gamble G, Reid IR. **The antiresorptive effects of a single dose of zoledronate persist for two years: a randomized, placebo-controlled trial in osteopenic postmenopausal women**. *Obstet Gynecol Surv.* (2009.0) **64** 462-463. DOI: 10.1097/01.ogx.0000351672.54929.75
182. Hooper MJ, Ebeling PR, Roberts AP, Graham JJ, Nicholson GC, D'Emden M. **Risedronate prevents bone loss in early postmenopausal women: a prospective randomized, placebo-controlled trial**. *Climacteric.* (2005.0) **8** 251-262. DOI: 10.1080/13697130500118126
183. Hosking D, Adami S, Felsenberg D, Andia JC, Valimaki M, Benhamou L. **Comparison of change in bone resorption and bone mineral density with once-weekly alendronate and daily risedronate: a randomised, placebo-controlled study**. *Curr Med Res Opin.* (2003.0) **19** 383-394. DOI: 10.1185/030079903125002009
184. Hosking D, Chilvers CE, Christiansen C, Ravn P, Wasnich R, Ross P. **Prevention of bone loss with alendronate in postmenopausal women under 60 years of age. Early Postmenopausal Intervention Cohort Study Group**. *N Engl J Med.* (1998.0) **338** 485-492. DOI: 10.1056/NEJM199802193380801
185. Lewiecki EM, Miller PD, McClung MR, Cohen SB, Bolognese MA, Liu Y. **Two-year treatment with denosumab (AMG 162) in a randomized phase 2 study of postmenopausal women with low BMD**. *J Bone Miner Res.* (2007.0) **22** 1832-1841. DOI: 10.1359/jbmr.070809
186. Li Y, Zhang Z, Deng X, Chen L. **Efficacy and safety of risedronate sodium in treatment of postmenopausal osteoporosis**. *J Huazhong Univ Sci Technol Med Sci.* (2005.0) **25** 527-529. DOI: 10.1007/BF02896007
187. Liberman UA, Weiss SR, Bröll J, Minne HW, Quan H, Bell NH. **Effect of oral alendronate on bone mineral density and the incidence of fractures in postmenopausal osteoporosis**. *N Engl J Med.* (1995.0) **333** 1437-1444. DOI: 10.1056/NEJM199511303332201
188. McClung M, Miller P, Recknor C, Mesenbrink P, Bucci-Rechtweg C, Benhamou CL. **Zoledronic acid for the prevention of bone loss in postmenopausal women with low bone mass: a randomized controlled trial**. *Obstet Gynecol.* (2009.0) **114** 999-1007. DOI: 10.1097/AOG.0b013e3181bdce0a
189. McClung MR, Geusens P, Miller PD, Zippel H, Bensen WG, Roux C. **Effect of risedronate on the risk of hip fracture in elderly women**. *N Engl J Med.* (2001.0) **344** 333-340. DOI: 10.1056/NEJM200102013440503
190. Mortensen L, Charles P, Bekker PJ, Digennaro J, Johnston CC. **Risedronate increases bone mass in an early postmenopausal population: two years of treatment plus one year of follow-up**. *J Clin Endocrinol Metab.* (1998.0) **83** 396-402. PMID: 9467547
191. Orwoll E, Teglbjaerg CS, Langdahl BL, Chapurlat R, Czerwinski E, Kendler DL. **A randomized, placebo-controlled study of the effects of denosumab for the treatment of men with low bone mineral density**. *J Clin Endocrinol Metab.* (2012.0) **97** 3161-3169. DOI: 10.1210/jc.2012-1569
192. Pitale S, Thomas M, Rathi G, Deshmukh V, Kumar P, Reddy S. **A randomized placebo-controlled trial of the efficacy of denosumab in Indian postmenopausal women with osteoporosis**. *Indian J Endocrinol Metab.* (2015.0) **19** 148-154. DOI: 10.4103/2230-8210.146871
193. Pols HAP, Felsenberg D, Hanley DA, Štepán J, Muñoz-Torres M, Wilkin TJ. **Multinational, placebo-controlled, randomized trial of the effects of alendronate on bone density and fracture risk in postmenopausal women with low bone mass: results of the FOSIT Study**. *Osteoporos Int.* (1999.0) **9** 461-468. DOI: 10.1007/PL00004171
194. Reid IR, Brown JP, Burckhardt P, Horowitz Z, Richardson P, Trechsel U. **Intravenous zoledronic acid in postmenopausal women with low bone mineral density**. *N Engl J Med.* (2002.0) **346** 653-661. DOI: 10.1056/NEJMoa011807
195. Reid IR, Horne AM, Mihov B, Stewart A, Garratt E, Wong S. **Fracture prevention with zoledronate in older women with osteopenia**. *N Engl J Med.* (2018.0) **379** 2407-2416. DOI: 10.1056/NEJMoa1808082
196. Välimäiki MJ, Farrerons-Minguella J, Halse J, Kröger H, Maroni M, Mulder H. **Effects of risedronate 5 mg/d on bone mineral density and bone turnover markers in late-postmenopausal women with osteopenia: a multinational, 24-month, randomized, double-blind, placebo-controlled, parallel-group, phase III trial**. *Clin Ther.* (2007.0) **29** 1937-1949. DOI: 10.1016/j.clinthera.2007.09.017
197. Yan Y, Wang W, Zhu H, Li M, Liu J, Luo B. **The efficacy and tolerability of once-weekly alendronate 70 mg on bone mineral density and bone turnover markers in postmenopausal Chinese women with osteoporosis**. *J Bone Miner Metab.* (2009.0) **27** 471-478. DOI: 10.1007/s00774-009-0057-7
198. 198.Zhu HM, Tang H, Cheng Q, He L, Li PQ, Xue QY, et al. Efficacy and safety of denosumab in Chinese postmenopausal women with osteoporosis at increased risk of fracture: Results from a 12-month, randomized, double-blind, placebo-controlled phase III study. J Bone Miner Res. 2017;31(Suppl 1):S160.
199. Boonen S, Adachi JD, Man Z, Cummings SR, Lippuner K, Törring O. **Treatment with denosumab reduces the incidence of new vertebral and hip fractures in postmenopausal women at high risk**. *J Clin Endocrinol Metab.* (2011.0) **96** 1727-1736. DOI: 10.1210/jc.2010-2784
200. Donaldson MG, Palermo L, Ensrud KE, Hochberg MC, Schousboe JT, Cummings SR. **Effect of alendronate for reducing fracture by FRAX score and femoral neck bone mineral density: the Fracture Intervention Trial**. *J Bone Miner Res.* (2012.0) **27** 1804-1810. DOI: 10.1002/jbmr.1625
201. Grey A, Bolland M, Wong S, Horne A, Gamble G, Reid IR. **Low-dose zoledronate in osteopenic postmenopausal women: a randomized controlled trial**. *J Clin Endocrinol Metab.* (2012.0) **97** 286-292. DOI: 10.1210/jc.2011-2081
202. Grey A, Bolland MJ, Horne A, Mihov B, Gamble G, Reid IR. **Duration of antiresorptive activity of zoledronate in postmenopausal women with osteopenia: a randomized, controlled multidose trial**. *CMAJ.* (2017.0) **189** E1130-E11E6. DOI: 10.1503/cmaj.161207
203. Hochberg MC, Thompson DE, Black DM, Quandt SA, Cauley J, Geusens P. **Effect of alendronate on the age-specific incidence of symptomatic osteoporotic fractures**. *J Bone Miner Res.* (2005.0) **20** 971-976. DOI: 10.1359/JBMR.050104
204. McCloskey EV, Johansson H, Oden A, Austin M, Siris E, Wang A. **Denosumab reduces the risk of osteoporotic fractures in postmenopausal women, particularly in those with moderate to high fracture risk as assessed with FRAX**. *J Bone Miner Res.* (2012.0) **27** 1480-1486. DOI: 10.1002/jbmr.1606
205. McClung MR, Boonen S, Törring O, Roux C, Rizzoli R, Bone HG. **Effect of denosumab treatment on the risk of fractures in subgroups of women with postmenopausal osteoporosis**. *J Bone Miner Res.* (2012.0) **27** 211-218. DOI: 10.1002/jbmr.536
206. McClung MR, Lewiecki EM, Cohen SB, Bolognese MA, Woodson GC, Moffett AH. **Denosumab in postmenopausal women with low bone mineral density**. *N Engl J Med.* (2006.0) **354** 821-831. DOI: 10.1056/NEJMoa044459
207. 207.NCTDenosumab China Phase III Study. Clincaltrials.gov2016. *Denosumab China Phase III Study. Clincaltrials.gov* (2016.0)
208. Silverman S, Viswanathan HN, Yang YC, Wang A, Boonen S, Ragi-Eis S. **Impact of clinical fractures on health-related quality of life is dependent on time of assessment since fracture: results from the FREEDOM trial**. *Osteoporos Int.* (2012.0) **23** 1361-1369. DOI: 10.1007/s00198-011-1720-0
209. Tucci JR, Tonino RP, Emkey RD, Peverly CA, Kher U, Santora AC. **Effect of three years of oral alendronate treatment in postmenopausal women with osteoporosis**. *Am J Med.* (1996.0) **101** 488-501. DOI: 10.1016/S0002-9343(96)00282-3
210. Chen LX, Ning GZ, Zhou ZR, Li YL, Zhang D, Wu QL. **The carcinogenicity of alendronate in patients with osteoporosis: evidence from cohort studies**. *PLOS ONE.* (2015.0) **10** E0123080. DOI: 10.1371/journal.pone.0123080
211. Crandall CJ, Newberry SJ, Diamant A, Lim YW, Gellad WF, Booth MJ. **Comparative effectiveness of pharmacologic treatments to prevent fractures: an updated systematic review**. *Ann Intern Med.* (2014.0) **161** 711-723. DOI: 10.7326/M14-0317
212. Davis S, Martyn-St James M, Sanderson J, Stevens J, Goka E, Rawdin A. **A systematic review and economic evaluation of bisphosphonates for the prevention of fragility fractures**. *Health Technol Assess (Winchester, England).* (2016.0) **20** 1-406. DOI: 10.3310/hta20780
213. Davis S, Simpson E, Hamilton J, Martyn-St James M, Rawdin A, Wong R. **Denosumab, raloxifene, romosozumab and teriparatide to prevent osteoporotic fragility fractures: a systematic review and economic evaluation**. *Health Technol Assess.* (2020.0) **24** 1-314. DOI: 10.3310/hta24290
214. Diedhiou D, Cuny T, Sarr A, Norou Diop S, Klein M, Weryha G. **Efficacy and safety of denosumab for the treatment of osteoporosis: a systematic review**. *Ann Endocrinol.* (2015.0) **76** 650-657. DOI: 10.1016/j.ando.2015.10.009
215. Fink HA, MacDonald R, Forte ML, Rosebush CE, Ensrud KE, Schousboe JT. *Long-term drug therapy and drug holidays for osteoporosis fracture prevention: a systematic review* (2019.0)
216. Kranenburg G, Bartstra JW, Weijmans M, de Jong PA, Mali WP, Verhaar HJ. **Bisphosphonates for cardiovascular risk reduction: a systematic review and meta-analysis**. *Atherosclerosis.* (2016.0) **252** 106-115. DOI: 10.1016/j.atherosclerosis.2016.06.039
217. Lv F, Cai X, Yang W, Gao L, Chen L, Wu J. **Denosumab or romosozumab therapy and risk of cardiovascular events in patients with primary osteoporosis: systematic review and meta- analysis**. *Bone.* (2020.0) **130** 115121. DOI: 10.1016/j.bone.2019.115121
218. 218.Tsourdi E, Zillikens MC, Meier C, Body J-J, Rodriguez EG, Anastasilakis AD, et al. Fracture risk and management of discontinuation of denosumab therapy: a systematic review and position statement by ECTS. J Clin Endocrinol Metab. 2020;dgaa756. 10.1210/clinem/dgaa756.
219. Tripto-Shkolnik L, Fund N, Rouach V, Chodick G, Shalev V, Goldshtein I. **Fracture incidence after denosumab discontinuation: real-world data from a large healthcare provider**. *Bone.* (2020.0) **130** 115150. DOI: 10.1016/j.bone.2019.115150
220. de Bekker-Grob EW, Essink-Bot ML, Meerding WJ, Pols HA, Koes BW, Steyerberg EW. **Patients’ preferences for osteoporosis drug treatment: a discrete choice experiment**. *Osteoporos Int.* (2008.0) **19** 1029-1037. DOI: 10.1007/s00198-007-0535-5
221. Fuzzell LN, Fraenkel L, Stark SL, Seehra SS, Nelson C, Keleman A. **A mixed methods study exploring older womens’ attitudes toward osteoporosis medications: adapting a health communication framework**. *Womens Health Rep (New Rochelle, NY).* (2020.0) **1** 102-113
222. Hudson B, Zarifeb A, Young L, Wells J. **Patients’ expectations of screening and preventive treatments**. *Ann Fam Med.* (2012.0) **10** 495-502. DOI: 10.1370/afm.1407
223. Neuner JM, Schapira MM. **Patient perceptions of osteoporosis treatment thresholds**. *J Rheumatol.* (2014.0) **41** 516-522. DOI: 10.3899/jrheum.130548
224. Si L, Tu L, Xie Y, Palmer AJ, Gu Y, Zheng X. **Chinese patients’ preference for pharmaceutical treatments of osteoporosis: a discrete choice experiment**. *Arch Osteoporos.* (2019.0) **14** 85. DOI: 10.1007/s11657-019-0624-z
225. LeBlanc A, Wang AT, Wyatt K, Branda ME, Shah ND, Van Houten H. **Encounter decision aid vs. clinical decision support or usual care to support patient-centered treatment decisions in osteoporosis: The Osteoporosis Choice Randomized Trial II**. *PLoS One.* (2015.0) **10** E0128063. DOI: 10.1371/journal.pone.0128063
226. Montori VM, Shah ND, Pencille LJ, Branda ME, Van Houten HK, Swiglo BA. **Use of a decision aid to improve treatment decisions in osteoporosis: the osteoporosis choice randomized trial**. *Am J Med.* (2011.0) **124** 549-556. DOI: 10.1016/j.amjmed.2011.01.013
227. Smallwood AJ, Schapira MM, Fedders M, Neuner JM. **A pilot randomized controlled trial of a decision aid with tailored fracture risk tool delivered via a patient portal**. *Osteoporos Int.* (2017.0) **28** 567-576. DOI: 10.1007/s00198-016-3767-4
228. Liu CS, Feasel AL, Kline GA, Billington EO. **Pharmacotherapy decisions among postmenopausal women attending a group medical consultation or a one-on-one specialist consultation at an osteoporosis center: an observational cohort study**. *Osteoporos Int.* (2021.0) **32** 1421-1427. DOI: 10.1007/s00198-021-05823-8
229. Hudson B, Toop L, Mangin D, Pearson J. **Risk communication methods in hip fracture prevention: a randomised trial in primary care**. *Br J Gen Pract.* (2011.0) **61** E469-E476. DOI: 10.3399/bjgp11X588439
230. Kalluru R, Petrie KJ, Grey A, Nisa Z, Horne AM, Gamble GD. **Randomised trial assessing the impact of framing of fracture risk and osteoporosis treatment benefits in patients undergoing bone densitometry**. *BMJ Open.* (2017.0) **7** e013703. DOI: 10.1136/bmjopen-2016-013703
231. Sheridan SL, Sutkowi-Hemstreet A, Barclay C, Brewer NT, Dolor RJ, Gizlice Z. **A comparative effectiveness trial of alternate formats for presenting benefits and harms information for low-value screening services: a randomized clinical trial**. *JAMA Intern Med.* (2016.0) **176** 31-41. DOI: 10.1001/jamainternmed.2015.7339
232. Rothmann MJ, Möller S, Holmberg T, Højberg M, Gram J, Bech M. **Non-participation in systematic screening for osteoporosis—the ROSE trial**. *Osteoporos Int.* (2017.0) **28** 3389-3399. DOI: 10.1007/s00198-017-4205-y
233. Rothmann MJ, Huniche L, Ammentorp J, Barkmann R, Glüer CC, Hermann AP. **Women’s perspectives and experiences on screening for osteoporosis (Risk-stratified Osteoporosis Strategy Evaluation, ROSE)**. *Arch Osteoporos.* (2014.0) **9** 192. DOI: 10.1007/s11657-014-0192-1
234. Jha S, Wang Z, Laucis N, Bhattacharyya T. **Trends in media reports, oral bisphosphonate prescriptions, and hip fractures 1996-2012: an ecological analysis**. *J Bone Miner Res.* (2015.0) **30** 2179-2187. DOI: 10.1002/jbmr.2565
235. Deutekom M, Vansenne F, McCaffery K, Essink-Bot M-L, Stronks K, Bossuyt PMM. **The effects of screening on health behaviour: a summary of the results of randomized controlled trials**. *J Public Health.* (2011.0) **33** 71-79. DOI: 10.1093/pubmed/fdq050
236. Condurache CI, Chiu S, Chotiyarnwong P, Johansson H, Shepstone L, Lenaghan E. **Screening for high hip fracture risk does not impact on falls risk: a post hoc analysis from the SCOOP study**. *Osteoporos Int.* (2020.0) **31** 457-464. DOI: 10.1007/s00198-019-05270-6
237. Masud T, Binkley N, Boonen S, Hannan MT. **Official Positions for FRAX® clinical regarding falls and frailty: can falls and frailty be used in FRAX®? From Joint Official Positions Development Conference of the International Society for Clinical Densitometry and International Osteoporosis Foundation on FRAX®**. *J Clin Densitometry* (2011.0) **14** 194-204. DOI: 10.1016/j.jocd.2011.05.010
238. Kanis JA, McCloskey E, Johansson H, Oden A, Leslie WD. **FRAX® with and without Bone Mineral Density**. *Calcif Tissue Int.* (2012.0) **90** 1-13. DOI: 10.1007/s00223-011-9544-7
239. Kanis JA, Borgstrom F, Zethraeus N, Johnell O, Oden A, Jönsson B. **Intervention thresholds for osteoporosis in the UK**. *Bone.* (2005.0) **36** 22-32. DOI: 10.1016/j.bone.2004.08.018
240. Leslie WD, Morin SN, Lix LM, McCloskey EV, Johansson H, Harvey NC. **Fracture prediction from FRAX for Canadian ethnic groups: a registry-based cohort study**. *Osteoporos Int.* (2021.0) **32** 113-122. DOI: 10.1007/s00198-020-05594-8
241. Wells GA, Hsieh SC, Zheng C, Peterson J, Tugwell P, Liu W. **Risedronate for the primary and secondary prevention of osteoporotic fractures in postmenopausal women**. *Cochrane Database Syst Rev.* (2022.0) **5** Cd004523. PMID: 35502787
242. Parkkari J, Kannus P, Palvanen M, Natri A, Vainio J, Aho H. **Majority of hip fractures occur as a result of a fall and impact on the greater trochanter of the femur: a prospective controlled hip fracture study with 206 consecutive patients**. *Calcified tissue international.* (1999.0) **65** 183-187. DOI: 10.1007/s002239900679
243. Symonds C, Kline G. **Warning of an increased risk of vertebral fracture after stopping denosumab**. *CMAJ.* (2018.0) **190** E485. DOI: 10.1503/cmaj.180115
244. Cummings SR, Ferrari S, Eastell R, Gilchrist N, Jensen JB, McClung M. **Vertebral fractures after discontinuation of denosumab: a post hoc analysis of the randomized placebo-controlled FREEDOM trial and its extension**. *J Bone Miner Res.* (2018.0) **33** 190-198. DOI: 10.1002/jbmr.3337
245. Yeam CT, Chia S, Tan HCC, Kwan YH, Fong W, Seng JJB. **A systematic review of factors affecting medication adherence among patients with osteoporosis**. *Osteoporos Int.* (2018.0) **29** 2623-2637. DOI: 10.1007/s00198-018-4759-3
246. Wilton-Clark MS, Feasel AL, Kline GA, Billington EO. **Autonomy begets adherence: decisions to start and persist with osteoporosis treatment after group medical consultation**. *Arch Osteoporos.* (2020.0) **15** 138. DOI: 10.1007/s11657-020-00809-1
247. Morrison A, Polisena J, Husereau D, Moulton K, Clark M, Fiander M. **The effect of English-language restriction on systematic review-based meta-analyses: a systematic review of empirical studies**. *Int J Technol Assess Health Care.* (2012.0) **28** 138-144. DOI: 10.1017/S0266462312000086
248. Waffenschmidt S, Knelangen M, Sieben W, Bühn S, Pieper D. **Single screening versus conventional double screening for study selection in systematic reviews: a methodological systematic review**. *BMC Med Res Methodol.* (2019.0) **19** 132. DOI: 10.1186/s12874-019-0782-0
|
---
title: BM-MSCs overexpressing the Numb enhance the therapeutic effect on cholestatic
liver fibrosis by inhibiting the ductular reaction
authors:
- Yan-nan Xu
- Wen Xu
- Xu Zhang
- Dan-yang Wang
- Xin-rui Zheng
- Wei Liu
- Jia-mei Chen
- Gao-feng Chen
- Cheng-hai Liu
- Ping Liu
- Yong-ping Mu
journal: Stem Cell Research & Therapy
year: 2023
pmcid: PMC10029310
doi: 10.1186/s13287-023-03276-w
license: CC BY 4.0
---
# BM-MSCs overexpressing the Numb enhance the therapeutic effect on cholestatic liver fibrosis by inhibiting the ductular reaction
## Abstract
### Background
Cholestatic liver fibrosis (CLF) is caused by inflammatory destruction of the intrahepatic bile duct and abnormal proliferation of the small bile duct after cholestasis. Activation of the Notch signaling pathway is required for hepatic stem cells to differentiate into cholangiocytes during the pathogenesis of CLF. Our previous research found that the expression of the Numb protein, a negative regulator of Notch signaling, was significantly reduced in the livers of patients with primary biliary cholangitis and CLF rats. However, the relationship between the *Numb* gene and CLF is largely unclear. In this study, we investigated the role of the *Numb* gene in the treatment of bile duct ligation (BDL)-induced CLF.
### Methods
In vivo, bone marrow-derived mesenchymal stem cells (BM-MSCs) with *Numb* gene overexpression or knockdown obtained using lentivirus transfection were transplanted into the livers of rats with BDL-induced CLF. The effects of the *Numb* gene on stem cell differentiation and CLF were evaluated by performing histology, tests of liver function, and measurements of liver hydroxyproline, cytokine gene and protein levels. In vitro, the *Numb* gene was overexpressed or knocked down in the WB-F344 cell line by lentivirus transfection, Then, cells were subjected immunofluorescence staining and the detection of mRNA levels of related factors, which provided further evidence supporting the results from in vivo experiments.
### Results
BM-MSCs overexpressing the *Numb* gene differentiated into hepatocytes, thereby inhibiting CLF progression. Conversely, BM-MSCs with Numb knockdown differentiated into biliary epithelial cells (BECs), thereby promoting the ductular reaction (DR) and the progression of CLF. In addition, we confirmed that knockdown of Numb in sodium butyrate-treated WB-F344 cells aggravated WB-F344 cell differentiation into BECs, while overexpression of Numb inhibited this process.
### Conclusions
The transplantation of BM-MSCs overexpressing Numb may be a useful new treatment strategy for CLF.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s13287-023-03276-w.
## Background
Cholestasis is a disease characterized by bile formation and excretion disorders caused by various factors. Its typical clinical manifestations are fatigue, itching and jaundice [1]. Primary sclerosing cholangitis (PSC) and, particularly, primary biliary cholangitis (PBC) are the most common cholestatic liver diseases in adults [2]. The pathology of cholestasis is characterized by inflammatory destruction of intrahepatic bile ducts and abnormal proliferation of small bile ducts, which eventually develops into cholestatic liver fibrosis (CLF) and cirrhosis [3]. However, an effective treatment for CLF is still lacking.
The Notch signaling pathway is a highly evolutionarily conserved intercellular signal transduction mechanism. It is necessary for embryonic development, maturation, cell specialization, and maintenance of stem cell characteristics. It plays a critical role in regulating cell fate and maintaining organ morphology [4]. The mammalian Notch signaling pathway consists of four transmembrane receptors (Notch-1/-2/-3/-4) and five ligands (Jagged (JAG)-1/-2 and delta-like (DLL)-1/-3/-4) [5]. Notch receptors are cleaved by γ-secretase, and the Notch intracellular domain (NICD) is released. After entering the nucleus, NICD binds to transcription factors, such as recombination signal binding protein Jκ (RBP-Jκ) and CBF1-Suppressor of Hairless-Lag-1 (CSL) and induces the transcriptional activation of Notch target genes such as Hairy and enhancer of split (Hes) and Hes-related with YRPW motif (Hey) [6]. Notch signaling is bidirectional and is a key regulator of the fate of stem cells [7]. At present, it has been recognized that human biliary diseases are closely related to the differentiation of HSCs mediated by the activated Notch signaling [8, 9].
In recent years, the role of Notch signaling in the pathogenesis of liver fibrosis has attracted extensive attention. Aimaiti et al. reported that the activation of Notch signaling in hepatic parenchymal cells or nonparenchymal cells activates hepatic stellate cells and promotes the progression of liver fibrosis [10], while the inhibition of JAG1/Notch3 signaling alleviates the activation of hepatic stellate cells and the progression of liver fibrosis [11]. Our previous studies confirmed that the activation of Notch signaling promotes the differentiation of HSCs into biliary epithelial cells (BECs) and the progression of rat CLF induced by bile duct ligation (BDL), whereas blocking Notch signaling inhibits this pathological process [12]. It was also found that the mRNA and protein expression levels of Numb, a negative regulator of Notch signaling, were significantly reduced in the livers of CLF rats [12]. This finding suggested that Numb may be involved in the pathological process of CLF.
Numb is an evolutionarily highly conserved gene that was first discovered in the precursor cells of the Drosophila sensory organ [13], in which Numb determines the fate of stem cells by antagonizing the membrane receptors of the Notch family through asymmetric mitosis [14, 15]. In recent years, many clinical trials of stem cells for liver disease have shown that stem cells might be a potential therapeutic approach [16]. In particular, mesenchymal stem cell-based therapy is currently considered to be an effective treatment strategy for hepatic disorders, such as liver cirrhosis and nonalcoholic fatty liver disease [17]. By comparison, bone marrow mesenchymal stem cells (BM-MSCs, a type of exogenous HSC) are considered to be the most capable stem cell population for liver cell regeneration [18]. Several clinical studies have shown that BM-MSCs are effective and feasible in the treatment of liver cirrhosis [19–21]. However, another study found that BM-MSCs have the potential to differentiate into myofibroblasts after transplantation [22]. Therefore, how to induce the precise differentiation of BM-MSCs into hepatocytes in the cirrhotic liver is a key scientific problem to be solved at present. However, researchers have not clearly determined whether Numb alters the differentiation of BM-MSCs or exerts a therapeutic effect on CLF. In the present study, we speculated that Numb may negatively regulate Notch signaling to determine the fate of HSCs and affect the progression or regression of CLF. We prepared a rat model of CLF induced by BDL and transplanted BM-MSCs with *Numb* gene editing into the rat liver to observe their directed differentiation in the liver and their effect on the progression of CLF as a method to confirm this hypothesis. The results are consistent with our hypothesis that Numb determines the fate of BM-MSCs and affects the progression or regression of CLF.
## Study design
The goal of this study was to confirm that the high expression of the *Numb* gene in the liver inhibits the activation of the Notch signaling pathway and then inhibits the differentiation of BM-MSCs into BECs and the ductular reaction (DR), which may become a new strategy for the treatment of CLF. First, rat BM-MSCs were isolated and purified, and Numb was overexpressed in these cells (BM-MSCsNumb-OE) by RNA cloning and transfection. Then, BM-MSCsNumb-OE were transplanted into the livers of rats with BDL-induced CLF to confirm that BM-MSCsNumb-OE differentiate into hepatocytes rather than BECs in the livers of CLF rats, thereby inhibiting the DR and the progression of CLF. Second, rat BM-MSCs were isolated and purified, Numb was knocked down (BM-MSCNumb-KD) by RNA interference, and then these cells were transplanted into the livers of CLF rats to confirm that BM-MSCsNumb-KD differentiate into BECs in the livers of rats, which promotes the DR and the progression of CLF. Third, we observed the effect of the Numb level on WB-F344 cell differentiation in vitro to further confirm that the Numb level determines the fate of HSC differentiation into hepatocytes or BECs by regulating the Notch signaling pathway as a method to provide direct evidence for in vivo experiments.
## Materials
The antibodies used for immunophenotype analysis of the BM-MSCs were mouse monoclonal antibody anti-CD10-fluorescein isothiocyanate (FITC) (MA5-14050; Thermo Fisher Scientific, Waltham, Massachusetts, USA); rabbit polyclonal antibody anti-CD14-FITC (17000-1-AP; Proteintech Group, Chicago, IL, USA); rabbit monoclonal antibody anti-CD34-FITC (ab81289; Abcam, Cambridge, UK); CD45-Alexa Fluor® 488-conjugated antibody (202210; BioLegend, Tokyo, Japan); phycoerythrin (PE)-labeled anti-rat CD29 antibody [562154], CD90-PE antibody [551401], goat anti-rabbit IgG-FITC antibody [554020] and goat anti-rabbit IgG-PE antibody [550083] (BD Biosciences, San Jose, CA, USA).
The types of media used to evaluate the differentiation potential of BM-MSCs were adipogenic differentiation medium (RASMx-90031) and osteogenic differentiation medium (RASMx-90021) for Sprague–Dawley (SD) rat BM-MSCs, both of which were purchased from Cyagen Biosciences Inc. (CA, USA).
The following antibodies were used for immunohistochemistry and immunoblot analysis: mouse monoclonal antibody anti-alpha smooth muscle actin (α-SMA; Clone 1A4; Sigma–Aldrich, St. Louis, MO, USA); rabbit polyclonal antibodies anti-cytokeratin 7 (CK7; 15539-1-AP), CK19 (10712-1-AP) and albumin (Alb; 16475-1-AP); mouse monoclonal antibody anti-Numb (60137-1-Ig) (Proteintech Group Inc., Chicago, IL, USA); rabbit polyclonal antibody anti-Numb (ab220362, Abcam Cambridge, UK); rabbit polyclonal antibody anti-Numb (YT5320, ImmunoWay Biotechnology Company, Newark, DE, USA); mouse monoclonal antibodies anti-hepatocyte nuclear factor 4 alpha (HNF4α; sc-374229) and Sox9 (E-9, sc-166505) (Santa Cruz Biotechnology, Inc., CA, USA); rabbit polyclonal antibody anti-RBP-Jκ (5313; Cell Signaling Technology, Danvers, MA, USA); rabbit polyclonal antibodies anti-Hes1 (ab108937) and EpCam (ab216832) (Abcam, Cambridge, UK); mouse monoclonal antibody anti-glyceraldehyde-3-phosphate dehydrogenase antibody (GAPDH, Chemicon International, Billerica, MA, USA), IRDye 800CW-conjugated donkey anti-mouse IgG (H + L) (LI-COR Bioscience, San Jose, CA, USA) and IRDye 680RD-conjugated donkey anti-rabbit IgG (H + L) (LI-COR Bioscience, San Jose, CA, USA).
## Isolation and culture of BM-MSCs
BM-MSCs were isolated using the method described by Wang et al. [ 23]. Briefly, male SD rats were sacrificed under aseptic conditions via cervical dislocation, and the femurs and tibiae were removed. The ends of the femur or tibia were cut open to expose the medullary cavity and repeatedly washed with Dulbecco's modified Eagle’s medium (DMEM, Life Technologies, Gibco, Carlsbad, CA, USA). BM mononuclear cells were isolated using density-gradient centrifugation (Histopaque-1077; Sigma–Aldrich, St. Louis, MO). Mononuclear cells were plated in 75-cm2 flasks (Falcon, Franklin Lakes, NJ) with low-glucose DMEM (Gibco, Grand Island, NY) containing $15\%$ fetal bovine serum (FBS; Gibco) and $1\%$ penicillin–streptomycin (Gibco) and cultured at 37 °C in a $5\%$ CO2 atmosphere for 24 h. After 24 h, one-half of the volume of the medium was replaced with fresh medium, with the medium was completely replaced at 48 h. After 5–7 days, nonadherent cells were removed by replacing the medium, and adherent cells were cultured for another 2–3 days. Colonized cells were detached with a trypsin/ethylene diamine tetra acetic acid solution (Gibco) and replated in 90-mm Petri dishes. When the cultures approached 70–$80\%$ confluence, the cells were serially subcultured through passaging every 3 to 5 days.
## Immunophenotype, cell cycle analysis and assessment of the differential potential of BM-MSCs
BM-MSC purity was determined by performing immunophenotyping. BM-MSCs were stained with the following antibodies conjugated to FITC or PE: anti-CD10-FITC, anti-CD14-FITC, anti-CD34-FITC, anti-CD45-ALexa Fluor® 488, anti-CD29-PE and anti-CD90-PE. The cells were analyzed using a FACScan flow cytometer (BD Biosciences). Briefly, 5 × 105 cells were resuspended in 0.2 mL of phosphate-buffered saline and incubated with antibodies for 20 min at room temperature (RT). Goat anti-rabbit IgG-FITC and goat anti-rabbit IgG-PE were used as isotype controls. The fluorescence intensity of the cells was evaluated using flow cytometry (BD Biosciences, San Jose, CA, USA) [24–26]. In addition, the cell cycle of BM-MSCs was evaluated using flow cytometry.
Adipogenic and osteogenic induction were performed using a reported method to assess the differentiation potential of BM-MSCs [23]. Briefly, cells were plated on Petri dishes in $15\%$ FBS/DMEM-L. For the induction of adipogenesis, lipogenic induction A solution was added when the cells were $100\%$ confluent. Three days later, the solution was replaced with lipogenic induction B solution (Cyagen Biosciences Inc.), and the solution was replaced with the lipogenic induction A solution after one day. This process was repeated. After 3 weeks of differentiation, the cells were fixed and sliced. Adipogenic differentiation was monitored by observing the red droplets after Oil Red O staining.
For the induction of osteogenic differentiation, the osteoblast induction solution (Cyagen Biosciences Inc.) was added when the cells were $60\%$ confluent, and the solution was changed every 3 days. After 4 weeks of differentiation, the cells were fixed and sliced. The osteogenic-induced culture was analyzed using Alizarin Red staining to visualize calcium deposits.
## Numb cloning and overexpression in BM-MSCs
Numb was overexpressed in BM-MSCs (BM-MSCsNumb-OE) by cloning and transfecting the *Numb* gene. Lentiviral vectors (LV) were labeled with enhanced green fluorescent protein (EGFP). LV-Numb-RNA (titer: 3 × 108 TU/ml, Shanghai Genechem Co., Ltd., Shanghai, China) was transfected into P3 BM-MSCs at a multiplicity of infection (MOI) = 80 with the addition of both polybrene and enhanced infection solution (ENi. S, Shanghai GeneChem Co., Ltd., Shanghai, China). The component sequence of LV-Numb-RNA [20910-4] is Ubi-MCS-3FLAG-SV40-EGFP-IRES-puromycin, and its target sequence is shown in Additional file 1: Text 1. The control BM-MSCs (BM-MSCoverexpression-empty vector, BM-MSCOE-EV) were transfected with CON238, an empty lentivirus vector (titer: 1 × 109 TU/ml, Shanghai Genechem Co., Ltd.), whose component sequence is Ubi-MCS-SV40-EGFP-IRES-puromycin. After transfection for 8–10 h, the medium was replaced with vector-free medium. Next, the cells were transfected with the lentivirus at an MOI = 80.
## Numb knockdown in BM-MSCs
Numb was knocked down in BM-MSCs (BM-MSCsNumb-KD) by RNA interference (RNAi). LV-Numb-RNAi (titer: 6 × 108 TU/ml, Shanghai Genechem Co., Ltd., Shanghai, China) was transfected using the method described above. The component sequence of LV-Numb-RNAi (52,618–1) is hU6-MCS-Ubiquitin-EGFP-IRES-puromycin, and its target sequence is 5′-AAGAGAGGAGATCATGAAACA-3′. Control BM-MSCs (BM-MSCsknockdown-empty vector, BM-MSCsKD-EV) were transfected with CON077, an empty lentivirus vector (titer: 8 × 108 TU/ml, Shanghai Genechem Co., Ltd.) whose component sequence is hU6-MCS-Ubiquitin-EGFP-IRES-puromycin. After transfection for 8–10 h, the medium was replaced with vector-free medium. Next, the cells were transfected with the lentivirus at an MOI = 80.
## Animals and experimental protocol
Male SD rats (160–180 g) were purchased from Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China). Animals were maintained in an environment with a constant temperature and supplied with laboratory chow and water ad libitum. All animal experimental protocols were approved by the Animal Research Committee at Shanghai University of Traditional Chinese Medicine (PZSHUTCM18111607), and the study protocols adhere to the ARRIVE guideline.
BDL was performed as previously described [12]. Briefly, 48 rats were randomly divided into the sham group ($$n = 6$$) and model group ($$n = 42$$). Model rats were anesthetized with pentobarbital sodium, and laparotomy was performed with sterile technique. The common bile duct and the left and right hepatic ducts were isolated. The left and right hepatic ducts and the hepatic portal and duodenal site of the common bile duct were ligated, and the abdomen was closed. In sham rats, the identical surgery was performed, except that the bile duct was not ligated. After BDL, model rats were randomly divided into the BDL ($$n = 6$$), BM-MSC ($$n = 6$$), BM-MSCOE-EV ($$n = 6$$), BM-MSCNumb-OE ($$n = 6$$), BM-MSCKD-EV ($$n = 6$$), BM-MSCNumb-KD ($$n = 6$$), and DAPT ($$n = 6$$, positive control drug) groups, and a single dose of 1 × 106 cells was injected into the spleen of each rat in the corresponding groups. The DAPT group was administered 50 mg/kg DAPT orally once per day for 4 weeks, and sham and BDL rats were administered the same volume of physiological saline. At the end of 4 weeks, all rats were euthanized by administering pentobarbital sodium at a dose of 60 mg/kg, and blood and hepatic tissue samples were obtained.
## Detection of biochemical markers in serum
Serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TBil), alkaline phosphatase (ALP), gamma-glutamyltransferase (GGT), total bile acid (TBA), and Alb levels were detected in the clinical laboratory center of Shuguang Hospital affiliated to Shanghai University of TCM.
## Hepatic hydroxyproline (Hyp) content
The Hyp content was determined using the method reported by Jamall et al. [ 27], with some modifications.
## Histopathological and immunohistochemical analyses
Paraformaldehyde-fixed specimens were cut into 4-μm-thick sections and stained with $0.1\%$ (w/v) Sirius Red or hematoxylin and eosin (H&E). Immunostaining was performed using previously published methods [28]. Briefly, sections were deparaffinized, washed, and preincubated with a blocking solution, followed by an incubation with antibodies against α-SMA (1:200), CK7 (1:100), CK19 (1:100), HNF4α (1:50), OV6 (1:40), Numb (1:50), RBP-Jκ (1:1,000), or Hes1 (1:100). Sections were then incubated with HRP-conjugated secondary antibodies (1:1,000) and washed. The samples were visualized using DAB with hematoxylin counterstaining and imaged with a Leica SCN400 scanner (Leica Microsystems Inc., Concord, ON, Canada).
For immunofluorescence staining, frozen specimens were cut into 7-μm-thick sections and subjected to immunofluorescence staining to detect the coexpression of EGFP (marked LV) and CK7 (1:50), EGFP and CK19 (1:50), EGFP and Alb (1:100), EGFP and HNF4α (1:100), EGFP and CD90 (1:100) and Alb, CK7 and OV6 (1:50), or CK19 and OV6. After an incubation with the primary antibodies, the samples were washed with PBST and incubated with Alexa Fluor 488-conjugated goat anti-mouse IgG (A11001; Invitrogen, Carlsbad, CA, USA) or Alexa Fluor 594-conjugated goat anti-rabbit IgG (AB6939; Abcam, Cambridge, UK) secondary antibodies. The nucleus was stained with 4′,6-diamidino-2-phenylindole (DAPI; 1:1,000), and images were captured using an FV10i confocal laser scanning microscope (Olympus, Japan).
## In vitro experimental protocol
In vitro studies were performed in WB-F344 cell lines, which have morphological and functional characteristics similar to those of freshly isolated hepatic progenitor cells [29].
## WB-F344 cell culture and treatment
Cells were divided into the normal group (N), sodium butyrate (SB) group (3.75 mM, Sigma, B5887-1G) [30], LV-Numb overexpression group (Numb-OE), overexpression-empty vector group (OE-EV), LV-Numb knockdown group (Numb-KD), and knockdown-empty vector group (KD-EV) ($$n = 3$$ per group).
Cell culture was performed using our previously reported methods [12]. Briefly, cells were cultured at 37 °C in a $5\%$ CO2 in air atmosphere with Ham’s F12 medium (Life Technologies) supplemented with $10\%$ fetal calf serum (Life Technologies). Chemically induced differentiation was induced by culturing WB-F344 cells on six-well Permanox Lab-Tek culture slides (NalgeNunc International, Naperville) at a density of 3 × 104 cells/well, starting 24 h after seeding. When the degree of confluence reached $30\%$, lentiviral transfection was performed (MOI = 50). LV-Numb-RNA transfection and LV-Numb-RNAi transfection were performed as described above. Then, the culture medium was changed to $10\%$ FBS/DMEM after 6 h of transfection and culture continued until 48 h. SB (3.75 mM) was added to the model group and each intervention group to induce differentiation. The culture medium was exchanged every 2 days, and the cells were collected on the 7th day. The immunofluorescence staining method was the same as described above.
## Real-time PCR (RT–PCR)
The mRNA expression levels of α-SMA, collagen I (Col[1]), Col[4], tumor necrosis factor alpha (TNF-α), transforming growth factor beta 1 (TGF-β1), CK7, CK19, Numb, Hes1, RBP-Jκ, Notch-1/-2-/3/-4, JAG-1/-2, DLL-1/-3/-4, Sox9, EpCam, ligase Numb protein X1 (LNX1), LNX2 and ITCH were assessed using RT–PCR. Total RNA was extracted from frozen hepatic tissues using Isogen (TOYOBO, Kita-ku, Osaka, Japan), and RNA from each sample was reverse transcribed using SuperScript II Reverse Transcriptase (Thermo Fisher Scientific, Waltham, MA, USA). The samples were then analyzed using fluorescence-based RT–PCR and SYBR Green Real-Time PCR Master Mix (TOYOBO) according to the manufacturer’s protocols. Primers and oligonucleotide probes were designed using Primer Express (Takara Chemical) and are listed in Table 1. Each PCR amplification was performed on samples from five rats in both the experimental and control groups. *Individual* gene expression was normalized to GAPDH. The conditions for the SYBR RT–PCR (Perfect Real Time) were as follows: an initial step of 15 min at 42 °C and 2 min at 95 °C and then 40 amplification cycles of denaturation at 95 °C for 15 s and annealing and extension at 60 °C for 1 min. Table 1Primer pairs and probes used for real-time PCRGenePrimer sequence (5′ → 3′)Noteα-SMAForwardAAT GGC TCT GGG CTC TGT AASYBR GreenReverseTCT CTT GCT CTG GGC TTC ATCol [1]ForwardACG TCC TGG TGA AGT TGG TCSYBR GreenReverseTCC AGC AAT ACC CTG AGG TCCol [4]ForwardTTT CCA GGG TTA CAA GGT GTSYBR GreenReverseAGT CCA GGT TCT CCA GCA TCTGF-β1ForwardATT CCT GGC GTT ACC TTG GSYBR GreenReverseAGC CCT GTA TTC CGT CTC CTTNF-αForwardGAC GTG GAA CTG GCA GAA GAGSYBR GreenReverseTTG GTG GTT TGT GAG TGT GAGCK7ForwardAGG AAC AGA AGT CAG CCA AGA GSYBR GreenReverseGCA ACA CAA ACT CAT TCT CAG CCK19ForwardGAT CTG CGT AGT GTG G-3′SYBR GreenReverseAAA ACC AAA CTG GGG ATG-3′NumbForwardGCT ACT TTC GAT GCC AGT AGA ACC ASYBR GreenReverseCTG TTG CCA GGA GCC ACT GARBP-JκForwardTTG CTT ACC TTC AGG CGT GTGSYBR GreenReverseGCC CAA TGA GTC TGC TGC AAHes1ForwardGAC GGC CAA TTT GCT TTCSYBR GreenReverseGAC ACT GCG TTA GGA CCCNotch1ForwardTGG ATG AGG AAG ACA AGC ATT ASYBR GreenReverseGAA AAG CCA CCG AGA TAG TCA GNotch2ForwardGAG GAA GAA GTG TCT CAASYBR GreenReverseGTG GCA TCA GAA ACA TAT GNotch3ForwardGAC AAG GAC CAC TCC CAC TACTSYBR GreenReverseATC CAC ATC ATC CTC ACA ACT GNotch4ForwardTGT CAG GAA CCA GTG TCA GAA CSYBR GreenReverseCCT GGG CTT CAC ATT CAT CTA TJAG1ForwardCCA TCA AGG ATT ATG AGA ACSYBR GreenReverseTGG TGC TTA TCC ATA TCAJAG2ForwardAAA TGA GTG GTC CGT GGC AGASYBR GreenReverseTGG TTG GAA GCC TTG TCT GCTDLL1ForwardGTG TGC AGA TGG TCC TTG CTT CSYBR GreenReverseCTG ACA TCG GCA CAG GTA GGA GDLL3ForwardCTG AGG TTA CAA GAC GGT GCTSYBR GreenReverseGTA AAT GGA AGG GGC TGG TAT GDLL4ForwardGCA GAA CCA CAC ACT GGA CTA TSYBR GreenReverseTGG CAC CTT CTC TCC TAA ACT CSox9ForwardGAA AGA CCA CCC CGA TTA CAA GSYBR GreenReverseAAG ATG GCG TTA GGA GAG ATG TGEpCamForwardTGT GGA CAT AGC TGA TGT GGC TTA CSYBR GreenReverseCAC CCT CAG GTC CAT GCT CTT ALNX1ForwardTGC TGC CAG GAG ACA TCA TSYBR GreenReverseCAT TGC TTC TGC TAC GGA ACT TLNX2ForwardACA CAG ATT GAG GGT GAA ACTSYBR GreenReverseGGT CCA CAC AGG AAG AGG TITCHForwardATG GGA GAT TTG TCA GTT TGT CSYBR GreenReverseCAG CGT CAT TCT GTG TAG CAAlbForwardAAG GCA CCC CGA TTA CTC CGSYBR GreenReverseTGC GAA GTC ACC CAT CAC CGHNF4αForwardCGG GCC ACT GGC AAA CACSYBR GreenReverseGTA ATC CTC CAG GCT CACGAPDHForwardGGC ACA GTC AAG GCT GAG AAT GSYBR GreenReverseATG GTG GTG AAG ACG CCA GTA
## Immunoblot analysis
Liver tissue was lysed in RIPA buffer containing a mixture of protease inhibitors and phosphatase inhibitors and then homogenized in ice-cold water. Protein concentrations were determined using a BCA protein assay kit (Thermo). Total proteins were resolved on SDS–PAGE gels, transferred onto PVDF membranes, and blocked with a $5\%$ (w/v) bovine serum albumin (Gibco) solution. The following dilutions of primary antibodies were used: α-SMA (1:1,000), CK7 (1:1,500), CK19 (1:1,500), Numb (1:300), RBP-Jκ (1:1,000), Hes1 (1:500), Sox9 (1:2,000), EpCam (1:1,000) and GAPDH (1:10,000). The following secondary antibodies were used: IRDye 800CW-conjugated donkey anti-mouse IgG (H + L) (1:10,000) and IRDye 680RD-conjugated donkey anti-rabbit IgG (H + L) (1:1,000). Finally, the data were analyzed using Odyssey 2.1 software.
## Statistical analysis
All data are presented as the means ± SD. Statistical analyses of multiple groups were performed using analysis of variance (ANOVA) with SPSS 24.0 software, and $P \leq 0.05$ was considered statistically significant.
## Identification of the purity and proliferation ability of BM-MSCs
When the BM-MSCs were cultured to the third generation, they showed a fusiform and vortex-like morphology (Fig. 1a). BM-MSCs were confirmed by flow cytometry, and showed the following: CD10 (−), CD14 (−), CD29 (+), CD34 (−), CD45 (−) and CD90 (+) (Fig. 1b). Furthermore, the proliferative capacity was evaluated by examining the cell cycle of BM-MSCs using flow cytometry, and the results showed that $80.2\%$ of the BM-MSCs were in G1 phase (Fig. 1c). Additionally, the BM-MSCs showed osteogenic and adipogenic abilities following the differentiation assay. A large number of calcium deposits were noted following osteogenic induction, and a large number of fat droplets were noted following adipogenic induction (Fig. 1d, e). These results demonstrate that the obtained BM-MSCs have a strong differentiation ability. Fig. 1Identification of the purity and proliferation ability of BM-MSCs. a BM-MSCs morphology (left: × 40; right: × 100). b Cell purity was detected by flow cytometry, and the results showed the following: CD10 (−), CD14 (−), CD29 (+), CD34 (−), CD45 (−) and CD90 (+). c The cell cycle was identified by flow cytometry, the result showed that $80.2\%$ of the BM-MSCs were in the G1 phase. d Adipogenic induction (× 200). e Osteogenic induction (× 200)
## BM-MSCsNumb-OE transplantation alleviates liver inflammation and fibrosis
First, we measured the Numb protein level in the livers of patients with PBC complicated with cirrhosis. Numb was widely expressed in the livers of healthy people, but its expression was clearly decreased in the livers of patients with PBC, as the Numb-positive staining area was reduced by $73\%$ in patients with PBC compared with the healthy population ($$P \leq 0.000$$) (Fig. 2a). This result suggests that the loss of Numb may be closely related to the pathogenesis of CLF.Fig. 2The transplantation of BM-MSCs overexpressing Numb inhibits the hepatic inflammatory response and liver fibrosis. a Expression of Numb in the livers from a healthy population ($$n = 10$$) and patients with PBC ($$n = 20$$): Numb immunostaining (× 200) and its positive area. b Experimental flow chart. c Lentivirus-transfected BMSCs (× 100) and Numb protein and mRNA expression levels in BM-MSCs overexpressing Numb (Full-length blot is presented in Additional file 2: Fig. 1). d H&E staining (× 200) and Sirius red collagen staining (× 100). e α-SMA immunostaining (× 200). f Serum levels of biochemical markers. g Hyp content in liver tissue. h α-SMA immunoblotting bands, gray-level integration and mRNA expression ($$n = 6$$/per group) (Full-length blot is presented in Additional file 2: Fig. 2). i The mRNA levels of cytokines related to liver fibrosis ($$n = 6$$/per group). * $P \leq 0.05$; **$P \leq 0.01$ Second, rat BM-MSCs and BM-MSCsNumb-OE were transplanted into rats subjected to BDL, and the effect of BM-MSCNumb-OE on CLF progression was observed (Fig. 2b; the experiment was repeated twice). As shown in Fig. 2c, when the MOI = 80, the transfection rate was greater than $80\%$ and cells maintained a normal morphology. The expression level of the Numb protein was significantly increased in the BM-MSCNumb-OE group compared with the BM-MSCOE-EV group ($P \leq 0.01$), which was 2 times higher than the expression in the BM-MSCOE-EV group, and the level of the Numb mRNA was consistent with the protein level (Full-length blot of *Numb is* presented in Additional file 2: Fig. 1). We performed immunofluorescence costaining for CD90 (a marker of BM-MSCs) and Numb. The results showed that CD90 and Numb were still co-expressed in the BM-MSCOE-EV and BM-MSCNumb-OE cells at the end of the 4th week after BDL (Additional file 1: Fig. S1), which confirmed that Numb was located in BM-MSCs in the livers of BDL rats. H&E staining showed that the inflammatory response and bile duct hyperplasia were clearly reduced in the BM-MSC and BM-MSCOE-EV groups compared with the BDL group, and the aforementioned pathological changes were further alleviated in the BM-MSCNumb-OE group compared with the BM-MSCOE-EV group (Fig. 2d).
Serum biochemical tests showed significant reductions in AST and ALP activities and the TBil content, revealing that BM-MSCs transplantation significantly improved liver function. Notably, compared with BM-MSCsOE-EV or BM-MSCs transplantation, BM-MSCsNumb-OE transplantation further improved serum indicators of liver function, as manifested by significant reductions in AST, ALP and GGT activities and the TBil content and a significant increase in the Alb content ($P \leq 0.05$ or $P \leq 0.01$) (Fig. 2f).
Sirius red staining revealed that proliferating BECs were surrounded by abundant collagen in the BDL group; however, collagen deposition was clearly reduced in the BM-MSC and BM-MSCOE-EV groups and was further reduced in the BM-MSCNumb-OE group compared with the BM-MSCOE-EV group (Fig. 2d). Consistent with the histopathology, the Hyp content in the liver tissue was significantly increased in the BDL group ($P \leq 0.01$) but significantly decreased in the BM-MSC and the BM-MSCOE-EV groups compared to the BDL group ($P \leq 0.05$), and it was further reduced in the BM-MSCNumb-OE group compared to the BM-MSC and BM-MSCOE-EV groups ($P \leq 0.01$ and $P \leq 0.05$, respectively) (Fig. 2g).
Immunostaining also confirmed that α-SMA expression (a myofibroblast marker) was detected around proliferating BECs in the BDL group; however, its levels were clearly reduced in the BM-MSC and BM-MSCOE-EV groups compared with the BDL group and further reduced in the BM-MSCNumb-OE group compared with the BM-MSCOE-EV group (Fig. 2e). Consistent with the immunostaining results, the α-SMA protein and mRNA expression levels were increased significantly in the BDL group ($P \leq 0.01$), whereas they were significantly reduced in the BM-MSC and BM-MSCOE-EV groups compared to the BDL group ($P \leq 0.01$) and further reduced in the BM-MSCNumb-OE group compared to the BM-MSC and BM-MSCOE-EV groups ($P \leq 0.01$) (Fig. 2h) (Full-length blot of α-SMA is presented in Additional file 2: Fig. 2).
In addition, the TGF-β1, TNF-α, Col[1], and Col[4] mRNA levels were significantly increased in the liver after BDL ($P \leq 0.01$), while the TNF-α, Col[1], and Col[4] mRNA levels were significantly reduced in the BM-MSC and BM-MSCOE-EV groups compared with the BDL group ($P \leq 0.05$ or $P \leq 0.01$). Furthermore, the Col[1] and Col[4] mRNA levels were further reduced in the BM-MSCNumb-OE group compared to the BM-MSCOE-EV group ($P \leq 0.05$ or $P \leq 0.01$). Although the mRNA levels of TGF-β1 and TNF-α in the BM-MSCNumb-OE group were not significantly decreased compared with those in the BM-MSC and BM-MSCOE-EV groups, they were significantly decreased compared with those in the BDL group ($P \leq 0.01$) (Fig. 2i). Based on these results, BM-MSCs transplantation exerts a good antifibrotic effect, and the intensity of this intervention effect becomes more significant when *Numb is* overexpressed in BM-MSCs.
## BM-MSCsNumb-OE transplantation suppressed the activation of Notch signaling in the livers from CLF rats and differentiation into hepatocytes
Immunostaining showed markedly reduced Numb expression in hepatocytes from the BDL group, while the expression levels of RBP-Jκ and Hes1 were markedly increased in the nuclei of proliferating BECs. After transplantation of BM-MSCs, BM-MSCsOE-EV or BM-MSCsNumb-OE, Numb expression was markedly increased, whereas the expression of RBP-Jκ and Hes1 was decreased, particularly in the BM-MSCNumb-OE group (Fig. 3a).Fig. 3BM-MSCsNumb-OE transplantation inhibits the activation of the Notch signaling pathway in the liver. a Numb, RBP-Jκ and Hes1 immunostaining (× 400). b Protein and mRNA expression levels of Numb, RBP-Jκ, and Hes1 ($$n = 6$$/per group) (Full-length blots are presented in Additional file 2: Figs. 3–5); c CK7 and CK19 immunostaining (× 200). d CK7 and CK19 immunoblotting bands, gray-level integration and mRNA expression ($$n = 6$$/per group) (Full-length blots are presented in Additional file 2: Figs. 6 and 7). e CK7/EGFP (labeled BM-MSCOE-EV and BM-MSCNumb-OE, the same below) immunofluorescence costaining (× 200). f CK19/EGFP immunofluorescence costaining (× 200). g Alb/EGFP immunofluorescence costaining (× 200), and the costaining area ratio of Alb/EGFP. h HNF4α/EGFP immunofluorescence costaining (× 200), and the costaining cells ratio of HNF4α/EGFP. i EGFP/CD90/Alb immunofluorescence costaining (× 400). * $P \leq 0.05$; **$P \leq 0.01$ Consistent with the immunostaining results, the Numb protein and mRNA expression levels were significantly reduced in the BDL group ($P \leq 0.01$), while the RBP-Jκ and Hes1 protein and mRNA levels were significantly increased ($P \leq 0.01$). Compared with the BDL group, Numb expression was significantly increased in the BM-MSC and BM-MSCOE-EV groups ($P \leq 0.01$), while RBP-Jκ and Hes1 levels were significantly decreased ($P \leq 0.01$). In particular, the Numb protein and mRNA levels were further increased, and RBP-Jκ and Hes1 levels were further decreased in the BM-MSCNumb-OE group compared to the BM-MSCOE-EV group ($P \leq 0.01$) (Fig. 3b) (Full-length blots of Numb, RBP-Jκ and Hes1 are presented in Additional file 2: Figs. 3, 4 and 5, respectively). In addition, we evaluated the mRNA expression levels of other related molecules in the Notch signaling pathway. BM-MSCsNumb-OE transplantation reduced the expression levels of the Notch-1/-3/-4 and JAG2 mRNAs compared with BM-MSCsOE-EV transplantation ($P \leq 0.05$ or $P \leq 0.01$) (Additional file 1: Fig. S2). Thus, BM-MSCs transplantation potentially suppresses the activation of Notch signaling, and this effect is more significant after the transplantation of Numb-overexpressing BM-MSCs.
Next, the mRNA expression of E3 ubiquitin ligases, including LNX-1/-2 (promote the proteasome-dependent degradation of Numb) [31] and ITCH (promotes the ubiquitination-dependent proteasomal degradation of the NICD) [32], was examined. LNX1 expression was increased significantly, and LNX2 and ITCH levels were decreased significantly in the BDL group compared with the sham group ($P \leq 0.01$). However, compared to the levels in the BDL group, LNX1 expression was decreased significantly in the BM-MSCOE-EV group ($P \leq 0.05$). Only ITCH expression was increased significantly in the BM-MSCNumb-OE group compared to the BM-MSCOE-EV group ($P \leq 0.05$) (Additional file 1: Fig. S3). Based on these results, BM-MSCsNumb-OE transplantation may increase the ubiquitination of Notch due to the increase in the Numb level [33], which leads to the suppression of Notch signaling in BM-MSCs. Then, the expression levels of other Notch signaling molecules upstream of Numb are decreased, which reduces the differentiation of BM-MSCs into BECs and inhibits CLF progression.
CK7 is considered a marker of hepatic progenitor cells [34], and CK19 is a recognized marker of BECs [35]. As shown in Fig. 3c, immunostaining revealed that CK7 and CK19 were widely expressed in proliferating BECs in the BDL group, while their expression was clearly reduced in the BM-MSC, BM-MSCOE-EV and BM-MSCNumb-OE groups, particularly in the BM-MSCNumb-OE group. Consistent with the immunostaining data, the expression levels of the CK7 and CK19 proteins were increased significantly in the BDL group ($P \leq 0.01$), whereas they were significantly reduced in the BM-MSC and BM-MSCOE-EV groups compared to the BDL group ($P \leq 0.01$). Additionally, the CK19 protein level was further reduced in the BM-MSCNumb-OE group compared with the BM-MSCOE-EV group ($P \leq 0.01$) (Fig. 3d) (Full-length blots of CK7 and CK19 are presented in Additional file 2: Figs. 6 and 7, respectively). The CK7 and CK19 mRNA expression levels were consistent with their protein expression levels (Fig. 3d). This finding suggests that BM-MSCNumb-OE transplantation inhibits the DR in CLF rats.
To determine the differentiation direction of BM-MSCs in CLF liver, we labeled cells with EGFP (labeled lentivirus, showing traces of BM-MSCOE-EV and BM-MSCNumb-OE) and costained them with antibodies against CK7, CK19, Alb (synthesized by mature hepatocytes) or HNF4α (a mature hepatocyte marker) to evaluate the oriented differentiation of BM-MSCsNumb-OE in the liver. As shown in Fig. 3e–h, very little coexpression of EGFP/CK7 or EGFP/CK19 was detected in the BM-MSCNumb-OE group, while EGFP/Alb and EGFP/HNF4α were widely coexpressed in hepatocytes. To evaluate the ability of BM-MSCsOE-EV or BM-MSCsNumb-OE to differentiate into hepatocytes, we analyzed the positive area ratio of EGFP/Alb immunofluorescence costaining and the positive cell ratio of EGFP/HNF4α immunofluorescence costaining. The results showed that the costaining area ratio of EGFP/Alb in the BM-MSCNumb-OE group was 1.6 times that in the BM-MSCsOE-EV group ($84.25\%$ vs. $57.70\%$, $$P \leq 0.000$$) (Fig. 3g histogram), and the costaining cell ratio of EGFP/HNF4α in the BM-MSCNumb-OE group was 1.4 times that in the BM-MSCsOE-EV group ($82.31\%$ vs. $57.75\%$, $$P \leq 0.003$$) (Fig. 3h histogram). In addition, we observed the ability of BM-MSCsNumb-OE to differentiate into hepatocytes in vitro. We performed immunofluorescence staining for EGFP (labeled BM-MSCOE-EV and BM-MSCNumb-OE), CD90 (a marker of BM-MSCs) and Alb (a marker of hepatocytes), and the results showed that there was extensive coexpression of EGFP/CD90/Alb in the BM-MSCNumb-OE cells compared with BM-MSCOE-EV cells on the 6th day of cultivation (Fig. 3i). Thus, BM-MSCs overexpressing Numb differentiate into hepatocytes rather than BECs in the liver of CLF rats.
## BM-MSCsNumb-KD transplantation promotes liver inflammation and fibrosis
We knocked down Numb in BM-MSCs by RNA interference (BM-MSCNumb-KD), injected the cells into the rat spleen at the same time as BDL, and obtained samples at the end of 4 w to determine whether the deletion of Numb in BM-MSCs promotes the progression of CLF (Fig. 4a, the experiment was repeated twice). As shown in Fig. 4b, when the MOI = 80, the transfection rate was greater than $80\%$ and the cells maintained a normal morphology. The expression level of the Numb protein was significantly decreased in the BM-MSCNumb-KD group compared with the BM-MSCKD-EV group ($P \leq 0.01$) to approximately $51.2\%$ of that in the BM-MSCKD-EV group (Full-length blot of *Numb is* presented in Additional file 2: Fig. 8), and the level of the Numb mRNA was consistent with the protein level. As mentioned above, BM-MSCs and BM-MSCsKD-EV transplantation alleviated liver inflammation, the DR, and collagen deposition; improved serum biochemical indexes; and decreased the Hyp content and the expression levels of proteins related to liver fibrosis, including α-SMA, TGF-β1 and Col[1]. However, when BM-MSCsNumb-KD were transplanted, liver inflammation, the DR and collagen deposition were markedly increased compared with those in the BM-MSCKD-EV group (Fig. 4c). In addition, serum ALT, AST and ALP activities (Fig. 4e), the Hyp content (Fig. 4f), α-SMA expression (Fig. 4d, g) (Full-length blot of α-SMA is presented in Additional file 2: Fig. 9) and the mRNA expression levels of proteins related to liver fibrosis, including TNF-α, TGF-β1 and Col[1] (Fig. 4h), were increased significantly in the BM-MSCNumb-KD group compared to the BM-MSCKD-EV group ($P \leq 0.05$ or $P \leq 0.01$). These results suggested that BM-MSCsNumb-KD transplantation aggravates the liver inflammatory response and hepatic stellate cell activation and therefore promotes the progression of CLF induced by BDL.Fig. 4BM-MSCsNumb-KD transplantation promotes the hepatic inflammatory response and liver fibrosis. a Experimental flow chart. b Lentivirus-transfected BMSCs (× 100) and Numb protein and mRNA expression levels in BM-MSCs with knockdown of Numb (Full-length blot is presented in Additional file 2: Fig. 8). c H&E staining (× 200) and Sirius red collagen staining (× 100). d α-SMA immunostaining (× 200). e Serum levels of biochemical markers and f the Hyp content in liver tissues. g α-SMA immunoblotting bands, gray-level integration and mRNA expression ($$n = 6$$/per group) (Full-length blot is presented in Additional file 2: Fig. 9); h The mRNA expression levels of TGF-β1, TNF-α, Col[1], and Col[4]. * $P \leq 0.05$; **$P \leq 0.01$
## BM-MSCsNumb-KD transplantation activates Notch signaling in the livers of CLF rats and promotes differentiation into BECs
As mentioned above, transplantation of BM-MSCs significantly inhibits the activation of the Notch signaling pathway in the livers of CLF rats. However, when BMSCs lacking Numb were transplanted, Notch signaling was significantly activated, as confirmed by the protein and mRNA expression levels of Numb, RBP-Jκ and Hes1; namely, the expression of Numb was decreased significantly and the expression levels of RBP-Jκ and Hes1 were increased significantly compared to those in the BM-MSCKD-EV group ($P \leq 0.01$). In addition, the Numb mRNA level in the BM-MSCNumb-KD group was significantly lower than that in the BDL group ($P \leq 0.05$), while significantly higher levels of the RBP-Jκ and Hes1 mRNAs were detected than those in the BDL group ($P \leq 0.01$) (Fig. 5a, b) (Full-length blots of Numb, RBP-Jκ and Hes1 are presented in Additional file 2: Figs. 10, 11 and 12, respectively). Furthermore, the mRNA expression levels of other related molecules in the Notch pathway, including Notch-2/-3/-4, JAG-1/-2, and DLL-1/-4, were also significantly increased in the BM-MSCNumb-KD group compared to the BM-MSCKD-EV group ($P \leq 0.05$ or $P \leq 0.01$) (Additional file 1: Fig. S4). In addition, we detected the mRNA expression levels of LNX-1/-2 and ITCH, and only ITCH expression was decreased significantly in the BM-MSCNumb-KD group compared to the BM-MSCKD-EV group ($P \leq 0.05$) (Additional file 1: Fig. S5). These results suggest that BM-MSCsNumb-KD transplantation may reduce the ubiquitination of Notch due to the decrease in the Numb level [33], which leads to the activation of Notch signaling in BM-MSCs, thus promotes the differentiation of BM-MSCs into BECs and aggravates CLF progression. Fig. 5BM-MSCNumb-KD transplantation activates Notch signaling in the livers of CLF rats and induces the differentiation of these cells into BECs. a Numb, RBP-Jκ and Hes1 immunostaining (× 400). b Protein and mRNA expression levels of Numb, RBP-Jκ, and Hes1 ($$n = 6$$/per group) (Full-length blots are presented in Additional file 2: Figs. 10–12). c CK7 and CK19 immunostaining (× 200). d Protein and mRNA expression levels of CK7 and CK19 (Full-length blots are presented in Additional file 2: Figs. 13 and 14). e CK7/EGFP (labeling BM-MSCKD-EV and BM-MSCNumb-KD) immunofluorescence costaining (× 200), and the costaining area ratio of CK7/EGFP. f CK19/EGFP immunofluorescence costaining (× 200), and the costaining area ratio of CK19/EGFP. * $P \leq 0.05$; **$P \leq 0.01$ As mentioned above, BM-MSCs transplantation significantly suppressed the DR in CLF livers. However, BM-MSCsNumb-KD transplantation significantly promoted the DR compared with BDL and BM-MSCsKD-EV transplantation, as confirmed by the protein and mRNA expression levels of CK7 and CK19 (Fig. 5c, d) (Full-length blots of CK7 and CK19 are presented in Additional file 2: Figs. 13 and 14, respectively).
We evaluated the oriented differentiation of BM-MSCsNumb-KD in the livers of CLF rats by observing cells with coexpression of EGFP (labeled lentivirus, showing traces of BM-MSCLV-EV and BM-MSCNumb-KD) and CK7 or CK19. The results showed very little coexpression of EGFP with CK7 or EGFP with CK19 in the BM-MSCKD-EV group, but extensive coexpression of EGFP with CK7 and CK19 was observed in the BM-MSCNumb-KD group (Fig. 5e, f). In addition, we analyzed the positive area ratio of CK7/EGFP and CK19/EGFP immunofluorescence costaining. The results showed that the costaining area ratio of EGFP/CK7 in the BM-MSCNumb-KD group was 14.8 times that in the BM-MSCsKD-EV group ($65.73\%$ vs. $4.44\%$, $$P \leq 0.000$$) (Fig. 5e histogram), and the costaining area ratio of EGFP/CK19 in the BM-MSCNumb-KD group was 18.5 times that in the BM-MSCsKD-EV group ($71.11\%$ vs. $3.84\%$, $$P \leq 0.000$$) (Fig. 5f histogram). The aforementioned results clearly indicate that BM-MSCs lacking Numb differentiated into BECs in the livers of CLF rats and promoted the DR.
## The Numb level determines the fate of HSCs in vitro
We overexpressed or knocked down Numb in WB-F344 cells (Numb-OE or Numb-KD) and stimulated them with SB to further clarify the regulatory effect of Numb on the differentiation fate of HSCs (Fig. 6a). When the lentivirus was added at an MOI = 50, the transfection rate was greater than $80\%$, and the cell morphology was normal (Fig. 6b). In addition, Numb protein expression was increased significantly in the Numb-OE group compared to the OE-EV group ($P \leq 0.01$), whereas it was decreased significantly in the Numb-KD group compared to the KD-EV group ($P \leq 0.01$), and the Numb mRNA level was consistent with the corresponding protein level (Fig. 6c, d) (Full-length blots of Numb are presented in Additional file 2: Figs. 15 and 16).Fig. 6Effect of Numb expression on the differentiation of WB-F344 cells. a Experimental flow chart. b Cell morphology and GFP expression after lentivirus transfection for 72 h (× 100). c, d Numb protein and mRNA levels (Full-length blots are presented in Additional file 2: Figs. 15 and 16). e, f CK19 immunofluorescence staining (× 600). g, h Immunofluorescence staining for Numb, RBP-Jκ and Hes1 (× 600). i, j The mRNA levels of CK19, Numb, RBP-Jκ and Hes1. * $P \leq 0.05$; **$P \leq 0.01$ Immunostaining showed a clear increase in CK19 expression in the SB, OE-EV and KD-EV groups, but its expression was decreased in the Numb-OE group and further increased in the Numb-KD group (Fig. 6e, f). The expression level of the CK19 mRNA was consistent with the immunostaining results (Fig. 6i, j).
In addition, immunostaining showed a clear decrease in Numb expression, and the expression of RBP-Jκ and Hes1 was increased in the SB, KD-EV and OE-EV groups. Compared with that in the OE-EV group, Numb expression was increased, and the expression of RBP-Jκ and Hes1 was decreased in the Numb-OE group (Fig. 6g). Conversely, the expression of Numb was further decreased, and RBP-Jκ and Hes1 levels were further increased in WB-F344 cells lacking Numb (Fig. 6h).
Consistent with the immunostaining results, Numb mRNA levels were decreased significantly ($P \leq 0.01$), and those of RBP-Jκ and Hes1 were increased significantly in the KD-EV and OE-EV groups ($P \leq 0.01$). However, the Numb mRNA levels were increased and those of RBP-Jκ and Hes1 were significantly decreased in the Numb-OE group compared to the OE-EV group ($P \leq 0.05$ or $P \leq 0.01$) (Fig. 6i). Conversely, the Numb mRNA expression levels were further reduced, and those of RBP-Jκ and Hes1 were further increased in the Numb-KD group compared to the KD-EV group ($P \leq 0.05$) (Fig. 6j).
Therefore, WB-F344 cells lacking Numb differentiate into BECs when stimulated with SB; conversely, the overexpression of Numb in WB-F344 cells suppresses this pathological process.
## Discussion
The liver is a very complex organ that is susceptible to multiple types of damage and dysfunction [36]. Biliary proliferation, also known as the DR, occurs when BECs are stimulated by persistent inflammation [37], and HSCs are an important participant in the DR [38, 39]. Proliferating BECs secrete a variety of profibrotic cytokines, such as TNF-α, TGF-β1, platelet-derived growth factor (PDGF), interleukin (IL)-1, -6, -8, and monocyte chemoattractant protein 1 (MCP1). They synergistically promote the activation of fibroblasts and hepatic stellate cells around portal veins into myofibroblasts, synthesize a large amount of extracellular matrix, and promote the occurrence and development of liver fibrosis [40]. Therefore, inhibition of the abnormal activation and proliferation of BECs may partially or even completely reverse CLF [33].
Activation of Notch signaling plays a critical role in the DR [41]. In our previous study, we confirmed that blocking the Notch signaling pathway significantly inhibits the differentiation of HSCs into BECs and the progression of CLF induced by BDL, and we found that the Numb mRNA and protein levels gradually decrease with CLF progression [12]. In addition, the expression of Numb in the livers of patients with PBC was only $26.95\%$ of the level in healthy subjects. However, the role of Numb in the occurrence and treatment of CLF has not been reported. This study is the first to focus on the effect of Numb on CLF.
## The Numb level in HSCs determines their fate in the livers of CLF rats
Numb negatively regulates Notch signaling and antagonizes the membrane receptors of the Notch family through asymmetric mitosis, which is an important determinant of cell fate [14]. In recent years, Numb has attracted extensive attention in tumor therapy; for example, the loss of Numb expression may increase Notch signaling activity in breast cancer cells [42]. Numb may be a therapeutic target for prostate cancer by inhibiting the activation of Notch signaling [43]. Moreover, the level of *Numb is* significantly decreased in human hepatocellular carcinoma, and miR-148a upregulates Numb expression to inhibit Notch signaling, thereby inhibiting hepatocellular carcinoma progression [44]. In the field of chronic liver disease, Numb may act as a "switch" of the Wnt-Notch signaling pathway, which determines the differentiation of HSCs into bile duct cells (activation of Notch signaling) or hepatocytes (activation of classical Wnt signaling) [45]. This finding highlights the importance of Numb in regulating the differentiation of HSCs.
Bone marrow is an important source of exogenous HSCs [46]. In a rat model of acute liver injury, transplantation of BM-MSCs significantly reduced the levels of liver injury markers [47]. In patients with alcoholic cirrhosis, transplantation of autologous BM-MSCs safely improved histologic fibrosis and liver function [20].
In this study, we first observed the effects of transplantation of BM-MSCs lacking or overexpressing Numb on CLF to directly observe the role of Numb expressed in HSCs in CLF pathogenesis. The results clearly showed that BM-MSCsNumb-OE transplantation effectively inhibits hepatic inflammation, the DR and CLF progression. The main mechanism is that BM-MSCs overexpressing Numb mainly differentiated into hepatocytes, as evidenced by the significantly increased levels of the Alb and HNF4α mRNAs and proteins in the liver and serum Alb content; immunofluorescence costaining also clearly suggested that BM-MSCsNumb-OE differentiated into hepatocytes and promoted the repair of liver injury.
In contrast, BM-MSCsNumb-KD transplantation significantly promotes the inflammatory response in the liver, hepatic stellate cell activation and CLF progression. The main mechanism is that the Notch signaling is activated in BM-MSCs that then differentiate into BECs due to the weakened negative regulation of Notch signaling after Numb loss, as manifested by CK7/EGFP, CK19/EGFP are widely coexpressed in proliferating BECs, and the DR is enhanced, thus promoting the progression of CLF. In addition, in vitro experiments confirmed that WB-F344 cells lacking Numb differentiate into a bile duct cell phenotype.
## The role of Numb in determining the fate of stem cells depends on its negative regulation of the Notch signaling pathway
The Notch signaling pathway is bidirectional and plays an important role in regulating the fate of stem cells [7]. For example, intestinal stem cells differentiate into intestinal cells and endocrine cells in adult fruit flies and mice. Inhibition of Notch signaling leads to the differentiation of intestinal stem cells into intestinal endocrine cells, while activation of Notch signaling promotes differentiation into intestinal cells [48]. In human diseases, inhibition of Notch signaling suppresses the self-renewal ability of lung adenocarcinoma stem cells and promotes their re-entry into asymmetric division [49]. Therefore, we speculate that Notch signaling may also regulate the mitotic state or pluripotency of stem cells in other organs.
As shown in our previous study, Notch signaling in the liver is gradually activated with the progression of CLF, while this pathological change is blocked by DAPT (a γ-secretase inhibitor) [12]. Thus, the inhibition of Notch signaling may be crucial for the treatment of CLF. In the present study, we first tested the effects of BM-MSCs with different Numb levels on hepatic Notch signaling after transplantation to clarify the mechanism by which changes in the Numb level in HSCs affect their differentiation. Consistent with our hypothesis, when Numb was deleted from BM-MSCs, the expression of Notch signaling factors downstream of RBP-Jκ and Hes1 was significantly increased. In addition, the mRNA expression levels of other components of Notch signaling, including Notch-2/-3/-4, JAG-1/-2 and DLL-1/-4, were also increased significantly and jointly promoted Notch signaling activation. When BM-MSCs overexpressed Numb, the expression levels of RBP-Jκ and Hes1 were significantly reduced. In addition, the mRNA expression levels of other components of Notch signaling, including Notch-2/-3/-4, JAG-1/-2 and DLL-3, were also significantly reduced and thus jointly inhibited the activation of Notch signaling. These results were also confirmed in vitro. Based on these results, the effect of Numb on determining the fate of HSCs depends on its negative regulation of Notch signaling, and this effect may be related to Numb-mediated promotion of Notch ubiquitination in rats with CLF.
In recent years, a large number of studies have shown that the biological antagonism between Numb and Notch controls the balance of stem cell proliferation and differentiation in development and homeostasis, and this biological antagonism depends on a series of ubiquitination processes. Mammalian *Numb is* clearly the substrate of E3 ubiquitin LNX, and wild-type LNX causes proteasome-dependent Numb degradation, which enhances the activity of Notch signaling [50]. On the other hand, Numb works with ITCH, another E3 ubiquitin ligase, in the cytoplasm to promote the ubiquitination of Notch in the cell membrane, thereby promoting the degradation of the NICD and avoiding its nuclear translocation and downstream target gene activation [51]. Thus, the balance of Numb–Notch ubiquitination may play an important role in maintaining liver homeostasis. In this study, after BM-MSCsNumb-KD transplantation, the level of the LNX mRNA did not change in the liver but that the level of ITCH mRNA was significantly reduced, suggesting that BM-MSCsNumb-KD transplantation attenuated the ubiquitination-mediated degradation of Notch. In contrast, after BM-MSCsNumb-OE transplantation, the level of LNX mRNA did not change, but the level of ITCH mRNA was significantly increased, suggesting that BM-MSCsNumb-OE transplantation promoted Notch ubiquitination. However, relevant evidence on methods to regulate the balance of Numb–Notch ubiquitination after hepatic Numb supplementation is still lacking.
In summary, Numb plays an important role in the occurrence and repair of CLF, and its key mechanism is to regulate Notch signaling and subsequently determine the differentiation of HSCs in livers of subjects with CLF. This study provides scientific evidence for improving the treatment of CLF by transplanting BM-MSCs with *Numb* gene editing. Of course, because BM-MSCs with Numb knockdown or overexpression were used for transplantation in this study, the method of conditional knockout or knock-in of liver Numb must be adopted to clarify the therapeutic value of Numb for CLF in the future.
## Conclusions
Numb is an important determinant of cell fate. In CLF, Numb determines the fate of HSCs, promotes their differentiation into hepatocytes and inhibits their differentiation into BECs by suppressing Notch signaling. Importantly, our results clearly indicated that the transplantation of BM-MSCs with Numb overexpression may be a useful new treatment strategy for CLF.
## Supplementary Information
Additional file 1. Supplementary Materials. Additional file 2. The original image of the immunoblotting.
## References
1. Hilscher MB, Kamath PS, Eaton JE. **Cholestatic liver diseases: a primer for generalists and subspecialists**. *Mayo Clin Proc* (2020) **95** 2263-2279. DOI: 10.1016/j.mayocp.2020.01.015
2. Polanco NP, Levy C, Martin EF. **Cholestatic liver diseases after liver transplant**. *Clin Liver Dis* (2017) **21** 403-420. DOI: 10.1016/j.cld.2016.12.011
3. Ghonem NS, Assis DN, Boyer JL. **Fibrates and cholestasis**. *Hepatology* (2015) **62** 635-643. DOI: 10.1002/hep.27744
4. Adams JM, Jafar-Nejad H. **The roles of Notch signaling in liver development and disease**. *Biomolecules* (2019) **9** 608. DOI: 10.3390/biom9100608
5. Lai EC. **Notch signaling: control of cell communication and cell fate**. *Development* (2004) **131** 965-973. DOI: 10.1242/dev.01074
6. Bray SJ. **Notch signalling: a simple pathway becomes complex**. *Nat Rev Mol Cell Biol* (2006) **7** 678-689. DOI: 10.1038/nrm2009
7. Liu K, Shen D, Shen J, Gao SM, Li B, Wong C. **The super elongation complex drives neural stem cell fate commitment**. *Dev Cell* (2017) **40** 537-551. DOI: 10.1016/j.devcel.2017.02.022
8. Sparks EE, Huppert KA, Brown MA, Washington MK, Huppert SS. **Notch signaling regulates formation of the three-dimensional architecture of intrahepatic bile ducts in mice**. *Hepatology* (2010) **51** 1391-1400. DOI: 10.1002/hep.23431
9. Carpino G, Cardinale V, Folseraas T, Overi D, Floreani A, Franchitto A. **Hepatic stem/progenitor cell activation differs between primary sclerosing and primary biliary cholangitis**. *Am J Pathol* (2018) **188** 627-639. DOI: 10.1016/j.ajpath.2017.11.010
10. Aimaiti Y, Yusufukadier M, Li W, Tuerhongjiang T, Shadike A, Meiheriayi A. **TGF-β1 signaling activates hepatic stellate cells through Notch pathway**. *Cytotechnology* (2019) **71** 881-891. DOI: 10.1007/s10616-019-00329-y
11. Li J, Dong S, Ye M, Peng G, Luo J, Wang C. **MicroRNA-489-3p represses hepatic stellate cells activation by negatively regulating the JAG1/Notch3 signaling pathway**. *Dig Dis Sci* (2021) **66** 143-150. DOI: 10.1007/s10620-020-06174-w
12. Zhang X, Du G, Xu Y, Li X, Fan W, Chen J. **Inhibition of notch signaling pathway prevents cholestatic liver fibrosis by decreasing the differentiation of hepatic progenitor cells into cholangiocytes**. *Lab Investig* (2016) **96** 350-360. DOI: 10.1038/labinvest.2015.149
13. Guo M, Jan LY, Jan YN. **Control of daughter cell fates during asymmetric division: interaction of Numb and Notch**. *Neuron* (1996) **17** 27-41. DOI: 10.1016/s0896-6273(00)80278-0
14. Loeffler D, Wehling A, Schneiter F, Zhang Y, Müller-Bötticher N, Hoppe PS. **Asymmetric lysosome inheritance predicts activation of haematopoietic stem cells**. *Nature* (2019) **573** 426-429. DOI: 10.1038/s41586-019-1531-6
15. Couturier L, Vodovar N, Schweisguth F. **Endocytosis by Numb breaks Notch symmetry at cytokinesis**. *Nat Cell Biol* (2012) **14** 131-139. DOI: 10.1038/ncb2419
16. Liu P, Mao Y, Xie Y, Wei J, Yao J. **Stem cells for treatment of liver fibrosis/cirrhosis: clinical progress and therapeutic potential**. *Stem Cell Res Ther* (2022) **13** 356. DOI: 10.1186/s13287-022-03041-5
17. He Q, Wang L, Zhao R, Yan F, Sha S, Cui C. **Mesenchymal stem cell-derived exosomes exert ameliorative effects in type 2 diabetes by improving hepatic glucose and lipid metabolism via enhancing autophagy**. *Stem Cell Res Ther* (2020) **11** 223. DOI: 10.1186/s13287-020-01731-6
18. Gazdic M, Arsenijevic A, Markovic BS, Volarevic A, Dimova I, Djonov V. **Mesenchymal stem cell-dependent modulation of liver diseases**. *Int J Biol Sci* (2017) **2017** 1109-1117. DOI: 10.7150/ijbs.20240.eCollection
19. Kharaziha P, Hellström PM, Noorinayer B, Farzaneh F, Aghajani K, Jafari F. **Improvement of liver function in liver cirrhosis patients after autologous mesenchymal stem cell injection: a phase I-II clinical trial**. *Eur J Gastroenterol Hepatol* (2009) **21** 1199-1205. DOI: 10.1097/MEG.0b013e32832a1f6c
20. Suk KT, Yoon JH, Kim MY, Kim CW, Kim JK, Park H. **Transplantation with autologous bone marrow-derived mesenchymal stem cells for alcoholic cirrhosis: phase 2 trial**. *Hepatology* (2016) **64** 2185-2197. DOI: 10.1002/hep.28693
21. Esmaeilzadeh A, Ommati H, Kooshyar MM, Jarahi L, Akhavan Rezayat K, Saberi S. **Autologous bone marrow stem cell transplantation in liver cirrhosis after correcting nutritional anomalies, a controlled clinical study**. *Cell J* (2019) **21** 268-273. DOI: 10.22074/cellj.2019.6108
22. Jia S, Liu X, Li W, Xie J, Yang L, Li L. **Peroxisome proliferator-activated receptor gamma negatively regulates the differentiation of bone marrow-derived mesenchymal stem cells toward myofibroblasts in liver fibrogenesis**. *Cell Physiol Biochem* (2015) **37** 2085-2100. DOI: 10.1159/000438567
23. Wang MY, Nestvold J, Rekda Ø, Kvalheim G, Fodstad Ø. **A novel rat fibrosarcoma cell line from transformed bone marrow-derived mesenchymal stem cells with maintained in vitro and in vivo stemness properties**. *Exp Cell Res* (2017) **352** 218-224. DOI: 10.1016/j.yexcr.2017.02.005
24. Mitchell JB, McIntosh K, Zvonic S, Garrett S, Floyd ZE, Kloster A. **Immunophenotype of human adipose-derived cells: temporal changes in stromal-associated and stem cell-associated markers**. *Stem Cells* (2006) **24** 376-385. DOI: 10.1634/stemcells.2005-0234
25. Pittenger MF, Mackay AM, Beck SC, Jaiswal RK, Douglas R, Mosca JD. **Multilineage potential of adult human mesenchymal stem cells**. *Science* (1999) **284** 143-147. DOI: 10.1126/science.284.5411.143
26. Gronthos S, Franklin DM, Leddy HA, Robey PG, Storms RW, Gimble JM. **Surface protein characterization of human adipose tissue-derived stromal cells**. *J Cell Physiol* (2001) **189** 54-63. DOI: 10.1002/jcp.1138
27. Jamall IS, Finelli VN, Que Hee SS. **A simple method to determine nanogram levels of 4-hydroxyproline in biological tissues**. *Anal Biochem* (1981) **112** 70-75. DOI: 10.1016/0003-2697(81)90261-x
28. Mu YP, Ogawa T, Kawada N. **Reversibility of fibrosis, inflammation, and endoplasmic reticulum stress in the liver of rats fed a methionine-choline-deficient diet**. *Lab Invest* (2010) **90** 245-256. DOI: 10.1038/labinvest.2009.123
29. Tsao MS, Smith JD, Nelson KG, Grisham JW. **A diploid epithelial cell line from normal adult rat liver with phenotypic properties of ‘oval’ cells**. *Exp Cell Res* (1984) **154** 38-52. DOI: 10.1016/0014-4827(84)90666-9
30. Couchie D, Holic N, Chobert MN, Corlu A, Laperche Y. **In vitro differentiation of WB-F344 rat liver epithelial cells into the biliary lineage**. *Differentiation* (2002) **69** 209-215. DOI: 10.1046/j.1432-0436.2002.690414.x
31. Nie J, McGill MA, Dermer M, Dho SE, Wolting CD, McGlade CJ. **LNX functions as a RING type E3 ubiquitin ligase that targets the cell fate determinant Numb for ubiquitin-dependent degradation**. *EMBO J* (2002) **21** 93-102. DOI: 10.1093/emboj/21.1.93
32. Di Marcotullio L, Greco A, Mazzà D, Canettieri G, Pietrosanti L, Infante P. **Numb activates the E3 ligase Itch to control Gli1 function through a novel degradation signal**. *Oncogene* (2011) **30** 65-76. DOI: 10.1038/onc.2010.394
33. McGill MA, McGlade CJ. **Mammalian numb proteins promote Notch1 receptor ubiquitination and degradation of the Notch1 intracellular domain**. *J Biol Chem* (2003) **278** 23196-23203. DOI: 10.1074/jbc.M302827200
34. Suskind DL, Muench MO. **Searching for common stem cells of the hepatic and hematopoietic systems in the human fetal liver: CD34**. *J Hepatol* (2004) **40** 261-268. DOI: 10.1016/j.jhep.2003.11.007
35. Mishra L, Banker T, Murray J, Byers S, Thenappan A, He AR. **Liver stem cells and hepatocellular carcinoma**. *Hepatology* (2009) **49** 318-329. DOI: 10.1002/hep.22704
36. Messina A, Luce E, Hussein M, Dubart-Kupperschmitt A. **Pluripotent-stem-cell-derived hepatic cells: hepatocytes and organoids for liver therapy and regeneration**. *Cells* (2020) **9** 420. DOI: 10.3390/cells9020420
37. Choi SS, Diehl AM. **Epithelial-to-mesenchymal transitions in the liver**. *Hepatology* (2009) **50** 2007-2013. DOI: 10.1002/hep.23196
38. Huang M, Chang A, Choi M, Zhou D, Anania FA, Shin CH. **Antagonistic interaction between Wnt and Notch activity modulates the regenerative capacity of a zebrafish fibrotic liver model**. *Hepatology* (2014) **60** 1753-1766. DOI: 10.1002/hep.27285
39. Fiorotto R, Raizner A, Morell CM, Torsello B, Scirpo R, Fabris L. **Notch signaling regulates tubular morphogenesis during repair from biliary damage in mice**. *J Hepatol* (2013) **59** 124-130. DOI: 10.1016/j.jhep.2013.02.025
40. Fabris L, Brivio S, Cadamuro M, Strazzabosco M. **Revisiting epithelial-to-mesenchymal transition in liver fibrosis: clues for a better understanding of the “reactive” biliary epithelial phenotype**. *Stem Cells Int* (2016). DOI: 10.1155/2016/2953727
41. Carpino G, Nevi L, Overi D, Cardinale V, Lu WY, Di Matteo S. **Peribiliary gland niche participates in biliary tree regeneration in mouse and in human primary sclerosing cholangitis**. *Hepatology* (2020) **71** 972-989. DOI: 10.1002/hep.30871
42. Colaluca IN, Basile A, Freiburger L, D'Uva V, Disalvatore D, Vecchi M. **A Numb-Mdm2 fuzzy complex reveals an isoform-specific involvement of Numb in breast cancer**. *J Cell Biol* (2018) **217** 745-762. DOI: 10.1083/jcb.201709092
43. Flores AN, McDermott N, Meunier A, Marignol L. **NUMB inhibition of NOTCH signalling as a therapeutic target in prostate cancer**. *Nat Rev Urol* (2014) **11** 499-507. DOI: 10.1038/nrurol.2014.195
44. Jung KH, Zhang J, Zhou C, Shen H, Gagea M, Rodriguez-Aguayo C. **Differentiation therapy for hepatocellular carcinoma: multifaceted effects of miR-148a on tumor growth and phenotype and liver fibrosis**. *Hepatology* (2016) **63** 864-879. DOI: 10.1002/hep.28367
45. Katoh M, Katoh M. **NUMB is a break of WNT-Notch signaling cycle**. *Int J Mol Med* (2006) **18** 517-521. DOI: 10.3892/ijmm.18.3.517
46. Wu XB, Tao R. **Hepatocyte differentiation of mesenchymal stem cells**. *HBPD Int* (2012) **11** 360-371. DOI: 10.1016/s1499-3872(12)60193-3
47. Cai Y, Zou Z, Liu L, Chen S, Chen Y, Lin Z. **Bone marrow-derived mesenchymal stem cells inhibits hepatocyte apoptosis after acute liver injury**. *Int J Clin Exp Pathol* (2015) **8** 107-116. PMID: 25755697
48. Guo Z, Ohlstein B. **Stem cell regulation. Bidirectional Notch signaling regulates Drosophila intestinal stem cell multipotency**. *Science* (2015) **350** aab0988. DOI: 10.1126/science.aab0988
49. Huang G, Wang M, Li X, Wu J, Chen S, Du N. **TUSC7 suppression of Notch activation through sponging MiR-146 recapitulated the asymmetric cell division in lung adenocarcinoma stem cells**. *Life Sci* (2019) **232** 116630. DOI: 10.1016/j.lfs.2019.116630
50. Lenihan JA, Saha O, Heimer-McGinn V, Cryan JF, Feng G, Young PW. **Decreased anxiety-related behaviour but apparently unperturbed NUMB function in ligand of NUMB protein-X (LNX) 1/2 doubles knockout mice**. *Mol Neurobiol* (2017) **54** 8090-8109. DOI: 10.1007/s12035-016-0261-0
51. Tarlow BD, Finegold MJ, Grompe M. **Clonal tracing of Sox9**. *Hepatology* (2014) **60** 278-289. DOI: 10.1002/hep.27084
|
---
title: 'Comprehensive evaluation of the influence of sex differences on composite
disease activity indices for rheumatoid arthritis: results from a nationwide observational
cohort study'
authors:
- Takahiro Nishino
- Atsushi Hashimoto
- Shigeto Tohma
- Toshihiro Matsui
journal: BMC Rheumatology
year: 2023
pmcid: PMC10029312
doi: 10.1186/s41927-023-00328-9
license: CC BY 4.0
---
# Comprehensive evaluation of the influence of sex differences on composite disease activity indices for rheumatoid arthritis: results from a nationwide observational cohort study
## Abstract
### Background
The effects and their magnitudes of sex on disease activity indices for rheumatoid arthritis are not clear. We aimed to comprehensively evaluate the influence of sex on disease activity indices in the real-world setting using a large observational database.
### Methods
We analyzed 14,958 patients registered in the National Database of Rheumatic Diseases in Japan (NinJa) in 2017. We evaluated the sex differences in the 28-joint disease activity score (DAS28) using erythrocyte sedimentation rate (ESR), DAS28 using C-reactive protein (DAS28-CRP), Simplified Disease Activity Index (SDAI), and Clinical Disease Activity Index by disease activity category using Cliff’s delta and regression analysis. Differences in the share of components of indices were evaluated using permutational multivariate analysis of variance. Correction equations were constructed to estimate the number of misclassification in male patients who achieve DAS28-ESR remission.
### Results
DAS28-ESR showed higher values in female patients than male patients in remission despite no obvious difference in other indices or disease activity categories. Among the components of DAS28-ESR, only ESR was higher in female patients than male patients in remission. In DAS28-CRP and SDAI, 28-tender joint count was higher and CRP was lower in female patients than male patients. In addition, the profiles in the components were different between female and male patients, especially among those with high disease activity. Using correction equations, almost $12\%$ of male patients with DAS28-ESR remission were estimated to be misclassified, mainly due to differences in ESR.
### Conclusion
Among the disease activity indices, significant sex difference was observed only in DAS28-ESR remission. The degree of misclassification in DAS28-ESR remission would be unignorable.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s41927-023-00328-9.
## Introduction
The concept of “Treat to Target” (T2T), which involves setting a goal and determining an appropriate treatment, has improved outcomes in patients with rheumatoid arthritis (RA) [1–3]. Assessing the disease activity is an important part of T2T. Disease activity is usually evaluated using scoring systems, such as 28-joint disease activity score (DAS28) using erythrocyte sedimentation rate (ESR), DAS28 using C-reactive protein (DAS28-CRP), Clinical Disease Activity Index (CDAI), and Simplified Disease Activity Index (SDAI) [4–7]. Although these indices are widely used and recommended to assess RA disease activity [8], they are influenced by sex, age, body mass index, and other factors [9–21]. These effects and their magnitude on disease activity indices are not clear; therefore, disease activity should be carefully interpreted based on the properties of each index and individual patient factors.
Sex differences in RA have been studied from multifaceted aspects, for example, incidence, phenotype, comorbidities, treatment response, and prognosis [22]. In addition, sex differences in disease activity indices have been studied, mainly in DAS28-ESR. Many studies have reported that DAS28-ESR is lower in male patients compared to female patients [10–16], and ESR is considered to contribute to the sex differences [10, 12, 13]. The association of sex difference with CRP level and discrepancy between DAS28-ESR and DAS28-CRP have been studied [16, 23–27], but few studies have evaluated the effect of sex differences on DAS28-CRP [16]. Although the effects of sex differences on CDAI and SDAI have not been thoroughly evaluated, these indices are influenced by pain perception and sex [9]. Therefore, the effect of sex differences on these indices cannot be ignored. Nevertheless, sex differences are not taken into account while assessing disease activity and it is not clear how sex differences influence the composite measure indices, which may lead to biased interpretation of these indices. The current problem is a lack of large-scale systematic evaluation of the effects of sex differences on disease activity indices in the era of biologics and molecular targeting therapy. Most previous studies that evaluated the effects of sex differences on disease activity indices were conducted in the 2000s and involved less than 1000 patients. Since then, the profile of drugs used in RA patients has changed. However, the effects of sex differences on disease activity indices are not well clear in patients treated with biologics or molecular targeting therapy. Furthermore, most previous studies only performed a simple comparison of the disease activity indices without stratification, which may introduce bias in the results due to the combined analysis of patients with varying disease activity. Thus, it is unclear whether the previous studies accurately evaluated the effects of sex differences on disease activity indices or the results were obtained due to differences in disease activity between the groups. For the aforementioned reasons, the effects of sex differences on disease activity indices, while taking into account the different drugs used, require a comprehensive evaluation. Therefore, we aimed to evaluate the influence of sex differences on composite measure indices and their clinical impact by analyzing the impact on each disease activity category using a large nationwide observational database. The results would enhance our understanding and allow more appropriate use of composite measure indices by taking into account the sex differences.
## Study population
We collected data from a nationwide observational cohort database of RA in Japan (National Database of Rheumatic Diseases in Japan; NinJa) [26, 28] in 2017. Forty-nine hospitals and institutions from all over Japan participated in the NinJa project in 2017. NinJa included RA patients diagnosed according to the standard diagnostic criteria for RA [29–32], regardless of disease duration, onset age, and treatment. Once a year, NinJa collects information about important events (e.g., hospitalization, surgical operation, malignancy, tuberculosis, herpes zoster, or childbirth) and data are arbitrarily collected at one point in the year for each patient, including 28-tender joint count (TJC28), 28-swollen joint count (SJC28), disease activity indices, Health Assessment Questionnaire-Disability Index (HAQ-DI), drug use, and joint destruction. The collected data are curated in National Hospital Organization Sagamihara National Hospital and verified in case of doubt about accuracy. Of a total of 15,185 patients registered in NinJa in 2017, 15,056 had onset of RA at age > 16 years. We analyzed 14,958 out of 15,056 patients, thereby excluding 98 patients in whom the drug used was unknown (94 patients) or not approved for RA in Japan (4 patients).
## Measures and disease activity categories
We evaluated the influence of sex differences on DAS28-ESR, DAS28-CRP, CDAI, and SDAI. Patients were classified into remission, low disease activity (LDA), moderate disease activity (MDA), and high disease activity (HDA) based on DAS28-ESR, DAS28-CRP, CDAI, and SDAI, in accordance with the updated American College of Rheumatology recommendations [8].
## Statistical analysis
Statistical analyses were performed using R version 4.0.3 software (R Foundation for Statistical Computing). The URLs and/or references of R packages and function are listed in the Additional file 1. The values and graphs are the results of available-case analysis (pairwise deletion), unless stated otherwise. Figures were generated using ggplot2 or car package or geom_flat_violin function. To compare continuous variables, differences in $25\%$ trimmed mean between female and male patients were calculated and its $95\%$ confidence intervals (CIs) were obtained by the percentile bootstrap method (5000 iterations) using simpleboot and boot packages. Cliff’s delta, non-parametric effect size, and its $95\%$ CIs were calculated using effsize package for comparing disease activity indices and their components between female and male patients. The magnitude of effect size based on Cliff’s delta for disease activity indices are not established, thus thresholds of magnitude for the absolute values were assessed that less than 0.147, 0.147 or more and less than 0.330, 0.330 or more and less than 0.474, and 0.474 or more corresponded to negligible, small, medium, and large according to the threshold values proposed by Romano et al. [ 33].
Permutational multivariate analysis of variance (PERMANOVA) was performed using vegan package for examining the difference in the share of components to disease activity indices between female and male patients in each disease activity category (number of permutations: 1000). The share of components is considered to be compositional data containing essential zeros, which cannot calculate Aitchison’s distance. Thus, Bray–Curtis dissimilarity was applied for the analysis of PERMANOVA. The threshold of p-value was not defined because this study used observational database without prespecified analysis plan and sample size design to control for type I error.
The results of regression analysis were confirmed by generalized linear model (GLM) with gamma distribution using identity link and quantile (median) regression (QR) model. Although the canonical link function of GLM with gamma distribution provides a reciprocal link, we used identity link because coefficients can be interpreted in terms of the effects of independent variables on disease activity indices at the original scale. Dependent variable in GLM with gamma distribution should not contain zero or negative values. In case that the remission dataset to be analyzed in GLM contained patients whose disease activity indices were zero, a small value (1.0 × 10–15) was added to the value for all patients in the dataset. QR was performed using quantreg package.
In regression models adjusting for patient-related factors, analysis using stacked dataset imputed by chained equations were also performed in addition to the available-case analysis to confirm the robustness of the results. Multivariate imputation by chained equations were conducted using mice package. Imputation method and models are described in the Supplementary Methods (see Additional file 1). As conventional procedure of multiple imputation, the coefficients of regression models were pooled by Rubin’s rule. However, imbalance of cases between imputed datasets occurs by stratifying disease activity categories due to imputation of the disease activity indices; therefore, the results cannot be combined by Rubin’s rule. Alternatively, we analyzed the stacked imputed dataset, which was deemed as the complete data. The point estimates calculated by the stacked method are unbiased, but the confidence intervals are invalid [34]. Thus, we present the coefficients and their confidence intervals in available-case analysis as the main results.
## Patient characteristics
Among a total of 14,958 patients analyzed in this study, 11,916 were female patients ($79.7\%$) and 3042 were male patients ($20.3\%$). Table 1 presents the patient characteristics. Male patients were older and had a higher age at onset, whereas female patients had longer disease duration, higher HAQ-DI score, and higher values of disease activity indices. Male patients had higher remission rate than female patients in all disease activity indices, especially DAS28-ESR. Of the 14,958 patients, 1043 were not treated with disease modifying anti-rheumatic drugs (DMARDs), 9593 were treated with conventional synthetic DMARDs (csDMARDs) alone, 2100 were treated with tumor necrotizing factor inhibitors (TNFi), 1144 were treated with interleukin 6 receptor inhibitors (IL-6i), 772 were treated with cytotoxic T lymphocyte-associated antigen 4 immunoglobulin (CTLA-4-Ig), and 306 were treated with Janus kinase inhibitors (JAKi). Additional file 1: Tables S1–S6 present the patient characteristics by sex for each treatment type (see Additional file 1).Table 1Patient characteristicsFemale ($$n = 11$$,916)Male ($$n = 3042$$)Δ$25\%$ trimmed mean ($95\%$ CI)Age, years68 (58–75)69 (62–76) − 1.8 (− 2.3 to − 1.3)Age at onset, years52 (41–62)59 (50–67) − 6.9 (− 7.5 to − 6.3)Disease duration, years11 (6–20)8 (4–14)3.6 (3.2 to 3.9)Number of artificial joints0 (Q1–Q3, 0–0; range, 0–9)0 (Q1–Q3, 0–0; range, 0–7)0.0 (0.0 to 0.0)Stage I2752 (25.3)980 (35.6)– II2908 (26.8)928 (33.7)– III2094 (19.3)479 (17.4)– IV3103 (28.6)363 (13.2)– Missing data1059292–Class I3762 (34.6)1247 (45.0)– II4960 (45.6)1168 (42.2)– III1848 (17.0)306 (11.0)– IV314 (2.9)50 (1.8)– Missing data1032271–BMI, kg/m221.92 (3.73)23.01 (3.38) − 1.29 (− 1.44 to − 1.15) Missing data1722422Steroid Regular use4110 (34.5)1137 (37.4)– Missing data01–NSAIDs Regular use3609 (30.3)959 (31.5)– Missing data10–RF, IU/mL44.0 (14.0–121.0)45.0 (10.0–160.2) − 5.69 (− 11.27 to − 0.67) Positive (> 15)6872 (73.7)1614 (68.6)– Missing data2598690–Anti-CCP, U/mL52.9 (3.6–266.0)60.5 (0.7–338.5) − 16.00 (− 34.00 to 0.72) Positive (≥ 4.5)2970 (74.1)806 (67.7)– Missing data79081851–HAQ-DI0.38 (0.00–1.00)0.00 (0.00–0.50)0.275 (0.249 to 0.301) Missing data2991777–*Smoking status* Never7734 (79.1)625 (24.7)– Former1386 (14.2)1322 (52.3)– Current654 (6.7)583 (23.0)– Missing data2142512–DAS28-ESR2.9 (2.2–3.7)2.5 (1.8–3.3)0.39 (0.33 to 0.45) Remission3552 (40.7)1183 (53.9)– Low1857 (21.3)386 (17.6)– Moderate2891 (33.1)531 (24.2)– High428 (4.9)94 (4.3)– Missing data3188848–DAS28-CRP2.1 (1.5–2.9)2.0 (1.5–2.8)0.09 (0.04 to 0.13) Remission6868 (67.1)1815 (69.5)– Low1509 (14.7)362 (13.9)– Moderate1691 (16.5)387 (14.8)– High170 (1.7)46 (1.8)– Missing data1678432–CDAI4.8 (1.9–9.2)3.7 (1.3–7.8)1.05 (0.82 to 1.29) Remission3537 (34.7)1110 (42.6)– Low4483 (43.9)1060 (40.7)– Moderate1851 (18.1)363 (13.9)– High333 (3.3)72 (2.8)– Missing data1712437–SDAI5.2 (2.1–9.9)4.2 (1.6–8.6)0.88 (0.62 to 1.12) Remission3684 (36.2)1101 (42.3)– Low4388 (43.1)1053 (40.5)– Moderate1853 (18.2)382 (14.7)– High247 (2.4)64 (2.5)– Missing data1744442–The values are n, n (%), mean (SD) or median (Q1–Q3). Number of artificial joints represents median, Q1–Q3 and range. SD Standard deviation; Q1 First quartile; Q3 Third quartile; $95\%$ CI $95\%$ confidence intervalIf categorical data contains missing data, the percentages are calculated with its denominator as the number subtracting the number of missing data from total number. If continuous data contains missing data, the representative values are the results of available-case analysisThe total of percentage may not equal to 100 due to roundingThe value of difference in $25\%$ trimmed mean (Δ$25\%$ trimmed mean) is calculated by subtracting the male value from the female value
## Sex differences in DAS28-ESR, DAS28-CRP, CDAI, and SDAI by disease activity category
A comparison of the distributions of DAS28-ESR, DAS28-CRP, CDAI, and SDAI values between female and male patients by disease activity category showed no difference, except for remission in DAS28-ESR (Fig. 1A). Cliff’s delta also showed sex difference only in DAS28-ESR remission, with higher values in female patients than male patients (Fig. 1B). Furthermore, we evaluated the sex difference in DAS28-ESR components and explored whether the components of these indices that showed no sex differences were less sensitive to sex difference. Cliff’s delta showed that ESR was higher in female patients than male patients in remission, while other components of DAS28-ESR and ESR in other disease activity categories showed no obvious sex differences (Cliff’s delta for each component of DAS28-ESR are shown in Fig. 1C and that of other indices are shown in Additional file 1: Fig. S1; the values of Cliff’s delta are shown in Additional file 1: Table S7). Therefore, we concluded that sex difference in DAS28-ESR remission was mainly caused by sex difference in ESR. Cliff’s delta was also calculated for each treatment group (Additional file 1: Table S7), which showed higher DAS28-ESR values and ESR for female patients in the DMARDs free, csDMARDs, TNFi, and IL-6i groups compared to male patients in remission; however, the results for CTLA-4-Ig and JAKi were equivocal. Fig. 1Comparison of DAS28-ESR, DAS28-CRP, CDAI, and SDAI between female and male patients. A Distributions of disease activity indices by sex. Jitter, box, and violin plots are depicted in each disease activity category by sex (red, female; blue, male). Dashed lines are drawn on the cutoff values for the disease activity categories (DAS28-ESR and DAS28-CRP: 2.6, 3.2, and 5.1. CDAI: 2.8, 10, and 22. SDAI: 3.3, 11, and 26). B, C Cliff’s delta for sex difference in disease activity indices. Positive values of Cliff’s delta indicate that the values of the indices and their components were higher in female patients compared to male patients, whereas negative values indicate the opposite. Black line is drawn at the value of 0.000. Blue, green, and red dashed lines are drawn on the values of 0.147, 0.330, and 0.474, respectively. Points and bars indicate the estimates and $95\%$ CIs of Cliff’s delta, respectively. Rem, remission; LDA, low disease activity; MDA, moderate disease activity; HDA, high disease activity Sex differences were also observed in TJC28 and CRP in DAS28-CRP and SDAI (Additional file 1: Fig. S1, Table S7). In DAS28-CRP and SDAI, TJC28 tended to be higher in female patients compared to male patients in LDA to MDA and showed obvious difference in HDA. In contrast, CRP tended to be higher in male patients compared to female patients in remission to MDA and showed obvious difference in HDA in DAS28-CRP and SDAI. No obvious differences were observed in the other components of DAS28-CRP and SDAI. In each treatment group, similar results were observed in DMARDs free and csDMARDs groups (Additional file 1: Table S7). However, in the other treatment groups, the sex differences in TJC28 and CRP were equivocal or could not be evaluated due to small sample size.
## Profile in the share of components of disease activity indices
Considering the contrasting dynamics of TJC28 and CRP as well as no obvious differences in the other components of DAS28-CRP and SDAI, DAS28-CRP and SDAI had no sex differences because TJC28 and CRP cancel each other. Therefore, the share of each component of the composite measures would be different between female and male patients, even if the values of the composite measures are the same. We compared the share of components between female and male patients. The share of each component to the total value of the composite measures was calculated using the procedure described by Radovits et al. [ 10], after excluding patients whose disease activity index value was zero. The share of components of each composite measure by disease activity category is presented in Fig. 2 and was similar between female and male patients, except for HDA in DAS28-CRP and SDAI. In HDA, the share of TJC28 in DAS28-CRP and SDAI was slightly lower in male patients compared to female patients. Distributions of the share supported that the share of TJC28 tended to be higher in female patients compared to male patients and that of CRP was the opposite in the two indices (Additional file 1: Fig. S2). Furthermore, we performed PERMANOVA to quantify the impact of sex on the share (Additional file 1: Table S8). In HDA, the R2 values in DAS28-CRP and SDAI were relatively higher than others, while sex only partly explained the variance (DAS28-CRP, R2 = 0.06577; SDAI, R2 = 0.02520). Thus, we concluded that the composition was similar between female and male patients, but the component profile in HDA slightly differed in DAS28-CRP and SDAI.Fig. 2Share of the components of disease activity indices by sex. The stacked bar plot shows the share in each patient. The horizontal axis indicates the number of patients and the vertical axis indicates percentage of share of each component. Rem, remission; LDA, low disease activity; MDA, moderate disease activity; HDA, high disease activity
## Adjustment of patient-related factors by regression models
We performed multivariable regression analysis by disease activity category to adjust patient-related factors using GLM and QR model. In both GLM and QR model, sex differences were only seen in remission in DAS28-ESR after adjustment of patient-related factors (Table 2 shows only the results of “Male” variable; the results of all variables are shown in Additional file 1: Table S9). The regression analysis using stacked dataset imputed by chained equations showed similar results (Additional file 1: Table S10).Table 2Partial regression coefficients of sex in GLM and QR models for adjustment of patient-related factorsRemissionLDAMDAHDAEstimate ($95\%$ CI)Estimate ($95\%$ CI)Estimate ($95\%$ CI)Estimate ($95\%$ CI)DAS28-ESR (GLM)− 0.281 (− 0.318 to − 0.244)− 0.003 (− 0.024 to 0.017)− 0.029 (− 0.079 to 0.023)0.011 (− 0.131 to 0.156)DAS28-ESR (QR)− 0.235 (− 0.285 to − 0.187)0.000 (− 0.022 to 0.025)− 0.033 (− 0.092 to 0.066)0.016 (− 0.131 to 0.149)DAS28-CRP (GLM)0.026 (0.002 to 0.051)0.004 (− 0.019 to 0.027)0.039 (− 0.022 to 0.100)0.136 (− 0.066 to 0.342)DAS28-CRP (QR)0.053 (0.010 to 0.094)0.008 (− 0.031 to 0.051)0.036 (− 0.040 to 0.142)0.036 (− 0.168 to 0.255)CDAI (GLM)− 0.017 (− 0.078 to 0.046)− 0.072 (− 0.219 to 0.079)0.126 (− 0.277 to 0.538)0.314 (− 1.941 to 2.681)CDAI (QR)0.033 (− 0.081 to 0.091)− 0.032 (− 0.231 to 0.221)0.140 (− 0.484 to 0.743)− 0.484 (− 2.355 to 1.840)SDAI (GLM)0.013 (− 0.055 to 0.084)− 0.058 (− 0.215 to 0.102)0.250 (− 0.231 to 0.743)0.821 (− 1.629 to 3.374)SDAI (QR)0.054 (− 0.042 to 0.129)0.013 (− 0.204 to 0.219)0.563 (− 0.007 to 1.247)0.184 (− 2.059 to 2.523)GLM Generalized linear model; QR Quantile regression; LDA Low disease activity; MDA Moderate disease activity; HDA High disease activity
## Estimation of the impact of sex difference on remission rate
To quantify the clinical impact of sex difference in DAS28-ESR remission, we constructed correction equations using GLM and QR model. Previous studies suggest that ESR levels change in a sex specific age-dependent manner [10, 35]. Assuming that tender and swollen joint count is surrogate index which accurately reflects disease activity under the condition ESR and DAS28-ESR may not accurately reflect disease activity, we used regression model with DAS28-ESR as a dependent variable and age, 0.56 × √(TJC28) + 0.28 × √(SJC28), sex, interaction term of sex and age, and interaction term of sex and 0.56 × √(TJC28) + 0.28 × √(SJC28) as independent variables for patients with DAS28-ESR remission. We calculated the coefficients using both GLM and QR model (Table 3). Using the results of regression analysis, we constructed the following correction equations (Eqs. 1 and 2).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.825 - 0.008 \times {\text{age}} - 0.177 \times \left({0.56 \times \surd \left({{\text{TJC}}28} \right) + 0.28 \times \surd \left({{\text{SJC}}28} \right)} \right)$$\end{document}0.825-0.008×age-0.177×0.56×√TJC28+0.28×√SJC282\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.021 - 0.011 \times {\text{age}} - 0.202 \times \left({0.56 \times \surd \left({{\text{TJC}}28} \right) + 0.28 \times \surd \left({{\text{SJC}}28} \right)} \right)$$\end{document}1.021-0.011×age-0.202×0.56×√TJC28+0.28×√SJC28Table 3Regression analysis for construction of correction equationsGLMQREstimate ($95\%$ CI)Estimate ($95\%$ CI)Intercept1.407 (1.325 to 1.490)1.503 (1.375 to 1.614)Age0.008 (0.006 to 0.009)0.007 (0.006 to 0.010)Joint0.426 (0.351 to 0.503)0.384 (0.333 to 0.449)Male− 0.825 (− 0.983 to − 0.664)− 1.021 (− 1.229 to − 0.686)Age: male0.008 (0.006 to 0.010)0.011 (0.006 to 0.014)Joint: male0.177 (0.040 to 0.318)0.202 (0.061 to 0.297)The variable “Joint” represents 0.56 × √(TJC28) + 0.28 × √(SJC28)The variable “Male” is dummy variable that female is used as referenceThe variable “Age: Male” represents interaction term of age and male sexThe variable “Joint: Male” represents interaction term of joint findings and male sexGLM Generalized linear model; QR Quantile regression If the values of equations are above zero, they were added to DAS28-ESR. If the values are zero or below, the original DAS28-ESR value was used. These equations were only applied to male patients. After correction using the above equations, sex difference was deemed to be negligible (Fig. 3A and Additional file 1: Fig. S3). Cliff’s delta indicated negligible sex difference after correction [Eq. [ 1], − 0.0530 ($95\%$ CI − 0.0887 to − 0.0172); Eq. [ 2], − 0.0446 ($95\%$ CI − 0.0800 to − 0.0090)]. Of the total 1183 male patients in DAS28-ESR remission (available-case analysis), 143 (almost $12.1\%$) and 141 (almost $11.9\%$) male patients in DAS28-ESR remission were classified as LDA by correction using Eqs. 1 and 2, respectively (Fig. 3B). Therefore, almost $12\%$ of male patients who achieve DAS28-ESR remission criteria are estimated to be misclassified. Fig. 3The effect of correction by equations on DAS28-ESR. A Relationship between DAS28-ESR, age, and joint findings before and after corrections. Three-dimensional scatter plot with fitted surface are depicted by sex (red, female; blue, male). “ Joint” represents 0.56 × √(TJC28) + 0.28 × √(SJC28). B Distributions of DAS28-ESR before and after corrections. Jitter, box, and violin plots are depicted for each disease activity category by sex (red, female; blue, male). Dashed lines are drawn at the value of 2.6 in DAS28-ESR
## Discussion
This study confirmed that among DAS28-ESR, DAS28-CRP, CDAI, and SDAI, sex difference in the composite measures is observed only in remission based on DAS28-ESR. In addition, sex difference in DAS28-ESR remission is mainly due to difference in ESR, which is consistent with previous studies [12, 13, 36]. However, our results showed no meaningful sex differences in other components of DAS28-ESR, which is not consistent with previous studies [12–14, 36]. In previous studies, tender joint count and patient global assessment values, as well as ESR, were consistently higher in female patients compared to male patients. The reason for this discrepancy is uncertain, but may result from differences in stratification and study population (e.g., remission rate or pain intensity). The previous studies compared female and male patients without stratification or with stratification by swollen joint count, while we stratified patients by disease activity based on each composite measure. Therefore, the populations analyzed in the previous studies may have included patients with various disease activities than our analysis. Both tender joint count and patient global assessment are affected by pain [13, 37]. Although the mechanisms underlying sex difference in pain perception are not fully clear, female sex is more sensitive to pain than male sex [38, 39]. Thus, pain may contribute to sex differences in composite measures. However, adjusting for individual differences in pain intensity is challenging because the most frequently used measures, Visual Analog Scale and Numeric Rating Scale, are subjective and dependent on individual pain tolerance and other factors [40].
We found that the values of composite measures do not differ between female and male patients, but profile in the share of components differ between the sexes in composite measures including CRP. Several studies reported that sex difference is not observed in CRP level in RA patients [13, 36], and it is widely believed that CRP is less sensitive to sex difference than ESR. However, some studies reported that the CRP level is higher in male patients compared to female patients, similar to the results of our study [14, 16, 23]. Although, this discrepancy needs to be evaluated in further studies, our findings suggest that results of clinical trials for patients whose disease activity is assessed as HDA by DAS28-CRP and SDAI may be carefully interpreted because of the possibility that study population is not homogeneous between female and male patients. Compared to DAS28-ESR, sex differences in DAS28-CRP, CDAI, and SDAI have not been well studied. This study revealed behaviors of DAS28-CRP, CDAI, and SDAI in sex difference using stratification by disease activity category and evaluation by effect size at the level of components. The importance of our results is that they clarify whether or not disease activity indices are affected by sex differences in each disease activity category. Evidence for robustness to sex differences is necessary for the appropriate selection of indices to be used in clinical trials and routine clinical practice. Based on the findings of our study, CDAI is the most robust to sex difference among the four indices. Our study provides a cautionary implication that composite measure indices should be carefully interpreted in light of confounders, such as sex difference, before making clinical decisions based on the values of composite measure indices.
In this study, we also found that almost $12\%$ of male patients may be misclassified as being in remission based on DAS28-ESR, suggesting that the criterion overestimates the remission rate in male patients. Generally, remission rate is lower in female patients compared to male patients, and male sex is a predictor of remission [41, 42]. In our study, sex difference had a greater effect on remission rate defined using the DAS28-ESR criteria compared to other criteria; however, to a lesser extent, the remission rates were higher in all disease activity indices in male patients compared to female patients. Previous studies showed that the remission rate was lower in female patients compared to male patients using DAS28-ESR, but not when other criteria were used, thereby suggesting that the sex difference in remission rate based on DAS28-ESR is due to sex difference in ESR [36, 43]. Although further study is needed to determine whether sex difference in remission rate may be explained by factors other than bias caused by composite measures themselves, e.g., pathophysiological or psychosocial differences, most part of the sex difference in remission rate based on DAS28-ESR would be explained by ESR [12, 36]. Overestimation of the remission rate due to sex difference in ESR is important in the context of determining treatment outcomes. In early RA, achieving remission confer favorable radiographic, quality of life and functional outcome compared to LDA [44]. Thus, misclassification of LDA into remission would have great impact on clinical decision making in T2T era, while further studies are needed to investigate the effects of misclassification of male patients with LDA into remission due to ESR on the radiographic appearance, quality of life, and functional outcome in both early and established RA using longitudinal data.
The present study had several limitations. First, we did not analyze all the factors that affect ESR and other components of the indices, such as hematocrit level, alcohol consumption, race, ethnicity, and history of fibromyalgia, hypergammaglobulinemia, Sjögren's syndrome, or other comorbidities. Second, the correction equations were not validated in an independent dataset. Therefore, correction using these equations is not yet suitable for clinical use, and estimated misclassification rates may fluctuate especially in applying to non-Japanese populations. Moreover, the equations and estimations are based on the assumption that tender and swollen joint count accurately reflects disease activity because no unbiased gold standard measure of disease activity has been established. The definitions of disease activity and remission are complex and controversial with respect to whether they should be defined solely by the state of inflammation (e.g., joint findings, ESR, or CRP) or by the combination of patients’ overall health status (e.g., patient global assessment) and state of inflammation [45, 46]. The dual target strategy, which separately manages inflammation measures as disease activity and patient-reported outcomes as disease impact, may avoid overtreatment and improve the quality of life of patients [46, 47]. Patient-reported outcomes are not only proposed as the pillar of the dual target strategy, but are also associated with functional outcomes and sustained remission [48–50]. Thus, it is unclear whether the assumption that joint findings are a proxy for disease activity is optimal. However, to the best of our knowledge, this is the first report to quantitatively determine the impact of bias due to sex difference on remission using correction equations, and these are candidates to correct sex difference in remission based on DAS28-ESR.
## Conclusion
Large-scale and detailed analysis, including the drug type used, of the disease activity indices showed significant sex differences only in DAS28-ESR remission, mainly due to differences in ESR. In DAS28-CRP and SDAI, the values of the composite measures showed no significant sex differences, but TJC28 was higher and CRP was lower in female patients than male patients. This sex difference in the components indicated that the profiles of male and female patients were different, especially in those with high disease activity. Furthermore, almost $12\%$ of male patients with DAS28-ESR remission were considered to be equivalent to LDA using equations to correct the effects of sex and age differences on ESR. Our results will help to understand the properties of composite measures of disease activity and allow the appropriate selection of indices based on the sex differences.
## Supplementary Information
Additional file 1: Methods S1. Imputation method and models for multiple imputation by chained equations. Methods S2. Predictor matrix. Fig. S1. Cliff’s delta for sex difference in each component in available-case analysis. Fig. S2. Distributions of the share of each component. Fig. S3. Relationships between DAS28-ESR and age or joint findings before and after correction. Table S1. Patient characteristics (DMARDs free). Table S2. Patient characteristics (csDMARDs). Table S3. Patient characteristics (TNFi). Table S4. Patient characteristics (IL-6i). Table S5. Patient characteristics (CTLA-4-Ig). Table S6. Patient characteristics (JAKi). Table S7. Cliff’s delta for sex difference in disease activity indices and their components by treatment type. Table S8. PERMANOVA for sex difference in the share of components. Table S9. Adjustment for patient-related factors with GLM and QR in available-case analysis. Table S10. Adjustment for patient-related factors with GLM and QR using stacked dataset imputed by chained equations.
## References
1. Stoffer MA, Schoels MM, Smolen JS. **Evidence for treating rheumatoid arthritis to target: results of a systematic literature search update**. *Ann Rheum Dis* (2016.0) **75** 16-22. PMID: 25990290
2. Smolen JS, Aletaha D, Bijlsma JWJ. **Treating rheumatoid arthritis to target: recommendations of an international task force**. *Ann Rheum Dis* (2010.0) **69** 631-637. PMID: 20215140
3. Aletaha D, Smolen JS. **Diagnosis and management of rheumatoid arthritis: a review**. *JAMA* (2018.0) **320** 1360-1372. PMID: 30285183
4. Prevoo MLL, van'T Hof MA, Kuper HH. **Modified disease activity scores that include twenty-eight-joint counts development and validation in a prospective longitudinal study of patients with rheumatoid arthritis**. *Arthritis Rheum* (1995.0) **38** 44-48. PMID: 7818570
5. Fransen J, Welsing PMJ, de Keijzer RMH. **Disease activity scores using C-reactive protein: CRP may replace ESR in the assessment of RA disease activity**. *Ann Rheum Dis* (2004.0) **62** 151
6. Smolen JS, Breedveld FC, Schiff MH. **A simplified disease activity index for rheumatoid arthritis for use in clinical practice**. *Rheumatology (Oxford)* (2003.0) **42** 244-257. PMID: 12595618
7. Aletaha D, Nell VPK, Stamm T. **Acute phase reactants add little to composite disease activity indices for rheumatoid arthritis: validation of a clinical activity score**. *Arthritis Res Ther* (2005.0) **7** 796-806
8. England BR, Tiong BK, Bergman MJ. **2019 update of the American College of Rheumatology recommended rheumatoid arthritis disease activity measures**. *Arthritis Care Res (Hoboken)* (2019.0) **71** 1540-1555. PMID: 31709779
9. Rintelen B, Haindl PM, Maktari A. **SDAI/CDAI levels in rheumatoid arthritis patients are highly dependent on patient’s pain perception and gender**. *Scand J Rheumatol* (2008.0) **37** 410-413. PMID: 18759162
10. Radovits BJ, Fransen J, van Riel PLCM. **Influence of age and gender on the 28-joint Disease Activity Score (DAS28) in rheumatoid arthritis**. *Ann Rheum Dis* (2008.0) **67** 1127-1131. PMID: 17965425
11. Tengstrand B, Ahlmén M, Hafström I. **The influence of sex on rheumatoid arthritis: a prospective study of onset and outcome after 2 years**. *J Rheumatol* (2004.0) **31** 214-222. PMID: 14760787
12. Sokka T, Toloza S, Cutolo M. **Women, men, and rheumatoid arthritis: analyses of disease activity, disease characteristics, and treatments in the QUEST-RA study**. *Arthritis Res Ther* (2009.0) **11** R7. PMID: 19144159
13. Leeb BF, Haindl PM, Maktari A. **Disease activity score-28 values differ considerably depending on patient’s pain perception and sex**. *J Rheumatol* (2007.0) **34** 2382-2387. PMID: 17985407
14. Ahlmén M, Svensson B, Albertsson K. **Influence of gender on assessments of disease activity and function in early rheumatoid arthritis in relation to radiographic joint damage**. *Ann Rheum Dis* (2010.0) **69** 230-233. PMID: 19158113
15. Shin S, Park EH, Kang EH. **Sex differences in clinical characteristics and their influence on clinical outcomes in an observational cohort of patients with rheumatoid arthritis**. *Joint Bone Spine* (2021.0) **88** 105124. PMID: 33346105
16. Iikuni N, Sato E, Hoshi M. **The influence of sex on patients with rheumatoid arthritis in a large observational cohort**. *J Rheumatol* (2009.0) **36** 508-511. PMID: 19208610
17. Matsui T, Kuga Y, Nishino J. **Comparison of composite disease activity indices for rheumatoid arthritis**. *Mod Rheumatol* (2011.0) **21** 134-143. PMID: 20972600
18. Jawaheer D, Olsen J, Lahiff M. **Gender, body mass index and rheumatoid arthritis disease activity: results from the QUEST-RA study**. *Clin Exp Rheumatol* (2010.0) **28** 454-461. PMID: 20810033
19. George MD, Giles JT, Katz PP. **Impact of obesity and adiposity on inflammatory markers in patients with rheumatoid arthritis**. *Arthritis Care Res (Hoboken)* (2017.0) **69** 1789-1798. PMID: 28393498
20. Ranzolin A, Brenol JCT, Bredemeier M. **Association of concomitant fibromyalgia with worse disease activity score in 28 joints, health assessment questionnaire, and short form 36 scores in patients with rheumatoid arthritis**. *Arthritis Care Res (Hoboken)* (2009.0) **61** 794-800
21. Ruiz-Medrano E, Espinosa-Ortega HF, Arce-Salinas CA. **The effect of concomitant hand osteoarthritis on pain and disease activity in patients with rheumatoid arthritis**. *Clin Rheumatol* (2019.0) **38** 2709-2716. PMID: 31076944
22. Favalli EG, Biggioggero M, Crotti C. **Sex and management of rheumatoid arthritis**. *Clin Rev Allergy Immunol* (2019.0) **56** 333-345. PMID: 29372537
23. Siemons L, Ten Klooster PM, Vonkeman HE. **How age and sex affect the erythrocyte sedimentation rate and C-reactive protein in early rheumatoid arthritis**. *BMC Musculoskelet Disord* (2014.0) **15** 1-7. PMID: 24387196
24. Hamann PDH, Shaddick G, Hyrich K. **Gender stratified adjustment of the DAS28-CRP improves inter-score agreement with the DAS28-ESR in rheumatoid arthritis**. *Rheumatology (Oxford)* (2019.0) **58** 831-835
25. Castrejón I, Ortiz AM, García-Vicuña R. **Are the C-reactive protein values and erythrocyte sedimentation rate equivalent when estimating the 28-joint disease activity score in rheumatoid arthritis?**. *Clin Exp Rheumatol* (2008.0) **26** 769-775. PMID: 19032807
26. Matsui T, Kuga Y, Kaneko A. **Disease Activity Score 28 (DAS28) using C-reactive protein underestimates disease activity and overestimates EULAR response criteria compared with DAS28 using erythrocyte sedimentation rate in a large observational cohort of rheumatoid arthritis patients**. *Ann Rheum Dis* (2007.0) **66** 1221-1226. PMID: 17369281
27. Siemons L, Vonkeman HE, Ten Klooster PM. **Interchangeability of 28-joint disease activity scores using the erythrocyte sedimentation rate or the C-reactive protein as inflammatory marker**. *Clin Rheumatol* (2014.0) **33** 783-789. PMID: 24562719
28. Saeki Y, Matsui T, Saisho K. **Current treatments of rheumatoid arthritis: from the “NinJa” registry**. *Expert Rev Clin Immunol* (2012.0) **8** 455-465. PMID: 22882220
29. Arnett FC, Edworthy SM, Bloch DA. **The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis**. *Arthritis Rheum* (1988.0) **31** 315-324. PMID: 3358796
30. Aletaha D, Neogi T, Silman AJ. **2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative**. *Arthritis Rheum* (2010.0) **62** 2569-2581. PMID: 20872595
31. Funovits J, Aletaha D, Bykerk V. **The 2010 American College of Rheumatology/European league against rheumatism classification criteria for rheumatoid arthritis: methodological report phase I**. *Ann Rheum Dis* (2010.0) **69** 1589-1595. PMID: 20699242
32. Neogi T, Aletaha D, Silman AJ. **The 2010 American College of Rheumatology/European League Against Rheumatism classification criteria for rheumatoid arthritis: phase 2 methodological report**. *Arthritis Rheum* (2010.0) **62** 2582-2591. PMID: 20872596
33. 33.Romano J, Kromrey JD, Coraggio J, et al.: Exploring methods for evaluating group differences on the NSSE and other surveys: are the t-test and Cohen’s d indices the most appropriate choices? Annu Meet South Assoc Institutional Res 2006:1–51.
34. van Buuren S. *Flexible imputation of missing data* (2018.0)
35. Miller A, Green M. **Simple rule for calculating normal erythrocyte sedimentation rate**. *Br Med J (Clin Res Ed)* (1983.0) **286(6361)** 266
36. Mäkinen H, Hannonen P, Sokka T. **Sex: A major predictor of remission as measured by 28-joint Disease Activity Score (DAS28) in early rheumatoid arthritis?**. *Ann Rheum Dis* (2008.0) **67** 1052-1053. PMID: 18556449
37. Ward MM, Guthrie LC, Dasgupta A. **Direct and indirect determinants of the patient global assessment in rheumatoid arthritis: differences by level of disease activity**. *Arthritis Care Res (Hoboken)* (2017.0) **69** 323-329. PMID: 27273981
38. Fillingim RB, King CD, Ribeiro-Dasilva MC. **Sex, gender, and pain: a review of recent clinical and experimental findings**. *J Pain* (2009.0) **10** 447-485. PMID: 19411059
39. Nasser SA, Afify EA. **Sex differences in pain and opioid mediated antinociception: modulatory role of gonadal hormones**. *Life Sci* (2019.0) **237** 116926. PMID: 31614148
40. Wagemakers SH, van der Velden JM, Gerlich AS. **A systematic review of devices and techniques that objectively measure patients’ pain**. *Pain Physician* (2019.0) **22** 1-13
41. Forslind K, Hafström I, Ahlmén M. **Sex: A major predictor of remission in early rheumatoid arthritis?**. *Ann Rheum Dis* (2007.0) **66** 46-52. PMID: 17158139
42. Yu C, Jin S, Wang Y. **Remission rate and predictors of remission in patients with rheumatoid arthritis under treat-to-target strategy in real-world studies: a systematic review and meta-analysis**. *Clin Rheumatol* (2019.0) **38** 727-738. PMID: 30341703
43. Maynard C, Mikuls TR, Cannon GW. **Sex differences in the achievement of remission and low disease activity in rheumatoid arthritis**. *Arthritis Care Res (Hoboken)* (2020.0) **72** 326-333. PMID: 30875461
44. Nikiphorou E, Norton SJ, Carpenter L. **Remission vs low disease activity: function, quality of life and structural outcomes in the Early Rheumatoid Arthritis Study and Network**. *Rheumatology (Oxford)* (2020.0) **59** 1272-1280. PMID: 31580448
45. Nikiphorou E, Radner H, Chatzidionysiou K. **Patient global assessment in measuring disease activity in rheumatoid arthritis: a review of the literature**. *Arthritis Res Ther* (2016.0) **18** 251. PMID: 27793211
46. Ferreira RJO, Duarte C, Ndosi M. **Suppressing inflammation in rheumatoid arthritis: Does patient global assessment blur the target? A practice-based call for a paradigm change**. *Arthritis Care Res (Hoboken)* (2018.0) **70** 369-378. PMID: 28544615
47. Ferreira RJO, Ndosi M, de Wit M. **Dual target strategy: a proposal to mitigate the risk of overtreatment and enhance patient satisfaction in rheumatoid arthritis**. *Ann Rheum Dis* (2019.0) **78** e109. PMID: 30127056
48. Studenic P, Felson D, de Wit M. **Testing different thresholds for patient global assessment in defining remission for rheumatoid arthritis: Are the current ACR/EULAR Boolean criteria optimal?**. *Ann Rheum Dis* (2020.0) **79** 445-452. PMID: 32024651
49. Doumen M, de Cock D, Pazmino S. **Psychosocial burden predicts sustained remission in early rheumatoid arthritis: unraveling the complex interplay of wellbeing and disease activity**. *Arthritis Care Res (Hoboken)* (2021.0). DOI: 10.1002/acr.24847
50. Pazmino S, Lovik A, Boonen A. **New indicator for discordance between patient-reported and traditional disease activity outcomes in patients with early rheumatoid arthritis**. *Rheumatology (Oxford)* (2022.0). DOI: 10.1093/rheumatology/keac213
|
---
title: Evaluating self-medication practices in Ethiopia
authors:
- Yabibal Berie Tadesse
- Abebe Tarekegn Kassaw
- Eyayaw Ashete Belachew
journal: Journal of Pharmaceutical Policy and Practice
year: 2023
pmcid: PMC10029313
doi: 10.1186/s40545-023-00553-0
license: CC BY 4.0
---
# Evaluating self-medication practices in Ethiopia
## Abstract
### Background
Self-medication with antibiotics has become an important factor driving antibiotic resistance and it masks the signs and symptoms of the underlying disease and hence complicates the problem, increasing drug resistance and delaying diagnosis. This study aimed to assess the extent of self-medication practice with antibiotics and its associated factors among adult patients attending outpatient departments (OPD) at selected public Hospitals, in Addis Ababa, Ethiopia.
### Methods
Facility-based cross-sectional study was employed. A systematic random sampling technique was used to include the study participants. Self-administered with structured questionnaires were applied among patients who visited outpatient departments at selected public Hospitals, in Addis Ababa. Data were entered into Epi-data version 4.6 and analyzed using SPSS version 26. Descriptive statistics such as frequencies and percentages were used for the present categorical data. The data are presented in pie charts, tables, and bar graphs. Furthermore, bivariable and multivariable binary logistic regression analyses were used to identify significant associations. Statistical significance was declared at p value < 0.05.
### Results
Out of 421 respondents interviewed, 403 ($95.7\%$) participants completed questionnaires. Among the respondents, $71\%$ had generally practiced self-medication. Among these, $48.3\%$ had self-medication with antibiotics, while $51.7\%$ had used other drugs. The most commonly cited indication for self-medication practice was abdominal pain $44.9\%$, followed by Sore throat $21\%$ commonly used antibiotics are amoxicillin ($57\%$), ciprofloxacin ($13\%$), amoxicillin/clavulanic ($10\%$), erythromycin ($8\%$), cotrimoxazole ($7\%$), and doxycycline ($5\%$).
### Conclusions
Self-medication with antibiotics was common among the study participants. The prevalence of general self-medication was $71\%$, whereas that of self-medication with antibiotics was $48.3\%$. *In* general, the potentially dangerous effects of SMP seem to be underestimated by patients with OPD patients.
## Introduction
Self-medication has traditionally been characterized as using medications, or home remedies on one’s own initiative or at the suggestion of another person without first contacting a medical professional [1]. The World Health Organization (WHO), defined self-medication as the selection and use of drugs to address self-identified diseases or symptoms. Self-medication is when a person obtains and consumes a medication without seeking medical assistance, whether for diagnosis, prescription, or monitoring of one’s own therapy or medication [2].
It usually involves over-the-counter (OTC) medications, but it can also include prescription-only medications (POM, purchasing drugs by reusing or resubmitting a previous prescription, taking medications on the advice of a relative or other, or eating leftover medications already on hand at home [3]. The FDA [2006] characterizes OTCs as a sedate item promoted for use by the buyer without the intercession of a well-being care proficient in arranging to get the item. With respect to the classification of drugs, it appears that individuals do not separate between medication and OTC drugs [4].
Prescription products are medications that require a doctor’s prescription [5]. Several studies conducted in indicated that self-medication with antibiotics is quite common, varies by community and social determinants of health, and is usually accompanied by the use of unsuitable drugs [6]. In developed countries, the use of antibiotics without prescription is the second highest prevalent next to antipain [7]. Antibiotics are not available over the counter, and they require a prescription before being dispensed. Over the counter products are also available at supermarkets and other community pharmacies in various countries, including Ethiopia [8].
Inappropriate self-medication waste resources raises the risk of drug resistance and leads to major health issues such adverse drug responses, treatment failure, prescription misuse, and drug addiction [9]. Despite this, self-medication may save money on health care and time spent waiting to visit a doctor for mild diseases.
Self-medication has several detrimental effects on one’s health. It may result in incorrect self-diagnosis and a delay in receiving urgent medical care. Moreover, it may lead to ineffective dosing, polypharmacy, and hazardous drug interactions [10]. This can lead to noncompliance with a drug regimen that can result in serious outcomes such as adverse drug reactions and reduction in the quality of treatment [10]. Moreover, currently, there is a worldwide concern about the emergence of antibiotic-resistant strains of micro-organisms, which might have been highly augmented by self-medication [11].
Self-medication has been reported to be on the rise around the world and has become a public health concern [9]. People in poor countries are self-medicating with not only non-prescription but also prescription medications without supervision. Although the WHO has stressed the importance of properly teaching and controlling self-medication, its use is nevertheless widespread [12].
A number of researches on various elements of self-medication have been undertaken internationally, and the prevalence of self-medication among adult outpatients has been found to be high [13–18]. The prevalence of self-medication in Greece was $77.9\%$ in [19], $98\%$ in Palestine [20], $71\%$ in India [21], and $76\%$ in Pakistan [21]. The rates are similar in Africa: *It is* $99.4\%$ in Nigeria [22], $56\%$ in Malawi [23], $53.5\%$ Kenya [24], $75.7\%$ in Uganda [25] and $50\%$ in Ethiopia [8]. Accordingly, individuals practiced self-medication for different purposes. Studies have reported that headache, fever, cough, gastrointestinal diseases, respiratory tract infections, maternal/menstrual, eye diseases, skin diseases, injury, and sexually transmitted diseases were common indications for self-medication practice [26].
A few studies have been undertaken in Ethiopia to investigate the usage of self-medication among the public and students, including medical students [27–29], and there are indications on the misuse of antibiotics by patients, even by health professionals [30]. However, there is no study conducted on self-medication practice with antibiotics of adult outpatients in Ethiopia. The findings of this study will fill the research and knowledge gap. In addition, this study will generate information that may be useful in policy development and review of policies on licensing of drugs. Furthermore, it can be used as a stepping stone for health professionals if there is any possibility of intervention. Finally, this research can be used as a base for other health professionals including pharmacists in understanding the situation of the case and extending their intervention or work to different institutions also the findings potentially assisting in the development of appropriate regulatory and administrative solutions in Ethiopian hospitals. As a result, this study aimed to assess self-medication practice with antibiotic-among adult patients in OPD at selected public hospitals in Addis Ababa, Ethiopia.
## Study design, setting and period
An institution-based cross-sectional study was conducted from February 2022 and March 2022, in selected public hospitals, Addis Ababa, Ethiopia. Addis *Ababa is* the capital city of Ethiopia, which contains 13 government hospitals (5 federals, 6 under Addis Ababa health bureau, one owned by the police force, and one owned by armed force) distributed throughout ten sub-cities. All Hospitals provide different OPD services. Four hospitals had been selected using simple random sampling by lottery method from the list of thirteen hospitals [31]. Tikur Anbessa specialized Hospital, St. Paul Hospital Millennium Medical College, Minillik II Hospital and St. Peter specialized hospital. Tikur Anbessa Specialized hospital and St. Paul’s Hospital both are the largest referral and teaching hospitals in Ethiopia and are operated under the ministry of health. St. Peter specialized *Hospital is* the other referral and teaching hospital among those operated under the federal ministry of health. However, Minillik II *Hospital is* among the six governmental referral hospitals that are managed under the Addis Ababa Administrative Health Office. Patients who attended outpatient clinics in the hospital were expected to provide information in respect to factors associated with SMA. The above selected hospitals outpatient clinics had a high volume of patients and this enabled the researcher to collect data from the subjects from the proposed sample size within the limited time of the survey. Patients visiting the hospital due to their illnesses may have had prior knowledge of SMA, unlike a person in the community who may not have experienced an illness.
## Study participants and eligibility criteria
All adult patients seeking treatment in OPD in the selected public Hospital during the study period and fulfilled the inclusion criteria were included. addition, patients attending out patients in selected PHs who were above 18 years of age, and those who had taken informed consent were included, whereas, participants who were admitted in the ward, unconscious, unable to speak and hear, critically ill patients and who were not willing to participate in the study were excluded in this study.
## Sample size determination and sampling techniques
Sample size (n) had been calculated on the basis of a single population proportion formula assuming that the prevalence of self-medication with antibiotics is taken from a previous similar study in Kenya to calculate the sample size 0.476 is taken [32]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = \frac{{z_{ \propto /2}^2\;p(1 - p)}}{{d^2}}$$\end{document}n=z∝/22p(1-p)d2 The assumptions used are: z value of 1.96 at $95\%$ confidence interval (CI) and margin of error (d) is $5\%$, non-response rate $10\%$: d marginal error of $5\%$; p proportion; n sample size.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=\frac{{(1.96)}^{2}*0.476\left(1-0.476\right)}{{(0.05)}^{2}}=383.$$\end{document}n=(1.96)2∗0.4761-0.476(0.05)2=383.
By adding $10\%$ for incomplete and non-responses, the total sample size required for this study was found to be 421. From 13 governmental Hospitals in Addis Ababa, Tikur Anbessa, St. Paul, Minillik II, St. Peter hospitals and St. Paul’s Hospital Millennium Medical College were selected randomly using a lottery method. To select 421 participants from adult medical and surgical OPD from a total of four selected PHs, first, all selected hospitals were listed down with their respective number of OPD patients per month. A stratified sampling method was performed with the strata being outpatient department, surgical, and medical outpatient clinics. Data were taken from each hospital with of monthly OPD patients report, and then, the number of OPD patients in each hospital was proportionally allocated for sample size, and then, finally the study participants for each hospital were selected and interviewed with a systematic random sampling method in every kth interval of each respective hospital until the required sample size was achieved (Fig. 1).Fig. 1Schematic representation of the sampling procedure to select study participants from Addis Ababa selected public hospitals, 2022 From the past 3-monthly report from November, 2021 to January 2022 of each hospital indicate that the average number of adult OPD cases at Tikur Anbessa specialized hospital, Minillik II, St. Peter hospitals and St. Paul’s Hospital Millennium Medical College, were 1200, 950, 862 and 1450, respectively.
The total sample size [421] was allocated proportionally for the four public hospitals based on the number of OPD patients seeking treatment in a month of each hospital was: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{n}_{j}}=\frac{{n}\times{N}_{j}}{N}$$\end{document}nj=n×NjN and interval size; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{k}_{j}}=\frac{Nj}{n}$$\end{document}kj=Njn where nj was the sample size of the jth hospital; Nj was population size of the jth hospital; n = nTASH + nSPMMC + n Minillik H + NSPSH was the total sample size [421];
N = NTASH + NSPMMC + NMinelikH + NSPSH was the total population size of hospitals = 1200 + 1450 + 950 + 862 = 4462; nTASH = 421 * $\frac{1200}{4462}$ = 113; $k = 1200$/113 = 11, SPMMC = 421 * $\frac{1450}{4462}$ = 137; $k = 1450$/137 = 11, Minillik II Hospital = 421 * $\frac{950}{4462}$ = 90; $k = 950$/90 = 11, SPSH = 421 * $\frac{862}{4462}$ = 81; $k = 862$/81 = 11.
Therefore, the initial sample was selected using the lottery metho; then, every 11th participant was selected until the calculated sample size was attained within the data collection period’s time frame.
## Variables of the study
Socio-demographic factors were age, sex, education, marital status, religion, and occupation. Behavioral and social factors such as: mild illness, prior experience, emergency use, stressful conditions, chronic illness, advice from others, laws controlling, policy factors, insurance, and knowledge on risks of use. Health facility factors are the availability of antibiotics, diagnosis, healthcare costs, ease of access, prescription, queues, and save time and attitudes, such as attitudes, distance, queues, policy factors, and laws. The dependent variable was self-medication practice with antibiotics. This was determined as the report of taking drugs without a prescription among adult outpatients.
## Data collection and procedures
The questionnaire was first prepared in the English language and translated into Amharic by a language expert for interviews. To check the accuracy and its consistency, the questionnaire was pretested on 21 participants ($5\%$ of the sample size) in one hospital in Addis Ababa, and this facility was excluded from the actual study before the start of the actual data collection period. Data were collected using structured and pretested questionnaires. Three BSc pharmacists participated in the data collection after 1 day of training were given on the objective, the relevance of the study, the confidentiality of information, respondent’s right, and informed consent. Frequent checks were made on the data collection process to ensure the completeness of principal investigators and supervisor.
## Data quality assurance
Data collectors were trained intensively by the principal investigator on the contents of the questionnaire, data collection methods, and ethical concerns. The filled questionnaire was checked daily for completeness by the principal investigator for uniformity and understandability of the checklist after which modification for its appropriateness and suitability was performed. Data collectors had trained on strict use of study criteria, an explanation of study objective, getting verbal consent from study participants, uniform interpretation of questions, and the collected data confidentiality.
## Data processing and analysis
After checking the collected data for its completeness and accuracy, codes were given to the questionnaire; then, the data were entered using *Epi data* 4.6 statistical software and analyzed using the SPSS version 26 statistical package. Binary logistic regression was used to determine the association between the explanatory and outcome variables, and multivariable logistic regressions were used to determine the association between dependent and independent variables, P value < 0.05 was considered as statistically significant.
## Operational definition
Self-medication: getting and using conventional medications for disease diagnosis, treatment, or prevention without a doctor’s prescription.
Antibiotics: this is a drug used in the treatment and prevention of bacterial infection.
Adverse reaction: this is an unwanted effect caused by administrating a drug.
Outpatient: patient who attends for treatment in an outpatient clinic without staying there overnight.
Outpatient department: is part of a hospital designed for treating outpatients for whom they have health problems but do not require a bed or to be admitted for overnight care.
Over the counter/non-prescribed drugs: are those drugs that can be legally purchased from a drug retail outlet without having a prescription from a licensed healthcare provider.
Self-medication practice: a person is said to practice self-medication if he/she self-medicated at least once [32].
## Ethical consideration
Ethical clearance was obtained from the Institutional Review Board RVU college of Health science and a support letter obtained from Addis Ababa health bureau administration for each Hospital. The objective and importance of the study was explained to the study participants; then, data were collected only after full informed verbal and written consent was obtained. The confidentiality of the information was maintained by excluding the participants’ name in the interview questionnaire.
## Results
Among 421 respondents approached for the study, 403 ($95.7\%$) were included in the final analysis. Around two-third ($69\%$) the respondents were females. The mean (± SD) age of the participants was 33 (± 21) years. More than one-third of the ($36.7\%$) participants was between 25 and 34 years. The majority of the respondents ($62.5\%$) were not employed, while those who were formally employed were $24.8\%$. Around three-fourth ($73\%$) of the respondents had no medical insurance scheme (Table 1).Table 1Socio-demographic characteristics of participants who used drugs for self-medication ($$n = 403$$)CharacteristicsFrequencyPercentageAge in years 18–245513.6 25–3414836.7 35–448721.6 > 4511328.1Sex Male12531.0 Female27869.0Marital status Married27067.0 Never married7719.1 Divorced153.7 Widowed4110.2Occupation Unemployed25262.5 Employed Formally employed10024.8 Business5112.7Religion Orthodox15939.5 Muslim7117.5 Protestants17343Healthcare insurance use Yes10927 No29473
## Education levels of the respondents
Education levels were categorized into four: those who had not gone to school, primary, secondary, and college/university. The respondents who had College/University were $6\%$ [24], secondary education was $51.1\%$ [206], those with primary education were $29.1\%$ [117], and those who had not gone to school were $13.8\%$ [56] (Fig. 2).Fig. 2Education levels of the respondents
## General prevalence of self-medication
The participants were asked whether they had ever taken any drug without prescription. Among the total respondents, majority of them ($71\%$), the respondents had generally practiced self-medication (Fig. 3).Fig. 3General prevalence of self-medication
## Prevalence of self-medication with antibiotics
The respondents were required to determine whether they had ever taken any antibiotic without prescription. Around half of the participants ($48.3\%$) had used antibiotics, while $51.7\%$ had used other drugs (Table 2).Table 2Prevalence of self-medication useSelf-medication with antibioticsResponsesPercentageYes13848.3No14851.7Total286100
## Number of times respondents self-medicated
More than half of the respondents, 59 ($42.8\%$) had used antibiotics above five times in the past year, 34 ($24.6\%$) three times, 26 ($18.8\%$) twice, 12 ($8.7\%$) four times, and 7 ($5.1\%$) once (Fig. 4).Fig. 4Number of times respondents self-medicated
## Factors associated with self-medication
The age of the respondents was categorized in groups ranging from 18 to 24 years, 25 to 34 years, 35 to 44 years, and those above 45 years. Age and medical insurance scheme were significantly associated with self-medication ($P \leq 0.05$) in the Chi-square test (Table 3).Table 3Factors associated with self-medicationSelf-medicatedNot self-medicatedNΧ2P valueAge of participants 18–2423 ($41.8\%$)32 ($58.2\%$)558.697P = 0.0004 25–3446 (31.1)102 ($68.9\%$)148 35–4432 ($36.8\%$)55 ($63.2\%$)87 > 4537 ($32.7\%$)76 ($67.3\%$)113 Total138265403Sex Male51 ($40.8\%$)74 ($59.2\%$)1250.79460.4782 Female87 ($31.3\%$)191 ($68.7\%$)278 Total138265403Marital status Married88 ($32.6\%$)182 ($67.4\%$)2700.63800.3280 Single Never married29 ($37.7\%$)48 ($62.3\%$)77 Divorced7 ($46.7\%$)8 ($53.3\%$)15 Widowed14 ($34.1\%$)27 ($65.9\%$)41 Total138265403Educational status Not gone to school14 ($25\%$)42 ($75\%$)563.43520.5674 Primary34 ($29.1\%$)83 ($70.9\%$)117 Secondary79 ($38.3\%$)127 ($61.7\%$)206 College/university11 ($45.8\%$)13 ($54.2\%$)24 Total138265403Occupation Unemployed78 ($31\%$)174 ($69.0\%$)2524.27600.643 Employed Formally employed46 ($46\%$)54 ($54\%$)100 Business person14 ($27.5\%$)37 ($72.5\%$)51 Subtotal60 ($39.7\%$)91 ($60.3\%$)151Total138265403Members of health insurance Yes21 ($19.3\%$)88 ($80.7\%$)1096.0240.0176 No117 ($39.8\%$)177 ($60.2\%$)294 Total138265403
## Independent predictors of self-medication
In the multivariable regression model, participants who had College and university students were 1.6 times likely to practice self-medication compared with those who never went to school or illiterate [AOR = 1.65, $95\%$ CI (1.3–2.4), $$P \leq 0.004$$]. The lack of medical insurance was also significantly associated with self-medication with antibiotics [AOR = 1.632, $95\%$ CI (1.21–2.63), $$P \leq 0.033$$] (Table 4).Table 4Independent predictors of self-medicationsVariableSMPNone SMPOR $95\%$ CIP valueAge 18–2423 ($41.8\%$)32 ($58.2\%$)1 25–3446 ($31.1\%$)102 ($68.9\%$)0.453 (0.51–0.67)0.801 35–4432 ($36.8\%$)55 ($63.2\%$)0.722 (0.731–0.82)0.057 > 4537 ($32.7\%$)76 ($67.3\%$)0.576 (0.60–0.84)0.831Sex Male51 ($40.8\%$)74 ($67.3\%$)0.955 (0.874–0.973)0.892 Female879 ($31.3\%$)191 ($68.7\%$)0.836 (0.76–0.98)0.61Education Illiterate14 ($25\%$)42 ($75\%$)1 Primary34 ($29.1\%$)83 ($70.9\%$)0.986 (0.9–1.97)0.236 Secondary79 ($38.3\%$)127 ($61.7\%$)1.241 (0.93–2.5)0.438 College and above11 ($45.8\%$)13 ($54.2\%$)1.65 (1.3–2.4)0.004Marital status Never married29 ($37.7\%$)48 ($62.3\%$)1 Married88 ($32.6\%$)182 ($67.4\%$)0.687 (0.63–1.84)0.39 Divorced7 ($46.7\%$)8 ($53.3\%$)1.614 (1.2–2.42)0.057 Widowed14 ($34.1\%$)27 ($65.9\%$)0.642 (0.36–1.31)0.634Occupation Unemployed78 ($31\%$)174 ($69.0\%$)0.984 (0.656–1.42)0.13 Formally employed46 ($46\%$)54 ($54\%$)1.129 (0.83–1.76)0.73 Business14 ($27.5\%$)37 ($72.5\%$)0.63 (0.33–1.240.603Medical insurance Yes21 ($19.3\%$)88 ($80.7\%$)1 No117 ($39.8\%$)177 ($60.2\%$)1.632 (1.21–2.63)0.033
## Reasons for indulging in self-medication
The majority of those who practiced self-medication with antibiotics ($53.6\%$) gave reasons for the practice as to reduce medical cost, $26.2\%$ said that there are long delays in health facility, while $11.6\%$ did so because of a busy day’s program (Table 5).Table 5Reasons for indulging in self-medicationReasons for SMPFrequencyPercentageLack of clinicians64.3Busy day program1611.6Cost cutting7453.6Long delays in health facility3626.2Previous experience of medical treatment of the same symptoms64.3Total138100
## Indication for SMP
Various respondents gave their complaints for taking antibiotics as follows: abdominal pain $44.9\%$, sore throat $21\%$, cough $16.7\%$, diarrhea and vomiting $8.0\%$, toothache $6.5\%$, and wound $2.9\%$ (Fig. 5).Fig. 5Complaints for SMP
## Patients who were advised on self-medication with antibiotics
Few participants ($21.1\%$) had been advised to take medication without prescription, while $78.9\%$ were advised but did not self-medicate. Advice had no significant association with self-medication ($P \leq 0.05$) (Table 6).Table 6Respondents advised on SMAdvisedSMNot SMNΧ2P valueYes23 ($21.1\%$)86 ($78.9\%$)1090.0280.7882No115 ($39.1\%$)179 ($60.9\%$)294Total138265403
## Respondents’ sources of advice for self-medication
Those advised to take an antibiotic the majority of the respondents ($47.4\%$) said their source of advice was from a colleague, $28\%$ from health workers, $12.3\%$ from a relative (Table 7).Table 7Sources of advice for SMPSources of adviceFrequencyPercentageColleague2747.4Health worker1628.0Relatives712.3Neighbour712.3Advertisement0–Total57100
## Commonly used antibiotics in SMA
The respondents were asked to mention the antibiotics that they had used without prescription. A list of antibiotics was provided to help the patients recall. The commonly used antibiotics are amoxicillin ($57\%$), ciprofloxacin ($13\%$), amoxicillin/clavulanic ($10\%$), erythromycin ($8\%$), cotrimoxazole ($7\%$), and doxycycline ($5\%$) (Fig. 6).Fig. 6Antibiotics used in self-medication
## Sources of antibiotics
From Table 8, the majority of respondents who practiced self-medication with antibiotics got the drugs from community pharmacy ($84.8\%$), others got drugs from health workers ($8.7\%$), while $6.5\%$ got the drugs from friends (Table 8).Table 8Sources of antibioticsSourcesFrequencyPercentageCommunity pharmacy11784.8Health worker128.7Given by a friend96.5Shops0–Total138100
## How respondents knew the dosage
As shown in the table above, it shows that $81.9\%$ had enquired the dosage from the seller, while $15.9\%$ used a previous prescription to know the dosage of the drug for self-medication (Table 9).Table 9How respondents established the dosageResponsesFrequencyPercentageEnquired from the seller11381.9Used a previous prescription2215.9Informed by a friend32.2Read on the packaging0–Total138100
## Discussion
Patients who attended outpatient clinics were surveyed over the course of a month and a half. The prevalence of SMP was found to be $48.3\%$ in the survey. Amoxicillin was the most widely used antibiotic for self-medication. Antibiotics were primarily obtained from local or community pharmacies. The primary motivation for using SMA was to save money. This study focused solely on antibiotic self-medication. Other drug classifications were excluded from the poll. The findings were still valid despite this constraint. To decrease the recall bias, a list of antibiotics was employed.
The majority of the respondents in this study were aged between 25 and 34 years at $31.1\%$. This was comparable with a study by Charles Kiragu Ngigi in Kenya, which had $27.7\%$ of the respondents with the same age who practiced self-medication with antibiotics [32] and nearly also comparable with the study conducted in India, which found that $39\%$ of respondents of the same age used antibiotics for self-medication [33]. According to this study, females account for $68.0\%$ of practiced self-medication. In contrast [34], found that $44\%$ of respondents in research in Saudi Arabia were female.
Self-medication with antibiotics was found to be $48.3\%$, which is lower than prior studies in Northern Nigeria, which found $56.8\%$ and $50.3\%$, respectively [35, 36]. Self-medication with antibiotics was reported by $41.8\%$ of respondents aged 18–24, which is comparable to data from [37] in Nigeria, who found $44\%$ of respondents in the same age range. SMA and education had no meaningful relationship. Around half (46.7)% of those who self-medicate with antibiotics are divorced and $38.3\%$ of those with secondary education. In this study, respondents who had not “gone to school” accounted for $25\%$ of the total, implying that educated people accounted for $75\%$ of the total, which is close to a study by Widayati et al. [ 38] that linked self-medication to a high percentage of education ($78\%$). Other similar study reported in Nigeria, with just $14\%$ of uneducated individuals self-medicating [39]. This link between self-medication and education could be due to the ease with which information can be obtained from various sources, including the internet.
Those who practiced self-medication with antibiotics had a smaller number of unemployed people ($31\%$). This contradicts a study conducted by Askarian et al. [ 40], which revealed that $7.4\%$ of the population was unemployed. SMA was practiced by $39.8\%$ of persons who did not have health insurance, which was lower than the 46.3 percent reported by Askarian et al. [ 40] in Southern Iran. Self-medication with antibiotics is significantly connected with a lack of medical insurance.
Amoxicillin was the most often used antibiotic, with 79 ($57\%$) of respondents using it. This was supported by studies conducted by Donkor et al. [ 41] in Accra, Ghana, and [42] in the United Arab Emirates, where amoxicillin had a high prevalence rate of $46\%$ in both countries. According to a study conducted in Europe by Ali et al. [ 43], Greece has one of the highest outpatient antibiotic usage rates in Europe, with cephalosporins and macrolides being the most commonly prescribed antibiotics. The argument for the regular use of Amoxicillin was reinforced by the drug’s low cost around the world and its widespread prescription by healthcare practitioners, suggesting that it is well-known to the general people.
The antibiotics were obtained by a majority of responders ($84.8\%$) from community pharmacy. This was contracted with a study conducted in Sudan, which found that $68.8\%$ of the medications were obtained from a community pharmacy [44]. It is similar with other studies in India, with $79.5\%$ of respondents getting their medications from pharmacies [45]. Other investigations in Palestine [20], Egypt [32], and other similar results [46]. The majority of respondents in all of these studies got their antibiotics from community pharmacies rather than through friends, health workers, or stores.
Antibiotics can be obtained from various sources: they are legally available over the counter, antibiotics originally prescribed by physicians can be saved and used without medical consultation, antibiotics can be obtained from friends or relatives, and they can be obtained via different sources.
## Limitations of the study
The research was done in a limited duration of 1 month. Patients who continued to practice the self-medication at home and did not visit the hospital during the period of study were not sampled. The identification of the actual antibiotic taken may not have been accurately recalled. The study was restricted to the practice of self-medication among adult patients, and patients below 18 years were excluded from the study. A list of antibiotics was provided to the patients to help them recall the drugs used.
## Recommendations
For policy makersHealth education initiatives on antibiotic self-medication should be directed at persons per age, gender, educational levels, and the public at large. Interventions aimed at minimizing antibiotic self-medication should focus on limiting access to antibiotics obtained without a prescription. Antibiotics will not be sold over the counter without a prescription at community pharmacies. The involvement of community pharmacists in reducing the prevalence of SMP should be significant. For further researchersMore research is needed:To determine whether widely self-medicated medications cause microorganisms to develop antibiotic resistance. To determine the prevalence of antibiotic self-medication in children under the age of 18.To investigate the public awareness and perceptions of antibiotic self-medication
## Conclusion
Antibiotic self-medication was prevalent among the study participants. Self-medication with antibiotics was 48.3 percent of the time, whereas general self-medication was $71\%$. Antibiotic self-medication is significantly associated with age, being a college or university student, and lack of health insurance. Self-medication with antibiotics is more common among adults aged 25–34 than in other age groups. The antibiotic amoxicillin was the most commonly self-medicated, followed by ciprofloxacin. Community pharmacies are a common source of antibiotics for self-medication. Self-medication with antibiotics, primarily amoxicillin and ciprofloxacin, was used to alleviate cough, abdominal pain, diarrhea, and vomiting. The majority of these drugs were gotten from a community pharmacy. *In* general, OPD patients appear to underestimate the possible hazardous effects of SMP. As a result, health officials should make an effort. As a result, health authorities such as the Drug and Therapeutic Committee, the Drug Regulatory Authority, hospital management, and other stakeholders should work together to ensure that antibiotics are used safely.
## References
1. Hernandez-Juyol M, Job-Quesada J. **Dentistry and self-medication: a current challenge**. *Medicina oral: organo oficial de la Sociedad Espanola de Medicina Oral y de la Academia Iberoamericana de Patologia y Medicina Bucal* (2002.0) **7** 344-347. PMID: 12415218
2. Mekuria AB. **Prevalence and predictors of self-medication practice among teachers’ education training college students in Amhara region, Ethiopia: a cross-sectional study**. *Front Pharmacol* (2021.0) **11** 2457. DOI: 10.3389/fphar.2020.593764
3. Helal R, Abou-ElWafa H. **Self-medication in university students from the city of Mansoura, Egypt**. *J Environ Public Health* (2017.0). DOI: 10.1155/2017/9145193
4. Björnsdóttir I, Almarsdóttir AB, Traulsen JM. **The lay public’s explicit and implicit definitions of drugs**. *Res Social Adm Pharm* (2009.0) **5** 40-50. DOI: 10.1016/j.sapharm.2008.04.003
5. Sweileh MW. **Self-medication and over-the-counter practices: a study in Palestine**. *Al-Aqsa Univ J (Nat Sci Ser)* (2004.0) **8** 1-9
6. Ocan M. **Household antimicrobial self-medication: a systematic review and meta-analysis of the burden, risk factors and outcomes in developing countries**. *MC Public Health* (2015.0) **15** 1-11
7. González-López JR, Rodríguez-Gázquez MDLÁ, Lomas-Campos MDLM. **Self-medication in adult Latin American immigrants in Seville**. *Acta Paulista de Enfermagem* (2012.0) **25** 75-81. DOI: 10.1590/S0103-21002012000900012
8. Andaulem T, Gebre-Mariam T. **Self-medication practices in Addis Ababa: a prospective study**. *Ethiop J Health Sci* (2004.0) **14** 1-11
9. Fuentes Albarrán K, Villa Zapata L. **Analysis and quantification of self-medication patterns of customers in community pharmacies in southern Chile**. *Pharm World Sci* (2008.0) **30** 863-868. DOI: 10.1007/s11096-008-9241-4
10. Ruiz ME. **Risks of self-medication practices**. *Curr Drug Saf* (2010.0) **5** 315-323. DOI: 10.2174/157488610792245966
11. Mihretie TM. *Self-medication practices with antibiotics among urban dwellers of Bahir Dar town, north west Ethiopia* (2014.0)
12. 12.World Health OrganizationGuidelines for the regulatory assessment of medicinal products for use in self-medication2000GenevaWorld Health Organization. *Guidelines for the regulatory assessment of medicinal products for use in self-medication* (2000.0)
13. Sharif SI, Bugaighis LM, Sharif RS. **Self-medication practice among pharmacists in UAE**. *Pharmacol Pharm* (2015.0) **6** 428. DOI: 10.4236/pp.2015.69044
14. Shankar P, Partha P, Shenoy N. **Self-medication and non-doctor prescription practices in Pokhara valley, Western Nepal: a questionnaire-based study**. *BMC Fam Pract* (2002.0) **3** 1-7. DOI: 10.1186/1471-2296-3-17
15. Shoaib MH. **Survey based study on the use of non-prescription drugs among pharmacists and non-pharmacists**. *Afr J Pharm Pharmacol* (2013.0) **7** 2652-2656. DOI: 10.5897/AJPP2013.3821
16. Mahardika AB, Suryawati S, Aji R. **Intervensi CBIA untuk Meningkatkan Pengetahuan, Sikap, dan Perilaku Penggunaan Antibiotik yang Rasional pada Anggota Bina Keluarga Balita**. *Jurnal Kedokteran Brawijaya* (2016.0) **29** 165-169. DOI: 10.21776/ub.jkb.2016.029.02.11
17. Jassim UT, Ebrahim SM. **Self-medication among nurses working in Basra teaching hospitals, Iraq**. *Turk J Physiother Rehabil* (2021.0) **32** 3
18. Kasulkar AA, Gupta M. **Self medication practices among medical students of a private institute**. *Indian J Pharm Sci* (2015.0) **77** 178. DOI: 10.4103/0250-474X.156569
19. Skliros E. **Self-medication with antibiotics in rural population in Greece: a cross-sectional multicenter study**. *BMC Fam Pract* (2010.0) **11** 1-3. DOI: 10.1186/1471-2296-11-58
20. Sawalha AF. **A descriptive study of self-medication practices among Palestinian medical and nonmedical university students**. *Res Soc Adm Pharm* (2008.0) **4** 164-172. DOI: 10.1016/j.sapharm.2007.04.004
21. Balamurugan E, Ganesh K. **Prevalence and pattern of self medication use in coastal regions of South India**. *Br J Med Pract* (2011.0) **4** a428
22. Gbagbo FY, Nkrumah J. **Self-medication among pregnant women in two municipalities in the Central Region of Ghana**. *Health Care Women Int* (2021.0) **42** 547-562. DOI: 10.1080/07399332.2020.1716235
23. Bradley C, Blenkinsopp A. **Over the counter drugs: the future for self medication**. *BMJ* (1996.0) **312** 835-837. DOI: 10.1136/bmj.312.7034.835
24. Nyambega JO. **Antibiotic use and misuse among adults in Magwagwa Ward, Nyamira County in Kenya**. *Age* (2017.0) **18** 26-35
25. Ocan M. **Patterns and predictors of self-medication in northern Uganda**. *PLoS ONE* (2014.0) **9** e92323. DOI: 10.1371/journal.pone.0092323
26. Parakh R. **Selfmedication practice among engineering students in an engineering college in North India**. *J Phytopharm* (2013.0) **2** 30-36. DOI: 10.31254/phyto.2013.2405
27. Abay S, Amelo W. **Assessment of Self-medication practices among medical, pharmacy, health science students in Gondar university**. *Ethiop J Young Pharm* (2010.0) **2** 306-310. DOI: 10.4103/0975-1483.66798
28. Gutema GB. **Self-medication practices among health sciences students: the case of Mekelle university**. *J Appl Pharm Sci* (2011.0) **1** 183
29. Bekele SA, Argaw MD, Yalew AW. **Magnitude and factors associated with self-medication practices among university students: the case of Arsi University, College of Health Science, Asella, Ethiopia: cross-sectional survey based study**. *Open Access Libr J* (2016.0) **3** 1-15
30. Kassa T. **Antibiotics self-medication practices among health care professionals in selected public hospitals of Addis Ababa, Ethiopia**. *Heliyon* (2022.0) **8** e08825. DOI: 10.1016/j.heliyon.2022.e08825
31. 31.Negera TT. Determinants of adherence to anti-retroviral therapy (ART): protocols in public hospitals in Addis Ababa, Ethiopia (Doctoral dissertation). 2020.
32. Ngigi CK, Mwagiru P, Wala J. **Self-medication with antibiotics prior to seeking treatment among adult patients attending outpatient department at Gatundu sub-county hospital, Kiambu County**. *Kenya. Imp J Interdiscip Res.* (2016.0) **2** 41-44
33. 33.Venkat MU, Nagesh A, Venkat MK. Community based survey of self-medication usage in Andhra Pradesh.
34. Alghanim S. **Self-medication practice among patients in a public health care system**. *East Mediterr Health J* (2011.0) **17** 409-416. DOI: 10.26719/2011.17.5.409
35. Olayemi O, Olayinka B, Musa A. **Evaluation of antibiotic self-medication pattern amongst undergraduate students of Ahmadu Bello university (main campus) Zaria**. *Res J Appl Sci Eng Technol* (2010.0) **2** 35-38
36. Fadare JO, Tamuno I. **Antibiotic self-medication among university medical undergraduates in northern Nigeria**. *J Public Health Epidemiol* (2011.0) **3** 217-220
37. Osemene K, Lamikanra A. **A study of the prevalence of self-medication practice among university students in southwestern Nigeria**. *Trop J Pharm Res* (2012.0) **11** 683-689. DOI: 10.4314/tjpr.v11i4.21
38. Widayati A. **Self medication with antibiotics in Yogyakarta City Indonesia: a cross sectional population-based survey**. *BMC Res Notes* (2011.0) **4** 1-8. DOI: 10.1186/1756-0500-4-491
39. Omolase C. **Self medication amongst general outpatients in a Nigerian community hospital**. *Ann Ibadan Postgrad Med* (2007.0) **5** 64-67
40. Askarian M. **A study of antibiotics self-medication at primary health care centers in Shiraz, Southern Iran**. *J Health Sci Surveill Syst* (2013.0) **1** 1-5
41. Donkor ES. **Self-medication practices with antibiotics among tertiary level students in Accra, Ghana: a cross-sectional study**. *Int J Environ Res Public Health* (2012.0) **9** 3519-3529. DOI: 10.3390/ijerph9103519
42. Sajith M. **Self-medication practices among health care professional students in a tertiary care hospital, Pune**. *Open Public Health J* (2017.0) **10** 63-68. DOI: 10.2174/1874944501710010063
43. Ali AN. **Self-medication practices among health care professionals in a Private University, Malaysia**. *Int Curr Pharm J* (2012.0) **1** 302-310. DOI: 10.3329/icpj.v1i10.11846
44. Awad A, Al-Rabiy S, Abahussain E. **Self-medication practices among diabetic patients in Kuwait**. *Med Princ Pract* (2008.0) **17** 315-320. DOI: 10.1159/000129612
45. Adhikary M. **Study of self-medication practices and its determinant among college students of Delhi University North Campus, New Delhi, India**. *Int J Med Sci Public Health* (2014.0) **3** 406-409. DOI: 10.5455/ijmsph.2014.260120146
46. Ibrahim NK. **Self-medication with analgesics among medical students and interns in King Abdulaziz University, Jeddah, Saudi Arabia**. *Pak J Med Sci* (2015.0) **31** 14. PMID: 25878607
|
---
title: 'Metabolomic biomarkers of the mediterranean diet in pregnant individuals:
A prospective study'
authors:
- Liwei Chen
- Jin Dai
- Zhe Fei
- Xinyue Liu
- Yeyi Zhu
- Mohammad L. Rahman
- Ruijin Lu
- Susanna D. Mitro
- Jiaxi Yang
- Stefanie N. Hinkle
- Zhen Chen
- Yiqing Song
- Cuilin Zhang
journal: Clinical nutrition (Edinburgh, Scotland)
year: 2023
pmcid: PMC10029322
doi: 10.1016/j.clnu.2023.01.011
license: CC BY 4.0
---
# Metabolomic biomarkers of the mediterranean diet in pregnant individuals: A prospective study
## SUMMARY
### Background and aims:
Metabolomic profiling is a systematic approach to identifying biomarkers for dietary patterns. Yet, metabolomic markers for dietary patterns in pregnant individuals have not been investigated. The aim of this study was to identify plasma metabolomic markers and metabolite panels that are associated with the Mediterranean diet in pregnant individuals.
### Methods:
This is a prospective study of 186 pregnant individuals who had both dietary intake and metabolomic profiles measured from the Fetal Growth Studies-Singletons cohort. Dietary intakes during the peri-conception/1st trimester and the second trimester were accessed at 8–13 and 16–22 weeks of gestation, respectively. Adherence to the Mediterranean diet was measured by the alternate Mediterranean Diet (aMED) score. Fasting plasma samples were collected at 16–22 weeks and untargeted metabolomics profiling was performed using the mass spectrometry-based platforms. Metabolites individually or jointly associated with aMED scores were identified using linear regression and least absolute shrinkage and selection operator (LASSO) regression models with adjustment for potential confounders, respectively.
### Results:
Among 459 annotated metabolites, 64 and 41 were individually associated with the aMED scores of the diet during the peri-conception/1st trimester and during the second trimester, respectively. Fourteen metabolites were associated with the Mediterranean diet in both time windows. Most Mediterranean diet-related metabolites were lipids (e.g., acylcarnitine, cholesteryl esters (CEs), linoleic acid, long-chain triglycerides (TGs), and phosphatidylcholines (PCs), amino acids, and sugar alcohols. LASSO regressions also identified a 10 metabolite-panel that were jointly associated with aMED score of the diet during the peri-conception/1st trimester (AUC: 0.74; $95\%$ CI: 0.57, 0.91) and a 3 metabolites-panel in the 2nd trimester (AUC: 0.68; $95\%$ CI: 0.50, 0.86).
### Conclusion:
We identified plasma metabolomic markers for the Mediterranean diet among pregnant individuals. Some of them have also been reported in previous studies among non-pregnant populations, whereas others are novel. The results from our study warrant replication in pregnant individuals by future studies.
### Clinical trial registration number:
This study was registered at ClinicalTrials.gov.
## Introduction
In the past two decades, growing evidence has consistently indicated the importance of dietary patterns as a measure of overall dietary quality, rather than individual nutrients or foods, in promoting health and preventing disease risk [1-3]. The Mediterranean diet, a traditional dietary pattern among people living in the Mediterranean Basin, featured higher intakes of vegetables, fruits, nuts, legumes, fish, cereals, and olive oil, but lower intakes of red and processed meats and sweets [4,5]. The favorable cardiometabolic and neurological effects of the Mediterranean diet have been demonstrated in both high-quality randomized controlled trials (RCT) [6] and systematic reviews and meta-analyses across different populations [7-9], including pregnant individuals [10,11]. Yet, molecular markers of the Mediterranean diet have not been elucidated in pregnant individuals.
Biomarkers of dietary intake can be applied as objective measures of dietary patterns and help to understand the underlying biological pathways between diet and health outcomes [12]. As compared to the traditional approach of examining biomarkers for a single nutrient or food separately (which likely overlooks the interactions among nutrients and food groups), the recent advance in high-throughput untargeted metabolomic profiling techniques permits a more comprehensive and systematic approach to identifying biomarkers for dietary patterns [13]. By measuring down-stream small molecules or metabolic products (<1.5 kDa, metabolomic markers), the metabolomics approach may provide more information on the interactions between nutrients/foods and genes for individuals and could identify novel biomarkers or biological pathways. Such an approach is ideal for identifying the complex, net impact of numerous nutrients and their metabolism in human bodies for a given dietary pattern.
Recent studies have investigated the metabolomic markers for the Mediterranean diet in non-pregnant populations, with each study having identified several blood metabolomic markers for the Mediterranean [14-23]. Yet, no studies have investigated the metabolomic markers for this healthy dietary pattern in pregnant individuals. Pregnancy can result in a series of dynamic physiological changes, including alterations of the maternal hormonal profile, basal metabolic rate, energy storage, and partition [24]. Thus, the metabolic responses to diet in pregnant individuals may not be the same as the non-pregnant populations. Furthermore, no previous studies have performed supervised methods to identify panels of dietary patterns related to metabolites. As dietary patterns are combinations of foods and nutrients, a panel of metabolites could better capture the multidimensionality and interrelations of nutrients and foods presented in the dietary patterns. Therefore, the primary aim of this study is to identify blood metabolomic markers and metabolite panels that are associated with the Mediterranean diet in pregnant individuals.
## Study population and design
This was a prospective study among racially diverse pregnant individuals who enrolled in the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Fetal Growth Studies-Singletons cohort (FGS). Details of the cohort have been described previously [25]. Briefly, a total of 2802 individuals aged 18–40 years with singletons were recruited between 8 and 13 weeks of gestation from 12 clinical sites in the United States between July 2009 and January 2013. Institutional Review Board approval was obtained for all participating clinical sites, the data coordinating center, and NICHD. The current analysis only included 186 individuals from a nested gestational diabetes mellitus (GDM) case–control study within the FGS who had both blood plasma metabolomic profiling data and dietary intakes [26]. The participant flow chart is presented in Supplementary Fig. 1.
## Assessment of dietary intakes and calculation of alternate mediterranean scores
We assessed individual's dietary intakes in the past 3 months (i.e., during the peri-conception and 1st trimester) at the baseline visit (8–13 weeks) using a semi-quantitative Food Frequency Questionnaire (FFQ) and during the second trimester (16–22 weeks) using the automated self-administered 24-h dietary recall (ASA24®; version Beta, 2011). Both dietary assessment tools were developed and validated by the National Cancer Institute, National Institutes of Health (NIH) [27-29]. We measured the adherence to the Mediterranean diet by calculating the alternate Mediterranean score (aMED) using the foods and nutrients data from the FFQ and ASA24. The aMED score was calculated from 8 food and nutrient components (i.e., fruits, vegetables, whole grains, nuts, legumes, fish, red and processed meats, and monounsaturated-to-saturated fat ratio), with the healthy components, scored 1 for above the median intake and 0 for below, and unhealthy components scored reversely [30]. Thus, a higher score indicated a better adherence to the Mediterranean diet. This method has been applied to calculate the alternate Mediterranean score in the same cohort in previous publications [10,11].
## Biospecimen collection and metabolomics profiling and data pretreatment
We collected the fasting blood samples in the 2nd trimester ($91.0\%$ of the blood draw were within 16–22 weeks). Plasma samples were immediately processed and stored at −80.0 °C until analysis. We performed the untargeted metabolomic profiling at the NIH West Coast Metabolomics Center at the University of California Davis using the two platforms: high-throughput liquid chromatography-quadrupole time of flight mass spectrometry (LC-QTOF-MS/MS) and gas chromatography-time of flight mass spectrometry (GC-TOF-MS) [31]. All samples were analyzed by the two platforms. Internal standards were used for the calibration of retention times. A total of 751 features were detected, with 459 annotated metabolites and 292 unknown features. In the current study, we only included the 459 annotated metabolites. All 459 metabolites had missing values < $20\%$ and missing values were imputed with half of the minimum value by batch, which is recommended for metabolomics data [32]. To correct day-to-day technical variation from the platform, metabolites were divided by their median values (i.e., re-scaled to a median of 1) and log-transformed within each batch [33].
Of the total 459 known metabolites, 353 were lipids or lipid-like molecules ($76.9\%$), followed by 55 organic acids and derivatives ($12.0\%$), 30 organic oxygen compounds ($6.5\%$), 12 organo-heterocyclic compounds ($2.6\%$), and 9 “others” (including 3 homogeneous non-metal compounds, 3 nucleosides, nucleotides and analogues, 2 benzenoids, and 1 organic nitrogen compound) (Supplementary Fig. 2).
## Covariates
We collected individuals’ sociodemographic characteristics, lifestyle factors, and reproductive and medical history using a detailed questionnaire at study enrollment. Pre-pregnancy body mass index (BMI) was calculated based on height measured at baseline and self-reported pre-pregnancy weight. We categorized individuals into normal weight (19.0–24.9 kg/m2), overweight (25.0–29.9 kg/m2), or obese (≥30.0 kg/m2) by their pre-pregnancy BMI. We assessed physical activity using a validated Pregnancy Physical Activity Questionnaire (PPAQ) [34] in the first trimester for measuring habitual physical activity in the past 12 months and second trimester for physical activity since the baseline visit.
## Statistical methods
We applied sampling weights in the statistical analyses to account for the oversampling of individuals with GDM in nested-case control samples and to represent the results in the full FGS sample. We described and compared individuals’ characteristics at the study enrollment across the aMED score tertiles. We presented the results as weighted percentages (%) and actual frequency (N) for categorical variables and weighted mean (standard errors, SE) for continuous variables. We calculated the P-values comparing individuals across aMED score tertiles by one-way Analysis of variance (ANOVA) tests for continuous variables and χ2-tests for categorical variables.
Several steps were performed to identify plasma metabolomic markers for aMED score. We used the aMED scores derived from the FFQs (assessed at baseline visit of 8–13 weeks) in the prospective analyses and the aMED scores derived from the ASA24® (assessed at 16–22 weeks) during the 2nd trimester in the cross-sectional analyses. We first identified plasma metabolites that were individually associated with aMED score using the linear regressions with adjustment for potential confounders, including maternal age (years), race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander), education (high-school degree or less, associate degree or more), pre-pregnancy BMI (kg/m2), and total physical activity (minutes/week). Metabolites were selected if they were significantly different, comparing the highest aMED tertile to the lowest tertile. We controlled for multiple comparisons by using the Benjamini-Hochberg method, with the overall false discovery rate (FDR) < 0.05 being considered statistically significant [35]. We also conducted the sensitivity analyses of the associations of individual metabolites with aMED score, stratified by the GDM status (i.e., within individuals who developed GDM and individuals who didn't develop GDM in the late 2nd semester). We performed the least absolute shrinkage and selection operator (LASSO) regressions to select the panel of plasma metabolites jointly associated (i.e., with) with aMED score [36]. The metabolites panels were selected if they had non-zero coefficients based on the criteria of lambda.1se. To avoid over-fitting, we performed the 10-fold cross-validation in LASSO regression with participants randomly divided into training and validation sets with a ratio of 2:1 and calculated the area under the curve (AUC) of each panel. All data analyses were conducted using SAS software (version 9.4; SAS Institute, Cary, NC, US) or R (version 4.0.2; R Studio: Integrated Development for R. R Studio, Inc., Boston, MA, US).
## Participants baseline characteristics
Among all individuals, $27.2\%$ were non-Hispanic White, $25.7\%$ were non-Hispanic Black, $24.4\%$ were Hispanic, and $22.7\%$ were Asian/Pacific Islander. The mean (SE) age of individuals at enrollment was 28.0 (0.4) years, and the mean BMI was 25.5 (0.4) kg/m2, with $50.4\%$ having a pre-pregnancy BMI in the normal range. The baseline characteristics of all individuals and by their aMED score tertiles are presented in Table 1. Compared to individuals in the lowest aMED score tertile (T1), individuals in the highest tertile (T3) were more likely to be non-Hispanic White, highly educated, nulliparous, and had lower pre-pregnancy BMI and a healthier profile of dietary intake.
## Prospective associations of aMED score reported in the first trimester
Among 459 annotated metabolites, 64 were individually associated with aMED score reported in the first trimester (FDR<0.05), adjusting for age, race/ethnicity, education, pre-pregnancy BMI, and physical activity (Table 2). Among 64 metabolites, 51 ($79.69\%$) had positive associations and 13 ($20.31\%$) had negative associations (Fig. 1A and Supplementary Fig. 3A). In the sensitivity analysis, all 64 metabolites remained significantly associated with aMED score in individuals without GDM (Supplementary Table 1).
A panel of 10 metabolites was jointly associated with aMED score (AUC: 0.74; $95\%$ CI: 0.57, 0.91) in LASSO regression, including positive associations of cholesterol ester (CE, 20:5) A, triglycerides (TG, 49:1), TG (58:4), phosphatidylcholines (PC, 33:1), and PC (40:7), and inverse associations of glutamic acid, aspartic acid, and 3-hydroxybutyric acid, and epsilon-caprolactam (Fig. 2). Except for the epsilon-caprolactam, all other 9 metabolites were also individually associated with aMED scores.
## Cross–sectional associations of metabolites with aMED score reported in the second trimester
After adjusting for the abovementioned covariates, 41 metabolites were significantly associated (FDR< 0.05) with aMED score reported in the second trimester, with 25 ($61.0\%$) having positive associations and 16 ($39.0\%$) having negative associations (Fig. 1B and Supplementary Fig. 3B). Among the 41 metabolites, 14 ($34\%$) were overlapped with the metabolites prospectively associated with aMED reported in the 1st trimester, including glycolic acid, aspartic acid, 3-hydroxybutyric acid, linoleic acid, acylcarnitine (C18:2) and (C18:0), TGs (58:4), TG (56:1) A and B, TG (60:2), PC (36:5) B, PC (42:6), and CE (20:5) A and B (Table 2). In the sensitivity analysis, all 41 metabolites remained significantly associated with aMED score in individuals without GDM (Supplementary Table 2).
A panel of 3 metabolites was jointly associated with aMED score with an AUC of 0.68 ($95\%$ CI: 0.50, 0.86) in the LASSO regression, including a positive association of CE (20:5) B and inverse associations of glycolic acid and acylcarnitine (C16:0). All 3 were also individually associated with the aMED score in linear regressions.
## Discussions
In this longitudinal study among pregnant individuals, we identified several plasma metabolomic markers for the Mediterranean diet in peri-conception and 1st trimester or recent diet in the 2nd trimester. Most Mediterranean diet-related metabolites are lipids species (e.g., acylcarnitines, linoleic acid, long-chain TGs, PCs, and CEs); others are amino acids and derivatives, and sugar alcohols. Of note, 14 metabolites were significantly related to the Mediterranean diet in both time windows. In addition to individual metabolites, we also identified multi-metabolite panels for the Mediterranean diet in pregnant individuals at two time windows. Such multi-metabolite panels with good-to-excellent predictability are promising to be considered the potential biomarkers of the Mediterranean diet because they significantly reduced the numbers of metabolites (i.e., as compared with all individually associated metabolites) we need to measure in future applications.
We are unaware of previous studies investigating the metabolomic markers for the Mediterranean diet in pregnant individuals. In the current study among pregnant individuals, we replicated some Mediterranean-related metabolites which were reported in previous studies among non-pregnant populations, including several long-chain TGs and PCs, glyceryl palmitate, free fatty acids (palmitoleic acid, oleic acid, linoleic acid), medium/long-chain acylcarnitines (C16:0; C18:0; C18:1; C18:2), CE (20:5), and amino acids (aspartic acid, 3-hydroxybutyric acid, glycolic acid) [14-23]. We also identified novel markers such as glycolic acid, sugar alcohols (i.e., lyxitol and xylitol), and organic acids (i.e., citric acid and isocitric acid). Cross-population consistent findings are promising, supporting the concept that metabolomics could be a new approach for identifying biomarkers for dietary patterns across populations. The novel metabolites may reflect the variability of the physiological conditions of study populations (e.g., pregnant vs. non-pregnant), but could also be due to the differences in study design, population, metabolomic profiling, and statistical analysis approach.
Consistent with previous studies in non-pregnant populations [14-23], we found most metabolites associated with the Mediterranean diet among pregnant individuals are lipids species (i.e., CE (20:5) A and B, long-chain acylcarnitines (C18:0) and (C18:2), long-chain TGs (TG (56:1) A and B, TG (58:4), TG (60:2) A, and PC (36:5) B, and linoleic acid). As the esterified form of cholesterol with a single fatty acid, CE is the major form of cholesterol in human plasma [37]. The plasma levels of CEs have been linked with dietary intake of monosaturated fatty acids (MUFA) [38], olive oil, and seafood [21,39]. The health effects of individual CEs are unclear, but available evidence tends to suggest that CEs with longer, more unsaturated acyl chains may be favorable for cardiovascular diseases (CVD) [16] and diabetes [40]. Acylcarnitines have been identified as markers of meat (positive associations) [1,41,42], and coffee intakes [43] (inverse associations). Long-chain acylcarnitines were also positively associated with Western dietary patterns in a Canadian study [44]. As intermediates of fatty acid oxidation, disturbance of plasma acylcarnitines has been linked with the development of diabetes [45,46] and CVD [47,48] in the non-pregnant population, as well as, GDM and fetal development in pregnant individuals [49,50]. Indeed, findings from PREDIMED trial data have suggested that the Mediterranean diet may mitigate the adverse associations of acylcarnitines with CVD [16,51]. Plasma long-chain TGs and PCs mainly come from the consumption of fish, nuts, and vegetable oils (e.g. olive oil) [52]. Long-chain PCs have also been identified as markers of coffee consumption [53]. Plasma profiles of TGs and PC have been suggested as important signatures of insulin sensitivity and CVD risk [40,54]. TGs and PCs with different acyl chains and double bonds may play different roles, but the exact contribution of each or the combination of TGs and PCs is still poorly understood.
Linoleic acid was inversely associated with the Mediterranean diet at two-time windows in our study and a previous study in male Finnish smokers from the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study cohort [23]. As an essential fatty acid to humans, the only resource of linoleic acid is from dietary intake (primarily from vegetable oils, nuts, and seeds) [55]. In the human body, linoleic acid is the parental n-6 polyunsaturated fatty acid (PUFA) and can be converted to arachidonic acid and subsequently metabolized to pro-inflammatory lipids mediators (i.e., eicosanoids). On the other hand, α-linolenic acid, an n-3 PUFA, can be converted to eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), which subsequently produce the anti-inflammatory lipid mediators (resolvins and protectins) [56]. Mediterranean diet has been linked with a well-balanced plasma linoleic/α-linolenic acid ratio, which may result in a lower risk of CVD through the inflammation responses [57,58].
Several amino acids and derivatives such as glutamic acid (a major excitatory neurotransmitter and a by-product of the branched-chain amino acids catabolism), aspartic acid (a metabolite involved in the urea cycle), and 3-hydroxybutyric acid (a by-product of short-chain fatty acids and branched-chain amino acids; also known as β-hydroxybutyric acid) were identified as metabolomic markers for Mediterranean diet in our study and two observations studies in non-pregnant populations [20,21]. These amino acid derivatives are mainly involved in the urea cycle during amino acid catabolism [59]. Previous studies have shown the association of glutamic and aspartic acids with central obesity [60-62], diabetic retinal disease [61,63], and risk of CVD [64]. More importantly, the glutamic acid/glutamate ratio has been identified as the single metabolite most strongly correlated with the visceral adipose tissue in several populations [62,65,66]. Taken together, it is possible that gluconeogenesis from amino acids could play important roles in the potential metabolic pathways that explain some health benefits of the Mediterranean diet.
We identified a couple of novel carbohydrate metabolites that were associated with the Mediterranean diet in pregnant individuals, including two sugar alcohols (i.e., lyxitol and xylitol), glycolic acid, and two organic acids (i.e., citric acid and isocitric acid). Xylitol and lyxitol are sugar alcohols that can be found naturally in many fruits and vegetables, or artificially produced [67]. As non-digestible carbohydrates, they have a sweet taste but produce much fewer calories in human bodies. Thus, they (mainly xylitol) are widely used as a sugar substitute in “sugar-free” products such as chewing gums, yogurt, cookies, and candies [67]. The plasma levels of these sugar alcohols are mainly from dietary sources through passive diffusion in the small intestine [68]. Therefore, the positive associations of xylitol and lyxitol with the Mediterranean diet may reflect the high consumption of fruit and vegetables, or some “commercially labeled “sugar-free” products, or both. Glycolic acid was inversely associated with the Mediterranean diet in both the first and second trimesters in our study of pregnant individuals, but not in previous studies in non-pregnant populations. Foods containing oxalic acid (e.g., spinach, rhubarb, and almond milk) or glyoxal (e.g. bread, cookies, yogurt, sardine oil, coffee, tea, beer, and wine) are likely the main exogenous sources of glycolic acid. However, most glycolic acids are produced endogenously from glycolaldehyde (a product of fructose and xylitol metabolism) during the synthesis of oxalate [69]. Clinically, glycolic acid is widely used in skin care products [70]. Urinary glycolic acid has also been inversely related to obesity and visceral fat tissue in healthy adults [71]. Citric acid and isocitric acid are tricarboxylic acid (TCA) cycle intermediates. TCA cycle is the final common oxidative pathway for carbohydrates, fats, and amino acids. It is the most important metabolic pathway for the energy supply in humans [72]. Citric acid is found in citrus fruits (e.g., oranges, lemons, and limes) and isocitric acid is rich in most berries (e.g., blackberries) and vegetables (e.g., carrots). Both of them are also commonly used as a flavoring (add sour taste) and preservative in food and beverages, especially soft drinks and candies. Urinary citric acid has been identified as the marker of dietary intake of wine and grape juice and lactovegetarian diet [73]. Serum isocitric acid has been linked to dementia in a small case–control study [74] and mild cognitive impairment in a large prospective study among Hispanic Community Health Study/Study of Latinos (HCHS/SOL) [75]. In pregnant individuals, first-trimester plasma citric acid has been identified as one of the top metabolites for predicting preeclampsia [76].
In the current study among pregnant individuals, we replicated some Mediterranean diet-related metabolites which were identified in previous studies among non-pregnant population. Among them, linoleic acid, aspartic acid, 3-hydroxybutyric acid, and CE (20:5) were associated with the Mediterranean diet in both time windows during the pregnancy, suggesting the robustness of the findings. The replicated findings are promising, supporting the concept that metabolomics could be a new approach for identifying biomarkers for dietary patterns across populations. The novel metabolites may reflect the variability of the physiological conditions of study populations (e.g., pregnant vs. non-pregnant), but could also be due to the differences in study design, population, metabolomic profiling, and statistical analysis approach.
Our study has several unique strengths. As the first study that investigated the metabolomic markers for Mediterranean diet in pregnant individuals, we recruited pregnant individuals from multiple racial and ethnic groups in 12 health centers in the US to enhance the generalizability of the study results to US pregnant women. We used longitudinal data with dietary intakes assessed prior to the plasma specimen collection to ensure the temporality of the association. Most previous studies only reported cross–sectional correlations [18-23,39]. Furthermore, we calculated the aMED score using a predefined method, which has been applied to both pregnant and non-pregnant populations, making our results to be easily compared with and replicated in other studies. In addition, we collected fasting blood samples that are less subjected to measurement variability and less influenced by other factors. As dietary patterns are combinations of foods and nutrients, a panel of metabolites could better capture the interrelations of nutrients and foods presented in dietary patterns with much smaller numbers of metabolites. In addition, with the availability of longitudinal dietary data before pregnancy and during early pregnancy, we were able to identify both long-term and short-term metabolite markers of Mediterranean diet. The multi-metabolite panels had fair-to-good predictability of adherence to Mediterranean diet in both first and second trimesters among pregnant individuals, suggesting the potential applications of using them as biomarkers of dietary intake.
Several limitations of this study should be considered when interpreting the results. First, our study is an observational study. Although we have examined and adjusted confounders rigorously, the residual confounding cannot be completely ruled out. Second, we only have one measure of 24-h recall at 2nd trimester, which may not be able to capture some foods that are usually consumed by study participants. Nevertheless, dietary assessments using one 24-h recall have been applied to derive the dietary patterns, including the Mediterranean diet, in both non-pregnant adult population [77] and pregnant individuals in large epidemiological studies [10]. Having meaningful numbers of overlapped metabolites from both the aMED scores derived from FFQ and one 24-h recall also suggested the usefulness of both dietary assessment methods. Future studies may need to quantify the levels of misclassification. Lastly, we only included the known metabolites, which could possibly miss some important metabolic features. However, these known metabolites are more reliable and can be used to compare with the results from other studies.
In conclusion, we detected a set of metabolites associated with the Mediterranean diet in pregnant individuals. Some of them were significantly related to the Mediterranean diet in both peri-conception and early-to-mid pregnancy time windows. These metabolites may deserve further investigations using the targeted approach to improve the accuracy of quantification and examining their relationships with pregnancy and fetal outcomes in future studies.
## Data sharing
Data described in the manuscript, code book, and analytic code will be available upon request pending application and approval of a data-sharing agreement.
## References
1. Schulze MB, Martinez-Gonzalez MA, Fung TT, Lichtenstein AH, Forouhi NG. **Food based dietary patterns and chronic disease prevention**. *BMJ* (2018.0) **361** k2396. DOI: 10.1136/bmj.k2396
2. Hu FB. **Dietary pattern analysis: a new direction in nutritional epidemiology**. *Curr Opin Lipidol* (2002.0) **13** 3-9. DOI: 10.1097/00041433-200202000-00002
3. Cespedes EM, Hu FB. **Dietary patterns: from nutritional epidemiologic analysis to national guidelines**. *Am J Clin Nutr* (2015.0) **101** 899-900. DOI: 10.3945/ajcn.115.110213
4. Martinez-Gonzalez MA, Gea A, Ruiz-Canela M. **The mediterranean diet and cardiovascular health**. *Circ Res* (2019.0) **124** 779-98. DOI: 10.1161/CIRCRESAHA.118.313348
5. Trichopoulou A, Costacou T, Bamia C, Trichopoulos D. **Adherence to a Mediterranean diet and survival in a Greek population**. *N Engl J Med* (2003.0) **348** 2599-608. DOI: 10.1056/NEJMoa025039
6. Estruch R, Ros E, Martinez-Gonzalez MA. **Mediterranean diet for primary prevention of cardiovascular disease**. *N Engl J Med* (2013.0) **369** 676-7. DOI: 10.1056/NEJMc1306659
7. Samadi M, Moradi S, Moradinazar M, Mostafai R, Pasdar Y. **Dietary pattern in relation to the risk of Alzheimer's disease: a systematic review**. *Neurol Sci* (2019.0) **40** 2031-43. DOI: 10.1007/s10072-019-03976-3
8. Galbete C, Schwingshackl L, Schwedhelm C, Boeing H, Schulze MB. **Evaluating Mediterranean diet and risk of chronic disease in cohort studies: an umbrella review of meta-analyses**. *Eur J Epidemiol* (2018.0) **33** 909-31. DOI: 10.1007/s10654-018-0427-3
9. Martinez-Gonzalez MA, Bes-Rastrollo M. **Dietary patterns, Mediterranean diet, and cardiovascular disease**. *Curr Opin Lipidol* (2014.0) **25** 20-6. DOI: 10.1097/MOL.0000000000000044
10. Li M, Grewal J, Hinkle SN, Yisahak SF, Grobman WA, Newman RB. **Healthy dietary patterns and common pregnancy complications: a prospective and longitudinal study**. *Am J Clin Nutr* (2021.0) **114** 1229-37. DOI: 10.1093/ajcn/nqab145
11. Yisahak SF, Mumford SL, Grewal J, Li M, Zhang C, Grantz KL. **Maternal diet patterns during early pregnancy in relation to neonatal outcomes**. *Am J Clin Nutr* (2021.0) **114** 358-67. DOI: 10.1093/ajcn/nqab019
12. Brennan L, Hu FB. **Metabolomics-based dietary biomarkers in nutritional epidemiology-current status and future opportunities**. *Mol Nutr Food Res* (2019.0) **63** e1701064. DOI: 10.1002/mnfr.201701064
13. Guasch-Ferre M, Bhupathiraju SN, Hu FB. **Use of metabolomics in improving assessment of dietary intake**. *Clin Chem* (2018.0) **64** 82-98. DOI: 10.1373/clinchem.2017.272344
14. Vazquez-Fresno R, Llorach R, Urpi-Sarda M, Lupianez-Barbero A, Estruch R, Corella D. **Metabolomic pattern analysis after mediterranean diet intervention in a nondiabetic population: a 1- and 3-year follow-up in the PREDIMED study**. *J Proteome Res* (2015.0) **14** 531-40. DOI: 10.1021/pr5007894
15. Bondia-Pons I, Martinez JA, de la Iglesia R, Lopez-Legarrea P, Poutanen K, Hanhineva K. **Effects of short- and long-term Mediterranean-based dietary treatment on plasma LC-QTOF/MS metabolic profiling of subjects with metabolic syndrome features: the Metabolic Syndrome Reduction in Navarra (RESMENA) randomized controlled trial**. *Mol Nutr Food Res* (2015.0) **59** 711-28. DOI: 10.1002/mnfr.201400309
16. Toledo E, Wang DD, Ruiz-Canela M, Clish CB, Razquin C, Zheng Y. **Plasma lipidomic profiles and cardiovascular events in a randomized intervention trial with the Mediterranean diet**. *Am J Clin Nutr* (2017.0) **106** 973-83. DOI: 10.3945/ajcn.116.151159
17. Galié S, García-Gavilán J, Papandreou C, Camacho-Barcía L, Arcelin P, Palau-Galindo A. **Effects of Mediterranean Diet on plasma metabolites and their relationship with insulin resistance and gut microbiota composition in a crossover randomized clinical trial**. *Clin Nutr* (2021.0) **40** 3798-806. DOI: 10.1016/j.clnu.2021.04.028
18. McCullough ML, Maliniak ML, Stevens VL, Carter BD, Hodge RA, Wang Y. **Metabolomic markers of healthy dietary patterns in US postmenopausal women**. *Am J Clin Nutr* (2019.0) **109** 1439-51. DOI: 10.1093/ajcn/nqy385
19. Walker ME, Song RJ, Xu X, Gerszten RE, Ngo D, Clish CB. **Proteomic and metabolomic correlates of healthy dietary patterns: the framingham heart study**. *Nutrients* (2020.0) **12**. DOI: 10.3390/nu12051476
20. Tong TYN, Koulman A, Griffin JL, Wareham NJ, Forouhi NG, Imamura F. **A combination of metabolites predicts adherence to the mediterranean diet pattern and its associations with insulin sensitivity and lipid homeostasis in the general population: the fenland study, United Kingdom**. *J Nutr* (2020.0) **150** 568-78. DOI: 10.1093/jn/nxz263
21. Li J, Guasch-Ferré M, Chung W, Ruiz-Canela M, Toledo E, Corella D. **The Mediterranean diet, plasma metabolome, and cardiovascular disease risk**. *Eur Heart J* (2020.0) **41** 2645-56. DOI: 10.1093/eurheartj/ehaa209
22. Kim H, Hu EA, K EW, Yu B, Steffen LM, Seidelmann SB. **Serum metabolites associated with healthy diets in african Americans and European Americans**. *J Nutr* (2021.0) **151** 40-9. DOI: 10.1093/jn/nxaa338
23. Playdon MC, Moore SC, Derkach A, Reedy J, Subar AF, Sampson JN. **Identifying biomarkers of dietary patterns by using metabolomics**. *Am J Clin Nutr* (2017.0) **105** 450-65. DOI: 10.3945/ajcn.116.144501
24. Soma-Pillay P, Nelson-Piercy C, Tolppanen H, Mebazaa A. **Physiological changes in pregnancy**. *Cardiovascular journal of Africa* (2016.0) **27** 89-94. DOI: 10.5830/CVJA-2016-021
25. Grewal J, Grantz KL, Zhang C, Sciscione A, Wing DA, Grobman WA. **Cohort profile: NICHD fetal Growth studies-singletons and twins**. *Int J Epidemiol* (2018.0) **47**. DOI: 10.1093/ije/dyx161
26. Zhu Y, Li M, Rahman ML, Hinkle SN, Wu J, Weir NL. **Plasma phospholipid n-3 and n-6 polyunsaturated fatty acids in relation to cardiometabolic markers and gestational diabetes: a longitudinal study within the prospective NICHD Fetal Growth Studies**. *PLoS Med* (2019.0) **16** e1002910. DOI: 10.1371/journal.pmed.1002910
27. Subar AF, Thompson FE, Kipnis V, Midthune D, Hurwitz P, McNutt S. **Comparative validation of the block, willett, and national cancer Institute food frequency questionnaires : the eating at America's table study**. *Am J Epidemiol* (2001.0) **154** 1089-99. DOI: 10.1093/aje/154.12.1089
28. Subar AF, Kipnis V, Troiano RP, Midthune D, Schoeller DA, Bingham S. **Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: the OPEN study**. *Am J Epidemiol* (2003.0) **158** 1-13. DOI: 10.1093/aje/kwg092
29. Moshfegh AJ, Rhodes DG, Baer DJ, Murayi T, Clemens JC, Rumpler WV. **The US Department of Agriculture Automated Multiple-Pass Method reduces bias in the collection of energy intakes**. *Am J Clin Nutr* (2008.0) **88** 324-32. DOI: 10.1093/ajcn/88.2.324
30. Fung TT, Rexrode KM, Mantzoros CS, Manson JE, Willett WC, Hu FB. **Mediterranean diet and incidence of and mortality from coronary heart disease and stroke in women**. *Circulation* (2009.0) **119** 1093-100. DOI: 10.1161/CIRCULATIONAHA.108.816736
31. Cajka T, Smilowitz JT, Fiehn O. **Validating quantitative untargeted lipidomics across nine liquid chromatography-high-resolution mass spectrometry platforms**. *Anal Chem* (2017.0) **89** 12360-8. DOI: 10.1021/acs.a-nalchem.7b03404
32. Wei R, Wang J, Su M, Jia E, Chen S, Chen T. **Missing value imputation approach for mass spectrometry-based metabolomics data**. *Sci Rep* (2018.0) **8** 663. DOI: 10.1038/s41598-017-19120-0
33. Rebholz CM, Lichtenstein AH, Zheng Z, Appel LJ, Coresh J. **Serum untargeted metabolomic profile of the Dietary Approaches to Stop Hypertension (DASH) dietary pattern**. *Am J Clin Nutr* (2018.0) **108** 243-55. DOI: 10.1093/ajcn/nqy099
34. Chasan-Taber L, Schmidt Md, Roberts DE, Roberts De, Hosmer D, Hosmer D, Markenson G, Markenson G, Freedson PS, Freedson PS. *Development and validation of a pregnancy physical activity questionnaire*
35. Benjamini Y, Hochberg Y. **Controlling the false discovery rate: a practical and powerful approach to multiple testing**. *J Roy Stat Soc B* (1995.0) **57** 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x
36. Ranstam J, Cook JA. **LASSO regression**. *Br J Surg* (2018.0) **105** 1348. DOI: 10.1002/bjs.10895
37. Lecerf JM, de Lorgeril M. **Dietary cholesterol: from physiology to cardiovascular risk**. *Br J Nutr* (2011.0) **106** 6-14. DOI: 10.1017/s0007114511000237
38. Lopes SM, Trimbo SL, Mascioli EA, Blackburn GL. **Human plasma fatty acid variations and how they are related to dietary intake**. *Am J Clin Nutr* (1991.0) **53** 628-37. DOI: 10.1093/ajcn/53.3.628
39. Kim H, Anderson CA, Hu EA, Zheng Z, Appel LJ, He J. **Plasma metabolomic signatures of healthy dietary patterns in the chronic renal insufficiency cohort (CRIC) study**. *J Nutr* (2021.0) **151** 2894-907. DOI: 10.1093/jn/nxab203
40. Rhee EP, Cheng S, Larson MG, Walford GA, Lewis GD, McCabe E. **Lipid profiling identifies a triacylglycerol signature of insulin resistance and improves diabetes prediction in humans**. *J Clin Invest* (2011.0) **121** 1402-11. DOI: 10.1172/jci44442
41. Schmidt JA, Rinaldi S, Ferrari P, Carayol M, Achaintre D, Scalbert A. **Metabolic profiles of male meat eaters, fish eaters, vegetarians, and vegans from the EPIC-Oxford cohort**. *Am J Clin Nutr* (2015.0) **102** 1518-26. DOI: 10.3945/ajcn.115.111989
42. Playdon MC, Sampson JN, Cross AJ, Sinha R, Guertin KA, Moy KA. **Comparing metabolite profiles of habitual diet in serum and urine**. *Am J Clin Nutr* (2016.0) **104** 776-89. DOI: 10.3945/ajcn.116.135301
43. Altmaier E, Kastenmüller G, Römisch-Margl W, Thorand B, Weinberger KM, Adamski J. **Variation in the human lipidome associated with coffee consumption as revealed by quantitative targeted metabolomics**. *Mol Nutr Food Res* (2009.0) **53** 1357-65. DOI: 10.1002/mnfr.200900116
44. Bouchard-Mercier A, Rudkowska I, Lemieux S, Couture P, Vohl MC. **The metabolic signature associated with the Western dietary pattern: a cross-sectional study**. *Nutr J* (2013.0) **12** 158. DOI: 10.1186/1475-2891-12-158
45. Adams SH, Hoppel CL, Lok KH, Zhao L, Wong SW, Minkler PE. **Plasma acylcarnitine profiles suggest incomplete long-chain fatty acid beta-oxidation and altered tricarboxylic acid cycle activity in type 2 diabetic African-American women**. *J Nutr* (2009.0) **139** 1073-81. DOI: 10.3945/jn.108.103754
46. Schooneman MG, Vaz FM, Houten SM, Soeters MR. **Acylcarnitines: reflecting or inflicting insulin resistance?**. *Diabetes* (2013.0) **62** 1-8. DOI: 10.2337/db12-0466
47. Kukharenko A, Brito A, Kozhevnikova MV, Moskaleva N, Markin PA, Bochkareva N. **Relationship between the plasma acylcarnitine profile and cardiometabolic risk factors in adults diagnosed with cardiovascular diseases**. *Clin Chim Acta* (2020.0) **507** 250-6. DOI: 10.1016/j.cca.2020.04.035
48. Gander J, Carrard J, Gallart-Ayala H, Borreggine R, Teav T, Infanger D. **Metabolic impairment in coronary artery disease: elevated serum acylcarnitines under the spotlights**. *Front Cardiovasc Med* (2021.0) **8** 792350. DOI: 10.3389/fcvm.2021.792350
49. Song Y, Lyu C, Li M, Rahman ML, Chen Z, Zhu Y. **Plasma acylcarnitines during pregnancy and neonatal anthropometry: a longitudinal study in a multiracial cohort**. *Metabolites* (2021.0) **11**. DOI: 10.3390/metabo11120885
50. Lin Y, Wu J, Zhu Y, Hinkle SN, Rawal S, Liang L. **A longitudinal study of plasma acylcarnitines throughout pregnancy and associations with risk of gestational diabetes mellitus**. *Clin Nutr* (2021.0) **40** 4863-70. DOI: 10.1016/j.clnu.2021.07.008
51. Guasch-Ferré M, Zheng Y, Ruiz-Canela M, Hruby A, Martínez-González MA, Clish CB. **Plasma acylcarnitines and risk of cardiovascular disease: effect of Mediterranean diet interventions**. *Am J Clin Nutr* (2016.0) **103** 1408-16. DOI: 10.3945/ajcn.116.130492
52. Abedi E, Sahari MA. **Long-chain polyunsaturated fatty acid sources and evaluation of their nutritional and functional properties**. *Food Sci Nutr* (2014.0) **2** 443-63. DOI: 10.1002/fsn3.121
53. Jacobs S, Kröger J, Floegel A, Boeing H, Drogan D, Pischon T. **Evaluation of various biomarkers as potential mediators of the association between coffee consumption and incident type 2 diabetes in the EPIC-Potsdam Study**. *Am J Clin Nutr* (2014.0) **100** 891-900. DOI: 10.3945/ajcn.113.080317
54. Stegemann C, Pechlaner R, Willeit P, Langley SR, Mangino M, Mayr U. **Lipidomics profiling and risk of cardiovascular disease in the prospective population-based Bruneck study**. *Circulation* (2014.0) **129** 1821-31. DOI: 10.1161/circulationaha.113.002500
55. Dasilva G, Medina I. **Lipidomic methodologies for biomarkers of chronic inflammation in nutritional research: omega-3 and omega-6 lipid mediators**. *Free Radic Biol Med* (2019.0) **144** 90-109. DOI: 10.1016/j.free-radbiomed.2019.03.017
56. Dasilva G, Medina I. **Lipidomic methodologies for biomarkers of chronic inflammation in nutritional research: ω-3 and ω-6 lipid mediators**. *Free Radic Biol Med* (2019.0) **144** 90-109. DOI: 10.1016/j.free-radbiomed.2019.03.017
57. Mazzocchi A, Leone L, Agostoni C, Pali-Schöll I. **The secrets of the mediterranean diet. Does [only] olive oil matter?**. *Nutrients* (2019.0) **11**. DOI: 10.3390/nu11122941
58. de Lorgeril M, Salen P. **The Mediterranean-style diet for the prevention of cardiovascular diseases**. *Publ Health Nutr* (2006.0) **9** 118-23. DOI: 10.1079/phn2005933
59. Sell DR, Monnier VM. **Ornithine is a novel amino acid and a marker of arginine damage by oxoaldehydes in senescent proteins**. *Ann N Y Acad Sci* (2005.0) **1043** 118-28. DOI: 10.1196/annals.1333.015
60. Cheng S, Rhee EP, Larson MG, Lewis GD, McCabe EL, Shen D. **Metabolite profiling identifies pathways associated with metabolic risk in humans**. *Circulation* (2012.0) **125** 2222-31. DOI: 10.1161/circulationaha.111.067827
61. Maltais-Payette I, Allam-Ndoul B, Pérusse L, Vohl MC, Tchernof A. **Circulating glutamate level as a potential biomarker for abdominal obesity and metabolic risk**. *Nutr Metabol Cardiovasc Dis* (2019.0) **29** 1353-60. DOI: 10.1016/j.numecd.2019.08.015
62. Boulet MM, Chevrier G, Grenier-Larouche T, Pelletier M, Nadeau M, Scarpa J. **Alterations of plasma metabolite profiles related to adipose tissue distribution and cardiometabolic risk**. *Am J Physiol Endocrinol Metab* (2015.0) **309** E736-46. DOI: 10.1152/ajpendo.00231.2015
63. Rhee SY, Jung ES, Park HM, Jeong SJ, Kim K, Chon S. **Plasma glutamine and glutamic acid are potential biomarkers for predicting diabetic retinopathy**. *Metabolomics* (2018.0) **14** 89. DOI: 10.1007/s11306-018-1383-3
64. Zheng Y, Hu FB, Ruiz-Canela M, Clish CB, Dennis C, Salas-Salvado J. **Metabolites of Glutamate Metabolism Are Associated With Incident Cardiovascular Events in the PREDIMED PREvención con DIeta MEDiterránea (PREDIMED) Trial**. *J Am Heart Assoc* (2016.0) **5**. DOI: 10.1161/jaha.116.003755
65. Yamakado M, Tanaka T, Nagao K, Ishizaka Y, Mitushima T, Tani M. **Plasma amino acid profile is associated with visceral fat accumulation in obese Japanese subjects**. *Clin Obes* (2012.0) **2** 29-40. DOI: 10.1111/j.1758-8111.2012.00039.x
66. Takashina C, Tsujino I, Watanabe T, Sakaue S, Ikeda D, Yamada A. **Associations among the plasma amino acid profile, obesity, and glucose metabolism in Japanese adults with normal glucose tolerance**. *Nutr Metab* (2016.0) **13** 5. DOI: 10.1186/s12986-015-0059-5
67. Grembecka M. **Natural sweeteners in a human diet**. *Rocz Panstw Zakl Hig* (2015.0) **66** 195-202. PMID: 26400114
68. Lenhart A, Chey WD. **A systematic review of the effects of polyols on gastrointestinal health and irritable bowel syndrome**. *Adv Nutr* (2017.0) **8** 587-96. DOI: 10.3945/an.117.015560
69. Crivelli JJ, Mitchell T, Knight J, Wood KD, Assimos DG, Holmes RP. **Contribution of dietary oxalate and oxalate precursors to urinary oxalate excretion**. *Nutrients* (2020.0) **13**. DOI: 10.3390/nu13010062
70. Murad H, Shamban AT, Premo PS. **The use of glycolic acid as a peeling agent**. *Dermatol Clin* (1995.0) **13** 285-307. PMID: 7600706
71. Schlecht I, Gronwald W, Behrens G, Baumeister SE, Hertel J, Hochrein J. **Visceral adipose tissue but not subcutaneous adipose tissue is associated with urine and serum metabolites**. *PLoS One* (2017.0) **12** e0175133. DOI: 10.1371/journal.pone.0175133
72. Akram M. **Citric acid cycle and role of its intermediates in metabolism**. *Cell Biochem Biophys* (2014.0) **68** 475-8. DOI: 10.1007/s12013-013-9750-1
73. Scalbert A, Brennan L, Manach C, Andres-Lacueva C, Dragsted LO, Draper J. **The food metabolome: a window over dietary exposure**. *Am J Clin Nutr* (2014.0) **99** 1286-308. DOI: 10.3945/ajcn.113.076133
74. Tsuruoka M, Hara J, Hirayama A, Sugimoto M, Soga T, Shankle WR. **Capillary electrophoresis-mass spectrometry-based metabolome analysis of serum and saliva from neurodegenerative dementia patients**. *Electrophoresis* (2013.0) **34** 2865-72. DOI: 10.1002/elps.201300019
75. He S, Granot-Hershkovitz E, Zhang Y, Bressler J, Tarraf W, Yu B. **Blood metabolites predicting mild cognitive impairment in the study of Latinos-investigation of neurocognitive aging (HCHS/SOL)**. *Alzheimers Dement (Amst)* (2022.0) **14** e12259. DOI: 10.1002/dad2.12259
76. Bahado-Singh RO, Syngelaki A, Akolekar R, Mandal R, Bjondahl TC, Han B. **Validation of metabolomic models for prediction of early-onset preeclampsia**. *Am J Obstet Gynecol* (2015.0) **213** 530 e1-e10. DOI: 10.1016/j.ajog.2015.06.044
77. Feart C, Samieri C, Rondeau V, Amieva H, Portet F, Dartigues JF. **Adherence to a Mediterranean diet, cognitive decline, and risk of dementia**. *JAMA* (2009.0) **302** 638-48. DOI: 10.1001/jama.2009.1146
|
---
title: 'The association between menstrual cycle characteristics and cardiometabolic
outcomes in later life: a retrospective matched cohort study of 704,743 women from
the UK'
authors:
- Kelvin Okoth
- William Parry Smith
- G. Neil Thomas
- Krishnarajah Nirantharakumar
- Nicola J. Adderley
journal: BMC Medicine
year: 2023
pmcid: PMC10029324
doi: 10.1186/s12916-023-02794-x
license: CC BY 4.0
---
# The association between menstrual cycle characteristics and cardiometabolic outcomes in later life: a retrospective matched cohort study of 704,743 women from the UK
## Abstract
### Background
Female reproductive factors are gaining prominence as factors that enhance cardiovascular disease (CVD) risk; nonetheless, menstrual cycle characteristics are under-recognized as a factor associated with CVD. Additionally, there is limited data from the UK pertaining to menstrual cycle characteristics and CVD risk.
### Methods
A UK retrospective cohort study (1995–2021) using data from a nationwide database (The Health Improvement Network). Women aged 18–40 years at index date were included. 252,325 women with history of abnormal menstruation were matched with up to two controls. Two exposures were examined: regularity and frequency of menstrual cycles; participants were assigned accordingly to one of two separate cohorts. The primary outcome was composite cardiovascular disease (CVD). Secondary outcomes were ischemic heart disease (IHD), cerebrovascular disease, heart failure (HF), hypertension, and type 2 diabetes mellitus (T2DM). Cox proportional hazards regression models were used to derive adjusted hazard ratios (aHR) of cardiometabolic outcomes in women in the exposed groups compared matched controls.
### Results
During 26 years of follow-up, 20,605 cardiometabolic events occurred in 704,743 patients. Compared to women with regular menstrual cycles, the aHRs ($95\%$ CI) for cardiometabolic outcomes in women with irregular menstrual cycles were as follows: composite CVD 1.08 ($95\%$ CI 1.00–1.19), IHD 1.18 (1.01–1.37), cerebrovascular disease 1.04 (0.92–1.17), HF 1.30 (1.02–1.65), hypertension 1.07 (1.03–1.11), T2DM 1.37 (1.29–1.45). The aHR comparing frequent or infrequent menstrual cycles to menstrual cycles of normal frequency were as follows: composite CVD 1.24 (1.02–1.52), IHD 1.13 (0.81–1.57), cerebrovascular disease 1.43 (1.10–1.87), HF 0.99 (0.57–1.75), hypertension 1.31 (1.21–1.43), T2DM 1.74 (1.52–1.98).
### Conclusions
History of either menstrual cycle irregularity or frequent or infrequent cycles were associated with an increased risk of cardiometabolic outcomes in later life. Menstrual history may be a useful tool in identifying women eligible for periodic assessment of their cardiometabolic health.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12916-023-02794-x.
## Background
Cardiovascular disease (CVD) is a major public health burden and remains the leading cause of mortality in women accounting for $35\%$ of the total deaths worldwide based on estimates from the global burden of disease study [1, 2]. Recent literature reviews and consensus statements from professional societies in the US and Europe have highlighted the association between female reproductive factors and risk of CVD in later life [3–6]. However, menstrual cycle history and its relation to CVD was not included despite evidence of its association with CVD risk [7–9].
The menstrual life course begins at menarche and ends at menopause. The regulation of menstrual cycles involves an intricate balance between hypothalamic, pituitary, and gonadal axis hormones. A disruption of this balance may result in changes in menstrual characteristics that may affect one or more of four menstrual cycle domains: frequency, regularity, duration, or volume of flow [10]. The years immediately after menarche and the menopausal transition period are characterized by irregular and unstable menstrual cycles [11]. When menstrual cycles are stable, a typical menstrual period will last for 3 to 5 days, while the average menstrual cycle will last for 28 days (range 21–35 days) [11]. Long or irregular menstrual cycles are associated with cardiovascular risk factors including hyperinsulinemia and dyslipidemia, hypertension, and diabetes mellitus [12, 13]. The American College of Obstetricians and Gynaecologists recommends the inclusion of menstrual cycle history as a vital sign to improve the timely identification of potential adverse health outcomes in later life [14]. However, the management of abnormal menstruation focuses primarily on addressing associated infertility challenges with other potential longer-term risks underappreciated.
The UK provides universal health care to all its residents. The first point of call for UK women with clinically significant changes in the menstrual cycle patterns will be the primary care practice. The present study will harness electronic health data from UK primary care to shed more light on the association between menstrual cycle characteristics and risk of cardiometabolic outcomes in the future.
## Study design
A population-based retrospective cohort study was conducted to evaluate the association between menstrual cycle characteristics and long-term risk of cardiometabolic outcomes. Only domains relating to the regularity and frequency of menstrual cycle were used in the present study (Additional file: Table S1) [10]. Therefore, two study cohorts were created. The first cohort was composed of women with irregular or no menstrual cycle (exposed group) and matched controls from the general population without a history of irregular menstrual cycles. The second cohort was composed of women with infrequent or frequent menstrual cycles (exposed group) and matched controls from the general population without a history of infrequent or frequent menstrual cycles. The study period was 1 January 1995 to 31 December 2021. The rates of cardiometabolic outcomes were compared in the exposed and control groups.
## Data source
IQVIA Medical Research Data (IMRD) incorporates data from The Health Improvement Network (THIN), a Cegedim database. Reference made to THIN is intended to be descriptive of the data asset licensed by IQVIA. The proposed study used de-identified data provided by patients as a part of their routine primary care. IMRD-UK (formerly THIN) is a nationwide UK-based database containing anonymized electronic health records contributed by 787 general practices. Registered practices contributing to the database are representative of the UK population [15, 16]. Participating practices collect patient data using an electronic health records software system known as the Vision software.
## Practice eligibility criteria
Practices were eligible for inclusion from the later of the date on which the practice met acceptable mortality reporting (a quality assurance standard) or 1 year after the practice began to use the Vision software system [17].
## Study population
The study population was composed of women aged 18–40 years at baseline. Participants entered the study at the latest of their 18th birthday, study start date (1 January 1995), or 1 year after joining the practice (to ensure sufficient time for recording of baseline information).
## Exposure
The coding of diagnoses and other health-related care processes in UK primary care is based on the Read code clinical terminology (computable phenotype) [18]. The exposures of interest were identified by the presence of a diagnostic Read code describing menstrual cycle irregularity or frequent or infrequent cycles as reported in primary care. Menstrual cycle characteristic self-report has been validated in other studies and is regarded as reliable [19, 20]. Where a patient had a diagnostic record for both irregularity and frequent or infrequent menstrual cycles, exposure status was assigned to the first ever recorded domain. Characteristics relating to the regularity of the menstrual cycle defined a composite exposure that included irregular cycles, amenorrhea, menometrorrhagia, and metropathia haemorrhagica. Attributes relating to menstrual cycle frequency defined a composite exposure that included too frequent (polymenorrhea, epimenorrhea) or infrequent (oligomenorrhea) cycles. Details are provided in Additional file: Table S2. Women with the exposure of interest were matched with up to two women without a record of the exposure (controls), randomly selected from a pool of eligible women. The exposed and unexposed groups were matched by age (± 1 year) and general practice. Women with a record of other menstrual related conditions including intermenstrual bleeding, menstrual disorders, and complications of duration or volume of flow were excluded from the study.
## Follow-up period
For newly (incident) diagnosed exposures (irregular cycles and frequent or infrequent cycles), the date of diagnosis served as the index date. For patients with a pre-existing record relating to complications in the regularity or frequency of menstrual cycles, the date the patient became eligible to participate in the study served as the index date. To mitigate immortal time bias, exposed patients were assigned the same index date as their corresponding controls and matched on this date [21]. Each exposed and matched control participant contributed follow-up time from the index to the exit date. The exit date was the earliest of (i) the outcome, (ii) death, (iii) study end date, and (iv) date of leaving the general practice or when the general practice stopped contributing to the database.
## Outcomes
The primary outcome was the incident diagnosis of cardiovascular disease, a composite of ischemic heart disease, heart failure, or cerebrovascular disease (stroke or transient ischemic attack). Secondary outcomes were the cardiovascular conditions separately, hypertension and type 2 diabetes mellitus. Participants with a diagnosis of the outcome of interest at baseline were excluded from the corresponding crude and adjusted regression analysis. Outcomes were identified using the relevant Read codes. The Read codes used in the present study were selected using a method comparable to that proposed by Davé and Peterson and Watson et al. [ 22, 23]. First, a list of pertinent medical terms associated with the outcomes was compiled. Using the medical terms identified in the first step, the description, and numeric fields (columns) of the Read code dictionary were searched for relevant diagnostic codes related to the outcomes of interest. Third, we compared the codes identified in the previous step with codes published in online Read code repositories (caliberresearch.org, clinicalcodes.org, Cambridge code lists index) [24–26], as well as codes published in supplementary material of existing literature [27]. Finally, we consulted with UK clinicians to determine the final set of codes to be used in the study. All the outcomes in this study are included in the United Kingdom’s Quality and Outcomes Framework (QOF), a pay-for-performance system. The QOF was established to improve chronic disease management by financially rewarding primary care practices for providing interventions associated with better health outcomes. Chronic conditions falling under the QOF domains are well documented in UK general practices. Validation studies demonstrate that the prevalence of chronic diseases in THIN databases is comparable to national estimates [15, 28, 29].
## Study covariates
The following potential confounders were included in the study: sociodemographic characteristics (age and Townsend index of deprivation), lifestyle characteristics (body mass index [BMI], smoking status, alcohol use), medical characteristics (current lipid medication, connective tissue disorders, migraine), and reproductive factors (current oral contraceptive pills use [COC], preeclampsia, gestation diabetes mellitus, pregnancy loss, pre-term delivery, polycystic ovary syndrome [PCOS], endometriosis, pelvic inflammatory disease and uterine fibroids). Age was calculated at index date. The Townsend deprivation index is a measure of material deprivation derived from census data and linked to residential area [30]. The Townsend deprivation index is computed using the following domains: unemployment as a percentage of economically active individuals aged 16 and older, car ownership as a percentage of all households, home ownership as a percentage of all households, and overcrowding. BMI was calculated as weight divided by height in meters squared and categorized using WHO criteria (< 18.5, 18.5–24.9. 24–29.9, and > 30 kg/m2) [31]. Smoking (non-smokers, current smokers, ex-smokers) and alcohol use (non-drinkers, drinkers with excess, drinker without excess, ex-drinker) were self-reported. Self-reported smoking status and self-reported alcohol use are reliably recorded in THIN database [32, 33]. Current lipid medication was defined as the prescription of lipid medication within 60 days of cohort entry. Connective tissue disorders included rheumatological diseases (systemic lupus erythematosus, polymyositis, mixed connective tissue disease, polymyalgia rheumatica, moderate to severe rheumatoid arthritis). Current combined oral contraceptive pills use was defined as contraceptive use within 1 year of cohort entry. For each of the covariates, the latest record of the variable prior to study entry was used.
## Analysis
Participant characteristics at baseline were reported using median (IQR) for continuous variables and counts (%) for categorical variables. The crude incidence rates of cardiometabolic outcomes were estimated for each exposure group. Unadjusted and adjusted Cox proportional hazard models were used to derive hazard ratios (HR) and $95\%$ confidence intervals ($95\%$ CI) for the associations between menstrual cycle characteristics (regularity or frequency) and incident cardiometabolic outcomes. In the multivariable models, adjustments were made for age, BMI, Townsend deprivation quintiles, smoking status, COC pills use, lipid-lowering drug use, alcohol use, connective tissue disorders, reproductive complications, and migraine. A separate category called missing was created for categorical data with missing data and incorporated in the regression analysis. For each model, the proportional hazards assumption was evaluated using the Schoenfeld residual test and graphical confirmation using the log–log survival curves.
## Sensitivity analysis
We performed several sensitivity analyses on the primary outcome to evaluate the robustness of our findings. Women with several reproductive characteristics, including polycystic ovary syndrome, amenorrhea, endometriosis, fibroids, and current contraceptive use, were excluded to evaluate whether these conditions drove any observed associations. We also examined, separately, the association between frequent or infrequent menstrual cycles and their relationship to cardiometabolic outcomes. Additionally, we evaluated any potential interaction between abnormal menstrual cycles (irregular and frequent or infrequent) and lifestyle characteristics (body mass index, smoking and alcohol consumption).
A two-tailed p-value of 0.05 was considered statistically significant. All analyses were conducted using Stata SE version 17.0.
## Results
Additional file: Figure S1 presents the study participants flow chart. There were 704,743 patients in the present study including 215,378 with a history of irregular menstrual cycles and 36,947 with a history of frequent or infrequent menstrual cycles (Table 1). By design, the median age of women in the exposed and unexposed groups was similar (approximately 27 years). Compared to women who had regular cycles, women with irregular menstrual cycles were more likely to be obese ($15.4\%$ versus $10.9\%$), be current smokers ($24.9\%$ versus $21.1\%$), be in the most deprived Townsend quintile ($14.8\%$ versus $13.1\%$), have migraine ($27.8\%$ versus $21.3\%$), have a current prescription for COC pills before cohort entry ($30.6\%$ versus $27.0.\%$), have a history of miscarriage ($9.2\%$ versus $6.2\%$), and have a diagnosis of polycystic ovary syndrome ($5.6\%$ versus $1.7\%$). A similar pattern in baseline differences was present in the group examining women with frequent or infrequent menstrual cycles compared to women with menstrual cycles of normal frequency. Table 1Baseline characteristics by menstrual characteristics statusCharacteristicsIrregular cycles ($$n = 215$$ 378)Regular cycles ($$n = 386$$ 825)Frequent/infrequent cycles ($$n = 36$$ 947)Normal cycle frequency ($$n = 65$$ 593)n (%)n (%)n (%)n (%)Age; median (IQR)27.5 (22.1–33.2)27.2 (22.0–32.7)27.5 (21.8–33.8)27.4 (21.9–33.4)Townsend deprivation quintile 1 (least deprived)36,877 (17.1)71,109 (18.4)6913 (18.7)12,912 (19.7) 233,111 (15.4)61,786 (16.0)5916 (16.0)10,949 (16.7) 338,970 (18.1)70,821 (18.3)6765 (18.3)12,014 (18.3) 439,909 (18.5)68,329 (17.7)6630 (17.9)11,310 (17.2) 5 (most deprived)31,911 (14.8)50,707 (13.1)5054 (13.7)8171 (12.5) Missing34,600 (16.1)64,073 (16.6)5669 (15.3)10,237 (15.6)BMI categories in kg/m2 18.5–2587,884 (40.8)157,658 (40.8)14,245 (38.6)26,492 (40.4) < 18.510,107 (4.7)15,996 (4.1)1561 (4.2)2589 (4.0) 25–3037,110 (17.2)61,591 (15.9)6233 (16.9)10,537 (16.1) > 3033,194 (15.4)41,996 (10.9)6234 (16.9)6925 (10.6) Missing47,083 (21.9)109,584 (28.3)8674 (23.5)19,050 (29.0)*Smoking status* Non-smokers121,477 (56.4)220,926 (57.1)20,922 (56.6)37,065 (56.5) Current smokers53,576 (24.9)81,706 (21.1)8629 (23.4)13,766 (21.0) Ex-smokers23,951 (11.1)38,887 (10.1)4012 (10.9)6434 (9.8) Missing16,374 (7.6)45,306 (11.7)3384 (9.2)8328 (12.7)*Alcohol status* Non-drinker40,821 (19.0)65,386 (16.9)6469 (17.5)10,345 (15.8) Drinker with excess4720 (2.2)5887 (1.5)737 (2.0)932 (1.4) Drinker no excess111,018 (51.6)192,483 (49.8)18,887 (51.1)32,934 (50.2) Ex-drinker2329 (1.1)3397 (0.9)348 (0.9)580 (0.9) Missing56,490 (26.2)119,672 (30.9)10,506 (28.4)20,802 (31.7) Current lipid medication456 (0.2)527 (0.1)87 (0.2)79 (0.1) Connective tissue disorders999 (0.5)1498 (0.4)168 (0.5)289 (0.4) Migraine59,873 (27.8)82,328 (21.3)10,550 (28.6)13,877 (21.2) Reproductive factors Current combined oral contraceptive pills65,820 (30.6)104,274 [27]10,266 (27.8)17,339 (26.4) Polycystic ovary syndrome11,970 (5.6)6448 (1.7)3755 (10.2)1017 (1.6) Pelvic inflammatory disease6283 (2.9)6941 (1.8)1100 (3.0)1193 (1.8) Endometriosis2353 (1.1)3730 (1.0)417 (1.1)639 (1.0) Fibroids698 (0.3)1268 (0.3)144 (0.4)210 (0.3) Miscarriage19,745 (9.2)23,954 (6.2)3083 [8]4225 [6] Gestational diabetes1280 (0.6)1603 (0.4)232 (0.6)253 (0.4) Pre-eclampsia812 (0.4)1081 (0.3)134 (0.4)203 (0.3) Pre-term births1452 (0.7)2403 (0.6)228 (0.6)378 (0.6)Baseline cardiovascular diseases Hypertension2631 (1.2)3181 (0.8)471 (1.3)562 (0.9) Diabetes1898 (0.9)2359 (0.6)334 (0.9)392 (0.6) Ischemic heart disease119 (0.1)123 (0.0)14 (0.0)22 (0.0) Stroke/TIA307 (0.1)379 (0.1)50 (0.1)73 (0.1) Heart failure46 (0.0)76 (0.0)12 (0.0)14 (0.0)BMI, body mass index; IQR, inter quartile range; Kg/m2, kilograms per meter square. There were no missing data for age. The total number (%) of missing data for Townsend deprivation quintile, BMI, and alcohol status were 114 579 ($16.3\%$), 184,391 ($26.2\%$), and 73,392 ($10.4\%$), respectively. For current lipid medication, connective tissue disorders, migraine, reproductive factors, and baseline cardiovascular diseases absence of a diagnostic code for these conditions was assumed to indicate absence of disease
## Menstrual cycle regularity
Between 1995 and 2021, 896 and 1056 composite CVD events were recorded among women with irregular versus regular menstrual cycles, respectively. Median (IQR) follow-up was 4.5 (1.7–9.6) years in the exposed and 3.8 (1.4–8.3) years in the unexposed group. The crude incidence rate (per 1000 years) of composite CVD was 0.67 in women with irregular menstrual cycles versus 0.50 in women with regular menstrual cycles. The HR for composite CVD comparing irregular with regular menstrual cycles were 1.26 ($95\%$ CI 1.15–1.38; $p \leq 0.001$) in the crude model and 1.08 ($95\%$ CI 1.00–1.19; $$p \leq 0.062$$) in the model adjusting for sociodemographic, lifestyle, medical, and reproductive characteristics (Figs. 1, 2, Additional file: Table S3).Fig. 1Forest plot showing the fully adjusted effect estimates and $95\%$ CI for cardiometabolic outcomes in women with history of irregular menstrual cycles or frequent or infrequent menstrual cyclesFig. 2Cumulative hazard estimates of cardiometabolic outcomes (A-F) in women with irregular cycles compared to those with regular cycles In the model comparing irregular to regular menstrual cycles, the adjusted HR for CVD subtypes were as follows: 1.18 ($95\%$ CI 1.01–1.37; $$p \leq 0.033$$) for ischemic heart disease, 1.04 ($95\%$ CI 0.92–1.17; $$p \leq 0.508$$) for cerebrovascular disease, and 1.30 ($95\%$ CI 1.02–1.65; $$p \leq 0.033$$) for heart failure (Figs. 1, 2, Additional file: Table S3).
During follow-up, the crude incidence rate (per 1000 person-years) of hypertension was 3.48 in women with irregular menstrual cycles versus 2.79 in controls with regular menstrual cycles. Compared to those with regular menstrual cycles, women with irregular menstrual cycles had a HR of subsequent hypertension of 1.19 ($95\%$ CI 1.14–1.24; $p \leq 0.001$) and 1.07 ($95\%$ CI 1.03–1.11; $$p \leq 0.001$$) in the unadjusted and adjusted models, respectively (Figs. 1, 2, and Additional file: Table S3).
The crude incidence rate (per 1000 person-years) of type 2 diabetes mellitus was 1.82 in women with irregular menstrual cycles and 1.05 in those with regular menstrual cycles. Compared to women with regular menstrual cycles, women with irregular menstrual cycles were more likely to develop type 2 diabetes mellitus in both the crude (HR 1.66; $95\%$ CI 1.34–1.49; $p \leq 0.001$) and adjusted (1.37; $95\%$ CI 1.29–1.45; $p \leq 0.001$) models (Figs. 1, 2, and Additional file Table S3).
The effect estimate for the association between menstrual cycle irregularity and composite CVD showed only minimal changes on exclusion of women with amenorrhea (aHR 1.09; $95\%$ CI 0.96–1.24; $$p \leq 0.173$$), polycystic ovary syndrome (aHR 1.09; $95\%$ CI 0.99–1.19; $$p \leq 0.080$$), endometriosis (aHR 1.09; $95\%$ CI 0.99–1.20; $$p \leq 0.068$$), fibroids (aHR 1.09; $95\%$ CI, 0.99–1.19; $$p \leq 0.067$$), or current oral contraceptive use (aHR 1.03; $95\%$ CI 0.94–1.15; $$p \leq 0.445$$) (Additional file: Table S4). Also, the effect estimate for the association between menstrual cycle irregularity and composite CVD showed only minimal changes (aHR 1.09 ($95\%$ CI, 1.00–1.20; $$p \leq 0.052$$) on excluding polycystic ovary syndrome, endometriosis, and fibroids as covariates included in the multivariable Cox proportional hazard model (Additional file: Table S5).
## Menstrual cycle frequency
During the study period, 205 versus 202 composite CVD events were recorded in women with frequent or infrequent menstrual cycles compared to controls with menstrual cycles of normal frequency, respectively. Median (IQR) follow-up was 5.1 (2.0–10.5) years in the exposed and 4.0 (1.5–8.8) years in the unexposed group. The crude incidence rate (per 1000 person years) of composite CVD was 0.83 in women with frequent or infrequent cycles compared to 0.53 in women with menstrual cycles of normal frequency with a crude HR of 1.46 ($95\%$ CI 1.20–1.78; $p \leq 0.001$) (Additional file: Table S3). In the adjusted model, the association between frequent or infrequent cycles and composite CVD was maintained (aHR 1.24, $95\%$ CI 1.02–1.52; $$p \leq 0.031$$) (Figs. 1, 3).Fig. 3Cumulative hazard estimates of cardiometabolic outcomes (A-F) in women with frequent or infrequent cycles compared to those with normal cycle frequency In the model comparing frequent or infrequent menstrual cycles to menstrual cycles of normal frequency, the adjusted HR for CVD subtypes were 1.13 ($95\%$ CI 0.81–1.57; $$p \leq 0.464$$) for ischemic heart disease, 1.43 ($95\%$ CI 1.10–1.87; $$p \leq 0.007$$) for cerebrovascular disease, and 0.99 ($95\%$ CI 0.57–1.75; $$p \leq 0.985$$) for heart failure (Figs. 1, 2, Additional file: Table S3).
The crude incidence rate (per 1000 person-years) of hypertension was 4.42 in women with frequent or infrequent menstrual cycles compared to 3.0 in women with menstrual cycles of normal frequency, with a crude HR of 1.41 ($95\%$ CI 1.30–1.54; $p \leq 0.001$). In the adjusted model, women with frequent or infrequent cycles were $32\%$ more likely to develop hypertension (HR 1.31; $95\%$ CI 1.21–1.43; $p \leq 0.001$) (Figs. 1, 3, and Additional file Table S3).
The crude incidence rate (per 1000 years) of type 2 diabetes mellitus was 2.38 in women with frequent or infrequent menstrual cycles versus 1.02 in women with menstrual cycles of normal frequency. Women with frequent or infrequent cycles were twice as likely to develop type 2 diabetes mellitus compared to women with menstrual cycles of normal frequency (crude HR 2.25; $95\%$ CI 1.96–2.53; $p \leq 0.001$). The association was maintained in the adjusted model (HR 1.74; $95\%$ CI 1.52–1.98; $p \leq 0.001$) (Figs. 1, 3, Additional file: Table S3).
The association between frequent or infrequent cycles and composite CVD was no longer maintained on exclusion of women with amenorrhea (aHR 1.18; $95\%$ CI 0.95–1.47; $$p \leq 0.130$$) and on current oral contraceptive use (aHR 1.14; $95\%$ CI 0.91–1.42; $$p \leq 0.259$$). The association between frequent or infrequent cycles and risk of composite CVD was sustained on exclusion of women with history of polycystic ovary syndrome (aHR 1.23; $95\%$ CI 1.01–1.51; $$p \leq 0.043$$), endometriosis (aHR 1.24; $95\%$ CI 1.02–1.52; $$p \leq 0.035$$), or uterine fibroids (aHR 1.26; $95\%$ CI 1.03–1.54; $$p \leq 0.025$$) (Additional file: Table S6). The effect estimate for composite CVD for the association between frequent or infrequent cycles and composite CVD was not materially affected (aHR 1.28 $95\%$ CI, 1.05–1.55 $$p \leq 0.016$$)) on exclusion of polycystic ovary syndrome, endometriosis, and fibroids as covariates included in adjusted Cox proportional hazard model (Additional file: Table S5). In the analysis comparing frequent menstrual cycles to menstrual cycles of normal frequency, the adjusted HRs for cardiometabolic outcomes were as follows: 1.42 ($95\%$ CI 1.09–1.85; $$p \leq 0.009$$) for composite CVD, 1.13 ($95\%$ CI 0.74–1.72; $$p \leq 0.570$$) for IHD, 1.88 ($95\%$ CI 1.33–2.67; $p \leq 0.001$) for cerebrovascular disease, 0.93 ($95\%$ CI, 0.42–2.06; $$p \leq 0.858$$) for heart failure, 1.37 ($95\%$ CI 1.22–1.54; $p \leq 0.001$) for hypertension, and 1.37 ($95\%$ CI 1.13–1.65; $p \leq 0.001$) for type 2 diabetes mellitus (Additional file: Table S7). For the analysis examining infrequent menstrual cycles versus menstrual cycles of normal frequency, the adjusted HRs for cardiometabolic outcomes were 1.06 ($95\%$ CI 0.78–1.45; $$p \leq 0.704$$) for composite CVD, 1.16 ($95\%$ CI 0.68–1.97; $$p \leq 0.582$$) for IHD, 1.01 ($95\%$ CI 0.66–1.53; $$p \leq 0.980$$) for cerebrovascular disease, 1.13 ($95\%$ CI 0.50–2.54; $$p \leq 0.770$$) for heart failure, and 1.24 ($95\%$ CI 1.85–2.72; $p \leq 0.001$) for type 2 diabetes mellitus (Additional file: Table S7).
There was no evidence of interaction between cycle dysfunction (irregular and frequent or frequent) and lifestyle characteristics including BMI, smoking, and alcohol consumption (Additional file: Figure S2).
## Main findings
In this nationwide cohort study of more than 700 thousand women from the UK, history of both irregular menstrual cycles and frequent or infrequent menstrual cycles were associated with an increased risk of several cardiometabolic outcomes. The associations were strongest for women with abnormal patterns in the frequency of their menstrual cycles with frequent or infrequent cycles being associated with a significant increase in hazard of composite CVD. History of menstrual cycle irregularity was associated with a borderline increase in the hazard of composite CVD. On examination by subtypes of CVD, menstrual cycle irregularity was associated with an increased risk of ischemic heart disease and heart failure but not cerebrovascular disease. Frequent or infrequent cycles were associated with an increased risk of cerebrovascular disease but not ischemic heart disease or heart failure. On examination by subtype of menstrual cycle frequency, frequent menstrual cycles were associated with an elevated risk of composite CVD and cerebrovascular disease but not ischemic heart disease or heart failure. No association was observed between infrequent menstrual cycles and composite CVD or any of the CVD subtypes. Both irregular menstrual cycles and frequent or infrequent cycles were linked with an increased risk of hypertension and type 2 diabetes mellitus.
## Comparison with previous literature
A summary of the study characteristics of selected existing literature are provided in Additional file: Table S8 [8, 9, 34–40]. Overall, results from our study support and expand existing literature that have examined the association between menstrual characteristics and cardiometabolic outcomes. A UK prospective cohort study of 40,896 premenopausal women aged 50 years and below at baseline examined the association between irregular menstrual cycles and risk of fatal and non-fatal cardiovascular disease [34]. During a median duration of follow-up of 6.9 years (IQR: 6.2 to 7.6), no relationship was found between irregular menstrual cycles compared to regular menstrual cycles and risk of fatal and non-fatal CVD outcomes (Additional file: Table S8). Three United States (US) prospective cohort studies that were all conducted by Wang et al. examined the relationship between menstrual cycle characteristics and cardiometabolic outcomes (CVD and diabetes mellitus) in a cohort of female nurses [8, 9, 35]. The studies by Wang et al. typically defined menstrual cycle regularity as very regular, regular, usually irregular, and always irregular or no period, while cycle length was defined as ≤ 25 days, 26–31 days, 32–39 days, and ≥ 40 days. The most recent study by Wang et al. [ 9] followed up 80,630 women for a period of 24 years to examine relationship between menstrual cycle regularity and risk of CVD (fatal and non-fatal). Compared to women who had very regular cycles at ages 14 to 17 years, 18 to 22 years, and 29 to 46 years, women with always irregular cycles or no periods at ages 18 to 22 and 29–46 age groups were at an elevated risk of CVD in later life (Additional file: Table S8). In the second prospective cohort study [8], Wang and colleagues followed up 79,505 premenopausal women for a period of 24 years to evaluate the association between menstrual cycle characteristics and risk of premature mortality. Always irregular cycle or no period at ages 18–46 was associated with mortality from CVD (Additional file: Table S8.) [ 35]. Wang et al. followed 75,456 participants followed for a period of 24 years to investigate the association between menstrual cycle characteristics and type 2 diabetes mellitus. Both irregular menstrual cycles and menstrual cycle length of ≥ 40 days were associated with an elevated risk of type 2 diabetes mellitus (Additional file: Table S8). A recent Australian study of 13,714 participants investigated the relationship between irregular menstrual cycles compared to regular menstrual cycles (never, sometimes, or rarely) and risk of non-fatal heart disease (myocardial infarction, angina) [36]. During the 20-year period of follow-up, irregular menstrual cycles compared to regular menstrual cycles were linked to higher risk of heart disease and diabetes mellitus (Additional file: Table S8). We observed a relationship between frequent (short) but not infrequent (long) menstrual cycles and CVD. This contrasted with a US cohort study which found an association between long menstrual cycles but not short menstrual cycles and CVD [9]. In the US study, compared to women with a menstrual cycle length of 26–31 days, the adjusted HRs for cardiovascular disease were as follows: 1.00 ($95\%$ CI 0.83–1.20) for cycle lengths of ≤ 25 days, 1.05 ($95\%$ CI 0.89–1.24) for cycle lengths of 32–39 days, and 1.30 ($95\%$ CI 1.09–1.57) for cycle lengths of ≥ 40 days or too irregular to estimate (Additional file: Table S8) [9]. In our study, menstrual cycle frequency was classified as either frequent or infrequent. Our study could not differentiate menstrual cycle frequency by cycle length in days. Therefore, our results should be interpreted with caution given that the relationship between infrequent (long) menstrual cycles and CVD appears to be greatest with increasing length (≥ 40 days) in menstrual cycles, as suggested by the findings from the US study [9]. The association between frequent (short) menstrual cycles and elevated CVD risk is biologically plausible. Frequent menstrual cycles are a marker of diminished ovarian reserve [41]. Previous studies have reported a relationship between diminished ovarian reserve and elevated CVD risk [42, 43].
An Iranian study followed up 2128 women aged 18–49 years at baseline to investigate the association between irregular menstrual cycles compared regular menstrual cycles and risk of cardiometabolic outcomes [37]. During the 15-year period of follow-up, irregular menstrual cycles compared to regular menstrual cycles were associated with higher risk of type 2 diabetes mellitus but not hypertension. The present study found that irregular menstrual cycles were associated with an increased risk of heart failure. However, due to the low number of events, we did not find any association between changes in menstrual cycle frequency and heart failure risk. Direct comparisons between the existing literature and the present study are challenging due to several differences which may partly explain some of the contrasting findings. The main methodological differences relate to the stratification of irregular menstrual cycles by severity into four categories (US studies) [8, 9, 35]: case definition of the exposure to include both regularity and frequency of menses as a single exposure (Iranian study) [37]; case definition of the unexposed group (regular menstrual cycles) as never, rarely, or sometimes (Australian study) [36]; restriction of the study participants exclusively to nurses (US studies) [8, 9, 35]; and inclusion of fatal CVD events in the outcomes (US and UK studies) [9, 34].
## Biological plausibility
Several mechanisms yet to be fully elucidated are suspected to play a role in the association between menstrual cycle characteristics and elevated risk of cardiometabolic outcomes. First, PCOS which a common cause of amenorrhea, irregular menstrual cycles, and oligomenorrhea is characterized by cardiovascular risk factors including metabolic syndrome, obesity, insulin resistance, dyslipidemia, and hypertension [44]. The present study found that the association between menstrual complications and cardiometabolic outcomes was independent of PCOS. That PCOS is associated with an increased risk of CVD is debatable. Some studies report an increased risk of CVD among women with PCOS, while other studies argue that any observed association is minimal or restricted to severe phenotypes of PCOS [45]. Second, other reproductive factors (endometriosis, fibroids) associated with changes in menstrual characteristics and linked to adverse cardiometabolic health may partly account for the observed association [46, 47]. However, exclusion of women with a record for endometriosis or fibroids in sensitivity analyses did not alter the observed effect estimates. Attenuation of the effect size on exclusion of women on current prescription for combined oral contraceptive (COC) suggests that increased CVD risk may be partly mediated by COC use [48]. Third, changes in menstrual cycle characteristics are strongly linked to hyperinsulinemia. Hyperinsulinemia suppresses the production of sex hormone-binding globulin resulting in elevated level of free testosterone. This hormonal environment is associated with higher risk of cardiometabolic outcomes [49–52]. Fourth, estrogen modulates vascular inflammation [53, 54]. Abnormal menstrual patterns may favor pro-inflammatory process which may result in atherosclerotic CVD. Fifth, differences in mechanistic pathways between menstrual cycle characteristics may partly account for the differences in findings. A longer cycle length may be indicative of fewer ovulations and, consequently, lower mean estrogen levels [39]. Higher levels of endogenous estradiol before menopause have been associated with a decreased risk of subclinical atherosclerosis after menopause [55]. Short menstrual cycle length may be an indicator of ovarian aging [41, 56]. Markers of diminished ovarian reserve including anti-Müllerian hormone (AMH) and elevated follicle stimulating hormone (FSH) have been associated with CVD risk factors [57, 58]. In addition, low AMH levels may act independently to promote atherogenesis [42, 59].
## Strengths and limitations
The main strength of the present study is the use of a large sample size that is representative of the UK population and a long duration of follow-up that allows sufficient time for the development of cardiometabolic outcomes. Unlike previous studies that relied on self-reports of the exposure several years after their occurrence, the present study relied on electronic health data documented at point of clinical consultation which helped to minimize recall bias. In addition, we adjusted for several key sociodemographic, lifestyle, medical, and reproductive characteristics. Several limitations should be acknowledged. Foremost, we could not characterize menstrual cycle characteristics by grades of severity or duration as this information was not coded in UK electronic health records. Second, the exposure of interest relies on self-report and is therefore susceptible to misclassification. Third, although we adjusted for several known and potential confounders, the possibility of unmeasured confounding remains; for instance, we were not able to adjust for dietary habits, physical activity, or family history of CVD as this information is not well recorded in UK primary care data. Fourth, where a patient had a diagnostic code for both irregularity and frequent or infrequent cycles, exposure status was assigned as the first ever recorded menstrual cycle characteristic domain. This makes the implicit assumption that the order in which these conditions are recorded is random; however, this may not be the case. Nevertheless, given that participants impacted by any potential classification bias will have had both menstrual cycle characteristics (and could therefore contribute to either exposure definition), and the direction of effect for most outcomes was similar for the two exposures, we expect this to have a limited impact on the findings. Fifth, there is potential for exposure misclassification among women who were included in the unexposed cohort but had abnormal menstrual cycle characteristics not recorded in primary care or who were on hormonal contraceptives. Also, we did not exclude women with abnormal cycle characteristic shortly after pregnancy or during lactation. Although history of breastfeeding compared to no breastfeeding is associated with reduced maternal risk CVD, hypertension, and diabetes mellitus [60, 61], we were not able to adjust for history of breastfeeding in the analysis. Sixth, a further drawback is the possibility for selection bias due to differential loss to follow-up between exposure groups: $40.7\%$ of women in the exposed groups and $46.0\%$ in the unexposed groups were lost to follow-up due to leaving the general practice.
## Implications for public health and research
Findings from the present study support calls for the inclusion of menstrual cycle history as an additional vital sign in the assessment of the overall health status of young women. Specifically, abnormal menstruation may act as a window into the future cardiometabolic health of women. Therefore, women with history of irregular menstrual cycles or frequent or infrequent menstrual cycles may benefit from periodic evaluation of their cardiometabolic health. Current UK guidelines should consider incorporating reproductive factors including menstrual cycle characteristics as risk enhancing factors for cardiometabolic disease given the low awareness about these factors among UK physicians [62]. Future research should determine the pathophysiological mechanisms linking menstrual cycle complications and adverse cardiometabolic health and the factors behind the differential impact of different menstrual cycle characteristics and poor cardiometabolic outcomes.
## Conclusions
History of irregular menstrual cycles or frequent or infrequent menstrual cycles is associated with increased risk of cardiometabolic outcomes in later-life. Research is needed to unravel the pathophysiological links behind changes in menstrual cycle and adverse cardiometabolic health. Incorporating reproductive history including menstrual cycle characteristics as part of routine medical evaluation may help identify potential candidates for periodic assessment of cardiometabolic health.
## Supplementary Information
Additional file1: Tables S1-S8 and Figures S1-S2. Table S1. Classification of menstrual cycle characteristics. Table S2. Diagnostic Read codes for menstrual cycle regularity and menstrual cycle frequency. Table S3. Incidence rates and hazard ratios of cardiometabolic outcomes. Table S4. Sensitivity analyses menstrual cycle regularity and composite CVD analyses (excluding women with amenorrhea, polycystic ovary syndrome, endometriosis, current hormonal contraceptive use, uterine fibroids). Table S5. Incidence rates and hazard ratio for cardiometabolic outcomes (sensitivity analyses excluding polycystic ovarian syndrome, endometriosis, and fibroids as covariates from adjusted Cox proportional hazard model). Table S6. Sensitivity analyses menstrual cycle frequency and composite CVD (excluding women with amenorrhea, polycystic ovary syndrome, endometriosis, current hormonal contraceptive use, uterine fibroids). Table S7. Incidence rates and hazard ratios for cardiometabolic outcomes among women with frequent (short) menstrual cycles and infrequent (Long) menstrual cycle. Table S8. Summary of selected existing literature examining the association between menstrual characteristics and cardiometabolic outcomes. Figure S1. Study participant flow chart. Figure S2. Interaction between (A) irregular menstrual cycles and (B) frequent or infrequent menstrual cycles and lifestyle factors (Body mass index, smoking, and alcohol use).
## References
1. Vogel B, Acevedo M, Appelman Y, BaireyMerz CN, Chieffo A, Figtree GA. **The Lancet women and cardiovascular disease Commission: reducing the global burden by 2030**. *Lancet* (2021.0) **397** 2385-2438. DOI: 10.1016/S0140-6736(21)00684-X
2. 2.VizHub - GBD Results. Available from: https://vizhub.healthdata.org/gbd-results/
3. Maas AHEM, Rosano G, Cifkova R, Chieffo A, van Dijken D, Hamoda H. **Cardiovascular health after menopause transition, pregnancy disorders, and other gynaecologic conditions: a consensus document from European cardiologists, gynaecologists, and endocrinologists**. *Eur Heart J* (2021.0) **42** 967-984. DOI: 10.1093/eurheartj/ehaa1044
4. Parikh NI, Gonzalez JM, Anderson CAM, Judd SE, Rexrode KM, Hlatky MA. **Adverse pregnancy outcomes and cardiovascular disease risk: unique opportunities for cardiovascular disease prevention in women: a scientific statement from the American Heart Association**. *Circulation* (2021.0) **143** E902-E916. DOI: 10.1161/CIR.0000000000000961
5. Agarwala A, Michos ED, Samad Z, Ballantyne CM, Virani SS. **The use of sex-specific factors in the assessment of women’s cardiovascular risk**. *Circulation* (2020.0) **141** 592-599. DOI: 10.1161/CIRCULATIONAHA.119.043429
6. 6.Okoth K, Chandan JS, Marshall T, Thangaratinam S, Thomas GN, Nirantharakumar K, et al. Association between the Reproductive Health of young women and cardiovascular disease in later life: Umbrella Review. BMJ. 2020;:m3502
7. Wang ET, Cirillo PM, Vittinghoff E, Bibbins-Domingo K, Cohn BA, Cedars MI. **Menstrual irregularity and cardiovascular mortality**. *J Clin Endocrinol Metab* (2011.0) **96** E114-8. DOI: 10.1210/jc.2010-1709
8. Wang YX, Arvizu M, Rich-Edwards JW, Stuart JJ, Manson JAE, Missmer SA. **Menstrual cycle regularity and length across the reproductive lifespan and risk of premature mortality: prospective cohort study**. *BMJ* (2020.0) **371** m3464. DOI: 10.1136/bmj.m3464
9. Wang YX, Stuart JJ, Rich-Edwards JW, Missmer SA, Rexrode KM, Farland LV. **Menstrual cycle regularity and length across the reproductive lifespan and risk of cardiovascular disease**. *JAMA Netw Open* (2022.0) **5** e2238513-e2238513. DOI: 10.1001/jamanetworkopen.2022.38513
10. Munro MG, Critchley HOD, Fraser IS, Haththotuwa R, Kriplani A, Bahamondes L. **The two FIGO systems for normal and abnormal uterine bleeding symptoms and classification of causes of abnormal uterine bleeding in the reproductive years: 2018 revisions**. *Int J Gynecol Obstet* (2018.0) **143** 393-408. DOI: 10.1002/ijgo.12666
11. 11.Laven, Joop SE; Lambalk NB. The normal and abnormal menstrual cycle. In: Text book of Obstetrics and Gynaecology, A life course approach. Bohn stafleu van loghum; 2019. p. 231–62.
12. Weiss DJ, Charles MA, Dunaif A, Prior DE, Lillioja S, Knowler WC. **Hyperinsulinemia is associated with menstrual irregularity and altered serum androgens in Pima Indian women**. *Metabolism* (1994.0) **43** 803-807. DOI: 10.1016/0026-0495(94)90257-7
13. Azevedo GD, Duarte JMBP, Souza MO, Costa-E-Silva TDN, Soares EMM, Maranhão TMO. **Menstrual cycle irregularity as a marker of cardiovascular risk factors at postmenopausal years**. *Arq Bras Endocrinol Metabol* (2006.0) **50** 876-883. DOI: 10.1590/S0004-27302006000500008
14. 14.(ACOG) TAC of O and G. ACOG Committee Opinion No. 651: menstruation in girls and adolescents: using the menstrual cycle as a vital sign. Obstet Gynecol. 2015;126(6):e143–6.
15. Blak BT, Thompson M, Dattani H, Bourke A. **Generalisability of The Health Improvement Network (THIN) database: demographics, chronic disease prevalence and mortality rates**. *Inform Prim Care* (2011.0) **19** 251-255. PMID: 22828580
16. 16.THIN : The Health Improvement Network - Cegedim Health Data. Available from: https://www.cegedim-health-data.com/cegedim-health-data/thin-the-health-improvement-network/
17. Maguire A, Blak BT, Thompson M. **The importance of defining periods of complete mortality reporting for research using automated data from primary care**. *Pharmacoepidemiol Drug Saf* (2009.0) **18** 76-83. DOI: 10.1002/pds.1688
18. 18.Booth Nick (Centre for coding and classification, Woodgate, Loughborough LL 2TG. What are the Read codes. Health Libr Rev. 1994;11(3):177–82.
19. Real FG, Svanes C, Omenaas ER, Antò JM, Plana E, Janson C. **Menstrual irregularity and asthma and lung function**. *J Allergy Clin Immunol* (2007.0) **120** 557-564. DOI: 10.1016/j.jaci.2007.04.041
20. Jukic AMZ, Weinberg CR, Wilcox AJ, McConnaughey DR, Hornsby P, Baird DD. **Accuracy of reporting of menstrual cycle length**. *Am J Epidemiol* (2008.0) **167** 25-33. DOI: 10.1093/aje/kwm265
21. 21.Yadav K, Lewis RJ. Immortal time bias in observational studies. Vol. 325, JAMA - Journal of the American Medical Association. American Medical Association; 2021. p. 686–7.
22. Davé S, Petersen I. **Creating medical and drug code lists to identify cases in primary care databases**. *Pharmacoepidemiol Drug Saf* (2009.0) **18** 704-707. DOI: 10.1002/pds.1770
23. Watson J, Nicholson BD, Hamilton W, Price S. **Identifying clinical features in primary care electronic health record studies: methods for codelist development**. *BMJ Open* (2017.0) **7** e019637. DOI: 10.1136/bmjopen-2017-019637
24. 24.caliberresearch.org. CALIBER. Available from: https://caliberresearch.org/portal/codelists
25. 25.clinicalcodes.org. ClinicalCodes Repository. Available from: https://clinicalcodes.rss.mhs.man.ac.uk/
26. 26.phpc.cam.ac.uk. Code Lists - Primary Care Unit. Available from: https://www.phpc.cam.ac.uk/pcu/research/research-groups/crmh/cprd_cam/codelists/
27. Wang X, Mobley AR, Tica O, Okoth K, Ghosh RE, Myles P. **Systematic approach to outcome assessment from coded electronic healthcare records in the DaRe2THINK NHS-embedded randomized trial**. *Eur Hear J - Digit Heal* (2022.0) **3** 426-436. DOI: 10.1093/ehjdh/ztac046
28. Ruigómez A, Martín-Merino E, García Rodríguez LA. **Validation of ischemic cerebrovascular diagnoses in the health improvement network (THIN)**. *Pharmacoepidemiol Drug Saf* (2010.0) **19** 579-585. DOI: 10.1002/pds.1919
29. Blak B, Thompson M. **Php88 how does the health improvement network (thin) data on prevalence of chronic diseases compare with national figures?**. *Value Heal* (2009.0) **12** A253. DOI: 10.1016/S1098-3015(10)74240-6
30. Townsend P, Phillimore P, Beattie A. *Health and deprivation: Inequality and the North* (1988.0)
31. 31.Consultation WHO. Obesity: preventing and managing the global epidemic. Report of a WHO consultation. World Heal Organ - Tech Rep Ser. 2000;894.
32. Marston L, Carpenter JR, Walters KR, Morris RW, Nazareth I, White IR. **Smoker, ex-smoker or non-smoker? The validity of routinely recorded smoking status in UK primary care: a cross-sectional study**. *BMJ Open* (2014.0) **4** e004958. DOI: 10.1136/bmjopen-2014-004958
33. Khadjesari Z, Marston L, Petersen I, Nazareth I, Walters K. **Alcohol consumption screening of newly-registered patients in primary care: a cross-sectional analysis**. *Br J Gen Pract* (2013.0) **63** e706. DOI: 10.3399/bjgp13X673720
34. Iliodromiti S, Nelson SM. **Irregular menstrual cycles are not associated with cardiovascular disease; a cohort study of 40,896 women**. *Fertil Steril* (2017.0) **108** e250. DOI: 10.1016/j.fertnstert.2017.07.750
35. Wang YX, Shan Z, Arvizu M, Pan A, Manson JAE, Missmer SA. **Associations of menstrual cycle characteristics across the reproductive life span and lifestyle factors with risk of type 2 diabetes**. *JAMA Netw Open* (2020.0) **3** e2027928-e2027928. DOI: 10.1001/jamanetworkopen.2020.27928
36. Kiconco S, Teede HJ, Earnest A, Loxton D, Joham AE. **Menstrual cycle regularity as a predictor for heart disease and diabetes: findings from a large population-based longitudinal cohort study**. *Clin Endocrinol (Oxf)* (2022.0) **96** 605-616. DOI: 10.1111/cen.14640
37. Dovom MR, Tehrani FR, Djalalinia S, Cheraghi L, Gandavani SB, Azizi F. **Menstrual cycle irregularity and metabolic disorders: a population-based prospective study**. *PLoS One* (2016.0) **11** e0168402. DOI: 10.1371/journal.pone.0168402
38. Solomon CG, Hu FB, Dunaif A, Rich-Edwards JE, Stampfer MJ, Willett WC. **Menstrual cycle irregularity and risk for future cardiovascular disease**. *J Clin Endocrinol Metab* (2002.0) **87** 2013-2017. DOI: 10.1210/jcem.87.5.8471
39. Solomon CG, Hu FB, Dunaif A, Rich-Edwards J, Willett WC, Hunter DJ. **Long or highly irregular menstrual cycles as a marker for risk of type 2 diabetes mellitus**. *J Am Med Assoc* (2001.0) **286** 2421-2426. DOI: 10.1001/jama.286.19.2421
40. Gast GCM, Grobbee DE, Smit HA, Bueno-De-Mesquita HB, Samsioe GN, Van Der Schouw YT. **Menstrual cycle characteristics and risk of coronary heart disease and type 2 diabetes**. *Fertil Steril* (2010.0) **94** 2379-2381. DOI: 10.1016/j.fertnstert.2010.03.044
41. Younis JS, Iskander R, Fauser BCJM, Izhaki I. **Does an association exist between menstrual cycle length within the normal range and ovarian reserve biomarkers during the reproductive years? A systematic review and meta-analysis**. *Hum Reprod Update* (2020.0) **26** 904-928. DOI: 10.1093/humupd/dmaa013
42. De Kat AC, Monique Verschuren W, Eijkemans MJC, Broekmans FJM, Van Der Schouw YT. **Anti-Müllerian hormone trajectories are associated with cardiovascular disease in women: results from the Doetinchem Cohort Study**. *Circulation* (2017.0) **135** 556-565. DOI: 10.1161/CIRCULATIONAHA.116.025968
43. Bleil ME, Gregorich SE, McConnell D, Rosen MP, Cedars MI. **Does accelerated reproductive aging underlie premenopausal risk for cardiovascular disease?**. *Menopause* (2013.0) **20** 1139-1146. DOI: 10.1097/GME.0b013e31828950fa
44. Osibogun O, Ogunmoroti O, Michos ED. **Polycystic ovary syndrome and cardiometabolic risk: opportunities for cardiovascular disease prevention**. *Trends Cardiovasc Med* (2020.0) **30** 399-404. DOI: 10.1016/j.tcm.2019.08.010
45. 45.Carmina E, Lobo RA. Is there really increased cardiovascular morbidity in women with polycystic ovary syndrome? https://home.liebertpub.com/jwh. 2018;27(11):1385–8.
46. Okoth K, Wang J, Zemedikun D, Thomas GN, Nirantharakumar K, Adderley NJ. **Risk of cardiovascular outcomes among women with endometriosis in the United Kingdom: a retrospective matched cohort study**. *BJOG* (2021.0) **128** 1598-1609. DOI: 10.1111/1471-0528.16692
47. Haan YC, Diemer FS, Van Der Woude L, Van Montfrans GA, Oehlers GP, Brewster LM. **The risk of hypertension and cardiovascular disease in women with uterine fibroids**. *J Clin Hypertens* (2018.0) **20** 718. DOI: 10.1111/jch.13253
48. Shufelt CL, BaireyMerz CN. **Contraceptive hormone use and cardiovascular disease**. *J Am Coll Cardiol* (2009.0) **53** 221. DOI: 10.1016/j.jacc.2008.09.042
49. Plymate SR, Matej LA, Jones RE, Friedl KE. **Inhibition of sex hormone-binding globulin production in the human hepatoma (Hep G2) cell line by insulin and prolactin**. *J Clin Endocrinol Metab* (1988.0) **67** 460-464. DOI: 10.1210/jcem-67-3-460
50. Rasmussen JJ, Selmer C, Frøssing S, Schou M, Faber J, Torp-Pedersen C. **Endogenous testosterone levels are associated with risk of type 2 diabetes in women without established comorbidity**. *J Endocr Soc* (2020.0) **4** bvaa050. DOI: 10.1210/jendso/bvaa050
51. Muka T, Nano J, Jaspers L, Meun C, Bramer WM, Hofman A. **Associations of steroid sex hormones and sex hormone-binding globulin with the risk of type 2 diabetes in women: a population-based cohort study and meta-analysis**. *Diabetes* (2017.0) **66** 577-586. DOI: 10.2337/db16-0473
52. Wei S, Schmidt MD, Dwyer T, Norman RJ, Venn AJ. **Obesity and menstrual irregularity: associations with SHBG, testosterone, and insulin**. *Obesity* (2009.0) **17** 1070-1076. DOI: 10.1038/oby.2008.641
53. Straub RH. **The complex role of estrogens in inflammation**. *Endocr Rev* (2007.0) **28** 521-574. DOI: 10.1210/er.2007-0001
54. Chakrabarti S, Lekontseva O, Davidge ST. **Estrogen is a modulator of vascular inflammation**. *IUBMB Life* (2008.0) **60** 376-382. DOI: 10.1002/iub.48
55. El Khoudary SR, Santoro N, Chen HY, Tepper PG, Brooks MM, Thurston RC. **Trajectories of estradiol and follicle-stimulating hormone over the menopause transition and early markers of atherosclerosis after menopause**. *Eur J Prev Cardiol* (2016.0) **23** 694-703. DOI: 10.1177/2047487315607044
56. Harris BS, Steiner AZ, Jukic AM. **Ovarian reserve biomarkers and menstrual cycle length in a prospective cohort study**. *J Clin Endocrinol Metab* (2021.0) **106** E3748-E3759. DOI: 10.1210/clinem/dgab204
57. Verit FF, Akyol H, Sakar MN. **Low antimullerian hormone levels may be associated with cardiovascular risk markers in women with diminished ovarian reserve**. *Gynecol Endocrinol* (2016.0) **32** 302-305. DOI: 10.3109/09513590.2015.1116065
58. Chu MC, Rath KM, Huie J, Taylor HS. **Elevated basal FSH in normal cycling women is associated with unfavourable lipid levels and increased cardiovascular risk**. *Hum Reprod* (2003.0) **18** 1570-1573. DOI: 10.1093/humrep/deg330
59. Li X, Chen W, Li P, Wei J, Cheng Y, Liu P. **Follicular stimulating hormone accelerates atherogenesis by increasing endothelial VCAM-1 expression**. *Theranostics* (2017.0) **7** 4671-4688. DOI: 10.7150/thno.21216
60. Rameez RM, Sadana D, Kaur S, Ahmed T, Patel J, Khan MS. **Association of maternal lactation with diabetes and hypertension: a systematic review and meta-analysis**. *JAMA Netw Open* (2019.0) **2** e1913401-e1913401. DOI: 10.1001/jamanetworkopen.2019.13401
61. 61.Tschiderer L, Seekircher L, Kunutsor SK, Peters SAE, O’keeffe LM, Willeit P. Breastfeeding is associated with a reduced maternal cardiovascular risk: systematic review and meta-analysis involving data from 8 studies and 1 192 700 parous women. J Am Heart Assoc. 2022;11(2):22746.
62. Wu P, Mamas MA, Gulati M. **Health care professional’s knowledge of pregnancy complications and women’s cardiovascular health: an international study utilizing social media**. *J Womens Health* (2022.0) **31** 1197-1207. DOI: 10.1089/jwh.2021.0298
|
---
title: 'Vision-Related Quality of Life in Patients With Keratoconus: A Nationwide
Study in Saudi Arabia'
journal: Cureus
year: 2023
pmcid: PMC10029697
doi: 10.7759/cureus.35178
license: CC BY 3.0
---
# Vision-Related Quality of Life in Patients With Keratoconus: A Nationwide Study in Saudi Arabia
## Abstract
Purpose: To evaluate the impact of keratoconus (KC) on quality of life and assess visual performance via the National Eye Institute Visual Functioning Questionnaire-25 (NEI-VFQ-25) in the Saudi population.
Patients and methods: A descriptive cross-sectional study was conducted using the NEI-VFQ-25 to evaluate the vision-related quality of life among previously diagnosed KC patients. An online questionnaire was used to distribute the validated survey through various social media networks. The data were extracted, reviewed, coded, and then analyzed using the Statistical Package for Social Sciences (SPSS) version 26 (IBM Corp., Armonk, NY).
Results: A total of 429 patients completed the questionnaire. The overall score of NEI-VFQ-25 was 58.6 (SD: 18.0). The visual performance was worse in females than males (with a score of 55.1), especially in patients aged less than 30 years. Visual function improved with the use of low-vision aids (spectacles and contact lenses) compared with those who did not use them.
Conclusion: Our study confirms the functional impairment in patients with KC, especially in females, patients aged less than 30 years, and those with no low-vision aids. Moreover, it suggests a significant role of these vision aids (spectacles and contact lenses) in improving the quality of life in patients with KC.
## Introduction
Keratoconus (KC) is a multifactorial ectatic corneal disease characterized by progressive corneal thinning and subsequent bulging of the corneal structure. This progression, if not halted, usually leads to the development of myopization and irregular astigmatism, resulting in impairment of visual acuity and quality of vision [1]. Typically, it begins in late childhood, clinically manifests in adolescence age, and persists till the third to fourth decade of life [2]. The prevalence of KC ranges from $0.05\%$ in the United States to $18.7\%$ in Saudi Arabia [3,4]. Depending on its severity, KC can be managed non-surgically by spectacles, lenses (either soft, rigid, or scleral), intrastromal corneal ring implants, or corneal collagen crosslinking. If still not effective, surgical management is usually performed (either lamellar or penetrating keratoplasty) [1,5,6]. In previous studies, it has been found that the management of choice in KC patients ultimately has a detrimental impact on their quality of life (QoL) [1,6,7].
The impact of chronic eye disease on daily activities can be measured by vision-related QoL, a good and measurable health outcome in patients with visual impairment [8]. The National Eye Institute Visual Functioning Questionnaire-25 (NEI-VFQ-25) has been found to be a validated instrument for measuring vision-related QoL [9]. This 25-item questionnaire has been used to assess patients’ QoL in a variety of eye-related disorders [10-14].
Evaluating the QoL in patients with KC serves as an insight into how the disease influences all aspects of their well-being and has become an important measure of disease management tailored to each patient. Therefore, this study aimed to explore the association between vision-related QoL and relevant sociodemographic and clinical variables in a group of patients with KC in Saudi Arabia, using the NEI-VFQ-25.
## Materials and methods
Study design and population This descriptive cross-sectional study was conducted using the NEI-VFQ-25 to evaluate the QoL in Saudi patients with KC. The validated questionnaire was distributed among the largest online Saudi KC support group (available on Twitter and Telegram) with 1161 participants using Google Forms (Google, Mountain View, CA). This study included adults previously diagnosed with KC, who electronically signed a voluntary informed consent to participate in this study. All individuals under the age of 18 years or over the age of 60 years were excluded. All procedures performed were in accordance with the ethical standards of the Institutional Research Committee at King Faisal University (KFU) and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The data variables that were collected were age, gender, education level, residence region, and living location.
Questionnaire The main theme variable was QoL related to vision and health obtained using NEI-VFQ-25 [15], a self-administered, vision-specific, patient-reported outcome measure that reports on real-world visual function and is globally validated for a variety of ocular conditions. The NEI-VFQ-25 mainly measures general health, visual acuity, eye pain, difficulty seeing near/far objects, and social functioning. The total score ranges from 0 to 100. Higher values indicate better results [15]. The NEI-VFQ-25 has adequate validity and reliability, with α = 0.831 ($95\%$ CI: 0.735-0.904) in the entire questionnaire and >0.70 in all subscales except in “driving,” which, as in its official version, obtained lower reliability because it is an activity that not everyone performs, thus reducing the response rate [16].
Statistical analysis Descriptive statistics of sample characteristics were calculated and presented as absolute frequencies and percentages for categorical variables and as mean and standard deviation (SD) for continuous variables. The normality of data was contrasted using the Kolmogorov-Smirnov test. Mann-Whitney or Kruskal-Wallis tests were used to assess associations between QoL and categorical variables related to health and vision. Two-tailed analysis with 0.05 was used as a cutoff for statistical significance. All data analyses were performed using the Statistical Package for Social Sciences (SPSS) version 26 (IBM Corp., Armonk, NY).
## Results
The present study consisted of 429 participants with the diagnosis of KC, who agreed to fill out our questionnaire. The socio-demographic characteristics of the patients are presented in Table 1. Among the participants, the most common age group was 30-40 years, with male gender predominance, i.e., nearly two-thirds ($65.5\%$). A large majority of study participants mostly were bachelor’s degree holders ($67.1\%$). Approximately $43.1\%$ of the patients were living in the central region of Saudi Arabia and most were urban residents ($86.9\%$).
**Table 1**
| Study data | N (%) |
| --- | --- |
| Age group | |
| ≤17 years | 03 (0.70%) |
| 18-25 years | 56 (13.1%) |
| 26-30 years | 89 (20.7%) |
| 31-35 years | 106 (24.7%) |
| 36-40 years | 110 (25.6%) |
| >40 years | 65 (15.2%) |
| Gender | |
| Male | 281 (65.5%) |
| Female | 148 (34.5%) |
| Educational level | |
| High school or below | 95 (22.2%) |
| Bachelor’s degree | 288 (67.1%) |
| Master’s degree or higher | 46 (10.7%) |
| Residence region | |
| Northern Region | 38 (08.9%) |
| Eastern Region | 42 (09.8%) |
| Central Region | 185 (43.1%) |
| Western Region | 90 (21.0%) |
| Southern Region | 74 (17.2%) |
| Living location | |
| Urban | 373 (86.9%) |
| Rural | 56 (13.1%) |
Regarding the characteristics of KC patients in our study, more than one-third ($35.4\%$) had a duration of KC of more than 15 years. The proportion of patients with a family history of KC was $33.3\%$. The commonly known type of correction was contact lens ($50.8\%$) while the most common obstacles encountered when using contact lenses were recurrent inflamed, allergy, or dry eye disease ($29.1\%$). Corneal cross-linking (CXL) has been the most frequent surgical procedure done for patients ($31.5\%$). Regarding the level of satisfaction toward corneal transplant, patients were either satisfied ($15.9\%$) or partially satisfied ($18.4\%$) with the procedure. Of those who underwent CXL, $9.5\%$ of them indicated that they repeated the procedure as it is required (Table 2).
**Table 2**
| Variables | N (%) |
| --- | --- |
| Time since keratoconus diagnosis | |
| 1 year or less | 25 (05.8%) |
| 1-5 years | 88 (20.6%) |
| 5-10 years | 85 (19.8%) |
| 10-15 years | 79 (18.4%) |
| More than 15 years | 152 (35.4%) |
| Family history of keratoconus | |
| Yes | 143 (33.3%) |
| No/unknown | 286 (66.7%) |
| Type of correction | |
| No optical correction | 91 (21.2%) |
| Spectacles | 95 (22.1%) |
| Contact lens | 218 (50.8%) |
| What obstacles did you face with contact lenses? | |
| | 59 (13.8%) |
| Adaption obstacles | 83 (19.3%) |
| Cost | 114 (26.6%) |
| Fear | 14 (03.3%) |
| Recurrent inflamed, allergic, dry eye | 125 (29.1%) |
| Not interested in contact lens | 34 (07.9%) |
| Surgical procedures | |
| | 175 (40.8%) |
| Corneal cross-linking (CXL) | 135 (31.5%) |
| Intrastromal corneal ring (ICR) | 41 (09.6%) |
| Implantable contact lenses | 06 (01.4%) |
| Partial corneal transplant | 19 (04.4%) |
| Total corneal transplant | 53 (12.4%) |
| Have you been managed by corneal transplantation? | |
| No | 170 (39.6%) |
| In one eye only | 113 (26.3%) |
| In both eyes | 146 (34.0%) |
| Level of satisfaction toward corneal transplantation | |
| Satisfied | 68 (15.9%) |
| Partially Satisfied | 79 (18.4%) |
| Not Satisfied | 94 (21.9%) |
| The recovery period is not completed yet | 188 (43.8%) |
| Was repeated corneal cross-linking (CXL) required if you had the first one already? (n = 232) | |
| Yes | 22 (09.5%) |
| No | 210 (90.5%) |
As shown in Figure 1, the most common associated eye diseases encountered in KC patients were astigmatism ($36.8\%$) and myopia ($30.5\%$).
**Figure 1:** *Other associated eye diseases*
In Table 3, the median values of NEI-VFQ-25 subscales in descending order were shown as color vision (82.5), general vision (77.1), general health (73.5), social functioning (65.2), driving (60.4), near activities (58.7), dependency (56.1), ocular pain (55.8), distance activities (55.2), peripheral vision (53.9), role difficulties (52.8), and mental health (44.7). The overall median score of NEI-VFQ-25 was 58.6 (SD: 18). See the distribution of the overall median score of NEI-VFQ-25 in Figure 2.
In Table 4, a statistical test revealed that a higher median mental health score and overall NEI-VFQ-25 score were associated with the older age group (>30 years). Male patients were more associated with a higher ocular pain median score and the overall NEI-VFQ-25 score. In addition, a higher median general vision score was more associated with patients who had a shorter duration of KC (≤10 years).
**Table 4**
| NEI-VFQ-25 | Age group | Age group.1 | Gender | Gender.1 | Time of diagnosis | Time of diagnosis.1 |
| --- | --- | --- | --- | --- | --- | --- |
| NEI-VFQ-25 | ≤30 years | >30 years | Male | Female | ≤10 years | >10 years |
| General health | 69.4 | 75.3 | 74.6 | 71.5 | 72.3 | 74.6 |
| General vision | 75.2 | 77.9 | 76.1 | 78.9 | 79.7* | 74.8* |
| Ocular pain | 52.8 | 57.1 | 57.6* | 52.4* | 54.9 | 56.5 |
| Near activities | 55.6 | 59.9 | 59.9 | 56.3 | 57.3 | 59.8 |
| Distance activities | 51.4 | 56.8 | 55.8 | 53.8 | 53.6 | 56.5 |
| Social functioning | 63.2 | 66.1 | 66.6 | 62.5 | 63.9 | 66.4 |
| Mental health | 40.6* | 46.5* | 45.7 | 42.9 | 44.3 | 45.0 |
| Role difficulties | 51.6 | 53.3 | 52.7 | 53.0 | 52.5 | 53.1 |
| Dependency | 53.3 | 57.3 | 56.3 | 55.6 | 56.9 | 55.4 |
| Driving | 58.9 | 60.9 | 59.9 | 64.3 | 58.3 | 61.8 |
| Color vision | 83.6 | 82.1 | 81.9 | 83.6 | 83.9 | 81.3 |
| Peripheral vision | 54.0 | 53.9 | 53.4 | 55.1 | 54.3 | 53.7 |
| Overall score | 55.8* | 59.9* | 60.5** | 55.1** | 57.8 | 59.3 |
As shown in Table 5, correction with spectacles was more associated with higher median scores in ocular pain, social functioning, color vision, and the overall NEI-VFQ-25. Additionally, correction with contact lenses was associated with a higher median score in the general vision subscale.
**Table 5**
| NEI-VFQ-25 | Unnamed: 1 | Unnamed: 2 | Unnamed: 3 |
| --- | --- | --- | --- |
| NEI-VFQ-25 | Type of correction | Type of correction | Type of correction |
| NEI-VFQ-25 | No correction | Spectacles | Contact lens |
| General health | 71.9 | 73.9 | 74.3 |
| General vision | 72.3 | 71.8 | 81.7** |
| Ocular pain | 50.4 | 64.2** | 53.9 |
| Near activities | 54.1 | 64.9 | 57.1 |
| Distance activities | 53.5 | 59.9 | 53.7 |
| Social functioning | 60.7 | 75.1** | 62.9 |
| Mental health | 38.1 | 51.8 | 42.9 |
| Role difficulties | 48.2 | 60.7 | 50.3 |
| Dependency | 52.9 | 64.5 | 52.3 |
| Driving | 57.1 | 63.3 | 58.3 |
| Color vision | 79.1 | 88.7* | 80.7 |
| Peripheral vision | 52.7 | 57.7 | 52.4 |
| Overall score | 54.7 | 63.7** | 57.5 |
In our study, there were no significant differences in the median scores of the type of surgical procedures in all the subscales of NEI-VFQ-25 and its overall score (all $p \leq 0.05$) (Table 6).
**Table 6**
| NEI-VFQ-25 | Type of surgical procedures | Type of surgical procedures.1 | Type of surgical procedures.2 | Type of surgical procedures.3 | Type of surgical procedures.4 |
| --- | --- | --- | --- | --- | --- |
| NEI-VFQ-25 | CXL | ICR | ICL | PCT | TCT |
| General health | 70.4 | 75.0 | 91.7 | 59.2 | 75.9 |
| General vision | 78.1 | 71.7 | 83.3 | 75.8 | 71.3 |
| Ocular pain | 55.3 | 60.9 | 54.2 | 51.3 | 54.7 |
| Near activities | 60.6 | 60.7 | 51.4 | 60.6 | 57.0 |
| Distance activities | 56.5 | 50.6 | 44.4 | 48.6 | 52.9 |
| Social functioning | 65.7 | 66.7 | 56.3 | 65.3 | 65.9 |
| Mental health | 46.6 | 47.4 | 34.4 | 38.5 | 43.9 |
| Role difficulties | 54.6 | 57.6 | 45.8 | 50.0 | 46.9 |
| Dependency | 56.5 | 60.6 | 61.1 | 46.9 | 54.6 |
| Driving | 63.4 | 60.8 | 58.3 | 49.4 | 56.0 |
| Color vision | 84.7 | 81.9 | 75.0 | 81.6 | 81.6 |
| Peripheral vision | 55.2 | 57.3 | 50.0 | 51.3 | 54.7 |
| Overall score | 59.3 | 60.4 | 58.0 | 54.5 | 57.3 |
Patients who have no other associated eye diseases were more associated with higher median scores in near activities, driving, and color vision while patients with concomitant hyperopia were more associated with a higher median score in mental health (Table 7).
**Table 7**
| NEI-VFQ-25 | Other eye diseases | Other eye diseases.1 | Other eye diseases.2 | Other eye diseases.3 | Other eye diseases.4 | Other eye diseases.5 |
| --- | --- | --- | --- | --- | --- | --- |
| NEI-VFQ-25 | Astigmatism | Myopia | Hyperopia | Amblyopia | Other | |
| General health | 71.5 | 72.7 | 87.5 | 74.3 | 72.2 | 82.4 |
| General vision | 77.2 | 78.5 | 90.0 | 74.3 | 78.1 | 76.3 |
| Ocular pain | 55.9 | 54.9 | 50.0 | 52.1 | 54.4 | 59.3 |
| Near activities | 57.4 | 59.6 | 45.8 | 50.5 | 57.5 | 66.7* |
| Distance activities | 55.5 | 54.4 | 50.0 | 47.5 | 51.7 | 60.7 |
| Social functioning | 66.4 | 63.9 | 50.0 | 58.5 | 62.9 | 68.4 |
| Mental health | 46.2 | 43.1 | 53.1* | 35.5 | 44.2 | 49.4 |
| Role difficulties | 54.4 | 51.1 | 50.0 | 43.6 | 53.8 | 56.5 |
| Dependency | 56.7 | 54.9 | 37.5 | 48.3 | 55.2 | 63.4 |
| Driving | 60.5 | 58.5 | 41.7 | 50.6 | 57.8 | 70.4* |
| Color vision | 81.5 | 83.5 | 75.0 | 73.6 | 84.4 | 85.4* |
| Peripheral vision | 52.8 | 53.4 | 37.5 | 52.1 | 50.0 | 59.7 |
| Overall score | 58.8 | 58.1 | 53.9 | 53.7 | 56.7 | 63.3 |
## Discussion
Due to remarkable advancements in medical care, patient-centered approaches prioritize well-being and QoL when tailoring therapies. In the case of KC, a known cause of social burden is early visual impairment and outcome unpredictability. Previous studies have shown a significant decrease in the QoL for these patients [17]. Identifying the factors linked to this decrease might aid in the development of action protocols aimed at not only lowering or eradicating KC symptoms but also enhancing the QoL and well-being of those affected. In our study, the total score was 58.6 ± 18.0. The highest NEI-VFQ-25 subscale score was attributed to color vision (82.5 ± 24.7) and the lowest scores were related to mental health (44.7 ± 27.9). Comparing our results with previous studies examining the impact of KC on vision-related QoL, we found that our participants’ overall score was lower than French [18], Iranian [19], and Turkish KC patients [9]. In addition, our study scores were lower than the Collaborative Longitudinal Evaluation of Keratoconus (CLEK) scores, except for general vision [17]. Compared to other chronic ocular diseases, the scores of near and distance activities in our study differ from the scores of glaucoma patients [11], but they are close to previously reported scores of age-related macular degeneration [13]. Given the fact that a higher median mental health score and overall NEI-VFQ-25 score were observed in the older age group (>30 years), this could mean that as the KC patients get older, they will develop perceptual adaptation toward the disease. This suggests that clinical indicators like visual acuity are not the only ones that influence patients' perceptions of their illness. Adopting patient-reported outcome measures also provides more detailed and useful information to lessen the KC burden.
Socio-demographic data and time since KC diagnosis exhibited an association with NEI-VFQ-25 scores. A higher median general vision score was more associated with patients who had a shorter duration of KC (≤10 years). In accordance with Gothwal et al., patients with KC disease duration of more than three years showed worse ratings on both the functional and emotional well-being measures [20]. Consistent with our findings, a previous study showed a considerable improvement in the vision-related QoL in KC patients equipped with implantable contact lenses (ICL), with a 19.5-point increase in the total score [1]. It is interesting to note that, as revealed in participant responses, there were no significant differences in the median scores of the type of surgical procedures in all the subscales of NEI-VFQ-25 and its overall score (all $p \leq 0.05$).
In our study, asking about the association between driving and the QoL in patients with KC is the main weakness since not all the participants drive, thus influencing the response rate.
## Conclusions
Our patients with KC had mental, visual, physical, and social impairment in their QoL. This perceived shortcoming was in excess compared with their clinical measures that confirm a previous theory about the disproportionate social burden of KC. Our findings showed that female patients, patients aged less than 30 years, and those with no low vision aids were associated with lower QoL scores using the NEI-VFQ-25. Strategies to improve the QoL in KC patients are warranted to maintain their QoL. Finally, our results provided evidence for the validity and reliability of the Saudi version of the NEI-VFQ-25, and potentially applicable in further investigations.
## References
1. Baudin F, Chemaly A, Arnould L, Barrénéchea E, Lestable L, Bron AM, Creuzot-Garcher C. **Quality-of-life improvement after scleral lens fitting in patients with keratoconus**. *Eye Contact Lens* (2021) **47** 520-525. PMID: 34261868
2. Dudeja L, Chauhan T, Vohra S. **Sequence of events leading to diagnosis of keratoconus and its impact on quality of life**. *Indian J Ophthalmol* (2021) **69** 3478-3481. PMID: 34826978
3. Torres Netto EA, Al-Otaibi WM, Hafezi NL, Kling S, Al-Farhan HM, Randleman JB, Hafezi F. **Prevalence of keratoconus in paediatric patients in Riyadh, Saudi Arabia**. *Br J Ophthalmol* (2018) **102** 1436-1441. PMID: 29298777
4. Al-Amri AM. **Prevalence of keratoconus in a refractive surgery population**. *J Ophthalmol* (2018) **2018** 5983530. PMID: 30271629
5. Pinto RD, Abe RY, Gomes FC. **Quality of life in keratoconus: evaluation with Keratoconus Outcomes Research Questionnaire (KORQ)**. *Sci Rep* (2021) **11** 12970. PMID: 34155238
6. Al-Qahtani BS, Al-Swailem SA, Ahmad K, Khandekar R. **Vision related quality of life of patients with keratoconus after keratoplasty at a tertiary eye hospital in Saudi Arabia**. *Eur J Ophthalmol* (2021) **31** 3500-3504. PMID: 34053313
7. Ortiz-Toquero S, Perez S, Rodriguez G, de Juan V, Mayo-Iscar A, Martin R. **The influence of the refractive correction on the vision-related quality of life in keratoconus patients**. *Qual Life Res* (2016) **25** 1043-1051. PMID: 26334843
8. Hashemi H, Mehravar F, Asgari S, Emamian MH, Fotouhi A. **Visual functions and disability in Iranian adults: a population-based study**. *BMC Ophthalmol* (2022) **22** 30. PMID: 35057773
9. Aydin Kurna S, Altun A, Gencaga T, Akkaya S, Sengor T. **Vision related quality of life in patients with keratoconus**. *J Ophthalmol* (2014) **2014** 694542. PMID: 24868455
10. Potic J, Bergin C, Giacuzzo C, Konstantinidis L, Daruich A, Wolfensberger TJ. **Application of modified NEI VFQ-25 after retinal detachment to vision-related quality of life**. *Retina* (2021) **41** 653-660. PMID: 32568989
11. Szegedi S, Boltz A, Scharinger EM, Vécsei-Marlovits PV. **Quality of life in patients with glaucoma assessed by 39-item National Eye Institute Visual Functioning Questionnaire (NEI VFQ-39)**. *Graefes Arch Clin Exp Ophthalmol* (2022) **260** 1623-1631. PMID: 34674029
12. Akpolat C, Demir M, Cevher S, Ozturk SZ, Yesiltas S. **The impact of phacoemulsification surgery on vision-related quality of life in senile cataract patients**. *Ther Adv Ophthalmol* (2022) **14**
13. Caballe-Fontanet D, Alvarez-Peregrina C, Busquet-Duran N, Pedemonte-Sarrias E, Andreu-Vázquez C, Sánchez-Tena MÁ. **Quality of life and anxiety in age macular degeneration patients: a cross-sectional study**. *Int J Environ Res Public Health* (2022) **19** 820. PMID: 35055640
14. Roberts-Martínez Aguirre I, Rodríguez-Fernández P, González-Santos J, Aguirre-Juaristi N, Alonso-Santander N, Mielgo-Ayuso J, González-Bernal JJ. **Exploring the quality of life related to health and vision in a group of patients with diabetic retinopathy**. *Healthcare (Basel)* (2022) **10** 142. PMID: 35052305
15. Wang LL, Liu WJ, Liu HY, Xu X. **Single-site baseline and short-term outcomes of clinical characteristics and life quality evaluation of Chinese wet age-related macular degeneration patients in routine clinical practice**. *Chin Med J (Engl)* (2015) **128** 1154-1159. PMID: 25947396
16. Alvarez-Peregrina C, Sánchez-Tena MA, Caballé-Fontanet D, Thuissard-Vasallo IJ, Gacimartín-García MB, Orduna-Magán C. **Crosscultural adaptation and validation into Spanish of the questionnaire National Eye Institute Visual Function Questionnaire 25**. *Arch Soc Esp Oftalmol (Engl Ed)* (2018) **93** 586-591. PMID: 30006104
17. Kymes SM, Walline JJ, Zadnik K, Gordon MO. **Quality of life in keratoconus**. *Am J Ophthalmol* (2004) **138** 527-535. PMID: 15488776
18. Mahdaviazad H, Bamdad S, Roustaei N, Mohaghegh S. **Vision-related quality of life in Iranian patients with keratoconus: National Eye Institute Vision Function Questionnaire-25**. *Eye Contact Lens* (2018) **44** 0-4
19. Panthier C, Moran S, Bourges JL. **Evaluation of vision-related quality of life in keratoconus patients, and associated impact of keratoconus severity indicators**. *Graefes Arch Clin Exp Ophthalmol* (2020) **258** 1459-1468. PMID: 32296993
20. Gothwal VK, Reddy SP, Fathima A, Bharani S, Sumalini R, Bagga DK, Sudharman PM. **Assessment of the impact of keratoconus on vision-related quality of life**. *Invest Ophthalmol Vis Sci* (2013) **54** 2902-2910. PMID: 23518765
|
---
title: Artificial intelligence-enabled electrocardiographic screening for left ventricular
systolic dysfunction and mortality risk prediction
authors:
- Yu-Chang Huang
- Yu-Chun Hsu
- Zhi-Yong Liu
- Ching-Heng Lin
- Richard Tsai
- Jung-Sheng Chen
- Po-Cheng Chang
- Hao-Tien Liu
- Wen-Chen Lee
- Hung-Ta Wo
- Chung-Chuan Chou
- Chun-Chieh Wang
- Ming-Shien Wen
- Chang-Fu Kuo
journal: Frontiers in Cardiovascular Medicine
year: 2023
pmcid: PMC10029758
doi: 10.3389/fcvm.2023.1070641
license: CC BY 4.0
---
# Artificial intelligence-enabled electrocardiographic screening for left ventricular systolic dysfunction and mortality risk prediction
## Abstract
### Background
Left ventricular systolic dysfunction (LVSD) characterized by a reduced left ventricular ejection fraction (LVEF) is associated with adverse patient outcomes. We aimed to build a deep neural network (DNN)-based model using standard 12-lead electrocardiogram (ECG) to screen for LVSD and stratify patient prognosis.
### Methods
This retrospective chart review study was conducted using data from consecutive adults who underwent ECG examinations at Chang Gung Memorial Hospital in Taiwan between October 2007 and December 2019. DNN models were developed to recognize LVSD, defined as LVEF <$40\%$, using original ECG signals or transformed images from 190,359 patients with paired ECG and echocardiogram within 14 days. The 190,359 patients were divided into a training set of 133,225 and a validation set of 57,134. The accuracy of recognizing LVSD and subsequent mortality predictions were tested using ECGs from 190,316 patients with paired data. Of these 190,316 patients, we further selected 49,564 patients with multiple echocardiographic data to predict LVSD incidence. We additionally used data from 1,194,982 patients who underwent ECG only to assess mortality prognostication. External validation was performed using data of 91,425 patients from Tri-Service General Hospital, Taiwan.
### Results
The mean age of patients in the testing dataset was 63.7 ± 16.3 years ($46.3\%$ women), and 8,216 patients ($4.3\%$) had LVSD. The median follow-up period was 3.9 years (interquartile range 1.5–7.9 years). The area under the receiver-operating characteristic curve (AUROC), sensitivity, and specificity of the signal-based DNN (DNN-signal) to identify LVSD were 0.95, 0.91, and 0.86, respectively. DNN signal-predicted LVSD was associated with age- and sex-adjusted hazard ratios (HRs) of 2.57 ($95\%$ confidence interval [CI], 2.53–2.62) for all-cause mortality and 6.09 (5.83–6.37) for cardiovascular mortality. In patients with multiple echocardiograms, a positive DNN prediction in patients with preserved LVEF was associated with an adjusted HR ($95\%$ CI) of 8.33 (7.71 to 9.00) for incident LVSD. Signal- and image-based DNNs performed equally well in the primary and additional datasets.
### Conclusion
Using DNNs, ECG becomes a low-cost, clinically feasible tool to screen LVSD and facilitate accurate prognostication.
## Introduction
Heart failure (HF) is a major health issue affecting over 26 million people worldwide. It causes a significant increase in both morbidity and mortality and imposes a financial burden on society [1]. Echouffo-Tcheugui et al. have classified left ventricular dysfunction into two categories: left ventricular systolic dysfunction (LVSD) and left ventricular diastolic dysfunction. LVSD is characterized by a reduced left ventricular ejection fraction (LVEF) and is associated with three times the risk of developing overt HF [2]. Early identification of individuals with asymptomatic LVSD can lead to effective interventions, such as lifestyle changes, and medications, including angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, mineralocorticoid receptor antagonists, and beta-blockers (3–7), which can delay the onset of HF, reduce the rate of cardiac events, and improve survival (8–10).
The most commonly used method to assess LVSD is the transthoracic echocardiogram (TTE), but its limitations, including portability, cost, and operator dependency, restrict its use as a screening tool. To address this, there is a need for more accurate and accessible screening tools to identify LVSD in asymptomatic patients, such as a weighted scoring model incorporating clinical characteristics and plasma natriuretic peptides. However, these tools lack the specificity to predict LVSD in asymptomatic populations [11, 12].
The electrocardiogram (ECG) is an inexpensive and widely available method that measures the collective electrical activity of the heart and may contain information related to LVSD. While ECG recording is a standardized process, the accuracy and consistency of human interpretation can vary widely based on the experience and expertise of the interpreter. In addition, subtle ECG features that are invisible to the human eye may be useful for LVSD detection and prognostication. To overcome these challenges, the use of deep neural networks (DNNs) is proposed.
In recent years, DNNs have been applied successfully in the healthcare industry, including image analysis [13], predictive modeling [14], natural language processing [15], and drug discovery [16]. They are superior to traditional pattern recognition methods [17] and form the foundation of clinical applications such as fracture detection [18], retinopathy grading [19], and lung nodule identification [20]. DNN tools can interpret ECGs with similar accuracy to experienced physicians. Attia et al. developed a DNN-based ECG screening tool to identify individuals with LVEF ≤$35\%$ [21]. A subsequent pragmatic clinical trial showed that a DNN-based intervention increased the likelihood of identifying patients with low LVEF during routine primary care [22]. However, the effectiveness of DNN-based models in predicting incident LVSD and mortality has not been studied in a large clinical setting.
With data from approximately 1.7 million individuals, we conducted this study to evaluate the feasibility of using DNN-based ECG interpretation as a screening tool for LVSD and to assess its utility in risk assessment. The primary outcome was the ability of the DNN model to accurately identify individuals with LVSD (defined as LVEF <$40\%$) based solely on the ECG. The secondary outcome was the ability of the DNN model to identify individuals at increased risk of death and at increased risk of developing LVSD.
## Data sources and study population
This study was conducted at Chang Gung Memorial Hospital (CGMH), the largest private hospital system in Taiwan. The study population included consecutive adult patients (age ≥ 18) who underwent standard 12-lead ECG at CGMH between October 2007 and December 2019 (1,777,039 individuals, 5,148,718 ECG tracings). ECGs with poor recording quality or unavailable leads were excluded. The ECG data were linked to the Chang Gung Research Database (CGRD), which included the electronic health records of all patients who visited any one of the following seven hospitals: Keelung, Taipei, Linkou (headquarters), Taoyuan, Yunlin, Chiayi, and Kaohsiung.
The patients’ survival status was confirmed by linking the CGRD to the National Death Registry. Valid internal patient record linkage was achieved by using unique patient identifiers, and these were encrypted before the data were released to researchers to protect patient confidentiality. This study was approved by the Institutional Review Board of CGMH and Tri-Service General Hospital. This study used anonymous and nontraceable data, so the need for patient consent was waived.
## Collection of data
Standard 12-lead ECGs with 10-s voltage-time traces were acquired at a sampling rate of 500 Hz using a MAC 5000, MAC 5500, or MAC5500HD ECG machine (GE Healthcare, Chicago, IL, United States) and stored using the Marquette Universal System for Electrocardiography (MUSE). Each standard 12-lead ECG was stored as a 12 × 5,000 matrix. Both the raw ECG signal data and processed ECG images at a 400 × 600-pixel resolution were obtained.
Transthoracic echocardiograms were performed and interpreted in accordance with the guidelines set forth by the American Society of Echocardiography and the American College of Cardiology/American Heart Association. Comprehensive two-dimensional (2D) or three-dimensional (3D) Doppler echocardiographic profiles and quantitative measurements were recorded in Chang Gung’s health information system. For this study, we only extracted LVEF values for analysis. LVEF was routinely measured using standardized methodologies. If different methods were used to measure LVEF in a report, the order of data preference was as follows: 3D echocardiogram, the Simpson biplane method, 2D method, linear measurement using M-mode. If multiple LVEF values were obtained using one method, the mean value was used for analysis.
To achieve proper correlation between ECG and TTE data, only TTEs obtained within 2 weeks of the index ECG were used for DNN model creation.
## Development of DNN models for identification of LVSD
In this study, we implemented two types of DNNs using the Pytorch framework and Python 3.6. All training was performed on an NVIDIA DGX-1 platform with 8 V100 GPUs and 32 GB of RAM per GPU. For the DNN that used signal inputs (DNN-signal), we used the deep residual network (ResNet) [23] modified to fit the signal input (Supplementary Figure 1). We used a wider kernel for the first convolution layer compared with the original ResNet framework as used for images. This architecture used skip connections, which allowed information to pass directly to the next layer to avoid the degradation caused by deeper neural networks. The network consisted of a convolution layer followed by eight residual blocks. Each residual block contained two convolution layers. The output of the last block was fed into hybrid pooling because combining max- and average-pooling methods improved the generalization ability while reducing dimensionality [24, 25]. The output of hybrid pooling was subsequently sent to a fully connected layer to perform the final classification. The output of each convolutional layer was followed by batch normalization for distribution normalization and fed into a rectified linear activation unit [26]. Cross-entropy loss with an Adam optimizer [27] was used in the model. Dropout was applied to reduce the overfitting by breakup co-adaptation on the training data [28].
For the DNN using the image inputs (DNN-image), we prepared a 400 × 600-pixel image similar to standard 12-lead ECG images (Supplementary Figure 2) using the signal data (12 × 5,000 matrix). The resolution was determined by a series of experiments using different image resolutions. The images were fed to ResNet-18 [23], and the output layer had two classes (Softmax function). The validation set was used to optimize the network architecture and network hyperparameters. The DNN-signal and DNN-image used the same training and validation sets for model building and were tested on the same testing set. A receiver operating characteristic (ROC) curve was plotted to assess the performance. The model with the highest area under the ROC curve (AUROC) was selected as the final model. We used the validation dataset ROC to select optimal threshold for the probability of LVSD by applying the Youden index (J) method.
We further assessed the network performance in different age, sex, and comorbidity strata. The odds ratio (OR), sensitivity, and specificity were calculated for each strata.
## Division of dataset
Among 1,684,298 adult patients with ECG tracings, 380,675 had at least one TTE data within 2 weeks of the index ECG during the study period (Figure 1). For patients with multiple ECG–TTE pairs, the earliest pair with the shortest ECG–TTE interval was selected for model development. Total 380,675 ECG–TTE paired datasets were used for the primary analysis. These ECG-TTE pairs were randomly allocated into a training, validation, or testing set using simple random sampling in which each dataset had an equal probability of selection without replacement. The final DNN development cohort included 133,225 patients in the training set, 57,134 in the validation set, and 190,316 in the testing set. No patient was allocated to more than one group (Figure 1).
**Figure 1:** *Data flow for ECG and TTE data pairing.*
We further conducted an external validation using paired ECG-TTE data from the Tri-service General Hospital. The external validation cohort included 91,425 consecutive adults between April 2010 and September 2021. The criteria of patient selection and echocardiographic performance methodology were the same as for the derivation cohort. Different from the ECG machine used at CGMH, ECGs from Tri-service General Hospital were obtained using the Philips system.
## Performance evaluation of the DNN models in predicting mortality
The ability of DNN to predict all-cause and cardiovascular mortality was assessed. According to the differences between the results of echocardiographic measurements and DNN predictions, we defined the following names: (i) ‘true positive’ DNN prediction represents both DNN-predicted and echo-measured LVEF <$40\%$; (ii) ‘true negative’ DNN prediction represents both DNN-predicted and echo-measured LVEF ≥$40\%$; (iii) ‘false positive’ DNN prediction represents DNN-predicted LVEF<$40\%$ and contemporaneous echo-measured LVEF ≥$40\%$; and (iv) ‘false negative’ DNN prediction represents DNN-predicted LVEF≥$40\%$ and contemporaneous echo-measured LVEF <$40\%$. The associations of different groups with all-cause or cardiovascular mortality were also assessed. The National Death Registry was linked to the study dataset. In Taiwan, it is mandatory for physicians to report deaths and causes of death to the Department of Health and Welfare. Therefore, death records within the National Death Registry are considered complete and accurate. A previous validation study estimated the effect of the misrecorded causes of death in the National Death Registry on cardiovascular mortality rates. The effect was less than $4\%$, suggesting accurate cause-of-death coding in Taiwan [29].
## Sensitivity analyses
We conducted sensitivity analyses in patients who were not included in the primary analysis. These patients were included in the following sub-analyses (Figure 1): (i) among patients with multiple TTE examinations in the original testing dataset (dataset A1, $$n = 49$$,564), the incidence of LVSD and mortality were compared in patients with ‘false-positive’ versus ‘true-negative’ predictions of LVSD; (ii) among patients who underwent TTE after more than 2 weeks of the index ECG (dataset B), the incidence of LVSD and mortality were compared in patients with positive versus negative predictions of LVSD; and (iii) among patients without echocardiographic data (dataset C), mortality rate was compared in patients with positive versus negative predictions of LVSD. Age- and sex-weighted Kaplan–*Meier analysis* was used to determine the incidence of LVSD or mortality. Cox proportional hazard regression was used to estimate the age- and sex-adjusted hazard ratios (HR; $95\%$ confidence intervals [CI]) for LVSD and mortality.
Table 2 summarizes the performance of the DNN models in additional datasets. Subset A1 included 49,564 patients with multiple echocardiograms. Within this subset, ‘false positive’ DNN-signal predictions were associated with HRs ($95\%$ CI) of 8.33 (7.71–9.00) for incident LVSD, 1.99 (1.92–2.06) for all-cause mortality, and 3.51 (3.25–3.80) for cardiovascular mortality compared to ‘true negative’ DNN-signal predictions. ‘ False positive’ DNN-image predictions were associated with HRs ($95\%$ CI) of 8.19 (7.57–8.87) for incident LVSD, 2.05 (1.98–2.12) for all-cause mortality, and 3.77 (3.49–4.07) for cardiovascular mortality compared to ‘true negative’ DNN-image predictions.
**Table 2**
| Datasets/predictions | Incident LVSD | Incident LVSD.1 | All-cause mortality | All-cause mortality.1 | Cardiovascular mortality | Cardiovascular mortality.1 |
| --- | --- | --- | --- | --- | --- | --- |
| | Rate (95% CI) | HR (95% CI) | Rate (95% CI) | HR (95% CI) | Rate (95% CI) | HR (95% CI) |
| Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms | Subset A1: 45,866 patients with preserved LVEF by TTE in the testing dataset with multiple echocardiograms |
| Negative by DNN-signal (n = 36,920) | 8.6 (8.1–9.1) | 1.00 (Reference) | 54.6 (105.7–112.3) | 1.00 (Reference) | 7.9 (7.5–8.3) | 1.00 (Reference) |
| Positive by DNN-signal (n = 8,946) | 75.9 (72.3–79.5) | 8.33 (7.71–9.00) | 109.0 (105.7–112.3) | 1.99 (1.92–2.06) | 27.7 (26.1–29.4) | 3.51 (3.25–3.80) |
| Negative by DNN-image (n = 35,604) | 8.0 (7.5–8.5) | 1.00 (Reference) | 51.9 (50.9–52.9) | 1.00 (Reference) | 7.1 (6.7–7.5) | 1.00 (Reference) |
| Positive by DNN-image (n = 10,262) | 69.9 (66.7–73.2) | 8.19 (7.57–8.87) | 114.4 (111.3–117.6) | 2.05 (1.98–2.12) | 28.7 (27.1–30.3) | 3.77 (3.49–4.07) |
| Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs | Subset B: 83,787 patients who had TTE > 14 days after index ECGs |
| Negative by DNN-signal (n = 74,928) | 1.5 (1.3–1.7) | 1.00 (Reference) | 22.9 (22.5–23.3) | 1.00 (Reference) | 1.9 (1.8–2.0) | 1.00 (Reference) |
| Positive by DNN-signal (n = 8,859) | 30.6 (28.2–32.9) | 19.23 (16.56–22.33) | 62.6 (60.5–64.8) | 2.18 (2.09–2.26) | 12.1 (11.2–13.1) | 5.20 (4.70–5.75) |
| Negative by DNN-image (n = 73,795) | 1.4 (1.2–1.5) | 1.00 (Reference) | 21.7 (21.3–22.1) | 1.00 (Reference) | 1.8 (1.6–1.9) | 1.00 (Reference) |
| Positive by DNN-image (n = 9,992) | 28.3 (26.2–30.5) | 19.52 (16.72–22.80) | 69.8 (67.7–72.0) | 2.32 (2.24–2.41) | 12.2 (11.3–13.1) | 4.99 (4.52–5.52) |
| Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE | Subset C: 1,194,982 patients without TTE |
| Negative by DNN-signal (n = 1,155,523) | – | – | 16.9 (16.8–17.0) | 1.00 (Reference) | 1.1 (1.0–1.1) | 1.00 (Reference) |
| Positive by DNN-signal (n = 39,459) | – | – | 100.3 (98.8–101.8) | 3.24 (3.19–3.29) | 14.3 (13.7–14.9) | 6.83 (6.51–7.16) |
| Negative by DNN-image (n = 1,151,691) | – | – | 16.3 (16.2–16.4) | 1.00 (Reference) | 1.0 (1.0–1.0) | 1.00 (Reference) |
| Positive by DNN-image (n = 43,291) | – | – | 120.5 (118.9–122.2) | 3.46 (3.40–3.51) | 15.9 (15.4–16.5) | 6.82 (6.51–7.14) |
Within subset B, including 83,787 patients, positive DNN-signal predictions were associated HRs ($95\%$ CI) of 19.23 (16.56–22.33) for incident LVSD, 2.18 (2.09–2.26) for all-cause mortality, and 5.20 (4.70–5.75) for cardiovascular mortality. Positive DNN-image predictions were associated HRs ($95\%$ CI) of 19.52 (16.72–22.80) for incident LVSD, 2.32 (2.24–2.41) for all-cause mortality, and 4.99 (4.52–5.52) for cardiovascular mortality.
Within subset C, including 1,194,982 patients, DNN signal-predicted LVSD was associated with a HR ($95\%$ CI) of 3.24 (3.19–3.29) for all-cause mortality and 6.83 (6.51–7.16) for cardiovascular mortality. DNN image-predicted LVSD was associated with a HR ($95\%$ CI) of 3.46 (3.40–3.51) for all-cause mortality and 6.82 (6.51–7.14) for cardiovascular mortality. Supplementary Figures 5–12 show Kaplan–Meier curves for incident LVSD, all-cause and cardiovascular mortality for subsets A1, B, and C.
## Statistical methods
Only the testing datasets were evaluated for performance measures. The model’s diagnostic performance was evaluated by calculating the AUROC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The F1 score, harmonic mean of the PPV, and sensitivity based on the selected threshold were also computed. Continuous variables are expressed as means ± standard deviation (SD). Categorical variables are expressed as numbers and percentages. Adjusted odds ratios (OR; $95\%$ CI) were calculated. For comparisons of population characteristics, the chi-square test was used for categorical variables and the unpaired Student’s t-test for continuous variables. Cox proportional hazards models were used to estimate hazard ratios (HR; $95\%$CI) for LVSD, all-cause, and cardiovascular mortality. A value of $p \leq 0.05$ was considered statistically significant. Statistical analyses were conducted using SAS 9.4 software.
## Results
The testing dataset contained 190,316 patients ($46.3\%$ females), and 8,216 patients ($4.3\%$) had LVSD. The mean age was 63.7 ± 16.3 years. The median follow-up time was 3.9 years (interquartile range 1.5–7.9 years) for testing dataset. Table 1 shows the characteristics of the patients in the training, validation, and testing sets. There were no significant differences between groups.
**Table 1**
| Characteristics | Training (n = 133,225) | Validation (n = 57,134) | Testing (n = 190,316) | P value | Additional A-1 (n = 49,564) | Additional B (n = 83,787) | Additional C (n = 1,194,982) |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Age years, mean ± SD | 63.7 ± 16.3 | 63.8 ± 16.3 | 63.7 ± 16.3 | 0.48 | 65.9 ± 14.5 | 57.1 ± 15.8 | 50.6 ± 16.1 |
| Age groups, n (%) | | | | 0.083 | | | |
| <40 | 12,483 (9.4) | 5,222 (9.1) | 17,717 (9.3) | | 2,687 (5.4) | 12,648 (15.1) | 326,015 (27.3) |
| 40–49 | 11,893 (8.9) | 5,224 (9.1) | 17,098 (9.0) | | 3,621 (7.3) | 11,796 (14.1) | 240,076 (20.1) |
| 50–59 | 21,276 (16.0) | 9,189 (16.1) | 30,251 (15.9) | | 7,687 (15.5) | 18,446 (22.0) | 242,396 (20.3) |
| 60–69 | 27,820 (20.9) | 11,741 (20.5) | 39,454 (20.7) | | 11,157 (22.5) | 16,700 (19.9) | 168,833 (14.1) |
| 70–79 | 27,298 (20.5) | 11,516 (20.2) | 38,732 (20.4) | | 11,742 (23.7) | 12,261 (14.6) | 97,981 (8.2) |
| 80+ | 32,455 (24.4) | 14,242 (24.9) | 47,064 (24.7) | | 12,670 (25.6) | 11,936 (14.2) | 119,681 (10.0) |
| Sex, n (%) | | | | 0.248 | | | |
| Female | 61,569 (46.2) | 26,233 (45.9) | 88,139 (46.3) | | 21,665 (43.7) | 43,483 (51.9) | 625,408 (52.3) |
| Male | 71,656 (53.8) | 30,901 (54.1) | 102,177 (53.7) | | 27,899 (56.3) | 40,304 (48.1) | 569,574 (47.7) |
| EF, mean ± SD | 66.8 ± 11.9 | 66.7 ± 12.0 | 66.8 ± 11.9 | 0.845 | 64.2 ± 13.9 | 69.4 ± 8.9 | |
| EF < 40%, n (%) | 5,745 (4.3) | 2,498 (4.4) | 8,216 (4.3) | 0.825 | 3,698 (7.5) | 899 (1.1) | |
| Medical history, n (%) | | | | | | | |
| Diabetes mellitus | 37,454 (28.1) | 15,898 (27.8) | 53,608 (28.2) | 0.277 | 18,730 (37.8) | 18,034 (21.5) | 127,042 (10.6) |
| Hyperlipidaemia | 5,087 (3.8) | 2,114 (3.7) | 7,074 (3.7) | 0.26 | 3,137 (6.3) | 3,517 (4.2) | 14,978 (1.3) |
| Renal disease | 20,471 (15.4) | 8,618 (15.1) | 29,100 (15.3) | 0.292 | 12,585 (25.4) | 8,073 (9.6) | 33,241 (2.8) |
| Hypertension | 71,628 (53.8) | 30,471 (53.3) | 102,060 (53.6) | 0.223 | 35,250 (71.1) | 41,790 (49.9) | 234,683 (19.6) |
| Coronary artery disease | 28,741 (21.6) | 12,309 (21.5) | 41,077 (21.6) | 0.98 | 20,394 (41.1) | 12,018 (14.3) | 20,569 (1.7) |
| Myocardial infarction | 10,186 (7.6) | 4,354 (7.6) | 14,484 (7.6) | 0.933 | 7,827 (15.8) | 1,406 (1.7) | 2,134 (0.2) |
| Without any | 41,416 (31.1) | 17,770 (31.1) | 59,113 (31.1) | 0.976 | 5,700 (11.5) | 32,207 (38.4) | 885,789 (74.1) |
| Death, n (%) | | | | | | | |
| Within 1 year | 17,379 (13.0) | 7,497 (13.1) | 24,950 (13.1) | 0.838 | 3,479 (7.0) | 2,747 (3.3) | 48,625 (4.1) |
| Within 3 years | 27,541 (20.7) | 11,786 (20.6) | 39,264 (20.6) | 0.954 | 8,868 (17.9) | 6,459 (7.7) | 80,156 (6.7) |
| Within 5 years | 33,060 (24.8) | 14,172 (24.8) | 47,189 (24.8) | 0.992 | 12,412 (25.0) | 9,301 (11.1) | 100,746 (8.4) |
| Anytime | 39,440 (29.6) | 16,917 (29.6) | 56,290 (29.6) | 0.981 | 16,727 (33.7) | 15,297 (18.3) | 139,905 (11.7) |
## Performance of the DNN models in identifying LVSD
The AUROC values of DNN-signal and DNN-image for identifying LVSD in the testing dataset were 0.95 and 0.94, respectively (Supplementary Figure 3). When selecting a threshold maximizing the Youden’s index, the overall accuracy of DNN-signal was 0.86, with a sensitivity of 0.91, specificity of 0.86, PPV of 0.22 and NPV of 0.995. The DNN-image model performed with similar robustness to DNN-signal (sensitivity, 0.91; specificity, 0.84; PPV, 0.20; NPV, 0.995). The similarly robust DNN performances across different age, sex, and comorbidity strata in both DNN-signal and DNN-image are shown in Figure 2. External validation using ECG obtained by the Philips system was conducted. The AUROC of the DNN-signal for data from Tri-service General Hospital was 0.95. The overall accuracy of DNN-signal was 0.87, with a sensitivity of 0.90, specificity of 0.87, PPV of 0.19 and NPV of 0.99. Supplementary Tables 1, 2 show the patient characteristics and the performance of DNN-signal using data from Tri-service General Hospital.
**Figure 2:** *Deep neural network sensitivity, specificity, and odds ratio for detecting LVSD across different subgroups. The neural network’s sensitivity and specificity for detecting LVSD is tabulated across subgroups. The odds ratio (OR), which is the ratio of the positive ratio [sensitivity / (1−specificity)] to the negative likelihood [(1−sensitivity) / specificity], with the 95% CI, are shown for the subgroups and overall study sample. (A) LVSD prediction using signal. (B) LVSD prediction using image.*
## Performance of the DNN models in predicting mortality
Age- and sex-weighted Kaplan–Meier curves for mortality of patients with DNN signal-predicted LVSD and echo-derived LVSD are shown in Figure 3. A total of 8,216 LVSD patients were identified using echocardiographic data, and 33,535 LVSD patients were identified using DNN-signal. DNN signal-predicted LVSD was associated with age- and sex-adjusted HRs ($95\%$ CI) of 2.57 (2.53–2.62) for all-cause mortality and 6.09 (5.83–6.37) for cardiovascular mortality at a median follow-up of 3.9 years. Echo-derived LVSD was associated with age- and sex-adjusted HRs ($95\%$ CI) of 2.68 (2.60–2.76) for all-cause mortality and 7.79 (7.39–8.22) for cardiovascular mortality. The DNN-image performed similarly to DNN-signal with age- and sex-adjusted HRs ($95\%$ CI) of 2.70 (2.66–2.75) for all-cause mortality and 6.47 (6.19–6.77) for cardiovascular mortality (Supplementary Figure 4).
**Figure 3:** *Associations of echocardiogram and DNN-signal predictions with all-cause and cardiovascular mortalities. Age- and sex-weighted Kaplan–Meier curves, death rates, and adjusted HRs (95% CI) stratified by (A) echo-derived LVSD for all-cause mortality (blue line, LVEF≥40%; yellow line, LVEF<40%), (B) DNN signal-predicted LVSD for all-cause mortality (blue line, LVEF≥40%; yellow line, LVEF<40%), (C) echo-derived LVSD for cardiovascular mortality (blue line, LVEF≥40%; yellow line, LVEF<40%), (D) DNN signal-predicted LVSD for cardiovascular mortality (blue line, LVEF≥40%; yellow line, LVEF<40%). a Adjusted K-M curves were adjusted by the inverse probability of treatment weighting, which calculated using sex and age. b The unit of incidence rate was 1,000 person-years. CI, confidence interval; DNN, deep neural network; LVEF, left ventricular ejection fraction.*
Compared with ‘true negative’ DNN predictions, ‘true positive’ DNN-signal predictions were associated with HRs ($95\%$ CI) of 3.27 (3.17–3.38) for all-cause mortality and 12.46 (11.75–13.21) for cardiovascular mortality. ‘ True positive’ DNN-image predictions were associated with HRs ($95\%$ CI) of 3.47 (3.36–3.58) for all-cause mortality and 13.8 (13.03–14.67) for cardiovascular mortality (Figure 4).
**Figure 4:** *Associations of DNN-signal and DNN-image predictions with all-cause and cardiovascular mortalities. Age- and sex-weighted Kaplan–Meier curves, death rates, and adjusted HRs (95% CI) stratified by both echocardiography and DNN (true negative: blue line, both echo-measured and DNN-predicted LVEF ≥40%; false negative: green line, echo-measured LVEF <40% and DNN-predicted LVEF ≥40%; true positive: red line, both echo-measured and DNN-predicted LVEF <40%; and false positive: yellow line, echo-measured LVEF ≥40% and DNN-predicted LVEF <40%) for all-cause and cardiovascular mortality (A) DNN-signal predictions and all-cause mortality, (B) DNN-image predictions and all-cause mortality, (C) DNN-signal predictions and cardiovascular mortality, and (D) DNN-image predictions and cardiovascular mortality. a Adjusted K-M curves were adjusted by the inverse probability of treatment weighting, which calculated using sex and age. b The unit of incidence rate was 1,000 person-years. CI, confidence interval; DNN, deep neural network; EF, ejection fraction; FN, false negative; FP, false positive; HR, hazard ratio; K-M, Kaplan–Meier; LVSD, left ventricular systolic dysfunction; No., number; TN, true negative; TP, true positive.*
Among patients with ‘false positive’ DNN prediction, a higher mortality rate was also observed during follow-up. ‘ False positive’ DNN-signal predictions were associated with HRs ($95\%$ CI) of 2.43 (2.38–2.47) for all-cause mortality and 4.78 (3.55–5.03) for cardiovascular mortality. ‘ False positive’ DNN-image predictions were associated with HRs ($95\%$ CI) of 2.57 (2.52–2.61) for all-cause mortality and 5.16 (4.92–5.42) for cardiovascular mortality (Figure 4).
## Discussion
The prevalence of LVSD ranges from 2 to $8\%$ in adults depending on the study population and cut-off value used (8–10). In both symptomatic and asymptomatic cases, LVSD is associated with increased morbidity and mortality. The Framingham cohort study showed that individuals with asymptomatic LVSD (LVEF <$40\%$) have around eight-fold increased risk of developing HF [30]. The combination of definite treatment and primary prevention of incident HF can reduce the disease burden. One such strategy is to screen for asymptomatic LVSD; however, the best method for this is unclear [11, 31, 32]. Our study demonstrated the potential of DNNs for screening asymptomatic LVSD. In addition, comprehensive real-world testing demonstrated the robustness of DNN to identify LVSD and patients at risk of future LVSD and mortality. Furthermore, we constructed DNN models based on both raw ECG signals and transformed images. In clinical settings in which raw ECG signals are not available, this method can digest ECG image tracing and provide similar performance. Consequently, the applicability of DNN-enabled ECG is broadened.
ECG is a ubiquitous and economical point-of-care diagnostic tool in cardiology. Previous research has demonstrated that LVSD might be characterized by specific ECG changes, such as Q-waves [33, 34], left bundle branch block [35], and wide QRS duration (>120 ms) [36]. However, no single feature had high enough predictive value to offer clinical utility. These various features seemed to interact in a non-linear fashion that could not be accounted for by traditional statistical methods or algorithmic approaches. DNNs afford the ability to consider complex datasets in the context of all of the contained data rather than preselected discrete data elements. Identifying these features may offer novel findings that can provide new diagnostic approaches or therapeutic targets. Finding ways to understand what drives the network’s interpretation is also the direction of future efforts.
We used DNN algorithms to perform binary classification of LVEF in a hospital-based population, with excellent performance (AUROC, 0.95) superior to known screening tests (e.g., natriuretic peptides) [11]. The DNN performed well across all age, sex, and comorbidity groups (Figure 2). In addition, the model performance was validated externally using data from the Phillips system, suggesting its robustness across different machine types. The diagnostic performance was characterized by a high NPV, which helps exclude LVSD with high confidence. The ‘false positive’ rates were high. However, we further demonstrated that ‘false positive’ DNN predictions were associated with an eight-fold increased risk of incident LVSD (confirmed by TTE), a two-fold increased risk of all-cause mortality, and a five-fold increased risk of cardiovascular mortality compared to ‘true negative’ DNN predictions. This means that DNN could detect early, subclinical, electrical or structural abnormalities shown on the ECG. These abnormalities may include cardiac arrhythmias, left ventricular deformation, valvular heart disease, or metabolic derangements and thus increase the risk of LVSD incidence and death. In this case, DNN-enabled ECG is an effective screening tool to identify patients at risk.
Several studies have demonstrated the potential of AI in turning ECGs into functional screening and diagnostic tools for various heart disorders. For instance, Mayo Clinic researchers have applied AI to automatically detect LVSD and even tried to identify atrial fibrillation through sinus rhythm. Compared with prior studies [21, 37], we not only verified the diagnostic effectiveness of AI-assisted ECG reading on LVSD screening, but also explored the use of ECGs as an outcome prediction tool with the assistance of AI. Individuals with a positive DNN prediction were associated with a two-fold increased risk of all-cause mortality and a six-fold increased risk of cardiovascular mortality at a median follow-up of 3.9 years. This finding suggested that some trivial electrical abnormalities due to metabolic or myocardial disturbances may precede LVSD. It was speculated that some of these disturbances might be irreversible or progressive, eventually causing long-term adverse effects.
While this study reveals that DNN-enabled ECG interpretation is a reliable method of detecting LVSD, the selection of target populations for screening remains to be addressed. Galasko et al. evaluated a variety of LVSD screening strategies and demonstrated that LVSD screening is more cost-effective in high-risk subjects than in the general population [38]. High-risk subjects were defined as those with hypertension, diabetes, atherosclerotic cardiovascular disease, and heavy alcohol consumpton [39]. Our research included individuals who visited the hospital for various reasons, not just for known heart disease. This hospital-based population did have higher prevalences of diabetes mellitus ($28.2\%$), hypertension ($53.6\%$), and coronary heart disease ($7.6\%$), which fits the definition of a high-risk group.
Based on this study, we propose a prototype approach for in-hospital LVSD screening. Step one involves ECG screening using the DNN-enabled classification of individuals who will undergo high-risk invasive treatment or those with pre-existing cardiovascular risk. Step two involves TTE evaluation of individuals identified as abnormal by DNN models. This DNN-enabled screening strategy offers an advantage, as ECG machines and internet services are widely available in modern hospitals, and the strategy is also financially sustainable. This DNN model also provides a potential complementary care approach to plasma natriuretic peptide measurement for primary LVSD screening. Further studies are needed to assess the impacts of the proposed DNN-enabled screening strategy on the incidence and prognosis of in-hospital HF-associated adverse events. Furthermore, a comprehensive analysis may be conducted to examine the cost-effectiveness of the proposed strategy.
In summary, DNN-enabled ECG is a valuable tool to screen for LVSD and predict outcomes. Given the low cost of DNN-enabled ECG, serial screening is possible, which also helps optimize screening strategy for LVSD without using invasive laboratory testing, particularly in settings with limited medical resources.
## Limitations of the study
There are several limitations to this study. First, some of the LVEF data used for analysis were measured using M-mode way. The major limitation of M-mode is its one dimensional nature and lack of direct spatial information. When regional LV deformation exists, the M-mode-derived LVEF is not reliable. Although most operators choose the 2D or 3D methods when performing LVEF measurements in patients with structural heart disease, we cannot completely rule out this potential bias. Second, echocardiographic parameters other than LVEF, such as left ventricular diameter, left ventricular diastolic function, right ventricular function or valvular heart disease, also affect mortality risk. However, the present study did not introduce these parameters to analyze and evaluate their impact on prognosis. Further research should be conducted to assess the differences between clinical characteristics of patients with DNN-predicted LVSD compared to those without DNN-predicted LVSD. Third, the study was conducted in an academic medical center in patients with more complex diseases. The primary analysis consisted of patients with a higher prevalence of HF and other cardiovascular comorbidities, whom clinicians identified as needing a TTE evaluation. Considering these cohort characteristics, the findings may not be generalizable to relatively healthy and truly asymptomatic populations. To verify the generalizability of our DNN models, we conducted multiple additional analyses in more than 1 million patients with different clinical characteristics. In addition, the stratified analysis of patients without known comorbidities showed a similar performance of the models. Finally, although the sensitivity and specificity were both satisfying in our study, we observed a relatively lower PPV. The performance of PPV is highly correlated to the proportion of positive subjects in the testing group. The low likelihood of LVSD ($4.3\%$) in testing dataset caused a low PPV. Despite this, an appropriate sensitivity is more critical in applying ECG as an LVSD screening tool. The purpose of this screening tool is to detect all potential subjects who are at risk of developing LVSD for following echocardiogram exams.
## Conclusion
The established DNN algorithms in this study enable rapid LVSD detection and represent an essential step in transforming the ECG into an effective, real-time screening tool. Its ability to predict LVSD incidence and long-term mortality may help stratify patient risk and initiate relevant interventions. With good accuracy and accessibility, DNN-enabled ECG has the potential to optimize the screening process for LVSD among at-risk populations and to advance HF care significantly.
## 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 Chang Gung Medical Foundation—Institutional Review Board. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
M-SW and C-FK conceived and designed the study. Y-CHu and Y-CHs did the literature search, acquired data, and wrote the manuscript. C-HL, RT, and J-SC did the statistical analyses. Y-CHs and Z-YL developed, trained, and applied the deep neural network. J-SC prepared the figures and tables. C-FK accessed and verified the data. H-TL, W-CL, H-TW, P-CC, C-CC, C-CW, and M-SW provided the commentary. All authors contributed to the interpretation of data and the revision of the manuscript, and approved the final manuscript.
## Funding
This work was supported by the Ministry of Science and Technology of Taiwan (grant number MOST 109-2321-B-182A-007, MOST 110-2314-B-182A-123, and MOST 110-2745-B-075A-001) and Chang Gung Memorial Hospital (grant number CLRPG3H0013, CORPG3L0161, and CORPG3L0461). We were also given methodological assistance from the University of Nottingham.
## 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.2023.1070641/full#supplementary-material
## References
1. Ponikowski P, Anker SD, AlHabib KF, Cowie MR, Force TL, Hu S. **Heart failure: preventing disease and death worldwide**. *ESC Heart Fail* (2014) **1** 4-25. DOI: 10.1002/ehf2.12005
2. Echouffo-Tcheugui JB, Erqou S, Butler J, Yancy CW, Fonarow GC. **Assessing the risk of progression from asymptomatic left ventricular dysfunction to overt heart failure: a systematic overview and meta-analysis**. *JACC Heart Fail* (2016) **4** 237-48. DOI: 10.1016/j.jchf.2015.09.015
3. Yusuf S, Pitt B, Davis CE, Hood WB, Cohn JN. **Effect of enalapril on mortality and the development of heart failure in asymptomatic patients with reduced left ventricular ejection fractions. [published correction appears in N Engl J Med 1992 Dec 10;327(24):1768]**. *N Engl J Med* (1992) **327** 685-91. DOI: 10.1056/NEJM199209033271003
4. Jong P, Yusuf S, Rousseau MF, Ahn SA, Bangdiwala SI. **Effect of enalapril on 12-year survival and life expectancy in patients with left ventricular systolic dysfunction: a follow-up study**. *Lancet* (2003) **361** 1843-8. DOI: 10.1016/S0140-6736(03)13501-5
5. Dahlöf B, Devereux RB, Kjeldsen SE, Julius S, Beevers G, de Faire U. **Cardiovascular morbidity and mortality in the losartan intervention for endpoint reduction in hypertension study (LIFE): a randomised trial against atenolol**. *Lancet* (2002) **359** 995-1003. DOI: 10.1016/S0140-6736(02)08089-3
6. Exner DV, Dries DL, Waclawiw MA, Shelton B, Domanski MJ. **Beta-adrenergic blocking agent use and mortality in patients with asymptomatic and symptomatic left ventricular systolic dysfunction: a post hoc analysis of the studies of left ventricular dysfunction**. *J Am Coll Cardiol* (1999) **33** 916-23. DOI: 10.1016/s0735-1097(98)00675-5
7. Colucci WS, Kolias TJ, Adams KF, Armstrong WF, Ghali JK, Gottlieb SS. **Metoprolol reverses left ventricular remodeling in patients with asymptomatic systolic dysfunction: the REversal of VEntricular remodeling with Toprol-XL (REVERT) trial**. *Circulation* (2007) **116** 49-56. DOI: 10.1161/CIRCULATIONAHA.106.666016
8. Wang TJ, Levy D, Benjamin EJ, Vasan RS. **The epidemiology of "asymptomatic" left ventricular systolic dysfunction: implications for screening**. *Ann Intern Med* (2003) **138** 907-16. DOI: 10.7326/0003-4819-138-11-200306030-00012
9. 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 [published correction appears in Eur Heart J. 2021 Oct 14;]**. *Eur Heart J* (2021) **42** 3599-726. DOI: 10.1093/eurheartj/ehab368
10. Heidenreich PA, Bozkurt B, Aguilar D, Allen LA, Byun JJ, Colvin MM. **2022 AHA/ACC/HFSA guideline for the management of heart failure: a report of the American College of Cardiology/American Heart Association joint committee on clinical practice guidelines [published correction appears in circulation. 2022 May 3;145(18):e1033] [published correction appears in circulation. 2022 Sep 27;146(13):e185]**. *Circulation* (2022) **145** e895-e1032. DOI: 10.1161/CIR.0000000000001063
11. Vasan RS, Benjamin EJ, Larson MG, Leip EP, Wang TJ, Wilson PWF. **Plasma natriuretic peptides for community screening for left ventricular hypertrophy and systolic dysfunction: the Framingham heart study**. *JAMA* (2002) **288** 1252-9. DOI: 10.1001/jama.288.10.1252
12. Betti I, Castelli G, Barchielli A, Beligni C, Boscherini V, de Luca L. **The role of N-terminal PRO-brain natriuretic peptide and echocardiography for screening asymptomatic left ventricular dysfunction in a population at high risk for heart failure. The PROBE-HF study**. *J Card Fail* (2009) **15** 377-84. DOI: 10.1016/j.cardfail.2008.12.002
13. Suganyadevi S, Seethalakshmi V, Balasamy K. **A review on deep learning in medical image analysis**. *Int J Multimed Inf Retr* (2022) **11** 19-38. DOI: 10.1007/s13735-021-00218-1
14. Li X, Zhu D, Levy P. **Predicting Clinical Outcomes with Patient Stratification via Deep Mixture Neural Networks**. *AMIA Jt Summits Transl Sci Proc* (2020) **2020** 367-76. PMID: 32477657
15. Collobert R, Weston J. **A unified architecture for natural language processing: deep neural networks with multitask learning**. (2008) 160-167
16. Grebner C, Matter H, Kofink D, Wenzel J, Schmidt F, Hessler G. **Application of deep neural network models in drug discovery programs**. *ChemMedChem* (2021) **16** 3772-86. DOI: 10.1002/cmdc.202100418
17. LeCun Y, Bengio Y, Hinton G. **Deep learning**. *Nature* (2015) **521** 436-44. DOI: 10.1038/nature14539
18. Lindsey R, Daluiski A, Chopra S, Lachapelle A, Mozer M, Sicular S. **Deep neural network improves fracture detection by clinicians**. *Proc Natl Acad Sci U S A* (2018) **115** 11591-6. DOI: 10.1073/pnas.1806905115
19. Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A. **Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs**. *JAMA* (2016) **316** 2402-10. DOI: 10.1001/jama.2016.17216
20. Ardila D, Kiraly AP, Bharadwaj S, Choi B, Reicher JJ, Peng L. **End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography [published correction appears in Nat Med. (2019);25(8):1319]**. *Nat Med* (2019) **25** 954-61. DOI: 10.1038/s41591-019-0447-x
21. Attia ZI, Kapa S, Lopez-Jimenez F, McKie PM, Ladewig DJ, Satam G. **Screening for cardiac contractile dysfunction using an artificial intelligence-enabled electrocardiogram**. *Nat Med* (2019) **25** 70-4. DOI: 10.1038/s41591-018-0240-2
22. Yao X, Rushlow DR, Inselman JW, McCoy RG, Thacher TD, Behnken EM. **Artificial intelligence-enabled electrocardiograms for identification of patients with low ejection fraction: a pragmatic, randomized clinical trial**. *Nat Med* (2021) **27** 815-9. DOI: 10.1038/s41591-021-01335-4
23. He K., Zhang X., Ren S., Sun J.. **Deep residual learning for image recognition**. (2016)
24. Tong Z, Aihara K, Tanaka G, Hirose A, Ozawa S, Doya K, Ikeda K, Lee M, Liu D. **A hybrid pooling method for convolutional neural networks**. *International conference on neural information processing. ICONIP 2016* (2016) 454-61
25. Tong Z, Tanaka G. **Hybrid pooling for enhancement of generalization ability in deep convolutional neural networks**. *Neurocomputing* (2019) **333** 76-85. DOI: 10.1016/j.neucom.2018.12.036
26. Gu J, Wang Z, Kuen J, Ma L, Shahroudy A, Shuai B. **Recent advances in convolutional neural networks**. *Pattern Recogn* (2018) **77** 354-77. DOI: 10.48550/arXiv.1512.07108
27. Kingma DP, Adam BJ. **A method for stochastic optimization**. *arXiv Preprint arXiv:1412.6980* (2014). DOI: 10.48550/arXiv.1412.6980
28. Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. **Dropout: a simple way to prevent neural networks from overfitting**. *J. Mach. Learn. Res.* (2014) **15** 1929-58
29. Lu TH, Lee MC, Chou MC. **Accuracy of cause-of-death coding in Taiwan: types of miscoding and effects on mortality statistics**. *Int J Epidemiol* (2000) **29** 336-43. DOI: 10.1093/ije/29.2.336
30. Wang TJ, Evans JC, Benjamin EJ, Levy D, LeRoy EC, Vasan RS. **Natural history of asymptomatic left ventricular systolic dysfunction in the community**. *Circulation* (2003) **108** 977-82. DOI: 10.1161/01.CIR.0000085166.44904.79
31. Cincin A, Ozben B, Erdogan O. **Diagnostic utility of specific electrocardiographical parameters in predicting left ventricular function**. *Exp Clin Cardiol* (2012) **17** 210-4. PMID: 23592938
32. Atherton JJ. **Screening for left ventricular systolic dysfunction: is imaging a solution?**. *JACC Cardiovasc Imaging* (2010) **3** 421-8. DOI: 10.1016/j.jcmg.2009.11.014
33. Nielsen OW, Hansen JF, Hilden J, Larsen CT, Svanegaard J. **Risk assessment of left ventricular systolic dysfunction in primary care: cross sectional study evaluating a range of diagnostic tests**. *BMJ* (2000) **320** 220-4. DOI: 10.1136/bmj.320.7229.220
34. Sheifer SE, Gersh BJ, Yanez ND, Ades PA, Burke GL, Manolio TA. **Prevalence, predisposing factors, and prognosis of clinically unrecognized myocardial infarction in the elderly**. *J Am Coll Cardiol* (2000) **35** 119-26. DOI: 10.1016/s0735-1097(99)00524-0
35. Boonman-de Winter LJ, Rutten FH, Cramer MJ, Landman MJ, Zuithoff NP, Liem AH. **Efficiently screening heart failure in patients with type 2 diabetes**. *Eur J Heart Fail* (2015) **17** 187-95. DOI: 10.1002/ejhf.216
36. Madias JE. **The resting electrocardiogram in the management of patients with congestive heart failure: established applications and new insights**. *Pacing Clin Electrophysiol* (2007) **30** 123-8. DOI: 10.1111/j.1540-8159.2007.00586.x
37. Attia ZI, Noseworthy PA, Lopez-Jimenez F, Asirvatham SJ, Deshmukh AJ, Gersh BJ. **An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction**. *Lancet* (2019) **394** 861-7. DOI: 10.1016/S0140-6736(19)31721-0
38. Galasko GI, Barnes SC, Collinson P, Lahiri A, Senior R. **What is the most cost-effective strategy to screen for left ventricular systolic dysfunction: natriuretic peptides, the electrocardiogram, hand-held echocardiography, traditional echocardiography, or their combination?**. *Eur Heart J* (2006) **27** 193-200. DOI: 10.1093/eurheartj/ehi559
39. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ. **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 [published correction appears in circulation. 2019 Sep 10;140(11):e649-e650] [published correction appears in circulation. 2020 Jan 28;141(4):e60] [published correction appears in circulation. 2020 Apr 21;141(16):e774]**. *Circulation* (2019) **140** e596-646. DOI: 10.1161/CIR.0000000000000678
|
---
title: 'Impact of ¡Míranos! on parent-reported home-based healthy energy balance-related
behaviors in low-income Latino preschool children: a clustered randomized controlled
trial'
authors:
- Sarah L. Ullevig
- Deborah Parra-Medina
- Yuanyuan Liang
- Jeffrey Howard
- Erica Sosa
- Vanessa M. Estrada-Coats
- Vanessa Errisuriz
- Shiyu Li
- Zenong Yin
journal: The International Journal of Behavioral Nutrition and Physical Activity
year: 2023
pmcid: PMC10029790
doi: 10.1186/s12966-023-01427-z
license: CC BY 4.0
---
# Impact of ¡Míranos! on parent-reported home-based healthy energy balance-related behaviors in low-income Latino preschool children: a clustered randomized controlled trial
## Abstract
### Background
Widespread establishment of home-based healthy energy balance-related behaviors (EBRBs), like diet, physical activity, sedentary behavior, screen time, and sleep, among low-income preschool-aged children could curb the childhood obesity epidemic. We examined the effect of an 8-month multicomponent intervention on changes in EBRBs among preschool children enrolled in 12 Head Start centers.
### Methods
The Head Start (HS) centers were randomly assigned to one of three treatment arms: center-based intervention group (CBI), center-based plus home-based intervention group (CBI + HBI), or control. Before and following the intervention, parents of 3-year-olds enrolled in participating HS centers completed questionnaires about their child’s at-home EBRBs. Adult-facilitated physical activity (PA) was measured by an index based on questions assessing the child’s level of PA participation at home, with or facilitated by an adult. Fruit, vegetable, and added sugar intake were measured via a short food frequency questionnaire, and sleep time and screen time were measured using 7-day logs. A linear mixed effects model examined the intervention’s effect on post-intervention changes in PA, intake of fruit, vegetable, and added sugar, sleep time, and screen time from baseline to post-intervention.
### Results
A total of 325 parents participated in the study (CBI $$n = 101$$; CBI + HBI $$n = 101$$; and control $$n = 123$$). Compared to control children, CBI and CBI + HBI parents reported decreases in children’s intake of added sugar from sugar-sweetened beverages. Both CBI and CBI + HBI parents also reported smaller increases in children’s average weekday screen time relative to controls. In addition, CBI + HBI parents reported CBI + HBI parents reported increases in children’s adult-facilitated PA, fruit and vegetable intake, and daily sleep time during weekdays (excluding weekends) and the total week from baseline to post-intervention, while children in the CBI increased sleep time over the total week compared to the children in the control group.
### Conclusions
Parent engagement strengthened the improvement in parent-reported EBRBs at home in young children participating in an evidence-based obesity prevention program in a childcare setting. Future studies should investigate equity-related contextual factors that influence the impact of obesity prevention in health-disparity populations.
### Trial registration
ClinicalTrials.gov, NCT03590834. Registered July 18, 2018, https://clinicaltrials.gov/ct2/show/NCT03590834
## Background
An increasingly sedentary society has contributed to an obesogenic environment [1] that predisposes young children to dysregulation of energy balance-related behaviors (EBRBs) and an imbalance of energy intake and expenditure [2, 3]. Consequences of this imbalance include the development of obesity (i.e., excessive weight gain) and increased risk for metabolic disorders and psychosocial-behavioral problems in youth and adulthood [4, 5]. For young children ages 3–5, primary EBRBs include dietary behaviors [6, 7], physical activity (PA; [8]), sedentary behavior [9], and sleep [10]. Early childhood is a critical stage in the formulation of healthy EBRBs [8, 11], which are heavily influenced by children’s care providers (parents, family members, and childcare providers) and surrounding sociocultural, policy, and physical environments [12, 13]. For instance, young children’s food preferences and eating habits are heavily influenced by parental feeding practices [14, 15], access to healthy foods, and the food environment [12, 16]. Children from low-income minority families are disproportionally predisposed to obesogenic environments that disadvantage favorable practice of EBRBs [6], and Latino children possess higher numbers of risk factors for dysregulation of EBRBs and obesity [17, 18].
Ongoing efforts to target EBRBs to prevent obesity in childcare settings have generated mixed results in young children ages 3–5 [19–21]. For example, a recent review of 18 studies targeting PA noted there were no indications of significant PA changes in 10 of the studies [22]. In another review of 18 studies targeting healthy eating in preschool settings, 13 reported at least one positive impact on specific foods or nutrients like fruits, vegetables, and sugar [23]. Although fewer intervention studies have focused on sleep and screen time in preschool-aged children, there is recent evidence suggesting interventions may have a positive effect on sleep or screen time in children under age five [24, 25]. Similarly, reported findings from childcare-based interventions targeting the social, cultural, and physical home environments have demonstrated promising impacts on children’s body mass index (BMI) and EBRBs [26, 27]. However, a gap remains in culturally tailored intervention to target American Latino children and their families [28, 29] who are at increased risk for obesity [30].
¡Míranos! Look at Us, We Are Healthy! (¡ Míranos!) is a multi-level obesity prevention intervention targeting multiple EBRBs that uses evidence-based strategies to reduce barriers and enhance enablers of obesity prevention for low-income Latino children enrolled in Head Start [31]. In this report, we examine the impact of ¡Míranos! on parent-reported, secondary outcomes (i.e., child’s PA, screen time, sleep, and diet), following the completion of the 8-month comprehensive intervention. The primary hypothesis of the study was that compared to children in the control group, children in the center-based intervention (CBI) or the center-based plus home-based intervention (CBI + HBI) would have significantly higher levels of parent-reported PA, sleep duration, and intake of fruit and vegetables, as well as lower levels of screen time and added sugar and sugar-sweetened beverage intake at the end of the intervention.
## Research design and intervention
This research study was a clustered randomized controlled trial conducted in 12 Head Start childcare centers from two agencies in San Antonio, Texas [32]. Head *Start is* a federally funded program that provides services in school readiness, health, and family support to children ages 3 to 5 from low-income families in the United States [33]. The study was powered on the primary outcome (BMI change) but not on the secondary outcomes (diet, PA, screen time, and sleep behaviors). The study sample included 12 Head Start Centers, 4 centers per group, with an average of 29 children per center ($$n = 444$$) at baseline to achieve $80\%$ power to detect a group difference of 0.53 in BMI change at the end of the intervention (i.e., mean change of -0.03 in the CBI group or the CBI + HBI group vs. mean change of 0.5 in the control group) using a two-sided t-test with a significance level of $5\%$, an intraclass correlation of 0.003, and a standard deviation (SD) of 1.147 (PASS Version 11).
Twelve Head Start centers from two Head Start agencies were randomly assigned to one of three conditions: the combined center-based plus home-based (CBI + HBI), the center-based only intervention (CBI), or the control condition in a 1:1:1 ratio. Agency #1 had four centers and agency #2 had eight centers. Treatment randomization was stratified by Head Start agency (agency one vs. agency two) and center size (small (≤ 2, 3-years-old classrooms) vs. large (≥ 3, 3 years-old classrooms)) and generated by the study biostatistician using R version 3.3.2 (R Development Core Team, Austria). All 3-year-old children enrolled in Head Start were eligible to participate in the study. Written parent or guardian consent was obtained for each child. All study protocols were approved by the institutional review board at the University of Texas at San Antonio (IRB# 18–187). Two cohorts of participants received the 8-month intervention (cohort 1: September 2018 to May 2019; cohort 2: September 2019 to May 2022).
The rationale and development of the intervention have been described previously [32]. Briefly, ¡Míranos! included multi-level systems changes to policy, education, and reinforcement of healthy behaviors among children and their families. In the CBI, center nutrition and PA policy modifications were implemented to ensure center environments were conducive to healthy behaviors and practices. Centers increased children’s exposure to new foods and other EBRBs by modifying meal patterns to include more fruits and vegetables, employing food tastings, adding supervised PA during transition periods, and implementing health contests (e.g., drinking water, reducing TV watching). Additionally, Head Start teachers provided age-appropriate health education throughout the center day and implemented an enhanced gross motor program. Head Start staff also participated in a voluntary staff wellness program designed to develop PA- and nutrition-related health literacy and promote a healthy lifestyle. Finally, parents received monthly newsletters containing tips to help children meet PA, nutrition, screen time, and sleep time recommendations at home and food tasting recipes that mirrored center-based food tastings.
CBI + HBI parents were invited to participate in 8 monthly education sessions with information on topics related to obesity prevention recommendations and practices and strategies to develop healthy EBRBs. Specific topics are detailed elsewhere [34]. Trained Head Start parents served as peer educators and delivered education sessions using wall posters scheduled for two days of the week during child pick-up time. Parents interacted with peer educators while viewing posters and completed a scavenger hunt form. Parents that completed the scavenger hunt received a take-home bag containing materials (e.g., health-themed storybook, bilingual family activities newsletter, developmentally appropriate interactive game) that reinforced the session topic. Following each education session, parents participated in a one-week family health challenge to improve one EBRB, of their choice, at home. Additionally, Head Start family service workers offered parenting skills training and support to promote the development of healthy EBRBs during three home visits by setting family goals and developing an action plan. The parents worked with the family service worker to evaluate and refine the action plan in follow-up visits.
Head Start staff at control centers implemented an obesity prevention program endorsed by Head Start entitled “I Am Moving, I Am Learning.” Parents of children enrolled in control centers were also invited to participate in a nutrition-themed literacy education program provided by a local grocery chain.
¡Míranos! was implemented in preschool classrooms with 3-year-olds over an 8 months (early fall to late spring). During the program implementation period, we conducted weekly, online surveys of center directors and teachers to collect specific implementation data. Via this survey, program staff kept researchers informed on various activities completed (e.g., distribution of program newsletters, health contests, food tastings). Teachers reported data on the number of children, parents, and staff who participated in the health contests, which indicated reach. Head Start staff who conducted the home visits completed a ¡Miranos! log to report the number of home visits and parent attendance at each educational session. Assessment of intervention fidelity indicated that $88\%$ of the centers distributed all newsletters to parents and caregivers each week from both CBI and CBI + HBI centers. In 6 of the 8 intervention centers ($67\%$), center directors reported that all 8 health contests were held during the year with the other 2 centers completing 4 to 5 health contests. On average, $67\%$ of parents/caregivers from the CBI + HBI centers received all 3 home visits, with $82\%$ of parents receiving at least 1 visit. On average, $92\%$ of parents/caregivers from the CBI + HBI centers attended education sessions with attendance for each session ranging from 86 to $96\%$.
## Data collection
Sources of child demographic and health history data included Head Start center records and self-administered parent questionnaires, which were available in English and Spanish. Estimates of family income and employment were derived using the child’s home address from the 2010 US Census block-level data. Parents reported child behaviors at home, including screen time, sleep, intake of fruit, vegetables, and sugar, and PA [35].
## Parent-Reported PA
Parents were asked 5 specific questions related to time spent with children in various activities during a typical day (i.e., played outside with me, played outside with neighborhood children, played outside with other family members, played on a sports team, walked with me). Responses included “None”, “30 Minutes”, “1 Hour”, “2 Hours”, “3 Hours”, and “ ≥ 4 Hours”. A factor score was generated to indicate the level of child’s participation in PA with facilitation and/or supervision of adults at home or in the community. This tool was used in a previous study but has not been validated [31].
## Screen and sleep time
Parents were instructed to document their child’s hours and minutes of screen time (i.e., video games, TV watching, phone/tablet) and sleep time, including nap time, at home for each day of the past week [36]. For each child, average daily screen time or sleep time in hours was calculated over seven days (total week), per weekday (Monday through Friday), and per weekend (Saturday and Sunday). These tools have not been validated but used previously in young children [36].
## Food intake
To assess children’s dietary intake at home, parents completed a modified version of the validated National Health and Nutrition Examination Survey (NHANES; [35]. Data were collected on 18 questions that assessed the frequency of fruits, vegetables, and sugar that children ate at home over the past month (i.e., never, 1–3 times in the last month, 1–2 per week, 3–4 per week, 5–6 times per week, 1–2 per day, 3–5 per day, or 6 or more times per day). To calculate the number of servings per day of fruit (cups), vegetables (cups), fruit and vegetables (cups), and added sugar (teaspoons), all frequencies were converted into servings per day and multiplied by standard serving sizes for age and sex [37]. The maximum allowable serving sizes set by NHANES were used for reported serving sizes that met or exceeded the maximum allowable serving size per day; this minimizes the likelihood of extreme and unrealistic serving sizes [38]. Fruits included fresh, canned, frozen fruit, and $100\%$ fruit juices. Vegetables consisted of leafy green vegetables, potatoes, beans, tomato and tomato sauce, and other vegetables. Total added sugar included added sugar from foods such as candy, cookies, cakes, pan dulce, ice cream, or other frozen desserts and sugar-sweetened beverages, including soda, honey added to coffee or tea, and sweetened fruit drinks.
Because the COVID-19 epidemic disrupted the program implementation and data collection in Spring 2020, this analysis included only data collected from August 2018 to June 2019. The COVID-19 pandemic did not affect the data collected during this period and what is reported in this manuscript.
## Statistical analysis
We employed descriptive statistics to summarize the demographic characteristics of both Head Start Centers and study participants as well as each outcome of interest (parent-reported PA, diet, screen time, and sleep time) measured at each time point (baseline vs. post-intervention). Baseline, post-intervention, and change scores (post-intervention – baseline) for each outcome of interest were compared among the three groups (CBI, CBI + HBI, Control) using the Kruskal–Wallis H test. For each outcome of interest, we used a 3-level (time nested within child, and child nested within center) linear mixed effects model (LMM) to examine the treatment effect. Two random effects were included in the LMM, one to account for the correlation among two measures nested within the same child, and the other to account for the correlation among children nested within the same center. We assumed data were missing at random. In LMM, time (baseline vs. post-intervention), treatment group (CBI vs. CBI + HBI vs. Control), the interaction between time and treatment group, and center size (large vs. small) were fixed design-related predictors that were always kept in the final model, regardless of whether they were statistically significant.
We considered the following confounders in the full LMM for each outcome of interest: child’s age at baseline, gender, race/ethnicity, asthma, mother’s education, language spoken most often at home, parent marital status, family history of diabetes, and child’s BMI status at baseline. We employed a backward model selection to remove one non-significant ($p \leq 0.05$) confounder at a time from the confounder list above, and used Akaike’s information criterion (AIC) and Bayesian information criterion (BIC) to guide the model selection process to select the final reduced model. All analyses were performed using Stata/SE (version 17).
## Results
Overall, 325 parents participated in the study and provided data on at least one of the outcomes of interest (see Fig. 1). Of these 325 parents, 311 ($88\%$) completed the survey at baseline, and 215 ($66\%$) completed the post-intervention survey. Children were primarily female ($57\%$) and Latino ($87\%$) with a mean age of 3.59 (SD = 0.29) years at baseline. Approximately $13\%$ of children had a diagnosis of asthma, $41\%$ had family members with diabetes, $65\%$ were normal weight, $15\%$ were overweight, and $17\%$ were obese (see Table 1). The majority of mothers reported attaining at least a high school degree, and English was the primary language spoken at home by the majority of the children. At baseline, there were no significant differences in children’s characteristics across the three conditions, with the exception that more children in the control group were enrolled in large-sized health centers compared to the other two groups. Table 1Demographics and characteristics of head start centers and study participatesVariablesH + CBI ($$n = 101$$)CBI ($$n = 101$$)Control ($$n = 123$$)Total ($$n = 325$$)P-valueCenter, n (%)0.02 Small65 (64.36)71 (70.3)65 (52.85)201 (61.85) Large36 (35.64)30 (29.7)58 (47.15)124 (38.15)Child age at baseline, yr0.47 Median [Q1, Q3]3.64 [3.38, 3.83]3.55 [3.36, 3.76]3.6 [3.31, 3.92]3.59 [3.34, 3.84] Mean ± SD3.6 ± 0.293.56 ± 0.263.6 ± 0.323.59 ± 0.29Child sex, n (%)0.14 Male44 (43.56)36 (35.64)60 (48.78)140 (43.08) Female57 (56.44)65 (64.36)63 (51.22)185 (56.92)Child race/ethnicity, n (%)0.15 Non-H AA7 (6.93)9 (8.91)4 (3.25)20 (6.15) Hispanic83 (82.18)86 (85.15)113 (91.87)282 (86.77) Other11 (10.89)6 (5.94)6 (4.88)23 (7.08)Asthma, n (%)8 (7.92)16 (15.84)17 (13.82)41 (12.62)0.21Mother education, n (%)0.72 Less than a High school degree11 (10.89)12 (11.88)13 (10.57)36 (11.08) High School Degree/GED46 (45.54)39 (38.61)58 (47.15)143 [44] College or Technical School Degree31 (30.69)41 (40.59)41 (33.33)113 (34.77) N/A or missing13 (12.87)9 (8.91)11 (8.94)33 (10.15)Language spoken most often at home, n (%)0.67 English58 (57.43)61 (60.4)61 (49.59)180 (55.38) Spanish or other23 (22.77)24 (23.76)31 (25.2)78 [24] English and Spanish equally13 (12.87)11 (10.89)22 (17.89)46 (14.15) Not reported7 (6.93)5 (4.95)9 (7.32)21 (6.46)Parents married, n (%)38 (37.62)36 (35.64)46 (37.4)120 (36.92)0.95Family members with diabetes, n (%)44 (43.56)44 (43.56)45 (36.59)133 (40.92)0.46BMI status0.98 Underweight3 (2.97)4 [4]4 (3.25)11 (3.4) Normal67 (66.34)62 [62]81 (65.85)210 (64.81) Overweight13 (12.87)16 [16]19 (15.45)48 (14.81) Obese18 (17.82)18 [18]19 (15.45)55 (16.98)P values are comparing the differences among the three groupsFig. 1Data Flow Diagram Table 2 reports the means and standard deviations of parent-reported PA, diet, screen time, and sleep time outcomes at baseline and post-intervention. Children in the control centers had the highest amount of screen time on weekdays (2.16 h/day) and the entire week (2.19 h/day), and the lowest amount of sleep time on weekdays (9.84 h/day) and the entire week (10.0 h/day) post-intervention. Children in the control centers also had the highest intake of total added sugar (3.39 tsp/day) at post-intervention. There were no significant differences between the three groups for any outcome at baseline or change scores. There were no significant differences in any outcome between children with completed data and children with missed post-intervention surveys (data not shown).Table 2Parent-reported child physical activity, diet, screen time, and sleep outcomes at baseline, post-intervention and change scoresVariables (per day)H + CBI ($$n = 101$$) M ± SDCBI ($$n = 101$$) M ± SDControl ($$n = 123$$) M ± SDTotal ($$n = 325$$) M ± SDP-valuePhysical ActivityAdult Facilitated PA Baseline1-0.21 ± 0.93 85-0.24 ± 0.96-0.08 ± 0.95-0.17 ± 0.950.31 Post-Intervention20.26 ± 1.120.07 ± 0.930.07 ± 1.090.14 ± 1.080.56 Change30.39 ± 1.030.30 ± 1.000.14 ± 1.050.29 ± 1.030.51DietFruit (cups) Baseline41.04 ± 0.861.60 ± 1.951.15 ± 1.071.24 ± 1.350.72 Post-Intervention51.35 ± 1.661.32 ± 1.241.39 ± 1.441.35 ± 1.460.93 Change60.42 ± 1.76-0.17 ± 1.910.14 ± 1.560.14 ± 1.770.89Vegetables (cups) Baseline70.57 ± 0.470.99 ± 1.670.64 ± 0.560.72 ± 1.011.00 Post-Intervention80.88 ± 1.340.73 ± 0.680.65 ± 0.470.76 ± 0.940.95 Change90.39 ± 1.52-0.19 ± 1.370.04 ± 0.590.01 ± 1.280.85Fruit and Vegetables (cups) Baseline101.59 ± 1.112.63 ± 3.281.84 ± 1.431.99 ± 2.110.66 Post-Intervention112.22 ± 2.692.07 ± 1.692.04 ± 1.672.12 ± 2.100.93 Change120.79 ± 2.95-0.45 ± 2.960.12 ± 1.800.19 ± 2.710.75Added Sugar (tsp) Baseline133.49 ± 4.093.86 ± 7.253.11 ± 4.153.45 ± 5.250.31 Post-Intervention142.40 ± 2.502.37 ± 2.063.39 ± 2.712.66 ± 2.440.03* Change15-0.68 ± 3.27-1.10 ± 5.58-0.50 ± 5.47-0.77 ± 4.800.28Added Sugar from food only (tsp) Baseline161.59 ± 2.522.03 ± 4.361.45 ± 2.361.66 ± 3.120.32 Post-Intervention171.41 ± 1.591.27 ± 1.351.83 ± 1.631.48 ± 1.530.06^ Change18-0.11 ± 2.50-0.50 ± 3.84-0.02 ± 3.02-0.21 ± 3.130.51Added Sugar from beverages only (tsp) Baseline191.79 ± 2.481.80 ± 3.261.71 ± 2.471.76 ± 2.720.3 Post-Intervention201.11 ± 1.641.09 ± 1.161.52 ± 1.551.22 ± 1.470.05^ Change21-0.54 ± 1.59-0.59 ± 2.45-0.53 ± 3.45-0.55 ± 2.510.33Screen timeWeekday (hrs) Baseline221.55 ± 0.901.55 ± 0.771.57 ± 0.811.56 ± 0.830.9 Post-Intervention231.55 ± 0.811.69 ± 0.852.16 ± 0.751.83 ± 0.84 < 0.001* Change240.03 ± 0.840.19 ± 0.850.60 ± 0.840.30 ± 0.88 < 0.001*Weekend (hrs) Baseline251.89 ± 1.201.86 ± 1.081.96 ± 0.961.91 ± 1.080.72 Post-Intervention261.68 ± 1.091.95 ± 1.152.25 ± 1.301.94 ± 1.190.06^ Change27-0.29 ± 1.230.11 ± 0.960.17 ± 1.49-0.03 ± 1.240.39Total Week (hrs)Baseline281.64 ± 0.931.62 ± 0.781.67 ± 0.801.64 ± 0.830.87 Post-Intervention291.60 ± 0.821.76 ± 0.872.19 ± 0.781.87 ± 0.86 < 0.001* Change30-0.30 ± 0.820.20 ± 0.800.52 ± 0.850.25 ± 0.85 < 0.001*SleepWeekday (hrs) Baseline319.83 ± 0.769.85 ± 0.669.86 ± 0.769.84 ± 0.730.99 Post-Intervention3210.1 ± 0.6910.1 ± 0.799.84 ± 0.6810.0 ± 0.720.01* Change330.27 ± 0.760.20 ± 0.780.06 ± 0.820.17 ± 0.790.19Weekend (hrs) Baseline3410.5 ± 1.110.7 ± 1.2110.9 ± 1.0910.7 ± 1.130.28 Post-Intervention3510.8 ± 1.2510.8 ± 0.9910.7 ± 1.4010.8 ± 1.200.95 Change360.23 ± 1.510.04 ± 1.36-0.06 ± 1.720.1 ± 1.510.94Total Week (hrs) Baseline3710.1 ± 0.7610.1 ± 0.7010.1 ± 0.7810.1 ± 0.750.97 Post-Intervention3810.3 ± 0.6510.3 ± 0.7710.0 ± 0.7010.2 ± 0.720.002* Change390.23 ± 0.690.15 ± 0.94-0.12 ± 0.970.08 ± 0.860.05^Entries are Mean ± SDP values are comparing the differences among the three groups based on Kruskal–Wallis H test^ 0.05 < = $p \leq 0.1$* $p \leq 0.05$Change = post-intervention—baseline1 Sample sizes are 84, 74 and 106 in the HBI&CBI, CBI and control groups, respectively2 Sample sizes are 71, 67 and 53 in the HBI&CBI, CBI and control groups, respectively3 Sample sizes are 62, 52 and 47 in the HBI&CBI, CBI and control groups, respectively4 Sample sizes are 90, 80 and 109 in the HBI&CBI, CBI and control groups, respectively5 Sample sizes are 75, 70 and 54 in the HBI&CBI, CBI and control groups, respectively6 Sample sizes are 69, 58 and 51 in the HBI&CBI, CBI and control groups, respectively7 Sample sizes are 81, 73 and 99 in the HBI&CBI, CBI and control groups, respectively8 Sample sizes are 72, 69 and 54 in the HBI&CBI, CBI and control groups, respectively9 Sample sizes are 60, 53 and 45 in the HBI&CBI, CBI and control groups, respectively10 Sample sizes are 81, 73 and 96 in the HBI&CBI, CBI and control groups, respectively11 Sample sizes are 72, 68 and 54 in the HBI&CBI, CBI and control groups, respectively12 Sample sizes are 60, 52 and 44in the HBI&CBI, CBI and control groups, respectively13 Sample sizes are 81, 79 and 102 in the HBI&CBI, CBI and control groups, respectively14 Sample sizes are 74, 71 and 54 in the HBI&CBI, CBI and control groups, respectively15 Sample sizes are 62, 58 and 49 in the HBI&CBI, CBI and control groups, respectively16 Sample sizes are 85, 79 and 105 in the HBI&CBI, CBI and control groups, respectively17 Sample sizes are 75, 71 and 55 in the HBI&CBI, CBI and control groups, respectively18 Sample sizes are 66, 58 and 51 in the HBI&CBI, CBI and control groups, respectively19 Sample sizes are 87, 81 and 109 in the HBI&CBI, CBI and control groups, respectively20 Sample sizes are 76, 71 and 55 in the HBI&CBI, CBI and control groups, respectively21 Sample sizes are 68, 60 and 52 in the HBI&CBI, CBI and control groups, respectively22 Sample sizes are 94, 97 and 118 in the HBI&CBI, CBI and control groups, respectively23 Sample sizes are 82, 80 and 102 in the HBI&CBI, CBI and control groups, respectively24 Sample sizes are 77, 76 and 97 in the HBI&CBI, CBI and control groups, respectively25 Sample sizes are 82, 70 and 83 in the HBI&CBI, CBI and control groups, respectively26 Sample sizes are 61, 59 and 52 in the HBI&CBI, CBI and control groups, respectively27 Sample sizes are 55, 47 and 37 in the HBI&CBI, CBI and control groups, respectively28 Sample sizes are 95, 98 and 118 in the HBI&CBI, CBI and control groups, respectively29 Sample sizes are 82, 80 and 102 in the HBI&CBI, CBI and control groups, respectively30 Sample sizes are 78, 77 and 97 in the HBI&CBI, CBI and control groups, respectively31 Sample sizes are 96, 98 and 116 in the HBI&CBI, CBI and control groups, respectively32 Sample sizes are 80, 79 and 96 in the HBI&CBI, CBI and control groups, respectively33 Sample sizes are 76, 78 and 91 in the HBI&CBI, CBI and control groups, respectively34 Sample sizes are 84, 70 and 89 in the HBI&CBI, CBI and control groups, respectively35 Sample sizes are 65, 64 and 49 in the HBI&CBI, CBI and control groups, respectively36 Sample sizes are 61, 48 and 35 in the HBI&CBI, CBI and control groups, respectively37 Sample sizes are 98, 98 and 117 in the HBI&CBI, CBI and control groups, respectively38 Sample sizes are 80, 79 and 96 in the HBI&CBI, CBI and control groups, respectively39 Sample sizes are 78, 78 and 92 in the HBI&CBI, CBI and control groups, respectively Findings from adjusted LMMs indicated that children in both the CBI (+ 0.31 [$95\%$ CI: 0.05, 0.57], $$p \leq 0.02$$) and the CBI + HBI (+ 0.44 [0.20, 0.68], $$p \leq 0.0001$$) groups significantly increased the index score of adult-facilitated PA participation from baseline to post-intervention while a small but not significant increase was found in control children (+ 0.14 [-0.12, 0.41], $$p \leq 0.29$$). There were no significant between-group differences in the change of adult-facilitated PA between CBI and control and CBI + HBI and control, although the trend favored both intervention groups (Table 3).Table 3Treatment effects on parent-reported change in child physical activity, diet, screen time, and sleep outcomes.1OutcomesH + CBI ($$n = 101$$)CBI ($$n = 101$$)Control ($$n = 123$$)Difference (H + CBI – Control)Difference (CBI – Control)Mean change (SE)Mean change (SE)Mean change (SE)Difference [$95\%$ CI]P valueDifference [$95\%$ CI]P valuePhysical Activity Adult Facilitated PA20.44 (0.12)*0.31 (0.13)*0.14 (0.14)0.30 [-0.06, 0.66]0.110.17 [-0.20, 0.54]0.38Diet Fruit (cups)30.35 (0.19)^-0.26 (0.20)0.20 (0.21)0.15 [-0.40, 0.71]0.59-0.46 [-1.02, 0.11]0.12 Vegetable (cups)40.31 (0.15)*-0.26 (0.16)0.03 (0.17)0.28 [-0.16, 0.73]0.21-0.29 [-0.74, 0.17]0.22 Fruit and Vegetable (cups)50.67 (0.31)*-0.56 (0.32)^0.15 (0.33)0.52 [-0.37, 1.41]0.25-0.70 [-1.61, 0.21]0.13 Added Sugar Total (tsp)6-0.89 (0.59)-1.37 (0.60)*0.18 (0.64)-1.08 [-2.78, 0.63]0.22-1.56 [-3.28, 0.16]0.08^ Added Sugar from food (tsp)7-0.17 (0.36)-0.73 (0.38)^0.31 (0.39)-0.48 [-1.53, 0.57]0.37-1.04 [-2.11, 0.04]0.06^ Added Sugar from Sugar Sweetened Beverages (tsp)8-0.60 (0.29)*-0.67 (0.31)*-0.26 (0.32)-0.34 [-1.19, 0.52]0.44-0.41 [-1.27, 0.46]0.36Screen Time Weekday (hrs)90.02 (0.09)0.16 (0.09)^0.59 (0.08)*-0.57 [-0.82, -0.33] < 0.001*-0.43 [-0.68, -0.18]0.006* Weekend (hrs)10-0.22 (0.15)0.10 (0.16)0.21 (0.17)-0.43 [-0.87, 0.02]0.06^-0.11 [-0.57, 0.35]0.64 Total Week (hrs)11-0.04 (0.09)0.16 (0.09)^0.52 (0.08)*-0.56 [-0.80, -0.31] < 0.001*-0.36 [-0.60, -0.11]0.004*Sleep Time Weekday (hrs)120.28 (0.09)*0.22 (0.09)*0.03 (0.08)0.25 [0.01, 0.48]0.04*0.19 [-0.04, 0.42]0.11 Weekend (hrs)130.23 (0.18)0.10 (0.19)-0.13 (0.20)0.36 [-0.16, 0.88]0.180.22 [-0.31, 0.76]0.41 Total Week (hrs)140.25 (0.10)*0.19 (0.10)*-0.10 (0.09)0.35 [0.09, 0.60]0.009*0.30 [0.04, 0.55]0.03*^ 0.05 < = $p \leq 0.1$* $p \leq 0.051$ All models take into account the correlations between multiple measures from the same child and multiple children from the same center and adjust for treatment, time, treatment × time, center size, and outcome-specific significant confounding variables as noted below2 Based on a linear mixed effects model of 455 observations (average observations per child = 1.5, average children per center = 37.9) adjusting for race/ethnicity; ICC = 0.47 for measures nested within children; ICC < 0.001 for children nested within centers3 Based on a linear mixed effects model of 478 observations (average observations per child = 1.6, average children per center = 39.8) adjusting for baseline asthma and mother’s education; ICC = < 0.001 for measures nested within children; ICC = 0.25 for children nested within centers4 Based on a linear mixed effects model of 448 observations (average observations per child = 1.5, average children per center = 37.3) no additional adjustments; ICC = 0.08 for measures nested within children; ICC = 0.05 for children nested within centers5 Based on a linear mixed effects model of 444 observations (average observations per child = 1.5, average children per center = 37.0) adjusting for asthma; ICC = 0.20 for measures nested within children; ICC = < 0.001 for children nested within centers6 Based on a linear mixed effects model of 461 observations (average observations per child = 1.6, average children per center = 38.4) adjusting for sex, mother’s education and language; ICC = 0.31 for measures nested within children; ICC < 0.001 for children nested within centers7 Based on a linear mixed effects model of 470 observations (average observations per child = 1.6, average children per center = 39.2) adjusting for sex and ethnicity; ICC = 0.22 for measures nested within children; ICC = 0.001 for children nested within centers8 Based on a linear mixed effects model of 479 observations (average observations per child = 1.6, average children per center = 39.9) adjusting for mother’s education and language; ICC = 0.36 for measures nested within children; ICC < 0.001 for children nested within centers9 Based on a linear mixed effects model of 572 observations (average observations per child = 1.8, average children per center = 47.8) adjusting for sex, ethnicity and mother’s language; ICC = 0.42 for measures nested within children; ICC < 0.001 for children nested within centers10 Based on a linear mixed effects model of 406 observations (average observations per child = 1.5, average children per center = 33.9) adjusting for sex and mother’s language; ICC = 0.37 for measures nested within children; ICC < 0.001 for children nested within centers11 Based on a linear mixed effects model of 575 observations (average observations per child = 1.8, average children per center = 47.9) adjusting for sex, ethnicity and mother’s language; ICC = 0.47 for measures nested within children; ICC = < 0.001 for children nested within centers12 Based on a linear mixed effects model of 565 observations (average observations per child = 1.8, average children per center = 47.1) adjusting for ethnicity and mother’s language; ICC = 0.38 for measures nested within children; ICC = 0.01 for children nested within centers13 Based on a linear mixed effects model of 421 observations (average observations per child = 1.5, average children per center = 34.1) adjusting for mother’s education and language; ICC = 0.16 for measures nested within children; ICC = 0.003 for children nested within centers14 Based on a linear mixed effects model of 568 observations (average observations per child = 1.8, average children per center = 47.3) adjusting for mother’s language; ICC = 0.22 for measures nested within children; ICC = 0.008 for children nested within centers There were no significant between-group differences for any diet outcome change score between control, CBI + HBI, or CBI, (Table 3). However, the directions of changes were in favor of children in CBI and CBI + HBI, except for fruit and vegetable intake among CBI children. However, there were significant within-group decreases in intake of total added sugar (-1.37 [$95\%$ CI: -2.55, -0.20] tsp, $$p \leq 0.022$$) and added sugar from beverages (-0.67 [-1.26, -0.07] tsp, $$p \leq 0.029$$) from baseline to post-intervention among children in the CBI. For children in the CBI + HBI, increases in vegetable (+ 0.31 [0.01, 0.61] cups, $$p \leq 0.043$$) and fruit and vegetable intake (+ 0.67 [0.06, 1.27] cups, $$p \leq 0.03$$) and a decrease in added sugar from beverages (-0.60 [-1.17, -0.02] tsp, $$p \leq 0.041$$) from baseline to post-intervention were found (Table 3).
Compared to control children (Table 3), weekday and total week screen time decreased significantly in both CBI + HBI (-0.57 [$95\%$ CI: -0.82, -0.33] hours and -0.56 [-0.80, -0.31] hours) and CBI groups (-0.43 [-0.68, -0.18] hours and -0.36 [-0.6, -0.11] hours). The difference in change in screen time from baseline to post-intervention between CBI + HBI and control groups approached significance for weekend screen time only (-0.43 [-0.87, 0.02] hours, $$p \leq 0.06$$). Within-group average weekday and total week screen time increased from baseline to post-intervention among children in CBI and control groups post-intervention. The overall trend for sleep time from baseline to post-intervention was similar, with children in both CBI + HBI and CBI groups increasing their weekday and total week sleep time, and sleep time among control children remaining the same or decreasing. Among CBI + HBI participants, there were significant increases in the change for weekday (0.25 [0.01, 0.48] hours, $$p \leq 0.04$$) and total week sleep time (0.35 [0.09, 0.60] hours, $$p \leq 0.009$$), and the increase in total week sleep time (0.30 [0.04, 0.55] hours, $$p \leq 0.03$$) reached significance in CBI children, compared to children in the control group.
## Discussion
¡Míranos! targeted low-income Latino children’s healthy EBRBs with evidence-based Head Start center policies and staff practices, and culturally tailored strategies for parental engagement [26, 27, 39]. The findings from this randomized controlled trial provide further evidence of the efficacy of early childhood obesity interventions in childcare settings on children’s EBRBs. Following the 8-month intervention, CBI + HBI parents reported positive changes in children’s at-home sleep and screen time, with similar results, but to a lesser extent in CBI children, compared to the control group. However, the effects of the intervention on changes in children’s PA and dietary outcomes were limited.
These findings are consistent with the conclusions of a recent systematic review [40] regarding the mixed results in PA and nutrition outcomes based on accelerometry and direct observations during center time. The impact of the CBI on the reported changes of EBRBs at home may be attributed to embedded intervention features. For example, childcare providers’ modeling of healthy behaviors and communication with parents, can influence children’s behaviors, including healthy eating and PA [41, 42]. In the ¡Míranos! program, teachers were not allowed to drink sugar-sweetened drinks in the presence of the children but were encouraged to taste all food at meals with children, thus modeling fruit and vegetable intake. Furthermore, various center-based activities required parental involvement or communication. The in-school fruit and vegetable intake contests, presence of healthy non-sugar sweetened beverages, and modeling of sleep and screen time included parent involvement. Food tastings conducted at school with new fruits and vegetables were accompanied by recipes provided to parents and stickers worn home by children communicated their involvement in the food tastings (e.g., “I tried carrots today!”). In the combined CBI + HBI group, parental involvement was associated with additional improvements in screen time, sleep on weekdays, and sleep for the entire week, and to a lesser extent in fruit and vegetable intake and decreased added sugar from beverages [31, 43]. The culturally tailored HBI with parent training and the parent-peer educator delivery of obesity prevention education in CBI + HBI contributed to the high level of fidelity of ¡Míranos! ( e.g., attendance in poster sessions and completion of home visits), in contrast to prior intervention efforts that have encountered barriers to engaging Latino parents [44, 45].
Family-focused interventions are efficacious in the prevention and management of childhood obesity in primary care settings [34, 46]. However, childhood obesity prevention initiatives involving families have been less effective and more challenging to implement in childcare settings, especially in low-income minority populations including Latino children [44, 45, 47]. Parent engagement and participation are critical to influencing children’s EBRBs at home in ¡Míranos!. The HBI constituted knowledge transference of all ¡Míranos!-focused behavioral outcomes through peer-led education sessions that were strengthened further in Head Start home visits to support behavior change with evidence-based strategies. Interventions with a parental component have shown better outcomes in PA, diet, and other non-anthropometric indices [48] as compared to those without parental involvement [23]. Of note, these strategies to engage the parents and families were built on the existing Head Start infrastructure and standard practice and could be scaled in other organized childcare settings.
Inadequate fruit and vegetable intake and excessive intake of added sugar are consistently documented in US children, including preschool-aged children. Approximately one-third of children aged 2–5 do not meet the fruit recommendation and 90–$98\%$ do not meet the vegetable recommendation [49]. Early childhood eating habits extend into adulthood and can contribute to long-term health effects. Head Start centers utilize the Child and Adult Care Food Program or the National School Lunch Program which mandates federal guidelines for nutritious meals and snacks. Because of these requirements, children who attend Head Start centers tend to have a higher quality diet and healthy eating habits as compared to non-Head Start preschoolers [50, 51]. Further improving the diet at home through parental involvement has the potential to significantly impact the current and future eating habits of children who attend Head Start. Children in the CBI + HBI group increased fruit intake by 0.35 cups, vegetable intake by 0.31 cups, and fruit and vegetable intake by 0.67 cups per day. A recent meta-analysis revealed multicomponent interventions increased both fruit and vegetable intake by 0.37 cups per day. Two out of the five studies contained parental components which specifically increased vegetable consumption by children [40, 52]. Children aged 2–5 have reported intakes of 1.2–1.4 cups of fruit and 0.56–0.66 cups of vegetables per day [49]. Increasing $\frac{2}{3}$ cups of fruits and vegetables per day at home, as found in this study, may aid in reaching the 2–3 cups of fruits and vegetables per day recommended by the Dietary Guidelines for Americans 2020 for this age group.
Sugar intake in children is associated with excessive weight gain. Children aged 1–5 years old reported having intakes of added sugar per day of greater than $10\%$ of daily total kilocalories from sugar [53, 54], which exceeds the World Health Organization and the Dietary Guidelines for Americans 2020 recommendations [55, 56]. Additionally, children in this age group report an average intake of 49 g per day (11.7 teaspoons) [53] which exceeds the American Heart Association recommendation of fewer than 6 tsp per day [57]. In this study, added sugar intake at home was assessed and does not account for food eaten at Head Start centers. However, menu regulations for Head Start restrict added sugar from beverages and foods, so we speculate it would likely be a minor contributor to added sugar intake. While both CBI + HBI (-0.89 tsp (3.74 g)) and CBI (-1.37 tsp (5.75 g)) reduced total added sugar from baseline to post-intervention, neither group significantly reduced total added sugar intake compared to the control.
Shorter sleep duration in preschool-aged children is associated with a risk of overweight/obesity [25]. Findings from ¡Míranos! Indicated increased weekday sleep among CBI participants, and increased weekday and total week sleep time among CBI + HBI, ranging from an additional 0.25–0.35 h/day (15–21 min/day). Interventions that improved sleep in preschool-aged children have shown an increase of 0.75 h/day (45 min/day) among children in a targeted at-home parental intervention [36] or documented improvements in the number of children who slept at least 11 h [58], while another reported no change in sleep time [59]. As indicated by the ¡Míranos! findings, improvements in sleep time as small as 15 min per day, can increase total sleep time per week by 105 min (1.75 h) and contribute to reaching the benchmark of 11 h per night among preschool-aged children.
Reducing screen time by 17 min per day in children under age 5 has been achieved through interventions in non-center settings [60]. In the current study, weekday screen time increased among all groups that is consistent with age-associated upward trends in screen time [61]. However, the increase in weekday screen time was smallest in the CBI + HBI group (0.02 h/day [1.2 min/day]) compared to CBI (0.16 h/day [9.6 min/day]) and control (0.59 h/day [35.4 min/day]). The difference in weekday screen time between CBI + HBI and control at post intervention was -0.57 h/day (34.2 min/day) and for CBI it was -0.43 h/day (25.8 min/day). Similar results were found in total week screen time for children in the CBI + HBI (-0.56 h/day, 33.6 min/day) and CBI (-0.36 h/day, 21.6 min/day) as compared to the control. Parents in both the CBI and CBI + HBI groups reported that at the end of the intervention, their children spent less than 2 h per day of screen time during the weekday, weekend, and entire week as compared to the control group which exceeded 2 h per day. Reduction in excessive screen time among children has been associated with sedentary behavior and risk for overweight/obesity [62, 63]. The outcomes of ¡Míranos! with both screen time reduction and increased sleep time provide support for parental rule-setting strategies aimed at curtailing young children’s behaviors that may have long-lasting detrimental lifestyle effects [25, 64].
However, the muted impact on screen time and sleep during the weekend days demonstrates unmet challenges in families from low socioeconomic status and racial/ethnic minorities [65]. Tandon and colleagues reported low SES families had less access to play equipment and more restrictive rules associated with PA, but more access to screens and more family screen time [65]. As Njoroge noted, parental attitudes about children’s screen time may explain racial/ethnic disparities [66]. Parents from low educational and low-income families were less likely to believe they could limit their children’s screen time and keep them busy without TV and more likely to believe preschool-aged children will benefit from educational TV programs [66]. Furthermore, increased screen time contributes to reduced sleep time or sleep quality in children [67, 68].
The lack of robust ¡Míranos! intervention effect on child PA and dietary outcomes highlights the challenges of efforts to modify the home environment for resource-dependent behaviors in low-income families [69]. Although the ¡Míranos! intervention incorporated evidenced-based strategies to build parents’ efficacy and skills to modify children’s PA and diet, the program did not provide increased access to, or financial support for, PA opportunities and healthy eating choices in the home or community, making it difficult for the low-income population to benefit from policy and environment intervention [70, 71]. For example, children in low-income families have less access to play equipment at home and to safe play environments, and at the same time have higher levels of at-home—media devices conducive to sedentary activities [65, 72]. Therefore, it is imperative to consider proportionate universalism in promoting the most appropriate solutions to address the resource-intensive challenges that are not commonly offered in obesity prevention initiatives [73].
Future research should test the feasibility if an equity-based approach to obesity prevention would address the root causes of obesity by providing population-specific interventions (i.e., removing financial barriers) aimed at ensuring all families have a fair and just opportunity to engage in PA and healthy eating practices [74, 75].
A major strength of ¡Míranos! is the high fidelity with which the multiple program components were implemented in both intervention groups. Other strengths include the incorporation of evidence-based strategies to train Head Start staff and parents in managing EBRBs guided by a sound theoretical framework [24]. Furthermore, ¡Míranos! was tailored to address the cultural, linguistic, and organizational needs of Head Start and study participants. Communications and interactions with participating families incorporated culturally and linguistically appropriate content and materials. Finally, although designed as an obesity prevention intervention, the key ¡Míranos! messages focused primarily on modifying EBRBs rather than weight reduction per se, thus avoiding stigma and victim blaming. [ 56].
The limitations of this study include participant parent-reported behaviors at home and the use of a dietary screener to collect dietary data. Measures of parent-reported behaviors at home including screen-time, sleep, and adult facilitated physical activity have been used in previous studies but not validated. Dietary screeners are short food frequency questionnaires designed to capture general dietary information aimed to the reduce respondent burden but limit the utility and interpretation of the dietary data. Dietary data are typically underreported, but some studies show that it can depend on the type of foods reported, with overreporting of healthy foods and underreporting can occur for unhealthy foods [76]. To minimize the impact of language and cultural barriers on non-English speaking Hispanic parents, survey and log questions were posed in both English and Spanish to aid Spanish-speaking parents. The COVID-19 pandemic resulted in the loss of Cohort 2 participants, reducing the study sample and potentially impacting the statistical power to detect an intervention effect at 8 months post-intervention. A further limitation was that Head Start staff and data collectors were not blinded to the study center conditions. Since multiple strategies were implemented to target each of the EBRBs in ¡Míranos!, we were not able to discern their effectiveness in affecting the behavioral outcome. Future research should examine the unique contribution and scalability of each strategy specific to the targeted population.
## Conclusion
¡Míranos!, multiple-component obesity prevention in early childcare centers with a culturally tailored home intervention was effective at improving the regulation of screen time and sleep but lacked a robust effect in modifying dietary and PA behaviors at home. The findings from this study support the role of parent education and training in changing young children’s health behaviors. However, future studies should investigate equity-related contextual factors that either enhance or mitigate the impact of obesity prevention initiatives in health-disparity populations.
## References
1. Reilly JJ. **Physical activity, sedentary behaviour and energy balance in the preschool child: opportunities for early obesity prevention**. *Proc Nutr Soc* (2008.0) **67** 317-325. DOI: 10.1017/S0029665108008604
2. De Craemer M, De Decker E, De Bourdeaudhuij I, Vereecken C, Deforche B, Manios Y. **Correlates of energy balance-related behaviours in preschool children: a systematic review**. *Obes Rev* (2012.0) **13** 13-28. DOI: 10.1111/j.1467-789X.2011.00941.x
3. Hesketh KD, Campbell KJ. **Interventions to prevent obesity in 0–5 year olds: an updated systematic review of the literature**. *Obesity* (2010.0) **18** S27-S35. DOI: 10.1038/oby.2009.429
4. McGinty SM, Osganian SK, Feldman HA, Milliren CE, Field AE, Richmond TK. **BMI trajectories from birth to young adulthood**. *Obesity (Silver Spring)* (2018.0) **26** 1043-1049. DOI: 10.1002/oby.22176
5. Simmonds M, Burch J, Llewellyn A, Griffiths C, Yang H, Owen C. **The use of measures of obesity in childhood for predicting obesity and the development of obesity-related diseases in adulthood: a systematic review and meta-analysis**. *Health Technol Assess* (2015.0) **19** 1-336. DOI: 10.3310/hta19430
6. Swinburn B, Caterson I, Seidell J, James W. **Diet, nutrition and the prevention of excess weight gain and obesity**. *Public Health Nutr* (2004.0) **7** 123-146. DOI: 10.1079/PHN2003585
7. Wang YC, Bleich SN, Gortmaker SL. **Increasing caloric contribution from sugar-sweetened beverages and 100% fruit juices among US children and adolescents, 1988–2004**. *Pediatrics* (2008.0) **121** e1604-e1614. DOI: 10.1542/peds.2007-2834
8. Reilly JJ. **Physical activity, sedentary behaviour and energy balance in the preschool child: opportunities for early obesity prevention**. *Proc Nutr Soc* (2008.0) **67** 317-325. DOI: 10.1017/S0029665108008604
9. LeBlanc AG, Spence JC, Carson V, Connor Gorber S, Dillman C, Janssen I. **Systematic review of sedentary behaviour and health indicators in the early years (aged 0–4 years)**. *Appl Physiol Nutr Metab* (2012.0) **37** 753-772. DOI: 10.1139/h2012-063
10. St-Onge MP. **The role of sleep duration in the regulation of energy balance: effects on energy intakes and expenditure**. *J Clin Sleep Med JCSM Publ Am Acad Sleep Med* (2013.0) **9** 73-80. DOI: 10.5664/jcsm.2348
11. Birch LL, Fisher JO. **Development of eating behaviors among children and adolescents**. *Pediatrics* (1998.0) **101** 539-549. DOI: 10.1542/peds.101.S2.539
12. Birch LL, Davison KK. **Family environmental factors influencing the developing behavioral controls of food intake and childhood overweight**. *Pediatr Clin North Am* (2001.0) **48** 893-907. DOI: 10.1016/S0031-3955(05)70347-3
13. Story M, Kaphingst K, French S. **The role of child care settings in obesity prevention**. *Future Child Center Future Chil David Lucile Packard Found* (2006.0) **16** 143-168. DOI: 10.1353/foc.2006.0010
14. 14.Clark HR, Goyder E, Bissell P, Blank L, Peters J. How do parents’ child-feeding behaviours influence child weight? Implications for childhood obesity policy. J Public Health. 2007;29(2):132–41.
15. Murashima M, Hoerr SL, Hughes SO, Kaplowitz SA. **Feeding behaviors of low-income mothers: directive control relates to a lower BMI in children, and a nondirective control relates to a healthier diet in preschoolers**. *Am J Clin Nutr* (2012.0) **95** 1031-1037. DOI: 10.3945/ajcn.111.024257
16. Birch LL. **Development of food acceptance patterns in the first years of life**. *Proc Nutr Soc* (1998.0) **57** 617-624. DOI: 10.1079/PNS19980090
17. Pope ML, Liu T, Getchell N. **Object-control skills in Hispanic preschool children enrolled in head start**. *Percept Mot Skills* (2011.0) **112** 193-200. DOI: 10.2466/10.11.17.24.PMS.112.1.193-200
18. Singh GK, Siahpush M, Kogan MD. **Rising social inequalities in US childhood obesity, 2003–2007**. *Ann Epidemiol* (2010.0) **20** 40-52. DOI: 10.1016/j.annepidem.2009.09.008
19. Volger S, Rigassio Radler D, Rothpletz-Puglia P. **Early childhood obesity prevention efforts through a life course health development perspective: a scoping review**. *PLoS ONE* (2018.0) **13** e0209787. DOI: 10.1371/journal.pone.0209787
20. Reilly JJ, Hughes AR, Gillespie J, Malden S, Martin A. **Physical activity interventions in early life aimed at reducing later risk of obesity and related non-communicable diseases: a rapid review of systematic reviews**. *Obes Rev* (2019.0) **20** 61-73. DOI: 10.1111/obr.12773
21. Kerr JA, Loughman A, Knox A, Koplin JJ, Allen KJ, Wake M. **Nutrition-related interventions targeting childhood overweight and obesity: a narrative review**. *Obes Rev* (2019.0) **20** 45-60. DOI: 10.1111/obr.12768
22. Jones RA, Sousa-Sá E, Peden M, Okely AD. **Childcare physical activity interventions: a discussion of similarities and differences and trends, issues, and recommendations**. *Int J Environ Res Public Health* (2019.0) **16** 4836. DOI: 10.3390/ijerph16234836
23. Ward DS, Welker E, Choate A, Henderson KE, Lott M, Tovar A. **Strength of obesity prevention interventions in early care and education settings: a systematic review**. *Prev Med* (2017.0) **95** S37-S52. DOI: 10.1016/j.ypmed.2016.09.033
24. Lewis L, Povey R, Rose S, Cowap L, Semper H, Carey A. **What behavior change techniques are associated with effective interventions to reduce screen time in 0–5 year olds? A narrative systematic review**. *Prev Med Rep* (2021.0) **23** 101429. DOI: 10.1016/j.pmedr.2021.101429
25. Miller MA, Bates S, Ji C, Cappuccio FP. **Systematic review and meta-analyses of the relationship between short sleep and incidence of obesity and effectiveness of sleep interventions on weight gain in preschool children**. *Obes Rev* (2021.0) **22** e13113. DOI: 10.1111/obr.13113
26. Morris H, Skouteris H, Edwards S, Rutherford L. **Obesity prevention interventions in early childhood education and care settings with parental involvement: a systematic review**. *Early Child Dev Care* (2015.0) **185** 1283-1313. DOI: 10.1080/03004430.2014.991723
27. Tomayko EJ, Tovar A, Fitzgerald N, Howe CL, Hingle MD, Murphy MP. **Parent involvement in diet or physical activity interventions to treat or prevent childhood obesity: an umbrella review**. *Nutrients* (2021.0) **13** 3227. DOI: 10.3390/nu13093227
28. Tamayo MC, Dobbs PD, Pincu Y. **Family-centered interventions for treatment and prevention of childhood obesity in hispanic families: a systematic review**. *J Community Health* (2021.0) **46** 635-643. DOI: 10.1007/s10900-020-00897-7
29. Wang X, Ammerman A, Orr CJ. **Family-based interventions for preventing overweight or obesity among preschoolers from racial/ethnic minority groups: a scoping review**. *Obes Sci Pract* (2022.0) **8** 371-386. DOI: 10.1002/osp4.578
30. 30.National Center for Health Statistics. (2021). National Health and Nutrition Examination Survey 2017–March 2020 Prepandemic Data Files Development of Files and Prevalence Estimates for Selected Health Outcomes. Hyattsville, MD Retrieved from https://stacks.cdc.gov/view/cdc/106273.
31. Yin Z, Parra-Medina D, Cordova A, He M, Trummer V, Sosa E. **Miranos! Look at us, we are healthy! An environmental approach to early childhood obesity prevention**. *Child Obes (Print)* (2012.0) **8** 429-439. DOI: 10.1089/chi.2011.0125
32. Yin Z, Ullevig SL, Sosa E, Liang Y, Olmstead T, Howard JT. **Study protocol for a cluster randomized controlled trial to test “¡ Míranos! Look at us, We are healthy!”–an early childhood obesity prevention program**. *BMC Pediatr* (2019.0) **19** 1-17. DOI: 10.1186/s12887-019-1541-4
33. 33.Office of Head Start. Head Start Program Performance Standards. US Department of Health and Human Services Administration for Children and Families. Washington, DC. 2016.
34. Brown CL, Perrin EM. **Obesity prevention and treatment in primary care**. *Acad Pediatr* (2018.0) **18** 736-745. DOI: 10.1016/j.acap.2018.05.004
35. Hewawitharana SC, Thompson FE, Loria CM, Strauss W, Nagaraja J, Ritchie L. **Comparison of the NHANES dietary screener questionnaire to the automated self-administered 24-hour recall for children in the healthy communities study**. *Nutr J* (2018.0) **17** 111. DOI: 10.1186/s12937-018-0415-1
36. Haines J, McDonald J, O’Brien A, Sherry B, Bottino CJ, Schmidt ME. **Healthy habits, happy homes: randomized trial to improve household routines for obesity prevention among preschool-aged children**. *JAMA Pediatr* (2013.0) **167** 1072-1079. DOI: 10.1001/jamapediatrics.2013.2356
37. 37.National Cancer Institute. Data Processing & Scoring Procedures Using Current Methods Bethesda, Maryland: U.S. Department of Health and Human Services National Institutes of Health; 2020 [May 5, 2021]. Available from: https://epi.grants.cancer.gov/nhanes/dietscreen/scoring/current/.
38. 38.National Cancer Institute. Identifying Extreme Exposure Values Bethesda, Maryland: U.S. Department of Health and Human Services National Institutes of Health; 2020 [May 5, 2021]. Available from: https://epi.grants.cancer.gov/nhanes/dietscreen/scoring/current/identify.html.
39. van de Kolk I, Goossens AJM, Gerards SMPL, Kremers SPJ, Manders RMP, Gubbels JS. **Healthy nutrition and physical activity in childcare: views from childcare managers, childcare workers and parents on influential factors**. *Int J Environ Res Public Health* (2018.0) **15** 2909. DOI: 10.3390/ijerph15122909
40. Nicklas T, Lopez S, Liu Y, Saab R, Reiher R. **Motivational theater to increase consumption of vegetable dishes by preschool children**. *Int J Behav Nutr Phys Act* (2017.0) **14** 16. DOI: 10.1186/s12966-017-0468-0
41. Benjamin-Neelon SE. **Position of the academy of nutrition and dietetics: benchmarks for nutrition in child care**. *J Acad Nutr Diet* (2018.0) **118** 1291-1300. DOI: 10.1016/j.jand.2018.05.001
42. 42.Ward S, Bélanger M, Donovan D, Carrier N. Systematic review of the relationship between childcare educators’ practices and preschoolers’ physical activity and eating behaviours. Obes Rev. 2015;16(12):1055–70.
43. 43.Cronk CE, Hoffmann RG, Mueller MJ, Zerpa-Uriona V, Dasgupta M, Enriquez F. Effects of a culturally tailored intervention on changes in body mass index and health-related quality of life of Latino children and their parents. Am J Health Promot: AJHP. 2011;25(4):e1–11.
44. Fitzgibbon ML, Stolley MR, Schiffer L, Van Horn L, KauferChristoffel K, Dyer A. **Hip-hop to health Jr. for Latino preschool children**. *Obesity* (2006.0) **14** 1616-1625. DOI: 10.1038/oby.2006.186
45. Barkin SL, Heerman WJ, Sommer EC, Martin NC, Buchowski MS, Schlundt D. **Effect of a behavioral intervention for underserved preschool-age children on change in body mass index: a randomized clinical trial**. *JAMA* (2018.0) **320** 450-460. DOI: 10.1001/jama.2018.9128
46. Seburg EM, Olson-Bullis BA, Bredeson DM, Hayes MG, Sherwood NE. **A review of primary care-based childhood obesity prevention and treatment interventions**. *Curr Obes Rep* (2015.0) **4** 157-173. DOI: 10.1007/s13679-015-0160-0
47. Strauss WJ, Nagaraja J, Landgraf AJ, Arteaga SS, Fawcett SB, Ritchie LD. **The longitudinal relationship between community programmes and policies to prevent childhood obesity and BMI in children: the healthy communities study**. *Pediatr Obes* (2018.0) **13** 82-92. DOI: 10.1111/ijpo.12266
48. Mehdizadeh A, Nematy M, Vatanparast H, Khadem-Rezaiyan M, Emadzadeh M. **Impact of parent engagement in childhood obesity prevention interventions on anthropometric indices among preschool children: a systematic review**. *Child Obes* (2019.0) **16** 3-19. DOI: 10.1089/chi.2019.0103
49. Demmer E, Cifelli CJ, Houchins JA, Fulgoni VL. **The pattern of complementary foods in American infants and children aged 0–5 Years Old—A cross-sectional analysis of data from the NHANES 2011–2014**. *Nutrients* (2018.0) **10** 827. DOI: 10.3390/nu10070827
50. 50.Lee R, Zhai F, Han W-J, Brooks-Gunn J, Waldfogel J. Head Start and children’s nutrition, weight, and health care receipt. Early Child Res Quar. 2013;28(4):723–33.
51. Erinosho T, Vaughn A, Hales D, Mazzucca S, Gizlice Z, Treadway C. **The quality of nutrition and physical activity environments of child-care centers across three states in the southern U.S**. *Prev Med* (2018.0) **113** 95-101. DOI: 10.1016/j.ypmed.2018.04.029
52. 52.Williams PA, Cates SC, Blitstein JL, Hersey J, Gabor V, Ball M, et al. Nutrition-education program improves preschoolers’ at-home diet: a group randomized trial. J Acad Nutr Diet. 2014;114(7):1001–8.
53. 53.Welsh JA, Wang Y, Figueroa J, Brumme C. Sugar intake by type (added vs. naturally occurring) and physical form (liquid vs. solid) and its varying association with children’s body weight, NHANES 2009–2014. Pediatr Obes. 2018;13(4):213–21.
54. Wang Y, Guglielmo D, Welsh JA. **Consumption of sugars, saturated fat, and sodium among US children from infancy through preschool age, NHANES 2009–2014**. *Am J Clin Nutr* (2018.0) **108** 868-877. DOI: 10.1093/ajcn/nqy168
55. 55.World Health Organization. Guideline: sugars intake for adults and children Geneva, Switzerland: World Health Organization; 2015 [cited 2021 May]. Available from: https://www.who.int/publications/i/item/9789241549028.
56. Mehdizadeh A, Nematy M, Vatanparast H, Khadem-Rezaiyan M, Emadzadeh M. **Impact of parent engagement in childhood obesity prevention interventions on anthropometric indices among preschool children: a systematic review**. *Child Obes (Print)* (2020.0) **16** 3-19. DOI: 10.1089/chi.2019.0103
57. Vos MB, Kaar JL, Welsh JA, Horn LVV, Feig DI, Anderson CAM. **Added sugars and cardiovascular disease risk in children: a scientific statement from the American heart association**. *Circulation* (2017.0) **135** e1017-e1034. DOI: 10.1161/CIR.0000000000000439
58. Woo Baidal JA, Nelson CC, Perkins M, Colchamiro R, Leung-Strle P, Kwass J-A. **Childhood obesity prevention in the Women, Infants, and children program: outcomes of the MA-CORD study**. *Obesity* (2017.0) **25** 1167-1174. DOI: 10.1002/oby.21865
59. Walton K, Filion AJ, Gross D, Morrongiello B, Darlington G, Simpson JR. **Parents and tots together: pilot randomized controlled trial of a family-based obesity prevention intervention in Canada**. *Can J Public Health* (2015.0) **106** e555-e562. DOI: 10.17269/CJPH.106.5224
60. Downing KL, Hnatiuk JA, Hinkley T, Salmon J, Hesketh KD. **Interventions to reduce sedentary behaviour in 0-5-year-olds: a systematic review and meta-analysis of randomised controlled trials**. *Br J Sports Med* (2018.0) **52** 314-321. DOI: 10.1136/bjsports-2016-096634
61. 61.Common Sense Media. The Common Sense census: media use by kids age zero to eight 2017. Common Sense Media 2017 [October 28, 2022]. Available from: https://www.commonsensemedia.org/research/the-common-sense-census-media-use-by-kids-age-zero-to-eight-2017. Accessed 28 Oct 2022.
62. Anderson SE, Whitaker RC. **Household routines and obesity in US preschool-aged children**. *Pediatrics* (2010.0) **125** 420-428. DOI: 10.1542/peds.2009-0417
63. 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** S221-S248. DOI: 10.1089/chi.2020.0139
64. 64.Jones A, Armstrong B, Weaver RG, Parker H, von Klinggraeff L, Beets MW. Identifying effective intervention strategies to reduce children’s screen time: a systematic review and meta-analysis. Int J Behav Nutr Phys Act. 2021;18(1):126.
65. Tandon PS, Zhou C, Sallis JF, Cain KL, Frank LD, Saelens BE. **Home environment relationships with children’s physical activity, sedentary time, and screen time by socioeconomic status**. *Int J Behav Nutr Phys Act* (2012.0) **9** 88. DOI: 10.1186/1479-5868-9-88
66. Njoroge WFM, Elenbaas LM, Garrison MM, Myaing M, Christakis DA. **Parental cultural attitudes and beliefs regarding young children and television**. *JAMA Pediatr* (2013.0) **167** 739-745. DOI: 10.1001/jamapediatrics.2013.75
67. Waller NA, Zhang N, Cocci AH, D’Agostino C, Wesolek-Greenson S, Wheelock K. **Screen time use impacts low-income preschool children’s sleep quality, tiredness, and ability to fall asleep**. *Child Care Health Dev* (2021.0) **47** 618-26. DOI: 10.1111/cch.12869
68. Lan Q-Y, Chan KC, Yu KN, Chan NY, Wing YK, Li AM. **Sleep duration in preschool children and impact of screen time**. *Sleep Med* (2020.0) **76** 48-54. DOI: 10.1016/j.sleep.2020.09.024
69. Perrin EM, Howard JB, Ward DS. **In the absence of clear causation, casting a wider net for prevention**. *Pediatrics* (2016.0) **138** e20162895. DOI: 10.1542/peds.2016-2895
70. Hämäläinen RM, Sandu P, Syed AM, Jakobsen MW. **An evaluation of equity and equality in physical activity policies in four European countries**. *Int J Equity Health* (2016.0) **15** 191. DOI: 10.1186/s12939-016-0481-y
71. Kumanyika SK. **A Framework for increasing equity impact in obesity prevention**. *Am J Public Health* (2019.0) **109** 1350-1357. DOI: 10.2105/AJPH.2019.305221
72. Maitland C, Stratton G, Foster S, Braham R, Rosenberg M. **A place for play? The influence of the home physical environment on children’s physical activity and sedentary behaviour**. *Int J Behav Nutr Phys Act* (2013.0) **10** 99. DOI: 10.1186/1479-5868-10-99
73. Carey G, Crammond B, De Leeuw E. **Towards health equity: a framework for the application of proportionate universalism**. *Int J Equity Health* (2015.0) **14** 81. DOI: 10.1186/s12939-015-0207-6
74. Amaro H. **The action is upstream: place-based approaches for achieving population health and health equity**. *Am J Public Health* (2014.0) **104** 964. DOI: 10.2105/AJPH.2014.302032
75. 75.Kumanyika SK. Getting to equity in obesity prevention: A new framework. Washington, DC: National Academy of Medicine; 2017.
76. Macdiarmid J, Blundell J. **Assessing dietary intake: Who, what and why of under-reporting**. *Nutr Res Rev* (1998.0) **11** 231-253. DOI: 10.1079/NRR19980017
|
---
title: 'Impact of a Mindfulness Mobile Application on Weight Loss and Eating Behavior
in People with Metabolic Syndrome: a Pilot Randomized Controlled Trial'
authors:
- Takaharu Matsuhisa
- Rieko Fujie
- Rie Masukawa
- Natsue Nakamura
- Norihisa Mori
- Kazuyuki Ito
- Yuki Yoshikawa
- Kentaro Okazaki
- Juichi Sato
journal: International Journal of Behavioral Medicine
year: 2023
pmcid: PMC10029796
doi: 10.1007/s12529-023-10173-2
license: CC BY 4.0
---
# Impact of a Mindfulness Mobile Application on Weight Loss and Eating Behavior in People with Metabolic Syndrome: a Pilot Randomized Controlled Trial
## Abstract
### Background
Weight-loss approaches involving mindfulness have been reported to reduce overeating behavior. We conducted a preliminary evaluation of the feasibility and effectiveness of a mindfulness mobile application (MMA) combined with a comprehensive lifestyle intervention (CLI) focused on weight loss and eating behaviors for people with metabolic syndrome based on post-intervention follow-up data.
### Method
Participants were randomly assigned (1:1) to a CLI group or a CLI + MMA group. Participants received weekly CLI for 13 weeks, followed by telephone counseling for 13 weeks. The CLI + MMA group also had access to the MMA. Feasibility was assessed by the number of people who refused to participate, rate of adherence to the MMA, follow-up rate, and participant satisfaction. The preliminary endpoint was weight change (at 26 weeks). Participants completed the Dutch Eating Behavior Questionnaire (DEBQ). A mixed linear model was used for efficacy analysis.
### Results
Eight of the 40 participants declined to participate. The MMA was used 4.4 ± 1.7 days per week, but the rate of adherence declined over time. The follow-up rate was $100\%$, and there was no difference in participant satisfaction between the groups. There was no significant group-by-time interaction for weight loss ($$p \leq 0.924$$), but there was a significant interaction for the DEBQ restrained eating score ($$p \leq 0.033$$).
### Conclusions
This study found that CLI plus MMA was highly feasible and moderately acceptable. There were no significant differences in weight loss between the groups, but the CLI + MMA group showed an increase in restrained eating. Further large-scale studies are needed.
### Trial Registration
Japanese University Hospital Medical Information Network (UMIN-ICDR). Clinical Trial identifier number UMIN000042626.
## Introduction
Increasing obesity rates are an international problem [1], including in Japan [2]. Obesity is a major cause of metabolic syndrome (MetS) [3]. Reducing obesity and improving MetS can improve hypertension, glucose intolerance, and dyslipidemia, and reduce the morbidity and mortality associated with cardiovascular diseases [4]. Aggressive lifestyle modification focused on weight loss and increased physical activity is the principal treatment for improving MetS [5]. Although intensive lifestyle interventions involving diet and exercise therapy are effective for people with obesity in the short term, maintaining long-term weight loss is difficult [6–8]. Different dysfunctional eating behaviors, such as binge eating, emotional eating, external eating, and eating in response to food cravings, have been linked to weight regain after successful weight loss [9]. This highlights the importance of investigating approaches to address associated psychological problems and potentially increase motivation and self-control among patients with obesity (e.g., to limit impulsive and inappropriate use of food) [10]. Psychological interventions, particularly behavioral and cognitive-behavioral strategies, have been reported to be beneficial for weight loss among adults with overweight and obesity, especially when combined with dietary and exercise strategies [11].
Mindfulness is a psychological process in which attention is paid to experiences that occur in the present moment. When developed through meditation and other disciplines, mindfulness can improve emotional control and reduce avoidant reactions to external and internal experiences [12]. Mindfulness-based interventions have been shown to have beneficial effects on weight loss and impaired eating behaviors, including improving the present-moment awareness of the sensory properties of food and reducing further food intake, and supporting decentering strategies that may help individuals resist desired foods [13]. Previous research suggested that mindfulness positively affected weight-related behaviors, such as reducing emotional and binge eating [9, 14–19], but had mixed effects on weight loss [14, 16–18, 20]. However, few studies have focused on evaluation at follow-up after an active intervention. A systematic review and meta-analysis found that a weight-loss approach involving mindfulness reduced overeating behavior and contributed to maintaining weight loss at follow-up [14]. A small-scale randomized controlled trial (RCT) showed that mindfulness combined with a standard behavioral weight loss program resulted in better weight loss, less overeating behavior, and better adherence to dietary restrictions at the 3-month follow-up after a 3-month active intervention compared with a standard behavioral weight loss program [21]. These studies suggested that mindfulness may be effective for maintaining long-term weight loss and healthy eating behavior after an active intervention.
Mindful eating (derived from mindfulness), when used to address unhealthy eating behavior, incorporates nonjudgmental awareness of physical and emotional sensations associated with eating [22]. Although research on mindful eating has been used in various ways, experimental studies are relatively limited, and it is not yet possible to conclude whether mindful eating strategies impact diet. However, some evidence suggests that certain mindful eating strategies may be promising, such as decentering and attention to the sensory properties of food [23]. A previous review showed that a mindful eating intervention reduced cravings and excess caloric intake and helped maintain continued weight loss [24].
Some studies have reported non-face-to-face weight loss interventions using mindfulness [25, 26]. One such study found that a telephone intervention using a mindfulness-based weight loss program did not improve weight loss compared with a standard weight loss program at the end of the intervention period and at the 6-month follow-up [25]. However, participants who used mindfulness in conjunction with the intervention showed decreased overeating behavior and improved mindful eating practices and mindfulness scale scores [25]. To date, the only published intervention study involving a mindfulness mobile application (MMA) for weight loss involved a general sample of students and reported a comparison with a behavioral self-monitoring electronic diary (e-diary) method [26]. Those results revealed no weight loss in the MMA group at follow-up, although improvements were observed in participants’ stress, eating behavior, mindfulness, and the frequency of mindful eating practices [26]. However, no intervention studies have examined changes in weight and eating behaviors when an MMA is used in addition to diet and exercise therapy. Mobile application-based interventions have been proposed as useful tools for weight loss [27, 28]. These interventions are portable and easy to practice anywhere [12], which may contribute to their potential role in MetS care [29]. Therefore, we aimed to explore the potential effectiveness and feasibility of using an MMA in conjunction with a comprehensive lifestyle intervention (CLI) focused on weight loss and eating behaviors, including at the post-intervention follow-up.
## Study Design
This open-label, parallel, pilot RCT included a 13-week CLI, which comprised a supportive workshop after a physical examination at a city general health center plus telephone counseling every 4 weeks for the following 13 weeks with and without MMA use. The application was provided to all participants, but the provision period was divided into two phases: 1–26 weeks and 27–52 weeks. We compared data between the two groups over 26 weeks. Participants were stratified by sex and randomly assigned to the two groups using a 1:1 ratio. After allocation, participants underwent baseline assessment, and those who did not meet the eligibility criteria were excluded from the analyses. The intervention was provided free of charge. This pilot trial was not designed to have sufficient statistical power to assess the effectiveness of the MMA intervention on weight loss and eating behavior.
## Participants
Eligibility criteria were used to select adults aged 20–75 years with MetS, as defined by the International Diabetes Federation (IDF) [30] that had smartphones with mobile applications available. The exclusion criteria were a history of serious heart disease or other conditions that prohibited exercise therapy, severe depression, severe anxiety disorder, severe somatoform disorder, and psychotic symptoms. We recruited 1031 participants (663 females) by mail who met the eligibility criteria from among individuals who underwent physical examinations at a general healthcare center in Kasugai City, Japan. Participants were recruited from April 2020 to January 2021. Three evaluation sessions (baseline, week 13, and week 26) were conducted at the general healthcare center.
## Randomization, Blinding, and Allocation Concealment
After orientation, participants who consented to participate in this study were randomly assigned to one of the two groups using the envelope method stratified by sex. The envelopes were opened in sequence in front of the participants. Because this was a pragmatic comparative study rather than a placebo-controlled trial, it was not possible to blind participants or the health workers providing the CLI. However, the risk for detection bias was minimized because all sessions were conducted independently, and the nutritionists and public health nurses in charge of the CLI were not involved in evaluating the results. The statistician always conducted outcome assessments in a separate room and was blinded to group allocation.
## Interventions: CLI
The CLI involved a 1-h lesson once a week for 13 weeks from April to June 2021 and was performed with all participants together at designated times. The sessions comprised lectures by public health nurses and nutritionists, nutritional guidance, and lectures and exercise guidance by health exercise instructors. Nutritional guidance was based on the portion control method using the Healthy Plate approach [31], with a daily diet of 1200–1500 kcal plus dairy products and fruits, with a target of $50\%$ carbohydrate, 25–$30\%$ protein, and 20–$25\%$ fat. The exercise instruction included open-eyed one-legged stands, heel lifts, squats, arm-leg crossing, push-ups, and sit-ups, with an increased load each time. Participants recorded their weight, exercise, meals, and snacks in a notebook each day. They were interviewed and given written feedback on their notebooks by public health nurses and nutritionists when they attended the sessions. As a state of emergency in response to COVID-19 was declared in Aichi Prefecture from May 12 to June 20, 2021 [32], six of the sessions scheduled during this time were canceled. Therefore, only six sessions were held. During the state of emergency, nutritionists and public health nurses called participants once a week to check on their condition. After completion of the active intervention, the nutritionists and public health nurses conducted telephone counseling every 4 weeks from July to September (weeks 14–26) to check whether participants were able to maintain the diet and exercise program and confirm that the MMA group was using the application. These healthcare workers provided instruction to participants based on the transtheoretical model of health behavior change [33].
## Experimental Intervention: CLI + MMA
The CLI + MMA group performed the CLI and practiced mindfulness every day using the mobile application. Participants’ MMA use was checked and counseling provided by the nutritionists and public health nurses (who were not experienced in mindfulness practices) when they delivered sessions or telephone counseling. We used the MMA developed by the Relook unit of ARETECO HOLDINGS LTD (https://relook.jp/) for this study, which comprised a 26-week MMA program for weight loss. The present researchers were not involved in the creation of this program. The application was provided free of charge and participants downloaded the application onto their smartphones. Details of the mobile application content are shown in Table 1. The CLI + MMA intervention comprised an average of 453 s (306–889 s) each day, and the application included a reminder notice once-a-day. Participants initially practiced basic mindfulness breathing exercises following audio navigation in Japanese using the mobile application. After practicing mindfulness breathing, participants practiced ways to be mindful in various daily situations. The MMA was provided free of charge to the CLI group for 26 weeks after the evaluation was completed to ensure equality between the two groups. Table 1Content of the mindfulness mobile applicationSessionThemeSession purposeContent of the session1st(weeks 1–6)Learn about and experience mindfulnessTo grasp a sense of mindfulness through both intellectual knowledge of mindfulness and experienceTo develop a habit of daily mindfulness meditationTo learn how to prepare for mindfulnessMindful eating (7 days), 3-min breathing exercise (3 days), 5-min breathing exercise (14 days), silent contemplation (14 days), body scan meditation (2 days), SOBER breathing space (2 days)2nd(weeks 7–12)Increase attention and awarenessTo become aware of thoughts, judgments, emotions, and feelingsTo pay attention more easily than in weeks 1–6Mindful eating (5 days), body scan meditation (5 days), walking meditation (7 days), silent contemplation (14 days), RAIN (6 days), SOBER breathing space (5 days)3rd(weeks 13–18)Practice mindfulness on a daily basisTo implement mindful coping strategies for stressTo expand the attention beyond breath and body, including sound, thought, and beingMindful eating (6 days), self-soothing touch (7 days), self-compassion break (14 days), RAIN (8 days), silent contemplation (7 days)4th(weeks 19–26)Review mindfulnessTo memorize the mindfulness mindsetTo know how to perform mindfulnessTo understand the importance of continued mindfulnessMindful eating (2 days), 3-min breathing exercise (4 days), 5-min breathing exercise (11 days), RAIN (3 days), SOBER breathing space (3 days), silent contemplation (11 days), body scan meditation (9 days), self-soothing touch (3 days), self-compassion break (5 days), walking meditation (5 days)SOBER Stop Observe Breathe Expand, and Respond, RAIN Recognize, Allow, Investigate, and Nurture
## Acceptability, Adherence, and Feasibility
These outcomes were measured by the number of people who refused to consent to the study after receiving orientation, the rate of adherence to the application (number of days application implementation was completed), follow-up rate, participant satisfaction, and number of adverse events. Data on patient adherence to the application were collected by ARETECO HOLDINGS LTD, which was performed automatically by the app and recorded every time a participant completed more than $90\%$ of an application session. ARETECO HOLDINGS LTD provided the research team with a list of codes, which allowed the researchers to see a log of the actual time the application was used. Satisfaction was assessed at week 26, with reference to previous studies [25], to compare the satisfaction of participants in the two groups regarding their overall impression of the program using a 5-point Likert scale (1 = “very dissatisfied” to 5 = “very satisfied”). We also assessed participants’ willingness to recommend the program to friends, and whether the program helped them consume a healthy diet using a 5-point Likert scale (1 = “very negative” to 5 = “very positive”). We systematically tracked adverse events at each weekly session and followed monthly telephone counseling.
After 6 months of evaluation, data were available for all 30 participants (follow-up rate: $100\%$). No adverse events were observed. The application use logs revealed the application was used 4.4 ± 1.7 days per week on average; however, the number of days of application use decreased over time (Fig. 2). In the first week, the average number of days of application use was 6.63 ± 0.78, which decreased to 4.06 ± 2.56 in week 13, and 2.00 ± 2.34 in week 26. The number of participants who did not use the MMA by intervention week was 0 at week 1, but increased to two of 16 ($17.5\%$) at week 13, and eight of 16 ($50\%$) at week 26. Sixteen CLI + MMA participants (response rate = $100\%$) and 14 CLI participants (response rate = $100\%$) reported satisfaction with their participation in the 6-month survey. The CLI + MMA and CLI groups had similar mean program satisfaction ratings for overall impressions of the program (CLI + MMA: 3.4 ± 0.9; CLI: 3.5 ± 1.2), recommending the program to friends who wanted to lose weight (CLI + MMA: 3.4 ± 1.0; CLI: 3.5 ± 1.0), and helping them to eat healthily (CLI + MMA: 3.8 ± 0.7; CLI: 3.4 ± 0.9).Fig. 2Average number of days the application was used per week. The average number of days spent using the application per week declined with time
## Assessment procedures
The primary, secondary, and exploratory outcomes were evaluated at weeks 0, 13, and 26.
## Primary Outcome Measures
The primary outcome of using the MMA was the rate of change in body weight. Body weight was measured in units of 0.1 kg, while wearing test clothes.
## Secondary Outcome Measures
Eating behavior.
The modified Japanese version of the Dutch Eating Behavior Questionnaire (DEBQ) was used to evaluate eating behavior. The DEBQ has three eating behavior subscales: restrained eating, emotional eating, and external eating [34]. The Japanese version of the DEBQ has been reviewed for reliability [35]. The 10-item restrained eating subscale assesses intentions and behaviors regarding restricting food intake because of weight concerns. The 13-item emotional eating subscale rates overeating behaviors triggered by negative emotions, such as anger, boredom, anxiety, and fear. The 10-item external eating subscale measures eating in response to food-related stimuli, such as the smell and taste of food, seeing other people eating, and seeing food being prepared. Participants responded to each item on a 5-point scale from “never” (1 point) to “always” (5 points). Higher scores indicated greater endorsement of that eating behavior.
## Exploratory Outcome Measures
As exploratory outcome measures, we assessed body mass index (BMI), body fat percentage, abdominal circumference, body blood pressure, and blood test parameters (total cholesterol, high-density lipoprotein, low-density lipoprotein, triglyceride, fasting blood sugar, hemoglobin A1c, blood urea nitrogen, creatinine, cystatin C, estimated glomerular filtration rate), self-reported physical activity using the Japanese version of the International Physical Activity Questionnaire (short version) [36], and the Motivation to *Live a* Healthy Diet scale [37]. Because of the large amount of data collected, this paper only describes the results for body weight and eating behavior, as no association between the other items and the effects of mindfulness was shown.
## Statistical Analysis
Statistical analyses were performed using SPSS version 27.0 (IBM Corporation, Armonk, NY, USA). After group allocation and before the intervention, one participant refused to participate in this study. Another participant was excluded because they did not meet the IDF criteria for MetS at the baseline assessment. Therefore, the modified intention-to-treat method was applied to investigate the treatment effects. Continuous variables measured at baseline were described using means and standard deviations and compared between treatment groups using two-sample t-tests. Categorical variables were described using frequencies and percentages, and comparisons between treatment groups were made using Fisher’s exact tests (two-sided). We used linear mixed models to analyze the effects of the intervention on body composition and DEBQ scores, including the endpoint of weight change. The model included fixed effects of group, time, and group-by-time interactions. The linear modeling analysis of participants who completed the MMA on at least 5 out of 7 days (per protocol) supported the results and the conclusions of the linear mixed modeling for the full sample. In this study, we only report the results of the linear mixed modeling for the full sample. It is considered more appropriate to use the standard deviation of the baseline values to reflect clinically meaningful differences based on the distribution of the population [38]. The effect size was calculated as the difference in mean changes at week 26 between the two groups, standardized to the pooled standard deviation of the baseline values. As an additional analysis, the rate of weight loss, changes in DEBQ score, and the frequency of application use were analyzed using Pearson’s product-moment correlation coefficients.
## Participant Flow: Screening and Study Acceptability
In total, 40 people attended the study orientation, of which 8 ($20\%$) declined to participate in this study. The remaining 32 participants ($80\%$) were randomly assigned to the study groups: 17 to the CLI + MMA group and 15 to the CLI group. One participant refused to participate in this study because of fear of infection with COVID-19 before the intervention started. Another participant did not meet the inclusion criteria for baseline measurements. These two participants were therefore excluded from the analyses, and background characteristics and results are reported for 30 participants. The enrollment, randomization, and retention processes are shown in Fig. 1.Fig. 1Flow diagram of trial participants
## Participants’ Characteristics
Differences between the two groups at baseline are shown in Table 2. Most participants ($83\%$, $$n = 25$$) were female and the mean age was 69.3 ± 5.0 years. Baseline measurements for the two groups showed a significant difference in weight ($$p \leq 0.035$$), with mean weights of 61.7 ± 7.5 kg and 69.3 ± 11.0 kg in the CLI + MMA and CLI groups, respectively. Table 2Baseline characteristics of participants randomly assigned to the comprehensive lifestyle intervention (CLI) and the CLI plus mindfulness mobile application group, assuming intention-to-treatCLI + mindfulness($$n = 16$$)CLI ($$n = 14$$)p-value* (two-tailed)Demographics Gender, female13 (81.3)12 (85.7)1.000 Age (years)69.9 ± 3.668.7 ± 6.30.535Baseline measures Height, cm154.9 ± 5.3158.1 ± 7.50.182 Weight, kg61.7 ± 7.569.3 ± 11.00.035 Body mass index, kg/m225.7 ± 2.227.6 ± 3.20.055 Abdominal circumference, cm93.6 ± 6.599.0 ± 9.00.065 Body fat percentage, %34.3 ± 4.438.2 ± 6.70.069 Systolic blood pressure, mmHg135.9 ± 17.5133.0 ± 11.90.608 Diastolic blood pressure, mmHg76.2 ± 11.177.6 ± 11.40.739 Total cholesterol, mg/dL211.9 ± 45.9208.6 ± 33.70.826 LDL cholesterol, mg/dL123.4 ± 36.8121.5 ± 32.80.881 HDL cholesterol, mg/dL68.0 ± 15.162.6 ± 13.00.305 Triglyceride, mg/dL171.4 ± 62.4195.4 ± 94.90.413 Blood sugar, mg/dL115.8 ± 23.3130.9 ± 53.90.316 Hemoglobin A1c, %6.0 ± 0.56.6 ± 1.20.107 *Blood urea* nitrogen, mg/dL17.5 ± 3.818.6 ± 3.60.393 Creatinine, mg/dL0.71 ± 0.110.72 ± 0.250.868 Cystatin C, mg/L0.85 ± 0.140.84 ± 0.160.835 eGFR, mL/min/1.73 m266.9 ± 8.868.9 ± 15.80.678 Restrained eating (DEBQ)3.27 ± 0.693.54 ± 0.530.251 Emotional eating (DEBQ)1.95 ± 0.752.10 ± 0.840.614 External eating (DEBQ)2.97 ± 0.732.84 ± 0.540.581 Total physical activity, MET-minutes/week2319.6 ± 3017.81487.2 ± 1402.50.353 Sedentary time, minutes384.4 ± 223.2345.0 ± 216.60.629Presence of chronic diseases None6 (37.5)4 (28.6)0.709 All10 (62.5)10 (71.4)The number of times the course was taken First10 (62.5)6 (42.9)0.464 Plural6 (37.5)8 (57.1)Data are presented as mean ± standard deviation or n (%). LDL, low-density lipoprotein; HDL, high-density lipoprotein; eGFR, estimated glomerular filtration rate; DEBQ, Dutch Eating Behavior Questionnaire; MET, metabolic equivalent*Group comparisons by two-sample t-tests for continuous data and Fisher’s exact test (two-sided) for categorical data
## Weight Loss
Compared with baseline, the mean weight loss at week 26 was significantly reduced by 2.1 kg (standard error (SE) 0.6, $$p \leq 0.004$$; $3.2\%$, SE 0.9) in the CLI group and 2.3 kg (SE 0.2, $p \leq 0.001$; $3.7\%$, SE 0.4) in the CLI + MMA group. There was no significant group-by-time interaction between mindfulness use and weight loss ($$p \leq 0.924$$). These results are shown in Table 3 and Fig. 3.Table 3Differences in primary and secondary outcomes (intention-to-treat) by group over time based on linear mixed modeling (CLI: $$n = 14$$; CLI + MMA: $$n = 16$$)Outcome variableBy intervention groupAssessmentsp-valuesEffect sizeWeek 0Week 13Week 26Group effectTime effectGroup-by-time interactionBaseline to 6 monthsMean (standard error)Primary outcome Body weight, kgCLI69.3 (2.9)67.6 (3.0)67.2 (3.2)0.039 < 0.0010.9240.02CLI + MMA61.7 (1.9)60.1 (2.0)59.5 (2.0)Secondary outcomes DEBQ Restrained eatingCLI3.54 (0.14)3.52 (0.15)3.42 (0.16)0.8500.3540.0330.78CLI + MMA3.27 (0.17)3.46 (0.13)3.64 (0.16) Emotional eatingCLI2.10 (0.22)1.95 (0.17)2.02 (0.25)0.6290.4120.4190.10CLI + MMA1.96 (0.19)1.98 (0.14)1.79 (0.15) External eatingCLI2.84 (0.15)2.72 (0.16)2.87 (0.18)0.9300.1880.2690.51CLI + MMA2.97 (0.18)2.73 (0.15)2.68 (0.16)DEBQ Dutch Eating Behavior Questionnaire, CLI comprehensive lifestyle intervention, MMA mindfulness mobile applicationFig. 3Change in body weight (%) (A) and restrained eating score (DEBQ) (B) for CLI versus CLI + MMA
## DEBQ
Analysis of DEBQ results showed the restrained eating score significantly increased from 3.27 (SE 0.17) to 3.64 (SE 0.16) at week 26 in the CLI + MMA group ($$p \leq 0.010$$), but no change was observed in the CLI group (baseline: 3.54, SE 0.14; week 26: 3.42, SE 0.16). The group-by-time interaction was significant ($$p \leq 0.033$$) (Fig. 3B and Table 3). The DEBQ emotional eating score showed a decreasing trend in the CLI + MMA group from 1.96 (SE 0.19) to 1.79 (SE 0.15) at week 26, but the difference was not significant ($$p \leq 0.311$$). There was no change in emotional eating score in the CLI group (baseline: 2.10, SE 0.22; week 26: 2.02, SE 0.25) and no significant group-by-time interaction ($$p \leq 0.419$$). The external eating score showed a decreasing trend in the CLI + MMA group from 2.97 (SE 0.18) to 2.68 (SE 0.16) at week 26, but this difference was not significant ($$p \leq 0.098$$). No changes were observed in the CLI group (baseline: 2.84, SE 0.15; week 26: 2.87, SE 0.18) and the group-by-time interaction was not significant ($$p \leq 0.269$$) (Table 3).
## Mechanisms/Moderators
We examined potential mechanisms for mindfulness treatment by correlating the change between baseline and follow-up at week 26 (follow-up score minus baseline score) for the rate of weight loss and restrained eating scores (Fig. 4A). Overall, an increase in restrained eating scores was correlated with higher rates of weight loss (Pearson’s correlation coefficient: 0.474; Fig. 4A), especially in the CLI + MMA group (Pearson’s correlation coefficient: 0.634; Fig. 4A, black dots) compared with the CLI group (Pearson’s correlation coefficient: 0.417; Fig. 4A, orange dots). We also investigated mindfulness engagement to test whether more engagement with the MMA was associated with more weight loss. In the CLI + MMA group, the number of days spent using the MMA was significantly and positively associated with increased weight loss (Pearson’s correlation coefficient: 0.598, $$p \leq 0.014$$; Fig. 4B).Fig. 4Correlation with weight loss: change in restrained eating score (DEBQ) (A) and application use days (B). A The CLI + MMA group: black dots; the CLI group: orange dots
## Discussion
The present study showed the combination of CLI and MMA did not affect participants’ satisfaction with the program over 26 weeks. The high follow-up rate in this study allowed us to successfully track participants, but the rate of adherence to the MMA was low. No adverse events were observed. The preliminary results indicated that the MMA intervention did not result in greater weight loss at 6 months, but significantly increased restrained eating behavior compared with the CLI group. A positive relationship was found between the rate of weight loss and increased restrained eating scores. There was also a positive relationship between the rate of weight loss and frequency of using MMA. These results suggested the MMA was potentially effective for people with MetS.
Weight loss results did not significantly differ between the CLI + MMA and CLI groups at follow-up from the end of the CLI. Overall, both groups exhibited modest weight loss ($3.5\%$ of body weight, 2.2 kg on average). These effects on body weight were consistent with the results of previous meta-analyses, which reported small effect sizes for weight loss with mindfulness-based interventions [17–19]. A meta-analysis demonstrated moderate effects on weight loss with these interventions, but the level of evidence was limited because of methodological weaknesses and variability [20]. The present study was designed with the CLI group as the target group, and the results were similar to those of the most recent meta-analysis in that the interventions in both groups resulted in the same amount of weight loss [17]. Regarding the persistence of intervention effects on weight loss, weight regain after an active intervention is common, with around half of the lost weight being regained within 2 years [39]. Carriere et al. conducted a systematic review and meta-analysis that revealed a difference in weight at follow-up with and without the intervention [14]. However, there is variability in the timing of weight-related assessments among studies [40], and it is unclear when weight regain occurs. A small-scale RCT examining a mindfulness intervention reported weight gain in the CLI group after 12 weeks of follow-up [21]. However, in the present study, both the CLI + MMA and CLI groups exhibited no weight gain. This discrepancy could be attributed to differences in study methodology, as well as differences in the average age and ethnicity of the sample. In this study, we continued regular telephone-based counseling after the active intervention. Continued biweekly or monthly behavioral counseling after initial weight loss is known to be an effective approach for preventing weight regain [41]. Therefore, the follow-up period in our study might have been insufficient to assess the impact of the MMA itself on weight, meaning the results do not preclude the possibility that the MMA intervention may help weight loss. Further large-scale, long-term follow-up studies are needed to confirm our findings.
Regarding eating behavior, we found restrained eating behavior was significantly increased in the CLI + MMA group. Several meta-analyses reported that mindfulness improved binge eating behavior [9, 14–19]. Daubenmier et al. measured the effect of mindfulness using the DEBQ as an evaluation item [42], and revealed a small effect on restrained eating behavior, but large effects on emotional eating and external eating behavior [42]. In the present study, the effects on restrained and external eating behavior were large, and the effects on emotional eating behavior were small. Compared with the previous study, participants in our study reported healthier eating behaviors in their baseline assessment (our study: restrained eating 3.27, emotional eating 1.95, external eating 2.97; Daubenmier et al. [ 42]: restrained eating 2.79, emotional eating 3.42, external eating 3.57). In addition, participants in the present study were older and had a lower BMI than those in the previous study [42]. Further research is needed to elucidate this issue in more depth.
The MMA used in this study appeared to be feasible, as most participants with MetS found it acceptable and the loss to follow-up was small. However, the average adherence to the application declined over time. The overall average length of time the MMA was used each day was 453 s (range 306–889 s) (Fig. 2). Maintaining adherence is important in weight management [27], and our sensitivity analysis showed a correlation between adherence to the MMA and weight loss (Fig. 4B). A systematic review and meta-analysis of home practice in mindfulness-based cognitive therapy and mindfulness-based stress reduction showed the pooled estimate for participants’ home practice was $64\%$ of the assigned amount, equating to about 30 min/day, 6 days/week [43]. Similarly, a systematic review of cancer survivors’ adherence to home mindfulness practice found the pooled adherence rate for participants’ home practice was $60\%$ of the assigned amount (27 min/day during the intervention period), although survivors tended to practice less as time passed [44]. A possible reason for this difference in adherence is that adherence to an application partly depends on users’ characteristics [45]. Participants in our study may not have been aware of the importance of mindfulness. In addition, the healthcare workers who checked MMA adherence were not familiar with mindfulness, which might have made it difficult to follow the transtheoretical model of health behavior change [33]. This study also included participants that were older than expected. Application-based studies with older participants have used devices with easy-to-use touch screens of about 7 in. [ 46] or incorporating multiple alarm functions per day [46, 47]. Creative strategies to increase adherence to such applications should be considered in further studies to comprehensively evaluate the benefit of the MMA for individuals with MetS.
The strength of the present study was our finding that increased restrained eating behavior may be associated with weight loss during CLI. Restraint theory suggests that cognitive control of human eating behavior leads to reduced sensitivity to internal cues for satiety, which can result in disinhibited eating (i.e., overeating) in situations where cognitive control is undermined [48]. However, there is little experimental evidence from non-clinical samples that increased eating restraint is related to disinhibited eating or an increased cognitive bias for food [49]. When those who reported trying to lose weight via dieting were compared with those who were not dieting, the former group showed a reduced frequency of binge eating [50]. Another study found current ongoing dieters did not show activation in the prefrontal cortex and orbitofrontal cortex regions related to cognitive control [51]. It is possible that participants in the present study had lost weight because they were ongoing dieters during CLI. In this study, the correlation between weight and restrained eating was higher in the CLI + MMA group compared with the CLI group (Fig. 4A). Mindfulness decreases default mode network and frontoparietal network connectivity, and changes in medial prefrontal-amygdala connectivity are critically implicated in the regulation of emotion, which may be related to improved emotional regulation [52, 53]. Although there is a possibility of confounding bias from other factors that contribute to weight loss, our findings regarding the MMA warrant further research.
The present study also had several limitations that should be considered. As expected in a pilot feasibility study, the sample size was too small to draw definitive conclusions. Furthermore, $92\%$ of the sample were women, which limited the generalizability of the findings to men. A third limitation was the choice of a control intervention. We compared a CLI group as the control group because of the feasibility of the study. Although this was a proprietary program in the public interest and not a scientifically proven effective program for weight loss, similar interventions throughout Japan have shown some effectiveness [54]. Fourth, the randomization was not successful as the control group was significantly heavier than the intervention group at baseline. Fifth, this study was conducted during the COVID-19 pandemic [32]. In addition to the fact that the planned interventions could not be implemented, the restrictions on daily life caused by COVID-19 measures also affected eating and physical activity [55], which could have influenced the outcomes in our study.
The present study showed that the MMA was highly feasible and moderately acceptable for use by people with MetS. The average duration of MMA use in the present study was 453 s (range 306–889 s) per day. This was a short duration compared with traditional mindfulness interventions, which typically involve face-to-face group exercises lasting 2–2.5 h per session [56]. We found the effects of the MMA included improved eating behavior, which suggested that this low-intensity application was effective for modifying some behaviors and attitudes related to problem eating behavior. Further large-scale studies are needed to confirm our results. We also found an association between adherence to the MMA and weight loss. This could be attributable to the characteristics of the application or the relationship between participants and the healthcare workers delivering the program. However, it was difficult to fully examine this important factor using the data collected in this study. Further studies focused on this aspect may inform development of more effective interventions.
## Conclusion
This study is the first pilot RCT to examine the efficacy of MMA combined with diet and exercise therapy for MetS. This pilot study found that MMA with CLI was highly feasible and moderately acceptable, but low adherence to the application suggests that the intervention should include additional steps to improve engagement with the application. We found no difference in the effects on weight loss between the study groups, although use of the MMA might have contributed to an increase in restrained eating. Further large-scale studies building on this study are needed to demonstrate the effectiveness of MMA interventions.
## References
1. 1.Collaboration NCDRF. Trends in adult body-mass index in 200 countries from 1975 to 2014: a pooled analysis of 1698 population-based measurement studies with 19.2 million participants. Lancet. 2016;387(10026):1377–96. 10.1016/S0140-6736(16)30054-X.
2. Funatogawa I, Funatogawa T, Nakao M, Karita K, Yano E. **Changes in body mass index by birth cohort in Japanese adults: results from the National Nutrition Survey of Japan 1956–2005**. *Int J Epidemiol* (2009.0) **38** 83-92. DOI: 10.1093/ije/dyn182
3. Muller MJ, Lagerpusch M, Enderle J, Schautz B, Heller M, Bosy-Westphal A. **Beyond the body mass index: tracking body composition in the pathogenesis of obesity and the metabolic syndrome**. *Obes Rev* (2012.0) **13** 6-13. DOI: 10.1111/j.1467-789X.2012.01033.x
4. Eberly LE, Prineas R, Cohen JD. **Metabolic syndrome: risk factor distribution and 18-year mortality in the multiple risk factor intervention trial**. *Diabetes Care* (2006.0) **29** 123-130. DOI: 10.2337/diacare.29.1.123
5. 5.Bassi N, Karagodin I, Wang S, et al. Lifestyle modification for metabolic syndrome: a systematic review. Am J Med. 2014;127(12):1242 e1–10. 10.1016/j.amjmed.2014.06.035.
6. Butryn ML, Webb V, Wadden TA. **Behavioral treatment of obesity**. *Psychiatr Clin North Am* (2011.0) **34** 841-859. DOI: 10.1016/j.psc.2011.08.006
7. Look ARG. **Eight-year weight losses with an intensive lifestyle intervention: the look AHEAD study**. *Obesity (Silver Spring)* (2014.0) **22** 5-13. DOI: 10.1002/oby.20662
8. 8.Bray GA, Kim KK, Wilding JPH, World Obesity F. Obesity: a chronic relapsing progressive disease process. A position statement of the World Obesity Federation. Obes Rev. 2017;18(7):715–23. 10.1111/obr.12551.
9. O'Reilly GA, Cook L, Spruijt-Metz D, Black DS. **Mindfulness-based interventions for obesity-related eating behaviours: a literature review**. *Obes Rev* (2014.0) **15** 453-461. DOI: 10.1111/obr.12156
10. 10.Pellegrini M, Carletto S, Scumaci E, Ponzo V, Ostacoli L, Bo S. The use of self-help strategies in obesity treatment. a narrative review focused on hypnosis and mindfulness. Curr Obes Rep. 2021;10(3):351–64. 10.1007/s13679-021-00443-z.
11. 11.Shaw K, O'Rourke P, Del Mar C, Kenardy J. Psychological interventions for overweight or obesity. Cochrane Database Syst Rev. 2005(2):CD003818. 10.1002/14651858.CD003818.pub2.
12. Creswell JD. **Mindfulness interventions**. *Annu Rev Psychol* (2017.0) **68** 491-516. DOI: 10.1146/annurev-psych-042716-051139
13. Tapper K. **Can mindfulness influence weight management related eating behaviors? If so, how?**. *Clin Psychol Rev* (2017.0) **53** 122-134. DOI: 10.1016/j.cpr.2017.03.003
14. Carriere K, Khoury B, Gunak MM, Knauper B. **Mindfulness-based interventions for weight loss: a systematic review and meta-analysis**. *Obes Rev* (2018.0) **19** 164-177. DOI: 10.1111/obr.12623
15. Godfrey KM, Gallo LC, Afari N. **Mindfulness-based interventions for binge eating: a systematic review and meta-analysis**. *J Behav Med* (2015.0) **38** 348-362. DOI: 10.1007/s10865-014-9610-5
16. Katterman SN, Kleinman BM, Hood MM, Nackers LM, Corsica JA. **Mindfulness meditation as an intervention for binge eating, emotional eating, and weight loss: a systematic review**. *Eat Behav* (2014.0) **15** 197-204. DOI: 10.1016/j.eatbeh.2014.01.005
17. 17.Daniela M, Robinson L, Gordon G, Werthmann J, Campbell IC, Schmidt U. The outcomes of mindfulness-based interventions for obesity and binge eating disorder: a meta-analysis of randomised controlled trials. Appetite. 2021;166:105464. 10.1016/j.appet.2021.105464.
18. Ruffault A, Czernichow S, Hagger MS. **The effects of mindfulness training on weight-loss and health-related behaviours in adults with overweight and obesity: a systematic review and meta-analysis**. *Obes Res Clin Pract* (2017.0) **11** 90-111. DOI: 10.1016/j.orcp.2016.09.002
19. Rogers JM, Ferrari M, Mosely K, Lang CP, Brennan L. **Mindfulness-based interventions for adults who are overweight or obese: a meta-analysis of physical and psychological health outcomes**. *Obes Rev* (2017.0) **18** 51-67. DOI: 10.1111/obr.12461
20. Olson KL, Emery CF. **Mindfulness and weight loss: a systematic review**. *Psychosom Med* (2015.0) **77** 59-67. DOI: 10.1097/PSY.0000000000000127
21. 21.Spadaro KC, Davis KK, Sereika SM, Gibbs BB, Jakicic JM, Cohen SM. Effect of mindfulness meditation on short-term weight loss and eating behaviors in overweight and obese adults: a randomized controlled trial. J Complement Integr Med. 2017;15(2). 10.1515/jcim-2016-0048.
22. Framson C, Kristal AR, Schenk JM, Littman AJ, Zeliadt S, Benitez D. **Development and validation of the mindful eating questionnaire**. *J Am Diet Assoc* (2009.0) **109** 1439-1444. DOI: 10.1016/j.jada.2009.05.006
23. Tapper K. **Mindful eating: what we know so far**. *Nutr Bull* (2022.0) **47** 168-185. DOI: 10.1111/nbu.12559
24. Dunn C, Haubenreiser M, Johnson M. **Mindfulness approaches and weight loss, weight maintenance, and weight regain**. *Curr Obes Rep* (2018.0) **7** 37-49. DOI: 10.1007/s13679-018-0299-6
25. Carpenter KM, Vickerman KA, Salmon EE, Javitz HS, Epel ES, Lovejoy JC. **A randomized pilot study of a phone-based mindfulness and weight loss program**. *Behav Med* (2019.0) **45** 271-281. DOI: 10.1080/08964289.2017.1384359
26. 26.Lyzwinski LN, Caffery L, Bambling M, Edirippulige S. The mindfulness app trial for weight, weight-related behaviors, and stress in university students: randomized controlled trial. JMIR Mhealth Uhealth. 2019;7(4):e12210. 10.2196/12210.
27. 27.Ghelani DP, Moran LJ, Johnson C, Mousa A, Naderpoor N. Mobile apps for weight management: a review of the latest evidence to inform practice. Front Endocrinol. (Lausanne). 2020;11:412. 10.3389/fendo.2020.00412.
28. 28.Flores Mateo G, Granado-Font E, Ferre-Grau C, Montana-Carreras X. Mobile phone apps to promote weight loss and increase physical activity: a systematic review and meta-analysis. J Med Internet Res. 2015;17(11):e253. 10.2196/jmir.4836.
29. Wong EML, Leung DYP, Tam HL, Wang Q, Yeung KW, Leung AYM. **The effect of a lifestyle intervention program using a mobile application for adults with metabolic syndrome, versus the effect of a program using a booklet: a pilot randomized controlled trial**. *Clin Interv Aging* (2021.0) **16** 633-644. DOI: 10.2147/CIA.S303920
30. Oda E. **Metabolic syndrome: its history, mechanisms, and limitations**. *Acta Diabetol* (2012.0) **49** 89-95. DOI: 10.1007/s00592-011-0309-6
31. Yamauchi K. **Evidence-based lifestyle improvement, efficacy and practice of the Japanese portion control method (Healthy Plate) with behavioral therapy**. *J Japan Ass Diabet Informatics* (2016.0) **14** 56-62
32. 32.Office for Novel Coronavirus Disease Control, Cabinet Secretariat, Government of Japan. Extension of the period for declaring a state of emergency for novel coronavirus infection. https://corona.go.jp/news/news_20200421_70.html. Accessed March 22, 2022. (in Japanese)
33. Prochaska JO, Velicer WF. **The transtheoretical model of health behavior change**. *Am J Health Promot* (1997.0) **12** 38-48. DOI: 10.4278/0890-1171-12.1.38
34. Van Strien T, Frijters JE, Bergers GP, Defares PB. **The Dutch Eating Behavior Questionnaire (DEBQ) for assessment of restrained, emotional, and external eating behavior**. *Int J Eat Disord* (1986.0) **5** 295-315. DOI: 10.1002/1098-108X(198602)5:2<295::AID-EAT2260050209>3.0.CO;2-T
35. Kato Y, Roth R. **The difference of eating behavior between Austrian and Japanese university students**. *J Yasuda Women’s University* (2009.0) **37** 209-220
36. Murase N, Katsumura T, Ueda C, Inoue S, Shimomitsu T. **Validity and reliability of Japanese version of International Physical Activity Questionnaire**. *Journal of Health Welfare Statistics* (2002.0) **49** 1-9
37. Kato Y, Nishida M, Tanaka Y, Kawabata T. **The relationship between motivation for healthy eating and attitudes toward childhood eating habits among university students**. *Japanese Journal of School Health* (2013.0) **54** 507-519
38. Morris SB. **Estimating effect sizes from pretest-posttest-control group designs**. *Organ Res Methods* (2008.0) **11** 364-386. DOI: 10.1177/1094428106291059
39. Stelmach-Mardas M, Mardas M, Walkowiak J, Boeing H. **Long-term weight status in regainers after weight loss by lifestyle intervention: status and challenges**. *Proc Nutr Soc* (2014.0) **73** 509-518. DOI: 10.1017/S0029665114000718
40. Mason AE, Epel ES, Aschbacher K. **Reduced reward-driven eating accounts for the impact of a mindfulness-based diet and exercise intervention on weight loss: data from the SHINE randomized controlled trial**. *Appetite* (2016.0) **100** 86-93. DOI: 10.1016/j.appet.2016.02.009
41. Alamuddin N, Bakizada Z, Wadden TA. **Management of obesity**. *J Clin Oncol* (2016.0) **34** 4295-4305. DOI: 10.1200/JCO.2016.66.8806
42. 42.Daubenmier J, Kristeller J, Hecht FM, et al. Mindfulness intervention for stress eating to reduce cortisol and abdominal fat among overweight and obese women: an exploratory randomized controlled study. J Obes. 2011;651936. 10.1155/2011/651936.
43. Parsons CE, Crane C, Parsons LJ, Fjorback LO, Kuyken W. **Home practice in mindfulness-based cognitive therapy and mindfulness-based stress reduction: a systematic review and meta-analysis of participants’ mindfulness practice and its association with outcomes**. *Behav Res Ther* (2017.0) **95** 29-41. DOI: 10.1016/j.brat.2017.05.004
44. Baydoun M, Moran C, McLennan A, Piedalue KL, Oberoi D, Carlson LE. **Mindfulness-based interventions in cancer survivors: a systematic review of participants’ adherence to home practice**. *Patient Prefer Adherence* (2021.0) **15** 1225-1242. DOI: 10.2147/PPA.S267064
45. 45.Zhao J, Freeman B, Li M. Can mobile phone apps influence people’s health behavior change? An evidence review. J Med Internet Res. 2016;18(11):e287. 10.2196/jmir.5692.
46. 46.Mira JJ, Navarro I, Botella F, et al. A Spanish pillbox app for elderly patients taking multiple medications: randomized controlled trial. J Med Internet Res. 2014;16(4):e99. 10.2196/jmir.3269.
47. 47.van Het Reve E, Silveira P, Daniel F, Casati F, de Bruin ED. Tablet-based strength-balance training to motivate and improve adherence to exercise in independently living older people: part 2 of a phase II preclinical exploratory trial. J Med Internet Res. 2014;16(6):e159. 10.2196/jmir.3055.
48. Herman CP, Mack D. **Restrained and unrestrained eating**. *J Pers* (1975.0) **43** 647-660. DOI: 10.1111/j.1467-6494.1975.tb00727.x
49. 49.Watson P, Le Pelley ME. A meta-analysis of the relationship between eating restraint, impaired cognitive control and cognitive bias to food in non-clinical samples. Clin Psychol Rev. 2021;89:102082. 10.1016/j.cpr.2021.102082.
50. 50.Lowe MR. Commentary on: What is restrained eating and how do we identify it?: unveiling the elephant in the room. Appetite. 2022;168:105221. 10.1016/j.appet.2021.105221.
51. Ely AV, Childress AR, Jagannathan K, Lowe MR. **Differential reward response to palatable food cues in past and current dieters: a fMRI study**. *Obesity (Silver Spring)* (2014.0) **22** E38-45. DOI: 10.1002/oby.20599
52. 52.Sezer I, Pizzagalli DA, Sacchet MD. Resting-state fMRI functional connectivity and mindfulness in clinical and non-clinical contexts: a review and synthesis. Neurosci Biobehav Rev. 2022;135:104583. 10.1016/j.neubiorev.2022.104583.
53. 53.Chumachenko SY, Cali RJ, Rosal MC, et al. Keeping weight off: mindfulness-based stress reduction alters amygdala functional connectivity during weight loss maintenance in a randomized control trial. PLoS One. 2021;16(1):e0244847. 10.1371/journal.pone.0244847.
54. 54.Nakao YM, Miyamoto Y, Ueshima K, et al. Effectiveness of nationwide screening and lifestyle intervention for abdominal obesity and cardiometabolic risks in Japan: the metabolic syndrome and comprehensive lifestyle intervention study on nationwide database in Japan (MetS ACTION-J study). PLoS One. 2018;13(1):e0190862. 10.1371/journal.pone.0190862.
55. Deschasaux-Tanguy M, Druesne-Pecollo N, Esseddik Y. **Diet and physical activity during the coronavirus disease 2019 (COVID-19) lockdown (March-May 2020): results from the French NutriNet-Sante cohort study**. *Am J Clin Nutr* (2021.0) **113** 924-938. DOI: 10.1093/ajcn/nqaa336
56. Mason AE, Epel ES, Kristeller J. **Effects of a mindfulness-based intervention on mindful eating, sweets consumption, and fasting glucose levels in obese adults: data from the SHINE randomized controlled trial**. *J Behav Med* (2016.0) **39** 201-213. DOI: 10.1007/s10865-015-9692-8
|
---
title: Positive end-expiratory pressure induced changes in airway driving pressure
in mechanically ventilated COVID-19 Acute Respiratory Distress Syndrome patients
authors:
- Mônica Rodrigues da Cruz
- Luciana Moisés Camilo
- Tiago Batista da Costa Xavier
- Gabriel Casulari da Motta Ribeiro
- Denise Machado Medeiros
- Luís Felipe da Fonseca Reis
- Bruno Leonardo da Silva Guimarães
- André Miguel Japiassú
- Alysson Roncally Silva Carvalho
journal: Critical Care
year: 2023
pmcid: PMC10029797
doi: 10.1186/s13054-023-04345-5
license: CC BY 4.0
---
# Positive end-expiratory pressure induced changes in airway driving pressure in mechanically ventilated COVID-19 Acute Respiratory Distress Syndrome patients
## Abstract
### Background
The profile of changes in airway driving pressure (dPaw) induced by positive-end expiratory pressure (PEEP) might aid for individualized protective ventilation. Our aim was to describe the dPaw versus PEEP curves behavior in ARDS from COVID-19 patients.
### Methods
Patients admitted in three hospitals were ventilated with fraction of inspired oxygen (FiO2) and PEEP initially adjusted by oxygenation-based table. Thereafter, PEEP was reduced from 20 until 6 cmH2O while dPaw was stepwise recorded and the lowest PEEP that minimized dPaw (PEEPmin_dPaw) was assessed. Each dPaw vs PEEP curve was classified as J-shaped, inverted-J-shaped, or U-shaped according to the difference between the minimum dPaw and the dPaw at the lowest and highest PEEP. In one hospital, hyperdistention and collapse at each PEEP were assessed by electrical impedance tomography (EIT).
### Results
184 patients (41 including EIT) were studied. 126 patients ($68\%$) exhibited a J-shaped dPaw vs PEEP profile (PEEPmin_dPaw of 7.5 ± 1.9 cmH2O). 40 patients ($22\%$) presented a U (PEEPmin_dPaw of 12.2 ± 2.6 cmH2O) and 18 ($10\%$) an inverted-J profile (PEEPmin_dPaw of 14,6 ± 2.3 cmH2O). Patients with inverted-J profiles had significant higher body mass index (BMI) and lower baseline partial pressure of arterial oxygen/FiO2 ratio. PEEPmin_dPaw was associated with lower fractions of both alveolar collapse and hyperinflation.
### Conclusions
A PEEP adjustment procedure based on PEEP-induced changes in dPaw is feasible and may aid in individualized PEEP for protective ventilation. The PEEP required to minimize driving pressure was influenced by BMI and was low in the majority of patients.
## Introduction
Hypoxemic respiratory failure is the leading cause of intensive care unit (ICU) admission in COVID-19, the majority of subjects meeting Acute Respiratory Distress *Syndrome criteria* (C-ARDS) [1]. Initially, it was observed that many patients presented a disparity between well-preserved lung mechanics and severe hypoxemia [2] and 2 different phenotypes in C-ARDS were proposed, which should be managed with different ventilatory strategies [2]. However, this was not confirmed in posterior published data, remaining recommendations to treat C-ARDS accordingly ARDS ventilation evidence-based [3]. Several hypotheses were proposed to the wide range of respiratory system compliance (Crs) observed in many C-ARDS series, including hypoxemia due to impaired perfusion in patients with higher compliance or lungs with high recruitability and lower compliance [2].
Optimal positive end-expiratory pressure (PEEP) has been pursued [4] and the question of how to recognize patients that get benefit from higher PEEP levels has led to new technologies like Electrical Impedance Tomography (EIT), a bedside tool to monitor ventilation distribution, allowing PEEP titration to reduce both collapse and hyperdistention [5].
Airway driving pressure (dPaw) is a simple parameter to monitor on the ventilator and, when diminished with increased PEEP was associated with reduced mortality risk in ARDS [6]. In C-ARDS, a lower dPaw was associated with better survival [7, 8].
In the present study, we aim to describe the profile of PEEP-induced changes in dPaw during a PEEP adjustment procedure as aid for individualized protective ventilation, including a group where it was done together with an EIT monitor.
## Patients
In this prospective observational physiologic study, adults patients admitted to the ICU of three hospitals with C-ARDS confirmed by positive nasopharyngeal polymerase chain reaction for SARS-CoV-2 and receiving invasive mechanical ventilation (MV) ≤ 48 h were analyzed. Patients with barotrauma assessed by computed tomography (CT), chronic pulmonary disease, and increased intracranial pressure were excluded.
## Mechanical ventilation settings
After analgesia and sedation adjustment, all subjects were initially ventilated in volume-controlled ventilation, tidal volume of 6 mL/kg with constant inspiratory flow, plateau pressure ≤ 30 cmH2O, FiO2 and PEEP adjusted to keep SaO2 > $90\%$ based on the ARDSNetwork table [9] and respiratory rate to maintain normal partial pressure of carbon dioxide (PaCO2). Fluids and vasopressors were provided to maintain mean arterial pressure above 60 mmHg and, neuromuscular blocking used to avoid ventilatory asynchronies.
## PEEP adjustment procedure
After initial ventilatory settings, PEEP was reduced, 2 cmH2O every thirty seconds [10], from 20 until 6 cmH2O while dPaw was assessed in each step, and the lowest PEEP that minimized dPaw (PEEPmin_dPaw) was identified. The posterior PEEP adjustment was at the discretion of the clinical team responsible for patient care.
## EIT assessment
In one of the hospitals, patients were investigated by EIT (Enlight 1800, Timpel, São Paulo, Brazil) during the PEEP adjustment procedure. Regional variations in impedance (∆Z) during ventilation, map the Vt distribution in the lung and creates a PEEP titration tool which was used to assess PEEP-induced pulmonary hyperdistention and collapse and its effects on dPaw during the PEEP adjustment procedure. The EIT optimal PEEP (PEEPEIT) was defined as the PEEP that represents the best compromise between hyperdistention and collapse estimated [5, 11].
## Evaluation of dPaw vs PEEP curve profile
After the PEEP adjustment procedure, each dPaw vs PEEP curve was recorded and retrospectively classified into one of three categories according to the difference between the minimum dPaw [12] and the dPaw at the lowest (ΔdPlow) and highest (ΔdPhigh) PEEP [4]. If ΔdPlow < 0.2 × ΔdPhigh, the curve was classified as J-shaped; if ΔdPhigh < 0.2 × ΔdPlow, the curve was classified as inverted-J-shaped; otherwise, the curve was U-shaped.
## Statistical analysis
Results are reported without imputation as mean (standard deviation), or count (percentage), after testing for normality using the Shapiro–Wilk test. One-way ANOVA was used for the comparison between the three groups. A Bonferroni-Holm post hoc test was applied to correct multiple testing. Hyperdistention and collapse curves at different PEEP levels were assessed by computing areas under the curves (AUCs) [13] by adding the areas under each pair of consecutive observations:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{AUC}} = \frac{1}{2}\mathop \sum \limits_{1}^{8} \left({{\text{PEEP}}_{i + 1} - {\text{PEEP}}_{i} } \right) \times \left({Y_{i + 1} + Y_{i} } \right),$$\end{document}AUC=12∑18PEEPi+1-PEEPi×Yi+1+Yi,where Y was the estimated hyperdistention or collapse. The AUCs were compared only between U-shaped and J-shaped PEEP vs dPaw groups, because just one patient with Inverted-J shape had EIT measurement.
Statistical analysis was performed in R (The R Foundation, Vienna, Austria), and a $p \leq 0.05$ was considered significant.
## Results
Between Jul 27th, 2020, and Feb 24th, 2021, a total of 184 patients were included, and a PEEP adjustment procedure was performed before 48 h on invasive MV. Table 1 shows clinical characteristics in each curve profile dPaw vs PEEP. Patients with inverted J-Shaped dPaw versus PEEP profile presented significantly higher body mass index (BMI) (Table 1) and lower partial pressure of arterial oxygen and fraction of inspired oxygen ratio (PaO2/FiO2) and Crs at baseline (Table 2).Table 1Characteristics of patients with C-ARDS enrolled in the PEEP titrationPatients’ characteristicsAll COVID-19J-shapedn = 126U-shapedn = 40Inverted J-shapedn = 18p valueAge, years, $$n = 18460$.04$ ± 15.8960.37 ± 15.9459.20 ± 17.3059.67 ± 12.680.915Male, n (%), $$n = 184127$$ ($69.02\%$)92 ($72.44\%$)25 ($62.50\%$)10 ($55.55\%$)0.160Body mass index, kg/m2, $$n = 18329$.02$ ± 6.4327.48 ± 6.65a30.16 ± 6.95a,b35.89 ± 8.67b< 0.001Comorbidities, n (%), $$n = 6944$$ ($63.8\%$)30 ($65.2\%$)11 ($68.8\%$)3 ($42.9\%$)0.500Hypertension, n (%)38 ($55.1\%$)24 ($52.2\%$)11 ($68.8\%$)3 ($42.9\%$)0.422Diabetes mellitus, n (%)29 ($42\%$)18 ($39.1\%$)8 ($50\%$)3 ($42.9\%$)0.807SOFA, $$n = 1435$.43$ ± 4.035.17 ± 3.045.69 ± 3.396.53 ± 3.060.262PEEP at baseline $$n = 11410$$ (10–14)10 (10–14)a,c12 (10–15.5)a15 (10.5–19.5)c< 0.05FiO2 at baseline, n (%), $$n = 18480$$ (60–100)70 (60–100)a100 (70–100)a95 (60–100)< 0.05PaCO2 at baseline, mmhg, $$n = 7551$.6$ ± 11.951.4 ± 11.154.2 ± 13.247.1 ± 14.40.410Respiratory rate at baseline, breaths/min, $$n = 7520$$ (20–25)20 (20–25)24 (20–24.5)20 (20–20)0.170Minute ventilation at baseline, L/min, $$n = 758$.1$ ± 2.08.3 ± 2.08.1 ± 1.76.8 ± 1.70.180Continuous variables are expressed as mean and standard deviation or median and interquartile range, according to normality distribution. A one-way ANOVA or the Kruskal–Wallis test was used for the comparison between three groups with a respective post hoc analysis. The letters a, b and c express values that are statistically different. COVID-19, coronavirus disease-19; SOFA, sequential organ failure assessment score; PEEP, positive end-expiratory pressure; dPaw, airway driving pressure; PaO2/FiO2, partial pressure of arterial oxygen and fraction of inspired oxygen ratio; FiO2, fraction of inspired oxygen; PaCO2, partial pressure of carbon dioxideTable 2Respiratory mechanics and EIT dataRespiratory mechanicsJ-shapedn = 126U-shapedn = 40Inverted J-shapedn = 18p valueTidal volume, mL/kg of IBW (mean, sd)6.03 ± 0.035.86 ± 0.925.97 ± 0.140.098Baseline Crs, mL/cmH2O (mean, sd)33.47 ± 7.25a29.24 ± 8.70a,b25.64 ± 8.45b< 0.001Baseline dPaw, cmH2O (mean, sd)12.65 ± 2.66a13.21 ± 3.94a,b15.03 ± 3.72b< 0.05Baseline PaO2/FiO2, mmHg (mean, sd)139.32 ± 52.67a120.72 ± 57.68a,b92.43 ± 40.43b< 0.05PEEPmin_dPaw, cmH2O (median, IIQ)7.52 ± 1.9a,c12.2 ± 2.64a,b14.6 ± 2.38b,c< 0.001EIT assessmentN = 28N = 12N = 1Hyperdistention at the optimal PEEP, % (mean, sd)1.58 ± 2.346.34 ± 10.221.3 ± 00.071AUC for hyperdistention, %.cmH2O (mean, sd)216.75 ± 81.44a116.17 ± 77.53ª6.3 ± 0< 0.001Collapse at the optimal PEEP, % (mean, sd)13.86 ± 13.3810.81 ± 10.380.0 ± 00.473AUC for collapse, %.cmH2O (mean, sd)96.95 ± 70.40a149.42 ± 95.54a418 ± 0< 0.001EIT optimal PEEP, cmH2O, (mean, sd)9.17 ± 2.53a12.96 ± 3.29ª14.22 ± 0< 0.001Continuous variables are expressed as mean and standard deviation or median and interquartile range, according to normality distribution. A one-way ANOVA or the Kruskal–Wallis test was used for the comparison between three groups with a respective post hoc analysis. A t-test was used for the comparison between pairs. The letters a, b and c express values that are statistically different. IBW, ideal body weight; Crs, respiratory system compliance; dPaw, airway driving pressure; PaO2/FiO2, ratio of partial pressure of arterial oxygen and fraction of inspired oxygen; PEEPmin_dPaw, lowest PEEP that minimized dPaw; PEEP, positive end-expiratory pressure; EIT, electrical impedance tomography; AUC, area under the curve
## Respiratory mechanics and PEEP titration
Based on the analysis of the dPaw vs PEEP profile, most of the COVID-19 patients ($$n = 126$$) exhibited a J-shaped dPaw vs PEEP profile with dPaw starting to increase for PEEPs ≥ 7.5 ± 1.9 cmH2O, only a few COVID-19 patients had mostly inverted-J profiles ($$n = 18$$), usually requiring higher levels of PEEP (PEEPmin_dPaw ranging from 14 to 20 cmH2O) (Table 2, Fig. 1). Only $21.7\%$ of COVID-19 patients presented the U-shaped profile with the PEEPmin_dPaw ranging from 10 to 14 cmH2O.Fig. 1Respiratory system mechanics associated with the percentage of collapse and hyperdistention at different levels of PEEP. In panels A, D, and G, data were obtained by electrical impedance tomography, where ● is the respiratory system compliance; Δ is the percentage of collapse and □ is the percentage of overdistension. Panels B, C, E, F, H, and I show the percentage change in driving pressure obtained by a mechanical ventilator for a representative patient (B, E, H) and all patients (C, F, I). Panels A–C correspond to the category of patients with J-shaped curves; panels D–F correspond to the category of patients with U-shaped curves, and panels G–I correspond to the category of patients with inverted J-shaped curves The J-shaped dPaw vs PEEP profile was associated with increased hyperdistention, and collapse reduction as PEEP increased and, in this group, PEEPmin_dPaw was lower than PEEP based on the ARDSNetwork table (Table 2). At the range of the PEEPmin_dPaw both hyperdistention and collapse were minimized independent of the dPaw vs PEEP profile (Table 2, Fig. 1).
## Discussion
Our study interpreted the dPaw vs PEEP curve profile among C-ARDS patients. The main findings were: [1] $90\%$ of C-ARDS-19 patients presented a J- or U-shaped dPaw vs PEEP curve profile usually requiring PEEPs < 12 cmH2O to minimize dPaw; [2] PEEPs > 15 cmH2O would be necessary in only $10\%$ of C-ARDS, and those patients presented an inverted-J dPaw vs PEEP curve profile and higher BMI; and [3] PEEPmin_dPaw was associated with a reduction of both alveoli collapse and hyperdistention. All these patients averaged PaO2/FiO2 below 150 which there is evidence of benefit from using higher levels of PEEP in ARDS [14].
ARDS and C-ARDS are heterogeneous conditions with uncertainty about to set PEEP [2, 3] commonly based by oxygenation targets [9]. However in C-ARDS this strategy frequently resulted in worse lung mechanics [15], and cardiac output impairment [16].
Our EIT data and an experimental CT study [4] show that, at constant VT, dPaw and compliance respond to both hyperdistention and collapse. $\frac{126}{184}$ of our patients presented a J-shaped curves, with the largest hyperdistention AUC, where increasing PEEP to improve oxygenation may not work. In U-shaped curves the balanced risk of collapse and hyperdistention was obtained with about 12 cmH2O PEEP. In these two groups, higher PEEPs would carry a greater risk of iatrogenesis. Finally, patients with an inverted-J-shaped required higher PEEPs to minimize dPaw and presented higher BMI and lower initial PaO2/FiO2 ratio. In the only patient with this profile on EIT, PEEP decreased collapsed areas without increasing hyperdistention up to 20 cmH2O. The interpretation of the PEEP with respiratory system mechanics or with the amount of recruitment and overdistension on EIT seems to give the same information.
At least one-third of patients were obese in C-ARDS different cohorts [3, 7, 8], even though the effect of obesity on respiratory mechanics is well known, a relationship between BMI and compliance has not been described as an explanation, at least in part, for the COVID-19 different phenotypes. Obesity reduce Crs with the major contribution coming from the lung and not the chest wall [17] in spite of no significant association between compliance and BMI has been detected in a large cohort study of C-ARDS [18]. Mezidi et al. comparing a group of obese vs non-obese in C-ARDS patients monitoring esophageal pressure in a decremental PEEP trial demonstrated a significant difference in PEEP level for the same transpulmonary driving pressure (∆PL) and dPaw [19]. ∆PL also did not enhance significant information concerning the prediction of outcome in ARDS patients compared to dPaw itself [20].
## Limitations
The observational nature of this study is its major limitation, and although data were acquired prospectively, they were interrogated retrospectively. The heavy workload upon COVID-19 pandemic made impossible to perform a clinical trial comparing clinical outcomes considering the observed profiles. The small proportion of patients investigated with EIT did not allow an appropriate comparison between the two methods, but data suggest a similar result to obtain the best PEEP for protective ventilation with a much simpler bedside procedure.
## Conclusion
The dPaw vs PEEP curve is a feasible method and provides individualized information. A range of compliance and PEEPmin_dPaw was observed in all 3 groups and its interpretation suggested that just in a minority of C-ARDS patients, higher PEEP improves compliance, and even in these cases, it appears that obesity, together with disease severity, determines this behavior. The overall influence of personalizing PEEP on clinical outcomes remains to be determined.
## References
1. Rubenfeld GD, Thompson T, Ferguson ND, Caldwell E, Fan E, Camporota L, Slutsky AS. **Acute Respiratory Distress Syndrome: the Berlin definition**. *JAMA* (2012.0). DOI: 10.1001/jama.2012.5669
2. Gattinoni L, Chiumello D, Caironi P, Busana M, Romitti F, Brazzi L. **COVID-19 pneumonia: different respiratory treatments for different phenotypes?**. *Intensive Care Med* (2020.0) **46** 1099-1102. DOI: 10.1007/s00134-020-06033-2
3. Ferrando C, Suarez-Sipmann F, Mellado-Artigas R, Hernández M, Gea A, Arruti E. **Clinical features, ventilatory management, and outcome of ARDS caused by COVID-19 are similar to other causes of ARDS**. *Intensive Care Med* (2020.0) **46** 2200-2211. DOI: 10.1007/s00134-020-06192-2
4. Carvalho AR, Spieth PM, Pelosi P, Vidal Melo MF, Koch T, Jandre FC. **Ability of dynamic airway pressure curve profile and elastance for positive end-expiratory pressure titration**. *Intensive Care Med* (2008.0) **34** 2291-2299. DOI: 10.1007/s00134-008-1301-7
5. Costa ELV, Borges JB, Melo A, Suarez-Sipmann F, Toufen C, Bohm SH. **Bedside estimation of recruitable alveolar collapse and hyperdistension by electrical impedance tomography**. *Intensive Care Med* (2009.0) **35** 1132-1137. DOI: 10.1007/s00134-009-1447-y
6. Amato MBP, Meade MO, Slutsky AS, Brochard L, Costa ELV, Schoenfeld DA. **Driving pressure and survival in the Acute Respiratory Distress Syndrome**. *N Engl J Med* (2015.0) **372** 747-755. DOI: 10.1056/NEJMsa1410639
7. Ferreira JC, Ho YL, Besen BAMP, Malbouisson LMS, Taniguchi LU, Mendes PV. **Protective ventilation and outcomes of critically ill patients with COVID-19: a cohort study**. *Ann Intensive Care* (2021.0) **11** 92. PMID: 34097145
8. Estenssoro E, Loudet CI, Ríos FG, Kanoore Edul VS, Plotnikow G, Andrian M. **Clinical characteristics and outcomes of invasively ventilated patients with COVID-19 in Argentina (SATICOVID): a prospective, multicentre cohort study**. *Lancet Respir Med* (2021.0) **9** 989-998. DOI: 10.1016/S2213-2600(21)00229-0
9. **Ventilation with Lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome**. *N Engl J Med* (2000.0) **342** 1301-1308. DOI: 10.1056/NEJM200005043421801
10. 10.Bergamini BC, Carvalho NS, Medeiros DM, Bozza FA, Giannella-Neto A, Carvalho AR. Time-Dependence influence on mechanical properties of the respiratory system in two strategies for decremental positive end-expiratory pressure titration. Em: A48 mechanical ventilation [Internet]. American Thoracic Society; 2012 [citado 17 de janeiro de 2023]. p. A1685–A1685. Disponível em. 10.1164/ajrccm-conference.2012.185.1_MeetingAbstracts.A1685
11. Perier F, Tuffet S, Maraffi T, Alcala G, Victor M, Haudebourg AF. **Electrical impedance tomography to titrate positive end-expiratory pressure in COVID-19 Acute Respiratory Distress Syndrome**. *Crit Care Lond Engl* (2020.0) **24** 678. DOI: 10.1186/s13054-020-03414-3
12. 12.Bergamini BC, Carvalho NS, Medeiros DM, Bozza FA, Giannella-Neto A, Carvalho AR. Time-dependence influence on mechanical properties of the respiratory system in two strategies for decremental positive end-expiratory pressure titration. Em: A48 mechanical ventilation [Internet]. American Thoracic Society; 2012 [citado 13 de janeiro de 2023]. p. A1685–A1685 (American Thoracic Society International Conference Abstracts). Disponível em. 10.1164/ajrccm-conference.2012.185.1_MeetingAbstracts.A1685
13. Matthews JN, Altman DG, Campbell MJ, Royston P. **Analysis of serial measurements in medical research**. *BMJ* (1990.0) **300** 230-235. DOI: 10.1136/bmj.300.6719.230
14. Briel M, Meade M, Mercat A, Brower RG, Talmor D, Walter SD. **Higher vs lower positive end-expiratory pressure in patients with acute lung injury and Acute Respiratory Distress Syndrome: systematic review and meta-analysis**. *JAMA* (2010.0) **303** 865-873. DOI: 10.1001/jama.2010.218
15. Protti A, Santini A, Pennati F, Chiurazzi C, Cressoni M, Ferrari M. **Lung response to a higher positive end-expiratory pressure in mechanically ventilated patients with COVID-19**. *Chest* (2022.0) **161** 979-988. DOI: 10.1016/j.chest.2021.10.012
16. Dell’Anna AM, Carelli S, Cicetti M, Stella C, Bongiovanni F, Natalini D. **Hemodynamic response to positive end-expiratory pressure and prone position in COVID-19 ARDS**. *Respir Physiol Neurobiol* (2022.0) **298** 103844. DOI: 10.1016/j.resp.2022.103844
17. Fumagalli J, Berra L, Zhang C, Pirrone M, Santiago RRDS, Gomes S. **Transpulmonary pressure describes lung morphology during decremental positive end-expiratory pressure trials in obesity**. *Crit Care Med* (2017.0) **45** 1374-1381. DOI: 10.1097/CCM.0000000000002460
18. Li Bassi G, Suen JY, Dalton HJ, White N, Shrapnel S, Fanning JP. **An appraisal of respiratory system compliance in mechanically ventilated covid-19 patients**. *Crit Care* (2021.0) **25** 199. DOI: 10.1186/s13054-021-03518-4
19. Mezidi M, Daviet F, Chabert P, Hraiech S, Bitker L, Forel JM. **Transpulmonary pressures in obese and non-obese COVID-19 ARDS**. *Ann Intensive Care* (2020.0) **10** 129. DOI: 10.1186/s13613-020-00745-w
20. Chen L, Grieco DL, Beloncle F, Chen GQ, Tiribelli N, Madotto F. **Partition of respiratory mechanics in patients with Acute Respiratory Distress Syndrome and association with outcome: a multicentre clinical study**. *Intensive Care Med* (2022.0) **48** 888-898. DOI: 10.1007/s00134-022-06724-y
|
---
title: 'Evaluation of Metabolic Parameters on Use of Newer Antiepileptics Versus Conventional
Antiepileptics in Patients of Generalised Tonic-Clonic Seizure: An Observational
Study'
journal: Cureus
year: 2023
pmcid: PMC10029831
doi: 10.7759/cureus.35181
license: CC BY 3.0
---
# Evaluation of Metabolic Parameters on Use of Newer Antiepileptics Versus Conventional Antiepileptics in Patients of Generalised Tonic-Clonic Seizure: An Observational Study
## Abstract
Background and objective *Epilepsy is* the commonest serious neurological condition and around 50 million people live with epilepsy (PWE). Primary and secondary generalised tonic-clonic seizures (GTCS) together constitute up to $50\%$ of adult and adolescent epilepsy. GTCS respond well to broad-spectrum AEDs like valproate, phenytoin, levetiracetam, lamotrigine, and topiramate. Carbamazepine and oxcarbazepine are considered alternatives. Metabolic derangements with the conventional AEDs (phenytoin causes loss of bone mass in women, phenytoin and carbamazepine produce increases in serum lipid and C-reactive protein, weight gain with valproate) are well documented. But, there is limited data regarding the effect of the newer AEDs on metabolic parameters. Thus, this study was undertaken to assess the effects of the newer AEDs on the metabolic profile of patients with epilepsy.
Material and methods A prospective observational study was conducted in the Department of Pharmacology, in collaboration with the Department of Neurology at S.C.B. Medical College and Hospital, Cuttack. 100 diagnosed patients with GTCS receiving monotherapy of either conventional or newer anti-epileptics were included in the study. Their metabolic parameters like total cholesterol, serum sodium, serum TSH and fasting blood glucose were collected at baseline, three months, and six months. ADRs were collected during the entire study period and causality assessment was done using WHO-UMC Causality Assessment Scale. All the data were analysed using SPSS 20.0 after applying appropriate statistical tests.
Results There was a significant increase in total cholesterol in all four groups ($$p \leq 0.002$$) but a pathological increase in the phenytoin and oxcarbazepine groups. There was a significant rise in the serum TSH levels in all groups except levetiracetam, but a pathological increase was seen with phenytoin and valproate, i.e., the conventional ones. Statistically significant hyponatremia was seen with valproate and oxcarbazepine. A rise in the FBS was seen with both phenytoin and valproate ($$p \leq 0.002$$) but a pathological rise was seen with phenytoin. Out of the total reported ADRs, $53.5\%$ were seen with conventional AEDs, and the rest $46.5\%$ were seen with newer ones.
Conclusion The advent of newer anti-epileptic drugs has unfolded wider horizons to the treatment of epilepsy. Each of these drugs has a unique mechanism of action, making it less prone to resistance. Metabolic derangements are a key determinant in the compliance of these drugs as they can predispose to other co-morbidities. Periodic monitoring of the various metabolic parameters is useful and together with patient counselling can improve the effectiveness of the anti-epileptic drugs.
## Introduction
Epilepsy is one of the most common and serious neurological conditions, affecting $0.5\%$-$1\%$ of the population and accounting for a notable proportion of the world’s disease burden. Today, there are an estimated 50 million people with epilepsy (PWE) [1], $80\%$ of whom live in developing countries. India contributes to one-sixth of the global burden of epilepsy [2]. Primary and secondary generalised tonic-clonic seizures (GTCS) constitute up to $50\%$ of adult and adolescent epilepsy [3]. At any given point in time, 4-10 people per 1,000 in a general population suffer active epilepsy, i.e., continuing seizures or needing treatment [1]. About $70\%$ of PWE become seizure-free with the appropriate use of medications [4]. Antiepileptic pharmacotherapy for adults can be put to an end after a seizure-free period of 2-5 years [5,6]. Antiepileptic drugs (AEDs) are a major treatment consideration for PWE, and therefore a primary concern in choosing the appropriate drug. GTCSs respond well to AEDs like valproate, phenytoin, levetiracetam, lamotrigine, and topiramate [3]. Carbamazepine and oxcarbazepine are considered alternatives.
Various treatment options have been prescribed for GTCSs, and adverse reactions related to AEDs are a major limiting factor. With the discovery of new AEDs, there has been a significant advancement in the treatment of epilepsy over the past decade. Newer AEDs offer the potential advantages of fewer drug interactions, unique mechanisms of action, and a broader spectrum of activity. They have also been used as an adjunctive treatment for refractory epilepsy [7]. Most new AEDs involve less teratogenicity, and they have a milder effect on the patient’s hormone secretion and bone and lipid metabolism [8].
Metabolic derangements with conventional AEDs are well documented. Phenytoin causes loss of bone mass in women; phenytoin and carbamazepine produce increases in serum lipid and C-reactive protein; and valproate causes weight gain [9]. But there is limited data regarding the effect of the newer AEDs on metabolic parameters. To address this research gap, this study was undertaken to assess the effects of the newer AEDs on the metabolic profile of patients with epilepsy.
## Materials and methods
This study was a prospective observational study conducted with the primary objective of evaluating the effect of the newer and conventional antiepileptics on metabolic parameters. The study was conducted over a period of two years from 2015 to 2017 in the Department of Pharmacology, in collaboration with the Department of Neurology, at S.C.B. Medical College and Hospital, Cuttack. Ethical clearance was provided by the Institutional Ethics Committee (IEC/IRB No.474) prior to the start of the study. Written, informed consent was established, and confidentiality was maintained throughout the study.
The study population consisted of diagnosed patients with GTCS attending the neurology Out-patient Department. Those receiving monotherapy of either conventional or newer antiepileptics were included in the study. The patients who denied consent for follow-up and those with psychiatric illnesses were excluded from the study. A total of 100 patients were observed during the study period. Their metabolic parameters, like total cholesterol, serum sodium, serum TSH, and fasting blood glucose were collected at baseline, three months, and six months. Each patient was followed up for a period of six months. Also, the patients were advised to report to the doctor in case of any adverse drug reaction or intolerance during the course of the treatment. The causality assessment was done using the WHO-UMC Causality Assessment Scale [10].
Drugs and doses GTCS patients receiving Phenytoin at a dose of 100 mg thrice daily were under group-A, patients receiving Valproate at a dose of 20-40 mg/kg of body weight were under group B, patients receiving Levetiracetam at a dose of 1,500-2,000 mg in divided doses were under group-C, and patients receiving Oxcarbazepine at a dose of 150-300 mg twice daily were under group-D. Twenty-five patients in each group, i.e., a total of 100 patients with GTCS, were observed during the study period. For evaluation of metabolic ADRs, reports of the total cholesterol, thyroid stimulating hormone (TSH), serum sodium, and fasting blood sugar (FBS) levels were collected from each group and studied at baseline, three-month, and six-month intervals.
The patient details were collected on a preformed case record form and later entered into a Microsoft Excel spreadsheet. All the data were analysed using SPSS 20.0. Numerical data were expressed in terms of mean and standard deviations. Metabolic parameters were described using the median and interquartile range, and comparison was done using the Wilcoxon Signed Rank test. Any difference was considered statistically significant at $p \leq 0.05.$
## Results
A total of 100 patients were evaluated. The mean age was 31 ± 14.1 years. The majority were males ($60\%$) and rest were females ($40\%$). Various adverse drug reactions (ADRs) were seen both with the conventional AEDs and newer AEDs (Table 1). The majority of ADRs ($53.5\%$) were seen with conventional AEDs, and the rest ($46.5\%$) were seen with the newer ones. CNS side-effects like impaired concentration, drowsiness, anxiety, depression, and sedation were the most common ($38\%$), followed by metabolic side-effects.
**Table 1**
| Age in years (mean±SD) | 31 ± 14.1 |
| --- | --- |
| Gender distribution [n=100] | Gender distribution [n=100] |
| Males (%) | 60 (60%) |
| Females (%) | 40(40%) |
| Conventional AEDs | Conventional AEDs |
| total number used (n=50) ADR caused (n=98) | 55 (56.1%) |
| Newer AEDs | Newer AEDs |
| total number used (n=50) ADR caused (n=98) | 43 (43.8%) |
| Types of ADRs (%) | Types of ADRs (%) |
| CNS | 38% |
| Metabolic | 25% |
| Dermatologic | 20% |
| Gastrointestinal | 22.5% |
| Genitourinary | 5% |
| Musculocutaneous | 5% |
| Constitutional | 1.5% |
| Ophthalmic | 1% |
| Causality assessment (n=100) | Causality assessment (n=100) |
| Possible | 60 (60%) |
| Probable/Likely | 11.5 (11.5%) |
| Unlikely | 28.5 (28.5%) |
For the evaluation of metabolic ADRs, reports of the total cholesterol, thyroid stimulating hormone (TSH), serum sodium, and fasting blood sugar (FBS) levels were studied, and conventional AEDs (phenytoin, valproate) were compared to newer AEDs (levetiracetam, oxcarbazepine) at baseline, three months, and six months (Table 2).
**Table 2**
| Metabolic parameters | Time intervals | Phenytoin (n=25) | P-value* | Valproate (n=25) | P-value*.1 | Levetiracetam (n=25) | P-value*.2 | Oxcarbazepine (n=25) | P-value*.3 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Total cholesterol (mg/dl) median (IQR) | Baseline | 160 (148.5–166.3) | | 160 (148–164.5) | | 151 (138.8–165) | | 155.5 (150–161.3) | |
| Total cholesterol (mg/dl) median (IQR) | 3 months | 174 (159–189) | 0.002 | 168 (156.5–170.5) | 0.002 | 159 (149– 176) | 0.002 | 159 (150.5–167.5) | 0.002 |
| Total cholesterol (mg/dl) median (IQR) | 6 months | 213 (207–220) | 0.002 | 172 (166.8–180) | 0.002 | 162.5 (158–181.3) | 0.002 | 210 (203.8– 211.3) | 0.002 |
| TSH(mIU/L) median (IQR) | Baseline | 2.5 (2–3) | | 2 (1–3) | | 2 (1–3) | | 2 (2–3) | |
| TSH(mIU/L) median (IQR) | 3 months | 3 (2–4) | 0.125 | 4 (2–4) | 0.004 | 2 (1–3) | 1.0 | 4 (2–5) | 0.02 |
| TSH(mIU/L) median (IQR) | 6 months | 7 (3–17.8) | 0.02 | 6.5 (3–10.3) | 0.002 | 2.5 (2–3) | 0.06 | 5 (4–8.8) | 0.002 |
| Sr sodium (mEq/L) median (IQR) | Baseline | 138.5 (137–140) | | 139.5 (138.5–142) | | 138 (136.8–139) | | 139 (135.8–141.3) | |
| Sr sodium (mEq/L) median (IQR) | 3 months | 138 (136.8–139.5) | 0.6 | 138 (137.3–131) | 0.5 | 139 (137.8–140) | 0.004 | 133 (130–137.5) | 0.002 |
| Sr sodium (mEq/L) median (IQR) | 6 months | 137 (135.8–140.5) | 0.2 | 130 (128–140.5) | 0.002 | 140 (139– 141) | 0.004 | 129.5 (125.8–136.3) | 0.002 |
| FBS (mg/dL) median (IQR)B | Baseline | 94 (87.5–99.3) | | 89.5 (78.8–95.8) | | 94 (80.5–100.5) | | 90 (84.5– 97.8) | |
| FBS (mg/dL) median (IQR)B | 3 months | 104 (93.8–109.3) | 0.002 | 94 (88.8–101.3) | 0.002 | 94.5 (80.5–109.3) | 0.03 | 89 (87.3–102.3) | 0.07 |
| FBS (mg/dL) median (IQR)B | 6 months | 113.5 (102–120.8) | 0.002 | 102 (95.8–108) | 0.002 | 93.5 (80.5–116.3) | 0.06 | 92.5 (87.8–104.3) | 0.05 |
Conventional drugs like phenytoin and valproate, along with newer AEDs like levetiracetam and oxcarbazepine, were considered for the evaluation of metabolic ADRs. A total of 100 patients were evaluated for any change in their in total cholesterol, TSH, serum sodium, and FBS. Twenty-five patients were given each of the drugs mentioned above. There was a significant increase in total cholesterol in all the four groups ($$p \leq 0.002$$) and a pathological increase in the phenytoin and oxcarbazepine group. There was a significant rise in the serum TSH levels in all groups except levetiracetam, but pathological increase was seen with phenytoin and valproate, i.e., the conventional ones. Statistically significant hyponatremia was seen with valproate and oxcarbazepine. Rise in the FBS was seen with both phenytoin and valproate ($$p \leq 0.002$$), but pathological rise was seen with phenytoin.
## Discussion
This study was designed to demonstrate the effects of phenytoin, valproate, levetiracetam, and oxcarbazepine on metabolic parameters such as total cholesterol, serum TSH, serum sodium, and FBS after three months and six months of treatment. The study showed an increase in total cholesterol with both the conventional and the newer antiepileptics, but a pathological rise was seen with phenytoin and oxcarbazepine. A total cholesterol level of <200mg/dl is considered to be desirable [11]. Previous studies have documented a rise in total cholesterol levels with phenytoin [12,13]. Mechanisms are poorly understood but may be caused by the activation of the pregnane X receptor (PXR), which is a major regulator of drug metabolism and drug-drug interactions contributing to hypercholesterolemia [14]. Similarly, there was a rise in serum total cholesterol, triglycerides, LDL-C/HDL-C apolipoprotein AI (ApoAI), and apolipoprotein B (ApoB) in children treated with sodium valproate [15]. On the contrary, a study by Kantoush et al. had varying results. In that study, children with epilepsy receiving valproate had lower total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), much lower-density lipoprotein cholesterol (VLDL-C) LDL-C/HDL-C ratio and a higher high-density lipoprotein Cholesterol (HDL-C) ratio than controls [16]. The mechanism behind the lipid profile changes due to valproic acid is still unclear. A possible mechanism may be insulin resistance and hyperinsulinemia, resulting in impaired lipid transport and lipogenesis [17]. Another possible mechanism may be the induction of cytochrome p (CYP) enzymes that play a role in metabolism and the biosynthesis of cholesterol. Studies have shown that there has been a significant rise in total cholesterol with oxcarbazepine, but levetiracetam showed no significant changes [18].
There was a significant increase in serum TSH levels in patients on valproate and phenytoin as compared to levetiracetam and oxcarbazepine. Similar results were seen in a study by Dahiya et al., where patients receiving phenytoin monotherapy developed hypothyroidism with a rise in TSH levels [19]. A study by Pattan et al. showed that with long-term phenytoin administration, there was a lowering of T4 level in the setting of normal TSH level [20]. Other studies showed that there was the development of subclinical hypothyroidism in children receiving AEDs, particularly valproate [21,22]. Thyroid hormones play an important role in the physiological process and a disturbance in them may lead to metabolic syndrome with multiple systems involved. Hence, it is important to keep watch on the hormone levels of patients on antiepileptics. According to the literature, it is most likely that AEDs affect hepatic microsomal enzyme systems and accelerate the metabolism of thyroid hormones. Human uridine diphosphate glucuronosyl transferase (UGT) could also be responsible for glucuronidation and thus metabolisation of thyroid hormones, as high levels of UGT have been observed after AED therapy. However, it was not established that valproate induced thyroid dysfunction. This could be due to enzyme induction.
In the present study, there was a significant pathological decrease in serum sodium in the valproate and oxcarbazepine group. These findings are similar to the studies by Patel et al. [ 23] and Branten et al. [ 24], which concluded that sodium valproate can cause hyponatremia with SIADH-like syndrome. Previous studies also showed hyponatremia with oxcarbazepine, presenting as nausea, fatigue, and worsening of GTCS [25,26]. Possible reasons could be a direct effect of oxcarbazepine on renal-collecting tubules or enhanced responsiveness to the circulating antidiuretic hormone (ADH).
There was a significant pathological increase in fasting blood glucose with phenytoin and valproate. A previous study by Rubeaan et al. showed that phenytoin-induced hyperglycemia could be primarily due to the inhibition of insulin release and increased insulin insensitivity [27]. In the present study, we found that hyperglycemia correlated with valproate. Similar results were found in a study performed on diabetic rabbits, where valproate resulted in hypoglycaemia [28]. In contrast, a study by Rakitins et al. showed that valproate was associated with weight gain, thus resulting in other metabolic changes and hyperinsulinemia [29].
## Conclusions
The advent of newer antiepileptic drugs has uncovered wider horizons for the treatment of epilepsy. Each of these drugs has a unique mechanism of action, making it less prone to resistance. Metabolic derangements are a key determinant for the compliance of these drugs, as they can result in a predisposition to other co-morbidities. For instance, a change in the lipid profile or blood sugar profile can put the epilepsy patient at risk of cardiovascular diseases or diabetes. Although there are studies indicating better side-effect profiles for the newer AEDs as compared to the conventional ones, further studies are needed to confirm their superiority. Periodic monitoring of the various metabolic parameters is useful, and, together with patient counselling, can improve the effectiveness of the antiepileptic drugs.
## References
1. **Epilepsy**. *Epilepsy. Published online* (2023)
2. Amudhan S, Gururaj G, Satishchandra P. **Epilepsy in India I: epidemiology and public health**. *Ann Indian Acad Neurol* (2015) **18** 263-277. PMID: 26425001
3. Gursahani R, Gupta N. **The adolescent or adult with generalized tonic-clonic seizures**. *Ann Indian Acad Neurol* (2012) **15** 81-88. PMID: 22566718
4. Newton RW. **When is drug treatment not necessary in epilepsy? Factors that should influence the decision to prescribe**. *J R Soc Med* (2004) **97** 15-19. PMID: 14702357
5. Hixson JD. **Stopping antiepileptic drugs: when and why?**. *Curr Treat Options Neurol* (2010) **12** 434-442. PMID: 20730110
6. Camfield P, Camfield C. **When is it safe to discontinue AED treatment?**. *Epilepsia* (2008) **49** 25-28
7. Brodie MJ, Sills GJ. **Combining antiepileptic drugs--rational polytherapy?**. *Seizure* (2011) **20** 369-375. PMID: 21306922
8. Reimers A. **New antiepileptic drugs and women**. *Seizure* (2014) **23** 585-591. PMID: 24908139
9. Mintzer S. **Metabolic consequences of antiepileptic drugs**. *Curr Opin Neurol* (2010) **23** 164-169. PMID: 20125012
10. Who-umc T, Who-umc T, Centres N, Monitoring ID. **The use of the WHO-UMC system for standardised case causality assessment. Published online 2018: 2-7**. (2013)
11. Horowitz DL. **Cholesterol and heart disease**. *Compr Ther* (1981) **7** 3-4
12. Murphy MJ, Dominiczak MH. **Efficacy of statin therapy: possible effect of phenytoin**. *Postgrad Med J* (1999) **75** 359-360. PMID: 10435174
13. Luoma PV, Reunanen MI, Sotaniemi EA. **Changes in serum triglyceride and cholesterol levels during long-term phenytoin treatment for epilepsy**. *Acta Med Scand* (1979) **206** 229-231. PMID: 495231
14. Karpale M, Hukkanen J, Hakkola J. **Nuclear receptor PXR in drug-induced hypercholesterolemia**. *Cells* (2022) **11** 1-22
15. Du R, Li Y, Yu X, Wang C. **Effect of sodium valproate on lipid metabolism in children with epilepsy and its mechanisms**. *Chinese Pharm J* (2014) **49** 982-985
16. Kantoush MM, El-shahawy AK, Sokker SS, Serag HR. **Effects of treatment with antiepileptic drugs on serum lipid profile in epileptic children**. *Alexandria J Pediatr* (1998) **12** 153-158
17. Jaeri S, Islamiyah WR. **The potential mechanisms of effect of valproic acid on lipid profiles: an updated review**. *J Epileptol* (2018) **26** 47-52
18. Manimekalai K, Visakan B, Salwe KJ, Murugesan S. **Evaluation of effect of antiepileptic drugs on serum lipid profile among young adults with epilepsy in a tertiary care hospital in Pondicherry**. *J Clin Diagn Res* (2014) **8** 0-9
19. Dahiya K, Kumar R, Bansal A, Dhankhar R, Ghalaut VS, Chugh K, Kumari S. **Phenytoin monotherapy in epilepsy and its effect on TSH**. *Int J Pharm Sci Res* (2019) **7** 9-25
20. Pattan V, Candula N, Adhikari R, Kashyap R. **Phenytoin- Medication That Warrants Deviation From Standard Approach for Thyroid Lab Interpretation**. *Cureus* (2020) **12** 0
21. Kim SH, Chung HR, Kim SH. **Subclinical hypothyroidism during valproic acid therapy in children and adolescents with epilepsy**. *Neuropediatrics* (2012) **43** 135-139. PMID: 22618302
22. Yılmaz U, Yılmaz TS, Akıncı G, Korkmaz HA, Tekgül H. **The effect of antiepileptic drugs on thyroid function in children**. *Seizure* (2014) **23** 29-35. PMID: 24091037
23. Patel KR, Meesala A, Stanilla JK. **Sodium valproate-induced hyponatremia: a case report**. *Prim Care Companion J Clin Psychiatry* (2010) **12** 0
24. Branten AJ, Wetzels JF, Weber AM, Koene RA. **Hyponatremia due to sodium valproate**. *Ann Neurol* (1998) **43** 265-267. PMID: 9485070
25. Ejikeme C, Elkattawy S, Kayode-Ajala F, Abuaita S, Khazai M. **Oxcarbazepine-induced hyponatremia: a case report and comprehensive literature review**. *Cureus* (2021) **13** 0
26. Sachdeo RC, Wasserstein A, Mesenbrink PJ, D'Souza J. **Effects of oxcarbazepine on sodium concentration and water handling**. *Ann Neurol* (2002) **51** 613-620. PMID: 12112108
27. al-Rubeaan K, Ryan EA. **Phenytoin-induced insulin insensitivity**. *Diabet Med* (1991) **8** 968-970. PMID: 1838051
28. Bari MA. **Sodium valproate effects on lipid profile and glucose level in normal and diabetic rabbits**. *Iraqi J Med Sci* (2018) **16** 247-257
29. Rakitin A, Kõks S, Haldre S. **Valproate modulates glucose metabolism in patients with epilepsy after first exposure**. *Epilepsia* (2015) **56** 0-5
|
---
title: Molecular characterization and expression of TGFβRI and TGFβRII and its association
with litter size in Tibetan sheep
authors:
- Junxia Zhang
- Mingming Li
- Ruizhe Sun
- Na He
- Xiaocheng Wen
- Xueping Han
- Zenghai Luo
journal: Veterinary Medicine and Science
year: 2023
pmcid: PMC10029873
doi: 10.1002/vms3.1013
license: CC BY 4.0
---
# Molecular characterization and expression of TGFβRI and TGFβRII and its association with litter size in Tibetan sheep
## Abstract
This study was undertaken to explore the characterization, expression analysis of TGFβR I and TGFβR II genes, and its association with litter size in Tibetan sheep. The TGFβ receptors (TGFβRI and TGFβR II) may play an important role in sheep reproduction.
### Backgrounds
Transforming growth factor‐β (TGF‐β) type I receptor (TGFβRI) and type II receptor (TGFβRII) are the members of the TGFβ superfamily, which are potent regulators of cell proliferation and differentiation in many organ systems, and they play key roles in multiple aspects of follicle development.
### Objectives
We aimed to explore the characterization, expression analysis of TGFβRI and TGFβRII genes, and the association with litter size in Tibetan sheep.
### Methods
In this study, we cloned the complete coding sequences of TGFβRI and TGFβRII genes in Tibetan sheep and analyzed their genomic structures.
### Results
The results showed that percentages of sequences homology of the two proteins in Tibetan sheep were the most similar to Ovis aries ($100\%$), followed by *Bos mutus* ($99\%$). The RT‐qPCR showed that two genes were expressed widely in the different tissues of Tibetan sheep. The TGFβRI expression was the highest in the lung ($p \leq 0.05$), followed by the spleen and ovary ($p \leq 0.05$). The TGFβRII expression was significantly higher in uterus than that in lung and ovary ($p \leq 0.05$). In addition, the χ 2 test indicated that all ewes in the population were in Hardy–Weinberg equilibrium, and the population was in medium or low polymorphic information content status. We also found four Single Nucleotide Polymorphism (SNPs), g.9414A > G, g.28881A > G, g.28809T > C, g.10429G > A in sheep TGFβRI gene and g.63940C > T, g.63976C > T, g.64538C > T, g.64504T > A in TGFβRII gene. Three genotypes, except for g.64504T > A, and three haplotypes were identified in each gene. linkage disequilibrium analysis indicated that there was strong linkage disequilibrium in each gene. The association analysis showed that the four SNPs of TGFβRI were associated with litter size ($p \leq 0.05$), and g.63940C > T of TGFβRII was confirmed to be associated with litter size ($p \leq 0.05$).
### Conclusions
Based on these preliminary results, we can assume that TGFβ receptors (TGFβRI and TGFβRII) may play an important role in sheep reproduction.
## INTRODUCTION
The TGFβ superfamily is a large and expanding group of regulatory polypeptides (Kumari et al., 2021). The molecular signalling pathway of the TGFβ superfamily has been conserved throughout the six hundred million years of metazoan evolution (Loveland & Hime, 2005), which is critical for regulating a variety of developmental events, including cell proliferation, differentiation, and matrix secretion (Elvin et al., 2000; Nong et al., 2019). The family members of the TGFβ superfamily are candidates for mediating important oocyte activity (Elvin et al., 2000; Lankford & Weber, 2010). TGFβ receptor type I (TGFβRI) and the TGFβ receptor typeII (TGFβRII) are important members of the TGFβ superfamily. TGFβ signalling, important in ovary development is mediated through TGFβRI and TGFβRII. These receptors are interdependent components of a heteromeric complex, as receptor I requires receptor II for TGFβ binding and receptor II requires receptor I for signalling (Attisano & Wrana, 2002; Knight & Glister, 2003; Sun et al., 2008). TGFβ ligands bind and activate TGFβ receptor complex composed of the type II (TGFβRII) and type I subunits (TGFβRI), which phosphorylate Smad2 and Smad3. Activated Smad$\frac{2}{3}$ forms transcriptional complexes with Smad4 and other transcriptional factors and regulates the transcription of genes (Serizawa et al., 2013). It has been reported that they play an important role in many aspects of follicular development, including activation of resting primordial follicles, proliferation and apoptosis of Granulosa cells and membrane cells, steroid formation, gonadotropin receptor expression, oocyte maturation, ovulation, and luteinization (Elvin et al., 2000). The various type I and type II receptors through which each of these ligands can signal are expressed by pre‐granulosa cells/granulosa cells of the corresponding early follicle stages, making these cells potential targets for paracrine signalling (Shimasaki et al., 2004).
A few genes of the TGFβ superfamily were investigated, and their association with reproductive performance has been observed in lines of sheep (Elvin et al., 2000; Jia et al., 2020; Shi et al., 2021; Shimasaki et al., 2004; Xu et al., 2010). However, little is known about the roles of other members of the TGF‐β superfamily in Tibetan sheep; thus, the potential interaction of members of the TGFβ superfamily and their relationship with lambing traits is unclear. Therefore, the objectives of this study were to characterize the complete or partial cDNA sequences of TGFβI and TGFβII, determine the expressing mRNA encoding TGFβRI and TGFβRII, and analyze the effects of TGFβI and TGFβII on litter size in Tibetan sheep.
## Animals
Tibetan sheep were obtained from sheep farm (Xiangkemeiduo Sheep Industry Co. Ltd., Qinghai, China), and the experimental group included 433 ewes, which were selected randomly. The health and reproduction records of the animals were kept by the farmers. Their litter size was obtained from reproduction records. All efforts were made to minimize discomfort during the blood collection. Blood samples were collected from the jugular vein under the supervision of qualified veterinarians. Genomic DNA was extracted from blood sample of each sheep using an EasyPure Blood Genomic DNA Kit (TransGen Biotech, Beijing, China). Three ewes were selected from purebred herds of the same farm in Qinghai province. The three selected ewe (6 months old) were healthy, similar in weight, and pastured in similar conditions of grassland. After slaughtered, and the tissues from hypothalamus, hypophysis, heart, liver, spleen, lung, kidney, ovary, oviduct, uterus, rumen, duodenum, and longissimus dorsi were collected and immediately frozen in liquid nitrogen, and then stored at −80°C. The RNA of tissues was extracted by TransZol (TransGen Biotech). Total RNA for each tissue was reverse‐transcribed to cDNA by TransScript One‐Step gDNA Removal.
## cDNA cloning and sequence analysis
The cDNA sequences of sheep TGFβRI and TGFβRII (GenBank Accession No. NM_001009224.1, XM_012179698.3, respectively) were used as templates. The primer pairs were designed using the coding regions of the two genes (Table 1). The PCR program was as follows: 94°C for 5 min; 30 cycles of 94°C for 30 s, Tm°C for 30 s and 72°C for 40 s, followed by one cycle at 72°C for 5 min. The above PCR products were electrophoresed on a $1\%$ agarose gel.
**TABLE 1**
| Gene name | Primer name | Primer sequences (5ʹ–3ʹ) | Size (bp) | Tm (°C) |
| --- | --- | --- | --- | --- |
| TGFβRI | TGFβRI‐CDS‐S TGFβRI‐CDS‐A | GAGGCGAGGCTTGTTGAG TGGCAGTTTCCTGGGTCT | 1751 | 55 |
| TGFβRII | TGFβRII‐CDS1‐S TGFβRII‐CDS1‐A | GCACGTTCCCAAGTCGGTT ATGTCCTTCTCCGTCTTCC | 801 | 61 |
| | TGFβRII‐CDS2‐S TGFβRII‐CDS2‐A | GCTGGTCATCTTCCAAGTGACA ACCTCTTTCCACTAGTATGGCTG | 1537 | 60 |
| TGFβRI | TGFβRI‐expression‐S TGFβRI‐expression‐A | TGGCAGAGCTGTGAAGCCTTG AGCCTAGCTGCTCCATTGGCAT | 77 | 63 |
| TGFβRII | TGFβRII‐expression‐S TGFβRII‐expression‐A | CTGGCCAACAGTGGGCAGGTG CGTCTGCTTGAAGGACTCGACATT | 99 | 63 |
| GAPDH | GAPDH‐expression‐S GAPDH‐expression‐A | GCGAGATCCTGCCAACATCAAGT CCCTTCAGGTGAGCCCCAGC | 105 | 63 |
The PCR product was purified using agarose gel DNA extraction kit (Takara, Dalian, China), and cloned into pMD19‐T vector (volume of 10 μl of 50 ng DNA, 50 ng pMD19‐T vector, 5 μl Solution I, incubated at 4°C overnight), then transformed into *Escherichia coli* DH5a (Takara) competent cell and grown in Luria‐Bertani (LB) agar plate with Amp. White colonies were selected (10 colonies for each sample) and cultured in liquid medium for 5 h, and then sequenced by Shanghai Sangon Biological Engineering Company. Alignments of multiple sequences were carried out with BLAST (NCBI, http://blast.ncbi.nlm.nih.gov/Blast.cgi). Open reading frame (ORF) Finder (http://www.ncbi.nlm.nih.gov/projects/gorf/) was used to determine the ORF and predict the amino acid sequence. ProtParam (http://web.expasy.org/protparam/) was used to predict the physical parameters of each protein. The hydrophilicity and hydrophobicity were analyzed using ProtScale (https://web.expasy.org/protscale/). Prediction of the secondary structure of each protein and its variants was analyzed using SOPMA (http://www.compbio.dundee.ac.uk/www‐jpred/). SWISS‐MODEL (http://swissmodel.expasy.org/) was used to predict the protein signal peptide and protein tertiary structure. Amino acid sequences alignments were conducted using DNAStar Lasergene (MegAlign), and a phylogenetic tree was established using the MEGA7 software.
## Tissue expression analysis of sheep TGFβRI and TGFβRII
The primers for real‐time PCR were designed according to mRNA sequences of TGFβRI and TGFβRII gene (GenBank accession No: XM_004004226.4 and XM_012099309.2) (Table 1). The reaction volume was 20 μl containing 10 μl of SYBR Premix ExTaq II, 0.4 μl (10 μmol/L) forward primer, 0.4 μl (10 μmol/L) reverse primer, 1 μl cDNA, and 8.2 μl ddH2O. The PCR cycle consisted of 94°C for 2 min; then, 45 cycles of 94°C for 10 s, 60°C for 20 s, and 72°C for 1 s; and an extension of 72°C for 5 min. The qPCR was performed using a CFX96 Touch Real‐Time PCR (BIO‐RAD, USA). All experiments were performed in triplicate, and GAPDH was used as the reference gene. The 2−∆∆CT method was used to analyze the data (Livak & Schmittgen, 2001).
## SNP identification and genotyping
TGFβRI and TGFβRII genes Single Nucleotide Polymorphism (SNPs) were screened Using the dbSNP database (http://www.ncbi.nlm.nih.gov/snp) and verified by DNA sequencing. Improved multiplex ligation detection reaction (iMLDRTM) was used for genotyping following the instrument operating guidelines. Genotypic, allelic frequencies, and genetic parameters were directly calculated following previous description (Zhao et al., 2013). The linkage disequilibrium (LD) was conducted using the Haploview software.
## Association analysis
The association analysis between genotypes and litter size of ewes was determined according to a general linear model (GLM) program. All statistical analyses were performed using SPSS 23.0. Results with $p \leq 0.05$ were considered significantly different. Based on the characteristics of sheep, the statistical model was as follows: yijn=μ+Pi+Gj+IPG+eijn, where yijn is the phenotypic value, μ is the population mean, *Pi is* the fixed effect of the ith parity ($i = 1$, 2, or 3), *Gj is* the fixed effect of the jth genotype ($j = 1$, 2, 3), IPG is the interaction effect of parity and genotype, and eijn is the random residual.
## Molecular cloning and sequence analysis of sheep TGFβRI and TGFβRII
In this study, 1751 bp of the sheep TGFβRI gene was cloned, which contained a calculated ORF of 1506 bp encoding a protein of 501 amino acid residues. Additionally, sheep TGFβRII contains ORFs of 1416 bp, and they encode proteins of 471 amino acid residues. The molecular weights of TGFβRI and TGFβRII are 55960.70 and 52879.55 Da, respectively, and the theoretical isoelectric points are 7.19 and 5.84, respectively. All of them include 20 types of amino acid composition. The total numbers of negatively charged residues (Asp + Glu) are 56 and 60, respectively, and the total numbers of positively charged residues (Arg + Lys) are 56 and 49, respectively. TGFβRI formula is C2470H3936N688O723S35. The total number of atoms is 7852. The Aliphatic index is 89.92; the grand average of hydropathicity (GRAVY) is −0.097; TGFβRII formula is C2339H3689N637O703S28. The total number of atoms is 7396. The Aliphatic index is 90.45; grand average of hydropathicity (GRAVY) is −0.170. A positive value indicates that the protein is hydrophobic, and a negative value indicates that it is hydrophilic, so all of them are hydrophilic. Subcellular localization of TGFβRI is $55.6\%$ in endoplasmic reticulum; it is $22.2\%$ in Golgi, $11.1\%$ in plasma membrane, $11.1\%$ in extracellular, including cell wall. And TGFβRII is $34.8\%$ in nuclear; it is $26.1\%$ in cytoplasmic, $21.7\%$ in mitochondrial, $4.3\%$ in endoplasmic reticulum, $4.3\%$ in peroxisomal, $4.3\%$ in vesicles of secretory system, $4.3\%$ in vacuolar. The proteins of TGFβRI and TGFβRII have signal peptides. WEBSEQUENCE Number of predicted Transmembrane *Helices is* 2 and 1. There were potential N‐glycosylation sites at amino acids 41, 148, 268, and 348. The potential values were 0.7000, 0.8358, 0.6267, and 0.4613 in Tibetan sheep TGFβRI. There were potential N‐glycosylation sites at amino acids 70, 94, and 266. The potential values were 0.5869, 0.6930, and 0.6757 in Tibetan sheep TGFβRII. There are 50 and 45 potential phosphate sites in sheep TGFβRI and TGFβRII, respectively. Amino acid sequence alignment and percentage of sequences homology of the two proteins in Ovis aries, Bos taurus, Bos mutus, Homo sapiens, Sus scrofa, Mus musculus, Maylandia zebra, *Canis lupus* familiaris, Pan troglodytes, Macaca mulatta, and *Gallus gallus* showed that Tibetan sheep TGFβRI and TGFβRII are most similar to O. aries ($100\%$), then B. mutus ($99\%$), and least similar to C. lupus familiaris ($82\%$), respectively (Figures 1 and 2).
**FIGURE 2:** *Percentage of sequences homology of TGFβRI (a) and TGFβRII (b)*
The structure prediction of sheep TGFβRI protein was performed by online protein analysis system SOPMA. The results showed that the extension chain composed of alpha‐helix, extended strand, beta turn, and random coil accounted for $39.32\%$, $11.38\%$, $3.39\%$, and $67.27\%$, respectively, and for TGFβRII protein, they were $33.76\%$, $15.92\%$, $3.82\%$ and $46.50\%$, respectively (Figures 3 and 4).
**FIGURE 3:** *Secondary structure of TGFβRI (a) and TGFβRII (b) protein. Blue represents alpha helix, green represents beta turn, purple represents random coil, and red represents extended strand.* **FIGURE 4:** *Structural features of TGFβRI (a) and TGFβRII (b)*
## Expression profile analysis
The RT‐qPCR was used to investigate the general tissue distributions of TGFβRI and TGFβRII. As shown in Figures 5 and 6, two genes were widely expressed in hypothalamus, pituitary, heart, liver, spleen, lung, kidney, ovary, oviduct, uterus, rumen, duodenum, and longissimus dorsi in Tibetan sheep. The TGFβRI was expressed with the highest level in the lung ($p \leq 0.05$), followed by the spleen, uterus and ovary ($p \leq 0.05$), and almost no expression in longissimus dorsi. The TGFβRII expression was the highest in uterus than in other tissues ($p \leq 0.05$), followed by lung, ovary, and spleen ($p \leq 0.05$). There were no significant differences among oviduct, duodenum, rumen, kidney, pituitary, liver, and heart ($p \leq 0.05$). Except for the hypothalamus, the expression of TGFβRII gene in longissimus dorsi was lower than that in the other tissues ($p \leq 0.05$).
**FIGURE 5:** *Expression of TGFßRI mRNA in different tissues of Tibetan sheep. Note: different superscripts indicate significant difference (p < 0.05).* **FIGURE 6:** *Expression of TGFßRII mRNA in different tissues of Tibetan sheep. Note: different superscripts indicate significant difference (p < 0.05).*
## Population genetic analysis of polymorphism in sheep TGFβRI and TGFβRII
In this study, four polymorphic nucleotide sites (SNPs) were identified in Tibetan sheep TGFβRI and TGFβRII genes, respectively. All mutations were synonymous mutations. Except for SNP g.64504T > A, the other SNPs were classified as three genotypes (Table 2), and three haplotypes were identified in each gene (Table 3). Linkage disequilibrium (r 2) block indicated strong linkage disequilibrium in two genes, respectively (Figure 7). In addition, Ho, He, Ne, and polymorphic information content (PIC) of Tibetan sheep TGFβRI were 0.72, 0.28, 1.40, and 0.24, respectively, and for TGFβRII, 0.76, 0.24, 1.31, and 0.21, respectively. Tibetan sheep were in medium PIC status at g.63940C > T and g.28809T > C sites, and the others have low PIC status (Table 4). The χ 2 test indicated that all ewes in the populations were in Hardy–Weinberg equilibrium.
## Association analysis of SNPs with litter size in Tibetan sheep
The effects of Tibetan sheep TGFβRI and TGFβRII SNPs on litter size of the experimental populations were studied. The results showed that the g.9414A > G, g.28881A > G, g.28809T > C, and g.10429G > A of sheep TGFβRI were associated with litter size ($p \leq 0.05$). In contrast, the TGFβRII g.63940C>T substitution was associated with litter size ($p \leq 0.05$). However, the SNPs, g.63976C > T, g.64538C > T, and g.64538C > T had no association with litter size (Table 5). All results indicated that TGFβRI and TGFβRII contributed to phenotype values.
**TABLE 5**
| Gene | Loci | Genotype | Number | Litter size |
| --- | --- | --- | --- | --- |
| TGFβR1 | g.9414A > G | AA | 383 | 1.07 ± 0.26b |
| | | AG | 45 | 1.00 ± 0.00b |
| | | GG | 5 | 1.40 ± 0.55a |
| | g.28881A > G | AA | 383 | 1.07 ± 0.26b |
| | | AG | 45 | 1.00 ± 0.00b |
| | | GG | 5 | 1.40 ± 0.55a |
| | g.28809T > C | TT | 154 | 1.09 ± 0.29b |
| | | TC | 217 | 1.03 ± 0.16b |
| | | CC | 62 | 1.16 ± 0.37a |
| | g.10429G > A | GG | 412 | 1.07 ± 0.26b |
| | | GA | 44 | 1.00 ± 0.00b |
| | | AA | 7 | 1.40 ± 0.55a |
| TGFβRII | g.63940C > T | CC | 175 | 1.05 ± 0.21b |
| | | CT | 188 | 1.07 ± 0.26ab |
| | | TT | 70 | 1.11 ± 0.32a |
| | g.63976C > T | CC | 336 | 1.07 ± 0.26 |
| | | CT | 90 | 1.07 ± 0.25 |
| | | TT | 7 | 1.00 ± 0.00 |
| | g.64538C > T | CC | 402 | 1.07 ± 0.25 |
| | | CT | 29 | 1.14 ± 0.35 |
| | | TT | 2 | – |
| | g.64504T > A | TT | 395 | 1.07 ± 0.25 |
| | | TA | 38 | 1.11 ± 0.31 |
| | | AA | 0 | – |
## DISCUSSION
TGFβ superfamily is evolutionarily conserved and plays fundamental roles in cell growth and differentiation (Attisano & Wrana, 1996; Hill, 1996). TGFβ superfamily signalling is essential for female reproduction (Li, 2014), and TGFβ superfamily affects the reproductive physiology of animals (Nie et al., 2014), for example, influencing the development of follicles by regulating the proliferation or apoptosis of Granulosa cells in the follicles and causing follicular atresia (Li, 2014; Nie et al., 2014; Ovchinnikov & Wolvetang, 2011). TGFβRI and TGFβRII, core components of TGF‐β superfamily, are important intraovarian growth factors (Ester et al., 1999), so TGFβRI and TGFβRII genes were used as candidate genes for reproductive traits to study. TGFβRI and TGFβRII are serine‐threonine kinases that signal through the Smad family of proteins (Ovchinnikov & Wolvetang, 2011; Sun et al., 2008). TGFβ1 binds to the TGFβRII, which in turn recruits the binding of TGFβRI to form a heterotetramer. TGFβRI then phosphorylates and activates the Smad2 protein (Li, 2014; Nie et al., 2014) after combining with Smad4, followed by translocation to the nucleus where the activated Smad complex. Then, it is involved in regulating transcriptional responses on target genes (Ikushima & Miyazono, 2010). At present, there are few studies on the structural characterization of TGFβRI and TGFβRII. In this study, we analyzed the homology of sheep TGFβRI and TGFβRII proteins with 10 other species, respectively. It was found that TGFβRI and TGFβRII have a higher percentage of sequences homology indicating that TGFβRI and TGFβRII were conserved across the above‐mentioned species.
Type I and type II TGFβ receptors appear to be ubiquitously expressed in most cell types (Knight & Glister, 2006). The tissue expression profiles revealed that TGFβRI and TGFβRII have broad expression patterns in Tibetan sheep. Ovarian cells have been shown to produce TGFβRI and TGFβRII, whose expression was first detected in preantral follicles and continues throughout the subsequent stages of follicular development (Knight & Glister, 2006). The mRNA and proteins of TGFβ receptors type I and II exist in the human oocyte, and receptor type I exists in blastocysts, indicating a selective expression of transcripts for TGFβ receptors in oocytes and blastocysts (Osterlund & Fried, 2000). Expression of TGFβRI mRNA was observed in the sheep ovary, while expression of TGFβRII mRNA within the follicle was limited to the theca (Juengel et al., 2004). The expression of TGFβR mRNA/protein in preantral follicles has been documented in several species including rodents, human, sheep, and cattle (Chow et al., 2001; Juengel et al., 2004; Osterlund & Fried, 2000; Roy, 2000). We found that both TGFβRI and TGFβRII were expressed in ovary, oviduct, uterus, hypothalamus, and hypophysis, as well as in other tissues. We also found that expression of TGFβRI was the highest in lung, followed by spleen, uterus, and ovary, and TGFβRII was higher in uterus than in the other tissues.
TGFβRI and TGFβRII are essential for regulating the growth and differentiation of ovarian follicles and thus fertility (Juengel et al., 2004). Osterlund and Fried [2000] reported that TGFβ receptor types I and II are present in human oocytes. Juengel et al. [ 2004] reported that the expression of mRNAs encoding TGF‐β1 and TGF‐β2 as well as both type I and II TGF‐β receptors were observed in the theca of small growing follicles indicating that TGF‐βs may be regulating thecal cell function in an autocrine manner. Expression of mRNA encoding TGF‐β type I and II receptors was observed in luteal cells, stroma, the vascular system, and surface epithelium suggesting that TGF‐βs may also regulate other cell types in the sheep ovary (Juengel et al., 2004). A similar pattern of expression for the TGFβRII mRNA was observed in mouse follicles, with expression most prominent in the theca and barely detectable in granulosa cells (Juengel et al., 2004). TGFβRI and TGFβRII are important cell regulators that play important regulatory roles in ovary development and animal reproduction. In this study, g.9414A > G, g.28881A > G, g.28809T > C, g.10429G > A in TGFβRI, and g.63940C > T in TGFβRII were associated with litter sizes in Tibetan sheep, and TGFβRI and TGFβRII can be used as molecular markers for improving the reproduction performance of Tibetan sheep. However, further studies on the association between the two genes and productive performance of different sheep breeds are required.
## CONCLUSIONS
In this study, we cloned cDNA sequences of TGFβRI and TGFβRII genes in Tibetan sheep and the sequences homology of the two genes was the most similar to O. aries, followed by B. mutus. We also found that TGFβRI and TGFβRII were expressed in the different tissues of Tibetan sheep, and the expression of TGFβRI was the highest in lung, followed by spleen, uterus, and ovary, and TGFβRII expression was higher in uterus than the other tissues. The g.9414A > G, g.28881A > G, g.28809T > C, g.10429G > A mutations of TGFβRI and g.63940C > T of TGFβRII were screened out, and three different genotypes as well as three different haplotypes were identified for each gene. The g.9414A > G, g.28881A > G, g.28809T > C, and g.10429G > A mutation of sheep TGFβRI had an association with litter size, and the TGFβRII g.63940C > T was associated with litter size. Thus, our results indicate that TGFβRI and TGFβRII can be used as candidate genes for the improvement of reproductive performance of Tibetan sheep during breeding.
## AUTHOR CONTRIBUTIONS
Formal analysis, methodology, validation, and writing—original draft, and writing—review and editing: Junxia Zhang. Data curation and investigation: Mingming Li, Na He, and Ruizhe Sun: Conceptualization, methodology, and writing—review and editing: Xiaocheng Wen. Data curation and validation: Xueping Han. Conceptualization, methodology, and writing—review and editing: Zenghai Luo.
## CONFLICTS OF INTEREST
The authors declare no conflict of interest.
## FUNDING INFORMATION
The funds of Science and Technology Planning Program of Qinghai (Science and Technology Department of Qinghai Province) (Grant No. 2020‐ZJ‐786); Outstanding Person of Kunlong: Rural Revitalization Program (Grant No. [ 2020]9).
## ETHICS STATEMENT
All experiments in this study were performed following the approved guidelines of the Regulation of the Standing Committee of Qinghai People's Congress. All experimental protocols and the collection of samples were approved by the Ethics Committee of Qinghai University under permission No. SL‐2021027.
## PEER REVIEW
The peer review history for this article is available at https://publons.com/publon/10.1002/vms3.1013.
## DATA AVAILABILITY STATEMENT
The data that support the findings of this study will be shared upon reasonable request to the corresponding author.
## References
1. Attisano L., Wrana J. L.. **Signal transduction by members of the transforming growth factor‐β superfamily**. *Cytokine & Growth Factor Reviews* (1996) **7** 327-339. DOI: 10.1016/S1359-6101(96)00042-1
2. Attisano L., Wrana J. L.. **Signal transduction by the TGF‐β superfamily**. *Science* (2002) **296** 1646-1647. DOI: 10.1126/science.1071809
3. Chow J. F., Lee K. F., Chan S. T., Yeung W. S.. **Quantification of transforming growth factor beta1 (TGFbeta1) mRNA expression in mouse preimplantation embryos and determination of TGFbeta receptor (type I and type II) expression in mouse embryos and reproductive tract**. *MHR Basic Science of Reproductive* (2001) **7** 1047-1056. DOI: 10.1093/molehr/7.11.1047
4. Elvin J. A., Yan C., Matzuk M. M.. **Oocyte‐expressed TGF‐beta superfamily members in female fertility**. *Molecular & Cellular Endocrinology* (2000) **159** 1-5. DOI: 10.1016/s0303-7207(99)00185-9
5. Ester P., Carl‐Henrik H., Peter T. D.. **Specificity, diversity, and regulation in TGF‐β superfamily signaling**. *FASEB Journal Official Publication of the Federation of American Societies for Experimental Biology* (1999) **13** 2105-2124. DOI: 10.1096/fasebj.13.15.2105
6. Hill C. S.. **Signalling to the nucleus by members of the transforming growth factor‐beta (TGF‐beta) superfamily**. *Cellular Signalling* (1996) **8** 533-544. DOI: 10.1016/S0898-6568(96)00122-2
7. Ikushima H., Miyazono K.. **Cellular context‐dependent “colors” of transforming growth factor‐beta signaling**. *Cancer Science* (2010) **101** 306-312. DOI: 10.1111/j.1349-7006.2009.01441.x
8. Jia J. L., Jin J. P., Chen Q., Yuan Z., Li H. Q., Bian J. H., Gui L. S.. **Eukaryotic expression, Co‐IP and MS identify BMPR‐1B protein‐protein interaction network**. *Biological Research* (2020) **53** 24. DOI: 10.1186/s40659-020-00290-7
9. Juengel J. L., Bibby A. H., Reader K. L., Lun S., Quirke L. D., Haydon L. J., McNatty K. P.. **The role of transforming growth factor‐beta (TGF‐beta) during ovarian follicular development in sheep**. *Reproductive Biology and Endocrinology* (2004) **2** 78. DOI: 10.1186/1477-7827-2-78
10. Knight P. G., Glister C.. **Local roles of TGF‐β superfamily members in the control of ovarian follicle development**. *Animal Reproduction Science* (2003) **78** 165-183. DOI: 10.1016/s0378-4320(03)00089-7
11. Knight P. G., Glister C.. **TGF‐beta superfamily members and ovarian follicle development**. *Reproduction* (2006) **132** 191-206. DOI: 10.1530/rep.1.01074
12. Kumari A., Shonibare Z., Monavarian M., Arend R. C., Lee N. Y., Inman G. J., Mythreye K.. **TGFβ signaling networks in ovarian cancer progression and plasticity**. *Clinical & Experimental Metastasis* (2021) **38** 139-161. DOI: 10.1007/s10585-021-10077-z
13. Lankford S. E., Weber G. M.. **Temporal mRNA expression of transforming growth factor‐beta superfamily members and inhibitors in the developing rainbow trout ovary**. *General & Comparative Endocrinology* (2010) **166** 250-258. DOI: 10.1016/j.ygcen.2009.09.007
14. Li Q. L.. **Transforming growth factor β signaling in uterine development and function**. *Journal of Animal Science and Biotechnology* (2014) **5** 52. DOI: 10.1186/2049-1891-5-52
15. Loveland K. L., Hime G.. **TGFbeta superfamily members in spermatogenesis: Setting the stage for fertility in mouse and Drosophila**. *Cell & Tissue Research* (2005) **322** 141-146. DOI: 10.1007/s00441-005-0008-0
16. Livak K. J., Schmittgen T. D.. **Analysis of relative gene expression data using real‐time quantitative PCR and the 2**. *Methods* (2001) **25** 402-408. DOI: 10.3892/ijo.2013.1860
17. Nie Y. N., Li S. Y., Yi Y. E., Su W. L., Chai X. L., Jia D. X., Wang Q.. **Effects of astragalus injection on the TGFβ/Smad pathway in the kidney in type 2 diabetic mice**. *BMC Complementary & Alternative Medicine* (2014) **14** 148. DOI: 10.1186/1472-6882-14-148
18. Nong X., Rajbanshi G., Chen L., Li J., Li J.. **Effect of artesunate and relation with TGF‐β**. *Archives for Dermatological Research* (2019) **311** 761-772. DOI: 10.1007/s00403-019-01960-7
19. Osterlund C., Fried G.. **TGFbeta receptor types I and II and the substrate proteins Smad 2 and 3 are present in human oocytes**. *Molecular Human Reproduction* (2000) **6** 498-503. DOI: 10.1093/molehr/6.6.498
20. Ovchinnikov D. A., Wolvetang E. J., Atwood C.. *Embryonic stem cells: The hormonal regulation of pluripotency and embryogenesis* (2011) 193. DOI: 10.5772/15585
21. Roy S. K.. **Regulation of transforming growth factor‐beta‐receptor type I and type II messenger ribonucleic acid expression in the hamster ovary by gonadotropins and steroid hormones**. *Biology of Reproduction* (2000) **62** 1858-1865. DOI: 10.1095/biolreprod62.6.1858
22. Serizawa M., Takahashi T., Yamamoto N., Koh Y.. **Combined treatment with erlotinib and a transforming growth factor‐β type I receptor inhibitor effectively suppresses the enhanced motility of erlotinib‐resistant non‐small‐cell lung cancer cells**. *Journal of Thoracic Oncology* (2013) **8** 259-269. DOI: 10.1097/JTO.0b013e318279e942
23. Shi A. D., Hillege M. M. G., Wüst R. C. I., Wu G., Jaspers R. T.. **Synergistic short‐term and long‐term effects of TGF‐β1 and 3 on collagen production in differentiating myoblasts**. *Biochemical and Biophysical Research Communications* (2021) **547** 176-182. DOI: 10.1016/j.bbrc.2021.02.007
24. Shimasaki S., Moore R. K., Otsuka F., Erickson G. F.. **The bone morphogenetic protein system in mammalian reproduction**. *Endocrine Reviews* (2004) **25** 72-101. DOI: 10.1210/er.2003-0007
25. Sun T., Xin Z., Jin Z., Wu Y., Gong Y.. **Effect of TGF‐beta/Smad signaling on sertoli cell and possible mechanism related to complete sertoli cell‐only syndrome**. *Molecular and Cellular Biochemistry* (2008) **319** 1-7. DOI: 10.1007/s11010-008-9869-3
26. Xu Y. F., Li E., Han Y. D., Chen L., Xie Z.. **Differential expression of mRNAs encoding BMP/Smad pathway molecules in antral follicles of high‐ and low‐fecundity Hu sheep**. *Animal Reproduction Science* (2010) **120** 47-55. DOI: 10.1016/j.anireprosci.2010.02.009
27. Zhao H. Y., Wu X. F., Cai H. F., Pan C. Y., Lei C., Chen H., Lan X. Y.. **Genetic variants and effects on milk traits of the caprine paired‐like homeodomain transcription factor 2 (PITX2) gene in dairy goats**. *Gene* (2013) **532** 203-210. DOI: 10.1016/j.gene.2013.09.062
|
---
title: Effect of Moringa leaf flavonoids on the production performance, immune system,
and rumen fermentation of dairy cows
authors:
- Ji Liu
- Yan Wang
- Ling Liu
- Guangming Ma
- Yonggen Zhang
- Jian Ren
journal: Veterinary Medicine and Science
year: 2022
pmcid: PMC10029909
doi: 10.1002/vms3.993
license: CC BY 4.0
---
# Effect of Moringa leaf flavonoids on the production performance, immune system, and rumen fermentation of dairy cows
## Abstract
In this study, we found that adding 50mg/BW moringa leaf flavonoids increased the protein content in milk, reduced the number of somatic cells, and had no effect on milk production, milk fat, total solids, and other indicators. At the same time, it improves the antioxidant capacity and immunity of dairy cows and does not affect blood physiological and biochemical indicators and rumen fermentation parameters.
### Background
Unreasonable use of antibiotics in animals is a major concern and will remain so, thus affecting people's health. However, the application of plant extracts can better solve this problem.
### Objectives
The purpose of this study was to study the effect of Moringa leaf flavonoids on the production performance, immunity, and rumen fermentation of dairy cows.
### Methods
Nine *Holstein multiparous* cows (average weight: 550 kg; days of lactation: 150 ± 6 days) were used in the experiment, using a 3 × 3 Latin square design. Cows were divided into three groups, each of which was supplemented with 0, 50, or 100 mg/body weight (BW) *Moringa oleifera* leaf flavonoids. Each experimental period consisted of three periods of 21 days, and the prefeeding period lasted 15 days.
### Results
Our results indicated that supplementation with Moringa leaf flavonoids increased the protein content and decreased the number of somatic cells in milk; had little effect on the biochemical indices of blood, the rumen fermentation, and serum biochemical indicators; and improved the activity of antioxidant enzymes, the antioxidant capacity, and immunity.
### Conclusions
Addition of 50 mg/BW Moringa leaf flavonoids to cow enhanced the antioxidant and immunity capacity in dairy cows but did not affect physiological levels of common biochemical parameters in blood or fermentation parameters in rumen.
## INTRODUCTION
Antibiotics are commonly used as supplement in ruminant feeds to improve production performance. However, the inappropriate use of antibiotics in feeds can lead to drug resistance in animals and drug residues in food, thus posing threats to human health (Byarugaba, 2004). The nontherapeutic use of antibiotics in animal feeds have been banned in the European Union and many other countries, due to the concerns of the spread of antibiotic resistance from animal products to consumers (Rojo et al., 2015). The use of some antibiotics has been banned worldwide. Therefore, there is a need to evaluate the potential of natural antibacterial agents (e.g., plant extracts) to improve the production capacity of animals (Kamel, 2006).
Flavonoids are polyphenolic compounds commonly found in plants as secondary metabolites. Flavonoids have proven abilities in enhancing immunity in animals, through their anti‐pathogenic, anti‐oxidation, and anti‐inflammatory functions. Flavonoids can affect rumen metabolism and rumen microbial fermentation (Linville et al. 2018) and promote the circulation of blood in dairy cows, thereby enhancing their metabolism, promoting the absorption of nutrients, and increasing milk production (Zhan et al., 2017). Flavonoids also have a weak estrogenic effect, which regulates the secretion of growth hormone and promotes breast development (Liu et al., 2020). Flavonoids can promote nitrogen metabolism in the rumen of dairy cows and can reduce methane production (Leake & Rankin, 1990).
Moringa is rich in nutrients and biologically active compounds, and thus has great potential to be used as a supplement in livestock feeds. The leaves, seeds, and bark of Moringa can be readily consumed by cows, sheep, goats, pigs, chickens, and rabbits (Radványi et al., 2013). Consumption of Moringa has been proven to improve the health, growth performance, milk production, and meat quality of livestock (Nardone & Valfrè, 1999). Moringa contains various flavonoids in the leaves, roots, flowers, and seed coats, and the contents of flavonoids vary depending the geographic origins of Moringa. The most common flavonoids in Moringa leaves are kaempferol, quercetin, isorhamnetin, and apigenin (Milugo et al., 2013).
So far, there have been few reports of the effects of Moringa leaves extracts rich in flavonoids on dairy cows. The aim of this study was to evaluate the effects of Moringa leaves flavonoids extract on the production performance, immune responses, and ruminal fermentation of dairy cows.
## Materials
Nine Holstein cows from Wonderson pasture in Harbin were used in the experiment. The cows were in milk for 150 ± 6 days, with average body weight (BW) of 550 ± 25 kg. The experiment was conducted in a 3 × 3 Latin square design, consisting of three 21‐day experimental periods, with 15 days for adaptation. Cows were divided into three groups and fed a total mixed ration (TMR) (Table 1) supplemented with 0, 50, and 100 mg/BW Moringa leaf flavonoids. Moringa leaf flavonoids (purity, $50\%$) were produced by Shaanxi Quanao‐Engineering Co. Ltd. (Xi′an, China) (See Table 2, 3, 4, 5, 6).
## Testing of milk samples
The milk samples were stored at 4°C and then submitted to Heilongjiang DHI Testing Center for analysis with a multifunctional dairy analyzer. The parameters measured included milk fat rate, milk total solids, lactose rate, milk protein rate, milk urea nitrogen, and somatic cell count.
## Blood biochemical test
On the 20th day of each phase of the test, blood was collected from the tail vein with a coagulation vacuum tube (containing inert separating gel) before and 2 h after ingestion. The blood was allowed to stand for 1 h at room temperature, followed by centrifugation to separate the serum, which was then aliquoted and stored at −20°C until further analysis.
The main blood biochemical indicators were total protein, albumin (ALB), globulin (GLB), glucose (GLU), high‐density lipoprotein (HDL), and low‐density lipoprotein (LDL). The main hormones were prolactin (PRL), triiodothyronine (T3), and tetraiodothyronine (T4).
## Volatile acid determination
The temperature of the inlet and detector was 220°C. The oven temperature scheme started with initial temperature at 120°C for 3 min, and then increased to 180°C at 10°C/min. The carrier gas was high‐purity nitrogen. The port pressure was maintained at 90 kpa; the hydrogen flow rate was 40 ml/min, the airflow rate was 400 ml/min, and the makeup flow rate was 45 ml/min.
## Determination of antioxidant and immune indicators
The immunoglobulin content (IgG, IgM, and IgA) was determined by radioimmunoassay, which was completed by the Beijing Huaying Institute of Biotechnology. Antioxidant indicators, total superoxide dismutase (T‐SOD), total antioxidant capacity kit (T‐AOC), glutathione peroxidase (GSH‐Px), catalase (CAT), and malondialdehyde (MDA) were analyzed using the test kit (Nanjing Jiancheng Institute of Biology) according to the manufacturer's instructions.
## Effect of Moringa leaf flavonoids on dairy cow performance
The protein content in the milk was increased by the diet supplemented with 50 mg/BW Moringa leaf flavonoids compared to the control ($p \leq 0.05$), which was the lowest in this latter diet among all three diets administered. There were no significant changes in milk fat, lactose, and total solids among the three diets. The number of somatic cells in the diet supplemented with 50 mg/BW Moringa leaf flavonoids decreased significantly and was significantly lower than the control, suggesting that the addition of 50 mg/BW Moringa leaf flavonoids significantly reduced the number of somatic cells. The above results show that the addition of an appropriate amount of Moringa leaf flavonoids to the diet can increase the protein content in milk and significantly reduce the number of somatic cells.
## Effect of Moringa leaf flavonoids on blood biochemical indicators
The blood total protein, ALB, GLB, low‐density lipoprotein (LDH), and GLU content increased with the increase of the amount of Moringa leaf flavonoids with some noticeable variations. The content of HDL increased as the amount of the Moringa leaf flavonoids increased, and HDL was the highest in the diet supplemented with 100 mg/BW. However, no significant differences were observed among treatments. The above data indicate that Moringa leaf flavonoids had no significant effect on blood biochemical indicators.
## Effect of Moringa leaf flavonoids on the volatile acids of milk
The content of acetic acid, propionic acid, and butyric acid in milk was lower in the treatments supplementation with Moringa leaf flavonoids than the control. The content of valeric acid, isobutyric acid, and isovaleric acid was the highest in the treatment with 50 mg/BW Moringa leaf flavonoids, but no significant differences were observed among diets. Based on the above results, the addition of Moringa leaf flavonoids in the diet did not show significant effects on rumen fermentation parameters.
## Effect of Moringa leaf flavonoids on antioxidant indexes
The addition of 50 mg/BW Moringa leaf flavonoids significantly reduced the content of MDA, significantly increased the content of T‐AOC and T‐SOD, and had no significant effect on the content of GSH‐PX and CAT.
## Effect of Moringa leaf flavonoids on immune performance indicators
The IgA content was significantly higher under supplementation with 50 mg/BW Moringa leaf flavonoids than under supplementation with 0 mg/BW and 100 mg/BW Moringa leaf flavonoids. The T3 content was significantly higher under supplementation with 50 mg/BW Moringa leaf flavonoids than under supplementation with 0 mg/BW and 100 mg/BW Moringa leaf flavonoids. Moringa leaves had no significant effect on the flavonoids PRL and T4 and other blood biochemical indicators. The above results indicate that Moringa leaf flavonoids affect serum biochemical indicators.
## DISCUSSION
The addition of appropriate amount of Moringa leaf flavonoids did not increase milk production in dairy cows. Some studies have shown that the addition of flavonoids in the form of grape pomace powder, green tea, and turmeric extracts does not increase milk production (Williamson et al., 2005). The results of our study are consistent with these findings. However, previous work has shown that the addition of propolis flavone as a feed additive has no effect on the number of somatic cells in the milk of dairy cows, which is inconsistent with the results of our study showing that the addition of Moringa leaf flavonoids decreases the number of somatic cells. This inconsistency can be explained by the different sources, types, and structures of flavonoids used (Zicker & Wedekind, 2005). Elevated somatic cells in milk indicate poor breast health and milk quality, which is a major problem in modern animal husbandry. Cows with elevated somatic cells produce less milk than healthy cows, which results in major economic losses to the dairy industry. Approximately $70\%$ of subclinical mastitis is related to a temporary or permanent decrease in milk production, which mainly stems from inflammatory damage to breast tissue (Eckersall et al., 2006). *In* general, nutritional factors play a key role in enhancing resistance to breast infections, and excessive amounts of micronutrients with effective antioxidant and immune‐enhancing properties added to the diet such as plant extracts have been reported to enhance breast health and reduce the somatic cell count (Scaletti et al., 2003). Milk production is lower in cows with higher initial somatic cell levels. Previous studies have shown that breast infections may significantly reduce milk production (Tikofsky et al., 2003). Previous work indicates that damage to secretory tissue and fibrous tissue can lead to breast infection, which causes a temporary or permanent decrease in milk production (Lauzon et al., 2006). The biologically active compounds in Moringa leaf flavonoids discovered in this study show anti‐inflammatory activity and reduce the number of somatic cells.
No effect of supplementation with Moringa leaf flavonoids on the concentration of HDL and LDL was observed. Hosoda et al. [ 2006] also observed no difference in the concentrations of HDL and LDL. Changes in blood biochemical indicators can provide insight into animal health, nutrient metabolism, and production performance. The level of serum total protein, the GLB content, and ALB content reflect the body's absorption of protein and protein metabolism. In this study, Moringa leaf flavonoids showed no effect on serum total protein, including GLB, ALB, HDL, GLU, and LDL.
Rumens are unique digestive organs of ruminants that contain a large number of microorganisms, which can obtain nutrients from feed through fermentation processes. Regulation of the rumen can improve the utilization rate of feed and promote the production performance of ruminants. The volatile acids produced by rumen fermentation of dairy cows are important energy sources for dairy cows, among which acetic acid, propionic acid, and butyric acid are the three main volatile acids accounting for more than $80\%$ of the total volatile acids (Foiklang & Toburan, 2011). Volatile fatty acids are the main product of carbohydrate degradation in the rumen, the main energy source for ruminants, and the main raw material for synthetic milk fat.
Xi et al. [ 2007] found that the addition of vegetable oils significantly reduced total volatile fatty acid (TVFA) production. Using in vitro culture methods and hay as the fermentation substrate, Vasta et al. [ 2008] showed that cinnamon oil has no significant effect on the concentration of volatile acids and the ratio of acetic acid and propionic acid.
Kwekkeboom et al. [ 1993] studied the effect of capsaicin on the rumen fermentation of lactating dairy cows and showed that these plant essential oils had no effect on TVFAs, their composition, and ammonia nitrogen. The findings of our study were similar to their results, as Moringa leaf flavonoids did not affect the rumen fermentation index.
Free radicals play an important role in immunity and signal transduction, but excessive free radicals can lead to the lipid peroxidation of cell membranes (Turner et al., 2004). Endogenous antioxidant enzymes, including SOD, CAT, and GSH‐Px, neutralize oxidative stress, which is the main form of intracellular defence (Manor et al., 1999). The antioxidant effect is achieved by the conversion of oxygen‐free radicals into weakly oxidized forms (Kamel, 2005). Supplementation of flavonoids can improve antioxidant capacity, improve non‐specific immunity, and reduce oxidative stress by increasing SOD and GSH‐Px activity while reducing the concentration of MDA. The underlying mechanism involves flavonoids acting as reducing agents and hydrogen donors to neutralize oxygen‐free radicals and remove hydrogen peroxide and superoxide ions (Kahkonen, 1999).
In this study, the addition of Moringa leaf flavonoids increased the activity of T‐SOD ($p \leq 0.01$) and T‐AOC ($p \leq 0.05$) in plasma and reduced the activity of MDA. These results are consistent with the observations that flavonoids induced antioxidant enzyme activity in rat liver and kidney. In addition, Campbell et al. [ 2013] observed that the flavonoids of Tartary buckwheat in the diet increased the antioxidant capacity of sheep plasma and liver. Similarly, Vasta and Luciano [2011] reported that adding tannins to the diet of lambs can improve the overall antioxidant status parameters of muscles. The possible reason for these findings is that flavonoids may selectively induce the expression of antioxidant enzyme genes by activating the nuclear factor E2‐related factor Nrf2. Yeh and Yen [2005] found that the mRNA abundance of SOD, GSH‐Px, and CAT in the liver of rats with phenolic compounds added was higher than that of Nrf2 protein in the control group and may play a key role in the activation of antioxidant genes induced by phenolic compounds.
Therefore, Moringa leaf flavonoids are antioxidants with diverse effects worthy of further study. The results of our study suggest that Moringa leaf flavonoids enhance the body's immune function by stimulating endogenous antioxidant enzyme activity and protecting the body from oxidative stress.
The immune response is closely related to animal health (Ingvartsen & Moyes, 2013). Determination of the serum immunoglobulin concentration is one of the most commonly used methods for evaluating immunity. The addition of 50 mg/BW Moringa leaf flavonoids significantly increased the content of IgA, and changes in IgG and IgM were not pronounced. Measurement of immune indicators revealed that Moringa leaf flavonoids have a regulatory effect on the immune system of cows and improve the body's immune function. Immunoglobulin is a globulin that can be used as an antibody and has a chemical structure similar to antibodies. Immunoglobulins play a role in antigen‐specific binding and regulation of the immune response.
The addition of 50 mg/BW Moringa leaf flavonoids significantly increased the content of T3, and changes in PRL and T4 were not pronounced.
Moringa leaf flavonoids can increase the concentration of immunoglobulin in the plasma of dairy cows. Moringa leaf flavonoids have been shown to play a role in immune function. Based on the results of previous research, we speculate that Moringa leaf flavonoids stimulate mitochondrial adenosine triphosphate production, which may be mediated by active oxidants and enhance cellular/mitochondrial antioxidant status. However, the specific mechanism by which Moringa leaf flavonoids enhance immune function requires further study (Leonard et al., 1984). The addition of Astragalus extract in the diet can reduce the plasma T3 level, which may lead to a decrease in the basal metabolic rate of heat‐stressed broilers and increase the energy required for broiler growth. These findings differ from the results of this experiment. This may be explained by the difference in the number of stomachs between ruminants and other animals. There was no difference in the concentration of T4 among the groups in this experiment. Cooke et al. [ 2011] showed that T3 in dairy cows is higher in diets containing tea pollen; however, tea pollen has no effect on the T4 concentration (Taraghijou et al., 2012). In this study, T3 increased significantly in dairy cows fed with 50 mg/BW Moringa flavonoids, which is consistent with the results of previous studies.
## CONCLUSIONS
In conclusion, the addition of 50 mg/BW Moringa leaf flavonoids to cow diet increased the protein content in milk, but had no effect on milk yield/day, mild fat, total solids, or other quality traits. Additionally, Moringa leaf flavonoids enhanced the antioxidant and immunity capacity in dairy cows but did not affect physiological levels of common biochemical parameters in blood or fermentation parameters in rumen.
## AUTHOR CONTRIBUTIONS
Conceptualization, resources, writing—original draft, and writing—review and editing: Ji Liu. Software and supervision: Yan Wang. Data curation and supervision: Ling Liu. Formal analysis and methodology: Guangming Ma. Funding acquisition: Yonggen Zhang. Data curation and funding acquisition: Jian Ren.
## CONFLICTS OF INTEREST
The authors declare no conflict of interest.
## ETHICS STATEMENT
All procedures used in this study were approved by the Institutional Animal Care and Use Committee of Northeast Agricultural University.
## PEER REVIEW
The peer review history for this article is available at https://publons.com/publon/10.1002/vms3.993.
## DATA AVAILABILITY STATEMENT
All data generated or analyzed during this study are included in this article.
## References
1. Busquet M., Calsamiglia S., Ferret A., Kamel C.. **Plant extracts affect in vitro rumen microbial fermentation**. *Journal of Dairy Science* (2006) **89** 761-771. PMID: 16428643
2. Byarugaba D. K.. **A view on antimicrobial resistance in developing countries and responsible risk factors**. *International Journal of Antimicrobial Agents* (2004) **24** 105-110. PMID: 15288307
3. Campbell M. A., López J. A., Sado T., Miya M.. **Pike and salmon as sister taxa: Detailed intraclade resolution and divergence time estimation of Esociformes+Salmoniformes based on whole mitochondrial genome sequences**. *Gene* (2013) **530** 57-65. PMID: 23954876
4. Cooke R. F., Bohnert D. W., Moriel P., Hess B. W., Mills R. R.. **Effects of polyunsaturated fatty acid supplementation on ruminal in situ forage degradability, performance, and physiological responses of feeder cattle**. *Journal of Animal Science* (2011) **89** 3677-3689. PMID: 21680784
5. Eckersall P. D., Young F. J., Nolan A. M., Knight C. H., McComb C., Waterston M. M., Hogarth C. J., Scott E. M., Fitzpatrick J. L.. **Acute phase proteins in bovine milk in an experimental model of**. *Journal of Dairy Science* (2006) **89** 1488-1501. PMID: 16606719
6. Foiklang S., Toburan W. W.. **Effects of various plant protein sources in high‐quality feed block on feed intake, rumen fermentation, and microbial population in swamp buffalo**. *Tropical Animal Health & Production* (2011) **43** 1517-1524. PMID: 21479843
7. Hosoda H., Kojima M., Kangawa K.. **Biological, physiological, and pharmacological aspects of ghrelin**. *The Japanese Pharmacological Society* (2006) **100** 398-410
8. Ingvartsen K. L., Moyes K.. **Nutrition, immune function and health of dairy cattle**. *Animal* (2013) **7** 112-122. PMID: 23031687
9. Kahkonen K.. **Multi‐character model of the construction project definition process**. *Automation in Construction* (1999) **8** 625-632
10. Kamel F. C.. **Effect of garlic oil and four of its compounds on rumen microbial fermentation**. *Journal of Dairy Science* (2005) **88** 4393-4404. PMID: 16291631
11. Kwekkeboom J., Boer M. D., Tager J. M., Groot C. D.. **CD40 plays an essential role in the activation of human B cells by murine EL4B5 cells**. *Immunology* (1993) **79** 439. PMID: 7691726
12. Lauzon K., Zhao X., Lacasse P.. **Deferoxamine reduces tissue damage during endotoxin‐induced mastitis in dairy cows**. *Journal of Dairy Science* (2006) **89** 3846-3857. PMID: 16960060
13. Leake D. S., Rankin S. M.. **Flavonoids inhibit the oxidative modification of low density lipoproteins by macrophages**. *Biochemical Pharmacology* (1990) **39** 1743-1750. PMID: 2344371
14. Leonard J. L., Enrique S. J., Kaplan M. M., Mellen S. A., Visser T. J., Reed L. P.. **Acute posttranscriptional regulation of cerebrocortical and pituitary iodothyronine 5′‐deiodinases by thyroid hormone**. *Endocrinology* (1984) **114** 998-1004. PMID: 6697974
15. Linville K., Casper D. P., Osorio J. S.. **377 In vitro analysis of rumen microbial fermentation at different temperatures**. *Journal of Animal Science* (2018) **95** 182-183
16. Liu J., Ma G., Wang Y., Zhang Y.. **Moringa oleifera leaf flavonoids protect bovine mammary epithelial cells from hydrogen peroxide‐induced oxidative stress in vitro**. *Reproduction in Domestic Animals* (2020) **55** 711-719. PMID: 32144827
17. Manor Y., Handsher R., Halmut T., Neuman M., Abramovitz B., Mates A., Mendelson E.. **A double‐selective tissue culture system for isolation of wild‐type poliovirus from sewage applied in a long‐term environmental surveillance**. *Applied & Environmental Microbiology* (1999) **65** 1794-1797. PMID: 10103285
18. Milugo T. K., Omosa L. K., Ochanda J. O., Owuor B. O., Ochieng J. W.. **Antagonistic effect of alkaloids and saponins on bioactivity in the quinine tree (**. *BMC Complementary & Alternative Medicine* (2013) **13** 285-285. PMID: 24160735
19. Nardone A., Valfrè F.. **Effects of changing production methods on quality of meat, milk and eggs**. *Livestock Production Science* (1999) **59** 165-182
20. Radványi D., Juhász R., Kun S., Szabó‐Nótin B., Barta J.. **Preliminary study of extraction of biologically active compounds from elderberry (**. *Acta Alimentaria* (2013) **42** 63-72
21. Rojo R., Kholif A. E., Salem A. Z. M., Elghandour M. M. Y., Odongo N. E., Montes De Oca R., Rivero N., Alonso M. U.. **Influence of cellulase addition to dairy goat diets on digestion and fermentation, milk production and fatty acid content**. *Journal of Agricultural Science* (2015) **153** 1514-1523
22. Scaletti R. W., Trammell D. S., Smith B. A., Harmon R. J.. **Role of dietary copper in enhancing resistance to Escherichia coli mastitis**. *Journal of Dairy Science* (2003) **86** 1240-1249. PMID: 12741549
23. Taraghijou P., Safaeiyan A., Mobasseri M., Ostadrahimi A.. **The effect of n‐3 long chain fatty acids supplementation on plasma peroxisome proliferator activated receptor gamma and thyroid hormones in obesity**. *Journal of Research in Medical Sciences* (2012) **17** 942-946. PMID: 23825994
24. Tikofsky L. L., Barlow J. W., Santisteban C., Schukken Y. H.. **A comparison of antimicrobial susceptibility patterns for**. *Microbial Drug Resistance* (2003) **9** S39-S45. PMID: 14633366
25. Turner R., Baron T., Wolffram S., Minihane A. M., Cassidy A., Rimbach G., Weinberg P. D.. **Effect of circulating forms of soy isoflavones on the oxidation of low density lipoprotein**. *Free Radical Research Communications* (2004) **38** 209-216
26. Vasta V., Luciano G.. **The effects of dietary consumption of plants secondary compounds on small ruminants' products quality**. *Small Ruminant Research* (2011) **101** 150-159
27. Vasta V., Nudda A., Cannas A., Lanza M., Priolo A.. **Alternative feed resources and their effects on the quality of meat and milk from small ruminants**. *Animal Feed Science & Technology* (2008) **147** 223-246
28. Williamson G., Barron D., Shimoi K., Terao J.. **In vitro biological properties of flavonoid conjugates found in vivo**. *Free Radical Research* (2005) **39** 457-469. PMID: 16036321
29. Xi D., Wanapat M., Deng W., He T., Mao H.. **Comparison of Gayal (**. *Asian Australasian Journal of Animal Sciences* (2007) **20** 1208-1214
30. Yeh C. T., Yen G. C.. **Induction of hepatic antioxidant enzymes by phenolic acids in rats is accompanied by the multidrug resistance‐associated protein 3 mRNA expression**. *Journal of Nutrition* (2005) **136** 11-15
31. Zhan J., Wu C., Liu M., Su X., Zhan K., Zhang C., Zhao G.. **Effects of alfalfa flavonoids on the production performance, immune system, and ruminal fermentation of dairy cows**. *Asian‐Australasian Journal of Animal Sciences* (2017) **30** 1416-1424. PMID: 28423878
32. Zicker S. C., Wedekind K. J.. (2005)
|
---
title: The innovative model based on artificial intelligence algorithms to predict
recurrence risk of patients with postoperative breast cancer
authors:
- Lixuan Zeng
- Lei Liu
- Dongxin Chen
- Henghui Lu
- Yang Xue
- Hongjie Bi
- Weiwei Yang
journal: Frontiers in Oncology
year: 2023
pmcid: PMC10029918
doi: 10.3389/fonc.2023.1117420
license: CC BY 4.0
---
# The innovative model based on artificial intelligence algorithms to predict recurrence risk of patients with postoperative breast cancer
## Abstract
### Purpose
This study aimed to develop a machine learning model to retrospectively study and predict the recurrence risk of breast cancer patients after surgery by extracting the clinicopathological features of tumors from unstructured clinical electronic health record (EHR) data.
### Methods
This retrospective cohort included 1,841 breast cancer patients who underwent surgical treatment. To extract the principal features associated with recurrence risk, the clinical notes and histopathology reports of patients were collected and feature engineering was used. Predictive models were next conducted based on this important information. All algorithms were implemented using Python software. The accuracy of prediction models was further verified in the test cohort. The area under the curve (AUC), precision, recall, and F1 score were adopted to evaluate the performance of each model.
### Results
A training cohort with 1,289 patients and a test cohort with 552 patients were recruited. From 2011 to 2019, a total of 1,841 textual reports were included. For the prediction of recurrence risk, both LSTM, XGBoost, and SVM had favorable accuracies of 0.89, 0.86, and 0.78. The AUC values of the micro-average ROC curve corresponding to LSTM, XGBoost, and SVM were 0.98 ± 0.01, 0.97 ± 0.03, and 0.92 ± 0.06. Especially the LSTM model achieved superior execution than other models. The accuracy, F1 score, macro-avg F1 score (0.87), and weighted-avg F1 score (0.89) of the LSTM model produced higher values. All P values were statistically significant. Patients in the high-risk group predicted by our model performed more resistant to DNA damage and microtubule targeting drugs than those in the intermediate-risk group. The predicted low-risk patients were not statistically significant compared with intermediate- or high-risk patients due to the small sample size (188 low-risk patients were predicted via our model, and only two of them were administered chemotherapy alone after surgery). The prognosis of patients predicted by our model was consistent with the actual follow-up records.
### Conclusions
The constructed model accurately predicted the recurrence risk of breast cancer patients from EHR data and certainly evaluated the chemoresistance and prognosis of patients. Therefore, our model can help clinicians to formulate the individualized management of breast cancer patients.
## Introduction
According to estimates from the Global Cancer Observatory (GLOBOCAN) in 2020, the incidence of female breast cancer ranked first, surpassing even lung cancer [1]. Meanwhile, in China, the incidence of breast cancer has risen to the fourth among all cancer types and shows a trend of younger age [2]. Breast cancer seriously harms women’s life and health. Accurately evaluating the recurrence risk of postoperative breast cancer patients can greatly improve their prognosis through appropriate treatment [3].
With the digitization of medical information, machine learning models have been applied in oncology (4–6). In 2021, artificial intelligence (AI) was used to predict the occurrence of breast cancer metastasis by learning from clinical electronic health record (EHR) data to support individualized diagnosis for patients [7]. EHRs contain numerous longitudinal records, including histopathology, molecular markers related to breast cancer, radiology, and clinical information. However, the manual integration of prognostic information from EHRs by clinical experts is time-consuming, laborious, and costly [8, 9]. Therefore, precisely assessing the recurrence risk and improving the efficiency of clinical evaluation plays a crucial role in controlling the disease burden of breast cancer.
Support vector machine (SVM) is a powerful learning algorithm that is capable of addressing various dimensions of data through different kernel functions. For example, breast cancer cells were classified in vitro with an accuracy of $93\%$ using linear and radial basis function (RBF) kernel SVMs [10]. Extreme gradient boosting (XGBoost) is a decision tree-based algorithm that is widely used in machine learning. It minimizes the loss function of the model through a gradient descent algorithm and implements the speed and performance of gradient-boosted decision trees [11]. Furthermore, artificial neural networks (ANN) comprise a fundamental component of deep learning algorithms, demonstrating great potential in building high prediction accuracy (12–15). Currently, AI algorithms have proven successful in processing clinical image data, obtaining desired prediction results (16–18). For example, a two-stage convolutional neural network (CNN) model was proposed to predict the occurrence of myocardial infarction and localize the site of infarction based on vectorcardiogram signals [19]. However, further research is needed to process clinical non-image data using machine learning.
In this study, we aimed to develop an artificial intelligence prediction model to regressively identify the recurrence risk of breast cancer patients after operation. We used SVM, XGBoost, and LSTM algorithms to integrate the histopathological and molecular characteristics of tumors in patients’ EHRs. We also validated the model’s performance in predicting risk categories for patients who received neoadjuvant and postoperative chemotherapy or postoperative chemotherapy alone, which can provide a precise assessment for personalized medicine for cancer patients. Our study made the following important contributions:
## Clinicopathological data of breast cancer patients
This retrospective study was designed to predict the risk of breast cancer patients who underwent surgery through automated models. The overall methodology of this study is illustrated in Figure 1. A total of 1,962 patients with breast cancer were recruited from the Third Affiliated Hospital of Harbin Medical University from $\frac{11}{05}$/2011, to $\frac{29}{12}$/2019. There were 121 ($6.1\%$) patients initially excluded because of incomplete pathological examination results or lack of clinical notes. Eventually, 1,841 patients were included in this retrospective analysis. A total of 432 patients underwent different treatment regimens following surgery and had complete treatment information, including radiotherapy, chemotherapy alone, combination therapy, endocrine therapy, and targeted therapy. Completed follow-up information of postoperative patients was collected, containing the surveillance of contralateral breast cancer, lymph node metastases, distant organ metastases, and other relevant monitoring. All study procedures were thoroughly reviewed and received ethical approval from the Harbin Medical University Ethics Committee. Informed written consent was obtained from each participant prior to their involvement in the study. A detailed description of the patient characteristics is found in Supplemental Table 1.
**Figure 1:** *Overall workflow of the study. Histopathological features of breast cancer were first extracted and annotated by retrospective retrieving EHR data of breast cancer patients. The preprocessed information was next generated as a feature set, and models were trained to predict the recurrence risk of patients. The model was further validated in patients who received neoadjuvant and postoperative chemotherapy or postoperative chemotherapy alone.*
## Data parsing and feature extraction
Data preprocessing plays an important role in the application of machine learning [20]. Since medical professionals have multiple expressions in medical reports, we first broke each note into blocks and standardized the reporting format, mainly regarding its clinical concepts and attributes. More details are explained in the Supplement Data. We further used natural language processing (NLP) based on the regular expression (regex) in Python to extract all key terms from EHRs [21]. The regular expression can quickly analyze large volumes of textual information and has a specialized syntax. We compiled the regular expression pattern for each feature according to this specified syntax, thus accurately matching specific strings [22]. An example below shows the feature extraction process: re.compile (r’ER\([\+\-].*?\)|ER\([\+\-].*?\)’,re. I) In addition, NegEx was used to identify whether a term had been negated, effectively rectifying false-positive cases [23]. For instance, “lymph nodes are not enlarged,” “lymph node-negative,” and “no evidence of lymphovascular invasion” were considered negative. After feature extraction, we combined all the features and created a dataset. The output values of all samples were displayed on the label with “=1” to match successfully; else, it was “=0” (Table 1). The missing values in our raw data were filled in “=0.” Eventually, the accuracy of feature extraction was estimated using the actual values in the original text snippets [24]. Correct extraction was considered true positive (TP) when the extracted values matched the actual values. A classification for the module was regarded as false positive (FP) when the extracted values did not match the actual values. Missed entities were considered false negative (FN) when actual values were available, but no extracted values were reported. It was regarded as a true negative (TN) when no extracted values were produced and there were no actual values. Supplementary Table 2 shows the confusion matrix for evaluated extraction.
**Table 1**
| Feature names | Feature descriptions | Illustrative example |
| --- | --- | --- |
| Patients | Patient ID | 616402.0 |
| Age | Years | 51.0 |
| Menopausal status | Pre = 0Post = 1 | 1.0 |
| ER | Estrogen receptor-positive = 1Estrogen receptor-negative = 0 | 1.0 |
| PR | Progesterone receptor-positive = 1Progesterone receptor-negative = 0 | 1.0 |
| HER2 | HER2/neu gene overexpressed or amplified = 1HER2/neu gene neither overexpressed nor amplified = 0 | 0.0 |
| Tumor size | Pathological tumor size ≤2cm = 0Pathological tumor size >2 cm = 1 | 0.0 |
| LNM | Positive lymph node metastasis = 1Negative lymph node metastasis = 0 | 0.0 |
| Number of LNM | The number of lymph node metastases | |
| G1 | Pathology grade I = 1Pathology grade II, pathology grade III = 0 | 1.0 |
| G2 | Pathology grade II = 1Pathology grade I, pathology grade III = 0 | 0.0 |
| G3 | Pathology grade III = 1Pathology grade I, pathology grade II = 0 | 0.0 |
| LVI | Lympho-vascular invasion (+) = 1Lympho-vascular invasion (-) = 0 | 0.0 |
| Ki-67 (%) | The median pathology of Ki-67 proliferative index | 5.0 |
| Distant organ metastasis | Distant organ metastasis = 1Non-distant organ metastasis = 0 | 0.0 |
| Label | Low risk = 0Intermediate risk = 1High risk = 2 | 0.0 |
## Model prediction and evaluation
The recurrence risk of postoperative breast cancer patients was according to the clinical guidelines for the diagnosis and treatment of Breast cancer in 2021 of Chinese Anti-Cancer Association, Committee of Breast Cancer Society (CACA-CBCS) (Supplemental Table 3) [25]. It has performed an important premise that Chinese clinicians base on to comprehensively assess and formulate treatment regimens.
Each prediction model was implemented through the Scikit-learn library in Python. First, the dataset was loaded into the Pandas dataframe and split into a training set ($70\%$) and a test ($30\%$) set with the train_test_split function. In order to avoid extreme values, the fillna() function was executed to fill the vacant values with default values and scale numerical variables for range adjustment.
SVM is a supervised learning algorithm commonly employed in binary classification and regression problems. The basic principle of SVM is to identify a decision boundary so that samples can be separated from different classes [26]. In this study, the sklearn.svm. SVC function was adopted to solve the three classification problems. The linear kernel was first selected to linearly classify the training set due to the significantly larger feature size than the sample size [27]. Since our data are linearly non-separable, slack variables were employed during the training process to improve the generalization ability of the model by allowing some sample points to be misclassified. Additionally, the decision hyperplane was determined by soft margin maximization and dual problem settlement. The application of multiclass classification utilized a one-vs-rest voting strategy, which means that three binary classifiers are trained [28]. Finally, samples from the test set were predicted separately and the category with the highest probability was subsequently assigned as the final prediction.
The XGBoost model contains K base learners in which each learner predicts the Xi outcome of the i-th input and then acquires the final classification result by pooling each output fK (Xi) [11]. The xgb. XGBClassifier function was adopted to build the model based on a set of relevant parameters such as learning rate, number of trees, and gamma. The grid search strategy was applied from the Sklearn interface to obtain the best-optimized hyperparameters, which optimizes the model’s performance and avoids overfitting issues [29]. Next, the XGBoost model was trained using the determined parameters and 10-fold cross-validation [30]. The most important features that were taken into consideration were as follows: distant organ metastasis, lymph node metastasis (including the number of lymph node metastases), HER-2, ER, PR, and Ki-67 expression; pathology grade; menopausal status; age; and lympho-vascular invasion. Eventually, values were predicted for the test set and evaluated by the module to obtain the reliability of the XGBoost model [31].
LSTM simulates the memory storage capacity of our brain, which develops novel artificial intelligence algorithms. Compared with traditional neural network algorithms, LSTM can precisely deal with more complex problems related to time series or sequential data [32, 33]. In this study, the LSTM model was constructed in Keras. After learning meaningful features, dense layers were used to map features from the high-dimensional data space to a low-dimension representation space and finally become a column vector, in which the number of columns is the same as risk categories [34]. Specifically, the first column corresponded to the low risk with “class 0,” the second column corresponded to the intermediate risk with “class 1,” and the third column corresponded to the high risk with “class 2.” Each patient would be obtained a column vector with a sum of 1 through the softmax_layer. For example, the predicted result for one patient was shown [0.2,0.7,0.1]. A predicted value of 0.2 represented the probability of class “0,” 0.7 represented the probability of class “1,” and 0.1 represented the probability of class “2.” This column vector indicated that this patient was finally classified as the maximum value of the predicted label “class-1” (intermediate-risk). Moreover, backpropagation was utilized to optimize the parameters of this model, thus minimizing the loss function [35]. Feature units were randomly dropped through dropout layers during each feedforward training to avoid overfitting issues and obtain a generalization model.
To determine the favorable model, the performance of each model was compared through the receiver operating characteristic curve (ROC) and the area under the curve (AUC). Since our dataset has an imbalanced distribution of samples, consisting of disparate sample sizes in each class. Precision (positive predictive value)–recall (sensitivity) curves were also applied as indicators to further assess each model’s performance [36]. Other important metrics for evaluation include accuracy, F1 score, macro-average, micro-average, and weight-average. A further explanation of these indicators is provided in the Supplement Data.
## Statistical analysis in patients
We divided the 85 patients treated with chemotherapy alone after surgery into chemo-sensitive and chemo-resistant groups based on each patient’s response to chemotherapy. The inclusion criteria of chemotherapeutic resistance are as follows [37, 38]: [1] An increase in tumor volume after postoperative chemotherapy was observed using B-ultrasound and MRI; [2] sustained increases in tumor marker levels and clinical symptoms did not relieve; [3] and patients were confirmed as having progressive disease (PD) according to Response Evaluation Criteria in Solid Tumors (RECIST version 1.1). Chemotherapy resistance is considered when one or all of the criteria are met. In order to retrospectively validate the predictive effectiveness of our model, we next used a binary logistic regression approach with chemotherapy resistance as the dependent variable and the risk categories predicted by our model as the covariate [39].
For patients treated with neoadjuvant chemotherapy, the endpoint was time to progression (TTP) because a death event was not observed at the cutoff in this study. TTP was defined as the date from registration to invalid treatment or disease progression [40, 41]. For subgroups only undergoing postoperative chemotherapy, the endpoint of interest was set as invasive disease-free survival (iDFS). iDFS is calculated as the time interval from the date of registration to the first recurrence of breast cancer, the development of contralateral primary breast cancer, or death from any cause [42].
Kaplan–*Meier analysis* and the log-rank test were used to assess survival outcomes in groups treated with neoadjuvant and postoperative chemotherapy as well as postoperative chemotherapy alone. All statistical analyses were implemented with the R software 3.5.0 (https://www.r-project.org/); a P value <0.05 was considered statistically significant.
## Training and test cohorts conducted
The included cohorts were randomly divided into training and test cohorts according to the ratio of 7($$n = 1$$,289):3($$n = 552$$) (Table 2) [43, 44]. The validation set was considered a part of the training cohort to fine-tune the hyperparameters in our models. Each group of information was evenly distributed without bias. Table 2 presents the characteristics of patients. Valuable information in EHRs was first segmented and annotated, including integrated pathological and clinical information from encounter notes and progress notes. Text snippets were further processed using feature extraction methods to extract specific string fields [45]. The extractor achieved $95\%$ accuracy, and each string was then matched against the numeric label “0” or “1”; all matched features of each patient were aggregated together to form a large dataset, which simplifies the learning process. This transformation process involved converting complex multiple input variables into a more manageable format, which greatly improved the classification performance of our model [46]. The standards for automatic extraction are shown in the methods.
**Table 2**
| Characteristic | Training set | Test set |
| --- | --- | --- |
| Number of patients | 1289 | 552 |
| Gender, %Female | 1,282 (99.5) | 549 (99.5) |
| Gender, %Male | 7 (0.5) | 3 (0.5) |
| Age, no. (%) | Age, no. (%) | Age, no. (%) |
| <35 | 12 (0.9) | 8 (1.4) |
| ≥35 | 1,277 (99.1) | 544 (98.6) |
| Menopausal status | Menopausal status | Menopausal status |
| Pre | 298 | 109 |
| Post | 984 | 440 |
| Molecular subtypes | Molecular subtypes | Molecular subtypes |
| Luminal A/luminal B | 609 | 248 |
| HER2+ | 429 | 178 |
| Triple negative | 251 | 126 |
| Histology | Histology | Histology |
| Invasive ductal carcinoma | 1067 | 443 |
| Invasive lobular carcinoma | 53 | 23 |
| Mixed (IDC and ILC) | 48 | 16 |
| DCIS/LCIS | 86 | 58 |
| Other types | 35 | 12 |
| Recurrence risk assessment | Recurrence risk assessment | Recurrence risk assessment |
| Low-risk | 102 | 86 |
| Intermediate-risk | 758 | 283 |
| High-risk | 429 | 183 |
## SVM, XGBoost, and LSTM models predicted the recurrence risk of postoperative breast cancer patients
After model development with the training subset, test samples were uploaded to predict recurrence risk, and this multi-classification task was conducted via a one-vs-the-rest method. Specifically, when one category was correctly predicted by the model, the remaining categories were considered negative [47], thus generating a confusion matrix for each category (Figure 2). We computed the evaluation metrics of each category based on the confusion matrix, such as accuracy, precision, recall, and the area under the receiver-operating characteristic curve (ROC-AUC) (Table 3 and Figure 3). In order to further compare the effectiveness of models, we averaged (macro-average, F1 score) and weighted (micro-average, weighted-average) the evaluation indicators of each category (Table 3) [47, 48]. Subsequently, we draw the ROC curve for each prediction category with the true positive rate (TPR) as the abscissa and the false positive rate (FPR) as the ordinate and explained the achievement of each model using a micro-average ROC curve (Figures 3A, C, E). The AUC values of the micro-average ROC curve corresponding to SVM, XGBoost, and LSTM were 0.92 ± 0.06, 0.97 ± 0.03, and 0.98 ± 0.01 (Figures 3A, C, E). Additionally, the area under the precision–recall curve (AUC-PR) is more suitable for assessing performance metrics on processing imbalanced data compared with the area under the receiver operating characteristic curve (AUC-ROC) (49–51). The SVM generated the smallest micro-average AUC-PR (0.86 ± 0.11), and the LSTM model demonstrated the largest micro-average AUC-PR (0.96 ± 0.02), which indicates that a great number of patients were correctly labeled (Figures 3B, D, F). Overall, the LSTM model accomplished superior performance on the test set, with a micro-averaged AUC-PR that represents an improvement of $10\%$ and $3\%$ compared with SVM and XGBoost. The LSTM model manifested a significantly higher accuracy (0.89), F1 score, macro-F1 score (0.87), and weighted-F1 (0.89) (Table 3).
**Figure 2:** *Normalized confusion matrix for the test set of each model. “Class 0,” “class 1,” and “class 2” correspond to low-risk, intermediate-risk, and high-risk categories. (A) SVM confusion matrix, (B) XGBoost confusion matrix, (C) LSTM confusion matrix.* TABLE_PLACEHOLDER:Table 3 **Figure 3:** *Predictive performance of models on the training set for multiclassification of breast cancer patients. The support vector machines (SVM), extreme gradient boosting (XGBoost), and long short-term memory (LSTM) recurrent neural network models were trained to classify patients with operated breast cancer from the feature label values. (A, C, E) Receiver-operating characteristics (ROC) curve and (B, D, F) Precision-recall (PR) curve for the test set was shown to quantify the performance of models. “Class 0,” “class 1,” and “class 2” correspond to low-risk, intermediate-risk, and high-risk categories.*
## Breast cancer patients at high recurrence risk are more likely to be resistant to chemotherapy after surgery
Chemotherapy resistance is the most crucial reason for recurrence of breast cancer patients after surgery [52]. In order to exclude the influence of other treatment options on the effect of chemotherapy, patients who received chemotherapy alone were included in the experiment. A binary logistic regression analysis was executed to identify the association between model-based predicted recurrence risk and chemotherapy resistance in breast cancer. The inclusion criteria for chemotherapy resistance in this study are described in the methods. A total of 432 patients received postoperative treatment, and 85 ($20\%$) patients underwent chemotherapy alone, which included DNA-damaging drugs such as anthracyclines and platinum and microtubule-targeting drugs like paclitaxel. There were 37 patients classified as high-risk by the LSTM model, 32 of which ($86\%$) were chemotherapy resistant. Among the 46 intermediate-risk patients predicted by the LSTM model, 29 ($63\%$) patients were chemotherapy resistant (Table 4). The results of binary logistic regression showed that the probability of DNA-damaging drug resistance in high-risk patients predicted by the LSTM model was 4.062 times more than in intermediate-risk patients ($P \leq 0.05$; Figure 4A). Meanwhile, the high-risk patients predicted by the LSTM model were more likely to be resistant to microtubule-targeted drugs than the intermediate-risk patients (high-risk: intermediate-risk = 5.667: 1; $P \leq 0.05$; Figure 4A). These results suggest that high-risk patients predicted by our model are more resistant to chemotherapy drugs after surgery and likely to perform more insensitively to paclitaxel. Consistent results were observed in the SVM and XGBoost models, but the P values are not significant (Figures 4B, C). We did not include the low-risk patients because the number of low-risk samples was insufficient to meet the minimum sample size ($$n = 10$$) required for binary logistic regression analysis.
## Our model can predict the neoadjuvant chemotherapy benefits and the survival of patients
Neoadjuvant therapy plays an important role in the clinical practice of systemic treatment for breast cancer patients [53]. Nevertheless, recent research has reported that neoadjuvant chemotherapy is not necessarily beneficial for patient survival. Patients who were refractory to neoadjuvant treatment can result in a higher local recurrence rate after surgery [54, 55]. Among our subgroups treated with neoadjuvant chemotherapy, 52 and 72 patients were predicted to be intermediate and high risk by LSTM, respectively (Table 5). Contrary to our anticipated outcome, the results indicate that the majority of patients who received neoadjuvant chemotherapy did not exhibit a low-risk profile as we had expected. Moreover, 43 and 23 patients treated with the postoperative chemotherapy alone were predicted as intermediate risk and high risk by LSTM. These results indicated that not all breast cancer patients should receive neoadjuvant chemotherapy before surgery. Our predictive model can be utilized to evaluate the benefit of patients receiving neoadjuvant chemotherapy.
**Table 5**
| Number of patients | Recurrence risk assessmentLow-risk (AUC ± SD) | Recurrence risk assessmentIntermediate-risk (AUC ± SD) | Recurrence risk assessmentHigh-risk (AUC ± SD) |
| --- | --- | --- | --- |
| Neoadjuvant and postoperative chemotherapy | 3 (0.93 ± 0.01) | 52 (0.91 ± 0.03) | 72 (0.95 ± 0.03) |
| Postoperative chemotherapy alone | 2 (0.89 ± 0.04) | 43 (0.87 ± 0.02) | 23 (0.89 ± 0.01) |
Data were next manually extracted on time to disease progression (TTP), which was considered a reliable surrogate endpoint in advanced cancer with medical therapy (Lee, Jang, Lee, Cho, Lee, Yu, Kim, Yoon, Kim, Han, Oh, Im and Kim 2016). For patients administered neoadjuvant chemotherapy, the intermediate-risk operated patients predicted by the LSTM model was shown to have a longer TTP than the high-risk ones ($P \leq 0.05$; Figure 5A). We compared invasive disease-free survival (IDFS) in the groups that received only postoperative chemotherapy and found that the high-risk patients acquired poorer IDFS than the intermediate-risk ones ($P \leq 0.05$; Figure 5B). Compared with intermediate-risk or high-risk, the low-risk sample size was insufficient to create reliable estimates. However, low-risk patients actually had better outcomes according to their clinical information. Therefore, our model can accurately predict the prognosis of breast cancer patients before treatment and suggest that clinicians provide the most appropriate treatment regimen for patients, such as whether to administrate patients with neoadjuvant chemotherapy or postoperative chemotherapy.
**Figure 5:** *Estimation of relative survival in classified patients treated with neoadjuvant therapy and postoperative chemotherapy alone by Kaplan–Meier curve analysis. Patients predicted to be classified as “Intermediate-risk” presented favorable TTP (A) and iDFS (B) than that patient identified as “high-risk.” The log-rank test was appropriate to assess performance.*
## Discussion
In this study, the advantages and limitations of our proposed model are as follows: (i) All models can seamlessly classify from labeled data with an accuracy of over $75\%$. ( ii) The linear SVM model generates a good non-linear mapping between input and output variables. It has good robustness and appears to have no effect on the model when non-supported vector samples are added and removed, thus avoiding the problems of leaf node selection in XGBoost and dimension disaster in LSTM. ( iii) The XGBoost model excited more parameters and performed more accurately than SVM. It illustrated a white box compared with ANN so that the model’s effectiveness can be intuitively evaluated. Moreover, the XGBoost model has presorted features based on the parameters before training, which were repeatedly utilized in subsequent iterations, significantly reducing the computation. ( iv) LSTM realized the highest accuracy among all models, attributed to the continuous optimization of gradient descent and backpropagation. ( v) The high recurrence risk predicted by the LSTM model was consistent with the chemotherapy resistance and the worse prognosis of postoperative patients, which corresponded to the actual situation. ( vi) The SVM algorithm is less sensitive to the handling of missing data. Clearly, vacant values were filled with the default value “0” during data preprocessing, which affects the linear separability in the feature space of SVM. Nevertheless, the XGBoost algorithm tries different methods at each node and identifies the best method to handle when missing data are encountered. LSTMs can learn complex correlations between features, including further details in default values. ( vii) The model uses only a single type of input information that converts textual clinical reports into labeled values. Once new variables emerge, we will manually develop and validate a new set of regular expressions for each specific task.
We established machine learning algorithms capable of extracting patient classification information from unstructured clinical notes. Benefitting from the application of technologies and frameworks of machine learning, our models for screening diagnostics with low-cost burden were favorable [56]. However, several unavoidable challenges with machine learning were posed. First, the data annotation and processing were complicated. In order to achieve data collection and annotation with high precision, including the term standardization of biological features, the variability of descriptive words, and the presence of negative phrases, we searched for each key term and encoded it with category encoders through feature engineering and natural language processing. High accuracy was achieved eventually for each feature of information abstracting. Secondly, for high-dimensional scene data exploration (such as medical time-series data), the XGBoost algorithm cannot effectively eliminate noise variables [57]. Therefore, we conducted a grid search to determine the algorithms of optimal dimensionality reduction and added randomness to improve robustness [58]. Additionally, an increasing fraction of the training time in the LSTM model would reduce the number of iterations within the same total training time [59]. We utilized forward calculation and backpropagation to continuously adjust the parameters for extracting the optimal features. Therefore, we provided a reproducible predicted tool to predict the recurrence risk of breast cancer patients after surgery.
To further guide clinical practice, our models maintained their performance in reflecting patient tolerance to chemotherapy drugs. We verified that high-risk patients tend to be more resistant to DNA damage and microtubule inhibition drugs than intermediate-risk patients. This result provides a basis for the clinical treatment application of different drugs to postoperative breast cancer patients. Chemotherapy resistance is not only an important risk factor for cancer recurrence but also a major cause of poor patient outcomes [52]. Meanwhile, our models also validated the prognosis of patients who underwent neoadjuvant chemotherapy and postoperative chemotherapy. Since the linkages between EHR data and death registries were rare, we used TTP or IDFS as surrogate endpoints to assess differences in survival outcomes of predicted categories. Our approach highlighted the importance of estimating the recurrence risk after neoadjuvant chemotherapy, indicating whether patients routinely receive preoperative chemotherapy is worth thought-provoking [60, 61]. Although patients classified as low-risk were predicted in our model, the recurrence was not statistically significant compared with the other two groups because of the rare number of samples.
Previous studies have applied natural language processing to abstract biological factors from medical records to predict breast cancer staging based on the American Joint Committee on Cancer (AJCC) staging manual [62]. In 2020, researchers also implemented artificial neural networks to predict breast cancer prognosis by selecting crucial survival factors, including tumor size, tumor staging, lymph node metastasis, and other related variables [63]. Moreover, deep learning has shown promise in predicting breast cancer risk rates by extracting factors such as age, race, and menstrual history [64]. In contrast, our approach significantly solved the bottleneck of extracting outcomes from a great number of clinical texts and achieved effective feature extraction in different scenes. Additionally, those included studies were predominantly conducted in the United States or Europe, but the data for Asian breast cancer patients remained unknown. Breast cancer incidence is strongly correlated with variations in geographic distribution [65, 66]. Because of differences in people’s diets and lifestyles, breast cancer is highly prevalent in the alpine region [67], such as the northeast of China. An accurate assessment of patients’ recurrence risk before tailored individual treatment plans can provide valuable guidance on improving patient outcomes. Our studies contribute to the development of screening strategies for breast cancer in the Asian population.
In conclusion, we developed AI-based models that integrate histopathological features of breast cancer and clinical information from preprocessed clinical notes to predict the recurrence risk of postoperative breast cancer patients. The performance and generalizability of our model have emphasized the potential application in the estimation of recurrence risk in breast cancer patients.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics Committee of Harbin 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
WY offered main direction and significant guidance of this manuscript. LZ, LL, DC and HL drafted the manuscript and illustrated the figures for the manuscript. LL provided the clinical data. YX and HB helped with the data analyzed. 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.2023.1117420/full#supplementary-material
## References
1. Huang J, Chan PS, Lok V, Chen X, Ding H, Jin Y. **Global incidence and mortality of breast cancer: A trend analysis**. *Aging (Albany NY)* (2021) **13**. DOI: 10.18632/aging.202502
2. Cao W, Chen HD, Yu YW, Li N, Chen WQ. **Changing profiles of cancer burden worldwide and in China: A secondary analysis of the global cancer statistics 2020**. *Chin Med J (Engl)* (2021) **134**. DOI: 10.1097/CM9.0000000000001474
3. Chlebowski RT. **Improving breast cancer risk assessment versus implementing breast cancer prevention**. *J Clin Oncol* (2017) **35**. DOI: 10.1200/JCO.2016.70.9386
4. Shimizu H, Nakayama KI. **Artificial intelligence in oncology**. *Cancer Sci* (2020) **111**. DOI: 10.1111/cas.14377
5. Sultan AS, Elgharib MA, Tavares T, Jessri M, Basile JR. **The use of artificial intelligence, machine learning and deep learning in oncologic histopathology**. *J Oral Pathol Med* (2020) **49**. DOI: 10.1111/jop.13042
6. Tran KA, Kondrashova O, Bradley A, Williams ED, Pearson JV, Waddell N. **Deep learning in cancer diagnosis, prognosis and treatment selection**. *Genome Med* (2021) **13** 152. DOI: 10.1186/s13073-021-00968-x
7. Jiang X, Xu C. **Deep learning and machine learning with grid search to predict later occurrence of breast cancer metastasis using clinical data**. *J Clin Med* (2022) **29** 11. DOI: 10.3390/jcm11195772
8. Evans RS. **Electronic health records: Then, now, and in the future**. *Yearb Med Inform* (2016) **Suppl 1**. DOI: 10.15265/IYS-2016-s006
9. Koleck TA, Dreisbach C, Bourne PE, Bakken S. **Natural language processing of symptoms documented in free-text narratives of electronic health records: A systematic review**. *J Am Med Inform Assoc* (2019) **26**. DOI: 10.1093/jamia/ocy173
10. Tripathy RK, Mahanta S, Paul S. **Artificial intelligence-based classification of breast cancer using cellular images**. *RSC Adv* (2014) **4**. DOI: 10.1039/C3RA47489E
11. Chen T, Guestrin C. *XGBoost: A Scalable Tree Boosting System* (2016)
12. Burke HB, Goodman PH, Rosen DB, Henson DE, Weinstein JN, Harrell FE. **Artificial neural networks improve the accuracy of cancer survival prediction**. *Cancer* (1997) **79**. DOI: 10.1002/(SICI)1097-0142(19970215)79:4<857::AID-CNCR24>3.0.CO;2-Y
13. Schmidhuber J. **Deep learning in neural networks: An overview**. *Neural Netw* (2015) **61** 85-117. DOI: 10.1016/j.neunet.2014.09.003
14. Tsai YC, Lai SH, Ho CJ, Wu FM, Henrickson L, Wei CC. **High accuracy respiration and heart rate detection based on artificial neural network regression**. *. Annu Int Conf IEEE Eng Med Biol Soc* (2020) **2020**. DOI: 10.1109/EMBC44109.2020.9175161
15. Zdolsek G, Chen Y, Bogl HP, Wang C, Woisetschlager M, Schilcher J. **Deep neural networks with promising diagnostic accuracy for the classification of atypical femoral fractures**. *Acta Orthop* (2021) **92** 394-400. DOI: 10.1080/17453674.2021.1891512
16. Dai G, Zhang X, Liu W, Li Z, Wang G, Liu Y. **Analysis of EPID transmission fluence maps using machine learning models and CNN for identifying position errors in the treatment of GO patients**. *Front Oncol* (2021) **11**. DOI: 10.3389/fonc.2021.721591
17. Jiang YQ, Cao SE, Cao S, Chen JN, Wang GY, Shi WQ. **Preoperative identification of microvascular invasion in hepatocellular carcinoma by XGBoost and deep learning**. *J Cancer Res Clin Oncol* (2021) **147**. DOI: 10.1007/s00432-020-03366-9
18. Kumar A, Vatsa A. **Untangling classification methods for melanoma skin cancer**. *Front Big Data* (2022) **5**. DOI: 10.3389/fdata.2022.848614
19. Karhade J, Ghosh SK, Gajbhiye P, Tripathy RK, Acharya UR. **Multichannel multiscale two-stage convolutional neural network for the detection and localization of myocardial infarction using vectorcardiogram signal**. *Applied Sci* (2021) **11** 7920. DOI: 10.3390/app11177920
20. Malley B, Ramazzotti D, Wu JT. *Data pre-processing. in: Secondary analysis of electronic health records* (2016)
21. Fu JT, Sholle E, Krichevsky S, Scandura J, Campion TR. **Extracting and classifying diagnosis dates from clinical notes: A case study**. *J BioMed Inform* (2020) **110** 103569. DOI: 10.1016/j.jbi.2020.103569
22. Veena G, Hemanth R, Hareesh J. *Relation extraction in clinical text using NLP based regular expressions* (2019)
23. Mehrabi S, Schmidt CM, Waters JA, Beesley C, Krishnan A, Kesterson J. **An efficient pancreatic cyst identification methodology using natural language processing**. *Stud Health Technol Inform* (2013) **192**. DOI: 10.3233/978-1-61499-289-9-822
24. Kitchenham B, Pfleeger SL, McColl B, Eagan SJ. **An empirical study of maintenance and development estimation accuracy**. *J Syst Softw* (2002) **64** 57-77. DOI: 10.1016/S0164-1212(02)00021-3
25. **[Guidelines for clinical diagnosis and treatment of breast cancer in China, (2021 edition)]. zhonghua**. *Zhong Liu Za Zhi* (2021) **31** 954-1040. DOI: 10.3760/cma.j.cn112152-20200817-00747
26. Kourou K, Exarchos TP, Exarchos KP, Karamouzis MV, Fotiadis DI. **Machine learning applications in cancer prognosis and prediction**. *Comput Struct Biotechnol J* (2015) **13** 8-17. DOI: 10.1016/j.csbj.2014.11.005
27. Jakkula V. **Tutorial on support vector machine (svm)**. *Comput Sci* (2006) **37** 3
28. Varpa K, Joutsijoki H, Iltanen K, Juhola M. **Applying one-vs-one and one-vs-all classifiers in k-nearest neighbour method and support vector machines to an otoneurological multi-class problem**. *User centred networked health care* (2011)
29. Nguyen G, Dlugolinsky S, Bobák M, Tran V, López García Á, Heredia I. **Machine learning and deep learning frameworks and libraries for large-scale data mining: a survey**. *Artif Intell Rev* (2019) **52** 77-124. DOI: 10.1007/s10462-018-09679-z
30. Ramraj S, Uzir N, Sunil R, Banerjee S. **Experimenting XGBoost algorithm for prediction and classification of different datasets**. *Int J Control Theory and Applications* (2016) **9**
31. Brownlee J. *XGBoost with python: Gradient boosted trees with XGBoost and scikit-learn* (2016)
32. Yao K, Cohn T, Vylomova K, Duh K, Dyer C. **Depth-gated LSTM**. (2015). DOI: 10.48550/arXiv.1508.03790
33. Yu Y, Si X, Hu C, Zhang J. **A review of recurrent neural networks: LSTM cells and network architectures**. *Neural Comput* (2019) **31**. DOI: 10.1162/neco_a_01199
34. Liu X, Zeng Z, Wunsch DC. **Memristor-based LSTM network with**. *Neural Netw* (2020) **131**. DOI: 10.1016/j.neunet.2020.07.035
35. Hou L, Zhu J, Kwok J, Gao F, Qin T, Liu T-y. **Normalization helps training of quantized lstm**. (2019) **32**. DOI: 10.5555/3454287.3454947
36. Liu Z, Bondell HD. **Binormal precision–recall curves for optimal classification of imbalanced data**. *Stat Biosci* (2019) **11**. DOI: 10.1007/s12561-019-09231-9
37. Jiang YZ, Liu Y, Xiao Y, Hu X, Jiang L, Zuo WJ. **Molecular subtyping and genomic profiling expand precision medicine in refractory metastatic triple-negative breast cancer: the FUTURE trial**. *Cell Res* (2021) **31**. DOI: 10.1038/s41422-020-0375-9
38. Chen Y, Feng X, Yuan Y, Jiang J, Zhang P, Zhang B. **Identification of a novel mechanism for reversal of doxorubicin-induced chemotherapy resistance by TXNIP in triple-negative breast cancer**. *Cell Death Dis* (2022) **13** 338. DOI: 10.1038/s41419-022-04783-z
39. Huang FL. **Alternatives to logistic regression models when analyzing cluster randomized trials with binary outcomes**. *Prev Sci* (2021) **22**. DOI: 10.1007/s11121-021-01228-5
40. Lee DW, Jang MJ, Lee KH, Cho EJ, Lee JH, Yu SJ. **TTP as a surrogate endpoint in advanced hepatocellular carcinoma treated with molecular targeted therapy: meta-analysis of randomised controlled trials**. *Br J Cancer* (2016) **115**. DOI: 10.1038/bjc.2016.322
41. Saito T, Murotani K. **Treatment of death events in the analysis of time to progression**. *Ther Innov Regul Sci* (2022) **56** 1-3. DOI: 10.1007/s43441-021-00343-3
42. Sparano JA, Gray RJ, Makower DF, Pritchard KI, Albain KS, Hayes DF. **Adjuvant chemotherapy guided by a 21-gene expression assay in breast cancer**. *N Engl J Med* (2018) **379**. DOI: 10.1056/NEJMoa1804710
43. Gholamy A, Kreinovich V, Kosheleva O. *Why 70/30 or 80/20 relation between training and testing sets: a pedagogical explanation* (2018)
44. Wagner J, Rapsomaniki MA, Chevrier S, Anzeneder T, Langwieder C, Dykgers A. **A single-cell atlas of the tumor and immune ecosystem of human breast cancer**. *Cell* (2019) **177** 1330-1345 e1318. DOI: 10.1016/j.cell.2019.03.005
45. Khalid S, Khalil T, Nasreen S. *A survey of feature selection and feature extraction techniques in machine learning* (2014)
46. Guyon I, Elisseeff A. **An introduction to feature extraction**. *Feature extraction* (2006) 1-25
47. Murphy KP. *Machine learning: A probabilistic perspective* (2012)
48. Pillai I, Fumera G, Roli F. *F-measure optimisation in multi-label classifiers* (2012)
49. Subbe CP, Slater A, Menon D, Gemmell L. **Validation of physiological scoring systems in the accident and emergency department**. *Emerg Med J* (2006) **23**. DOI: 10.1136/emj.2006.035816
50. Churpek MM, Yuen TC, Winslow C, Meltzer DO, Kattan MW, Edelson DP. **Multicenter comparison of machine learning methods and conventional regression for predicting clinical deterioration on the wards**. *Crit Care Med* (2016) **44**. DOI: 10.1097/CCM.0000000000001571
51. Ferrari D, Milic J, Tonelli R, Ghinelli F, Meschiari M, Volpi S. **Machine learning in predicting respiratory failure in patients with COVID-19 pneumonia-challenges, strengths, and opportunities in a global health emergency**. *PLos One* (2020) **15**. DOI: 10.1371/journal.pone.0239172
52. Prihantono M. **Breast cancer resistance to chemotherapy: When should we suspect it and how can we prevent it**. *Ann Med Surg (Lond)* (2021) **70** 102793. DOI: 10.1016/j.amsu.2021.102793
53. Montemurro F, Nuzzolese I, Ponzone R. **Neoadjuvant or adjuvant chemotherapy in early breast cancer**. *Expert Opin Pharmacother* (2020) **21**. DOI: 10.1080/14656566.2020.1746273
54. Vaidya JS, Massarut S, Vaidya HJ, Alexander EC, Richards T, Caris JA. **Rethinking neoadjuvant chemotherapy for breast cancer**. *BMJ* (2018) **360** j5913. DOI: 10.1136/bmj.j5913
55. Asaoka M, Gandhi S, Ishikawa T, Takabe K. **Neoadjuvant chemotherapy for breast cancer: Past, present, and future**. *Breast Cancer (Auckl)* (2020) **14** 1178223420980377. DOI: 10.1177/1178223420980377
56. Aliabadi A, Sheikhtaheri A, Ansari H. **Electronic health record-based disease surveillance systems: A systematic literature review on challenges and solutions**. *J Am Med Inform Assoc* (2020) **27**. DOI: 10.1093/jamia/ocaa186
57. Euh S, Lee H, Kim D, Hwang D. **Comparative analysis of low-dimensional features and tree-based ensembles for malware detection systems**. *Institute of Electrical and Electronics Engineers (IEEE)* (2020) **8**. DOI: 10.1109/ACCESS.2020.2986014
58. Paleczek A, Grochala D, Rydosz A. **Artificial breath classification using XGBoost algorithm for diabetes detection**. *Sensors (Basel)* (2021) **18** 21. DOI: 10.3390/s21124187
59. You Y, Hseu J, Ying C, Demmel J, Keutzer K, Hsieh C-J. *Large-Batch training for LSTM and beyond* (2019). DOI: 10.48550/arXiv.1901.08256
60. Duchesneau ED, An SJ, Strassle PD, Reeder-Hayes KE, Gallagher KK, Ollila DW. **Sociodemographic and clinical predictors of neoadjuvant chemotherapy in cT1-T2/N0 HER2-amplified breast cancer**. *Ann Surg Oncol* (2022) **29**. DOI: 10.1245/s10434-021-11260-y
61. Fujita N, Fujita K, Kim SJ, Iguchi C, Nomura T, Aono T. **Response-guided omission of anthracycline in patients with HER2-positive early breast cancer treated with neoadjuvant taxane and trastuzumab: 5-year follow-up of prognostic study using propensity score matching**. *Oncology* (2022) **100**. DOI: 10.1159/000522384
62. Deshmukh PR, Phalnikar R. **Information extraction for prognostic stage prediction from breast cancer medical records using NLP and ML**. *Med Biol Eng Comput* (2021) **59**. DOI: 10.1007/s11517-021-02399-7
63. Boeri C, Chiappa C, Galli F, De Berardinis V, Bardelli L, Carcano G. **Machine learning techniques in breast cancer prognosis prediction: A primary evaluation**. *Cancer Med* (2020) **9**. DOI: 10.1002/cam4.2811
64. Stark GF, Hart GR, Nartowt BJ, Deng J. **Predicting breast cancer risk using personal health data and machine learning models**. *PLos One* (2019) **14**. DOI: 10.1371/journal.pone.0226765
65. 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
66. Su H, Li X, Lv Y, Qiu X. **Breast cancer epidemiology and survival analysis of shenyang in northeast China: A population-based study from 2008 to 2017**. *Breast J* (2022) **282** 6168832. DOI: 10.1155/2022/6168832
67. Liu B, Lao X, Feng Y, Liu J, Jiao M, Zhao M. **Cancer prevalence among the rural poverty-stricken population in northeast China**. *Cancer Manag Res* (2019) **11**. DOI: 10.2147/CMAR.S205867
|
---
title: Association between urinary albumin creatinine ratio and cardiovascular disease
authors:
- Yoo Jin Kim
- Sang Won Hwang
- Taesic Lee
- Jun Young Lee
- Young Uh
journal: PLOS ONE
year: 2023
pmcid: PMC10030008
doi: 10.1371/journal.pone.0283083
license: CC BY 4.0
---
# Association between urinary albumin creatinine ratio and cardiovascular disease
## Abstract
### Introduction
The association between microalbuminuria and cardiovascular disease (CVD) is accumulating in various patient populations. However, when stratified by sex, the relationship between microalbuminuria and CVD remains unclear.
### Method
We obtained data from the 2011–2014 and 2019–2020 Korea National Health and Nutrition Examination Survey (KNHANES). Microalbuminuria was measured based on spot urine albumin-creatinine ratio (UACR). The Framingham risk score (FRS) model was implemented to evaluate the CVD risk. Linear and logistic regression models were used to identify the associations of microalbuminuria status with cardiometabolic predictors and CVD status determined by the FRS score.
### Results
Among 19,340 representative Korean participants, the (UACR) in Korean women and men with history of CVD was higher than in those without history of CVD. Among patients without history of CVD, multivariate regression analysis showed that a high UACR was related to older age, lower high-density lipoprotein cholesterol level, higher total cholesterol level, higher systolic blood pressure, higher prevalence of current smoking, higher prevalence of diabetes, and higher anti-hypertensive medication use in both women and men. The UACR showed a positive linear correlation with the Framingham risk score in both women and men.
### Conclusion
The presence of microalbuminuria was significantly associated with the cardiometabolic risk factors and the increased risk of CVD evaluated by FRS model in both women and men in a nationally representative sample of Korea.
## Introduction
Cumulative evidence indicates that albuminuria is associated with increased risk of cardiovascular diseases (CVDs) [1, 2]. The Heart Outcomes Prevention Evaluation (HOPE) study concluded that any degree of albuminuria (e.g., microalbuminuria) is a risk factor for cardiovascular (CV) events; particularly, a 0.4-mg/mmol increase in the albumin-creatinine ratio (ACR) was related to a $5.9\%$ increased risk for CV events [1]. The Prevention of Renal and Vascular End Stage Disease (PREVEND) study, conducted among inhabitants of the city of Groningen (the Netherlands), reported that a 2-fold increase in albuminuria was associated with a 1.29 and 1.12-times increased risk for CV mortality and non-CV mortality, respectively [2]. The Prevention of Events with an ACE inhibitor (PEACE) trial showed that albuminuria, even at low levels within the normal range, is an independent predictor of CV mortality (hazard ratio per log ACR:1.74) [3]. This evidence comes from cohort studies not only from individuals at high risk of CVDs (patients with diabetes, hypertension, older adults, or stable coronary artery disease), but also from the general population [2–5]. Several studies have shown this association, even in patients with microalbuminuria [1, 6]. However, most studies pertain to sex-adjusted CV risk and no study has revealed sex-specific CV risk.
The Korea National Health and Nutrition Examination Survey (KNHANES) is a nationwide cohort in Korea that has collected urine albumin levels of participants since 2011. Initial studies that used the urinary albumin data from KNHAENS focused on identifying the risk factors related to albuminuria and/or microalbuminuria [7, 8]. Eventually, Korean nephrologists analyzed the 2011–2013 KNHANES, and reported clinical predictors related to albuminuria and/or chronic kidney disease (CKD) [9]. The accumulation of urine albumin data in KNHANES has diversified research topics. Body composition-related health problems have been studied as candidate association factors with albuminuria. Several studies using the KNHANES have reported the independent association between sarcopenia (also referred to as low skeletal muscle mass status) and albuminuria [10, 11]. A study analyzing urine albumin and body composition data in the KNHANES demonstrated that urine albumin level is related to bone mineral density of total hip in postmenopausal women [12].
Meanwhile, only a few studies analyzing KNHANES data have pinpointed the association between albuminuria and CVD [13]. Ahn et al. [ 13] obtained urine albumin data from the 2011–2013 KNHANES, and demonstrated that albuminuria could reflect CVD risk as measured by the Framingham risk score (FRS) [14]. However, this study only reported findings among postmenopausal women without diabetes [14]. Taken together, the current study aimed to evaluate the relationship between urine albumin-creatinine ratio (UACR) and CVD according to sex.
## Ethics statement
All the participants enrolled in the KNHANES signed an informed consent form. The KNHANES data and their analyses in the present study were performed in compliance with the Declaration of Helsinki. The present study protocol was approved by the Institutional Review Board of Wonju Severance Christian Hospital (IRB No. CR321375).
## Study population
This study analyzed data obtained from the 2011–2014 and 2019–2020 KNHANES. The KNHANES is conducted annually by the Division of Chronic Disease Surveillance of the Korea Centers for Disease Control and Prevention in the Ministry of Health and Welfare to assess and monitor the general and medical health and nutrition status in South Korea [15, 16]. The KNHANES includes three main components, a health interview, health examination, and nutrition survey. The KNHANES implements a complex, multi-stage probability sample design to obtain nationally representative data [15, 16]. The KNHANES is publicly available data (https://knhanes.kdca.go.kr/knhanes/sub03/sub03_02_05.do). Among 47,613 participants in the KNHANES, we excluded those aged under 40 years ($$n = 24$$,982), and those with missing information on demographics, lifestyle, medical, anthropometric, and laboratory variables ($$n = 3$$,291). After the exclusion, a total of 19,340 participants were analyzed.
## Measurement of urine albumin and creatinine
The gold standard for measuring urine albumin excretion is 24-h urine collection. However, the National Kidney Foundation recommends the use of spot urine albumin-creatinine ratio to detect microalbuminuria, which is more convenient and accurate than 24-h urine collection. However, the cutoff value to diagnose microalbuminuria is different across different races and sex [17]. Therefore, we estimated sex-specific UACR and CV risk using data from the KNHANES.
## Covariates
Old age, diet (low intake of vegetables, fruits, and whole grain; and high intake of processed red meats, refined carbohydrates, and sweetened beverages), low or irregular physical activity, diabetes, and tobacco use are widely known risk factors for atherosclerotic CVD (ASCVD) or its related mortality [18]. Moreover, other factors, such as serum lipid profiles, air pollution, and genetic factors, are related to ASCVD [19–21].
From representative models [22, 23], we determined seven predictors: age, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), systolic blood pressure (SBP), antihypertensive medication (AHM), current smoking (CS), and diabetes as covariates.
## Measurement of CVD risk score based on Framingham risk score
CVD risk score was measured using the FRS model [22]. The FRS model was established based on the Cox proportional hazards model which is widely used in the medical field [24]. The Cox model includes a linear unit and a non-linear unit (termed to survival function [22, 24]). The FRS includes an interaction term AHM×SBP, for which, the coefficient is 0.06106 (2.82263–2.76157) in women and 0.06578 (1.99881–1.93303) in men. The equation for calculating the FRS is as follows.
Linear predictor (LP)women = ln(age)×2.32888 + ln(TC) ×1.20904 + ln(HDL-C) ×(–0.70833) + ln(SBP)×2.76157 + ln(SBP)×AHM (0:no; 1: yes)×(2.82263–2.76157) + current smoking status × 0.52873 + diabetes×0.69154–26.1931 FRSwomen [22] = 1–0.95012exp(LPwomen) FRSmen [22] = ln(age)×3.06117+ ln(TC) ×1.12370 + ln(HDL-C) ×(–0.93263) + ln(SBP)×1.93303 + ln(SBP)×AHM (0:no; 1: yes)×(1.99881–1.93303) + current smoking status × 0.65451 + diabetes×0.57367–23.9802 FRSmen [22] = 1–0.88936exp(Lpmen)
## Statistics
R language (version 4.0.1) [25] was implemented to reconstruct and preprocess the dataset and perform statistical analysis. Continuous variables, such as demographics (e.g., age) and laboratory values, were analyzed using ANOVAR. For categorical variables, the chi-square test was utilized. To evaluate the linear trends of categorical or continuous variables based on UACR tertile, we determined the median UACR levels of each tertile group as continuous variables when using the Chi-square test and one-way ANOVA.
To estimate the total population that the data would represent, we employed the sampling weights determined by the data constructors. After adopting the weight values, we analyzed the association between UACR and cardiometabolic risk factors included in the FRS equation. Multivariate linear or logistic regressions were performed using the following equation: cardiometabolic risk factors (dependent variable) ~ UACR (independent variable) + covariates. A p-value of < 0.05 was determined to be statistically significant.
## Results
The distribution of urine albumin (S1 Fig) among Korean women did not show abnormal distribution; instead, the distribution skewed left, indicating the existence of several outliers of urine albumin levels. After inclusion of cases with urine albumin less than 10 mg/dL, the patterns still did not show normal distribution; instead, they showed gamma or log-normal distributions. In contrast, urine creatinine levels in Korean women slightly skewed to the left, but generally followed a normal distribution. The distribution of UACR values in Korean men was similar to that of their urine albumin levels. These findings were consistent among Korean men (S2 Fig).
When comparing UACR levels according to the CVD status obtained from questionnaires, both mean and median values of UACR among Korean women were higher in the CVD group than in the non-CVD group (median [non-CVD/CVD groups] = $\frac{5.99}{8}$ mg/g; mean = $\frac{20.8}{60.8}$ mg/g, Fig 1A). However, in the non-CVD group, many outliers had extremely high UACRs because most participants were in the non-CVD group (Fig 1A). These results were also exhibited among Korean men because the prevalence of CVD was low in both Korean men and women (Fig 1A). Among the self-reported responses, we found that the UACR was high in CVD status; therefore, as the next step, we evaluated the association between UACR, and CVD status measured by FRS score after excluding CVD patients.
**Fig 1:** *UACRs according to CVD status in Korean women (A) and men (B). Left side boxplots (grey and brown colored boxes) indicate median-based summary statistics; specifically, the middle, upper, and lower lines describe median, 75, and 25 percentile values, respectively. Right side boxplots indicate mean-based summary statistics, in which the middle, upper, and lower lines illustrate mean, one standard deviation values, respectively.*
*Different* general characteristics were shown among Korean women according to increasing UACR: older age, higher SBP, higher TC, lower HDL-C, greater AHM use, higher prevalence in diabetes, and higher levels of FRS. For Korean men, most risk factors, except for the serum levels of TC, exhibited similar bio-signatures compared to those of Korean women (Table 1). However, as the UACR increased from tertile 1 to tertile 3, the number of current smokers did not show any significant difference between men and women. Note that all participants analyzed in Table 1 were not diagnosed with CVD.
**Table 1**
| Unnamed: 0 | Korean women | Korean women.1 | Korean women.2 | Korean women.3 |
| --- | --- | --- | --- | --- |
| Variable | T1 | T2 | T3 | p-value |
| Unweighted participants, n | 2325 | 2324 | 2332 | |
| Age, years | 55.3 ± 0.22 | 57.6 ± 0.23 | 62.5 ± 0.23 | <0.001 |
| Systolic BP, mmHg | 116.1 ± 0.32 | 119.6 ± 0.35 | 128.8 ± 0.35 | <0.001 |
| Antihypertensive medication, n | 424 (18.2) | 596 (25.6) | 982 (42.1) | <0.001 |
| Diabetes, n | 120 (5.2) | 180 (7.7) | 374 (16) | <0.001 |
| Current smoker, n | 79 (3.4) | 84 (3.6) | 83 (3.6) | 0.917 |
| Total cholesterol, mg/dL | 197.7 ± 0.74 | 197.1 ± 0.75 | 199.3 ± 0.75 | <0.001 |
| HDL-cholesterol, mg/dL | 52.9 ± 0.25 | 52.5 ± 0.25 | 50.3 ± 0.25 | <0.001 |
| Urine albumin, mg/dL | 0.2 ± 0 | 0.7 ± 0.01 | 4.7 ± 0.01 | <0.001 |
| Urine creatinine, mg/dL | 101.6 ± 1.15 | 129.2 ± 1.45 | 115.5 ± 1.44 | <0.001 |
| UACR, mg/g | 1.5 ± 0.02 | 5.1 ± 0.03 | 52.4 ± 0.03 | <0.001 |
| FRS | 0.065 ± 0.001 | 0.079 ± 0.001 | 0.126 ± 0.001 | <0.001 |
| | Korean men | Korean men | Korean men | Korean men |
| Unweighted participants, n | 1669 | 1669 | 1675 | |
| Age, years | 56.6 ± 0.27 | 57.9 ± 0.27 | 62.5 ± 0.27 | <0.001 |
| Systolic BP, mmHg | 119.3 ± 0.36 | 121.8 ± 0.38 | 128.8 ± 0.38 | <0.001 |
| Antihypertensive medication, n | 323 (19.4) | 403 (24.1) | 675 (40.3) | <0.001 |
| Diabetes, n | 98 (5.9) | 135 (8.1) | 414 (24.7) | <0.001 |
| Current smoker, n | 556 (33.3) | 617 (37) | 594 (35.5) | 0.085 |
| Total cholesterol, mg/dL | 189.5 ± 0.84 | 190.1 ± 0.81 | 186.4 ± 0.81 | <0.001 |
| HDL-cholesterol, mg/dL | 47.4 ± 0.28 | 47.3 ± 0.28 | 46.3 ± 0.28 | <0.001 |
| Urine albumin, mg/dL | 0.2 ± 0 | 0.7 ± 0.01 | 9.9 ± 0.01 | <0.001 |
| Urine creatinine, mg/dL | 142 ± 1.66 | 176.2 ± 1.95 | 155.3 ± 1.94 | 0.999 |
| UACR, mg/g | 1.1 ± 0.02 | 4 ± 0.03 | 80.9 ± 0.03 | <0.001 |
| FRS | 0.161 ± 0.003 | 0.185 ± 0.003 | 0.272 ± 0.003 | <0.001 |
We analyzed the association of the UACR with seven risk factors used in the calculation of FRS based on univariate analysis. In both Korean women and men, the following robust and significant signatures were associated with increase in the UACR: older age, lower TC, lower HDL-C, higher SBP, higher ratio of hypertensive medication, and higher prevalence of diabetes (Fig 2).
**Fig 2:** *Relationship between UACR and cardiometabolic risk factors in Korean women (A) and men (B). Beta values were measured by linear regression after setting continuous variables, including age, TC, HDL-C, and SBP as dependent variables and UACR subgroups as independent variables. In case of features exhibiting binomial distribution, such as AHM use, smoking, and diabetes, the ratio of presence of disease or status was set as the dependent variable in the linear regression for the calculation of the Beta value. Abbreviations: UACR, urinary albumin-creatinine ratio; Beta, beta-coefficient; AHM, anti-hypertensive medication; HDL-C. high-density lipoprotein-cholesterol; SBP, systolic blood pressure; TC, total cholesterol.*
The FRS included seven cardiometabolic predictors, including age, TC, HDL-C SBP, AHM, smoking, and diabetes. Among six features, four were continuous variables set as dependent variables in each model using multivariate linear regression. The UACR was the independent variable, and the other six variables were covariates (Fig 3). In case of dichotomous variables, including AHM, smoking, and diabetes, logistic regression was used to evaluate their association with UACR after adjusting for the other remaining six predictors (Fig 3). As a result, in both Korean women and men, the high levels of UACR were related to older age, higher TC, lower HDL-C, higher SBP, greater smoking levels, greater AHM use, and diabetes were related to FRS (Fig 3).
**Fig 3:** *Relationship between albuminuria and cardiometabolic risk factors in Korean women (A) and men (B). Top four graphs (i.e., age, TC, HDL-C, SBP) were obtained by multivariate linear regression after setting the four predictors arranged separately as dependent variables. UACR was determined as the independent variable, and other remnant six predictors as covariates. The lower three graphs (i.e., AHM, smoking, diabetes) were obtained by multivariate logistic regression set to the same conditions as the multivariate linear regression. All x-axes indicate beta-coefficients obtained from the multivariate linear or logistic regressions. UACR levels were log-transformed for the associational analyses. Abbreviations: TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; SBP, systolic blood pressure; HTN Med, hypertension medication; DM, diabetes mellitus.*
We compared the relationship between UACR and the combined effect of seven cardiometabolic predictors, in the form of an equation, referred to as the FRS (Fig 4). In both women and men, as UACR increased, the FRS exhibited monotonic elevated patterns. Moreover, all the increasing characteristics showed exponential distributions, indicating that the albuminuria groups (Q16 –Q20 in Fig 4) were directly proportional to extremely high risk of CVD.
**Fig 4:** *Relationship between albuminuria and FRS in Korean women (A) and men (B). Urine albumin-to-creatine ratio was categorized into 20 groups (x-axes) based on ascending order. FRS was calculated based on the equation provided by a study [22]. Abbreviation: FRS, Framingham risk score.*
## Discussion
Our study showed that both Korean women and men with CVD history had higher UACR level than those with no previous history of CVD. In particular, for those without CVD history, multivariate adjusted analysis showed that higher UACR was associated with CV risk factors such as older age, higher TC, lower HDL, higher SBP, higher proportion of HTN, higher proportion of current smoking, and higher proportion of diabetes in both Korean women and men. In both women and men without CVD history, UACR showed positive correlation with FRS. In correlation analyses between UACR and individual cardiometabolic risk factors, several non-linear correlations were shown: age, TC, HDL-C, SBP, smoking status in Korean women and men (Fig 2). Moreover, gender-specific associational findings could be observed (Fig 2). For example, in Korean men, UACR levels were negatively related to HDL-C levels (beta-coefficients: -0.1834; p-value < 0.001), besides, in Korean women, this trend slightly were diluted (beta-coefficient: -0.0738; p-value: 0.0011).
The HOPE and PEACE studies showed that albuminuria is associated with a higher risk of CVD incidence and mortality among high-risk patients with CVD [1, 3]. Moreover, albuminuria was associated with the risk of CVD among healthy individuals in the general population, without history of CVD [5, 26, 27]. The Multi-Ethnic Study of Atherosclerosis (MESA) study showed that UACR was associated with an $11\%$ increase in the risk of CVD events [5]. The Framingham cohort study reported that without CVD, low level of UACR predicted the development of CVD among normotensive and nondiabetic individuals [28]. The Strong Heart Study also showed that a lower UACR than the normal value predicted CVD [27]. The Prevention of Renal and Vascular End Stage Disease Intervention Trial (PREVEND IT) study showed that the FRS is correlated with microalbuminuria [29].
The risk of CVD differs according to sex. Our results showed that the prevalence of smoking was more common among men than women, and the prevalence of obesity was higher among women than men. Moreover, women tended to have better levels of cholesterol and blood pressure. These differences have been attributed to the differences in lifestyle, health awareness, and sex hormones (such as estrogen). Recently, this difference has been decreasing; however, the difference in cholesterol and body mass index among the different sexes remains significant [30]. Because there are still differences in the control of high blood pressure, diabetes, and hyperlipidemia among different sexes, it is necessary to stratify and analyze microalbuminuria as a risk factor for CVD by sex. Our results showed that microalbuminuria could be considered an important predictor of CVD regardless of sex.
The precise pathophysiological mechanism of microalbuminuria as a CV risk factor remains unknown. The association between microalbuminuria and CVD is explained by endothelial dysfunction or chronic low-grade inflammation. Endothelial dysfunction could increase glomerular pressure and glomerular barrier permeability which increases endothelial permeability. Increased microalbuminuria could be a marker of generalized endothelial dysfunction which could predispose to an atherogenic lipoprotein accumulation in the subendothelial cell space [31, 32]. Microalbuminuria is also associated with chronic low-grade inflammation which could be both cause and consequence of endothelial dysfunction. Furthermore, endothelial dysfunction and low-grade inflammation can not only lead to atherothrombosis but can also be independently associated as a risk for CVD [32, 33].
Diabetic patients have an increased risk of microalbuminuria and 20–$30\%$ of patients with diagnosed diabetes have been found to have microalbuminuria [34]. Those patients have abnormal insulin resistance and increased serum glucose level that makes serum insulin level increase. Insulin stimulates store-operated Ca entry via Orai-1 channel in podocytes that makes actin remodeling and transepithelial albumin leakage resulting in microalbuminuria [35]. Recently over 8 years follow up study from Korean Genome and Epidemiology Study (KOGES) showed that microalbuminuria could be used as an early marker of progression to diabetes even in the non-prediabetic population [36]. Through microalbuminuria we could predict abnormal insulin resistance and diabetes, which were major risk factors for cardiovascular disease [37, 38].
This study has several limitations. First, this was a cross-sectional designed study, therefore, a causal relationship between the exposure (i.e., UACR) and outcome (i.e., CVD status) could not be established, but only association between them could only be identified. To establish causation between UACR and CVD status, a longitudinal study design, intervention study design, or study using mendelian randomization analysis [39] is required. Second, we implemented the FRS to categorize subjects into binomial groups, including high- and low-risk CVD groups. Two reasons not to use the real CVD status obtained from a questionnaire for individuals’ current condition or diagnosis with CVD: the prevalence of CVD in KNHANES is extremely low, therefore, could yield the biased results; the real CVD status only reflects a subject’s current condition, besides, the FRS could predict their future risk of the incidence of CVD. Third, we could not consider the use of medications such as angiotensin-converting enzyme inhibitors or angiotensin receptor blockers which may reduce the degree of microalbuminuria. Fourth, we used a single urine spot sample to assess the UACR rather than the 24-hour urine collection or multiple samples. Nevertheless, we investigated the correlation between microalbuminuria and CVD in a single representative group by using nationally notarized data.
In conclusion, our study showed that UACR level was associated with FRS in both women and men with no previous history of CVD. In addition to FRS, measuring UACR is a cost-effective tool for predicting and preventing CVD in both sexes.
## References
1. Gerstein HC, Mann JF, Yi Q, Zinman B, Dinneen SF, Hoogwerf B. **Albuminuria and risk of cardiovascular events, death, and heart failure in diabetic and nondiabetic individuals**. *Jama* (2001) **286** 421-6. DOI: 10.1001/jama.286.4.421
2. Hillege HL, Fidler V, Diercks GF, van Gilst WH, de Zeeuw D, van Veldhuisen DJ. **Urinary albumin excretion predicts cardiovascular and noncardiovascular mortality in general population**. *Circulation* (2002) **106** 1777-82. DOI: 10.1161/01.cir.0000031732.78052.81
3. Solomon SD, Lin J, Solomon CG, Jablonski KA, Rice MM, Steffes M. **Influence of albuminuria on cardiovascular risk in patients with stable coronary artery disease**. *Circulation* (2007) **116** 2687-93. DOI: 10.1161/CIRCULATIONAHA.107.723270
4. Cirillo M, Laurenzi M, Panarelli P, Mancini M, Zanchetti A, De Santo NG. **Relation of urinary albumin excretion to coronary heart disease and low renal function: role of blood pressure**. *Kidney Int* (2004) **65** 2290-7. DOI: 10.1111/j.1523-1755.2004.00652.x
5. Carter CE, Katz R, Kramer H, de Boer IH, Kestenbaum BR, Peralta CA. **Influence of urine creatinine concentrations on the relation of albumin-creatinine ratio with cardiovascular disease events: the Multi-Ethnic Study of Atherosclerosis (MESA).**. *Am J Kidney Dis* (2013) **62** 722-9. DOI: 10.1053/j.ajkd.2013.05.010
6. Mogensen CE. **Microalbuminuria predicts clinical proteinuria and early mortality in maturity-onset diabetes**. *N Engl J Med* (1984) **310** 356-60. DOI: 10.1056/NEJM198402093100605
7. Won JC, Lee YJ, Kim JM, Han SY, Noh JH, Ko KS. **Prevalence of and factors associated with albuminuria in the Korean adult population: the 2011 Korea National Health and Nutrition Examination Survey.**. *PLoS One.* (2013) **8** e83273. DOI: 10.1371/journal.pone.0083273
8. Kim CH, Kim KJ, Kim BY, Jung CH, Mok JO, Kang SK. **Prediabetes is not independently associated with microalbuminuria in Korean general population: the Korea National Health and Nutrition Examination Survey 2011–2012 (KNHANES V-2,3).**. *Diabetes Res Clin Pract* (2014) **106** e18-21. DOI: 10.1016/j.diabres.2014.09.004
9. Park JI, Baek H, Jung HH. **Prevalence of chronic kidney disease in Korea: the Korean National Health and Nutritional Examination Survey 2011–2013.**. *J Korean Med Sci* (2016) **31** 915-23. DOI: 10.3346/jkms.2016.31.6.915
10. Han E, Lee YH, Kim G, Kim SR, Lee BW, Kang ES. **Sarcopenia is associated with albuminuria independently of hypertension and diabetes: KNHANES 2008–2011**. *Metabolism* (2016) **65** 1531-40. DOI: 10.1016/j.metabol.2016.07.003
11. Yoon HE, Nam Y, Kang E, Hwang HS, Shin SJ, Hong YS. **Gender-Specific associations between low skeletal muscle mass and albuminuria in the middle-aged and elderly population.**. *Int J Med Sci* (2017) **14** 1054-64. DOI: 10.7150/ijms.20286
12. Yu TY, Kim HY, Lee JM, Lee DH, Cho CG. **Association between bone mineral density and albuminuria: cross-sectional analysis of data from the 2011 Korea National Health and Nutrition Examination Survey V-2.**. *Endocrinol Metab (Seoul).* (2018) **33** 211-8. DOI: 10.3803/EnM.2018.33.2.211
13. Ahn HJ, Moon DS, Kang DY, Lee JI, Kim DY, Kim JH. **Urinary albumin excretion reflects cardiovascular risk in postmenopausal women without diabetes: the 2011 to 2013 Korean National Health and Nutrition Examination Survey.**. *Endocrinol Metab.* (2016) **31** 537-46. DOI: 10.3803/EnM.2016.31.4.537
14. **Third Report of the National Cholesterol Education Program (NCEP) expert panel on detection, evaluation, and treatment of high blood cholesterol in adults (Adult Treatment Panel III) final report.**. *Circulation* (2002) **106** 3143-421. DOI: 10.1161/circ.106.25.3143
15. Kweon S, Kim Y, Jang MJ, Kim Y, Kim K, Choi S. **Data resource profile: the Korea National Health and Nutrition Examination Survey (KNHANES).**. *Int J Epidemiol* (2014) **43** 69-77. DOI: 10.1093/ije/dyt228
16. Kim Y.. **The Korea National Health and Nutrition Examination Survey (KNHANES): current status and challenges**. *Epidemiol Health.* (2014) **36** e2014002. DOI: 10.4178/epih/e2014002
17. Mattix HJ, Hsu CY, Shaykevich S, Curhan G. **Use of the albumin/creatinine ratio to detect microalbuminuria: implications of sex and race**. *J Am Soc Nephrol* (2002) **13** 1034-9. DOI: 10.1681/ASN.V1341034
18. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ. **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) **140** e596-e646. DOI: 10.1161/CIR.0000000000000678
19. Cholesterol Treatment Trialists C, Mihaylova B, Emberson J, Blackwell L, Keech A, Simes J. **The effects of lowering LDL cholesterol with statin therapy in people at low risk of vascular disease: meta-analysis of individual data from 27 randomised trials**. *Lancet (London, England).* (2012) **380** 581-90. DOI: 10.1016/S0140-6736(12)60367-5
20. Franklin BA, Brook R, Arden Pope C. **Air pollution and cardiovascular disease.**. *Curr Probl Cardiol* (2015) **40** 207-38. DOI: 10.1016/j.cpcardiol.2015.01.003
21. Lee T, Lee H. **Identification of disease-related genes that are common between Alzheimer’s and cardiovascular disease using blood genome-wide transcriptome analysis.**. *Biomedicines* (2021) **9** 1525. DOI: 10.3390/biomedicines9111525
22. D’Agostino RB, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM. **General cardiovascular risk profile for use in primary care: the Framingham Heart Study**. *Circulation* (2008) **117** 743-53. DOI: 10.1161/CIRCULATIONAHA.107.699579
23. Goff DC, Lloyd-Jones DM, Bennett G, Coady S, D’Agostino RB, Gibbons R. **2013 ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines**. *Circulation* (2014) **129** S49-73. DOI: 10.1161/01.cir.0000437741.48606.98
24. Cox DR. **Regression models and life-tables**. *Journal of the Royal Statistical Society Series B (Methodological).* (1972) **34** 187-220
25. 25R Core Team. R: A language and environment for statistical computing [Internet]. Vienna, Austria:
R Foundation for Statistical Computing; 2019. Available from: https://www.R-project.org/.. (2019)
26. Romundstad S, Holmen J, Kvenild K, Hallan H, Ellekjaer H. **Microalbuminuria and all-cause mortality in 2,089 apparently healthy individuals: a 4.4-year follow-up study. The Nord-Trøndelag Health Study (HUNT), Norway**. *Am J Kidney Dis* (2003) **42** 466-73. DOI: 10.1016/s0272-6386(03)00742-x
27. Xu J, Knowler WC, Devereux RB, Yeh J, Umans JG, Begum M. **Albuminuria within the "normal" range and risk of cardiovascular disease and death in American Indians: the Strong Heart Study.**. *Am J Kidney Dis* (2007) **49** 208-16. DOI: 10.1053/j.ajkd.2006.10.017
28. Arnlöv J, Evans JC, Meigs JB, Wang TJ, Fox CS, Levy D. **Low-grade albuminuria and incidence of cardiovascular disease events in nonhypertensive and nondiabetic individuals: the Framingham Heart Study**. *Circulation* (2005) **112** 969-75. DOI: 10.1161/CIRCULATIONAHA.105.538132
29. Asselbergs FW, Hillege HL, van Gilst WH. **Framingham score and microalbuminuria: combined future targets for primary prevention?**. *Kidney Int Suppl* (2004) S111-4. DOI: 10.1111/j.1523-1755.2004.09227.x
30. Peters SAE, Muntner P, Woodward M. **Sex Differences in the Prevalence of, and Trends in, Cardiovascular Risk Factors, Treatment, and Control in the United States, 2001 to 2016**. *Circulation* (2019) **139** 1025-35. DOI: 10.1161/CIRCULATIONAHA.118.035550
31. Deckert T, Feldt-Rasmussen B, Borch-Johnsen K, Jensen T, Kofoed-Enevoldsen A. **Albuminuria reflects widespread vascular damage. The Steno hypothesis**. *Diabetologia* (1989) **32** 219-26. DOI: 10.1007/BF00285287
32. Jensen JS, Borch-Johnsen K, Jensen G, Feldt-Rasmussen B. **Microalbuminuria reflects a generalized transvascular albumin leakiness in clinically healthy subjects**. *Clin Sci (Lond)* (1995) **88** 629-33. DOI: 10.1042/cs0880629
33. Nosadini R, Velussi M, Brocco E, Abaterusso C, Piarulli F, Morgia G. **Altered transcapillary escape of albumin and microalbuminuria reflects two different pathogenetic mechanisms**. *Diabetes* (2005) **54** 228-33. DOI: 10.2337/diabetes.54.1.228
34. Jones CA, Francis ME, Eberhardt MS, Chavers B, Coresh J, Engelgau M. **Microalbuminuria in the US population: third National Health and Nutrition Examination Survey**. *Am J Kidney Dis* (2002) **39** 445-59. DOI: 10.1053/ajkd.2002.31388
35. Kim J-H, Hwang K-H, Dang BTN, Eom M, Kong ID, Gwack Y. **Insulin-activated store-operated Ca2+ entry via Orai1 induces podocyte actin remodeling and causes proteinuria.**. *Nat Commun* (2021) **12** 6537. DOI: 10.1038/s41467-021-26900-w
36. Jung D-H, Byun Y-S, Kwon Y-J, Kim G-S. **Microalbuminuria as a simple predictor of incident diabetes over 8 years in the Korean Genome and Epidemiology Study (KoGES).**. *Sci Rep.* (2017) **7** 15445. DOI: 10.1038/s41598-017-15827-2
37. Adeva-Andany MM, Martínez-Rodríguez J, González-Lucán M, Fernández-Fernández C, Castro-Quintela E. **Insulin resistance is a cardiovascular risk factor in humans.**. *Diabetes & Metabolic Syndrome: Clinical Research & Reviews.* (2019) **13** 1449-55. DOI: 10.1016/j.dsx.2019.02.023
38. Dal Canto E, Ceriello A, Rydén L, Ferrini M, Hansen TB, Schnell O. **Diabetes as a cardiovascular risk factor: An overview of global trends of macro and micro vascular complications.**. *Eur J Prev Cardiol* (2020) **26** 25-32. DOI: 10.1177/2047487319878371
39. Liu X, Tong X, Zou Y, Lin X, Zhao H, Tian L. **Mendelian randomization analyses support causal relationships between blood metabolites and the gut microbiome**. *Nature Genetics* (2022) **54** 52-61. DOI: 10.1038/s41588-021-00968-y
|
---
title: 'A co-designed website (FindWays) to improve mental health literacy of parents
of children with mental health problems: Protocol for a pilot randomised controlled
trial'
authors:
- Daniel Peyton
- Greg Wadley
- Naomi Hackworth
- Anneke Grobler
- Harriet Hiscock
journal: PLOS ONE
year: 2023
pmcid: PMC10030009
doi: 10.1371/journal.pone.0273755
license: CC BY 4.0
---
# A co-designed website (FindWays) to improve mental health literacy of parents of children with mental health problems: Protocol for a pilot randomised controlled trial
## Abstract
### Background
Mental health problems, such as behavioural and emotional problems, are prevalent in children. These problems can have long lasting, detrimental effects on the child, their parents and society. Most children with a mental health problem do not receive professional help. Those that do get help can face long wait times. While waiting, parents want to learn how they can help their child. To address this need, we co-designed a new website to help parents find ways of helping their child’s mental health problem while waiting to get specialist help.
### Objectives
To assess the acceptability and feasibility of a new co-designed website, FindWays, through a pilot randomised controlled trial. The protocol is registered with ISRCTN (ISRCTN64605513).
### Methods
This study will recruit up to 60 parents of children aged two-twelve years old referred to a paediatrician for behavioural and/or emotional problems. Participants will be randomly allocated by computer generated number sequence to either the intervention or control group. Intervention group participants will receive access to the FindWays website to help them manage their child’s mental health problem while they wait to see the paediatrician. Acceptability and feasibility will be assessed over the 4-month intervention through mixed methods including: recruitment, adherence, retention, net promoter score (quantitative measures) and semi-structured interviews to gain an in-depth understanding of parents’ experience and potential adverse effects (qualitative measure). Secondary outcomes measured by parent survey at 4-months post randomisation include child mental health, parent mental health, impact of the child’s mental health problem on their functioning and family, and health service use and associated costs.
### Results
Recruitment commenced June 2022 with publication expected in October 2023.
### Conclusion
This study will provide novel data on the acceptability and feasibility of a new website co-designed with parents to help them find ways of managing their child’s behaviour and emotions.
## Mental health problems and their long term effects
Diagnosed mental health disorders, such as disruptive behaviour, anxiety and mood problems, affect one in seven Australian children [1]. However, mental health problems, which can still impact the child but may not reach diagnostic criteria for a mental health disorder. These more common behavioural and emotional problems likely affect around one in four children [1]. These common problems impact children’s short and long term outcomes including difficulty with relationships [2], reduced educational achievement [2], increased risk of criminality [3, 4], alcohol and drug problems [5] and lower earning potential [3].
Half of all adult mental health disorders start in childhood [6]. Evidence-based treatments including parenting interventions have been shown to improve child behavioural and emotional problems, including face to face therapy, group parenting programs (e.g., Tuning in to Kids) and online parenting interventions (e.g., computer assisted cognitive behaviour therapy, such as BRAVE online) [7, 8]. Often, interventions target the parents given that parenting is a key modifiable factor in child mental health [9, 10].
Despite the availability of a range of evidence-based interventions for mental health problems, most children with a mental health disorder do not receive professional help [11–14]. Reasons for this are multifactorial and include accessibility barriers, stigma, and poor parental (and professional) mental health literacy [11]. Mental health literacy is the “knowledge and beliefs about mental disorders which aid their recognition, management or prevention” [15][p182]. This can include knowing where a child can go to get mental health help, or strategies parents can implement to help alleviate their child’s problem.
Qualitative studies have shown that parents often face difficulty navigating the maze of services for mental health [16, 17]. Parents can find it hard to know where to go or who can help. Many will end up seeing a paediatrician to help their child [18].
In Australia, the paediatrician is the largest provider of longitudinal care of children with a mental health disorder [11]. Paediatricians are often an ‘early port of call’ for children with mental health disorders and play a crucial role in diagnosis, exclusion of medical causes of child mental health problems and prescription of medication. *Paediatricians* generally do not deliver behaviour management therapy nor are they trained in cognitive behavioural therapy. *They* generally refer to other providers for these interventions.
With few publicly funded paediatric services available, anecdotal reports suggest wait times in excess of six to twelve months to access a paediatrician in Australia. Pre-COVID [2019], mean wait time to access a paediatrician in the states of Victoria and South Australia was 44 days with four out of 10 providers closing their books to new patients mid-year [19]. This problem of accessibility has worsened due to the COVID-19 pandemic, especially for families who cannot afford private care.
## Parents want support and information while they wait to see a mental health specialist
Parents want information while they wait to access mental health care [20]. This includes a better understanding of how the mental health system works and how to access other forms of treatment to help their child while they wait, such as group therapy, online interventions or telephone support [20].
There is little qualitative data on parent’s information preferences while waiting to see a specialist mental health professional. From first author’s qualitative interviews with 16 parents waiting to get help from a paediatrician (unpublished data), inductive content analysis revealed that parents want to understand whether their child’s problem is normal or not, what they could try while they waited to see the paediatrician, which services could help their child’s problem, and a list of available professional help in their community.
## Why a digital health intervention might help
Digital health interventions (DHIs) may be able to address these information gaps identified by parents. DHIs are desirable for the following reasons: they facilitate the rapid exchange of tailored information to consumers, are scalable, are liked by parents, can have vast reach, are accessible on demand, and generally thought to be cost-effective [21].
Our recent systematic review identified only five studies examining the effects of a DHI on parents’ mental health literacy or help-seeking for their child’s mental health problem [22]. There is some evidence that DHIs, such as websites and emailed PowerPoint presentations, may help improve mental health literacy [23–25]. However, the review included only one randomised controlled trial [26]. A Finnish study measured help-seeking behaviours of parents of four year old children with a recognised behaviour problem at a routine four year old check with a universal service. The randomised controlled trial found an information website, coupled with a telephone coach decreased uptake of services. However, this decrease in uptake of services was associated with improved child behaviour using a validated measure. This study showed a DHI, with a telephone coach, can affect help-seeking [26]. However, no quantitative studies have shown DHIs can improve uptake of services for children with a mental health problem.
## A new co-designed website to improve help-seeking and child mental health outcomes
To design a new website to improve parents’ knowledge of treatments and reduce barriers to accessing evidence-based therapies for their child, we followed a framework proposed by Gemert-Pijen et al. [ 27]. This is a user-centred, iterative approach that refines the DHI to match the needs of the user, during every phase of the development cycle [27]. The website was developed in collaboration with parents and clinicians. As part of this process, we undertook a contextual enquiry with parents, three co-design workshops and six one-on-one usability studies. Through this process, we designed and developed a website with the content detailed in the intervention section below.
## Evaluating the feasibility and acceptability of the DHI
The World Health Organisation recommends evaluating the feasibility of a new DHI, after usability testing and prior to efficacy and effectiveness evaluation [28]. The measurement of feasibility is complicated by the absence of a singular definition of feasibility testing, and the synonymous use of feasibility studies and pilot studies [29, 30]. Feasibility trials typically involve 20–100 participants, and evaluate the suitability of outcome measures, participant acceptance of the intervention (recruitment and retention rates) and DHI usage (e.g. adherence) [28–32].
From our systematic review of five included studies, we found single cohort trials limited the strength of evidence of effectiveness as this methodology is inherently prone to confounding bias [33]. None of the included studies in the systematic review identified or adjusted for the presence of confounding. To address this, we plan to conduct a randomised controlled pilot study with a “routine care” control group. This also allows us to describe differences in efficacy of secondary outcomes between the two groups, accounting for confounding and natural history of child mental health problems over time.
## Aim
We aim to assess the acceptability and feasibility of a co-designed website, FindWays, through a pilot randomised controlled trial using one-to-one allocation in parallel arms. The study will also assess efficacy for outcomes related to child mental health, parent mental health, family impact and health service use and costs.
## Hypothesis
We hypothesise that FindWays is an acceptable and feasible DHI with the potential to help parents find ways to improve their child’s mental health problems while they wait to see a paediatrician.
## Trial design
We will conduct a pilot randomised controlled trial with participants randomly assigned to receive either the FindWays website (intervention group) or routine care (control group). We will use a mixed methods evaluation to determine acceptability and feasibility.
## Setting
Participants will be recruited from three general paediatric private outpatient clinics in Geelong, Australia. Geelong is the largest regional city in the state of Victoria. 51,804 children under the age of 15 live in Geelong [34]. At the start of recruitment, there were no public general paediatric outpatient clinics. The list of recruitment sites is available from the website https://tiny.one/findways.
## Eligibility
Participants in the study must be a parent of a child referred by a general practitioner (family doctor) to a paediatrician at one of three private paediatric clinics, for a first appointment to evaluate or manage their child’s behavioural or emotional problem. Participants will be assigned to a randomised trial intervention only if they meet all the inclusion criteria and none of the exclusion criteria. Only one parent of a child will be enrolled.
## Inclusion criteria
Each participant must meet all the following criteria to be enrolled in this trial:
## Exclusion criteria
Participants meeting any of the following criteria will be excluded from the trial:
## Purpose
The intervention consists of a new, co-designed website. This website, FindWays, offers parents relevant and specific information on behavioural and emotional problems. This information focuses on ways parents can manage a behavioural or emotional problem.
## Theory
Human-centred and participatory design methods were used to inform the website’s design, as was persuasive technology [35]. These methods align with those suggested by existing theoretical framework for DHIs [27]. In order to encourage behaviour change, the UK Behavioural Insights Team EAST behavioural insights were applied to the design and content of website [36]. The behavioural insights team recommend four principles to encourage a behaviour: make is easy, attractive, social and timely [36].
## Contextual enquiry
The content of the website was informed by inductive content analysis of 16 qualitative interviews completed in early 2020 by DP with parents recently referred to a paediatrician to manage their child’s behavioural or emotional problem.
## Prototyping
We conducted a further three co-design sessions in 2022 with five parents to inform the functional requirements and look and feel of the website. We designed the intervention using an iterative process, sketching wireframes and mock-ups of the website using free sketch on pen and paper, PowerPoint, Adobe XD and finally in Figma, and asking parents for feedback on content and usability. The development of the prototype in Figma and the website in Webflow was assisted by digital technology design company Curve Tomorrow [37]. The prototype was built in Webflow utilising the Content Management System. Usability testing was completed with a further six parents. These usability tests were completed as one-on-one interviews, either face-to-face in a paediatric clinic or over Zoom.
## Technical
This static website contains 60 pages of content, 47 videos and a list of 63 local and online services for child mental health, hosted on Webflow. The parent can select specific ages (preschool or primary school aged) and problems (15 in total) to find tailored information relevant to their child.
## Content
The website contains the following elements: i) strategies for parents to try at home based on a program logic that identifies evidence-based strategies associated with positive child mental health outcomes, ii) descriptions of the roles of professionals and programs that can help a child’s mental health problem, and iii) lists of local professionals and available programs. The website content includes videos, written content, and links to relevant external content and services, including local psychologists, occupational therapists and evidence-based programs. The content is tailored to the child’s age and specific sub-problem (e.g., separation anxiety, tantrums), as identified by the parent. The parent can choose what help they want for their child, but the website guides parents, who self-identify through the website that they want more help, to a local professional or program that can help their child’s specific sub-problem.
The parenting strategies provided by the website reference existing information sources such as the Raising Children Network, Beyond Blue and the Diagnostic Standard Manual-V (DSM-V) [38]. The Raising Children Network and Beyond Blue are both Australian websites providing mental health content for consumers, funded in part by the Australian Government. DSM-V is a diagnostic tool, published by the American Psychiatric Association (APA) and used by clinicians as a standardised guide to classify mental health disorders [38].
All content on FindWays was written by first author DP, a general paediatrician. The content underwent a quality review process including review by another paediatrician (HH). Video scripts were reviewed and approved by the paediatricians in Geelong. The content was also reviewed by NH (a psychologist and Director, Raising Children Network) to ensure the tone is appropriate for families.
## Prompts
Throughout the intervention, parents will receive scheduled prompts to use the website. These prompts, delivered by SMS and email, will contain static information drawn from the website, and a link to the website. These will be delivered at regular, pre-scheduled intervals and are not tailored. Parents will be able to request the prompts stop at any time, either by email or text message.
## Control group
The control group will receive routine care. This care is provided by usual providers (e.g., GP, existing online resources, teachers at school). The control group will not be provided access to the FindWays website. The FindWays website is not listed on search engines, and the URL is not publicly available.
## Outcomes
Outcomes of this trial will be measured using mixed methods. Data will be collected at baseline and at four months post randomisation, except for DHI usage data which will be measured continuously by Google Analytics and downloaded manually every month during the trial. See Table 1 for a summary of the outcome measures and data collection time points.
**Table 1**
| Demographic data, Primary and Secondary Objectives | Data Sources | Method of collection | Data Collection Time points | Outcomes |
| --- | --- | --- | --- | --- |
| Primary outcomes | Primary outcomes | Primary outcomes | Primary outcomes | Primary outcomes |
| Feasibility–recruitment and retention | Clinic database | Print copy of database | Recorded at weekly intervals for the duration of recruitment | How many eligible patients were flagged by the paediatrician, how many were contacted by the clinic, and how many agreed to pass on their details to the researcher and how many were randomised to the intervention/control, and completed their survey at four months. |
| Adherence and usage (intervention group only) | Google Analytics | Data extracted manually from Google Analytics | Baseline and Weekly intervals for 4 months | Number of page views, individual sessions, time spent on each page for each participant. |
| Adherence and usage (intervention group only) | Google Analytics | Data extracted manually from Google Analytics | Baseline and Weekly intervals for 4 months | Strategies viewed, provider profiles viewed, programs viewed. |
| Acceptability–usability and safety (intervention group only) | Parent qualitative Interview | Qualitative interview with a subsample of approximately 20 parents from Intervention group | 4 months | Understanding of the parent’s experience of using the platform, facilitators, and barriers to use, and any perceived adverse outcomes |
| Adherence–task completion (intervention group only) | Parent Survey | Online via REDCap | 4 months | Whether parents reported viewing any of the resources within the website, and whether this changed the parent’s behaviour (e.g., help-seeking, implementing new strategies) |
| Acceptability—recommendation (intervention group only) | Parent Survey | Online via REDCap | 4 months | Net Promoter Score measured via 10-point Likert scale. |
| Secondary outcomes | Secondary outcomes | Secondary outcomes | Secondary outcomes | Secondary outcomes |
| Secondary outcome: child behaviour and emotions | Parent Survey | Control and intervention group surveys completed online via REDCap. | Baseline and 4 months | Strengths and Difficulties Questionnairea (SDQ) |
| Family impact of child’s behaviour and emotions | Parent survey | Control and intervention group surveys completed online via REDCap. | Baseline and 4 months | SDQ Impact Supplementa measuring the impact of the child’s behaviour and emotions on their functioning and the family. |
| Secondary outcome: Parent mental health and distress | Parent Survey | Control and intervention group surveys completed online via REDCap. | Baseline and 4 months | Depression, Anxiety and Stress Scalea (DASS-21) |
| Secondary outcome: Health service use | Parent Survey | Control and intervention group surveys completed online via REDCap. | Baseline and 4 months | Survey of services used in the past 4 months for the child’s emotional or behavioural problems, including number of times accessed, distance travelled to service, and any out of pocket costs ($AUD). |
## Primary outcomes measures
Despite the lack of a singular definition of feasibility, they typically involve measures of recruitment rates, attrition rates, evidence of harm, user satisfaction and DHI usage [28, 29] and will be assessed in the following ways:
## Secondary outcome measures
Secondary outcomes measured by parent-reported surveys at baseline and 4-months post randomisation include: Demographic characteristics of the user will be recorded by parent survey at baseline.
The feasibility of these secondary outcomes will also be assessed by reviewing how many parents complete the secondary outcome measures. This will be used to inform the design of a later trial powered to assess the effectiveness of the FindWays website.
## Participant timelines
Participants will be recruited over an estimated three-month period in mid 2022 using a rolling recruitment strategy. After participants have consented and completed their baseline measures, they will be randomised and followed over 4 months. At the end of the 4 months, participants will fill out a second survey. Those in the intervention group will be invited to participate in a semi-structured interview. A visual summary of the timeline and assessment points can be found in Fig 1.
**Fig 1:** *SPIRIT schedule of enrolment, interventions, and assessments.*
## Sample size
There is no consistent advice regarding the number of participants required to assess acceptability and feasibility. Typically, between 20–100 participants are recruited into similar trials assessing acceptability, feasibility and usability [28, 30, 32]. For this trial, we anticipate recruiting up to 60 participants, within a three month time frame. As a pilot RCT, the trial is not powered to evaluated effectiveness of the intervention compared to control for parent, child or health service-related outcomes.
## Recruitment
The study design, including recruitment and allocation, is summarised in Fig 2.
**Fig 2:** *CONSORT flow diagram.Study design flowchart of the FindWays trial.*
Screening eligible referrals. Paediatricians will screen incoming referrals for eligibility. Paediatricians will flag referrals as potentially eligible if they appear to meet inclusion criteria and do not meet exclusion criteria. These potentially eligible participants will then be contacted via telephone by the administration staff at each clinic. The administration staff will request consent to pass on the parent’s contact details to DP to hear more about the study. Those who do not consent to passing on their contact details will have their non-consent recorded without any identifying information, to help ascertain recruitment rates.
## Phone parents direct
For those parents who wish to hear more about the study, the administration staff will take and record verbal consent from the parent to pass on their contact details (parent name, email address, phone number and postcode) to DP. At this time, the clinic will also post the parents a copy of the participant information statement. DP will then contact the parents to tell them more about the study, check they received the participant information statement and answer any questions.
## Sequence generation
A statistician, not involved in the analysis of the trial results, will prepare the randomisation schedule. The randomisation schedule will be created using computer-generated random numbers before the first participant has been recruited, in a one-to-one ratio. The participant cohort will be stratified by child age (two-six year olds and seven-twelve year olds) and clinic (clinic one, clinic two and clinic three). Within each stratum, permuted block randomisation will be used to ensure balance between the intervention and control group. A randomly generated sequence of block sizes containing two, four, or six participants will be used. This will help prevent any predictability when randomising participants to intervention or control [43].
## Allocation concealment
The schedule will be held by the independent statistician, and allocation will not be revealed prematurely to DP. Because of these procedures, the research team will be unable to predict which group participants will be allocated to.
## Implementation
When a participant has consented to participate, and completed their baseline measures, DP will contact the independent statistician so they can reveal the participant’s randomisation status. Both DP and the participant will become aware of which trial group they were allocated to after randomisation.
## Blinding
The participant and researcher will not be blind to their intervention status because it is impossible to blind a novel website intervention to participants. The administration staff and paediatricians will not be notified by the researchers of the intervention status of the participants.
## Withdrawing from the trial
Participants are free to withdraw from the trial at any time upon request. If known, a brief reason will be recorded on the participant database. All parents in the intervention group who withdraw will be asked if they are still willing to participate in the final interview to understand their experience using the platform. Withdrawing from the trial will not affect their access to standard treatment or their relationship their paediatrician. If they withdraw, they will be asked to no longer access the FindWays website. For participants that do withdraw, we will report on whether they continued to access the FindWays website using their unique link by reviewing Google Analytics data.
## Ancillary and post-trial care
Participants in the intervention group will have ongoing access to the FindWays website after the trial is over. Participants in the control group will also have access to the FindWays website after completion of their final survey four months after randomisation. Participants in both the intervention and control group will be free to utilise any treatments (other than the intervention) available to them.
## Data collection and management
Data will be collected and entered using electronic data collection forms which will be completed by the participant. Website usage data will be collected by Google Analytics and manually transferred across to the relevant participant record on REDCap, by DP. Google Analytics collects website usage data. For this trial, each participant will be sent a unique link with an individual Google Analytics tracking code. Google Analytics will then be able to show individual usage data for each participant.
Participants in the intervention group will be asked if they are willing to participate in an optional telephone semi-structured interview. DP will conduct all the phone interviews as per the interview guide. The interview will last approximately 20–30 minutes and is designed to better understand the parent’s experience using the platform, enquire about any adverse events (including safety issues), barriers to use, or ways the website could be improved. Participant interviews will be recorded via dictaphone and all audio recordings will be transcribed verbatim.
Hard copy data will be stored in a locked cabinet in a secure location, accessible to the research team only. Electronic data will be securely stored in MCRI’s REDCap database system and in files stored in MCRI’s network file servers, which are backed up nightly.
REDCap is hosted on MCRI infrastructure and is subject to the same security and backup regimen as other systems (e.g., the network file servers).
## Statistical analysis
The baseline characteristics of the intervention and control groups will be summarised and presented separately. The primary outcome data will be summarised and presented as percentages.
For secondary outcomes linear regression and logistic regression will be conducted to estimate mean differences (and $95\%$ confidence intervals) for continuous outcomes, and odds ratios (and $95\%$ confidence intervals) for binary outcomes, respectively, between trial groups. Analyses will be adjusted for baseline scores of the outcome measure. This pilot RCT will be reported in accordance with the CONSORT e-health statement and we intend to complete a multiple imputation intention-to-treat analysis at the level of the child [44].
Qualitative interviews will be analysed using an inductive content analysis approach. This approach employs three main phases: i) open coding; ii) creating categories by cross referencing, and grouping the data; and, iii) abstraction [45, 46]. This analysis will be conducted using NVivo software. Data will be coded by DP and discussed with supervising authors on a regular basis. The coding framework will be reviewed by each of the co-authors. The analysis will describe key themes and events in parents’ experience using the FindWays website.
## Ethical considerations
This protocol, the informed consent document and any subsequent amendments was reviewed and approved prior to commencing the research. Ethics approval granted by The Royal Children’s Hospital Human Research Ethics Committee. HREC/75854/RCHM-2021 on $\frac{22}{4}$/2022. A letter of protocol approval by HREC was obtained prior to the commencement of the trial, as well as approval for other trial documents requiring HREC review. Amendments will be communicated to investigators, ISRCTN and publishing journals.
The protocol is registered on ISRCTN (ISRCTN64605513). As this is a pilot study, a data monitoring committee is not necessary and interim analysis will not be conducted. Safety is being measured by recording individual level quantitative participant data on mental health and health service-related outcomes, as well as through qualitative interviews offered to all participants in the intervention group.
DP will conduct the informed consent discussion and will check that the parent comprehends the information provided. DP will answer any questions about the trial.
The parent will be invited to provide verbal consent. Consent will be voluntary and free from coercion. Optional consent will be obtained by DP from participants in the intervention group, who choose to participate in the semi-structured interview four months after randomisation.
Participant confidentiality is strictly held in trust by the participating investigators, research staff, and the sponsor. This confidentiality is extended to cover the health information of the participants and will not be released without written permission of the participant, except as necessary for monitoring by HREC or regulatory agencies, or as required by law.
## Dissemination policy and results
Upon completion of the study, results will be disseminated via four methods: i) Publication of results in a peer-reviewed journal; ii) presentation of results at conferences; iii) presentation of results to local paediatricians in Geelong; and iv) plain language summary of results distributed to parents agreeing to receive pilot results.
Up to January 2022, 32 participants have been recruited. Participants are expected to be enrolled until February 2022. The final outcome measures are expected to be collected in June 2023, with publication of results expected in December 2023.
The authors do not intend to use professional writers for any part of the publication.
## Discussion
This pilot RCT will determine whether a co-designed DHI, designed to help parents find strategies and services to help their child, is feasible and acceptable among parents referred to a paediatrician for their child’s behavioural or emotional problems. This will be the first evaluation of its kind of a DHI for parents targeting mental health problems in young children. It will also be the first randomised trial to measure the effects of a self-directed website targeting parents on uptake of mental health services.
Limitations of this protocol include parents being recruited from a single regional city, and the study not sufficiently powered to evaluate for efficacy of the website. However, following the pilot RCT, we plan to conduct a fully powered RCT to determine the effectiveness and cost-effectiveness of this intervention in reducing the need for paediatrician services and improving child and parent outcomes The recruitment and retention rates in this pilot will inform the later RCT design, to allow for adequate time to recruit a fully powered sample. Further, the first author will conduct the qualitative interviews and the participants are aware of their involvement in the design of the research. This could possibly result in a bias towards positive feedback.
In the long term, the FindWays website is potentially impactful because: i) it may help families improve their child’s behaviour and emotions, without needing to see a paediatrician, or ii) they can find and engage with available evidence-based services earlier.
Mental health workers may also find the website useful for identifying strategies, programs and providers that are known to help children with a particular issue and available in their community. These mental health workers could include wellbeing officers at schools, general practitioners, well child/family nurses, and even paediatricians.
If this DHI is a feasible and acceptable way of linking some families to the right treatments, without the need to consult a paediatrician, it could increase the efficiency of the health system. With fewer families on waitlists, children with more complex or severe problems may be seen sooner. The overall capacity of the health system may be increased as we expect families to access group parenting programs and scalable online programs, where there is capacity to see more children than through individual face-to-face appointments with a specialist.
## References
1. Lawrence D, Johnson S, Hafekost J, de Haan KB, Sawyer M, Ainley J. *The mental health of children and adolescents: report on the second Australian child and adolescent survey of mental health and wellbeing* (2015.0)
2. Duchesne S, Vitaro F, Larose S, Tremblay RE. *J Youth Adolesc* (2008.0) **37** 1134-46
3. Asendorpf JB, Denissen JJA, van Aken MAG. **Inhibited and Aggressive Preschool Children at 23 Years of Age: Personality and Social Transitions into Adulthood.**. *Dev Psychol.* (2008.0) **44** 997-1011. DOI: 10.1037/0012-1649.44.4.997
4. Stevenson J, Goodman R. **Association between behaviour at age 3 years and adult criminality.**. *British Journal of Psychiatry.* (2001.0) **179** 197-202. DOI: 10.1192/bjp.179.3.197
5. Woodward LJ, Fergusson DM. **Life course outcomes of young people with anxiety disorders in adolescence.**. *J Am Acad Child Adolesc Psychiatry* (2001.0) **40** 1086-93. DOI: 10.1097/00004583-200109000-00018
6. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. **Lifetime Prevalence and Age-of-Onset Distributions of DSM-IV Disorders in the National Comorbidity Survey Replication.**. *Arch Gen Psychiatry [Internet].* (2005.0) **62** 593. PMID: 15939837
7. Havighurst SS, Wilson KR, Harley AE, Prior MR, Kehoe C. **Tuning in to Kids: improving emotion socialization practices in parents of preschool children–findings from a community trial**. *Journal of Child Psychology and Psychiatry* (2010.0) **51** 1342-50. DOI: 10.1111/j.1469-7610.2010.02303.x
8. March S, Spence SH, Donovan CL. **The Efficacy of an Internet-based CBT Intervention for Child Anxiety Disorders**. *J Pediatr Psychol* (2009.0) **34** 474-87. PMID: 18794187
9. Kiernan KE, Mensah FK. **Poverty, family resources and children’s early educational attainment: The mediating role of parenting.**. *Br Educ Res J* (2011.0) **37** 317-36
10. Gardner F, Shaw DS. **Behavioral problems of infancy and preschool children (0–5).**. *Rutter’s child and adolescent psychiatry.* (2008.0) **5** 882-92
11. Tuominen M.. **Comparison of anileridine and pethidine in preventing pain responses during nitrous oxide-oxygen anaesthesia**. *Ann Chir Gynaecol Fenn [Internet].* (1975.0) **64** 299-306. PMID: 2091
12. Sadler K, Vizard T, Ford T, Marchesell F, Pearce N, Mandalia D. *Mental health of children and young people in England, 2017.* (2018.0)
13. 13US Department of Health and Human Services. Mental health: A report of the Surgeon General. 1999;
14. Jensen PS, Goldman E, Offord D, Costello EJ, Friedman R, Huff B. **Overlooked and underserved:“action signs” for identifying children with unmet mental health needs.**. *Pediatrics* (2011.0) **128** 970-9. DOI: 10.1542/peds.2009-0367
15. Jorm AF, Korten AE, Jacomb PA, Christensen H, Rodgers B, Pollitt P. **“Mental health literacy”: A survey of the public’s ability to recognise mental disorders and their beliefs about the effectiveness of treatment.**. *Medical Journal of Australia* (1997.0) **166** 182-6. DOI: 10.5694/j.1326-5377.1997.tb140071.x
16. Reardon T, Harvey K, Baranowska M, O’Brien D, Smith L, Creswell C. **What do parents perceive are the barriers and facilitators to accessing psychological treatment for mental health problems in children and adolescents? A systematic review of qualitative and quantitative studies.**. *Eur Child Adolesc Psychiatry* (2017.0) **26** 623-47. DOI: 10.1007/s00787-016-0930-6
17. Boulter E, Rickwood D. **Parents’ experience of seeking help for children with mental health problems.**. *ADVANCES IN MENTAL HEALTH* (2013.0) **11** 131-42
18. Johnson SE, Lawrence D, Hafekost J, Saw S, Buckingham WJ, Sawyer M. **Service use by Australian children for emotional and behavioural problems: Findings from the second Australian Child and Adolescent Survey of Mental Health and Wellbeing**. *Australian & New Zealand Journal of Psychiatry.* (2016.0) **50** 887-98. DOI: 10.1177/0004867415622562
19. Mulraney M, Lee C, Freed G, Sawyer M, Coghill D, Sciberras E. **How long and how much? Wait times and costs for initial private child mental health appointments.**. *J Paediatr Child Health.* (2021.0) **57** 526-32. DOI: 10.1111/jpc.15253
20. Cunningham CE, Chen Y, Deal K, McGrath P, Reid G, Lipman E. **The Interim Service Preferences of Parents Waiting for Children’s Mental Health Treatment: A Discrete Choice Conjoint Experiment.**. *J Abnorm Child Psychol.* (2013.0) **41** 865-77. DOI: 10.1007/s10802-013-9728-x
21. 21World Health Organisation. WHO guideline: recommendations on digital interventions for health system strengthening. Geneva; 2019.
22. Peyton D, Goods M, Hiscock H. **The Effect of Digital Health Interventions on Parents’ Mental Health Literacy and Help Seeking for Their Child’s Mental Health Problem: Systematic Review.**. *J Med Internet Res* (2022.0) **24** e28771. DOI: 10.2196/28771
23. Sapru I, Khalid-Khan S, Choi E, Alavi N, Patel A, Sutton C. **Effectiveness of online versus live multi-family psychoeducation group therapy for children and adolescents with mood or anxiety disorders: A pilot study**. *Int J Adolesc Med Health* (2018.0) **30**
24. Ryan GS, Haroon M, Melvin G. **Evaluation of an educational website for parents of children with ADHD. Int J Med Inform**. *Internet]* (2015.0) **84** 974-81. DOI: 10.1016/j.ijmedinf.2015.07.008
25. Montoya A, Hernández S, Massana M, Fernández-Jaén A, Herreros O, Diez JA. **Evaluating internet information on attention-deficit/hyperactivity disorder (ADHD) treatment: Parent and expert perspectives.**. *Education for Health* (2013.0) **26** 48-53. DOI: 10.4103/1357-6283.112801
26. Sourander A, McGrath PJ, Ristkari T, Cunningham C, Huttunen J, Hinkka-Yli-Salomäki S. **Two-Year Follow-Up of Internet and Telephone Assisted Parent Training for Disruptive Behavior at Age 4.**. *J Am Acad Child Adolesc Psychiatry* (2018.0) **57** 658-668. DOI: 10.1016/j.jaac.2018.07.001
27. van Gemert-Pijnen JEWC, Nijland N, van Limburg M, Ossebaard HC, Kelders SM, Eysenbach G. **A holistic framework to improve the uptake and impact of eHealth technologies**. *J Med Internet Res* (2011.0) 13. DOI: 10.2196/jmir.1672
28. 28Monitoring and evaluating digital health interventions: a practical guide to conducting research and assessment [Internet]. Geneva: World Health Organization; 2016. Available from: http://apps.who.int/bookorders.. *Monitoring and evaluating digital health interventions: a practical guide to conducting research and assessment* (2016.0)
29. Eldridge SM, Lancaster GA, Campbell MJ, Thabane L, Hopewell S, Coleman CL. **Defining feasibility and pilot studies in preparation for randomised controlled trials: development of a conceptual framework**. *PLoS One* (2016.0) **11** e0150205. DOI: 10.1371/journal.pone.0150205
30. Karimi N, Crawford D, Opie R, Maddison R, O’Connell S, Hamblin PS. **EatSmart, a web-based and mobile healthy eating intervention for disadvantaged people with type 2 diabetes: Protocol for a pilot mixed methods intervention study**. *JMIR Res Protoc* (2020.0) **9** e19488. DOI: 10.2196/19488
31. Hilliard ME, Eshtehardi SS, Minard CG, Saber R, Thompson D, Karaviti LP. **Strengths-based behavioral intervention for parents of adolescents with type 1 diabetes using an mHealth app (type 1 doing well): Protocol for a Pilot Randomized Controlled trial.**. *JMIR Res Protoc.* (2018.0) **7** e9147. DOI: 10.2196/resprot.9147
32. Sanci L, Kauer S, Thuraisingam S, Davidson S, Duncan AM, Chondros P. **Effectiveness of a mental health service navigation website (link) for young adults: randomized controlled trial.**. *JMIR Ment Health.* (2019.0) **6** e13189. DOI: 10.2196/13189
33. Sterne JAC, Hernán MA, Reeves BC, Savović J, Berkman ND, Viswanathan M. **ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions**. *bmj* (2016.0) 355. DOI: 10.1136/bmj.i4919
34. 34Australian Bureau of Statistics. Geelong: 2016 Census All persons QuickStats [Internet].
2016 [cited 2022 Jul 29]. Available from: https://www.abs.gov.au/census/find-census-data/quickstats/2016/203. *Geelong: 2016 Census All persons QuickStats [Internet].* (2016.0)
35. Fogg BJ. **Persuasive technology: using computers to change what we think and do.**. *Ubiquity* (2002.0) **2002** 5
36. Hallsworth M, Snijders V, Burd H, Prestt J, Judah G, Huf S. *Applying behavioral insights: simple ways to improve health outcomes* (2016.0)
37. 37Curve Tomorrow. Curve [Internet]. 2022 [cited 2022 Jul 29]. Available from: https://www.curvetomorrow.com.au/
38. 38American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. 3rd ed. Arlington, VA: American Psychiatric Publishing; 2013.. *Diagnostic and Statistical Manual of Mental Disorders* (2013.0)
39. Goodman R.. *J Am Acad Child Adolesc Psychiatry* (2001.0) **40** 1337-45. PMID: 11699809
40. Goodman R.. **The extended version of the Strengths and Difficulties Questionnaire as a guide to child psychiatric caseness and consequent burden**. *Journal of child psychology and psychiatry* (1999.0) **40** 791-9. PMID: 10433412
41. Antony MM, Bieling PJ, Cox BJ, Enns MW, Swinson RP. **Psychometric properties of the 42-item and 21-item versions of the Depression Anxiety Stress Scales in clinical groups and a community sample.**. *Psychol Assess* (1998.0) **10** 176
42. Crawford J, Cayley C, Lovibond PF, Wilson PH, Hartley C. **Percentile norms and accompanying interval estimates from an Australian general adult population sample for self‐report mood scales (BAI, BDI, CRSD, CES‐D, DASS, DASS‐21, STAI‐X, STAI‐Y, SRDS, and SRAS).**. *Aust Psychol* (2011.0) **46** 3-14
43. Broglio K.. **Randomization in clinical trials: permuted blocks and stratification**. *JAMA* (2018.0) **319** 2223-4. DOI: 10.1001/jama.2018.6360
44. Eysenbach G. **CONSORT-EHEALTH: improving and standardizing evaluation reports of Web-based and mobile health interventions**. *J Med Internet Res* (2011.0) **13**. DOI: 10.2196/jmir.1923
45. Elo S, Kyngäs H. **The qualitative content analysis process.**. *J Adv Nurs* (2008.0) **62** 107-15. DOI: 10.1111/j.1365-2648.2007.04569.x
46. Braun V, Clarke V. **Using thematic analysis in psychology.**. *Qual Res Psychol* (2006.0) **3** 77-101
|
---
title: Are portable ankle brachial pressure index measurement devices suitable for
hypertension screening?
authors:
- Justyna Janus
- Jennifer K. Nicholls
- Edward Pallett
- Matthew Bown
- Emma M. L. Chung
journal: PLOS ONE
year: 2023
pmcid: PMC10030014
doi: 10.1371/journal.pone.0283281
license: CC BY 4.0
---
# Are portable ankle brachial pressure index measurement devices suitable for hypertension screening?
## Abstract
### Objective
In a large-scale population cardiovascular screening programme, peripheral artery disease (PAD) and hypertension would ideally be rapidly assessed using a single device. The ankle-brachial pressure index (ABPI) is calculated by comparing the ankle and brachial blood pressure (BP). However, it is currently unclear whether brachial BP measurements provided by automated PAD screening systems are sufficiently accurate for simultaneous hypertension screening.
### Methods
Two portable PAD screening devices, the MESI ABPI MD and Huntleigh’s Dopplex ABIlity, were evaluated following the European Society of Hypertension International Protocol (ESH-IP) Revision 2010 using a mercury-free sphygmomanometer as a reference device.
### Results
On average, the MESI slightly underestimated brachial systolic blood pressure (BP) with a bias and standard deviation (SD) of -3.5 (SD: 3.3) mmHg and diastolic BP with a bias of -1.5 (SD: 2.3) mmHg. For systolic BP estimates, the Dopplex was more accurate than the MESI with a lower bias of -0.5 (SD: 4.2) mmHg but less precise. The MESI successfully fulfilled all the requirements of the ESH-IP for hypertension screening. The Dopplex device failed the ESH-IP due to the absence of DBP measurements.
### Conclusions
The MESI device appears to be suitable for simultaneous PAD and hypertension screening as part of a preventative care programme. Huntleigh’s Dopplex ABIlity failed to pass the ESH-IP validation test. Further clinical trials are underway to assess the use of the MESI for simultaneous screening for hypertension and PAD in a population screening setting.
## Introduction
Peripheral arterial disease (PAD) is a chronic disease resulting from the narrowing of the arteries in the legs due to atherosclerosis [1]. PAD is underdiagnosed worldwide, with at least $50\%$ of PAD patients being asymptomatic [2, 3]. PAD is often associated with hypertension, so early diagnosis of PAD, and better management of high blood pressure (HBP) can improve preventive cardiovascular care, reducing the burden on healthcare providers.
To detect PAD, systolic BP is measured at the posterior tibial artery of the ankle and compared to the systolic BP measured at the brachial artery in the upper arm; the ratio of these measurements is known as the ankle-brachial pressure index (ABPI). If the ABPI is abnormal (<0.9), then the presence of PAD is indicated [4, 5]. Traditionally, ABPI is measured using a handheld continuous-wave Doppler instrument [6] to guide the user in accurately determining the systolic BP. However, this method is time consuming and requires a skilled operator [7].
The recent introduction of automated PAD screening devices means that ABPI measurements can be completed faster than measurements using traditional hand-held Doppler, after minimal user training [8]. This may make it economically feasible to add PAD screening to existing national screening programmes, such as the abdominal aortic aneurysm (AAA) screening programme, which is one of 11 population screening programmes offered in the United Kingdom (UK).
As automated PAD devices use Blood Pressure measurements to estimate the ABPI, automated PAD screening devices could potentially be used to simultaneously check for high BP. However, as separate validated Blood Pressure monitoring devices are readily available, portable PAD screening devices developed for ABPI measurements are generally not licensed or approved for identifying high BP. Consequently, it is unclear whether PAD devices are suitable for identifying high BP, and the accuracy of PAD screening devices for hypertension screening has yet to be assessed.
In this study, two automated PAD devices are assessed for suitability and accuracy for detecting hypertension. The results of this study will inform the design of a combined AAA-PAD-HBP population screening test to be offered to all men in the UK at the age of 65. When designing and evaluating the cost effectiveness of large-scale population screening programmes, it is essential to keep the screening test as short as possible and to minimise equipment costs. If a single PAD device is suitable for combined PAD and hypertension screening, this would be more time and cost effective than conducting separate blood pressure and PAD measurements using different devices.
## PAD devices
Two brand new (box fresh) automated PAD detection devices, with valid manufacturer calibration certificates, were tested in this study (S1 Fig and S1 Table). Systems were tested within 3 months of delivery and were purchased in June 2021. Throughout this manuscript, these are referred to as `the MESI’ (MESI ABPI MD system, MESI Ltd., Slovenia, EU), and `the Dopplex’ (Huntleigh Dopplex ABIlity Automatic ABPI system, Huntleigh Healthcare Ltd., Cardiff, Wales).
These devices were purchased as part of a larger UK screening trial [9]; “Peripheral arterial disease, High blood pressure and Aneurysm Screening Trial” (PHAST) designed to test the feasibility of adding combined PAD and hypertension screening to the UK’s existing abdominal aortic aneurysm screening programme.
Both devices were validated for hypertension screening following the European Society of Hypertension International Protocol (ESH-IP, Revision 2010) [10]. Neither of these two PAD devices had previously been evaluated for hypertension screening.
## Nissei DM-3000
A mercury-free sphygmomanometer, Nissei DM-3000 (Nissei DM-3000, Nissei Japan Precision Instruments, Gunma, Japan), acted as a reference device due to the phasing out of the usage of mercury devices within the UK National Health Service (NHS) due to environmental concerns [11]. The Nissei DM-3000 is a validated BP measurement system [10]. It has been previously found to comfortably pass all ESH-IP validation requirements with a similar level of accuracy as a mercury sphygmomanometer, confirming its use as a reliable alternative reference device [12].
The Nissei has two modes; in the automated oscillometric mode, inflation and deflation of the cuffs is fully automated [12]. The automated oscillometric BP mode was selected with the deflation rate set to 2.5 mmHg per second to reduce any variability associated with manual measurements.
The Nissei device has a liquid crystal display, which displays systolic BP, diastolic BP and pulse rate. Two cuff sizes were available for use: standard and large. The University Hospitals of Leicester Clinical Engineering Scientific Services team regularly checked the device’s accuracy using a calibrated pressure meter [13].
The manufacturer’s instructions for the PAD and reference devices were strictly followed to eliminate factors that could impact measurement accuracy [13–15].
## Participants and recruitment
Participants were recruited according to a protocol approved by the University of Leicester Medicine and Biological Sciences Research Ethics Committee and following the Declaration of Helsinki [2013] [16]. All participants provided written informed consent. Each validation study required 33 participants aged over 25, comprising at least 10 men and 10 women. Participants with a history (or family history) of venous thromboembolic disease, limb ulceration, Parkinson’s disease, severe PAD, lymphedema, or clinical evidence of cellulitis were excluded. Participants were excluded if they could not remain still or lie flat, or reported any condition preventing both arms from being measured. Participants taking medication for pulmonary hypertension were included in this study.
Healthy volunteers were asked to refrain from consuming caffeine or nicotine for at least 1 hour before their appointment to avoid transient BP changes. Participants were also asked to avoid vigorous exercise an hour before the study and to wear loose clothing to allow access to their lower limbs and upper arms. The age and sex of the participants were recorded. The correct cuff size was selected by measuring the circumference of the arm. Socks and shoes were removed before measurements. Cuffs were then applied according to the device manufacturer’s instructions. If no suitable cuff size was available, participants were excluded from the study.
## Study protocol
Each participant lay supine on a couch with their back straight and legs uncrossed, resting for 10–15 minutes before the first measurement. Care was taken to ensure that the participant’s heels rested fully on the couch, as placing weight on the calf may affect measurement results. Subjects were asked to remain still and avoid talking. Phones and other devices were removed to avoid interruptions [10, 17].
The validation team included two individuals (an observer and a supervisor) trained in taking BP measurements. The same individuals acted as observer and supervisor for all observations. The accuracy of each PAD device (MESI or Dopplex) was estimated based on comparing BP readings with those from a reference sphygmomanometer (Nissei). Two entry BP measurements were obtained to determine the participant’s suitability for inclusion in the study. One measurement using the reference device (left arm) and one measurement using one of the two automated PAD devices being trialled. For the Nissei measurements, the lower end of the cuff was placed 2 cm above the antecubital fossa and tightness was assessed by placing a finger between the arm and the cuff; two fingers should be able to fit but would be snug. The cuff was inflated to 180 mmHg, and the BP was recorded.
For the MESI device, three cuffs were placed on the participant. One cuff was placed on the left upper arm 2 cm above the antecubital fossa, so the cuff was lined up against the brachial artery, and the other two cuffs were placed 2 cm above the ankle, lined up against the dorsalis pedis artery.
For the Dopplex device, each upper arm chamber was secured in the same way as for the standard sphygmomanometer, with the additional lower chamber attached below the elbow, on the forearm. Using the lower chamber, leg cuffs were secured above the ankle and around the foot. Arrows on all cuffs were pointed upwards to determine the correct orientation of each cuff.
Depending on the entry measurement results from the sphygmomanometer and the device, and the required BP range, participants were included or excluded in the study. If participants were included, the left arm cuff was switched between the sphygmomanometer and the trialled device. Seven BP measurements were taken, alternating between the reference device (four times) and MESI or Dopplex device (three times).
The interval between BP measurements was at least 30–60 seconds to avoid congestion, but no longer than 60 seconds apart as natural variations in BP are likely to occur over extended periods [10]. The total time for obtaining all seven measurements was approximately 60 minutes.
Mean values obtained from the reference device were used to classify each subject’s systolic and diastolic BP as low, medium, or high (S2 Table). Subjects were excluded if the test and reference device failed to record a measurement after three successive attempts. If high BP was identified and the subject was unaware, a consultation with their general practitioner was advised.
## Data analysis
BP readings were analysed as outlined by the ESH-IP validation protocol [10]. This involved comparing the test device BP measurement with measurements made before and after the test device using the reference device. The smallest difference between the reference and the test device was taken forward for further analysis. This resulted in three pairs of reference and test device readings relating to the systolic BP. In the case of the MESI, the diastolic BP of each participant also resulted in three pairs of readings. Each of these 6 readings was classified into 4 groups; within 0–5 mmHg, 6–10 mmHg, 11–15 mmHg, or >15 mmHg of the reference reading [10].
## Statistical analysis
Statistical analysis was performed with GraphPad Prism 9.1.2 Software (GraphPad Software, Inc., San Diego, CA). Continuous parameters were checked for normality and are reported as a mean and Standard Deviation (SD). Bland-*Altman analysis* of the systolic and diastolic BP readings were used to estimate the bias and $95\%$ limits of agreement of the test device relative to the reference device. The relationship between the test and reference measurements was summarised by fitting a straight line using simple linear regression. Pearson’s coefficient of correlation (R2) was calculated and considered to indicate a high correlation for values >0.9.
## Results
Thirty-five participants were screened using the MESI, and thirty-four were screened using the Dopplex device. Two participants screened using the MESI device and one using the Dopplex device were excluded. For the MESI device, one of the participants was excluded due to persistent inflation errors. The other participant was excluded due to the absence of displayed values after inflation. For the Dopplex device, one participant was excluded due to absent values for all 4 limbs. This resulted in a total of 33 participants with readings suitable for further analysis (S3 Table). Although studies for the 2 devices were conducted separately, 30 participants volunteered for both studies, and demographic characteristics were almost identical for evaluation of both test devices (Table 1).
**Table 1**
| Unnamed: 0 | MESI ABPI | Dopplex ABIlity |
| --- | --- | --- |
| Male: Female | 18: 15 | 18: 15 |
| Mean Age (SD, range) | 55 (19, 25: 87)a | 54 (20, 25: 87)a |
| Mean Arm circumference (cm) (SD, range) | 30.0 (3.4, 24: 38) | 30.0 (3.4, 24: 38) |
| Cuff for the test device (Standard) | 29 | 29 |
| Cuff for the test device (Large) | 4 | 4 |
## Device agreement
Ninety-nine systolic BP and diastolic BP MESI measurements (3 measurements for each of the 33 subjects), and 99 Dopplex systolic BP measurements, were available for further analysis. Table 2 compares mean BP values for the MESI and Dopplex with the corresponding Nissei reference value. The number of measurements differed from the Nissei reference by 5, 10 and 15 mmHg for systolic and diastolic BP, according to the ESH-IP, are summarised in Table 3 (MESI) and Table 4 (Dopplex). Based on these measurements, the MESI successfully passed part 1 of ESH-IP, Table 3. The Dopplex device passed the systolic BP requirements but failed part 1 due to a lack of diastolic BP readings, Table 4.
The MESI passed part 2 of the ESH-IP requirements (Table 3). For the Dopplex, $\frac{21}{33}$ participants had a minimum of 2 out of 3 measurements within 5 mmHg of the reference device. This was below the required target (Table 4). Therefore, the device failed this part of the ESH-IP protocol.
Part 3 of the ESH-IP combines the outcomes from parts 1 and 2 of the protocol. All of the requirements were satisfied for the MESI device (Table 3). The Dopplex device failed part 3 of the ESH-IP as it did not fulfil the requirements of either part 1 or part 2 (Table 4).
Linear correlation analysis assessed the association between test devices and reference measurements. As expected, systolic and diastolic BP estimates from the PAD and reference device were strongly correlated. Bland-*Altman analysis* revealed that the MESI underestimated BP with a mean bias (SD) of -3.5 (SD: 3.3) mmHg for systolic BP measurements and -1.5 (SD: 2.3) mmHg for diastolic BP, respectively (Fig 1). Bland-*Altman analysis* for the Dopplex device showed a mean bias of 0.5 (SD: 4.2) mmHg for systolic BP measurements (Fig 2).
**Fig 1:** *Agreement between the MESI and Nissei reference data, based on 33 pairs of systolic BP (A and B) and diastolic BP (C and D) measurements.The MESI tended to underestimate SBP by -3.5 (SD: 3.3) mmHg (95% LoA: -10.0, 3.0) and DBP by -1.5 (SD: 2.3) mmHg (95% LoA: −5.9, 3.0). The solid red line in A and C represents a linear fit to the measured data (circles), compared to the line of perfect agreement (black line).* **Fig 2:** *Agreement between the Dopplex and Nissei reference data, based on 33 pairs of SBP measurements.The Dopplex ABIlity was in close agreement with the Nissei reference device, with a bias of -0.53 (SD: 4.2) mmHg, but measurements were more variable with a wider 95% LoA ranging from −8.7 to 7.6 mmHg.*
## Discussion
This study is the first to validate the suitability of automated PAD devices for simultaneous PAD and hypertension screening. Both automated PAD devices trialled in this study have previously been reported to have high accuracy for PAD [14, 18]. Although PAD devices available on the market offer BP measurements, they are not licenced or validated for hypertension screening.
This study explored the suitability of PAD devices for hypertension screening by following the ESH-IP validation protocol. A mercury-free Nissei sphygmomanometer was used as a calibrated reference device which has a similar level of accuracy as that of a standard mercury sphygmomanometer.
Huntleigh’s Dopplex ABIlity did not meet the requirements of the ESH-IP as it failed to pass parts 1 and 2 for its SBP measurements and subsequently failed part 3. This device does not provide DBP measurements, preventing it from passing the accuracy criteria for DBP.
The MESI met the requirements of the ESH-IP. Bland-*Altman analysis* showed that the device slightly underestimates BP with a bias of -3.5 (SD: 3.3) mmHg ($95\%$ LoA: −10.0, 3.0) for systolic BP and bias of -1.5 ± 2.3 mmHg ($95\%$ LoA: −5.9, 3.0) for diastolic BP. The MESI BP data closely correlated with reference device estimates (SBP: R2 = 0.94 and DBP: R2 = 0.92, respectively). The MESI would be a suitable device for simultaneous PAD and hypertension screening in a national screening programme as an alternative to performing separate ABPI and hypertension screening tests.
## Limitations and future work
Our study had several limitations. Firstly, only one MESI device and one Dopplex ABIlity device were tested, which may not represent the performance of these devices overall. Two identical devices from the same manufacturer could potentially exhibit differences in readings. However, differences between devices would not impact our main finding that the Dopplex system would not pass the ESH-IP protocol. The next phase of this clinical trial involves assessing the suitability and accuracy of PAD and HBP screening when using multiple MESI devices in a population-based screening setting.
Since this was a healthy volunteer study, few participants exhibited raised BP. Most participants with clinically diagnosed hypertension were taking antihypertensive medication to control their BP. Further clinical trials using patients being screened for hypertension would be valuable for assessing the accuracy of this device in the very high BP range. Our findings are, therefore, only valid for medium to low BP values. Finally, the ban on mercury sphygmomanometers meant we had to use a mercury-free reference standard (the Nissei DM-3000). Using this device deviates from the ESH-IP validation protocol, however, previous researchers have suggested the Nissei to be a reliable, mercury-free alternative [12]. Two Nissei devices were used for this study. To confirm their accuracy, they were frequently compared and calibrated by the University Hospitals of Leicester NHS Trust Clinical Engineering Scientific Services team.
## Conclusion
The MESI ABPI MD device was sufficiently accurate for use in hypertension screening, according to the ESH-IP validation protocol. Huntleigh’s Dopplex ABIlity failed to pass the ESH-IP validation and is not currently suitable for hypertension screening.
## References
1. Brevetti G, Giugliano G, Brevetti L, Hiatt WR. **Inflammation in Peripheral Artery Disease.**. *Circulation* (2010.0) **122** 1862-75. DOI: 10.1161/CIRCULATIONAHA.109.918417
2. Høyer C, Sandermann J, Petersen LJ. **The toe-brachial index in the diagnosis of peripheral arterial disease**. *Journal of Vascular Surgery* (2013.0) **58** 231-8. DOI: 10.1016/j.jvs.2013.03.044
3. Hur KY, Jun JE, Choi YJ, Lee YH, Kim DJ, Park SW. **Color Doppler Ultrasonography Is a Useful Tool for Diagnosis of Peripheral Artery Disease in Type 2 Diabetes Mellitus Patients with Ankle-Brachial Index 0.91 to 1.40**. *Diabetes & metabolism journal* (2018.0) **42** 63-73. DOI: 10.4093/dmj.2018.42.1.63
4. Aboyans V, Criqui MH, Abraham P, Allison MA, Creager MA, Diehm C. **Measurement and interpretation of the ankle-brachial index: a scientific statement from the American Heart Association**. *Circulation* (2012.0) **126** 2890-909. DOI: 10.1161/CIR.0b013e318276fbcb
5. Casey S, Lanting S, Oldmeadow C, Chuter V. **The reliability of the ankle brachial index: a systematic review**. *Journal of foot and ankle research* (2019.0) 12. PMID: 30809275
6. Jayaraj A, Blomberg J, Kohler T. **Comparison of Automated Oscillometric Measurement of Ankle Brachial Index with Standard Doppler Measurement as a Screening Tool for Peripheral Artery Disease.**. *Journal for Vascular Ultrasound* (2013.0) **37** 71-5
7. Scott J, Lecouturier J, Rousseau N, Stansby G, Sims A, Wilson L. **Nurses’ and patients’ experiences and preferences of the ankle-brachial pressure index and multi-site photoplethysmography for the diagnosis of peripheral arterial disease: A qualitative study**. *PloS one* (2019.0) **14** e0224546. DOI: 10.1371/journal.pone.0224546
8. Davies JH, Williams EM. **Automated plethysmographic measurement of the ankle-brachial index: a comparison with the doppler ultrasound method**. *Hypertens Res* (2016.0) **39** 100-6. DOI: 10.1038/hr.2015.124
9. Watson EL, Patel B, Katsogridakis E, Pepper CJ, Messeder SJ, Saratzis A. **Selecting Portable Ankle/Toe Brachial Pressure Index Systems for a Peripheral Arterial Disease Population Screening Programme: a Systematic Review, Clinical Evaluation Exercise, and Consensus Process**. *Eur J Vasc Endovasc Surg* (2022.0) **64** 693-702. DOI: 10.1016/j.ejvs.2022.08.008
10. O’Brien E, Atkins N, Stergiou G, Karpettas N, Parati G, Asmar R. **European Society of Hypertension International Protocol revision 2010 for the validation of blood pressure measuring devices in adults**. *Blood Press Monit* (2010.0) **15** 23-38. DOI: 10.1097/MBP.0b013e3283360e98
11. O’Brien E.. **Replacing the mercury sphygmomanometer. Requires clinicians to demand better automated devices**. *BMJ (Clinical research ed).* (2000.0) **320** 815-6. DOI: 10.1136/bmj.320.7238.815
12. Tasker F, De Greeff A, Shennan AH. **Development and validation of a blinded hybrid device according to the European Hypertension Society protocol: Nissei DM-3000.**. *J Hum Hypertens* (2010.0) **24** 609-16. DOI: 10.1038/jhh.2009.113
13. 13Nissei. Mercury column image digital sphygmomanometer DM-3000 instruction manual Medaval: Nissei; [Available from: https://medaval.ie/device/nissei-dm-3000/.
14. 14Huntleigh. HUNTLEIGH Dopplex ABIlity INSTRUCTIONS FOR USE [User Manual]. Huntleigh Healthcare Ltd. 2011 [Available from: https://irp-cdn.multiscreensite.com/a7c96b08/files/uploaded/771438EN-11_Dopplex_Ability_Instructions_for_Use-English.pdf
15. 15MESI. MESI ABPI MD INSTRUCTIONS FOR USE [User manual]. MESI; 2017 [Available from: https://www.mesimedical.com/app/uploads/2019/11/ABPIMD_IFU_ENG_v7-2_2019-07-05_web.pdf.
16. **World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects**. *Jama* (2013.0) **310** 2191-4. DOI: 10.1001/jama.2013.281053
17. O’Brien E, Pickering T, Asmar R, Myers M, Parati G, Staessen J. **Working Group on Blood Pressure Monitoring of the European Society of Hypertension International Protocol for validation of blood pressure measuring devices in adults**. *Blood Press Monit* (2002.0) **7** 3-17. DOI: 10.1097/00126097-200202000-00002
18. Špan M, Geršak G, Millasseau SC, Meža M, Košir A. **Detection of peripheral arterial disease with an improved automated device: comparison of a new oscillometric device and the standard Doppler method.**. *Vascular health and risk management* (2016.0) **12** 305-11. DOI: 10.2147/VHRM.S106534
|
---
title: A retrospective study of risk factors, causative micro-organisms and healthcare
resources consumption associated with prosthetic joint infections (PJI) using the
Clinical Practice Research Datalink (CPRD) Aurum database
authors:
- Stefano Perni
- Bsmah Bojan
- Polina Prokopovich
journal: PLOS ONE
year: 2023
pmcid: PMC10030031
doi: 10.1371/journal.pone.0282709
license: CC BY 4.0
---
# A retrospective study of risk factors, causative micro-organisms and healthcare resources consumption associated with prosthetic joint infections (PJI) using the Clinical Practice Research Datalink (CPRD) Aurum database
## Abstract
### Background
Prosthetic joint infection (PJI) is a serious complication after joint replacement surgery and it is associated with risk of mortality and morbidity along with high direct costs.
### Methods
The Clinical Practice Research Datalink (CPRD) data were utilized to quantify PJI incidence after hip or knee replacement up to 5 years after implant and a variety of risk factors related to patient characteristics, medical and treatment history along with characteristics of the original surgery were analyzed through Cox proportional hazard.
### Results
221,826 patients (individual joints 283,789) met all the inclusion and exclusion criteria of the study; during the study follow-up period (5 years), 707 and 695 PJIs were diagnosed in hip and knee, respectively. Patients undergoing joint replacement surgery during an unscheduled hospitalization had greater risk of PJI than patients whose surgery was elective; similarly, the risk of developing PJI after a secondary hip or knee replacement was about 4 times greater than after primary arthroplasty when adjusted for all other variables considered. A previous diagnosis of PJI, even in a different joint, increased the risk of a further PJI. Distribution of average LoS per each hospitalization caused by PJI exhibited a right skewed profile with median duration [IQR] duration of 16 days [8–32] and 13 days [7.25–32] for hip and knee, respectively. PJIs causative micro-organisms were dependent on the time between initial surgery and infection offset; early PJI were more likely to be multispecies than later (years after surgery); the identification of Gram- pathogens decreased with increasing post-surgery follow-up.
### Conclusions
This study offers a contemporary assessment of the budgetary and capacity (number and duration of hospitalizations along with the number of Accident and Emergency (A&E) visits) posed by PJIs in UK for the national healthcare system (NHS). The results to provide risk management and planning tools to health providers and policy makers in order to fully assess technologies aimed at controlling and preventing PJI. The findings add to the existing evidence-based knowledge surrounding the epidemiology and burden of PJI by quantifying patterns of PJI in patients with a relatively broad set of prevalent comorbidities.
## 1 Introduction
Total joint arthroplasty (TJA) has been reported to be one of the widely successful treatments for damaged hip or knee joints [1, 2]. However, TJA failure are observed as consequence of periprosthetic joint infections (PJIs) [3, 4]. The reported rate of PJI is one to two percent after primary TJA despite prevention and management policies [5, 6]. PJI prevention mainly relys on antimicrobial prophylactic therapy either systemically or in situ from antibiotic laden bone cement; Management can involve debridement, removal of the implant and his replacement or, in extreme cases, amputation; furthermore, PJI can also lead to death [7]. Therefore, this devastating complication is associated with repeated hospital admissions, severe pain, long term functional deficit and poor health outcomes along with a significant economic burden [8–10].
Hence, it becomes increasingly important to understand the risk factors, both modifiable and non-modifiable, for PJI incidence and outcomes to optimize medical management in patients, scheduled to undergo TJA, at high risk of periprosthetic infection [11, 12]. Numerous risk factors have been identified for PJI after TJA including obesity [13, 14], diabetes mellitus [15, 16], rheumatoid arthritis (RA) [17], urinary tract infections [18], operation time [14]. However, not all studies have demonstrated a similar association between these variables and PJI [19] and, generalization of the conclusions obtained from patients of a certain country may not be applicable to another [20–24].
Contemporary UK specific incidence rates, specific treatment pathways and healthcare related resources consumption have not been well established yet. This information is critical in the assessment of new approaches to prevention and management of PJI thus improving both patient satisfaction post procedure and return on the limited resources of the national healthcare system (NHS).
In this study, our purpose was to identify nationally representative, and current, risk factors for developing PJI after total hip or knee arthroplasty from a large pool of potential covariates that covers type and characteristics of the joint replacement surgery along with patients’ demographic, medical and treatment history. Clinical Practice Research Datalink (CPRD) [25–27] with linkage with Hospital episode statistics (HES) offers the opportunity to explore a wider range of the risk factors for PJI than those recorded in only secondary care [28–30]. Moreover, we employed these data to analyze the reported causative micro-organisms of PJI and to determine the present-day burden posed by PJIs on the NHS in terms of overall cost, length of stay and number of hospital admissions and Accident and Emergency (A&E) visits.
## 2.1 Data sources
Data were obtained from the UK Clinical Practice Research Datalink (CPRD) with linkage to Inpatient and Outpatient Hospital Episode Statistics (HES) secondary care data and Office for National Statistics (ONS) mortality data.
## 2.2 Study design
This was a retrospective study of patients undergoing hip or knee replacement; the index date for inclusion was the day of joint replacement surgery. Patients were included if relevant inclusion were met: Aged at least 30 years at index date and surgery after $\frac{01}{01}$/2007.
## 2.3 Sample size considerations
This research was largely descriptive rather than inferential in nature in that it aimed to characterise pattens of PJI. A sample size and power calculation were, therefore, not undertaken with respect to identifying differences in outcomes between groups. However, the number of patients identified suggests that the analysed population will be large enough to allow for sufficiently robust conclusions to be drawn from the study.
## 2.4 Participants
Patients undergoing hip or knee replacement surgery were identified through OPCS codes for such procedures in the HES database (S1 Table).
Patients were excluded if they meet any of the following criteria: aged <30 years at index date, registered on the CPRD for < 6 months prior to index date and most recent CPRD up-to-standard (UTS) date > 6 months prior to index date. Laterality of the procedure was determined through OPCS code (Z94.2: Right sided operation and Z94.3: Left sided operation); when the code reported a bilateral procedure (Z94.1: Bilateral operation) two separate entries were created, one for the left joint and one for the left. A specific entry was created for each patient, index date, joint (hip or knee) and side (left or right).
Eligible patients were followed-up from index date and their records extracted for all observations up to and including the first occurrence of: joint replaced, death, loss to follow-up or end of study period (5-years after last joint replacement surgery).
Arthroplasty were defined as primary if this specifically stated in the OPCS procedure code, the procedures were categorized as secondary when specifically stated in the OPCS procedure code or if an arthroplasty procedure (S1 Table) was observed in the patient record in the same joint; otherwise the procedure primary or secondary property was classified as “unknown”.
The date of replacement was determined as the date a subsequent arthroplasty procedure was recorded on the same joint. This also corresponded to the index date of a further entry.
Loss to follow-up was be defined as the earliest date a patient was transferred out of the practice or the date that the practice left the CPRD database.
Covariate related to patient characteristics were extracted from the CPRD database; joint replacement surgery properties were derived from the HES database according to OPCS codes (for type of fixation, laterality fixation, grafts…) and admission codes. Medical history was determined by the presence/absence of disease specific MedCodeID codes in the CPRD database or of ICD-10 codes in the HES database before index date.
Prescription history was assessed from relevant BNF codes reported in the CPRD database.
## 2.5 Primary and secondary outcomes
The primary outcome of the study was the determination of risk factors for PJI: PJI occurrence determined through the presence of the ICD-10 code M84.5 in the HES database. In order to ascertain that PJI diagnosed were affecting the joint of interest as patients could have multiple joints replaced, only a diagnosis of PJI and a record any procedure (OPCS) in the joint of interest during the hospitalization were considered (S2 Table).
The date of PJI offset was determined as the hospitalization date corresponding to the first diagnosis of PJI in the joint (hip or knee and left or right). The time to PJI was calculated as difference between index date and date of PJI occurrence.
Secondary outcomes of the study were the quantification of:
## 2.6 Comparison group(s) or controls
No comparison or control groups are specified. Patients will be characterised by levels of demographic, clinical and treatment characteristics.
## 2.7 Ethics
This study protocol (19_009) was reviewed by the Independent Scientific Advisory Committee (ISAC) and received approval in January 2020; a minor amendment was approved in December 2021. The research team had access only to data de-identified before delivery thus approval from an institutional review board was not sought.
## 2.8 Statistical analysis
Descriptive analyses were generated, characterizing patient demographics, clinical and treatment characteristics. Summary statistics (for example, mean, standard deviation, standard error, median, inter-quartile range, minimum, and maximum) were calculated for continuous variables, and number and proportion/percentage for categorical variables. The number and proportion of patients with missing data was also reported for each of the variables of interest.
Where to statistically describe differences between patient subgroups was appropriate, univariate methods were employed. The type of test used was dependent on the type/distribution of the outcome variable. t-test was used for numeric variables and chi-squared tests was be used for (unordered) categorical variables. A p-value < 0.05 was considered statistically significant.
Risk factors for PJI were studies using the Cox proportional hazard model both in its univariate and multivariate form.
LoS and number of hospitalizations were fitted with negative binomial and Poisson distributions, along the zero inflated and zero truncated form, respectively. The final model selection was based on the observation of goodness of fitting parameters such as AIC, BIC and Log-likelihood.
All data collection, analysis and visualization were performed using R (ver 4.0) and relevant packages [32, 33].
## 3 Results
In total 235,249 patients, corresponding to 330,173 joints, were detected in the CPRD database after linkage to HES and ONS who underwent hip or knee replacement with non-missing laterality of the surgery. 221,826 patients (individual joints 283,789) met all the inclusion and exclusion criteria of the study (S1 Table).
Most of the patients completed the observational period of 5 years; the most common reason for patients ($$n = 76$$,800) to be lost to follow-up was the last collection date in the database being earlier than 5 years from surgery. The second and third most common reason for not completing the 5 years follow-up period were patients transferring out of CPRD ($$n = 35$$,831) and death ($$n = 32$$,981), respectively (S4 Table).
## 3.1 Baseline characteristics
Patients in the cohort were predominantly female, with age ranging from 31 to 109 years and a median of 72 years. Majority of patients had a BMI between 26 and 30 with a median BMI record of 28.74 at the time of surgery to replace the hip or knee joint; for about $30\%$ of patients no record of BMI was available (Table 1). Gender distribution was different between the patients who developed PJI and those who did not ($p \leq 0.001$), majority of patients diagnosed with PJI in the joint replaced were male. Age and BMI were also parameters with different distribution in the two cohorts regardless of considering these variable continuous or categorical. Patient developing PJI were generally younger and with higher BMI ($p \leq 0.001$).
**Table 1**
| Variable | All | No PJI | PJI | p value |
| --- | --- | --- | --- | --- |
| Gender | Gender | Gender | Gender | Gender |
| Female | 173,961 (61.30%) | 173,291 (61.37%) | 670 (47.79%) | <0.001 |
| Male | 109,828 (38.70%) | 109,096 (38.63%) | 732 (52.21%) | <0.001 |
| Age | Age | Age | Age | Age |
| ≤ 45 | 4,074 (1.44%) | 4,027 (1.43%) | 47 (3.35%) | <0.001 |
| 46–55 | 17,512 (6.17%) | 17,392 (6.16%) | 120 (8.56%) | <0.001 |
| 56–65 | 54,632 (19.25%) | 54,279 (19.22%) | 353 (25.18%) | <0.001 |
| 66–75 | 96,747 (34.09%) | 96,262 (34.09%) | 485 (34.59%) | <0.001 |
| 76–85 | 81,892 (28.86%) | 81,540 (28.88%) | 352 (25.11%) | <0.001 |
| > 85 | 28,932 (10.19%) | 28,887 (10.23%) | 45 (3.21%) | <0.001 |
| Mean (SD) | 71.88 | 71.90 (10.98) | 68.40 (10.86) | <0.001 |
| Median | 72 | 72 | 69 | <0.001 |
| IQR | 65–80 | 65–80 | 62–76 | <0.001 |
| Min, max | 31, 109 | 31, 109 | 31, 98 | <0.001 |
| BMI | BMI | BMI | BMI | BMI |
| < 20 | 7,694 (2.71%) | 7,675 (2.72%) | 19 (1.36%) | <0.001 |
| 20–25 | 39,004 (13.74%) | 38,873 (13.77%) | 131 (9.34%) | <0.001 |
| 26–30 | 70,293 (24.77%) | 69,987 (24.78%) | 306 (21.83%) | <0.001 |
| 31–35 | 50,075 (17.65%) | 49,776 (17.63%) | 299 (21.33%) | <0.001 |
| 36–50 | 30,963 (10.91%) | 30,736 (10.88%) | 227 (16.19%) | <0.001 |
| > 50 | 698 (0.25%) | 691 (0.24%) | 7 (0.50%) | <0.001 |
| Unknown | 85,062 (29.97%) | 84,649 (29.98%) | 413 (29.46%) | <0.001 |
| Mean (SD) | 29.37 | 29.36 (6.02) | 31.20 (6.28) | <0.001 |
| Median | 28.74 | 28.73 | 30.6 | <0.001 |
| IQR | 25.30–32.86 | 25.30–32.81 | 27.00–34.80 | <0.001 |
| Min, max | 10.46, 99.90 | 10.46, 99.90 | 13.10, 60.80 | <0.001 |
| Smoking status | Smoking status | Smoking status | Smoking status | Smoking status |
| Non-smoker | 153,344 (54.03%) | 152,682 (54.07%) | 662 (47.22%) | <0.001 |
| Cigar | 223 (0.08%) | > 218 | < 5 | <0.001 |
| Current | 12,098 (4.26%) | 12,028 (4.26%) | 70 (4.99%) | <0.001 |
| Ex-smoker | 88,563 (31.21%) | 88,051 (31.18%) | 512 (36.52%) | <0.001 |
| Heavy | 307 (0.11%) | > 302 | < 5 | <0.001 |
| Light | 782 (0.28%) | > 778 | < 5 | <0.001 |
| Moderate | 704 (0.25%) | > 699 | < 5 | <0.001 |
| Quitting | 886 (0.31%) | 881 (0.31%) | 5 (0.36%) | <0.001 |
| Unknown | 26,882 (9.47%) | 26,738 (9.47%) | 144 (10.27%) | <0.001 |
| Alcohol consumption | Alcohol consumption | Alcohol consumption | Alcohol consumption | Alcohol consumption |
| Non-drinker | 4,826 (1.70%) | 4,810 (1.70%) | 16 (1.14%) | 0.09 |
| Ex drinker | 418 (0.15%) | > 413 | < 5 | 0.09 |
| Heavy drinker | 1,580 (0.56%) | 1,569 (0.56%) | 11 (0.78%) | 0.09 |
| Light drinker | 52,731 (18.58%) | 52,446 (18.57%) | 285 (20.33%) | 0.09 |
| Moderate drinker | 9,282 (3.27%) | 9,223 (3.27%) | 59 (4.21%) | 0.09 |
| Social drinker | 4,781 (1.68%) | 4,752 (1.68%) | 29 (2.07%) | 0.09 |
| Very heavy | 561 (0.20%) | > 556 | < 5 | 0.09 |
| Other | 1,732 (0.61%) | 1,723 (0.61%) | 9 (0.64%) | 0.09 |
| Unknown | 207,878 (73.25%) | 206,892 (73.27%) | 986 (70.33%) | 0.09 |
More than half of patients were non-smokers and about $30\%$ had quit the habit at the time of joint replacement surgery; less the $10\%$ had no record regarding their smoking habit. The smoking status was different among patients developing PJI compared to patients not experiencing PJI in the 5 years post implant surgery ($p \leq 0.001$). Patients attitude toward alcohol consumption did not appear to impact the risk of PJI ($p \leq 0.05$) (Table 1).
Similar number of patients underwent total hip or knee replacement, replacement of only the femur head was the least common procedure among those considered and the resulted in the lowest risk of developing PJI ($p \leq 0.001$). The implanted device was fixed with bone cement in $65.3\%$ of patients while $23.4\%$ of devices implanted were uncemented; hybrid fixation was used in $7.9\%$ of procedures; the fixation method distribution among patients who developed PJI was different than those who did not ($p \leq 0.001$). Over $90\%$ of recorded procedures were primary arthroplasty, however PJI were more likely to develop after replacement procedures ($p \leq 0.001$). $82.8\%$ of procedures were elective and bone grafts were recorded in under $2\%$ of surgeries and in knee replacement patellar resurfacing was observed in $14.9\%$ of joints, both surgical characteristics distribution among the cohorts of patients with PJI and without-PJI were statistically different ($p \leq 0.001$) (Table 2).
**Table 2**
| Variable | All | No PJI | PJI | p value |
| --- | --- | --- | --- | --- |
| Joint replaced | Joint replaced | Joint replaced | Joint replaced | Joint replaced |
| Femur | 38,159 (13.45%) | 38,102 (13.49%) | 57 (4.07%) | <0.001 |
| Hip | 124,265 (43.79%) | 123,615 (43.78%) | 650 (46.36%) | <0.001 |
| Knee | 121,365 (42.77%) | 120,670 (42.73%) | 695 (49.57%) | <0.001 |
| Laterality of operation | Laterality of operation | Laterality of operation | Laterality of operation | Laterality of operation |
| Left | 133,269 (46.96%) | 132,578 (46.95%) | 691 (49.29%) | 0.08 |
| Right | 150,520 (53.04%) | 149,809 (53.05%) | 711 (50.71%) | 0.08 |
| Type of fixation | Type of fixation | Type of fixation | Type of fixation | Type of fixation |
| Cemented | 185,253 (65.28%) | 184,327 (65.27%) | 926 (66.05%) | <0.001 |
| Hybrid | 22,349 (7.88%) | 22,261 (7.88%) | 88 (6.28%) | <0.001 |
| Non cemented | 66,507 (23.44%) | 66,202 (23.44%) | 305 (21.75%) | <0.001 |
| Unknown | 9,680 (3.41%) | 9,597 (3.40%) | 83 (5.92%) | <0.001 |
| Primary arthroplasty | Primary arthroplasty | Primary arthroplasty | Primary arthroplasty | Primary arthroplasty |
| Yes | 262,247 (92.41%) | 261,304 (92.53%) | 943 (67.26%) | <0.001 |
| No | 20,817 (7.34%) | 20,371 (7.21%) | 446 (31.81%) | <0.001 |
| Unknown | 725 (0.26%) | 712 (0.25%) | 13 (0.93%) | <0.001 |
| Admission type | Admission type | Admission type | Admission type | Admission type |
| Elective | 234,961 (82.79%) | 233,778 (82.79%) | 1,183 (84.38%) | 0.03 |
| A&E | 48,156 (16.97%) | 47,944 (16.98%) | 212 (15.12%) | 0.03 |
| Unknown | 672 (0.24%) | 665 (0.24%) | 7 (0.50%) | 0.03 |
| Patella resurfacing (* only for knee replacement surgeries) | Patella resurfacing (* only for knee replacement surgeries) | Patella resurfacing (* only for knee replacement surgeries) | Patella resurfacing (* only for knee replacement surgeries) | Patella resurfacing (* only for knee replacement surgeries) |
| No | 103,244 (85.07%) | 102,605 (85.03%) | 639 (91.94%) | <0.001 |
| Yes | 18,121 (14.93%) | 18,065 (14.97%) | 56 (8.06%) | <0.001 |
| Graft | Graft | Graft | Graft | Graft |
| No | 279,681 (98.55%) | 278,322 (98.56%) | 1,359 (96.93%) | <0.001 |
| Autograft | 1,857 (0.65%) | 1,841 (0.65%) | 16 (1.14%) | <0.001 |
| Autograft + other | 76 (0.03%) | > 71 | < 5 | <0.001 |
| Other graft | 2,175 (0.77%) | 2,149 (0.76%) | 26 (1.85%) | <0.001 |
The most common comorbidity observed in the patients cohort was osteoarthritis (OA) ($$n = 199$$,551–$70.32\%$) followed by hypertension ($$n = 157$$,086–$55.35\%$) and cancer ($$n = 135$$,577–$47.77\%$); other diseases affecting more than $10\%$ of the included patients at the time of arthroplasty were: rheumatoid arthritis ($$n = 72$$,090, $25.40\%$), CKD ($$n = 45$$,916–$16.18\%$), Type1 or Type2 diabetes ($$n = 42$$,259–$14.89\%$), anemia (40,062–$14.12\%$) and ischemic heart disease ($$n = 31$$,192–$10.99\%$). Of these, CKD, diabetes, OA and RA frequency on patients affected by PJI was statistically different ($p \leq 0.01$) than those who did not develop PJI. Patients affected by DVT or dementia at the time of joint replacement surgery were more likely to have developed PJI ($p \leq .001$); while osteoporosis at the time of surgery was more common in patients that did not develop PJI ($p \leq 0.001$). The majority of patients had been diagnosed with 2 or 3 of the comorbidities considered in this work at the time of surgery, more than a quarter of the cohort had been diagnosed with 4 or 5 of the diseases considered. The overall number of diagnosed comorbidities did not differ between the patients group developing PJI compared to those who did not ($p \leq 0.05$). A previous diagnosis of PJI before surgery was more common in the patients that developed a further PJI even if the infection was in a different joint ($p \leq 0.001$) (Table 3).
**Table 3**
| Variable | All | No PJI | PJI | p value |
| --- | --- | --- | --- | --- |
| AF | AF | AF | AF | AF |
| No | 258,233 (90.99%) | 256,981 (91.00%) | 1,252 (89.30%) | 0.03 |
| Yes | 25,556 (9.01%) | 25,406 (9.00%) | 150 (10.70%) | 0.03 |
| Liver failure | Liver failure | Liver failure | Liver failure | Liver failure |
| No | 283,035 (99.73%) | 281,638 (99.73%) | 1,397 (99.64%) | 0.69 |
| Yes | 754 (0.27%) | 749 (0.27%) | 5 (0.36%) | 0.69 |
| CKD | CKD | CKD | CKD | CKD |
| No | 237,873 (83.82%) | 236,657 (83.81%) | 1,216 (86.73%) | <0.01 |
| Yes | 45,916 (16.18%) | 45,730 (16.19%) | 186 (13.27%) | <0.01 |
| PE | PE | PE | PE | PE |
| No | 277,125 (97.65%) | 275,767 (97.66%) | 1,358 (96.86%) | 0.06 |
| Yes | 6,664 (2.35%) | 6,620 (2.34%) | 44 (3.14%) | 0.06 |
| DVT | DVT | DVT | DVT | DVT |
| No | 272,228 (95.93%) | 270,911 (95.94%) | 1,317 (93.94%) | <0.001 |
| Yes | 11,561 (4.07%) | 11,476 (4.06%) | 85 (6.06%) | <0.001 |
| Diabetes | Diabetes | Diabetes | Diabetes | Diabetes |
| No | 241,530 (85.11%) | 240,385 (85.13%) | 1,145 (81.67%) | <0.001 |
| Yes | 42,259 (14.89%) | 42,002 (14.87%) | 257 (18.33%) | <0.001 |
| OA | OA | OA | OA | OA |
| No | 84,238 (29.68%) | 83,868 (29.70%) | 370 (26.39%) | <0.01 |
| Yes | 199,551 (70.32%) | 198,519 (70.30%) | 1,032 (73.61%) | <0.01 |
| RA | RA | RA | RA | RA |
| No | 211,699 (74.60%) | 210,702 (74.61%) | 997 (71.11%) | <0.01 |
| Yes | 72,090 (25.40%) | 71,685 (25.39%) | 405 (28.89%) | <0.01 |
| AC | AC | AC | AC | AC |
| No | 148,212 (52.23%) | 147,443 (52.21%) | 769 (54.85%) | 0.05 |
| Yes | 135,577 (47.77%) | 134,944 (47.79%) | 633 (45.15%) | 0.05 |
| HF | HF | HF | HF | HF |
| No | 276,416 (97.40%) | 275,045 (97.40%) | 1,371 (97.79%) | 0.41 |
| Yes | 7,373 (2.60%) | 7,342 (2.60%) | 31 (2.21%) | 0.41 |
| MI | MI | MI | MI | MI |
| No | 256,242 (90.29%) | 255,002 (90.30%) | 1,240 (88.45%) | 0.02 |
| Yes | 27,547 (9.71%) | 27,385 (9.70%) | 162 (11.55%) | 0.02 |
| IHD | IHD | IHD | IHD | IHD |
| No | 252,597 (89.01%) | 251,370 (89.02%) | 1,227 (87.52%) | 0.08 |
| Yes | 31,192 (10.99%) | 31,017 (10.98%) | 175 (12.48%) | 0.08 |
| Hypertension | Hypertension | Hypertension | Hypertension | Hypertension |
| No | 126,703 (44.65%) | 126,102 (44.66%) | 601 (42.87%) | 0.19 |
| Yes | 157,086 (55.35%) | 156,285 (55.34%) | 801 (57.13%) | 0.19 |
| COPD | COPD | COPD | COPD | COPD |
| No | 262,628 (92.54%) | 261,329 (92.54%) | 1,299 (92.65%) | 0.92 |
| Yes | 21,161 (7.46%) | 21,058 (7.46%) | 103 (7.35%) | 0.92 |
| Hemorrhagic stroke | Hemorrhagic stroke | Hemorrhagic stroke | Hemorrhagic stroke | Hemorrhagic stroke |
| No | 277,864 (97.91%) | 276,488 (97.91%) | 1,376 (98.15%) | 0.6 |
| Yes | 5,925 (2.09%) | 5,899 (2.09%) | 26 (1.85%) | 0.6 |
| Ischemic stroke | Ischemic stroke | Ischemic stroke | Ischemic stroke | Ischemic stroke |
| No | 276,570 (97.46%) | 275,198 (97.45%) | 1,372 (97.86%) | 0.38 |
| Yes | 7,219 (2.54%) | 7,189 (2.55%) | 30 (2.14%) | 0.38 |
| Dementia | Dementia | Dementia | Dementia | Dementia |
| No | 273,956 (96.54%) | 272,571 (96.52%) | 1,385 (98.79%) | <0.001 |
| Yes | 9,833 (3.46%) | 9,816 (3.48%) | 17 (1.21%) | <0.001 |
| Thyroidism | Thyroidism | Thyroidism | Thyroidism | Thyroidism |
| No | 281,401 (99.16%) | 280,008 (99.16%) | 1,393 (99.36%) | 0.5 |
| Yes | 2,388 (0.84%) | 2,379 (0.84%) | 9 (0.64%) | 0.5 |
| Anemia | Anemia | Anemia | Anemia | Anemia |
| No | 243,727 (85.88%) | 242,523 (85.88%) | 1,204 (85.88%) | 1 |
| Yes | 40,062 (14.12%) | 39,864 (14.12%) | 198 (14.12%) | 1 |
| Osteoporosis | Osteoporosis | Osteoporosis | Osteoporosis | Osteoporosis |
| No | 259,856 (91.57%) | 258,531 (91.55%) | 1,325 (94.51%) | <0.001 |
| Yes | 23,933 (8.43%) | 23,856 (8.45%) | 77 (5.49%) | <0.001 |
| PJI before in same joint | PJI before in same joint | PJI before in same joint | PJI before in same joint | PJI before in same joint |
| No | 282,212 (99.44%) | 280,936 (99.49%) | 1,276 (91.01%) | <0.001 |
| Yes | 1,577 (0.56%) | 1,451 (0.51%) | 126 (8.99%) | <0.001 |
| PJI before in another joint (* not only hip or knee) | PJI before in another joint (* not only hip or knee) | PJI before in another joint (* not only hip or knee) | PJI before in another joint (* not only hip or knee) | PJI before in another joint (* not only hip or knee) |
| No | 280,664 (98.90%) | 279,402 (98.94%) | 1,262 (90.01%) | <0.001 |
| Yes | 3,125 (1.10%) | 2,985 (1.06%) | 140 (9.99%) | <0.001 |
| Total number of comorbidities | Total number of comorbidities | Total number of comorbidities | Total number of comorbidities | Total number of comorbidities |
| 0–1 | 53,909 (19.00%) | 53,652 (19.00%) | 257 (18.33%) | 0.15 |
| 2–3 | 127,623 (44.97%) | 127,018 (44.98%) | 605 (43.15%) | 0.15 |
| 4–5 | 74,689 (26.32%) | 74,303 (26.31%) | 386 (27.53%) | 0.15 |
| 6–7 | 22,085 (7.78%) | 21,958 (7.78%) | 127 (9.06%) | 0.15 |
| 8 or more | 5,483 (1.63%) | 5,456 (1.63%) | 27 (1.85%) | 0.15 |
The only medications used by more than half of patients at any time before the surgery were antibacterial and NSAID; antifungal drugs were used by about $20\%$ of patients before arthroplasty and steroids were prescribed to about $10\%$ of the full cohort; the most common anticoagulant treatment observed was warfarin. Majority of patients receiving an intra-articular injection in the joint of interest did receive only 1 injection and more likely more than 6 months before surgery; some patients had received over 5 intra-articular injections in the joint of interest. Patients in the cohort had also received intra-articular injections in other joints, if they received these injections, the most common number of procedures undergone was 3 and the last recorded more than 6 months before the index surgery (Table 4).
**Table 4**
| Variable | All | No PJI | PJI | p value |
| --- | --- | --- | --- | --- |
| Steroids injection | Steroids injection | Steroids injection | Steroids injection | Steroids injection |
| No use | 259,053 (91.28%) | 257,798 (91.29%) | 1,255 (89.51%) | 0.11 |
| Use < 3 months | 1,368 (0.48%) | 1,359 (0.48%) | 9 (0.64%) | 0.11 |
| Use 3–6 months | 2,568 (0.90%) | 2,551 (0.90%) | 17 (1.21%) | 0.11 |
| Use > 6 months | 20,800 (7.33%) | 20,679 (7.32%) | 121 (8.63%) | 0.11 |
| Chondroitin glucosamine | Chondroitin glucosamine | Chondroitin glucosamine | Chondroitin glucosamine | Chondroitin glucosamine |
| No use | 280,998 (99.02%) | 279,613 (99.02%) | 1,385 (98.79%) | 0.73 |
| Use < 3 months | 493 (0.17%) | > 488 | < 5 | 0.73 |
| Use 3–6 months | 135 (0.05%) | > 130 | < 5 | 0.73 |
| Use > 6 months | 2,163 (0.76%) | 2,149 (0.76%) | 14 (1.00%) | 0.73 |
| NSAID | NSAID | NSAID | NSAID | NSAID |
| No use | 124,717 (43.95%) | 124,184 (43.98%) | 533 (38.02%) | <0.001 |
| Use < 3 months | 55,388 (19.52%) | 54,966 (19.46%) | 422 (30.10%) | <0.001 |
| Use 3–6 months | 15,509 (5.46%) | 15,421 (5.46%) | 88 (6.28%) | <0.001 |
| Use > 6 months | 88,175 (31.07%) | 87,816 (31.10%) | 359 (25.61%) | <0.001 |
| Methotrexate | Methotrexate | Methotrexate | Methotrexate | Methotrexate |
| No use | 277,084 (97.64%) | 275,730 (97.64%) | 1,354 (96.58%) | 0.04 |
| Use < 3 months | 4,304 (1.52%) | 4,270 (1.51%) | 34 (2.43%) | 0.04 |
| Use 3–6 months | 236 (0.08%) | > 231 | < 5 | 0.04 |
| Use > 6 months | 2,165 (0.76%) | 2,152 (0.76%) | 13 (0.93%) | 0.04 |
| DMARD | DMARD | DMARD | DMARD | DMARD |
| No use | 279,519 (98.50%) | 278,140 (98.50%) | 1,379 (98.36%) | 0.03 |
| Use < 3 months | 2,311 (0.81%) | 2,305 (0.82%) | 6 (0.43%) | 0.03 |
| Use 3–6 months | 231 (0.08%) | < 226 | < 5 | 0.03 |
| Use > 6 months | 1,728 (0.61%) | 1,714 (0.61%) | 14 (1.00%) | 0.03 |
| Antibacterial | Antibacterial | Antibacterial | Antibacterial | Antibacterial |
| No use | 83,025 (29.26%) | 82,630 (29.26%) | 395 (28.17%) | 0.03 |
| Use < 3 months | 54,748 (19.29%) | 54,454 (19.28%) | 294 (20.97%) | 0.03 |
| Use 3–6 months | 24,826 (8.75%) | 24,680 (8.74%) | 146 (10.41%) | 0.03 |
| Use > 6 months | 121,190 (42.70%) | 120,623 (42.72%) | 567 (40.44%) | 0.03 |
| Antifungal | Antifungal | Antifungal | Antifungal | Antifungal |
| No use | 224,414 (79.08%) | 223,345 (79.09%) | 1,069 (76.25%) | 0.02 |
| Use < 3 months | 7,530 (2.65%) | 7,478 (2.65%) | 52 (3.71%) | 0.02 |
| Use 3–6 months | 4,388 (1.55%) | 4,363 (1.55%) | 25 (1.78%) | 0.02 |
| Use > 6 months | 47,457 (16.72%) | 47,201 (16.72%) | 256 (18.26%) | 0.02 |
| DOAC | DOAC | DOAC | DOAC | DOAC |
| No use | 278,363 (98.09%) | 277,001 (98.09%) | 1,362 (97.15%) | 0.01 |
| Use < 3 months | 3,725 (1.31%) | 3,694 (1.31%) | 31 (2.21%) | 0.01 |
| Use 3–6 months | 303 (0.11%) | > 298 | < 5 | 0.01 |
| Use > 6 months | 1,398 (0.49%) | 1,389 (0.49%) | 9 (0.64%) | 0.01 |
| Heparin | Heparin | Heparin | Heparin | |
| No use | 279,872 (98.62%) | 278,494 (98.62%) | 1,378 (98.29%) | 0.06 |
| Use < 3 months | 587 (0.21%) | 582 (0.21%) | 5 (0.36%) | 0.06 |
| Use 3–6 months | 261 (0.09%) | > 256 | < 5 | 0.06 |
| Use > 6 months | 3,069 (1.08%) | 3,054 (1.08%) | 15 (1.07%) | 0.06 |
| Warfarin | Warfarin | Warfarin | Warfarin | Warfarin |
| No use | 265,204 (93.45%) | 263,922 (93.46%) | 1,282 (91.44%) | 0.01 |
| Use < 3 months | 12,094 (4.26%) | 12,021 (4.26%) | 73 (5.21%) | 0.01 |
| Use 3–6 months | 1,081 (0.38%) | 1,073 (0.38%) | 8 (0.57%) | 0.01 |
| Use > 6 months | 5,410 (1.91%) | 5,371 (1.90%) | 39 (2.78%) | 0.01 |
| Intra-articular injections (n) | Intra-articular injections (n) | Intra-articular injections (n) | Intra-articular injections (n) | Intra-articular injections (n) |
| 0 | 271,827 (95.78%) | 270,497 (95.79%) | 1,330 (94.86%) | 0.16 |
| 1 | 8,389 (2.96%) | 8,337 (2.95%) | 52 (3.71%) | 0.16 |
| 2 | 2,197 (0.77%) | 2,188 (0.77%) | 9 (0.64%) | 0.16 |
| 3 | 721 (0.25%) | 716 (0.25%) | 5 (0.36%) | 0.16 |
| 4 | 264 (0.09%) | > 259 | < 5 | 0.16 |
| 5 | 178 (0.06%) | > 173 | < 5 | 0.16 |
| > 5 | 213 (0.08%) | > 208 | < 5 | 0.16 |
| Last intra-articular injection | Last intra-articular injection | Last intra-articular injection | Last intra-articular injection | Last intra-articular injection |
| No use | 271,827 (95.78%) | 270,497 (95.79%) | 1,330 (94.86%) | 0.37 |
| Use < 3 months | 434 (0.15%) | > 429 | < 5 | 0.37 |
| Use 3–6 months | 1,805 (0.64%) | 1,794 (0.64%) | 11 (0.78%) | 0.37 |
| Use > 6 months | 9,723 (3.43%) | 9,664 (3.42%) | 59 (4.21%) | 0.37 |
| Intra-articular injection in other joints (n) | Intra-articular injection in other joints (n) | Intra-articular injection in other joints (n) | Intra-articular injection in other joints (n) | Intra-articular injection in other joints (n) |
| 0 | 249,945 (88.07%) | 248,747 (88.09%) | 1,198 (85.45%) | <0.01 |
| 1 | 294 (0.10%) | > 289 | < 5 | <0.01 |
| 2 | 3,901 (1.37%) | 3,873 (1.37%) | 28 (2.00%) | <0.01 |
| 3 | 9,606 (3.38%) | 9,564 (3.39%) | 42 (3.00%) | <0.01 |
| 4 | 6,541 (2.30%) | 6,497 (2.30%) | 44 (3.14%) | <0.01 |
| 5 | 2,090 (0.74%) | 2,078 (0.74%) | 12 (0.86%) | <0.01 |
| > 5 | 11,412 (4.02%) | 11,335 (4.01%) | 77 (5.49%) | <0.01 |
| Last intra-articular injection in other joints | Last intra-articular injection in other joints | Last intra-articular injection in other joints | Last intra-articular injection in other joints | Last intra-articular injection in other joints |
| No use | 249,945 (88.07%) | 248,747 (88.09%) | 1,198 (85.45%) | <0.01 |
| Use < 3 months | 1,266 (0.45%) | 1,256 (0.44%) | 10 (0.71%) | <0.01 |
| Use 3–6 months | 4,031 (1.42%) | 4,012 (1.42%) | 19 (1.36%) | <0.01 |
| Use > 6 months | 28,547 (10.06%) | 28,372 (10.05%) | 175 (12.48%) | <0.01 |
## 3.2 Primary outcome—PJI risk factors
During the study follow-up period, 707 and 695 PJIs were diagnosed in hip and knee respectively. The cumulative incidence of PJI is reported in the Kaplan-Meier curve in Fig 1a; $0.21\%$ of the considered joints developed PJI in the first 6 months after surgery, $0.30\%$ in the first year, $0.41\%$ in the first 2 years, $0.54\%$ in the first 4 years and $0.58\%$ in the first 5 years. Therefore, about of a third of the recorded PJI were diagnosed in the first 6 months and half in the first year. Examples of Kaplan- Meier estimates are presented in Fig 1; older patients hazard of developing PJI was lower than younger ones (log-rank $p \leq 0.001$); similarly, male patients developed PJI at a higher rate than female (log-rank $p \leq 0.001$) (Fig 1b and 1c). The rate of PJI development in total hip and knee joints was not statistically different (log-rank $p \leq 0.05$), but greater than when only femur head was replaced (log-rank $p \leq 0.001$) (Fig 1d). Joint after primary replacement had a lower rate of PJI development than after revision (log-rank $p \leq 0.001$) or surgeries where the primary or replacement characteristic could not be ascertained (Fig 1e). In joints with an unknown fixation method the rate of PJI development was greater than any of method of fixation (log-rank $p \leq 0.001$) (Fig 1f).
**Fig 1:** *Example of Kaplan-Meier curves of cumulative risk of PJI for the entire cohort (a) and stratified based on age (b), gender (c), joint (d), primary/replacement surgery (e) and type of fixation (f).*
When patients’ risk of developing PJI was fitted with the Cox regression proportional hazard model, male patients had a greater risk of developing PJI than female, also the older the patient at the time of surgery the lower the risk of been diagnosed with a PJI after the procedure. Other patients baseline characteristics such as BMI were not a risk factor for PJI when adjusted for other variables (Fig 2 and S1 Fig).
**Fig 2:** *Forest plot of Hazard ratios (HR) of multivariate Cox proportional regression model for variables related to patient characteristics (a), arthroplasty surgery (b), medical history (c) and drug history (d).*
Patients undergoing joint replacement surgery during a hospitalization after been admitted to A&E had greater risk of PJI than patients whose surgery was elective; similarly, the risk of developing PJI after a secondary hip or knee replacement was about 4 times greater than after primary arthroplasty when adjusted for all other variables considered. Patellar resurfacing was associated to a reduced the risk of PJI after knee replacement while the type of fixation method did not have a statistically significant impact on the adjusted hazard rate estimated by the Cox regression model (Fig 2 and S1 Fig).
Numerous comorbidities, such as AF, RA, PE and DVT, were risk factors for PJI following hip or knee replacement when the considered individually in the Cox regression (S1 Fig); however only diabetes and hypertension remained statistically significant in the fully adjusted model (Fig 2). A previous diagnosis of PJI, even in a different joint, increased the risk of a further PJI.
Prescription of DOACs or antifungal in the 3 months before joint replacement surgery increased the risk of patients developing PJI while a prescription of DMARDs in the same period reduced the risk of PJI. Intra-articular injections before implant surgery or prescription of other anticoagulant drugs, such as warfarin and heparin, did not have a statistically significant impact on the adjusted hazard ratio estimated by the Cox model (Fig 2 and S1 Fig).
## 3.3.1 PJI health resource consumption and outcomes
Distribution of average LoS per each hospitalization caused by PJI exhibited a right skewed profile with mean values of 26.5 and 18.4 days for hip and knee respectively; median [IQR] duration were 16 [8–32] and 13 [7.25–32] for hip and knee respectively.
The average length of stay (LoS) of each hospitalization following a PJI diagnosis after index surgery was modelled with Poisson and negative binomial distributions and their relative zero inflated version. The fitting of the data with the zero inflated version of the distributions was better than the plain formulation of the distribution as described by greater Log-likelihood and lower AIC and BIC coefficients (S5 Table); negative binomial zero inflated gave the overall better performance.
In hospitalizations related to PJI diagnosis, male patients had on average a LoS $10\%$ shorter than women when all other factors were considered; LoS increased with age and was statistically significantly higher for patients with BMI > 50. Average LoS of each hospitalization increased with greater number of admissions through A&E. Patients with PJI also had longer LoS if affected by DVT, HF, anemia or osteoporosis at the time of the joint replacement surgery. When the final outcome of the PJI involved a replacement of the affected device, LoS was longer compared to outcomes with retention of the original device (Fig 3).
**Fig 3:** *Coefficients for regression of LoS with zero inflated negative binomial distribution for variables related to patient characteristics (a), arthroplasty surgery (b), medical history (c), drug history (d) and PJI characteristics (e). Rootogram of actual LoS and zero inflated negative binomial model predictions (f).*
The mean number of hospitalizations resulting from the development of PJI was not dependent ($p \leq 0.05$) on patients’ characteristics, medical and treatment history or attributes of the initial arthroplasty surgery. LoS regression coefficient regarding PJI outcome were statistically significant ($p \leq 0.05$) (S2 Fig).
The diagnosis of PJI resulted in 4,129 procedures associated to its management; the most common recorded procedure was the removal and replacement of the original devices, performed in 1,466 cases; other common procedures performed were aspiration or debridement, amputation was carried out in two cases (Table 5). PJI was diagnosed in a total of 1,654 hospitalizations with 602 following a visit A&E.
**Table 5**
| Procedure | Occurrence (n) |
| --- | --- |
| Device Removal/replacement | 1466 |
| Aspiration | 516 |
| Insertion of spacers | 516 |
| Debridement/Irrigation | 280 |
| Irrigation | 233 |
| Superficial debridement | 208 |
| Other | 199 |
| Drainage | 98 |
| Debridement | 86 |
| Intra-articular injection | 19 |
| Amputation | 2 |
The most common final outcome of the recorded PJIs was the revision of the device (performed in one stage for 743 patients and in two stages for 476 patients); the original devise was retained in 95 patients (Table 6). 5.4 and $9.3\%$ of PJI were resolved with DAIR in hip and knee, respectively. The distribution of the direct costs (based on the cost of each procedure performed, type of admission and patients’ comorbidities) associated to each of the recorded PJI revealed a clear not Gaussian profile for both hip and knee devices. The total cost of PJI ranged between £1,146 and £165,824 for hip PJI and between £2,261 and £140,201 for knee PJI; the mean cost of a PJI after hip replacement was £23,337 while after knee £26,523 while the median costs were £17,668 and £20,399 for hip and knee PJI respectively (Fig 4).
**Fig 4:** *Histograms of direct costs associated with PJIs developed in hips and knees.* TABLE_PLACEHOLDER:Table 6 PJI in hips resulting in the replacement of the device had a mean cost of £24,078 against £10,361 when managed only through DAIR; similarly, in knee affected by PJI requiring device replacement, the cost was £28,192 against £8,815 when the device was retained.
## 3.3.2 PJIs causative micro-organisms
Gram+ bacteria were the common pathogens recorded in PJI at any time after the implant surgery with increasing frequency in late PJI as Gram+ represented $71\%$ of PJI diagnosed in the first 3 months after surgery and over $85\%$ of those diagnosed after more than 12 months from the initial surgery. Regardless of the multi-species or single-species PJI nature, the most common species were S. aureus and other Staphylococcus spp.; E. coli and P. aeruginosa were observed in $12.2\%$ and $6.1\%$, respectively, of PJIs diagnosed in the first 3 months following surgery and in $6.1\%$ and $2.6\%$, respectively, of PJIs diagnosed more than 12 months following surgery. Streptococci represented about $15\%$ of species observed in PJIs at any point after surgery. Multispecies PJIs were observed more frequently in early infections (first 3 months after surgery) than in later PJI (Fig 5).
**Fig 5:** *Causative micro-organisms of PJIs observed after different periods of time following arthroplasty aggregated by individual species (a), mono and multi species (b) and Gram (c).*
## 4 Discussion
Most the patients (over 200,000) reached the end of the study period (5 years) and the most common reason for censoring was the termination of the data collection before the end of the study period as expected any patient undergoing arthroplasty after 2015 would not be able to be followed-up for at least 5 years as the data cut for the database extraction was August 2020 The observed cumulative PJI occurrence after 5 years from joint replacement surgery was $0.58\%$ (Fig 1); this is in agreement with the generally reported range of 0.2–$1\%$ [34–37]. With the majority of cases developing in the first 6–12 months post device implant and about $\frac{2}{3}$ of all cases diagnosed in the first 2 years post-surgery, similarly to other reports [38]. A previous study has employed the England and Wales register along with (HES) database to establish risk factor for PJI [34]. However, this study employed only secondary care information related to patient medical history as HES collects data only in secondary care. This approach limited the covariates considered as information related to patients’ medications and characteristics such as alcohol and smoking habits are collected in primary care in the UK.
## 4.1 PJI risk factors
Male gender and obesity were found risk factors for the development of PJI (Fig 2 and S1 Fig) as also previously reported [34, 38–43]. We observed a decreasing PJI risk with increasing age [34] (Fig 2 and S1 Fig) as also found in other studies [34]; however the impact of age is controversial as an increased risk of PJI for older patients was reported [44] while other studies reported no relation [45, 46].
In terms of medical history, the impact of diabetes on increased rates of PJI was also expected [38, 40–42] as poor glycemic control is generally linked to higher infection rates [47]. When adjusted for other cofactors, the most impactful medical history in PJI risk is a history of previous PJI in the same joint or in another joint, confirming previous findings [48, 49]. Regardless of a previous PJI, a secondary joint replacements was 4 times more likely to develop PJI and in line with previous reported PJI risks after aseptic replacement [50, 51]; this is possibly linked to the higher complexity of the surgery [52]. Moreover, the higher risk of PJI in devices implanted during an unscheduled hospitalization (through A&E) is likely a consequence of the higher complexity of the surgery, tissue damage, greater likelihood of foreign microorganisms entering the patients and often suboptimal pre- and perioperative care in a post-traumatic THA/TKA [53, 54].
The number of intra-articular injections prior to surgery was not included in the multivariable Cox proportional hazard analysis because of the multicollinearity observed with the last injection (no use is equivalent to 0 injections). Despite the modulating/suppressive activity on the immune system, the use of DMARDs was not observed to contribute to PJI development; on the contrary, a prescription of DMARDs in the 3 months prior surgery was associated to a lower risk of PJI (Fig 2); these findings support the recommendation of the International Consensus Meeting on PJI to suspend DMARDs prior to an elective joint arthroplasty based on their half-life [55] and the non-significant role of TNF-α inhibitors prior to knee/hip replacement [43]. We found that steroids use, that also reduce immune system responses, statistically significantly contributed to PJI only when the last prescription was more than 6 months before surgery (Fig 2), partially supporting the concern related to the causality between steroids use and surgical site infections [56]. In contrast with previous finding linking intra-articular injections to higher PJI risk regardless of the time elapsed [57, 58], we did not observe such procedure as a risk factor for PJI. DOACs are normally prescribed for AF or DVT/PE, while such diagnosis was not found to be a risk factor for PJI, only the recent prescription of a DOAC (less than 3 months before arthroplasty) was observed to increase the risk of PJI. Coagulopathy is associated to greater risk of hematoma during surgery and PJI [59]; therefore, a distant prescription is suggestive of a resolved pathology.
The fixation method did not significantly impact the risk of developing PJI in line with previous finding that questioned the long-term efficacy of antibiotic laden bone cement in preventing PJI mainly because of the short-term drug release from such material [60–62].
## 4.2 PJI health resource consumption and outcomes
Zero inflated model returned the best fitting of the LoS associated to a PJI as some patients were more likely to be discharged on the same day they were admitted while zero truncated models performed better in predicting the number of hospitalizations resulting from PJI because no patient with PJI had no hospitalization.
LoS distribution (Fig 3) showed a similar pattern to those of at Nuffield Orthopaedic Centre, Oxford ‘most complex/salvage’ knee with a median LoS = 11 days (IQR 7–19) [5]. Furthermore, we observed longer LoS for hip than knee PJI in agreement with previous [63]. The critical parameters correlated to LoS were gender, age and BMI as generally expected [64–66]. PJI not resulting in device revision had generally shorter LoS because of the less invasive procedures carried out; while patients with history of DVT, anemia and HF spent longer in hospital in line with the general trend of longer LoS in patents with diseases of the circulatory system [67]. Also, attendance to A&E increased LoS likely as in indication of more acute cases.
We observed slightly higher overall mean and median costs for the management of PJI affecting hips than knees. However, the opposite (knee more expensive than hip) was recorded when considering only PJI resulting in the replacement of the original device; this was similar to a previous reports [63, 68]; PJIs that were resolved with DAIR had lower costs because of the reduced complexity of the treatment. The costs observed were similar to those reported in USA [63] but lower compared to other reports of £33,000 [12] and £30,011 [69]. This could be associated to the inclusion of only direct hospital management costs in this study and not the costs related to the diagnosis of PJI and out-patients costs rehabilitation.
Most of the PJI recorded in this study resulted in the replacement of the device or of one the components; however, we observed a higher ratio of single stage revision then previous reports [51]. The observation of final outcomes such as the first stage of a two stages revision were likely the results of two stages replacement in patients lost to follow-up (Table 6).
## 4.3 PJIs causative micro-organisms
Gram+ were the main pathogens observed in PJIs; however, Gram- became more frequent with increasing follow-up time, this is in line with the main difference among early and late PJI. The former infections are predominantly caused by highly virulent pathogens (e.g., Staphylococcus aureus, streptococci and enterococci) while the later are mainly caused by low-virulent organisms (e.g., Propionibacterium acnes, S. lugdunensis) [36].
Among the Gram+ pathogens recorded, staphylococci were responsible for more than $50\%$ of PJIs, streptococci and enterococci accounted for about $10\%$ of all PJIs and fungi for about 1–$2\%$; in line with the expected proportions of causative micro-organisms [4, 45, 46, 70, 71].
More than one pathogen species was reported in about a quarter of the recorded PJIs in line with [72] with the greater frequency of polymicrobial species observed in the early PJI as possibly caused by the route of such infection being the surgical wound [4].
## 4.4 Strengths and Limitations of study design, data sources, and analytic methods
This study employed data from the CPRD and HES (admitted patient care/outpatients) databases, as well as death registrations from the ONS allowing for clinical events (for example, diagnoses of PJI, type of management (debridement, revision surgery and amputation) and risk factors (through outpatient/inpatient admissions) to be captured more fully than using either primary care data from the CPRD or secondary care from HES alone. Furthermore, ONS death registrations data ensured accurate and complete recording of all-cause mortality among study patients. A further strength of this study is the large samples of patients reducing uncertainty in the effect size estimated.
This study was retrospective in nature and reliant upon routinely captured data. As with current publications assessing incidence of PJI associated with patient factors, it is likely that a number of plausible confounding variables were unavailable, either because they are not routinely recorded (for example, nutritional status, genetic factors) or unavailable (for example, hospital prescribing and over the counter medication use). The records in the database do not differentiate the presence of antimicrobials in bone cement used during fixation.
Additionally, as we are using register-based data, the possibility of error in the recording of data or codes may be a source of information bias.
As the CPRD is a retrospective observational database, clinical measurements were collected through routine primary care rather than being mandated through a study protocol, as in a prospective study. Missing values, therefore, were present in covariates at baseline.
The study population was restricted to patients with complete data at baseline for certain variables inviting bias, as the patient profile of those with complete data may be different to those without. However, a workable dataset was required in order to fulfil the study objectives and, therefore, this practical approach was adopted to identify a suitable study population.
## 4.5 Recommendation and future directions
The impact of each risk factors on the probability of developing PJI is unlikely to be identical, therefore risk equations capturing the contribution of each factor on the likelihood of PJI would provide further scope for the identification of patients at risk.
Incorporation of time varying covariates would allow a better representation of the patients’ status through the follow-up period particularly for medical history and drug use.
As some potentially relevant risk factors were not available through the chosen data sources, a study containing such information would fill such knowledge gap.
## 5 Conclusions
The knowledge about risks factors for PJI, specific to England and Wales, generated in this study reinforces the role of some commonly accepted drivers while also showed that other sometimes considered influential are not shaping the probability of developing PJI. Thus, this work can provide reasons to update local guidelines.
Additionally, the quantification of the contemporary impact of PJI on resource utilization such as hospital capacity (LoS and A&E admissions) along with the budgetary impact will provide the necessary information for the evaluation of future intervention for the prevention or managements of PJI.
Gram+ were the most common causative microorganisms of PJI; early PJI were more likely to be multispecies than later (years after surgery); the presence of Gram- decreased with increasing post-surgery follow-up.
## References
1. Learmonth ID, Young C, Rorabeck C. **The operation of the century: total hip replacement**. *Lancet (London, England)* (2007.0) **370** 1508-19. DOI: 10.1016/S0140-6736(07)60457-7
2. Edwards RR, Campbell C, Schreiber KL, Meints S, Lazaridou A, Martel MO. **Multimodal prediction of pain and functional outcomes 6 months following total knee replacement: a prospective cohort study**. *BMC Musculoskelet Disord* (2022.0) **23** 302. DOI: 10.1186/s12891-022-05239-3
3. Katz JN. **Total joint replacement in osteoarthritis**. *Best Pract Res Clin Rheumatol* (2006.0) **20** 145-53. DOI: 10.1016/j.berh.2005.09.003
4. Tande AJ, Patel R. **Prosthetic joint infection**. *Clinical microbiology reviews* (2014.0) **27** 302-45. DOI: 10.1128/CMR.00111-13
5. Sabah SA, von Fritsch L, Khan T, Shearman AD, Rajasekaran RB, Beard DJ. **Revision total knee replacement case-mix at a major revision centre**. *J Exp Orthop* (2022.0) **9** 34. DOI: 10.1186/s40634-022-00462-2
6. Osmon DR, Berbari EF, Berendt AR, Lew D, Zimmerli W, Steckelberg JM. **Diagnosis and management of prosthetic joint infection: clinical practice guidelines by the Infectious Diseases Society of America**. *Clinical infectious diseases: an official publication of the Infectious Diseases Society of America* (2013.0) **56** e1-e25. DOI: 10.1093/cid/cis803
7. Zmistowski B, Karam JA, Durinka JB, Casper DS, Parvizi J. **Periprosthetic Joint Infection Increases the Risk of One-Year Mortality**. *JBJS* (2013.0) **95** 2177-84. DOI: 10.2106/jbjs.L.00789
8. Moore AJ, Blom AW, Whitehouse MR, Gooberman-Hill R. **Deep prosthetic joint infection: a qualitative study of the impact on patients and their experiences of revision surgery**. *BMJ Open* (2015.0) **5** e009495. DOI: 10.1136/bmjopen-2015-009495
9. Vanhegan IS, Malik AK, Jayakumar P, Ul Islam S, Haddad FS. **A financial analysis of revision hip arthroplasty: the economic burden in relation to the national tariff**. *The Journal of bone and joint surgery British volume* (2012.0) **94** 619-23. DOI: 10.1302/0301-620X.94B5.27073
10. Okafor C, Hodgkinson B, Nghiem S, Vertullo C, Byrnes J. **Cost of septic and aseptic revision total knee arthroplasty: a systematic review**. *BMC Musculoskelet Disord* (2021.0) **22** 706. DOI: 10.1186/s12891-021-04597-8
11. Premkumar A, Kolin DA, Farley KX, Wilson JM, McLawhorn AS, Cross MB. **Projected Economic Burden of Periprosthetic Joint Infection of the Hip and Knee in the United States**. *The Journal of arthroplasty* (2021.0) **36** 1484-9.e3. DOI: 10.1016/j.arth.2020.12.005
12. Garfield K, Noble S, Lenguerrand E, Whitehouse MR, Sayers A, Reed MR. **What are the inpatient and day case costs following primary total hip replacement of patients treated for prosthetic joint infection: a matched cohort study using linked data from the National Joint Registry and Hospital Episode Statistics**. *BMC Med* (2020.0) **18** 335. DOI: 10.1186/s12916-020-01803-7
13. Dowsey MM, Choong PF. **Early outcomes and complications following joint arthroplasty in obese patients: a review of the published reports**. *ANZ journal of surgery* (2008.0) **78** 439-44. DOI: 10.1111/j.1445-2197.2008.04554.x
14. Maoz D, Mannucci F, Nelemans G. **Observational Clues to the Progenitors of Type Ia Supernovae**. *Annual Review of Astronomy and Astrophysics* (2014.0) **52** 107-70. DOI: 10.1146/annurev-astro-082812-141031
15. Malinzak RA, Ritter MA, Berend ME, Meding JB, Olberding EM, Davis KE. **Morbidly obese, diabetic, younger, and unilateral joint arthroplasty patients have elevated total joint arthroplasty infection rates**. *The Journal of arthroplasty* (2009.0) **24** 84-8. DOI: 10.1016/j.arth.2009.05.016
16. Everhart JS, Altneu E, Calhoun JH. **Medical comorbidities are independent preoperative risk factors for surgical infection after total joint arthroplasty**. *Clinical orthopaedics and related research* (2013.0) **471** 3112-9. DOI: 10.1007/s11999-013-2923-9
17. Cipriano CA, Brown NM, Michael AM, Moric M, Sporer SM, Della Valle CJ. **Serum and synovial fluid analysis for diagnosing chronic periprosthetic infection in patients with inflammatory arthritis**. *The Journal of bone and joint surgery American volume* (2012.0) **94** 594-600. DOI: 10.2106/JBJS.J.01318
18. Pulido L, Ghanem E, Joshi A, Purtill JJ, Parvizi J. **Periprosthetic joint infection: the incidence, timing, and predisposing factors**. *Clinical orthopaedics and related research* (2008.0) **466** 1710-5. DOI: 10.1007/s11999-008-0209-4
19. Berbari EF, Hanssen AD, Duffy MC, Steckelberg JM, Ilstrup DM, Harmsen WS. **Risk factors for prosthetic joint infection: case-control study**. *Clinical infectious diseases: an official publication of the Infectious Diseases Society of America* (1998.0) **27** 1247-54. DOI: 10.1086/514991
20. Pedersen AB, Svendsson JE, Johnsen SP, Riis A, Overgaard S. **Risk factors for revision due to infection after primary total hip arthroplasty. A population-based study of 80,756 primary procedures in the Danish Hip Arthroplasty Registry**. *Acta orthopaedica* (2010.0) **81** 542-7. DOI: 10.3109/17453674.2010.519908
21. Schrama JC, Espehaug B, Hallan G, Engesaeter LB, Furnes O, Havelin LI. **Risk of revision for infection in primary total hip and knee arthroplasty in patients with rheumatoid arthritis compared with osteoarthritis: a prospective, population-based study on 108,786 hip and knee joint arthroplasties from the Norwegian Arthroplasty Register**. *Arthritis care & research* (2010.0) **62** 473-9. DOI: 10.1002/acr.20036
22. 22Guido Schwarzer JRC, Gerta Rücker. Meta-Analysis with R: Springer Cham; 2015.
23. 23NICE Decision Support Unit (DSU) Technical Support Document. The use of real world data for the estimation of treatment effects in nice decision making. 2016.
24. 24National Academies of Sciences E, and Medicine; Health and Medicine Division; Board on Health Sciences Policy; Forum on Drug Discovery, Development, and Translation. Real-World Evidence Generation and Evaluation of Therapeutics. Washington (DC): National Academies Press (US); 2017.
25. Wolf A, Dedman D, Campbell J, Booth H, Lunn D, Chapman J. **Data resource profile: Clinical Practice Research Datalink (CPRD) Aurum**. *International Journal of Epidemiology* (2019.0) **48** 1740-g. DOI: 10.1093/ije/dyz034
26. Herrett E, Gallagher AM, Bhaskaran K, Forbes H, Mathur R, van Staa T. **Data Resource Profile: Clinical Practice Research Datalink (CPRD)**. *Int J Epidemiol* (2015.0) **44** 827-36. DOI: 10.1093/ije/dyv098
27. Persson R, Vasilakis-Scaramozza C, Hagberg KW, Sponholtz T, Williams T, Myles P. **CPRD Aurum database: Assessment of data quality and completeness of three important comorbidities**. *Pharmacoepidemiol Drug Saf* (2020.0) **29** 1456-64. DOI: 10.1002/pds.5135
28. Persson R, Sponholtz T, Vasilakis-Scaramozza C, Hagberg KW, Williams T, Kotecha D. **Quality and Completeness of Myocardial Infarction Recording in Clinical Practice Research Datalink Aurum**. *Clin Epidemiol* (2021.0) **13** 745-53. DOI: 10.2147/CLEP.S319245
29. Morgan A, Sinnott SJ, Smeeth L, Minassian C, Quint J. **Concordance in the recording of stroke across UK primary and secondary care datasets: a population-based cohort study**. *BJGP Open* (2021.0) **5**. DOI: 10.3399/BJGPO.2020.0117
30. 30CPRD. Data highlights 2022 [cited 2022 16 April]. https://cprd.com/data-highlights.
31. 31Digital N. 2019–20 Reference Costs Collection 2022. https://digital.nhs.uk/data-and-information/data-collections-and-data-sets/data-collections/reference-costs.
32. 32R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria.2019.
33. 33Therneau T. A Package for Survival Analysis in S. version 2.38. 2015.
34. Lenguerrand E, Whitehouse MR, Beswick AD, Kunutsor SK, Burston B, Porter M. **Risk factors associated with revision for prosthetic joint infection after hip replacement: a prospective observational cohort study**. *The Lancet Infectious Diseases* (2018.0) **18** 1004-14. DOI: 10.1016/S1473-3099(18)30345-1
35. Kurtz SM, Ong KL, Lau E, Bozic KJ, Berry D, Parvizi J. **Prosthetic joint infection risk after TKA in the Medicare population**. *Clinical orthopaedics and related research* (2010.0) **468** 52-6. DOI: 10.1007/s11999-009-1013-5
36. Izakovicova P, Borens O, Trampuz A. **Periprosthetic joint infection: current concepts and outlook**. *EFORT Open Rev* (2019.0) **4** 482-94. DOI: 10.1302/2058-5241.4.180092
37. Venäläinen MS, Panula VJ, Klén R, Haapakoski JJ, Eskelinen AP, Manninen MJ. **Preoperative Risk Prediction Models for Short-Term Revision and Death After Total Hip Arthroplasty: Data from the Finnish Arthroplasty Register**. *JBJS Open Access* (2021.0) **6** e20.00091. DOI: 10.2106/JBJS.OA.20.00091
38. **Risk Factors for Periprosthetic Joint Infection Following Primary Total Hip Arthroplasty: A 15-Year, Population-Based Cohort Study**. *JBJS* (2020.0) **102** 503-9. DOI: 10.2106/JBJS.19.00537
39. Li T, Zhang H, Chan PK, Fung WC, Fu H, Chiu KY. **Risk factors associated with surgical site infections following joint replacement surgery: a narrative review**. *Arthroplasty* (2022.0) **4** 11. DOI: 10.1186/s42836-022-00113-y
40. Blanco JF, Díaz A, Melchor FR, da Casa C, Pescador D. **Risk factors for periprosthetic joint infection after total knee arthroplasty**. *Arch Orthop Trauma Surg* (2020.0) **140** 239-45. DOI: 10.1007/s00402-019-03304-6
41. Iannotti F, Prati P, Fidanza A, Iorio R, Ferretti A, Pèrez Prieto D. **Prevention of Periprosthetic Joint Infection (PJI): A Clinical Practice Protocol in High-Risk Patients**. *Trop Med Infect Dis* (2020.0) **5**. DOI: 10.3390/tropicalmed5040186
42. Tan T, Xu C, Kuo FC, Ghanem E, Higuera C, Parvizi J. **Risk Factors for Failure and Optimal Treatment of Total Joint Arthroplasty for Septic Arthritis**. *The Journal of arthroplasty* (2021.0) **36** 892-6. DOI: 10.1016/j.arth.2020.09.020
43. Salt E, Wiggins AT, Rayens MK, Morris BJ, Mannino D, Hoellein A. **Moderating effects of immunosuppressive medications and risk factors for post-operative joint infection following total joint arthroplasty in patients with rheumatoid arthritis or osteoarthritis**. *Semin Arthritis Rheum* (2017.0) **46** 423-9. DOI: 10.1016/j.semarthrit.2016.08.011
44. Liu X, Dong Z, Li J, Feng Y, Cao G, Song X. **Factors affecting the incidence of surgical site infection after geriatric hip fracture surgery: a retrospective multicenter study**. *J Orthop Surg Res* (2019.0) **14** 382. DOI: 10.1186/s13018-019-1449-6
45. Baier C, Adelmund S, Schwab F, Lassahn C, Chaberny IF, Gossé F. **Incidence and risk factors of surgical site infection after total knee arthroplasty: Results of a retrospective cohort study**. *Am J Infect Control* (2019.0) **47** 1270-2. DOI: 10.1016/j.ajic.2019.04.010
46. Muilwijk J, Walenkamp GH, Voss A, Wille JC, van den Hof S. **Random effect modelling of patient-related risk factors in orthopaedic procedures: results from the Dutch nosocomial infection surveillance network ’PREZIES’**. *J Hosp Infect* (2006.0) **62** 319-26. DOI: 10.1016/j.jhin.2005.08.006
47. Knapp S. **Diabetes and infection: is there a link?—A mini-review**. *Gerontology* (2013.0) **59** 99-104. DOI: 10.1159/000345107
48. Chalmers BP, Weston JT, Osmon DR, Hanssen AD, Berry DJ, Abdel MP. **Prior hip or knee prosthetic joint infection in another joint increases risk three-fold of prosthetic joint infection after primary total knee arthroplasty: a matched control study**. *Bone Joint J* (2019.0) **101-b** 91-7. DOI: 10.1302/0301-620X.101B7.BJJ-2018-1189.R1
49. Komnos GA, Manrique J, Goswami K, Tan TL, Restrepo C, Sherman MB. **Periprosthetic Joint Infection in Patients Who Have Multiple Prostheses in Place: What Should Be Done with the Silent Prosthetic Joints**. *JBJS* (2020.0) **102** 1160-8. DOI: 10.2106/JBJS.18.01500
50. Quinlan ND, Werner BC, Brown TE, Browne JA. **Risk of Prosthetic Joint Infection Increases Following Early Aseptic Revision Surgery of Total Hip and Knee Arthroplasty**. *The Journal of arthroplasty* (2020.0) **35** 3661-7. DOI: 10.1016/j.arth.2020.06.089
51. Lenguerrand E, Whitehouse MR, Beswick AD, Toms AD, Porter ML, Blom AW. **Description of the rates, trends and surgical burden associated with revision for prosthetic joint infection following primary and revision knee replacements in England and Wales: an analysis of the National Joint Registry for England, Wales, Northern Ireland and the Isle of Man**. *BMJ Open* (2017.0) **7** e014056. DOI: 10.1136/bmjopen-2016-014056
52. Sabah SA, Alvand A, Price AJ. **Revision knee replacement for prosthetic joint infection: Epidemiology, clinical outcomes and health-economic considerations**. *Knee* (2021.0) **28** 417-21. DOI: 10.1016/j.knee.2020.12.024
53. Gallardo-Calero I, Larrainzar-Coghen T, Rodriguez-Pardo D, Pigrau C, Sánchez-Raya J, Amat C. **Increased infection risk after hip hemiarthroplasty in institutionalized patients with proximal femur fracture**. *Injury* (2016.0) **47** 872-6. DOI: 10.1016/j.injury.2015.12.032
54. Crego-Vita D, Aedo-Martín D, Garcia-Cañas R, Espigares-Correa A, Sánchez-Pérez C, Berberich CE. **Periprosthetic joint infections in femoral neck fracture patients treated with hemiarthroplasty—should we use antibiotic-loaded bone cement?**. *World J Orthop* (2022.0) **13** 150-9. DOI: 10.5312/wjo.v13.i2.150
55. Yeganeh MH, Kheir MM, Shahi A, Parvizi J. **Rheumatoid Arthritis, Disease Modifying Agents, and Periprosthetic Joint Infection: What Does a Joint Surgeon Need to Know?**. *The Journal of arthroplasty* (2018.0) **33** 1258-64. DOI: 10.1016/j.arth.2017.11.031
56. Ismael H, Horst M, Farooq M, Jordon J, Patton JH, Rubinfeld IS. **Adverse effects of preoperative steroid use on surgical outcomes**. *The American Journal of Surgery* (2011.0) **201** 305-9. DOI: 10.1016/j.amjsurg.2010.09.018
57. Avila A, Acuña AJ, Do MT, Samuel LT, Kamath AF. **Intra-articular injection receipt within 3 months prior to primary total knee arthroplasty is associated with increased periprosthetic joint infection risk**. *Knee Surg Sports Traumatol Arthrosc* (2022.0). DOI: 10.1007/s00167-022-06942-3
58. Cheok T, Jennings M, Aprato A, Jayasekera N, Jaarsma RL. **Safety of intraarticular corticosteroid injection preceding hip and knee arthroplasty: a systematic review and meta-analysis amid resolving COVID-19 arthroplasty restrictions**. *J Hip Preserv Surg* (2021.0) **8** 215-24. DOI: 10.1093/jhps/hnab064
59. Parvizi J, Ghanem E, Joshi A, Sharkey PF, Hozack WJ, Rothman RH. **Does "excessive" anticoagulation predispose to periprosthetic infection?**. *The Journal of arthroplasty* (2007.0) **22** 24-8. DOI: 10.1016/j.arth.2007.03.007
60. Schiavone Panni A, Corona K, Giulianelli M, Mazzitelli G, Del Regno C, Vasso M. **Antibiotic-loaded bone cement reduces risk of infections in primary total knee arthroplasty? A systematic review**. *Knee Surg Sports Traumatol Arthrosc* (2016.0) **24** 3168-74. DOI: 10.1007/s00167-016-4301-0
61. King JD, Hamilton DH, Jacobs CA, Duncan ST. **The Hidden Cost of Commercial Antibiotic-Loaded Bone Cement: A Systematic Review of Clinical Results and Cost Implications Following Total Knee Arthroplasty**. *The Journal of arthroplasty* (2018.0) **33** 3789-92. DOI: 10.1016/j.arth.2018.08.009
62. Rodriguez-Merchan EC. **Antibiotic-loaded bone cement in primary total knee arthroplasty: does it reduce the risk of periprosthetic joint infection?**. *Hosp Pract (1995)* (2020.0) **48** 188-95. DOI: 10.1080/21548331.2020.1769417
63. Bozic KJ, Kamath AF, Ong K, Lau E, Kurtz S, Chan V. **Comparative Epidemiology of Revision Arthroplasty: Failed THA Poses Greater Clinical and Economic Burdens Than Failed TKA**. *Clinical orthopaedics and related research* (2015.0) **473** 2131-8. DOI: 10.1007/s11999-014-4078-8
64. Zizza C, Herring AH, Stevens J, Popkin BM. **Length of hospital stays among obese individuals**. *Am J Public Health* (2004.0) **94** 1587-91. DOI: 10.2105/ajph.94.9.1587
65. Khosravizadeh O, Vatankhah S, Bastani P, Kalhor R, Alirezaei S, Doosty F. **Factors affecting length of stay in teaching hospitals of a middle-income country**. *Electron Physician* (2016.0) **8** 3042-7. DOI: 10.19082/3042
66. Agboado G, Peters J, Donkin L. **Factors influencing the length of hospital stay among patients resident in Blackpool admitted with COPD: a cross-sectional study**. *BMJ Open* (2012.0) **2** e000869. DOI: 10.1136/bmjopen-2012-000869
67. Baek H, Cho M, Kim S, Hwang H, Song M, Yoo S. **Analysis of length of hospital stay using electronic health records: A statistical and data mining approach**. *PLoS One* (2018.0) **13** e0195901. DOI: 10.1371/journal.pone.0195901
68. Yao JJ, Hevesi M, Visscher SL, Ransom JE, Lewallen DG, Berry DJ. **Direct Inpatient Medical Costs of Operative Treatment of Periprosthetic Hip and Knee Infections Are Twofold Higher Than Those of Aseptic Revisions**. *JBJS* (2021.0) **103** 312-8. DOI: 10.2106/JBJS.20.00550
69. Kallala R. F., Vanhegan I. S, Ibrahim M. S., Sarmah S., H FS. **Financial analysis of revision knee surgery based on NHS tariffs and hospital costs**. *The Bone & Joint Journal* (2015.0) **97-B** 197-201. DOI: 10.1302/0301-620x.97b2.33707
70. Chisari E, Lin F, Fei J, Parvizi J. **Fungal periprosthetic joint infection: Rare but challenging problem**. *Chin J Traumatol* (2022.0) **25** 63-6. DOI: 10.1016/j.cjtee.2021.12.006
71. Guo H, Xu C, Chen J. **Risk factors for periprosthetic joint infection after primary artificial hip and knee joint replacements**. *J Infect Dev Ctries* (2020.0) **14** 565-71. DOI: 10.3855/jidc.11013
72. Li H, Fu J, Niu E, Chai W, Xu C, Hao LB. **The risk factors of polymicrobial periprosthetic joint infection: a single-center retrospective cohort study**. *BMC Musculoskelet Disord* (2021.0) **22** 780. DOI: 10.1186/s12891-021-04664-0
|
---
title: Comparative metabolomics analysis reveals alkaloid repertoires in young and
mature Mitragyna speciosa (Korth.) Havil. Leaves
authors:
- Rubashiny Veeramohan
- Arief Izzairy Zamani
- Kamalrul Azlan Azizan
- Hoe-Han Goh
- Wan Mohd Aizat
- Mohd Fauzi Abd Razak
- Nur Sabrina Mohd Yusof
- Sharif Mahsufi Mansor
- Syarul Nataqain Baharum
- Chyan Leong Ng
journal: PLOS ONE
year: 2023
pmcid: PMC10030037
doi: 10.1371/journal.pone.0283147
license: CC BY 4.0
---
# Comparative metabolomics analysis reveals alkaloid repertoires in young and mature Mitragyna speciosa (Korth.) Havil. Leaves
## Abstract
The fresh leaves of *Mitragyna speciosa* (Korth.) Havil. have been traditionally consumed for centuries in Southeast Asia for its healing properties. Although the alkaloids of M. speciosa have been studied since the 1920s, comparative and systematic studies of metabolite composition based on different leaf maturity levels are still lacking. This study assessed the secondary metabolite composition in two different leaf stages (young and mature) of M. speciosa, using an untargeted liquid chromatography-electrospray ionisation-time-of-flight-mass spectrometry (LC-ESI-TOF-MS) metabolite profiling. The results revealed 86 putatively annotated metabolite features (RT:m/z value) comprising 63 alkaloids, 10 flavonoids, 6 terpenoids, 3 phenylpropanoids, and 1 of each carboxylic acid, glucoside, phenol, and phenolic aldehyde. The alkaloid features were further categorised into 14 subclasses, i.e., the most abundant class of secondary metabolites identified. As per previous reports, indole alkaloids are the most abundant alkaloid subclass in M. speciosa. The result of multivariate analysis (MVA) using principal component analysis (PCA) showed a clear separation of $92.8\%$ between the young and mature leaf samples, indicating a high variance in metabolite levels between them. Akuammidine, alstonine, tryptamine, and yohimbine were tentatively identified among the many new alkaloids reported in this study, depicting the diverse biological activities of M. speciosa. Besides delving into the knowledge of metabolite distribution in different leaf stages, these findings have extended the current alkaloid repository of M. speciosa for a better understanding of its pharmaceutical potential.
## Introduction
The *Mitragyna genus* from the Rubiaceae family encompasses 10 species, of which six are Asian and four are African. The most prevalent species in the Malay Peninsula are Mitragyna speciosa, Mitragyna diversifolia, Mitragyna hirsuta, Mitragyna parvifolia, Mitragyna rotundifolia, and Mitragyna tubulosa, which are known to contain indole alkaloids with pharmacological properties [1,2]. Among the species, M. speciosa has the most documented narcotic properties as an opium substitute with controversial debate on its legal usage and potential abuse. Furthermore, it is easily obtained through the internet in many Western countries like the United Kingdom (UK) and the United States (US) [3–5] and certain Asian countries like Japan [3,6]. M. speciosa is widely grown in Southeast Asian nations such as Indonesia, Malaysia, and Thailand, mostly for its leaves [7,8]. Indonesia is known to cultivate M. speciosa for global exportation, especially to Europe and North America [9,10]. In Malaysia, the trees of M. speciosa are often grown by villagers in their backyards for consumption [11,12]. It is generally known as kratom in Thailand and ketum or biak-biak in Malaysia. The fresh mature leaves of M. speciosa have been traditionally utilised for therapeutic purposes [13] by chewing or consumed as tea for stimulating effects that increase energy and work productivity [11,14,15]. It is also widely used in Southeast Asian countries as an aphrodisiac, to improve blood circulation, to endure physical fatigue, and to treat diarrhoea, fever, diabetes, chronic pain, and opiate withdrawal syndrome [12,15–19]. The leaf extracts of M. speciosa have been reported to show various biological activities, including antibacterial, antioxidant [20], antimutagenic [21], anti-inflammatory [22], antitussive [23], anaesthetic [24], antipsychotic [25], and antinociceptive [22,26,27] effects. These pharmacological actions are mostly linked to alkaloids in the extracts. However, the pharmacological and safety profiles of M. speciosa remain poorly understood and warrant further investigations [11].
To date, at least 58 alkaloids have been reported in different plant organs (leaf, bark, stem bark, stem, root, fruit, etc.), since the 1920s (S1 Table). Mitragynine (MG) was the first alkaloid to be isolated [28], followed by mitraspecine [29], and the rest were identified between 1963 to 2020. Most pharmacological studies of M. speciosa constituents have been extensively focused on MG and 7-hydroxymitragynine (7-OHMG), known as opioid antinociceptive agents [30–32]. MG and 7-OHMG function as partial agonists of the human mu (μ)-opioid receptor, which also acts on kappa (κ)- and delta (δ)-opioid receptors as competitive antagonists [33,34]. 7-OHMG was shown to be 46- and 13-fold more potent than MG and morphine, exhibiting greater affinity for the μ-opioid receptor [35]. Moreover, 7-OHMG was proposed to pose a higher risk of addiction and toxicity with M. speciosa consumption [36,37].
Previous investigations on the distribution of indole and oxindole alkaloids in M. speciosa using thin-layer chromatography (TLC) have shown that the occurrence and abundance of alkaloids vary between young plants and old trees, different organs (leaf, twig, stem bark, and root bark), as well as locality and time of collection [38–40]. Other factors that influence the variability in M. speciosa constituents are environmental factors [41], variety [42], and leaf maturity [43]. However, Houghton et al. [ 43] collected the young and mature leaf samples from different geographical areas of Malaysia; young leaves were collected from trees grown in Selangor, whereas mature leaves were collected from Perlis. Four isolated alkaloids, mitragynaline, corynantheidinaline, mitragynalinic acid, and corynantheidinalinic acid, were more abundant in young leaves, with a minute amount in mature leaves [43]. Furthermore, many other studies focused on investigating the alkaloids in either mature leaves [44–46], commercial products [47,48], or leaf samples of unspecified age [7,49,50]. Although biochemical and physiological characteristics between young and mature leaves generally vary [51–55], systematic comparison studies on the influence of different leaf stages on metabolite composition and abundance are limited. As the constituents of M. speciosa can also vary geographically and with sampling time [39], it is necessary to conduct sampling at the same locality and time point. In this study, a systematic study was employed using an untargeted metabolomics approach of liquid chromatography-electrospray ionisation-time-of-flight-mass spectrometry (LC-ESI-TOF-MS) to compare the composition and abundance of secondary metabolites in the young and mature leaves of M. speciosa.
To date, only a few targeted indole and oxindole alkaloids were already isolated and characterised from M. speciosa. Enhanced metabolite identifications using nuclear magnetic resonance (NMR) [7,47,49], gas chromatography-mass spectrometry (GC-MS) [42,56,57], high-performance liquid chromatography (HPLC) [42], and liquid chromatography-mass spectrometry (LC-MS) [50,57,58] were performed on M. speciosa leaf samples from various locations, with the isolated and characterised metabolites mostly targeted. Phytochemical characterisation of M. speciosa leaves was conducted using NMR and TLC, which isolated 18 compounds [41] without distinguishing different leaf tissues. At the current stage, apart from the targeted metabolites, a complete metabolite profile of the plant is yet to be studied. On this basis, untargeted metabolite profiling using LC-ESI-MS serves as a practical tool for finding bioactive compounds by analysing the metabolites in plant extracts and linking them to their biological activities [59,60]. Only one untargeted LC-MS metabolomics was employed recently to profile 53 commercial kratom products in the US to determine alkaloid variations, followed by targeted studies of MG, 7-OHMG, and speciofoline with in vitro evaluation of their biological effects [48]. However, annotation of all the overall metabolites was not reported. To our knowledge, this study presents the first systematic metabolite profiling and comparative analysis of young and mature M. speciosa leaves. The findings will substantially enhance existing knowledge of M. speciosa leaves and set the groundwork for subsequent research on this plant.
## Chemicals and reagents
Analytical-grade methanol (CH3OH) was acquired from Merck, Germany, while umbelliferone (C9H6O3, purity $99\%$) was acquired from Sigma-Aldrich, St. Louis, USA. Deionised water was filtered using the Milli-Q Reagent Water System (Millipore Billerica, MA, USA). Mitragynine (C23H30N2O4, purity ≥ $95\%$) and 7-hydroxymitragynine (C23H30N2O5, purity $97.9\%$) reference standards were obtained from Cayman Chemical (Ann Arbor, Michigan, USA) and Cerilliant (Round Rock, Texas, USA).
## Plant materials
Young (freshly expanding top two leaves from the shoot tip) and mature (seventh to tenth leaves from the top) leaves (S1 Fig) were obtained from the Centre for Drug Research (CDR), Universiti Sains Malaysia (USM), Penang. The leaves were collected from the same tree at the same time point at Kuala Kedah. The plant was identified by Dr Farah Alia, from Universiti Sains Malaysia (USM), and a voucher specimen number 11869 was deposited at the herbarium of the School of Biological Sciences, USM. The authentication of plant material is included in the supplementary materials, and the results are summarised in the S1 Text. The leaves of M. speciosa were flash-frozen using liquid nitrogen and kept in a -80°C freezer for metabolite extractions.
## Metabolite extraction
Sample extraction was performed as previously described [61]. The leaves were individually pulverised in liquid nitrogen with mortar and pestle prior to weighing and put into separate Falcon tubes. About 100 mg of powdered samples were immersed in freshly made ice-cold methanol (5mL), vortexed, and incubated for 8 hours on dry ice before overnight incubation in a high-capacity incubator shaker at 20°C. The mixture was centrifuged at 4°C and 6,000 rpm for 10 min. Next, a 0.2-μm polytetrafluoroethylene (PTFE) syringe filter was used to filter the supernatant. To prevent degradation, 1 mL of the extract was pipetted into vials and kept in a -80°C freezer. Later, the extracts were spiked with an internal standard (100 ppm of umbelliferone) before LC-MS analysis [61]. An internal standard is crucial in metabolite profiling studies because it serves as a reference for relative quantification and validation of chromatographic and MS system performance [62–64]. Pooled young (2–3 leaves) and individual mature leaf samples were used to prepare five biological and five technical replicates.
## LC-ESI-TOF-MS analysis
A Thermo Scientific C18 column (AcclaimTM Polar Advantage II, 3 × 150 mm, 3 μm particle size) on an Ultimate UHPLC system (Dionex) was used to perform chromatographic separation of M. speciosa leaf extracts as described previously [61]. Gradient elution with mobile phases of $0.1\%$ formic acid in water (A) and $100\%$ acetonitrile (ACN, B) was performed at 40°C with a flow rate of 0.4 mL/min. The total run time was 15 min. A sample injection volume of 1 μL was used, and the gradient was initiated at $5\%$ solvent B (0–0.5 min), increased to $90\%$ solvent B (0.5–6 min) and maintained at $90\%$ solvent B (6–10 min). The gradient was then returned to $5\%$ solvent B (10–12 min) and finally maintained at $5\%$ solvent B (12–15 min). A MicroTOF-Q III Bruker Daltonics was used to perform high-resolution mass spectrometry in positive ionisation mode with electrospray ionisation (ESI) source settings of 4,500 V capillary voltage, 1.2 bar of nebuliser pressure, 8 L/min of drying gas flow rate at 200°C, and mass range of scan spectra from 50 to 1000 m/z.
MS/MS analysis for young and mature leaf samples of M. speciosa was done according to Rosli et al. [ 62] by pooling replicates of all extracts at equal amounts. Tandem mass spectra were acquired in Auto-MS/MS and multiple reaction monitoring (MRM) mode to facilitate compound identification.
## LC-MS data processing
The LC-MS raw data was processed as described by Veeramohan et al. [ 61] using Bruker Compass DataAnalysis version 4.1 (Bruker Daltonic GmbH) for peak detection and deconvolution of the total ion current chromatogram (TIC). This subsequently generated a list of retention time (RT) to mass per charge ratio (m/z) peaks linked to the detected compounds and intensity values via Find Molecular Features (FMF) algorithm [65]. The processed data was then aligned using Bruker Compass ProfileAnalysis version 2.1 (Bruker Daltonic, Germany) for bucket generation. Next, the generated dataset was tabulated and changed to.xlsx and.csv formats for subsequent analysis. Each bucket (peak) in the table represents a metabolite feature (RT:m/z value; m/z value up to three decimal places as default setting by the software), representing a metabolite. Data filtering was conducted by filtering out metabolite features that are not present in at least $50\%$ of the samples in at least one leaf age group. Missing intensity values for several features in the dataset were manually added via manual peak picking. Furthermore, metabolite features with a coefficient of variation (CV) of > $30\%$ in both leaf age groups were filtered out before metabolite profiling, relative quantification, and statistical analysis to minimise intensity value variation between technical replicates.
For the Auto-MS/MS data, raw data were analysed using DataAnalysis Viewer 4.2 (Bruker Daltonics) to visualise the fragmentation pattern of each RT:m/z value pairs detected. Targeted metabolite features were subjected to MRM mode to obtain their fragmentation profiles. The acquisition parameters of the targeted metabolites are shown in S2 Table.
## Identification of metabolites
The RT:m/z values detected in at least three out of five biological replicates in either group were selected for annotation. Metabolite identification for LC-MS data was attained by employing searches based on mass (m/z values), followed by manual verification similar to Rosli et al. [ 62], with some modifications. The m/z values were looked up in previous studies and online databases, such as METLIN [66], MetFrag [67], MassBank [68], and KNApSAcK [69]. All metabolite annotations were based on only protonated molecule ions of [M+H]+. Metabolites from the Auto-MS/MS and MRM data were identified based on the m/z value, RT in min, molecular formula, and fragmentation profiles. Since several metabolite databases were searched to identify candidates, isomeric features may match numerous candidates [70]. Hence, only metabolites with molecular weights within a 20 ppm mass error of the query m/z value were acquired and annotated from the databases [62] to decrease the number of candidates and increase the confidence in identification. Although these databases help with annotations, it is not always possible to narrow down the results of an observed metabolite feature to a single candidate. Therefore, multiple annotations are provided for such metabolite features in this manuscript. Additionally, the identities of MG and 7-OHMG were validated with authentic standards. Chromatograms of MG and 7-OHMG are shown in S2 Fig.
The level of identification (ID level) for the metabolites identified in this study was determined according to the criteria previously disclosed with some changes [71,72]. Level 1 was attributed to the metabolites validated with authentic standards, whereas level 2 was assigned to putatively identified metabolites with fragmentation profiles. Level 3 was assigned to putatively identified metabolites using parent ions due to the absence of fragmentation profiles. Exact m/z values (up to 4 decimal places) were reported for metabolite features identified with levels 1 and 2, while m/z values up to 3 decimal places (as a default setting by ProfileAnalysis software) were reported for metabolite features annotated with level 3 identification.
## Statistical analyses and relative quantification
MetaboAnalyst 5.0 [73] was used for peak intensity data normalisation by reference feature (internal standard), log transformation, and Pareto scaling, along with statistical analysis (fold change analysis [|Log2FC| > 2] and t-test [false discovery rate (FDR)-adjusted p-value < 0.05]). FDR correction was automatically done in MetaboAnalyst 5.0. Normalisation by reference feature was caried out to account for systematic variations between samples. Log transformation and Pareto scaling were applied to reduce high variations of intensity values between metabolite features. Multivariate analysis by unsupervised PCA and supervised partial least squares discriminant analysis (PLS-DA) was executed via SIMCA-P 14.1 (Umetrics, Sweden) using the normalised data generated by MetaboAnalyst 5.0. Metabolites were further organised according to their significance in projecting the variations in the PLS-DA model. Metabolites that significantly contribute to the discrimination of young and mature leaf groups were shown to have variable importance in the projection (VIP) values > 1.0. The Venn diagram was created using an online application (http://bioinformatics.psb.ugent.be/webtools/Venn/).
The relative abundance of metabolites was calculated using a semi-quantitative method [74,75]. The signal intensities of each putative metabolite were divided by the signal intensity values of the spiked internal standard (umbelliferone, 100 mg/L) from each biological and technical replicate. Means were calculated from each sample replicate and denoted as relative abundance. A heatmap denoting the relative abundance of putatively identified metabolites across young and mature leaves of M. speciosa was generated using MetaboAnalyst 5.0 [73]. The comparative relative abundance of each putatively identified indole alkaloid was graphed using GraphPad Prism 7 (GraphPad Software, Inc.). The mean for each group with standard error of the mean (SEM) was displayed in each graph.
## Overall metabolites in M. speciosa
Untargeted LC-ESI-TOF-MS profiling was performed to obtain metabolite features (RT:m/z values) corresponding to various compounds in young and mature leaves of M. speciosa. The predominant ion was the molecular ion ([M+H]+, m/z 399.2286) of alkaloid MG (Fig 1).
**Fig 1:** *The representative base peak chromatograms (BPC) of young (A) and mature (B) M. speciosa leaf methanol extract obtained by LC-ESI-TOF-MS. Peak numbering designates the identified compounds.*
The metabolites of the young and mature M. speciosa leaves were putatively identified using LC-ESI-TOF-MS, tandem mass spectrometry, previous studies, and online databases. The results revealed a total of 86 metabolites putatively identified in young and mature leaves of M. speciosa (S3 and S4 Tables). They were further categorised into different classes of secondary metabolites, including 63 alkaloids (S3 Table) and 23 other secondary metabolites consisting of a carboxylic acid, a glucoside, a phenol, 3 phenylpropanoids, 6 terpenoids, 10 flavonoids, and a phenolic aldehyde (S4 Table). Alkaloids make up most of the total metabolites in M. speciosa, followed by flavonoids and terpenoids.
The Venn diagram illustrates the uniqueness and overlapping of the overall identified metabolite features among young and mature leaves of M. speciosa (Fig 2). Both young and matured leaves shared high similarity (76 metabolites) in the metabolite profiles. Five metabolites are only found in young leaves, while another five metabolites are uniquely present in mature leaves (Fig 2). The five metabolites uniquely present in young leaves include four alkaloids and one terpenoid, while the five exclusively present in mature leaves include four alkaloids and one flavonoid. The distribution of each metabolite feature can be found in S3 and S4 Tables.
**Fig 2:** *Venn diagram representing the number of overall identified metabolite features shared between or unique to young and mature leaves of M. speciosa.*
## Metabolomics revealed differences in metabolite composition between young and mature leaves of M. speciosa
To analyse the differences in metabolite composition between young and mature leaves, the normalised data matrices acquired from the LC-ESI-TOF-MS analysis were subjected to MVA. With each point denoting a distinct sample, an unsupervised PCA displays the projections of each sample in a multidimensional space. The differences in metabolite compositions are connected to the sample dispersions, and samples with greater similarities are located together, while samples with greater differences are located farther away [62]. A score plot clusters samples according to their metabolite composition, whereas a loading plot represents the metabolites contributing to the variances amongst samples on the score plot [76]. The PCA score plot in Fig 3A shows that the young (Y1–Y5) and mature (M1–M5) groups formed discrete clusters, separated from one another with a total variance (R2X[cumulative] and Q2[cumulative]) of $92.8\%$ and $87.9\%$.
**Fig 3:** *Multivariate analysis of M. speciosa’s metabolite profile.(A) PCA score scatter plot of young (lime green) and mature (dark green) leaves in five biological replicates. Circles are labelled relative to the leaf age groups. (B) PLS-DA score scatter plot of young and mature leaves in five biological replicates. Circles are labelled relative to the leaf age groups. (C) PLS-DA loading scatter plot projecting metabolite features that influenced the clustering and separation of young and mature leaf groups.*
A supervised PLS-DA was used to identify metabolite features that vary significantly between young and mature leaves. The PLS-DA score plot shows that the difference between young and mature leaf metabolome was separated with a good fit of R2X(cumulative) of $92.8\%$, R2Y(cumulative) of $99.7\%$, and Q2(cumulative) of $99.4\%$ (Fig 3B). The PLS-DA loading scatter plot (Fig 3C) shows the distribution of metabolite features corresponding to the separation of the leaf samples in the PLS-DA score scatter plot (Fig 3B), where the metabolites projected further from the centre contribute more to the separation. Subsequently, variable importance in projection (VIP) scores were used to identify metabolite features contributing to the discrimination between young and mature leaf samples. A total of 34 metabolite features with VIP > 1.0 were identified and included in the S3 and S4 Tables. Metabolites with the greatest influence include 11-methoxy-vinorine (alkaloid), vomicine (alkaloid), hirsuteine (alkaloid), 5′′-O-β-D-glucosylpyridoxine (glucoside), and 3-methoxytyramine-betaxanthine (phenol).
Additionally, a heatmap is used to visualise the metabolite expression of all 86 metabolite features annotated to their respective secondary metabolite classes (Fig 4). The heatmap demonstrates varying expression levels of metabolites between young (Y1–Y5) and mature (M1–M5) leaves. Most secondary metabolites, including alkaloids, carboxylic acid, flavonoids, glucoside, phenol, phenolic aldehyde, phenylpropanoids, and terpenoids in M. speciosa showed higher expression in mature leaves (up-regulation) compared to the young ones. Conversely, 26 metabolite features annotated as alkaloids, a phenylpropanoid, and 2 terpenoids were observed to show higher expression in young leaves (down-regulation) compared to the mature leaves (Fig 4).
**Fig 4:** *Heatmap showing the relative abundance of 86 metabolite features [RT(min):m/z] in young and mature leaves of M. speciosa.The scale bar of the heatmap indicates the relative abundance of the annotated metabolite features, with the lighter green colour representing lower intensity and the brighter red colour representing higher intensity. Annotation of each metabolite feature can be found in the S3 and S4 Tables.*
## M. speciosa leaves contain alkaloids of various subclasses
A total of 57 annotated alkaloid features were further categorised into 14 subclasses, i.e., indole, quinoline, isoquinoline, quinolizidine, quinazoline, terpenoid, cyclopeptide, peptide, harmala, phenanthridine, piperamide, piperidine, purine, and pyrazine alkaloids. Meanwhile, six metabolite features were categorised as unclassified due to several annotation hits from different alkaloid subclasses for similar metabolite features. For example, the metabolite feature m/z 225.195 at RT 7.77 is annotated to anapheline (piperidine alkaloid) and cuscohygrine (pyrrolidine alkaloid), with the same mass error (Δppm -5.1; S3 Table). Indole alkaloids are the major subclass, with 30 metabolite features putatively identified (ID level 2 and 3) and 2 metabolite features validated with authentic standards (ID level 1), i.e., MG and 7-OHMG (S2 Fig), followed by 6 isoquinolines, 4 quinolines, 3 piperidines, 2 metabolite features in each subclass of cyclopeptides and purines, 1 metabolite feature in the other remaining subclasses, and 6 unclassified alkaloids (S3 Table). It is noteworthy that most of these alkaloid subclasses have not been reported in M. speciosa thus far. These results putatively revealed 46 more alkaloids than the previously reported list (S1 Table) (14 alkaloids putatively identified with ID level 2 and 32 alkaloids annotated with ID level 3). From the 58 alkaloids previously reported in M. speciosa, 17 metabolite features in this study are annotated to the known indole alkaloids (MG, 7-OHMG, speciogynine, corynantheidine, paynantheine, isopaynantheine, speciociliatine, and ajmalicine) and oxindole alkaloids (mitraphylline, isomitraphylline, rynchophylline, isorynchophylline, corynoxine, corynoxine B, speciofoline, isospeciofoleine, and javaphylline) of M. speciosa (S3 Table).
## Differential expression and abundance of alkaloids found in young and mature leaves of M. speciosa
Since the pharmacological actions of M. speciosa are associated with alkaloids, their expression in young and mature leaves was further assessed. Of the 63 putatively identified alkaloids, 38 are significantly (FDR < 0.05) different (S3 Table). Fold change (FC) analysis revealed 22 significantly different (|Log2FC| > 2) alkaloids between young and mature leaves, with 14 and 8 alkaloids exhibiting at least 4-fold higher expressions in young and mature leaves (Table 1).
**Table 1**
| No. | RT:m/z | Metabolite | Alkaloid subclass | Log2(FC) |
| --- | --- | --- | --- | --- |
| 1 | 8.53min:381.1808m/z | Vomicine | Indole | -8.0 |
| 2 | 5.34min:365.1858m/z | 11-Methoxy-vinorine | Indole | -7.3 |
| 3 | 8.38min:351.171m/z | Perakine/vomilenine/polyneuridine aldehyde/19-epi-cathenamine/cathenamine | Indole | -6.6 |
| 4 | 5.39min:367.2031m/z | Hirsuteine | Indole | -6.0 |
| 5 | 8.53min:349.1558m/z | Alstonine | Indole | -4.5 |
| 6 | 5.18min:243.0857m/z | Lumichrome | Pyrazine | 3.8 |
| 7 | 5.21min:385.2113m/z | Rynchophylline | Indole | -3.4 |
| 8 | 5.31min:383.196m/z | Akuammine/aricine/cabucine/lochnerinine | Indole | -3.4 |
| 9 | 4.90min:531.2313m/z | 3-α(S)-Strictosidine | Indole | -3.1 |
| 10 | 5.55min:395.196m/z | Brucine | Indole | -3.1 |
| 11 | 4.91min:413.205m/z | (-)-Alstolucine A | Indole | 3.1 |
| 12 | 5.18min:399.229m/z | Speciogynine | Indole | 3.0 |
| 13 | 1.94min:130.087m/z | L-Pipecolic acid/D-pipecolic acid | Piperidine | -2.9 |
| 14 | 4.22min:161.1076m/z | Tryptamine | Indole | -2.8 |
| 15 | 1.78min:181.071m/z | Theophylline/theobromine/paraxanthine | Purine | 2.8 |
| 16 | 1.84min:322.147m/z | Acronycine/2-[4-(3,4-Methylenedioxyphenyl)butyl]-4(1H)-quinolinone | Unclassified | 2.8 |
| 17 | 4.53min:372.1430m/z | (+)-N-(methoxycarbonyl)-N-norboldine | Isoquinoline | 2.8 |
| 18 | 5.35min:353.185m/z | Ajmalicine | Indole | -2.7 |
| 19 | 2.10min:261.084m/z | 1,2,3,4-Tetrahydro-β-carboline-1,3-dicarboxylic acid | Harmala | 2.7 |
| 20 | 5.31min:369.2198m/z | Corynantheidine | Indole | -2.6 |
| 21 | 4.57min:369.180m/z | Mitraphylline/isomitraphylline/strictosidine aglycone/horhammericine/dialdehyde | Indole | 2.5 |
| 22 | 4.87min:353.1839m/z | Akuammidine | Indole | -2.3 |
The relative abundance of alkaloids quantified in M. speciosa is tabulated according to their respective subclasses to compare their expression in young and mature leaves (S3 Table). According to the heatmap, most annotated alkaloids of various subclasses are more abundant in mature leaves than in young leaves (Fig 4). Many alkaloids of M. speciosa belonging to the isoquinoline, cyclopeptide, harmala, piperidine, terpenoid, peptide, phenanthridine, pyrazine, and quinolizidine subclasses are more abundant in the mature leaves, whereas alkaloids from the indole, quinoline, piperamide, and quinazoline subclasses are more abundant in the young leaves. Meanwhile, two purine alkaloids showed similar abundance in young and mature leaves (S3 Table and Fig 5).
**Fig 5:** *Bar chart representing the distribution of 63 annotated alkaloid features categorised to their respective subclasses.Alkaloid subclasses with a higher abundance in young leaves are coloured blue, while subclasses with a higher abundance in mature leaves are coloured red.*
Since indoles are the most prominent category of alkaloids in M. speciosa, further comparisons were focused on these compounds (Fig 6). Among the 32 annotated indole alkaloids, 13 are significantly more abundant in young leaves, while 3 are more abundant in mature leaves (S3 Table and Fig 6). Interestingly, two of the most studied indole alkaloids of M. speciosa, MG and its derivative, 7-OHMG, showed a higher expression in mature leaves than young leaves. Relative quantification analysis showed that MG is 1.2-fold higher in mature leaves than young leaves, while 7-OHMG is 3.3-fold higher in mature leaves (S3 Table and Fig 6).
**Fig 6:** *Relative abundance of 32 indole alkaloids putatively identified in young and mature leaves of M. speciosa.Relative intensity values for all indole alkaloids are provided in the S3 Table. Asterisks (*) indicate highly differential metabolites (* FDR < 0.05, |Log2FC| > 2). The graphs are arranged in alphabetical order. RT:m/z values are provided instead of compound names for metabolite features annotated to more than one compound, which can be referred to in the S3 Table.*
## Untargeted metabolite profiling of M. speciosa leaves using LC-ESI-TOF-MS
Untargeted metabolomics provides comprehensive and unbiased qualitative and quantitative analyses of each metabolite in an organism [77]. LC-MS is a typical approach for analysing a wide spectrum of plant metabolites in untargeted metabolomics [62,78–81]. Additionally, the combination of LC and ESI-TOF-MS in this investigation improves mass accuracy, allowing for faster metabolite identification and quantification [62,81,82]. Recent untargeted metabolomics followed by a targeted quantification study of M. speciosa profiled 53 commercial kratom products. Compound identification was only focused on three targeted alkaloids, hampering the identification of other potentially bioactive compounds [48].
In this study, we focused on untargeted metabolite profiling to explore the secondary metabolite composition in the young and mature leaves of M. speciosa. Metabolite annotation was manually conducted using mass-based (m/z values) searches against several online databases and putatively identified 86 metabolites, of which 2 were identified using authentic standards (ID level 1), 39 were successfully identified using MS/MS data (ID level 2), and 45 annotated using MS data (ID level 3). In addition to alkaloids, 23 other secondary metabolites are categorised into flavonoids, terpenoids, and phenylpropanoids, while the rest comprises a carboxylic acid, a glucoside, a phenol, and a phenolic aldehyde (S3 and S4 Tables). Previously, León et al. [ 41] isolated and identified 10 other phytochemicals on top of 8 known M. speciosa alkaloids. The 10 phytochemicals isolated comprise a flavonoid, a saponin, two triterpenoid saponins, two monoaryl glycosides, and two cyclohexanone glycosides [41]. Several years later, Charoonratana et al. [ 83] determined the metabolite profiles in M. speciosa via NMR and HPLC analyses to identify the metabolites involved in the biosynthesis of MG. Besides MG, 15 metabolites were identified, including flavonoids, iridoids, triterpenes, organic acids, phenolic acids, amino acids, and sugar [83]. In our study, only two previously reported compounds, epicatechin (flavonoid) (ID level 2) and secologanin (terpenoid) (ID level 3), were annotated. The rest of the 21 secondary metabolites are reported for the first time in M. speciosa (20 metabolites putatively identified with ID level 2 and 1 metabolite annotated with ID level 3) (S4 Table). These metabolite annotations are only putative due to inadequate data on M. speciosa metabolites in public databases. Due to the various acquisition of metabolic data with different analytical instruments and methods, a complete spectral database for LC-MS is nearly unfeasible [62,70,84,85]. Nonetheless, a comprehensive metabolite profiling reported in the present study grants a future search of more potential metabolites in the plant.
## Variation of secondary metabolites between young and mature leaves
Various findings have shown that the composition of secondary metabolites differs in young and mature leaves [52,54,55,77,86]. Recent studies have found that *Gingko biloba* L. [52] and sugarcane [55] contain higher secondary metabolites in older leaves than in immature leaves. However, the younger leaves of *Melicope ptelefolia* [77] and Inga trees [87] had higher concentrations of secondary metabolites than mature leaves. Thus, secondary metabolite expressions in young and mature leaves differ between species. Previous research on M. speciosa mostly involved targeted isolations and structural elucidation of alkaloids. Initial investigations on the young and mature leaves of M. speciosa collected from different geographical regions isolated and revealed only four alkaloids with a higher abundance in young leaves than mature leaves using TLC examination [43]. A recent study on Malaysian M. speciosa leaves published targeted isolations of 10 indole and oxindole alkaloids and profiled the alkaloids of leaves gathered from various locations in the northern states [7]. However, the study of expression levels of many other secondary metabolites, including alkaloids, in young and mature leaves is lacking.
In this study, various secondary metabolites in the young and mature leaves of M. speciosa were comprehensively compared using untargeted metabolomics. Overall, the multivariate analysis showed that leaf age is a vital separating factor that led to clear discrimination among young and mature leaves collected from the same tree at the same time point (Fig 3). Our study also found that mature M. speciosa leaves expressed higher levels of secondary metabolites than young leaves (Fig 4), further indicating that metabolites are distributed differently in the young and mature leaves of the same tree. Furthermore, many alkaloids of M. speciosa belonging to various subclasses are more abundant in mature leaves than young leaves (Fig 5), implying that the concentrations of these metabolites increased with age. Most alkaloids are toxic and are typically distributed in sections of plants most threatened by the attacks of herbivores, insects, and/or microorganisms [88–90]. Generally, susceptible organs and tissues like seeds, plantlets, buds, and young leaves require more protection than aged ones; hence, more defence substances are synthesised or sequestered [90,91]. However, the distribution of alkaloids as defensive substances among young and mature leaves follows different trends in herbs and trees. A “phenological defence” is given to simultaneously occurring flushes of new leaves in trees, so there is more toxin accumulation in mature leaves because of the protection needed by mature leaves until fresh shoots of young leaves are formed [90,92]. Some metabolites are only found in either young or mature leaves; the diversity found among these specific metabolites in leaves of different maturity presumably reflects the evolution of metabolites towards various roles during leaf growth.
On the other hand, apart from a few key indole alkaloids in M. speciosa like MG and its derivative 7-OHMG that showed higher abundance in mature leaves, most of the putatively identified indole alkaloids in M. speciosa, such as corynantheidine, hirsuteine, rynchophylline, vomicine, tryptamine, and 3-α(S)-strictosidine among others, are significantly abundant in young leaves than mature leaves (Fig 6), implying that these indole alkaloids are highly synthesised in young leaves. Similar findings were observed on several indole alkaloid-producing plants, i.e., Camptotheca acuminata, Catharanthus roseus, Rauvolfia serpentina, and *Uncaria tomentosa* [93–97]. A previous study suggested that indole alkaloids found to be significantly abundant in young M. speciosa leaves may be the precursors of mitragynine and its derivatives [43]. Indole and oxindole alkaloids are usually produced via complex and divergent enzymatic steps of the monoterpenoid indole alkaloid (MIA) biosynthesis pathway [7]. The MIA biosynthesis is generally initiated with the condensation of the key precursor, 3-α(S)-strictosidine, from tryptamine (indole precursor) and secologanin (terpenoid precursor), catalysed by strictosidine synthase (STR) [98,99]. Two of the important MIA biosynthesis precursors, tryptamine and 3-α(S)-strictosidine, annotated in this study, are significantly greater in young leaves (Fig 6), further supporting the hypothesis of Houghton et al. [ 43]. Therefore, this study also identifies possible intermediates of the missing steps in the MIA pathway of M. speciosa. In short, our metabolomics analysis has provided insights into the different compositions of secondary metabolites in the young and mature leaves of M. speciosa.
It is also important to mention that the major obstacle faced during the identification of M. speciosa alkaloids was due to the chemical similarity of the indole and oxindole alkaloids and their small differences in molecular weights [100]. Often, MS/MS data are insufficient to distinguish structural and stereoisomers [70]. Hence, further validation of the potential indole and oxindole alkaloids using authentic standards or NMR is needed to support the current observation.
## New putatively identified secondary metabolites show broader therapeutic potentials of M. speciosa
Despite the wide consumption of M. speciosa in Southeast Asian countries for health and well-being, studies to assess the complete spectrum of biological activities in M. speciosa are still lacking. Several studies deduced that the methanol extracts of M. speciosa contain a mixture of compounds that possibly share similar pharmacological activities, e.g., muscle relaxation [101] and antidiabetic activity [102], to be more effective or potent than a single compound like MG. Through our untargeted metabolomics analysis, several unreported alkaloids that are probable contributors to the known and unknown medicinal properties of M. speciosa are shortlisted (Table 2). Fold change analysis identified three alkaloids (vomicine, hirsuteine, and alstonine) among the top ten most significant alkaloids. Vomicine isolated from the seeds of Strychnos nux-vomica exhibited anti-diabetic activity in albino rats [103]. Hirsuteine isolated from *Uncaria sinensis* showed neuroprotective activity in rats via Ca2+ influx suppression [104]. Likewise, alstonine, commonly found in the Apocynaceae plant family, showed antimutagenic properties in mice bearing YC8 lymphoma cells and Ehrlich ascitic carcinoma cells [105]. Besides, alstonine also showed a dose-dependent and potent antipsychotic profile in mice models [106]. Furthermore, other secondary metabolites like flavonoids and terpenoids also contribute to antioxidant properties in plants [55,78]. Therefore, the collective effects of all the newly annotated alkaloids of M. speciosa may be responsible for several aforementioned biological activities reported in M. speciosa leaf extracts [14,15,20–27].
**Table 2**
| Compounds | Molecular Formula | Biological activity | Reference |
| --- | --- | --- | --- |
| Indole alkaloids | Indole alkaloids | Indole alkaloids | Indole alkaloids |
| *Akuammidine | C21H24N2O3 | Anti-asthmaticAnti-inflammatoryAnalgesicAntitussiveAntidepressant | [107][108][109][110] |
| *Alstonine | C21H20N2O3 | AntipsychoticAnticancer | [106,111][105] |
| *Hirsuteine | C22H26N2O3 | Neuroprotective | [104] |
| *Tryptamine | C10H12N2 | Neurotransmitter & neuromodulatorVasoconstrictor & vasodilatorAntimicrobial & antibacterialAntioxidant & antifungal agents | [112] |
| *Vomicine | C22H24N2O4 | Antidiabetic | [103] |
| Yohimbine | C21H26N2O3 | Aids weight lossAphrodisiac (love drug)Mydriatics (induces dilation of the pupil)AntidiabeticAntifungal | [113][114][115] |
| Pyrroloindole alkaloid | Pyrroloindole alkaloid | Pyrroloindole alkaloid | Pyrroloindole alkaloid |
| Eseramine | C16H22N4O3 | Anticholinesterase | [116] |
| Isoquinoline alkaloids | Isoquinoline alkaloids | Isoquinoline alkaloids | Isoquinoline alkaloids |
| *(+)-N-(methoxycarbonyl)-N-norboldine | C20H21NO6 | Antimicrobial agent | [117] |
| Quinazoline alkaloids | Quinazoline alkaloids | Quinazoline alkaloids | Quinazoline alkaloids |
| Vasicinol | C11H12N2O2 | Anticholinesterase | [118] |
| Terpenoid alkaloids | Terpenoid alkaloids | Terpenoid alkaloids | Terpenoid alkaloids |
| Daphniphylline | C32H49NO5 | Central nervous system depressant | [119] |
| Piperidine alkaloids | Piperidine alkaloids | Piperidine alkaloids | Piperidine alkaloids |
| Prosopinine | C18H35NO3 | Sedative | [120] |
| Purine alkaloids | Purine alkaloids | Purine alkaloids | Purine alkaloids |
| Caffeine | C8H10N4O2 | Stimulant | [121] |
In this study, although the young and mature leaves from the same M. speciosa tree shared many similar metabolites (Fig 2), individual metabolites vary in abundance (S3 and S4 Tables; Figs 4 and 6), suggesting that the metabolomics analysis of different leaf organs aids in determining the part with the most potent medicinal effects. Most of the annotated compounds with notable biological activities like vomicine, hirsuteine, alstonine, akuammidine, and tryptamine are indole alkaloids, which showed a significant abundance in the young leaves (Table 1 and Fig 6), warranting further studies.
## Conclusions
This study reports a comprehensive metabolome from young and mature leaves of M. speciosa via LC-ESI-TOF-MS. In total, 86 metabolites were annotated as alkaloids, flavonoids, terpenoids, phenylpropanoids, carboxylic acid, glucoside, phenol, and phenolic aldehyde. Diverse alkaloid profiles were also identified and classified into 14 subclasses, with 13 subclasses of alkaloids that have not been reported in M. speciosa. These alkaloids are potentially associated with the physiological and biochemical progressions during leaf maturity and plant defence against herbivores, insects, and pathogens. Our findings support M. speciosa as a prominent source of biologically active alkaloids, which can be potentially used for myriads of therapeutic uses, including managing depression, cancer, and diabetes. Interestingly, this study also found that young leaves contain more significantly abundant and medicinally beneficial alkaloids, all of which have prominent bioactivities. Altogether, this work adds to our understanding of metabolite composition in the young and mature leaves of M. speciosa, paving the way for future identification of new biologically active compounds.
## References
1. Brown PN, Lund JA, Murch SJ. **A botanical, phytochemical and ethnomedicinal review of the genus Mitragyna korth: Implications for products sold as kratom**. *J Ethnopharmacol* (2017) **202** 302-325. DOI: 10.1016/j.jep.2017.03.020
2. Gong F, Gu H, Xu Q, Kang W. **Genus**. *Phytopharmacology* (2012) **3** 263-272
3. Cinosi E, Martinotti G, Simonato P, Singh D, Demetrovics Z, Roman-urrestarazu A. **Following “the roots” of kratom (**. *Biomed Res Int* (2015) **2015** 1-11. DOI: 10.1155/2015/968786
4. Veltri C, Grundmann O. **Current perspectives on the impact of kratom use.**. *Subst Abuse Rehabil* (2019) **10** 23-31. DOI: 10.2147/SAR.S164261
5. Singh D, Müller CP, Murugaiyah V, Hamid SBS, Vicknasingam BK, Avery B. **Evaluating the hematological and clinical-chemistry parameters of kratom (**. *J Ethnopharmacol* (2018) **214** 197-206. DOI: 10.1016/j.jep.2017.12.017
6. Maruyama T, Kawamura M, Kikura-Hanajiri R, Takayama H, Goda Y. **The botanical origin of kratom (**. *J Nat Med* (2009) **63** 340-344. DOI: 10.1007/s11418-009-0325-9
7. Chear NJ-Y, León F, Sharma A, Kanumuri SRR, Zwolinski G, Abboud KA. **Exploring the chemistry of alkaloids from Malaysian**. *J Nat Prod* (2021) 1-10. DOI: 10.1021/acs.jnatprod.0c01055
8. Singh A, Desgagné-Penix I. **Transcriptome and metabolome profiling of**. *Sci Rep* (2017) **7** 17356. DOI: 10.1038/s41598-017-17724-0
9. Tanguay P.. **Kratom in Thailand**. *SSRN Electron J* (2012). DOI: 10.2139/ssrn.1908849
10. Singh D, Brown PN, Cinosi E, Corazza O, Henningfield JE, Garcia-Romeu A. **Current and future potential impact of COVID-19 on Kratom (**. *Front Psychiatry* (2020) **11** 1-4. DOI: 10.3389/fpsyt.2020.574483
11. Domnic G, Narayanan S, Mohana-Kumaran N, Singh D. **Kratom (**. *J Subst Use* (2021) **27** 1-6. DOI: 10.1080/14659891.2021.1885515
12. Vicknasingam B, Narayanan S, Beng GT, Mansor SM. **The informal use of ketum (**. *Int J Drug Policy* (2010) **21** 283-288. DOI: 10.1016/j.drugpo.2009.12.003
13. Atiqah S, Ismail I, Che-Othman MH. **Investigating the roles of miRNA in secondary metabolite production in**. *Transactions of the Malaysian Society of Plant Physiology* (2020) **27** 216-222
14. Grewal KS. **Observations on the pharmacology of mitragynine**. *J Pharmacol Exp Ther* (1932) **46** 251-271
15. Jansen KLR, Prast CJ. **Ethnopharmacology of kratom and the**. *J Ethnopharmacol* (1988) **23** 115-119. PMID: 3419199
16. Ahmad K, Aziz Z. *J Ethnopharmacol* (2012) **141** 446-450. DOI: 10.1016/j.jep.2012.03.009
17. Chittrakarn S, Sawangjaroen K, Prasettho S, Janchawee B, Keawpradub N. **Inhibitory effects of kratom leaf extract (**. *J Ethnopharmacol* (2008) **116** 173-178. DOI: 10.1016/j.jep.2007.11.032
18. Hassan Z, Muzaimi M, Navaratnam V, Yusoff NHM, Suhaimi FW, Vadivelu R. **From kratom to mitragynine and its derivatives: physiological and behavioural effects related to use, abuse, and addictio**. *Neurosci Biobehav Rev* (2013) **37** 138-151. DOI: 10.1016/j.neubiorev.2012.11.012
19. Suwanlert S.. **A study of kratom eaters in Thailand**. *Bull Narc* (1975) **27** 21-27. PMID: 1041694
20. Parthasarathy S, Azizi J Bin, Ramanathan S, Ismail S, Sasidharan S, Mohd MI. **Evaluation of antioxidant and antibacterial activities of aqueous, methanolic and alkaloid extracts from**. *Molecules* (2009) **14** 3964-3974. DOI: 10.3390/molecules14103964
21. Ghazali AR, Abdullah R, Ramli N, Rajab NF, Ahmad-Kamal MS, Yahya NA. **Mutagenic and antimutagenic activities of**. *J Med Plants Res* (2011) **5** 1345-1348
22. Mossadeq WMS, Sulaiman MR, Tengku Mohamad TA, Chiong HS, Zakaria ZA, Jabit ML. **Anti-inflammatory and antinociceptive effects of**. *Med Princ Pract* (2009) **18** 378-384. DOI: 10.1159/000226292
23. Macko E, Weisbach JA, Douglas B. **Some observations on the pharmacology of mitragynine**. *Arch Int Pharmacodyn Ther* (1972) **198** 145-161. PMID: 4626477
24. Vermaire DJ, Skaer D, Tippets W. **Kratom and general anesthesia: a case report and review of the literature**. *A&A Pract* (2019) **12** 103-105. DOI: 10.1213/XAA.0000000000000857
25. Vijeepallam K, Pandy V, Kunasegaran T, Murugan DD, Naidu M. *Front Pharmacol* (2016) **7** 1-11. DOI: 10.3389/fphar.2016.00464
26. Reanmongkol W, Keawpradub N, Sawangjaroen K. **Effects of the extracts from**. *Songklanakarin J Sci Technol* (2007) **29** 39-48
27. Watanabe K, Yano S, Horie S, Yamamoto LT. **Inhibitory effect of mitragynine, an alkaloid with analgesic effect from Thai medicinal plant**. *Life Sci* (1997) **60** 933-942. DOI: 10.1016/S0024-3205(97)00023-4
28. Field E.. **Mitragynine and mitraversine, two new alkaloids from species of**. *J Chem Soc Trans* (1921) **119** 887-891
29. Denis P.. **Mitraspecine a new alkaloid from**. *Bull la Cl des Sci* (1938) **24** 653-658
30. Matsumoto K, Horie S, Ishikawa H, Takayama H, Aimi N, Ponglux D. **Antinociceptive effect of 7-hydroxymitragynine in mice: discovery of an orally active opioid analgesic from the Thai medicinal herb**. *Life Sci* (2004) **74** 2143-2155. DOI: 10.1016/j.lfs.2003.09.054
31. Matsumoto K, Mizowaki M, Takayama H, Sakai S, Aimi N, Watanabe H. **Suppressive effect of mitragynine on the 5-methoxy-N,N-dimethyltryptamine-induced head-twitch response in mice**. *Pharmacol Biochem Behav* (1997) **57** 319-323. DOI: 10.1016/s0091-3057(96)00314-0
32. Takayama H, Ishikawa H, Kurihara M, Kitajima M, Aimi N, Ponglux D. **Studies on the synthesis and opioid agonistic activities of mitragynine-related indole alkaloids: discovery of opioid agonists structurally different from other opioid ligands**. *J Med Chem* (2002) **45** 1949-1956. DOI: 10.1021/jm010576e
33. Kruegel AC, Gassaway MM, Kapoor A, Varadi A, Majumdar S, Filizola M. **Synthetic and receptor signaling explorations of the**. *J Am Chem Soc* (2016) **138** 6754-6764. DOI: 10.1021/jacs.6b00360
34. Kruegel AC, Grundmann O. **The medicinal chemistry and neuropharmacology of kratom: a preliminary discussion of a promising medicinal plant and analysis of its potential for abuse**. *Neuropharmacology* (2018) **134** 108-120. DOI: 10.1016/j.neuropharm.2017.08.026
35. Takayama H.. **Chemistry and pharmacology of analgesic indole alkaloids from the rubiaceous plant,**. *Chem Pharm Bull (Tokyo)* (2004) **52** 916-928. DOI: 10.1248/cpb.52.916
36. Hassan Z, Suhaimi FW, Ramanathan S, Ling KH, Effendy MA, Müller CP. **Mitragynine (kratom) impairs spatial learning and hippocampal synaptic transmission in rats.**. *J Psychopharmacol* (2019) **33** 908-918. DOI: 10.1177/0269881119844186
37. Hemby SE, McIntosh S, Leon F, Cutler SJ, McCurdy CR. **Abuse liability and therapeutic potential of the**. *Addict Biol* (2018). DOI: 10.1111/adb.12639
38. Shellard EJ, Houghton PJ, Resha M. **The**. *Planta Med* (1978) **34** 253-263. DOI: 10.1055/s-0028-1097448
39. Shellard EJ, Houghton PJ, Resha M. **The**. *Planta Med* (1978) **34** 26-36
40. Shellard EJ, Houghton PJ, Resha M. **The**. *Planta Med* (1978) **34** 253-263
41. León F, Habib E, Adkins JE, Furr EB, McCurdy CR, Cutler SJ. **Phytochemical characterization of the leaves of**. *Nat Prod Commun* (2009) **4** 907-10. PMID: 19731590
42. Boffa L, Ghè C, Barge A, Muccioli G, Cravotto G. **Alkaloid profiles and activity in different**. *Nat Prod Commun* (2018) **13** 1111-1116. DOI: 10.1177/1934578x1801300904
43. Houghton PJ, Latiff A, Said IM. **Alkaloids from**. *Phytochemistry* (1991) **30** 347-350
44. Takayama H, Kurihara M, Kitajima M, Said IM, Aimi N. **New indole alkaloids from the leaves of Malaysian**. *Tetrahedron* (1998) **54** 8433-8440. DOI: 10.1016/S0040-4020(98)00464-5
45. Takayama H, Kitajima M, Kogure N. **Chemistry of indole alkaloids related to the corynanthe-type from**. *Curr Org Chem* (2005) **9** 1445-1464. DOI: 10.2174/138527205774370559
46. Zhang M, Sharma A, León F, Avery B, Kjelgren R, McCurdy CR. **Plant growth and phytoactive alkaloid synthesis in kratom [**. *PLoS One* (2022) **17** 1-16. DOI: 10.1371/journal.pone.0259326
47. Flores-Bocanegra L, Raja HA, Graf TN, Augustinovic M, Wallace ED, Hematian S. **The chemistry of kratom [**. *J Nat Prod* (2020) **83** 2165-2177. DOI: 10.1021/acs.jnatprod.0c00257
48. Todd DA, Kellogg JJ, Wallace ED, Khin M, Flores-Bocanegra L, Tanna RS. **Chemical composition and biological effects of kratom (**. *Sci Rep* (2020) **10** 1-13. DOI: 10.1038/s41598-020-76119-w
49. Ali Z, Demiray H, Khan IA. **Isolation, characterization, and NMR spectroscopic data of indole and oxindole alkaloids from**. *Tetrahedron Lett* (2014) **55** 369-372. DOI: 10.1016/j.tetlet.2013.11.031
50. Avula B, Sagi S, Wang YH, Wang M, Ali Z, Smillie TJ. **Identification and characterization of indole and oxindole alkaloids from leaves of**. *J AOAC Int* (2015) **98** 13-21. DOI: 10.5740/jaoacint.14-110
51. Ahmad P, Ahanger MA, Singh VP, Tripathi DK, Alam P, Alyemeni MN. *Plant metabolites and regulation under environmental stress* (2018). DOI: 10.1016/C2016-0-03727-0
52. Guo J, Wu Y, Wang G, Wang T, Cao F. **Integrated analysis of the transcriptome and metabolome in young and mature leaves of**. *Ind Crops Prod* (2020) **143** 111906. DOI: 10.1016/j.indcrop.2019.111906
53. Xue Z, Gao H, Zhao S. **Effects of cadmium on the photosynthetic activity in mature and young leaves of soybean plants.**. *Environ Sci Pollut Res* (2014) **21** 4656-4664. DOI: 10.1007/s11356-013-2433-z
54. Xia X, Chen C, Yang L, Wang Y, Duan A, Wang D. **Analysis of metabolites in young and mature**. *PeerJ* (2022) **10** e12844. DOI: 10.7717/peerj.12844
55. Rao MJ, Ahmed U, Ahmed MH, Duan M, Wang J, Wang Y. **Comparison and quantification of metabolites and their antioxidant activities in young and mature leaves of sugarcane.**. *ACS Food Sci Technol* (2021) **1** 362-373. DOI: 10.1021/acsfoodscitech.0c00133
56. Beng GT, Hamdan MR, Siddiqui MJ, Mordi MN, Mahsufi SM. **A simple and cost effective isolation and purification protocol of mitragynine from**. *Malaysian J Anal Sci* (2011) **15** 54-60
57. Casey CR, Conley T, Heise A, Thomas T, Ayres PR. **Quantitative and qualitative analysis of mitragynine in kratom (**. *J Regul Sci* (2015) **02** 1-14
58. Kikura-Hanajiri R, Kawamura M, Maruyama T, Kitajima M, Takayama H, Goda Y. **Simultaneous analysis of mitragynine, 7-hydroxymitragynine, and other alkaloids in the psychotropic plant “kratom” (**. *Forensic Toxicol* (2009) **27** 67-74. DOI: 10.1007/s11419-009-0070-5
59. Erika P RC, R M, Avila F, Batista D, Costa C. **Metabolomic profiling of**. *Phytochemistry* (2019) **168** 1-13. DOI: 10.1016/j.phytochem.2019.112128
60. Kumar R, Bohra A, Pandey AK, Pandey MK, Kumar A. **Metabolomics for plant improvement: status and prospects**. *Front Plant Sci* (2017) **8** 1-27. DOI: 10.3389/fpls.2017.01302
61. Veeramohan R, Azizan KA, Aizat WM, Goh HH, Mansor SM, Yusof NSM. **Metabolomics data of Mitragyna speciosa leaf using LC-ESI-TOF-MS.**. *Data Br.* (2018) **18** 1212-1216. DOI: 10.1016/j.dib.2018.04.001
62. Rosli MAF, Mediani A, Azizan KA, Baharum SN, Goh H. **UPLC-TOF-MS/MS-based metabolomics analysis reveals species-specific metabolite compositions in pitchers of**. *Front Plant Sci* (2021) **12** 1-13. DOI: 10.3389/fpls.2021.655004
63. Colegate SM, Gardner DR. **LC-MS of alkaloids: qualitative profiling, quantitative analysis, and structural Identification.**. *Modern Alkaloids: Structure, Isolation, Synthesis and Biology.* (2007). DOI: 10.1002/9783527621071.ch13
64. Liu Z, Rochfort S. **Recent progress in polar metabolite quantification in plants using liquid chromatography-mass spectrometry**. *J Integr Plant Biol* (2014) **56** 816-825. DOI: 10.1111/jipb.12181
65. Krug D, Zurek G, Schneider B, Garcia R, Müller R. **Efficient mining of myxobacterial metabolite profiles enabled by liquid chromatography-electrospray ionisation-time-of-flight mass spectrometry and compound-based principal component analysis**. *Anal Chim Acta* (2008) **624** 97-106. DOI: 10.1016/j.aca.2008.06.036
66. Smith CA O, ’Maille G, Want EJ, Qin C, Trauger SA, Brandon T R. **METLIN: A metabolite mass spectral database.**. *Ther Drug Monit* (2005) **27** 747-51. DOI: 10.1097/01.ftd.0000179845.53213.39
67. Wolf S, Schmidt S, Müller-Hannemann M, Neumann S. **In silico fragmentation for computer assisted identification of metabolite mass spectra**. *BMC Bioinformatics* (2010) 11. DOI: 10.1186/1471-2105-11-148
68. Horai H, Arita M, Kanaya S, Nihei Y, Ikeda T, Suwa K. **MassBank: a public repository for sharing mass spectral data for life sciences**. *J Mass Spectrom* (2010) **45** 703-14. DOI: 10.1002/jms.1777
69. Afendi FM, Okada T, Yamazaki M, Hirai-Morita A, Nakamura Y, Nakamura K. **KNApSAcK family databases: Integrated metabolite-plant species databases for multifaceted plant research**. *Plant Cell Physiol* (2012) **53** 1-12. DOI: 10.1093/pcp/pcr165
70. Schrimpe-Rutledge AC, Codreanu SG, Sherrod SD, McLean JA. **Untargeted metabolomics strategies—challenges and emerging directions**. *J Am Soc Mass Spectrom* (2016) **27** 1897-1905. DOI: 10.1007/s13361-016-1469-y
71. Sumner LW, Amberg A, Barrett D, Beale MH, Beger R, Daykin CA. **Proposed minimum reporting standards for chemical analysis: chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI).**. *Metabolomics* (2007) **3** 211-221. DOI: 10.1007/s11306-007-0082-2
72. Blaženović I, Kind T, Ji J, Fiehn O. **Software tools and approaches for compound identification of LC-MS/MS data in metabolomics.**. *Metabolites* (2018) **8**. DOI: 10.3390/metabo8020031
73. Pang Z, Chong J, Zhou G, Anderson De Lima Morais D, Chang L, Barrette M. **MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights**. *Nucleic Acids Res* (2021) 49. DOI: 10.1093/nar/gkab382
74. Mazlan O, Aizat WM, Baharum SN, Azizan KA. **Metabolomics analysis of developing**. *Sci Hortic (Amsterdam)* (2018) **233** 323-330. DOI: 10.1016/j.scienta.2018.01.061
75. Wesseling A-M, Demetrowitsch TJ, Schwarz K, Ober D. **Variability of pyrrolizidine alkaloid occurrence in species of the grass subfamily Pooideae (Poaceae).**. *Front Plant Sci* (2017) **8** 1-8. DOI: 10.3389/fpls.2017.02046
76. Azizan KA, Ghani NHA, Nawawi MF. **GC-MS based metabolomics and multivariate statistical analysis of**. *Plant Omi J* (2015) **8** 537-543
77. Shuib NH, Shaari K, Khatib A, Maulidiani R, Kneer S. **Discrimination of young and mature leaves of**. *Food Chem* (2011) **126** 640-645. DOI: 10.1016/j.foodchem.2010.10.043
78. Alara OR, Abdurahman NH, Ukaegbu CI, Azhari NH, Kabbashi NA. **Metabolic profiling of flavonoids, saponins, alkaloids, and terpenoids in the extract from**. *J Liq Chromatogr Relat Technol* (2018) **41** 722-731. DOI: 10.1080/10826076.2018.1511995
79. Commisso M, Strazzer P, Toffali K, Stocchero M, Guzzo F. **Untargeted metabolomics: an emerging approach to determine the composition of herbal products.**. *Comput Struct Biotechnol J* (2013) **4** e201301007. DOI: 10.5936/csbj.201301007
80. Mamat SF, Azizan KA, Baharum SN, Noor NM, Aizat WM. **GC-MS and LC-MS analyses reveal the distribution of primary and secondary metabolites in mangosteen (**. *Sci Hortic (Amsterdam)* (2020) **262** 109004. DOI: 10.1016/j.scienta.2019.109004
81. Mazlan O, Aizat WM, Aziz Zuddin NS, Baharum SN, Noor NM. **Metabolite profiling of mangosteen seed germination highlights metabolic changes related to carbon utilization and seed protection**. *Sci Hortic (Amsterdam)* (2019) **243** 226-234. DOI: 10.1016/j.scienta.2018.08.022
82. Chawla G, Ranjan C. **Principle, instrumentation, and applications of UPLC: A novel technique of liquid chromatography.**. *Open Chem J* (2016) **3** 1-16. DOI: 10.2174/1874842201603010001
83. Charoonratana T, Wungsintaweekul J, Pathompak P, Georgiev MI, Choi YH, Verpoorte R. **Limitation of mitragynine biosynthesis in**. *Zeitschrift fur Naturforsch—Sect C J Biosci* (2013) **68** 394-405. DOI: 10.5560/ZNC.2013.68c0394
84. Aretz I, Meierhofer D. **Advantages and pitfalls of mass spectrometry based metabolome profiling in systems biology**. *Int J Mol Sci* (2016) **17** 632. DOI: 10.3390/ijms17050632
85. Matsuda F.. **Rethinking mass spectrometry-based small molecule identification strategies in metabolomics**. *Mass Spectrom.* (2014) **3** S0038. DOI: 10.5702/massspectrometry.S0038
86. Sedio BE, Archibold AD, Echeverri JCR, Debyser C, Boya P, CA SJ. **A comparison of inducible, ontogenetic, and interspecific sources of variation in the foliar metabolome in tropical trees.**. *PeerJ* (2019) **2019** 1-25. DOI: 10.7717/peerj.7536
87. Wiggins NL, Forrister DL, Endara MJ, Coley PD, Kursar TA. **Quantitative and qualitative shifts in defensive metabolites define chemical defense investment during leaf development in**. *Ecol Evol.* (2016) **6** 478-492. DOI: 10.1002/ece3.1896
88. Hopkins WG, Hűner NPA. *John Wiley & Sons, Inc* (2009). DOI: 10.2134/agronj1951.00021962004300010013x
89. Mu HM, Wang R, Li XD, Jiang YM, Peng F, Xia B. **Alkaloid accumulation in different parts and ages of**. *Zeitschrift fur Naturforsch—Sect C J Biosci* (2010) **65** 458-462. DOI: 10.1515/znc-2010-7-807
90. Sengupta G, Gaurav A, Tiwari S. *Synthesis of Medicinal Agents from Plants* (2018) 47-74. DOI: 10.1016/b978-0-08-102071-5.00003–9
91. Godschalx AL, Stady L, Watzig B, Ballhorn DJ. **Is protection against florivory consistent with the optimal defense hypothesis?**. *BMC Plant Biol* (2016) **16** 1-9. DOI: 10.1186/s12870-016-0719-2
92. Mckey D.. **Adaptive patterns in alkaloid physiology**. *Am Nat* (1974) **108** 305-320
93. Dugé De Bernonville T, Carqueijeiro I, Lanoue A, Lafontaine F, Sánchez Bel P, Liesecke F. **Folivory elicits a strong defense reaction in**. *Sci Rep* (2017) **7** 1-14. DOI: 10.1038/srep40453
94. Flores-Sanchez IJ, Paniagua-Vega D, Vera-Reyes I, Cerda-García-Rojas CM, Ramos-Valdivia AC. **Alkaloid biosynthesis and metabolic profiling responses to jasmonic acid elicitation in**. *Metabolomics* (2016) **12**. DOI: 10.1007/s11306-016-0999-4
95. Lopez-Meyer M, Nessler CL, McKnight TD. **Sites of accumulation of the antitumor alkaloid camptothecin in**. *Planta Med* (1994) **60** 558-560. DOI: 10.1055/s-2006-959571
96. Sadre R, Magallanes-Lundback M, Pradhan S, Salim V, Mesberg A, Jones AD. **Metabolite diversity in alkaloid biosynthesis: a multilane (diastereomer) highway for camptothecin synthesis in**. *Plant Cell* (2016) **28** 1926-1944. DOI: 10.1105/tpc.16.00193
97. Yan X-F, Wang Y, Yu T, Zhang Y-H, Dai S-J. **Variation in camptothecin content in**. *Bot Bull Acad Sin* (2003) **44** 99-105
98. Brose J, Lau KH, Dang TTT, Hamilton JP, Martins L do V, Hamberger B. **The**. *G3 Genes|Genomes|Genetics.* (2021). DOI: 10.1093/g3journal/jkab058
99. Gogineni V, Leon F, Avery BA, McCurdy C, Cutler SJ, Raffa RB. **Phytochemistry of**. *Kratom and Other Mitragynines: The Chemistry and Pharmacology of Opioids From a Non-Opium Source.* (2015) 77-94. DOI: 10.1201/b17666-7
100. Ramanathan S, McCurdy CR. **Kratom (**. *Curr Opin Psychiatry* (2020) **33** 312-318. DOI: 10.1097/YCO.0000000000000621
101. Chittrakarn S, Keawpradub N, Sawangjaroen K, Kansenalak S, Janchawee B. **The neuromuscular blockade produced by pure alkaloid, mitragynine and methanol extract of kratom leaves (**. *J Ethnopharmacol* (2010) **129** 344-349. DOI: 10.1016/j.jep.2010.03.035
102. Purintrapiban J, Keawpradub N. **Study on glucose transport in muscle cells by extracts from**. *Nat Prod Res* (2011) **25** 1379-1387. DOI: 10.1080/14786410802267627
103. Bhati R, Singh A, Saharan V, Ram VA, Bhandari A. *J Ayurveda Integr Med* (2012) **3** 80-84. DOI: 10.4103/0975-9476.96523
104. Shimada Y, Goto H, Itoh T, Sakakibara I, Kubo M, Sasaki H. **Evaluation of the protective effects of alkaloids isolated from the hooks and stems of**. *J Pharm Pharmacol* (1999) **51** 715-722. DOI: 10.1211/0022357991772853
105. Beljanski M, Beljanski MS. **Three alkaloids as selective destroyers of cancer cells in mice**. *Oncology* (1986) **43** 198-203. PMID: 3703465
106. Costa-Campos L, Lara DR, Nunes DS, Elisabetsky E. **Antipsychotic-like profile of alstonine**. *Pharmacol Biochem Behav* (1998) **60** 133-141. DOI: 10.1016/s0091-3057(97)00594-7
107. Hou Y, Cao X, Wang L, Cheng B, Dong L, Luo X. **Microfractionation bioactivity-based ultra performance liquid chromatography/quadrupole time-of-flight mass spectrometry for the identification of nuclear factor-κB inhibitors and β2 adrenergic receptor agonists in an alkaloidal extract of the folk herb**. *J Chromatogr B* (2012) **908** 98-104. DOI: 10.1016/j.jchromb.2012.10.004
108. Menzies JRW, Paterson SJ, Duwiejua M, Corbett AD. **Opioid activity of alkaloids extracted from**. *Eur J Pharmacol* (1998) **350** 101-108. DOI: 10.1016/s0014-2999(98)00232-5
109. Dapaah G, Koffuor GA, Mante PK, Ben IO. **Antitussive, expectorant and analgesic effects of the ethanol seed extract of**. *and H. Durand. Res Pharm Sci* (2016) **11** 100-112
110. Perviz S, Khan H, Pervaiz A. **Plant alkaloids as an emerging therapeutic alternative for the treatment of depression.**. *Front Pharmacol* (2016) **7** 1-7. DOI: 10.3389/fphar.2016.00028
111. Herrmann AP, Lunardi P, Pilz LK, Tramontina AC, Linck VM, Okunji CO. **Effects of the putative antipsychotic alstonine on glutamate uptake in acute hippocampal slices**. *Neurochem Int* (2012) **61** 1144-1150. DOI: 10.1016/j.neuint.2012.08.006
112. Kousara S, Anjuma SN, Jaleela F, Khana J, Naseema S. **Biomedical significance of tryptamine: a review.**. *J Pharmacovigil* (2017) **5** 1-6. DOI: 10.4172/2329-6887.1000239
113. Ostojic SM. **Yohimbine: the effects on body composition and exercise performance in soccer players**. *Res Sport Med* (2006) **14** 289-299. DOI: 10.1080/15438620600987106
114. Tang Y, Axelsson AS, Spégel P, Andersson LE, Mulder H, Groop LC. **Genotype-based treatment of type 2 diabetes with an alpha 2A-adrenergic receptor antagonist.**. *Sci Transl Med.* (2014) **6** 257ra139
115. Almagro L, Fernández-Pérez F, Pedreño MA. **Indole alkaloids from**. *Molecules* (2015) **20** 2973-3000. DOI: 10.3390/molecules20022973
116. Liu Z, Carpenter SB, Bourgeois WJ, Yu Y, Constantin RJ, Falcon MJ. **Variations in the secondary metabolite camptothecin in relation to tissue age and season in**. *Tree Physiol* (1998) **18** 265-270. DOI: 10.1093/treephys/18.4.265
117. Feng T, Xu Y, Cai XH, Du ZZ, Luo XD. **Antimicrobially active isoquinoline alkaloids from**. *Planta Med* (2009) **75** 76-79. DOI: 10.1055/s-0028-1088344
118. Liu W, Shi X, Yang Y, Cheng X, Liu Q, Han H. *PLoS One.* (2015) **10** 122366. DOI: 10.1371/journal.pone.0122366
119. Al-Abd NM, Nor ZM, Al-Adhroey AH, Suhaimi A, Sivanandam S. **Recent advances on the use of biochemical extracts as filaricidal agents.**. *Evidence-based Complementary and Alternative Medicine.* (2013) 1-13. DOI: 10.1155/2013/986573
120. Kohmoto K, Yoder O. **Molecular Genetics of Host-Specific Toxins in Plant Disease**. *Developments in Plant Pathology* (1998)
121. Nehlig A.. **Effects of coffee/caffeine on brain health and disease: what should I tell my patients?**. *Pract Neurol* (2016) **16** 89-95. DOI: 10.1136/practneurol-2015-001162
|
---
title: Impact of high intensity interval and moderate continuous training on plasma
ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 after coronary artery bypass
grafting surgery
authors:
- Sara Zare Karizak
- Majid Kashef
- Abbas Ali Gaeini
- Mostafa Nejatian
journal: Frontiers in Physiology
year: 2023
pmcid: PMC10030057
doi: 10.3389/fphys.2023.1114813
license: CC BY 4.0
---
# Impact of high intensity interval and moderate continuous training on plasma ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 after coronary artery bypass grafting surgery
## Abstract
Background: ProBNP1-108/BNP1-32, and NT-pro-BNP1-76/BNP1-32 ratios are significant indices for predicting complications after coronary artery bypass grafting (CABG) surgery. However, the effect of aerobic training types on these biomarkers has not been fully understood. So, the current study aimed to determine the impact of aerobic interval and continuous training programs on plasma ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 after coronary artery bypass grafting surgery.
Method: 36 patients were selected purposive (27 men and 9 women with mean of age 60.32 ± 5.81 years, height 164.64 ± 9.25 cm, weight 73.86 ± 14.23 kg, fat 32.30 ± 4.28, SBP 142.67 ± 6.49, DBP 84.5 ± 5.16 mmHg in seated position at rest situation and functional capacity of 7.08 ± 2.49 METs) and then divided randomly into three groups: control (C) group (without training program) moderate continuous training (MCT) and high intensity interval training (HIIT) (exercise training program was performed 3 days/week for 8 weeks) with intensities $65\%$–$80\%$ and $80\%$–$95\%$ of reserve heart rate in order. Blood samples were taken 48 h before the first session and 48 h after the last training session to measure the plasma levels of ProBNP1–108, corin enzyme, BNP1-32, and NT-pro-BNP1-76 using the enzyme-linked immunosorbent assay (ELISA) technique. Wilcoxin and kruskal wallis tests were used for analyzing data.
Results: *The plasma* corin enzyme was increased, and the ratios of proBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 were reduced in both training groups in compared with control group ($$p \leq 0.004$$, $$p \leq 0000$$, $$p \leq 0.016$$, $$p \leq 0.003$$, $$p \leq 0.009$$, and $$p \leq 0.016$$) when there was no significant difference was found between training groups ($$p \leq 0.074$$, $$p \leq 450$$, and $$p \leq 0.295$$).
Conclusion: Both high intensity interval training and moderate continuous training in compared with inactivity have positive effects on ratios of ProBNP1-108/BNP1-32, NT-pro-BNP1-76/BNP1-32 and could be effective to promote the health of coronary arteries and prevention of HF in post-CABG patients.
## Introduction
Coronary artery bypass grafting surgery (CABG) is an important therapeutic strategy for patients with coronary heart disease (Serruys et al., 2001). Although CABG can decrease the rates of morbidity and mortality in patients, by the improvement of blood circulation, in the heart, it is also accompanied by the emergence of some side effects. Postoperative heart failure (HF) is one of the significant outcomes that usually occurs after CABG (Siribaddana, 2012). Several lines of evidence indicate that the complications of CABG include inflammation and the production of free radicals due to reperfusion of the ischemic heart. These factors are the initiator of the fibrosis signaling pathway and negative remodeling of the heart. So, they can weaken the myocardial contractility and cause HF in the long term (Siribaddana, 2012; Cockburn et al., 2013). Brain natriuretic-related peptides (proBNP1-108, BNP1-32, and NT-pro-BNP1-76) are critical markers applied for the diagnosis of HF after surgery (Fox et al., 2011; Preeshagul et al., 2013; Fu et al., 2018a). For example, Fox et al. [ 2011] showed a direct and strong association between the increase of BNP1-32 after CABG and the development of HF in the next 5 years. Fu et al. ( 2018a) also reported that impairment in processing and degradation of BNP1-32 and NT-pro-BNP1-76 are effective in the occurrence of HF. Brain natriuretic peptide (BNP1-32) is a hormone released from the cardiac ventricles in response to ischemia and myocardial wall stress due to volume or pressure overload (Fu et al., 2018b). BNP1-32 is primarily synthesized from an inactive prohormone (proBNP1-108) that is cleaved into the active hormone (BNP1-32) and the inactive N-terminal fragment (NT-pro-BNP1-76) by the enzyme named corin. BNP1-32 is decomposed into the body after carrying out its biological function (Ichiki et al., 2011; Ichiki et al., 2013). BNP1-32 reduces blood pressure directly and indirectly by relaxing vascular smooth muscles, blocking the cardiac sympathetic nervous system, and increasing the diuretic and natriuretic effects. BNP1-32 also inhibits the renin-angiotensin-aldosterone system and has anti-proliferative and anti-fibrotic effects on the myocardium (Chopra et al., 2013). Although BNP1-32 is a cardiovascular protective peptide, it is measured in patients’ blood as a cardiovascular stress index. Recent studies indicated that in addition to BNP1-32 indices, namely, proBNP1-108, NT-pro-BNP1-76, like the ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 play a major role in predicting the development of HF (Jensen et al., 2010; Jensen et al., 2012; Miller et al., 2012; Barnet et al., 2015; Huntley et al., 2015; Fu et al., 2018a). *The* generation and degradation of BNP1-32 is a significant problem in cardiac patients. For example, Barnet et al. [ 2015] have shown that the amount of corin enzyme involved in the conversion of proBNP1-108 to BNP1-32 decreases after CABG. Whereas proBNP1-108 is increased simultaneously, indicating a gap in the production process of BNP1-32, particularly after CABG (Barnet et al., 2015). In other words, the enzymatic activity is impaired in the myocardium of patients, especially after CABG. Hence, the ProBNP1-108/BNP1-32 ratio could be increased in these patients and may result in an increased risk of HF (Barnet et al., 2015). On the other hand, proper breakdown of BNP1-32 and NT-pro-BNP1-76 is essential for heart functions; the accumulation of BNP1-32 leads to the saturation of its receptor and inefficiency of BNP1-32 in patients with cardiovascular diseases (Huntley et al., 2015). Some studies indicated that the impairment of BNP1-32 processing and its degradation system could reduce the levels of BNP1-32 and increase the levels of inactive ProBNP1-108 and NT-pro-BNP1-76. This condition, a paradox of BNP1-32, is medically considered a dilemma in patients with cardiovascular diseases (Huntley et al., 2015). Numerous BNP-related peptides are circulated in the bloodstream; however, they have no beneficial effect on the health status of patients with cardiovascular disorders, as most of these peptides are found inactive and have no functionality (Huntley et al., 2015; Zarekarizak et al., 2017). Hence, the ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 could be regarded as two important indices for BNP1-32 processing and degradation, which are defective in patients with cardiovascular diseases. The dysfunction of BNP1-32 processing and its degradation retain a fundamental role in the development of HF in the long-term period (Jensen et al., 2010; Jensen et al., 2012; Barnet et al., 2015; Huntley et al., 2015). For example, Suzuki and Sugiyama, [2018] showed a significant risk of HF in patients with a high ratio of NT-proBNP1-76/BNP1-32 according to Kaplan-*Meier analysis* (Suzuki and Sugiyama, 2018). In other words, the positive effects of exercise on cardiac rehabilitation, especially ACT, have been addressed and can act as a therapeutic strategy to prevent secondary problems and mortality after surgery (Dendale et al., 2005; Conraads and Beckers, 2010; Valkeinen et al., 2010; Ghashghaei et al., 2012; Meyer et al., 2013; Nishitani et al., 2013). High Intensity Interval Training (HIIT) is also an exercise training employed for cardiovascular adaptations and improvement of functional capacity (Wisloff et al., 2007; Gibala et al., 2012; Guiraud et al., 2012). HIIT is characterized by repeated bouts of high-intensity exercise interspersed by rest periods or low-intensity exercise for recovery (Gibala et al., 2012). HIIT is recommended for patients with coronary heart disease because of the higher tolerability of this type of exercise than continuous exercise programs (Guiraud et al., 2012). In this regard, studies have been performed exercise training on BNP1-32 and NT-proBNP1-76 levels in cardiac patients. For example, Pearson et al. [ 2018] reported a positive effect of exercise on BNP1-32 and NT-proBNP1-76 scores among patients with HF (Pearson et al., 2018). Wisløff et al. [ 2007] also examined the impacts of eight weeks of high-intensity interval training upon moderate-intensity continuous training. They showed a greater reduction of BNP1-32 in high-intensity interval training than in continuous training in HF patients (Wisloff et al., 2007). Additionally, many studies have reported the beneficial impacts of HIIT, but there are conflicts about the optimal training program compared to MCT in patients with coronary heart disease. Besides, the effects of these two training programs remained unclear on the processing and degradation of BNP1-32, as shown by the ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32. Therefore, the purpose of this study was to determine the impact of HIIT and MCT on plasma ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 in patients who underwent CABG. The chemical structure and production processing of BNP1-32 and NT-pro-BNP1-76 is depicted in Figure 1.
**FIGURE 1:** *Chemical structure and production processing of BNP1-32 and NT-pro-BNP1-76.*
## Subjects
The statistical population were all post CABG patients referred to Tehran Heart Center Hospital. Sample size were 36 patients because according to studies there must be at least chosen 30 samples to semi-experimental studies (Dietz and Kalof, 2009). so first of all we selected 36 patients purposive (27 men’s and 9 women) according our criteria and then allocated them to three groups according to homogeneity and gender randomly (9 men and 3 women in each group). The mean of their age were 60.32 ± 5.81 years old, height 164.64 ± 9.25 cm, weight 73.86 ± 14.23 kg, body mass index (BMI) 27.24 ± 3.90 kg/m2, systolic blood pressure (SBP) 142.67 ± 6.49, diastolic blood pressure 9 (DBP) 84.5 ± 5.16 mmHg, and functional capacity 7.08 ± 2.49 (METs). Two patients withdrew from the study during the experiment, and 34 patients were remained in our research. Table 3 illustrates patients’ characteristics for each group.
## Inclusion and exclusion criteria
All inclusion and exclusion criteria were considered according to patients’ medical records.
Inclusion criteria were as follows passing one month after following CABG surgery, low cardiovascular risk stratification (Thow, 2006), level 1 hypertension (SBP 140–159 mmHg or DBP 90–99 mmHg or both (Giles et al., 2005), and simultaneous consumption of aspirin, beta-blockers, anti-hypertensive drugs, and statins. Since all subjects were selected with low cardiovascular risk stratification and level 1 hypertension, the dosage of the drugs used by patients was almost the same. Furthermore, all patients had concentric pathologic hypertrophy (LVMI 1 more than116 g/m2 and RWT 2 was more than 0.42) (Lang et al., 2006).
Exclusion criteria were myocardial infarction, heart valve surgery history, ejection fraction <$30\%$, and movement limitation.
## Procedures
All procedures in this study have been endorsed by the Ethics Committee of Shahid Rajaee Teacher Training University of Tehran (IRSRTTU.SSF.2020.104). Informed consent was obtained from all participants. All measurements were performed as a pre-test after one week of familiarization and 48 h before the first exercise training session. After 2 months and 48 h after the last session of exercise training, all measurements were conducted as a post-test in similar conditions in which the pre-test was carried out. Exercise training was performed along with other routine rehabilitation programs such as psychological rehabilitation (counseling sessions to decrease anxiety and depression of patients), pharmacological treatments (aspirin, beta-blockers, anti-hypertensive drugs, and statins); lifestyle correction counseling (encouragement to increase the physical activity in daily life); diet modification and cessation of cigarette smoking performed in the cardiac rehabilitation clinic.
## Exercise training protocol
A comprehensive cardiac rehabilitation program (CR) includes six main aspects: 1) initial patient assessment, 2) nutritional counseling and weight management, 3) ongoing management of coronary risk factors, 4) psychological management, 5) physical activity counseling, and 6) exercise training (de Waard et al., 2021). In the cardiac rehabilitation clinic, exercise training programs were conducted three days a week over 8 weeks training groups performed exercise training on the treadmill (HP Cosmos, Germany) under the supervision of a physician, nurses, and the exercise physiologist. The rate of perceived exertion, rhythm, and arrhythmia was measured by Borg scale, ECG (United States; MHC 1,200), and remote control system (Iran, Avicenna Company; Telemetry), respectively, during each exercise training session. HIIT included 30 min exercises that consisted of 7 min warm-up (walking at $50\%$–$55\%$ of heart rate reserve (HRR), 4 intense intervals for 2.30 min at $80\%$–$95\%$ of HRR, 4 light intervals for 2.30 min at $65\%$–$80\%$ HRR between intense intervals, and finally 3 min cool-down (walking at $50\%$–$55\%$ of HRR). MCT included 33 min exercise on a treadmill that consisted of 7 min warm-up (walking at $50\%$–$55\%$ of HRR), followed by 23 min exercise on a treadmill at $65\%$–$80\%$ HRR, and, finally, 3 min cool-down (walking at $50\%$–$55\%$ of HRR). These protocols were similar to those used in a study by Wisloff et al. [ 2007] with slight modifications. We considered a similar training load (volume × intensity) to redesign the protocols for HIIT and MCT groups (Figure 2). Furthermore, the mode of intensity overload in the continuous and interval training group have been shown in Tables 1, 2. The maximum heart rate was defined according to the modified Bruce test result. The training heart rate was calculated by the Karvonen equation, as mentioned below (Sagiv, 2012). HRR=MHR attained−RHR THR=HRR * %intensity+RHR HRR, heart rate reserve; MHR, maximum heart rate attained at a peak stress test; and RHR, resting heart rate; THR, training heart rate.
**FIGURE 2:** *High Intensity Interval Training (HIIT) and Moderate Continuous Training (MCT) protocols.* TABLE_PLACEHOLDER:TABLE 1 TABLE_PLACEHOLDER:TABLE 2
## Laboratory measurements
All participants’ exact height (cm) and weight (kg) were measured for their BMI. Lange Skinfold caliper, 3-site formula, and Siri equation were used to estimate body fat percentage. Resting blood pressure was also measured by the digital blood pressure system (medical space labs made in the United States) from the brachial artery at a seating position. Blood samples (10 mL) were collected into the standard EDTA-containing tubes and immediately transferred to the laboratory to measure values at baseline and after the experiment. The sample plasma was separated by centrifugation at 3,000 rpm for about 10 min, then stored at −80°C until analysis. Corin, BNP1-32, proBNP1-108, and NT-pro-BNP1-76 were measured using the ELISA technique (kits: BOSTER Company made in United States Cat. No: EK1283, BIOMEDICA Company made in Austria Cat. No: BI-20852W and EASTBIOPHARM Company made in china Cat. No: CK-E90422, CK-E10219).
## Statistical analysis
The data were presented as mean and standard deviation (mean ± SD). The Kolmogorov-Simonov test was used to specify the normality of data distribution. We assessed normal distributed data (witness variables: blood pressure and subcutaneous fat, the ratios of proBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32) with the paired sample test and one-way ANOVA to determine within-group variations and between-group differences in the delta of means from pre-test to post-test, respectively. Kruskal-Wallis and Mann-Whitney U tests were used for non-normal distributed data (proBNP1-108, Corin, BNP1-32, and NT-pro-BNP1-76) to analyze within-group variations and between-group differences in the delta of means, from pre-test to post-test, respectively. A p-value of less than 0.05 was statistically considered significant for tests.
## Results
The plasma corin enzyme was increased, and the ratios of proBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 were reduced in both training groups in compared with control group ($$p \leq 0.004$$, $$p \leq 0000$$, $$p \leq 0.016$$, $$p \leq 0.003$$, $$p \leq 0.009$$, and $$p \leq 0.016$$) when there was no significant difference was found between training groups ($$p \leq 0.074$$, $$p \leq 450$$, and $$p \leq 0.295$$).
The statistical results are depicted in Tables 3–5. Within and between groups, changes in PROBNP1-108, CORIN, BNP1-32, NT-PROBNP1-76, ProBNP/BNP ratio and NTProBNP/BNP ratio are shown in Figures 3A–F.
## Discussion
The results of the present study indicated that 8-week HITT, in comparison with MCT, significantly decreased the plasma levels of ProBNP1-108, BNP1-32, and NT-pro-BNP1-76, on the other hand, increased plasma corin enzyme. Our findings mirror those of similar studies. For instance, Smart and Still [2010] showed the role of resistance and aerobic exercise in reducing BNP1-32 and NT-pro-BNP1-76 in HF patients (Smart and Steele, 2010). Santoso et al., 2020 have also demonstrated that aerobic exercise can reduce NT-pro-BNP1-76 and increase the left ventricular function in HF patients (Santoso et al., 2020). With this line, some studies have shown no effect of exercise on BNP1-32 dependent peptides, such as NT-pro-BNP1-76; for example, Zdrenghea et al. [ 2014] have investigated the effect of two methods, including exercise on the Ergometer and isometric handgrip test on NT-pro-BNP1-76 level of HF patients and reported any change. Perhaps the reason for the inconsistency of the mentioned study with the present study was the difference in the kind of patients HF patients versus post-CABG patients (in fact, HF is a secondary adverse consequence of CABG in long term. Although HF is a multifactorial problem, it seems hypertension, due to the pressure and volume overload on the heart, and with the direct effect on the structure and function of cardiomyocytes and fibroblasts, causes the gradual development of the fibrotic process, and pathologic hypertrophy. Finally, these structural changes lead to a rise in diastolic heart failure (impairment in optimal ventricular filling), systolic heart failure (impairment in Ejection fraction), and possibly patient mortality (Zile et al., 2011; Zare Karizak et al., 2017). So the improvement of the BNP processing system is important as an effective factor in blood pressure and consequently pathologic hypertrophy and Hf.), type, and duration of exercise (Isotonic and isometric exercise in single sessions versus eight weeks of aerobic interval and continuous training). As previously mentioned, BNP1-32 is primarily synthesized by the converting enzyme (corin) from an inactive prohormone (proBNP1-108) that is cleaved into the active hormone (BNP1-32) and inactive N-terminal fragment (NT-pro-BNP1-76). So in the present study probably, the reduction of prohormone (proBNP1-108) and increase of corin enzyme indicates an improvement in the production system of BNP1-32 due to an increase of converting enzyme (corin) in training groups. These results indicated an enhancement in the production system of BNP1-32; the amount of BNP1-32 and NT-pro-BNP1-76, as the final products, decreased in both training groups, especially in the HIIT group. This event may be owing to the regulatory role of BNP1-32, which could be increased as a compensatory mechanism in response to pressure overload, volume overload, and ischemic conditions (Fu et al., 2018b). Hence, its concentration is initially elevated in patients with cardiovascular disorders, while it could be decreased after 8-week HIIT. This phenomenon stems from the reduction of stress conditions. Consequently, studies indicated that body adaptation with HIIT could diminish the pressure overload by decreasing vasoconstriction factors, such as angiotensin II, endothelin, and increment in vasodilator agents, such as nitric oxide and prostaglandins (Passino et al., 2006) and decrease of adrenomedullin as a stress index on heart (Zarekarizak et al., 2021). HIIT also decreases volume overload by inhibiting the renin-angiotensin-aldosterone system (Gademan et al., 2007). Other studies have also indicated that HIIT can attenuate the ischemic condition by incrementing the quantity and quality of micro-vessels and improving oxygen delivery to the heart (Macheret et al., 2011). Besides, our findings confirmed that HIIT could dramatically decrease blood pressure compared to other groups. Furthermore, while body fat reduction is associated with an increased ratio of the biological natriuretic peptide receptors (NPR-A to NPR-C), it has been suggested that the body fat percentage influences the sensitivity of the human body to BNP1-32 (Dessì-Fulgheri et al., 1997; Marney et al., 2014; Arora et al., 2015). The reduction in body fat percentage after exercise training appears to correlate with the reduction of BNP1-32 and an increase in the body’s sensitivity to BNP1-32. Our results showed that the rate of the reduction of the body fat percentage is significantly higher at HIIT in comparison with MCT. On the other hand, the ratio of ProBNP1-108/BNP1-32 was increased in the control group compared with other groups. As previously mentioned, the ratio of ProBNP1-108/BNP1-32 is an indicator of the production system of BNP1-32. The increment ratio in the control group is associated with the impaired production system of BNP1-32, which may result from the reduction of the corin enzyme of the control group in the present study. Furthermore, the reduction of degradation enzymes of proBNP1-108 has disrupted the BNP1-32 production system in the control group likely and increased the ProBNP1-108/BNP1-32 ratio in this group. In contrast, the ratio of ProBNP1-108/BNP1-32 was remarkably decreased in plasma levels of both training groups compared with the control group. However, since ProBNP1-108/BNP1-32 decreased in both training groups, the rate of change in the ratio ProBNP1-108/BNP1-32 would remain unchanged. Therefore, no significant difference was found in that ratio between training groups. The statistical analysis demonstrated a significant difference in the ratio of ProBNP1-108/BNP1-32 between the HIIT and control groups and between the MCT and control groups. The ratio of NT-pro-BNP1-76/BNP1-32 was also increased in the group of control but decreased in the training groups. The ratio of NT-pro-BNP1-76/BNP1-32 is an indicator of the degradation system of BNP1-32. Like the production system of BNP1-32, its degradation system was impaired in patients and exacerbated, resulting from the inactivity of the control group. BNP1-32 is cleaved in several ways, such as guanylyl cyclase receptor type C and glomerular filtration (Barnet et al., 2015; Huntley et al., 2015), while the breakdown of NT-pro-BNP1-76 is exclusively performed by glomerular filtration (Holm, 2013). Also, inflammation is associated with a decreased breakdown of NT-pro-BNP1-76 compared to BNP1-32 and an increased ratio of NT-pro-BNP1-76/BNP1-32 (Jensen et al., 2010). Several lines of evidence indicated that exercise training decreases inflammation in post-CABG patients (Goto, 2010), so it could be an influential factor in the degradation of BNP1-32 and NT-pro-BNP1-76. However, this possible mechanism was not examined in our study and was considered a limitation of the experiment. There was no significant difference between the training groups between ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 ratios. Therefore, both HIIT and MCT may be useful to regulate production and degradation systems for BNP1-32 in underwent CABG patients. However, due to the multiplicity of cardiac protection pathways and HF prevention, the lack of more mechanisms was the limitation of this study. For example, BNP1-32 is cleaved by several enzymes, including dipeptidyl peptidase-4 (DPPIV), neprilysin (NEP), and insulin-degrading enzyme (IDE) so further studies are required to survey the effects of MCT and HIIT on these mechanisms. Furthermore, the number of patients was small and this limits the reliability of the results and was limitation of our study. it should be consider more patients for future studies.
## Conclusion
The inactivity of post-CABG patients harms the corin enzyme; ProBNP1-108/BNP1-32 ratios and NT-pro-BNP1-76/BNP1-32, while both HIIT and MCT have a positive effect on ratios of ProBNP1-108/BNP1-32 and NT-pro-BNP1-76/BNP1-32 and could be effective to promote the health of coronary arteries and prevention of HF in post-CABG patients.
## Data availability statement
The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by the Ethics Committee of Shahid Rajaee Teacher Training University of Tehran (IRSRTTU.SSF.2020.104). Informed consent was obtained from all participants. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
All authors have been involved in gathering information, writing, and reviewing the 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.
## References
1. Arora P., Reingold J., Baggish A., Guanaga D. P., Wu C., Ghorbani A.. **Weight loss, saline loading, and the natriuretic peptide system**. *J. Am. Heart Assoc.* (2015) **4** e001265. DOI: 10.1161/JAHA.114.001265
2. Barnet C. S., Liu X., Body S. C., Collard C. D., Shernan S. K., Muehlschlegel J. D.. **Plasma corin decreases after coronary artery bypass graft surgery and is associated with postoperative heart failure: A pilot study**. *J. Cardiothorac. Vasc. Anesth.* (2015) **29** 374-381. DOI: 10.1053/j.jvca.2014.11.001
3. Chopra S., Cherian D., Verghese P. P., Jacob J. J.. **Physiology and clinical significance of natriuretic hormones**. *Indian J. Endocrinol. metabolism* (2013) **17** 83-90. DOI: 10.4103/2230-8210.107869
4. Cockburn J., Blows L., Cohen A., Holmberg S., Hyde J., Lewis M.. **Acute ischemic complications of PCI and CABG: Who should cover whom for coronary revascularization?**. *J. Interventional Cardiol.* (2013) **26** 372-377. DOI: 10.1111/joic.12045
5. Conraads V. M., Beckers P. J.. **Exercise training in heart failure: Practical guidance**. *Heart* (2010) **96** 2025-2031. DOI: 10.1136/hrt.2009.183889
6. de Waard D., Fagan A., Minnaar C., Horne D.. **Management of patients after coronary artery bypass grafting surgery: A guide for primary care practitioners**. *Can. Med. Assoc. J.* (2021) **193** E689-E694. DOI: 10.1503/cmaj.191108
7. Dendale P., Berger J., Hansen D., Vaes J., Benit E., Weymans M.. **Cardiac rehabilitation reduces the rate of major adverse cardiac events after percutaneous coronary intervention**. *Eur. J. Cardiovasc. Nurs.* (2005) **4** 113-116. DOI: 10.1016/j.ejcnurse.2004.11.003
8. Dessì-Fulgheri P., Sarzani R., Tamburrini P., Moraca A., Espinosa E., Cola G.. **Plasma atrial natriuretic peptide and natriuretic peptide receptor gene expression in adipose tissue of normotensive and hypertensive obese patients**. *J. Hypertens.* (1997) **15** 1695-1699. DOI: 10.1097/00004872-199715120-00074
9. Dietz T., Kalof L.. *Introduction to social statistics: The logic of statistical reasoning* (2009)
10. Fox A. A., Marcantonio E. R., Collard C. D., Thoma M., Perry T. E., Shernan S. K.. **Increased peak postoperative B-type natriuretic peptide predicts decreased longer-term physical function after primary coronary artery bypass graft surgery**. *Anesthesiol. J. Am. Soc. Anesthesiol.* (2011) **114** 807-816. DOI: 10.1097/ALN.0b013e31820ef9c1
11. Fu S., Ping P., Wang F., Luo L.. **Synthesis, secretion, function, metabolism and application of natriuretic peptides in heart failure**. *J. Biol. Eng.* (2018) **12** 2. DOI: 10.1186/s13036-017-0093-0
12. Fu S., Ping P., Zhu Q., Ye P., Luo L.. **Brain natriuretic peptide and its biochemical, analytical, and clinical issues in heart failure: A narrative review**. *Front. physiology* (2018) **9** 692. DOI: 10.3389/fphys.2018.00692
13. Gademan M. G., Swenne C. A., Verwey H. F., Van Der Laarse A., Maan A. C., Van De Vooren H.. **Effect of exercise training on autonomic derangement and neurohumoral activation in chronic heart failure**. *J. cardiac Fail.* (2007) **13** 294-303. DOI: 10.1016/j.cardfail.2006.12.006
14. Ghashghaei F. E., Sadeghi M., Marandi S. M., Ghashghaei S. E.. **Exercise-based cardiac rehabilitation improves hemodynamic responses after coronary artery bypass graft surgery**. *ARYA Atheroscler.* (2012) **7** 151-156. PMID: 23205048
15. Gibala M. J., Little J. P., MacDonald M. J., Hawley J. A.. **Physiological adaptations to low-volume, high-intensity interval training in health and disease**. *J. physiology* (2012) **590** 1077-1084. DOI: 10.1113/jphysiol.2011.224725
16. Giles T. D., Berk B. C., Black H. R., Cohn J. N., Kostis J. B., Izzo J. L.. **Expanding the definition and classification of hypertension**. *J. Clin. Hypertens.* (2005) **7** 505-512. DOI: 10.1111/j.1524-6175.2005.04769.x
17. Goto Y.. **Exercise training in post-CABG patients at low prognostic risk–beyond recovery from surgery–**. *Circulation J.* (2010) **74** 2548-2549. DOI: 10.1253/circj.cj-10-1061
18. Guiraud T., Nigam A., Gremeaux V., Meyer P., Juneau M., Bosquet L.. **High-intensity interval training in cardiac rehabilitation**. *Sports Med.* (2012) **42** 587-605. DOI: 10.2165/11631910-000000000-00000
19. Holm J.. *Markers of hemodynamic state and heart failure as predictors for outcome in cardiac surgery: With special reference to mixed venous oxygen saturation and natriuretic peptides* (2013)
20. Huntley B. K., Sandberg S. M., Heublein D. M., Sangaralingham S. J., Burnett J. C., Ichiki T.. **Pro–B-Type natriuretic peptide-1-108 processing and degradation in human heart failure**. *Circ. Heart Fail.* (2015) **8** 89-97. DOI: 10.1161/CIRCHEARTFAILURE.114.001174
21. Ichiki T., Boerrigter G., Huntley B. K., Sangaralingham S. J., McKie P. M., Harty G. J.. **Differential expression of the pro-natriuretic peptide convertases corin and furin in experimental heart failure and atrial fibrosis. American Journal of Physiology-Regulatory**. *Integr. Comp. Physiology* (2013) **304** R102-R9
22. Ichiki T., Huntley B. K., Heublein D. M., Sandberg S. M., McKie P. M., Martin F. L.. **Corin is present in the normal human heart, kidney, and blood, with pro–B-type natriuretic peptide processing in the circulation**. *Clin. Chem.* (2011) **57** 40-47. DOI: 10.1373/clinchem.2010.153908
23. Jensen J., Ma L-P., Bjurman C., Hammarsten O., Fu M. L.. **Prognostic values of NTpro BNP/BNP ratio in comparison with NTpro BNP or BNP alone in elderly patients with chronic heart failure in a 2-year follow up**. *Int. J. Cardiol.* (2012) **155** 1-5. DOI: 10.1016/j.ijcard.2011.01.083
24. Jensen J., Ma L-P., Fu M. L., Svaninger D., Lundberg P-A., Hammarsten O.. **Inflammation increases NT-proBNP and the NT-proBNP/BNP ratio**. *Clin. Res. Cardiol.* (2010) **99** 445-452. DOI: 10.1007/s00392-010-0140-z
25. Lang R. M., Bierig M., Devereux R. B., Flachskampf F. A., Foster E., Pellikka P. A.. **Recommendations for chamber quantification**. *Eur. J. Echocardiogr.* (2006) **7** 79-108. DOI: 10.1016/j.euje.2005.12.014
26. Macheret F., Boerrigter G., McKie P., Costello-Boerrigter L., Lahr B., Heublein D.. **Pro–B-Type natriuretic peptide1–108 circulates in the general community: Plasma determinants and detection of left ventricular dysfunction**. *J. Am. Coll. Cardiol.* (2011) **57** 1386-1395. DOI: 10.1016/j.jacc.2011.01.005
27. Marney A. M., Brown N. J., Tamboli R., Abumrad N.. **Changes in B-type natriuretic peptide and BMI following Roux-en-Y gastric bypass surgery**. *Diabetes care* (2014) **37** e70-e71. DOI: 10.2337/dc13-2449
28. Meyer P., Gayda M., Juneau M., Nigam A.. **High-intensity aerobic interval exercise in chronic heart failure**. *Curr. heart Fail. Rep.* (2013) **10** 130-138. DOI: 10.1007/s11897-013-0130-3
29. Miller W. L., Grill D. E., Jaffe A. S.. **Comparison of novel pro-BNP1–108 and standard BNP assays in heart failure patients**. *Clin. Chim. acta* (2012) **413** 920-926. DOI: 10.1016/j.cca.2012.02.007
30. Nishitani M., Shimada K., Masaki M., Sunayama S., Kume A., Fukao K.. **Effect of cardiac rehabilitation on muscle mass, muscle strength, and exercise tolerance in diabetic patients after coronary artery bypass grafting**. *J. Cardiol.* (2013) **61** 216-221. DOI: 10.1016/j.jjcc.2012.11.004
31. Passino C., Severino S., Poletti R., Piepoli M. F., Mammini C., Clerico A.. **Aerobic training decreases B-type natriuretic peptide expression and adrenergic activation in patients with heart failure**. *J. Am. Coll. Cardiol.* (2006) **47** 1835-1839. DOI: 10.1016/j.jacc.2005.12.050
32. Pearson M. J., King N., Smart N. A.. **Effect of exercise therapy on established and emerging circulating biomarkers in patients with heart failure: A systematic review and meta-analysis**. *Open heart* (2018) **5** e000819. DOI: 10.1136/openhrt-2018-000819
33. Preeshagul I., Gharbaran R., Jeong K. H., Abdel-Razek A., Lee L. Y., Elman E.. **Potential biomarkers for predicting outcomes in CABG cardiothoracic surgeries**. *J. Cardiothorac. Surg.* (2013) **8** 176. DOI: 10.1186/1749-8090-8-176
34. Sagiv M. S.. *Exercise cardiopulmonary function in cardiac patients: Springer science and business media* (2012)
35. Santoso A., Maulana R., Alzahra F., Prameswari H. S., Ambari A. M., Hartopo A. B.. **The effects of aerobic exercise on N-terminal pro-B-type natriuretic peptide and cardiopulmonary function in patients with heart failure: A meta-analysis of randomised clinical trials**. *Lung Circulation* (2020) **29** 1790-1798. DOI: 10.1016/j.hlc.2020.05.098
36. Serruys P. W., Unger F., Sousa J. E., Jatene A., Bonnier H. J., Schönberger J. P.. **Comparison of coronary-artery bypass surgery and stenting for the treatment of multivessel disease**. *N. Engl. J. Med.* (2001) **344** 1117-1124. DOI: 10.1056/NEJM200104123441502
37. Siribaddana S.. **Cardiac dysfunction in the CABG patient**. *Curr. Opin. Pharmacol.* (2012) **12** 166-171. DOI: 10.1016/j.coph.2012.01.010
38. Smart N. A., Steele M.. **Systematic review of the effect of aerobic and resistance exercise training on systemic brain natriuretic peptide (BNP) and N-terminal BNP expression in heart failure patients**. *Int. J. Cardiol.* (2010) **140** 260-265. DOI: 10.1016/j.ijcard.2009.07.004
39. Suzuki S., Sugiyama S.. **The molar ratio of N-terminal pro-B-type natriuretic peptide/B-type natriuretic peptide for heart failure-related events in stable outpatients with cardiovascular risk factors**. *Intern. Med.* (2018) **57** 2621-2630. DOI: 10.2169/internalmedicine.0471-17
40. Thow M.. *Exercise leadership in cardiac rehabilitation: An evidence-based approach* (2006)
41. Valkeinen H., Aaltonen S., Kujala U.. **Effects of exercise training on oxygen uptake in coronary heart disease: A systematic review and meta-analysis**. *Scand. J. Med. Sci. sports* (2010) **20** 545-555. DOI: 10.1111/j.1600-0838.2010.01133.x
42. Wisloff U., Stoylen A., Loennechen J. P., Bruvold M., Haram P. M., Tjonna A. E.. **Superior cardiovascular effect of aerobic interval-training versus moderate continuous training in elderly heart failure patients: 651May 31 8: 15 AM-8: 30 am**. *Med. Sci. Sports Exerc.* (2007) **39** S32. DOI: 10.1249/01.mss.0000273010.06226.99
43. Zare Karizak S., Kashef M., Gaeini A. A., Nejatian M.. **The comparison of two protocol of interval and continues aerobic training on level of concentric pathologic hypertrophy and cardiac function in patients after coronary artery bypass grafting surgery**. *J. Pract. Stud. Biosci. Sport* (2017) **5** 9-20
44. Zarekarizak S., Kashef M., Gaeini A. A.. **The comparison of eight weeks interval and continuous training on PRO-BNP/CORIN system in coronary artery disease patients after CABG surgery**. *Sport Physiol.* (2017) **9** 131-152
45. Zarekarizak S., Kashef M., Nejatian M., Hamid R.. **Variations of plasma adrenomedullin, ventricular ejection fraction and resting rate pressure product, following rehabilitation programs of interval and continuous training after coronary artery bypass grafting surgery**. *Journal for Research in Sport Rehabilitation* (2021) **8** 15-23
46. Zdrenghea D. T., Ilea M., Bodizs G., Sitar-Tăut A., Zdrenghea M., Pop D.. **NT-pro-BNP during isotonic and isometric exercise in heart failure patients with preserved LV ejection fraction**. *Clin. Lab.* (2014) **60** 2055-2061. DOI: 10.7754/clin.lab.2014.140417
47. Zile M. R., DeSantis S. M., Baicu C. F., Stroud R. E., Thompson S. B., McClure C. D.. **Plasma biomarkers that reflect determinants of matrix composition identify the presence of left ventricular hypertrophy and diastolic heart failure**. *Circ. Heart Fail.* (2011) **4** 246-256. DOI: 10.1161/CIRCHEARTFAILURE.110.958199
|
---
title: 'A pseudo-outbreak of Cyberlindnera fabianii funguria: Implication from whole
genome sequencing assay'
authors:
- Xin Fan
- Rong-Chen Dai
- Timothy Kudinha
- Li Gu
journal: Frontiers in Cellular and Infection Microbiology
year: 2023
pmcid: PMC10030058
doi: 10.3389/fcimb.2023.1130645
license: CC BY 4.0
---
# A pseudo-outbreak of Cyberlindnera fabianii funguria: Implication from whole genome sequencing assay
## Abstract
### Background
Although the yeast *Cyberlindnera fabianii* (C. fabianii) has been rarely reported in human infections, nosocomial outbreaks caused by this organism have been documented. Here we report a pseudo-outbreak of C. fabianii in a urology department of a Chinese hospital over a two-week period.
### Methods
Three patients were admitted to the urology department of a tertiary teaching hospital in Beijing, China, from Nov to Dec 2018, for different medical intervention demands. During the period Nov 28 to Dec 5, funguria occurred in these three patients, and two of them had positive urine cultures multiple times. Sequencing of rDNA internal transcribed spacer (ITS) region and MALDI-TOF MS were applied for strain identification. Further, sequencing of rDNA non-transcribed spacer (NTS) region and whole genome sequencing approaches were used for outbreak investigation purpose.
### Results
All the cultured yeast strains were identified as C. fabianii by sequencing of ITS region, and were $100\%$ identical to the C. fabianii type strain CBS 5640T. However, the MALDI-TOF MS system failed to correctly identify this yeast pathogen. Moreover, isolates from these three clustered cases shared $99.91\%$-$100\%$ identical NTS region sequences, which could not rule out the possibility of an outbreak. However, whole genome sequencing results revealed that only two of the C. fabianii cases were genetically-related with a pairwise SNP of 192 nt, whilst the third case had over 26,000 SNPs on its genome, suggesting a different origin. Furthermore, the genomes of the first three case strains were phylogenetically even more diverged when compared to a C. fabianii strain identified from another patient, who was admitted to a general surgical department of the same hospital 7 months later. One of the first three patients eventually passed away due to poor general conditions, one was asymptomatic, and other clinically improved.
### Conclusion
In conclusion, nosocomial outbreaks caused by emerging and uncommon fungal species are increasingly being reported, hence awareness must be raised. Genotyping with commonly used universal gene targets may have limited discriminatory power in tracing the sources of infection for these organisms, requiring use of whole genome sequencing to confirm outbreak events.
## Introduction
Emerging fungal infections have become a global health concern in the past few decades due to their notable morbidity and mortality, especially among immunosuppressed patients admitted to intensive care units (ICUs), or undergoing invasive medical interventions (Pappas et al., 2018; Hoenigl et al., 2022; World Health Organization, 2022). Although Candida albicans remains the most predominant yeast pathogen, the incidence of uncommon yeast species causing human infections has increased enormously in recent years (Pappas et al., 2018; Chen et al., 2021). Uncommon yeast species often exhibit decreased susceptibility to commonly used antifungal agents, making them difficult to manage in clinical settings. Moreover, there are increasing incidences of nosocomial infections and outbreak events reported due to transmission of uncommon or emerging yeast species worldwide (Pappas et al., 2018; Chen et al., 2021). For instance, Candida auris, which was first described in 2009, has caused a number of outbreaks in different continents (Chow et al., 2018; Hoenigl et al., 2022; World Health Organization, 2022).
In the investigations and tracing of fungal nosocomial transmissions, molecular genotyping could provide essential genetic evidence. Hence, a wide variety of molecular typing assays have been evaluated and implemented in the study of outbreaks, including band-pattern-based DNA analysis like random amplified polymorphic DNA (RAPD) and pulsed field gel electrophoresis (PFGE), traditional DNA sequencing-based phylogenetic methods like single gene analysis, microsatellite analysis and multilocus sequence typing (MLST), protein spectrum analysis by matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS), and whole genome sequencing (WGS) techniques (Reiss et al., 1998; Pulcrano et al., 2012; Mikosz et al., 2014; Xiao et al., 2014; Litvintseva et al., 2015; Oliveira and Azevedo, 2022). Of these methods, WGS has become increasingly used due to its outstanding discriminatory power (Litvintseva et al., 2015; Bougnoux et al., 2018; Desnos-Ollivier et al., 2020; Oliveira and Azevedo, 2022).
In this study, we report on three clustered funguria cases caused by a rare fungal pathogen, Cyberlindnera fabianii, which occurred over a two-week period within the same urology department, which was initially considered as a nosocomial outbreak event. Using WGS, this event was finally confirmed as a pseudo-outbreak caused by C. fabianii from two diverged genetic lineages.
## Ethics
This study was approved by the Human Research Ethics Committee of the Beijing Chaoyang Hospital (No. KE332). Written informed consent was obtained from all participants involved.
## Routine isolation of the microorganisms and MALDI-TOF MS identification.
C. fabianii strains were isolated from urine samples of three patients (number 1-3); on four different occasions for patient number 1, only once for patient number 2, and on three different occasions for patient number 3 (Figure 1 and Table 1). After these three cases, no further C. fabianii cases were detected in the same hospital for over seven months, till a new C. fabianii isolate, cultured from an ascites sample of a patient admitted to general surgery department (recorded as patient number 4), was detected, and this isolate was used as a comparator.
**Figure 1:** *Clinical features, treatment regimens, and outcomes of three clustered cases with *Cyberlindnera fabianii* funguria. Abbreviations: CC, CFU (colony forming unit) count; LOS, length of stay; Culture +: culture positive for C. fabianii.* TABLE_PLACEHOLDER:Table 1 Routine culture of specimens was carried out as per standard laboratory protocols. Generally, for urine samples, 1 μl of the sample was inoculated on Blood *Agar media* and incubated at 35 °C for 24 h. Thereafter, the number of colonies growing on the media plate was counted to ensure that they met the criterion for a urinary tract infection. A brief identification protocol revealed that the cultured isolates were yeast. Thus, Sabouraud glucose agar (SDA) was used to subculture these isolates for further identification testing. Attempts were made to identify the cultured yeasts by using a Vitek MS MALDI-TOF MS system (bioMérieux, Marcy l’Etoile, France, with IVD database version 2.1), following manufacturer’s instructions. For each run, *Escherichia coli* strain ATCC 8739 was used to calibrate and control the method. Unfortunately, this system was unable to identify the yeast strains.
## Molecular identification and phylogenetic analysis by rDNA gene spacer regions
As all the suspected yeast isolates could not be identified using the MALDI-TOF MS systems, sequencing of rDNA internal transcribed spacer (ITS) regions was carried out. Briefly, DNA extraction of the isolates was performed using a QIAamp DNA Mini Kit (Qiagen, Hilden, Germany). The universal primer pair ITS1 and ITS4 was used for amplification and sequencing of the ITS region for each strain (Xiao et al., 2014), and a species identification was carried out by querying against the Westerdijk Fungal Biodiversity Institute’s database using a web-based pairwise alignment tool (https://wi.knaw.nl/page/Pairwise_alignment).
Further, to investigate the potential relatedness of these cases, the first yeast isolate of each patient case was chosen for further testing, and the rDNA non-transcribed spacer region 1 (NTS-1) was amplified with a forward primer NTS1-F (5’-GGGATAAATCATTTGTATACGAC-3’) and a reverse primer NTS1-R (5’-TTGCGGCCATATCCACAAGAAA-3’) as described previously (Al-Sweih et al., 2019), and sequenced from both directions. A phylogenetic tree of NTS-1 sequences was generated by Mega X (version 10.2, https://www.megasoftware.net/) using neighbor-joining method with bootstrap value of 1000. NTS-1 sequences from C. fabianii type strain CBS 5640T, and C. fabianii reference genome strain JOY008, were also downloaded from GenBank and included in the analysis. In addition, NTS-1 sequence extracted from the genome of *Cyberlindnera jadinii* strain NBRC 0988 was selected as an outgroup.
## Whole genome sequencing and analysis of C. fabianii strains
Whole-genome sequencing was performed on each of the first yeast strain from each of patients 1 to 4. Generally, a 350-bp DNA library was prepared using NEB Next Ultra DNA library prep kits (NEW ENGLAND BioLabs Inc., MA, USA), following the manufacturer’s instructions. Library integrity was evaluated with an Agilent 2100 Bioanalyzer (Agilent Technologies, CA, USA). Sequencing was performed on an Illumina NovaSeq using PE150 in a commercial company (Novogene Co., Ltd., Beijing, China).
For genome analysis, the complete reference genome of C. fabianii strain JOY008 (GenBank accession no. GCA_022641835.1) was used for read mapping. Single-nucleotide polymorphism (SNP) analysis was carried out by Burrows-Wheeler Aligner (version 0.7.7), SAMtools (version 1.2), and Genome Analysis Toolkit (GATK) (v.3.3-0) per GATK Best Practices (Li and Durbin, 2009; Li et al., 2009; Mckenna et al, 2010). The filtered reads were compared to the reference genome by SAMtools to generate BAM files. Then, variants were marked by GATK MarkDuplicates for each sample, and single-sample GVCF files were created by GATK HaplotypeCaller with the option –emitRefConfidence GVCF. The GVCF files were aggregated by GATK CombineGVCFs tool. After that, the GVCF files were jointly genotyped with the GATK GenotypeGVCFs to produce a single VCF file containing variants data on every strain. Finally, the VCF file was selected using GATK SelectVariants with the option -select-type SNP and filtered using the following parameters: VariantFiltration, QD < 2.0, ReadPosRankSum < −8.0, FS > 60.0, MQRankSum < −12.5, MQ < 40.0 and HaplotypeScore > 13.0.
Specifically, in this study, the term “pseudo-outbreak” was used to describe inappropriate artifactual clustering of real infections as an outbreak event, due to limitation of investigation tools.
## Antifungal susceptibility testing
Minimum inhibitory concentrations (MICs) of all the isolates were determined by Sensititre YeastOne YO10 kits (Thermo Scientific, OH, USA) following the manufacturer’s instructions, and with Candida krusei ATCC 6258 and *Candida parapsilosis* ATCC 22019, used as quality control strains.
## Data availability
DNA sequences of rDNA ITS and NTS-1 regions for each of the first yeast strain isolated from each individual has been deposited into NCBI GenBank database (accession nos. OP904191-OP904194 for ITS region and OP912967-OP912970 for NTS-1 region). Their WGS reads data is also now available in National Microbiology Data Center (NMDC) database (Bioproject accession no. PRJNA907923).
## Patients
Information pertaining to each of the 4 patients included in this case study is summarized in Figure 1 and Table 1.
Patient 1 was a 65-year-old female admitted to the urology department of Beijing Chao-Yang hospital on Nov 22, 2018, due to presence of fever for two weeks and a parastomal fistula after ileal replacement due to bladder cancer in 2016. Upon admission, the patient had fever for over a week. On day 6 after admission, a yeast strain was isolated from her urine sample and the colony count (CC) was 8×104 CFU/ml. The same urine culture also grew *Enterococcus faecium* (5×104 CFU/ml) as a mixed culture with the yeast. The routine MALDI-TOF MS identification system failed to identify the yeast isolate. Her urine samples collected on days 8, 13 and 15 after admission also yielded yeasts (CC of 8 to >10×104 CFU/ml). Follow-up ITS sequencing assigned all the strains as C. fabianii. The patient was given fluconazole at 200 mg/day for 18 days after which her condition improved notably, and she was finally discharged from the hospital on day 33 of admission.
Patient 2 was an 83-year-old male admitted to the urology department of the hospital on Dec 04, 2018, for follow-up of bladder cancer electrosurgery performed eight and four months before his admission. On day 1 after admission, a urine sample was collected for routine screening, which was reported positive for yeasts with a CC of 5×104 CFU/ml. The yeast strain was identified as C. fabianii by ITS sequencing. This patient didn’t present with any symptoms of infection, and hence antifungal therapy was not given. Later, he received a transurethral resection of bladder tumor on day 3, and was discharged on day 7 after admission.
Patient 3 was a 65-year-old male admitted to the urology department of the hospital on Dec 03, 2018, due to presence of high fever with backaches. Nine months before this admission, the patient hand undergone nephroureterectomy of the left kidney. He received nephrostomy on the right kidney immediately on day 1 after admission. His urine sample collected on day 2 was culture positive for a yeast (CC: 8×104 CFU/ml), which was identified as C. fabianii by ITS sequencing. However, no antifungal agents were prescribed for him and only a broad-spectrum antibiotic was given. On days 26 and 27, two urine samples were collected consecutively and both were positive for C. fabianii with a CC of > 10×104 CFU/ml. Of note, all his urine samples also grew E. faecium (>10×104 CFU/ml) as part of a mixed culture with the yeast. Though fluconazole therapy (200 mg/day) was initiated on day 27 after admission, the patient passed away on the same day.
Patient 4 was a 54-year-old female admitted to the general surgery department of the hospital on Jul 18, 2019, which was over seven months after the patient 1, 2 and 3 case clusters. She was hospitalized due to pancreatic cancer, and received radical pancreatoduodenectomy on day 12 after admission; later with pancreatic intestinal anastomotic fistula. The patient’s ascites sample collected on day 20 was reported positive for C. fabianii and E. faecium. However, she didn’t have any other culture positive results for fungi after that, nor received any antifungal treatment, and was discharged from the hospital on day 52.
## C. fabianii identification
All the yeast strains could not be identified using the Vitek MS MALDI-TOF MS system IVD 2.0 database, nor were the isolates misidentified as something else (identification confidence values <60.0). This is not surprising as C. fabianii is not currently included in the system’s spectrum database.
By sequencing of the ITS region, all the yeast strains from the four patients were unambiguously assigned to C. fabianii, with their respective ITS sequences $100\%$ ($\frac{602}{602}$ bp) identical to that of C. fabianii type strain CBS 5640T.
## Phylogenetic analysis by rDNA NTS-1 region
Since C. fabianii is a rare yeast species identified in human patients, and the fact that the clustered cases (patients 1 to 3) described here were identified within a two-week period from the same department, an investigation was carried out to assess the possibility of this being a nosocomial outbreak. Owing to the high sequence similarity of the ITS region among the strains, sequence analysis based on rDNA NTS-1 region was further carried out, which was assumed to have higher discriminatory power and has previously been used to confirm a C. fabianii outbreak in Kuwait (Al-Sweih et al., 2019).
Using C. jadinii (strain NBRC 0988) as an outgroup, the phylogenetic tree based on the NTS-1 region clustered all the C. fabianii isolates together (Figure 2), and inter-species variation between C. jadinii and C. fabianii in NTS-1 region was >$43.5\%$. Amongst C. fabianii strains, some intra-species variation was observed (Figure 2). However, sequence variations amongst the strains from patients 1 to 3 was inconspicuous, as these strains exhibited the same sequence type, while the strain from patient 2 had only one SNP (identity $\frac{1179}{1180}$, $99.92\%$). In contrast, strains from the three clustered cases were quite diverged from the strain from patient 4 which was isolated seven months later, which had an overall 7-bp insertions and 4 additional SNPs in its NTS-1 region (identity $99.07\%$).
**Figure 2:** *Phylogenetic trees generated based on rDNA non-transcribed spacer (NTS) region-1 sequences and whole genome sequencing (WGS) SNPs, and heatmaps revealing pairwise differences of SNPs amongst four patients’ strains collected in this study.*
## Whole genome sequencing results
Genome sequencing of yeast strains from patients 1 to 4 produced 2.6 to 3.7 Gb of clean data, and average depths of sequencing were all above 200×. The average genome size obtained was 12.94 Mb. Their genomes had an average GC content of $44.4\%$ to $45.1\%$, with N50 of 13,739 bp to 202,514 bp. Comparative genomic analysis was performed for all strains. The pairwise differences between genome reference strain JOY008, which originated from a soil environment in USA, and our four patients’ clinical strains, were 29,810-53,490 SNPs (Figure 2).
We carried out a review of previous outbreak reports caused by different yeast species, and the number of pairwise SNPs described varied from less than ten to several hundred (Table 2). The yeast strains from patient 1 and 3 had only 192 SNPs identified, suggesting that they were probably closely related (Figure 2). However, there were over 26,000 SNPs identified between the strain from patient 2 and strains from patients 1 and 3 (Figure 2), which was considered as a high-level genomic variation. These findings suggested that the C. fabianii strain from patient 2 was from a different origin. In addition, the yeast strain from patient 4 was even more diverged, with >43,000 of SNPs compared to all strains from patients 1 to 3, and the reference genome (Figure 2). Lastly, the phylogenetic tree constructed based on whole genome SNPs also support the same conclusion (Figure 2).
**Table 2**
| Species | Reference Genomesize (Mb) | Country | Patient population | Ward | No. of caseswith WGS | No. of SNPswithin each event | Reference |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Candida albicans | 14.3 | Spain | Neonate | ICU | 2-11 | 134-769 | (Guinea et al., 2021) |
| Candida parapsilosis | 13.0 | Spain | Neonate | ICU | 2-4 | 49-241 | (Guinea et al., 2021) |
| Candida auris | 12.7 | US | Adults | Not specified | 26 | 2-50 | (De St Maurice et al., 2022) |
| | | India | Adults | Medical wards | 2-2 | ≤7 | (Yadav et al., 2021) |
| | | UK | Adults | ICU, high dependency units,surgical admission ward | 5-17 | ≤134 | (Rhodes et al., 2018) |
| | | UK | Adults | ICU, neurosciences wards | 37 | ≤215 | (Eyre et al., 2018) |
| | | Colombia | Not specified | Not specified | 5 | ≤40 | (Escandon et al., 2019) |
| | | USA | Not specified | Not specified | 10 | ≤12 | (Chow et al., 2018) |
| Dirkmeia churashimaensis | 21.0 | India | Neonate | ICU | 6 | 1621 | (Chowdhary et al., 2020a) |
| Candida blankii | 14.8 | India | Neonate | ICU | 6 | ≤277 | (Chowdhary et al., 2020b) |
| Malassezia pachydermatis | 8.2 | USA | Neonate | ICU | 5 | ≤14 | (Chow et al., 2020) |
| Cyberlindnera fabianii | 12.3 | China | Adults | Urology department | 2 | 192 | This study |
## Antifungal susceptibilities
All the C. fabianii strains isolated in this study showed good susceptibility to all the nine antifungal agents tested (Table 1), with geometric minimum inhibitory concentration (GM MIC) of 0.84 mg/L to fluconazole, 0.02 mg/L to voriconazole, 0.08 mg/L to itraconazole, 0.12 mg/L to posaconazole, 0.04 mg/L to caspofungin, 0.03 mg/L to micafungin, 0.02 mg/L to anidulafungin, 0.08 mg/L to 5-flucytosine, and finally, 0.35 mg/L to amphotericin B. If using clinical breakpoints or epidemiological cut-off values of C. albicans as references, all these strains could be classified as susceptible or of wild-type phenotype to all antifungal agents tested.
## Discussion
C. fabianii, basionym Hansenula fabianii, homotypic synonyms Candida fabianii, *Lindnera fabianii* and Pichia fabianii, is an ascomycetous yeast that has a close relationship with human activities (Kato et al., 1997; Arastehfar et al., 2019; Van Rijswijck et al., 2019). This yeast species is commonly seen in fermented food products like alcohols (Arastehfar et al., 2019; Van Rijswijck et al, 2019), and has also been used for treatment of waste water with a long history (Kato et al., 1997). The species has now been assigned within the Wickerhamomycetaceae clade (Kidd et al., 2023). Within this clade, there are several other species that have been reported to cause human infections, such as *Wickerhamomyces anomalus* and *Cyberlindnera jadinii* (Treguier et al., 2018; Zhang et al., 2021).
Generally, detection of C. fabianii in clinical settings is rare. According to previous surveillance reports on human fungal diseases, the prevalence of C. fabianii is generally <$0.1\%$ (Pfaller et al., 2019; Xiao et al., 2020). However, this yeast species is also an opportunistic pathogen that can cause a broad-range of infections, including lethal fungemia (Al-Sweih et al., 2019; Arastehfar et al., 2019). A previous research suggests that C. fabianii only has low virulence attributes (Arastehfar et al., 2019), although Nouraei et al. observed that this fungal species was one of the uncommon yeasts with high-level production of hemolysin, phospholipase and proteinase (Nouraei et al., 2020). In addition, C. fabianii has been observed to have a strong capacity for biofilm formation, which may contribute to its persistence and resistance to antifungal therapies (Hamal et al., 2008; Nouraei et al., 2020).
Of note, several studies have revealed difficulties in the accurate identification of C. fabianii using conventional methods, which may influence precision clinical recognition and management of infections caused by this organism (Svobodova et al., 2016; Al-Sweih et al., 2019). MALDI-TOF MS has been reported as a powerful tool for identification of yeasts, but the system’s identification capacity largely relies on the peptide mass fingerprint database that is incorporated into the system. Some of the MALDI-TOF MS systems, such as Biotyper (Bruker Daltonics, Germany, with IVD library version 8) and MicroIDSys (ASTA, Korea, with database version 1.23.2), have demonstrated capacity to accurately identify C. fabianii strains (Park et al., 2019; Teke et al., 2021). In contrast to this, C. fabianii is still absent from the Vitek MS IVD database (up to version 3.2), hence this system failed to identify any of C. fabianii isolates in this study. Similar findings were also reported by Teke et al. ( Teke et al., 2021).
Although nosocomial transmission of fungal pathogens is less frequently encountered in clinical practice compared to bacterial pathogens, fungal outbreaks are more common than publicly appreciated, and are mostly associated with medical products or contamination of the hospital environment (Kanamori et al., 2015; Litvintseva et al., 2015; Magill et al., 2018). For instance, Candida parapsilosis, one of the most prevalent human pathogenic yeast species, is well-known for causing catheter-related bloodstream infections. There have been a large number of nosocomial outbreaks caused by *Candida parapsilosis* worldwide, including several recently reported cases caused by fluconazole-resistant clones that raised more public health concerns (Arastehfar et al., 2020; Zhang et al., 2020; Thomaz et al., 2022). Moreover, reports of outbreaks caused by unusual fungal pathogens, such as the recently emerged C. auris, are increasing (Litvintseva et al., 2015; Chow et al., 2018). Of note, during the COVID-19 pandemic period, fungal outbreaks caused higher medical burdens to healthcare facilities and patients (Hoenigl et al., 2022; Thomaz et al., 2022), and hence are beginning to receive more attention.
Of note, a recent outbreak of C. fabianii in Kuwait was described by Al-Sweih et al., which involved a total of 10 fungemia cases in neonates (Al-Sweih et al., 2019). Furthermore, previous reviews on C. fabianii cases have demonstrated that the elderly population is the second most vulnerable population after neonates overall, with funguria being the first to second commonest clinical symptom (Al-Sweih et al., 2019; Arastehfar et al., 2019; Park et al., 2019). This agrees with our three-clustered C. fabianii funguria cases which all occurred in elderly patients, and with funguria as the common clinical symptom, though not every patient had symptomatic urinary tract infection.
Published literature have emphasized that presence of indwelling urinary catheter is the most important risk factor and transmission route for nosocomial urinary tract infections, especially when catheter care quality is poor. However, a variety of additional risk factors have also been described, including female gender, increased age, diabetes, bladder instrumentation, urinary outflow obstruction, amongst others. ( Pearson-Stuttard et al., 2016; Mody et al., 2017; Odabasi and Mert, 2020). Of the four patients described in this study, only one carried an indwelling urinary catheter, and all of them had undergone abdominal surgeries prior to the onset of funguria. Besides, three of the four patients had E. faecium detected concurrently with C. fabianii in the same urine sample. Enterococcus species, including E. faecium, are well-known ubiquitous inhabitants of the human gut microbiota and could lead to urinary tract infections (Magruder et al., 2019). Moreover, C. fabianii has also been identified in the human intestinal microbiota (Zhai et al., 2020), and previously Mathy et al. hypothesized that translocation of C. fabianii from the gut was responsible for a ventriculoperitoneal shunt case (Mathy et al., 2020). Therefore, it is possible that C. fabianii funguria cases identified in our study may have resulted from gut microbiota translocations, and abdominal surgeries might serve as triggers or risk factors.
As widely-acknowledged, application of ITS sequencing could allow accurate identification of yeast species but with insufficient discriminatory power for intra-species typing (Stielow et al., 2015; Al-Sweih et al., 2019). Al-Sweih et al. applied sequencing of NTS-1 regions, a gene locus that is considered to have a higher discriminatory power, in C. fabianii outbreak investigation, and found that all outbreak strains in Kuwait shared $100\%$ identical NTS-1 sequences (Al-Sweih et al., 2019). In comparison, we found a single SNP within NTS-1 region on patient 2’s strain versus strains from patients 1 and 3 in this study. However, further solid evidence was still needed to rule out the possibility of a potential outbreak.
To address concerns on readiness and limitations in discriminatory power of molecular typing methods in outbreak investigations, WGS has been recommended as a valuable alternative (Litvintseva et al., 2015; Bougnoux et al., 2018; Desnos-Ollivier et al., 2020). In this study, SNP-based analysis based on results acquired from WGS data clearly suggested that the genome of patient 2’s strain was quite divergent amongst the three clustered cases, which indicated a pseudo-outbreak event. Of note, the phrase “pseudo-outbreaks” could refer to either clustering of false infections, or artifactual clustering of real infections (Wallace et al., 1998). Clustering of false infections was more widely-noted, which may be associated with e.g. medical device or clinical laboratory contaminations (Kirby et al., 2017; Abdolrasouli et al., 2021). However, as indicated in our study, artifactual misinterpretation of “outbreaks” due to limitation of investigation methodologies (such as inadequate discriminatory power of molecular typing assays) should also be avoided.
Although WGS has made significant contributions in epidemiological studies, some limitations still remain. One major issue, as noted in outbreak investigations of all microbes including bacteria and fungi, is lack of consensus for data interpretation. Specifically, setting-up pairwise SNP-based cut-off values for assigning transmission events is still cumbersome, which has limited the wide utility of WGS in epidemiological studies (Coll et al., 2020; Guinea et al., 2021). In review of previous reports for outbreaks caused by yeast species that were characterized by WGS, it can be seen that the number of pairwise SNPs described in different studies of diverged species varied significantly, from less than ten to over hundreds. In the present study, genomic evidence clearly supported that patient 2’s C. fabianii strain was from a different origin, compared to others (with >26,000 SNPs compared to strains from patients 1 and 3). However, the 192 pairwise-SNP between strains from patient 1 and 3 may suggest that these two patients could have acquired the yeasts from a common source in the same ward but through different routes, rather than a direct transmission between the 2 patients, in which case the number of SNPs would be expected to be much less. But the hypothesis needs additional evaluation in a larger population and with more cases.
Due to the possibility of nosocomial transmission of this yeast in the described ward, surveillance infection control cultures were obtained to screen for C. fabianii in the department’s environment and amongst related health-care staff, but no C. fabianii was detected. Additional infection control strategies implemented further included enhancing environmental cleaning and hand hygiene practices, as well as providing education of fungal nosocomial infections to all healthcare staff.
One limitation of the study is that, antifungal susceptibility testing was not carried out using the standard broth microdilution methods, though YeastOne has proved equally efficient with good correlation in testing of yeasts (Cuenca-Estrella et al., 2010). Furthermore, with the limited number of cases studied, our base-line understanding for intra-species variation of C. fabianii genomes was still limited. Interpretation of any outbreak events shouldn’t simply rely on WGS result alone. It warrants a comprehensive analysis of different aspects of the cases, including patients’ clinical characteristics and epidemiological data, as well as the pathogens’ phenotypic and molecular characteristics.
In conclusion, as there are increasing reports of nosocomial outbreaks caused by emerging and uncommon fungal species, increased awareness of these rare organisms is warranted in public health. Conventional genotyping methods may have limited discriminatory power in investigating outbreaks due to these rare organisms; WGS has proven to be a good typing method for supporting investigation of such rare outbreak events.
## 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: Genbank: OP904191-OP904194 for ITS region, OP912967-OP912970 for NTS-1 region. WGS reads data can be found in NCBI database under Bioproject accession no. PRJNA907923.
## Ethics statement
The studies involving human participants were reviewed and approved by Human Research Ethics Committee of the Beijing Chaoyang Hospital. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.
## Author contributions
XF, R-CD and LG conceived the work. XF, and R-CD performed the experiments and data analysis. XF, TK, and LG drafting 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. Abdolrasouli A., Gibani M. M., De Groot T., Borman A. M., Hoffman P., Azadian B. S.. **A pseudo-outbreak of rhinocladiella similis in a bronchoscopy unit of a tertiary care teaching hospital in London, united kingdom**. *Mycoses* (2021) **64** 394-404. DOI: 10.1111/myc.13227
2. Al-Sweih N., Ahmad S., Khan S., Joseph L., Asadzadeh M., Khan Z.. **Cyberlindnera fabianii fungaemia outbreak in preterm neonates in Kuwait and literature review**. *Mycoses* (2019) **62** 51-61. DOI: 10.1111/myc.12846
3. Arastehfar A., Daneshnia F., Hilmioglu-Polat S., Fang W., Yasar M., Polat F.. **First report of candidemia clonal outbreak caused by emerging fluconazole-resistant candida parapsilosis isolates harboring Y132F and/or Y132F+K143R in Turkey**. *Antimicrob. Agents Chemother.* (2020) **64**. DOI: 10.1128/AAC.01001-20
4. Arastehfar A., Fang W., Al-Hatmi A. M. S., Afsarian M. H., Daneshnia F., Bakhtiari M.. **Unequivocal identification of an underestimated opportunistic yeast species, cyberlindnera fabianii, and its close relatives using a dual-function PCR and literature review of published cases**. *Med. Mycol* (2019) **57** 833-840. DOI: 10.1093/mmy/myy148
5. Bougnoux M. E., Brun S., Zahar J. R.. **Healthcare-associated fungal outbreaks: New and uncommon species, new molecular tools for investigation and prevention**. *Antimicrob. Resist. Infect. Control* (2018) **7** 45. DOI: 10.1186/s13756-018-0338-9
6. Chen S. C., Perfect J., Colombo A. L., Cornely O. A., Groll A. H., Seidel D.. **Global guideline for the diagnosis and management of rare yeast infections: an initiative of the ECMM in cooperation with ISHAM and ASM**. *Lancet Infect. Dis.* (2021) **21** e375-e386. DOI: 10.1016/S1473-3099(21)00203-6
7. Chow N. A., Chinn R., Pong A., Schultz K., Kim J., Gade L.. **Use of whole-genome sequencing to detect an outbreak of malassezia pachydermatis infection and colonization in a neonatal intensive care unit-california-2016**. *Infect. Control Hosp Epidemiol.* (2020) **41** 851-853. DOI: 10.1017/ice.2020.73
8. Chow N. A., Gade L., Tsay S. V., Forsberg K., Greenko J. A., Southwick K. L.. **Multiple introductions and subsequent transmission of multidrug-resistant candida auris in the USA: a molecular epidemiological survey**. *Lancet Infect. Dis.* (2018) **18** 1377-1384. DOI: 10.1016/S1473-3099(18)30597-8
9. Chowdhary A., Sharada K., Singh P. K., Bhagwani D. K., Kumar N., De Groot T.. **Outbreak of dirkmeia churashimaensis fungemia in a neonatal intensive care unit, India**. *Emerg. Infect. Dis.* (2020) **26** 764-768. DOI: 10.3201/eid2604.190847
10. Chowdhary A., Stielow J. B., Upadhyaya G., Singh P. K., Singh A., Meis J. F.. **Candida blankii: an emerging yeast in an outbreak of fungaemia in neonates in Delhi, India**. *Clin. Microbiol. Infect.* (2020) **26 648** e645-648e648. DOI: 10.1016/j.cmi.2020.01.001
11. Coll F., Raven K. E., Knight G. M., Blane B., Harrison E. M., Leek D.. **Definition of a genetic relatedness cutoff to exclude recent transmission of meticillin-resistant staphylococcus aureus: a genomic epidemiology analysis**. *Lancet Microbe* (2020) **1** e328-e335. DOI: 10.1016/S2666-5247(20)30149-X
12. Cuenca-Estrella M., Gomez-Lopez A., Alastruey-Izquierdo A., Bernal-Martinez L., Cuesta I., Buitrago M. J.. **Comparison of the vitek 2 antifungal susceptibility system with the clinical and laboratory standards institute (CLSI) and European committee on antimicrobial susceptibility testing (EUCAST) broth microdilution reference methods and with the sensititre YeastOne and etest techniques for**. *J. Clin. Microbiol.* (2010) **48** 1782-1786. DOI: 10.1128/JCM.02316-09
13. Desnos-Ollivier M., Maufrais C., Pihet M., Aznar C., Dromer F., French Mycoses Study G.. **Epidemiological investigation for grouped cases of trichosporon asahii using whole genome and IGS1 sequencing**. *Mycoses* (2020) **63** 942-951. DOI: 10.1111/myc.13126
14. De St Maurice A., Parti U., Anikst V. E., Harper T., Mirasol R., Dayo A. J.. **Clinical, microbiological, and genomic characteristics of clade-III candida auris colonization and infection in southern california-2022**. *Infect. Control Hosp Epidemiol.* (2022) **1-9**. DOI: 10.1017/ice.2022.204
15. Escandon P., Chow N. A., Caceres D. H., Gade L., Berkow E. L., Armstrong P.. **Molecular epidemiology of candida auris in Colombia reveals a highly related, countrywide colonization with regional patterns in amphotericin B resistance**. *Clin. Infect. Dis.* (2019) **68** 15-21. DOI: 10.1093/cid/ciy411
16. Eyre D. W., Sheppard A. E., Madder H., Moir I., Moroney R., Quan T. P.. **And its control in an intensive care setting**. *N Engl. J. Med.* (2018) **379** 1322-1331. DOI: 10.1056/NEJMoa1714373
17. Guinea J., Mezquita S., Gomez A., Padilla B., Zamora E., Sanchez-Luna M.. **Whole genome sequencing confirms candida albicans and candida parapsilosis microsatellite sporadic and persistent clones causing outbreaks of candidemia in neonates**. *Med. Mycol* (2021) **60**. DOI: 10.1093/mmy/myab068
18. Hamal P., Ostransky J., Dendis M., Horvath R., Ruzicka F., Buchta V.. **A case of endocarditis caused by the yeast pichia fabianii with biofilm production and developed**. *Med. Mycol* (2008) **46** 601-605. DOI: 10.1080/13693780802078180
19. Hoenigl M., Seidel D., Sprute R., Cunha C., Oliverio M., Goldman G. H.. **COVID-19-associated fungal infections**. *Nat. Microbiol.* (2022) **7** 1127-1140. DOI: 10.1038/s41564-022-01172-2
20. Kanamori H., Rutala W. A., Sickbert-Bennett E. E., Weber D. J.. **Review of fungal outbreaks and infection prevention in healthcare settings during construction and renovation**. *Clin. Infect. Dis.* (2015) **61** 433-444. DOI: 10.1093/cid/civ297
21. Kato M., Iefuji H., Miyake K., Iimura Y.. **Transformation system for a wastewater treatment yeast, hansenula fabianii J640: isolation of the orotidine-5’-phosphate decarboxylase gene (URA3) and uracil auxotrophic mutants**. *Appl. Microbiol. Biotechnol.* (1997) **48** 621-625. DOI: 10.1007/s002530051105
22. Kidd S. E., Abdolrasouli A., Hagen F.. **Fungal nomenclature: Managing change is the name of the game**. *Open Forum Infect. Dis.* (2023) **10** ofac559. DOI: 10.1093/ofid/ofac559
23. Kirby J. E., Branch-Elliman W., Lasalvia M. T., Longhi L., Mackechnie M., Urman G.. **Investigation of a candida guilliermondii pseudo-outbreak reveals a novel source of laboratory contamination**. *J. Clin. Microbiol.* (2017) **55** 1080-1089. DOI: 10.1128/JCM.02336-16
24. Li H., Durbin R.. **Fast and accurate short read alignment with burrows-wheeler transform**. *Bioinformatics* (2009) **25** 1754-1760. DOI: 10.1093/bioinformatics/btp324
25. Li H., Handsaker B., Wysoker A., Fennell T., Ruan J., Homer N.. **The sequence Alignment/Map format and SAMtools**. *Bioinformatics* (2009) **25** 2078-2079. DOI: 10.1093/bioinformatics/btp352
26. Litvintseva A. P., Brandt M. E., Mody R. K., Lockhart S. R.. **Investigating fungal outbreaks in the 21st century**. *PloS Pathog.* (2015) **11** e1004804. DOI: 10.1371/journal.ppat.1004804
27. Magill S. S., O’leary E., Janelle S. J., Thompson D. L., Dumyati G., Nadle J.. **Changes in prevalence of health care-associated infections in U.S. hospitals**. *N Engl. J. Med.* (2018) **379** 1732-1744. DOI: 10.1056/NEJMoa1801550
28. Magruder M., Sholi A. N., Gong C., Zhang L. S., Edusei E., Huang J.. **Gut uropathogen abundance is a risk factor for development of bacteriuria and urinary tract infection**. *Nat. Commun.* (2019) **10** 5521. DOI: 10.1038/s41467-019-13467-w
29. Mathy V., Chousterman B., Munier A. L., Cambau E., Jacquier H., De Ponfilly G. P.. **First reported case of postneurosurgical ventriculoperitonitis due to kocuria rhizophila following a ventriculoperitoneal shunt placement**. *Infect. Dis. Clin. Pract.* (2020) **28** 169-170. DOI: 10.1097/IPC.0000000000000829
30. Mckenna A., Hanna M., Banks E., Sivachenko A., Cibulskis K., Kernytsky A.. **The genome analysis toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data**. *Genome Res.* (2010) **20** 1297-1303. DOI: 10.1101/gr.107524.110
31. Mikosz C. A., Smith R. M., Kim M., Tyson C., Lee E. H., Adams E.. **Fungal endophthalmitis associated with compounded products**. *Emerg. Infect. Dis.* (2014) **20** 248-256. DOI: 10.3201/eid2002.131257
32. Mody L., Greene M. T., Meddings J., Krein S. L., Mcnamara S. E., Trautner B. W.. **A national implementation project to prevent catheter-associated urinary tract infection in nursing home residents**. *JAMA Intern. Med.* (2017) **177** 1154-1162. DOI: 10.1001/jamainternmed.2017.1689
33. Nouraei H., Pakshir K., Zareshahrabadi Z., Zomorodian K.. **High detection of virulence factors by candida species isolated from bloodstream of patients with candidemia**. *Microb. Pathog.* (2020) **149** 104574. DOI: 10.1016/j.micpath.2020.104574
34. Odabasi Z., Mert A.. **Candida urinary tract infections in adults**. *World J. Urol* (2020) **38** 2699-2707. DOI: 10.1007/s00345-019-02991-5
35. Oliveira M., Azevedo L.. **Molecular markers: An overview of data published for fungi over the last ten years**. *J. Fungi (Basel)* (2022) **8** 803. DOI: 10.3390/jof8080803
36. Pappas P. G., Lionakis M. S., Arendrup M. C., Ostrosky-Zeichner L., Kullberg B. J.. **Invasive candidiasis**. *Nat. Rev. Dis. Primers* (2018) **4** 18026. DOI: 10.1038/nrdp.2018.26
37. Park J. H., Oh J., Sang H., Shrestha B., Lee H., Koo J.. **Identification and antifungal susceptibility profiles of cyberlindnera fabianii in Korea**. *Mycobiology* (2019) **47** 449-456. DOI: 10.1080/12298093.2019.1651592
38. Pearson-Stuttard J., Blundell S., Harris T., Cook D. G., Critchley J.. **Diabetes and infection: assessing the association with glycaemic control in population-based studies**. *Lancet Diabetes Endocrinol.* (2016) **4** 148-158. DOI: 10.1016/S2213-8587(15)00379-4
39. Pfaller M. A., Diekema D. J., Turnidge J. D., Castanheira M., Jones R. N.. **Twenty years of the SENTRY antifungal surveillance program: Results for**. *Open Forum Infect. Dis.* (2019) **6** S79-S94. DOI: 10.1093/ofid/ofy358
40. Pulcrano G., Roscetto E., Iula V. D., Panellis D., Rossano F., Catania M. R.. **MALDI-TOF mass spectrometry and microsatellite markers to evaluate candida parapsilosis transmission in neonatal intensive care units**. *Eur. J. Clin. Microbiol. Infect. Dis.* (2012) **31** 2919-2928. DOI: 10.1007/s10096-012-1642-6
41. Reiss E., Tanaka K., Bruker G., Chazalet V., Coleman D., Debeaupuis J. P.. **Molecular diagnosis and epidemiology of fungal infections**. *Med. Mycol* (1998) **36 Suppl 1** 249-257
42. Rhodes J., Abdolrasouli A., Farrer R. A., Cuomo C. A., Aanensen D. M., Armstrong-James D.. **Genomic epidemiology of the UK outbreak of the emerging human fungal pathogen candida auris**. *Emerg. Microbes Infect.* (2018) **7** 43. DOI: 10.1038/s41426-018-0098-x
43. Stielow J. B., Levesque C. A., Seifert K. A., Meyer W., Iriny L., Smits D.. **One fungus, which genes? development and assessment of universal primers for potential secondary fungal DNA barcodes**. *Persoonia* (2015) **35** 242-263. DOI: 10.3767/003158515X689135
44. Svobodova L., Bednarova D., Ruzicka F., Chrenkova V., Dobias R., Mallatova N.. **High frequency of candida fabianii among clinical isolates biochemically identified as candida pelliculosa and candida utilis**. *Mycoses* (2016) **59** 241-246. DOI: 10.1111/myc.12454
45. Teke L., Baris A., Bayraktar B.. **Comparative evaluation of the bruker biotyper and vitek MS matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) systems for non-albicans candida and uncommon yeast isolates**. *J. Microbiol. Methods* (2021) **185** 106232. DOI: 10.1016/j.mimet.2021.106232
46. Thomaz D. Y., Del Negro G. M. B., Ribeiro L. B., Da Silva M., Carvalho G., Camargo C. H.. **A Brazilian inter-hospital candidemia outbreak caused by fluconazole-resistant candida parapsilosis in the COVID-19 era**. *J. Fungi (Basel)* (2022) **8** 100. DOI: 10.3390/jof8020100
47. Treguier P., David M., Gargala G., Camus V., Stamatoullas A., Menard A. L.. **Cyberlindnera jadinii (teleomorph candida utilis) candidaemia in a patient with aplastic anaemia: a case report**. *JMM Case Rep.* (2018) **5**. DOI: 10.1099/jmmcr.0.005160
48. Van Rijswijck I. M. H., Van Mastrigt O., Pijffers G., Wolkers-Rooijackers J. C. M., Abee T., Zwietering M. H.. **Dynamic modelling of brewers’ yeast and cyberlindnera fabianii co-culture behaviour for steering fermentation performance**. *Food Microbiol.* (2019) **83** 113-121. DOI: 10.1016/j.fm.2019.04.010
49. Wallace R. J., Brown B. A., Griffith D. E.. **Nosocomial outbreaks/pseudo-outbreaks caused by nontuberculous mycobacteria**. *Annu. Rev. Microbiol.* (1998) **52** 453-490. DOI: 10.1146/annurev.micro.52.1.453
50. World Health Organization (2022). WHO fungal priority pathogens list to guide research, development and public health action.
WHO. Available at: https://www.who.int/publications/i/item/9789240060241.. *WHO fungal priority pathogens list to guide research, development and public health action.* (2022)
51. Xiao M., Chen S. C., Kong F., Xu X. L., Yan L., Kong H. S.. **Distribution and antifungal susceptibility of candida species causing candidemia in China: An update from the CHIF-NET study**. *J. Infect. Dis.* (2020) **221** S139-S147. DOI: 10.1093/infdis/jiz573
52. Xiao M., Wang H., Lu J., Chen S. C., Kong F., Ma X. J.. **Three clustered cases of candidemia caused by candida quercitrusa and mycological characteristics of this novel species**. *J. Clin. Microbiol.* (2014) **52** 3044-3048. DOI: 10.1128/JCM.00246-14
53. Yadav A., Singh A., Wang Y., Haren M. H. V., Singh A., De Groot T.. **Colonisation and transmission dynamics of candida auris among chronic respiratory diseases patients hospitalised in a chest hospital, Delhi, India: A comparative analysis of whole genome sequencing and microsatellite typing**. *J. Fungi (Basel)* (2021) **7** 81. DOI: 10.3390/jof7020081
54. Zhai B., Ola M., Rolling T., Tosini N. L., Joshowitz S., Littmann E. R.. **High-resolution mycobiota analysis reveals dynamic intestinal translocation preceding invasive candidiasis**. *Nat. Med.* (2020) **26** 59. DOI: 10.1038/s41591-019-0709-7
55. Zhang L., Xiao M., Arastehfar A., Ilkit M., Zou J., Deng Y.. **Investigation of the emerging nosocomial wickerhamomyces anomalus infections at a Chinese tertiary teaching hospital and a systemic review: Clinical manifestations, risk factors, treatment, outcomes, and anti-fungal susceptibility**. *Front. Microbiol.* (2021) **12**. DOI: 10.3389/fmicb.2021.744502
56. Zhang L., Yu S. Y., Chen S. C., Xiao M., Kong F., Wang H.. **Molecular characterization of candida parapsilosis by microsatellite typing and emergence of clonal antifungal drug resistant strains in a multicenter surveillance in China**. *Front. Microbiol.* (2020) **11**. DOI: 10.3389/fmicb.2020.01320
|
---
title: 'Policy diffusion theory, evidence-informed public health, and public health
political science: a scoping review'
authors:
- Katrina Fundytus
- Cristina Santamaria-Plaza
- Lindsay McLaren
journal: Canadian Journal of Public Health = Revue Canadienne de Santé Publique
year: 2023
pmcid: PMC10030077
doi: 10.17269/s41997-023-00752-x
license: CC BY 4.0
---
# Policy diffusion theory, evidence-informed public health, and public health political science: a scoping review
## Abstract
### Objectives
Our aim was to synthesize published scholarship that applies policy diffusion—a theory of the policy process that considers the interdependence of government-level public health policy choices. We paid particular attention to the role of scientific evidence in the diffusion process, and to identifying challenges and gaps towards strengthening the intersection of public health, public policy, and political science.
### Methods
We systematically searched 17 electronic academic databases. We included English-language, peer-reviewed articles published between 2000 and 2021. For each article, we extracted the following information: public health policy domain, geographic setting, diffusion directions and mechanisms, the role of scientific evidence in the diffusion process, and author research discipline.
### Synthesis
We identified 39 peer-reviewed, primary research articles. Anti-smoking and tobacco control policies in the United States ($$n = 9$$/39) were the most common policy domain and geographic context examined; comparatively fewer studies examined policy diffusion in the Canadian context ($$n = 4$$/39). In terms of how policies diffuse, we found evidence of five diffusion mechanisms (learning, emulation, competition, coercion, and social contagion), which could moreover be conditional on internal government characteristics. The role of scientific evidence in the diffusion process was unclear, as only five articles discussed this. Policy diffusion theory was primarily used by public policy and political science scholars ($$n = 19$$/39), with comparatively fewer interdisciplinary authorship teams ($$n = 6$$/39).
### Conclusion
Policy diffusion theory provides important insights into the intergovernmental factors that influence public health policy decisions, thus helping to expand our conceptualization of evidence-informed public health. Despite this, policy diffusion research in the Canadian public health context is limited.
## Objectifs
Nous avons voulu faire une synthèse des travaux d’érudition publiés sur la diffusion des politiques—une théorie du processus d’élaboration des politiques qui prend en considération l’interdépendance des choix de politiques de santé publique au niveau gouvernemental. Nous nous sommes intéressés en particulier au rôle des preuves scientifiques dans le processus de diffusion et à la mise au jour des difficultés et des lacunes associées au renforcement de l’intersection entre la santé publique, les politiques publiques et les sciences politiques.
## Méthode
Nous avons systématiquement interrogé 17 bases de données électroniques universitaires. Nous avons inclus les articles en anglais évalués par les pairs publiés entre 2000 et 2021. Pour chaque article, nous avons extrait les informations suivantes: le domaine de politique de santé publique, le lieu géographique, les orientations et les mécanismes de diffusion, le rôle des preuves scientifiques dans le processus de diffusion et la discipline de recherche des auteurs.
## Synthèse
Nous avons recensé 39 articles de recherche primaire évalués par les pairs. La lutte contre le tabagisme et les politiques antitabac aux États-Unis ($$n = 9$$/39) étaient les domaines de politiques et le contexte géographique les plus couramment abordés; comparativement moins d’études portaient sur la diffusion des politiques dans le contexte canadien ($$n = 4$$/39). En ce qui concerne la façon dont les politiques se diffusent, nous avons relevé cinq mécanismes de diffusion (apprentissage, émulation, compétition, coercition et contagion sociale), qui peuvent de plus dépendre des caractéristiques internes du gouvernement. Le rôle des preuves scientifiques dans le processus de diffusion n’était pas clair, car seulement cinq articles en parlaient. La théorie de la diffusion des politiques était principalement utilisée par les théoriciens des politiques publiques et des sciences politiques ($$n = 19$$/39), avec comparativement moins d’équipes d’auteurs interdisciplinaires ($$n = 6$$/39).
## Conclusion
La théorie de la diffusion des politiques apporte des éclairages utiles sur les facteurs intergouvernementaux qui influencent les décisions en matière de politiques de santé publique, ce qui contribue à élargir notre conceptualisation de la santé publique éclairée par les données probantes. Malgré cela, la recherche sur la diffusion des politiques dans le contexte de la santé publique canadienne est limitée.
Policy diffusion theory has relevance to public health policy scholarship for two key reasons. First, and more generally, the use of political science and policy process theory in public health scholarship is rare, and focusing on policy diffusion provides one example of the richness and nuance that can come from applying a theory of the policy process to public health policy scholarship. Second, policy diffusion specifically is informative for public health policy because it can lead to both positive and negative consequences for public health outcomes, which may be missed if the primary focus is on scientific evidence (as per evidence-informed public health, for example). It illuminates the policy decisions of other governments as a key source of information, which may be in addition to, or instead of, scientific evidence and internal factors. The effect of policy diffusion can be positive if, for example, governments learn about effective public health policies from other governments, which can save both time and resources (Place Research Lab, n.d.). Conversely, through policy diffusion processes, the wrong lessons can be learned from others’ experiences, or governments may feel pressured to conform to the policy decisions of other “like-minded” governments even if they are “ineffective”, or they may seek to establish a competitive advantage over others (Shipan & Volden, 2012). Thus, the study of how and why, via the key mechanisms, policies diffuse has relevance to understanding what factors, aside from scientific evidence, contribute to public health policy decision-making and ultimately to public health outcomes such as population health status and health inequities.
## Introduction
There is long-standing, yet under-mobilized, recognition that governments can influence the distribution of the social determinants of health and health inequities (i.e., unfair and avoidable differences in health outcomes) by enacting public policies in domains such as housing, employment, and environment (Hancock, 1985; Raphael, 2020; World Health Organization, 2010). Public policies broadly refer to the decisions (both action and inaction) of a government, and can include statutes, regulations, procedures, programs, and executive decisions (Weible, 2014).
Public policy decision-making is complex, and one approach to better understand the intricacies of policymaking is to consider theories of the policy process (Cairney & Oliver, 2017; Fafard, 2015; Fafard & Cassola, 2020). The present study focuses on policy diffusion, where policy decisions in one jurisdiction influence policymaking in other jurisdictions (Berry & Berry, 2014). Policy diffusion is anchored in the recognition that policy adoption is inherently interdependent, and rarely occurs as a result of internal factors alone (Berry & Berry, 2014; Petridou, 2014).
Policy diffusion is a distinct class of studies within a broader literature on innovation and diffusion (Shipan & Volden, 2012). It draws heavily from Everett Rogers’ diffusion of innovations theory (Rogers, 1962, 2003), which examines the spread of non-policy innovations (i.e., individual- or organization-level interventions) via communication channels over a range of areas (e.g., teaching practices in school systems, medical/health ideas in hospitals). Scholarship in policy diffusion has evolved to incorporate new approaches and techniques that build upon Roger’s original framework (Berry & Berry, 2014, 2018; Karch, 2022). The present work is situated within this contemporary scholarship as described next.
Policy diffusion theory has been used to study whether, how, and why policies spread across government jurisdictions. This can occur in four directions: horizontal, diffusion across the same government level (e.g., provincial-to-provincial); bottom-up vertical, occurs from lower- to higher-level governments (e.g., local-to-provincial); top-down vertical, policy spreads from a higher- to lower-level government (e.g., provincial-to-local); and, replication, where a single government applies existing policy ideas to a new analogous policy domain (e.g., policy ideas spread across different domains within the same government) (Shipan & Volden, 2006; Train & Snow, 2019). In addition, five key mechanisms of diffusion have been identified (Berry & Berry, 2014; Maggetti & Gilardi, 2016; Pacheco, 2012; Shipan & Volden, 2008). Briefly, learning is when policymaking in one jurisdiction is influenced by the observed consequences of policies in other jurisdictions; the more successful a policy, the more likely its adoption elsewhere. Unlike learning, emulation is not contingent on whether a policy “works”; policy decisions are instead influenced by the normative environment or social acceptability. Coercion occurs when one government pressures others to take policy action via threat or incentive. Competition occurs when policy decisions are made to gain economic advantage (or avoid disadvantage) over other jurisdictions. Finally, social contagion refers to policy learning at the citizen level (as opposed to the government level), and the corresponding policy responsiveness of government officials.
Although there is a large literature on policy diffusion theory in political science and policy studies (Berry & Berry, 2014; Graham et al., 2013), its application to public health policy is not well studied (Breton & de Leeuw, 2011; Moloughney, 2012). This presents an important knowledge gap, which is perhaps indicative of a broader interdisciplinary research challenge identified by scholars working at the intersection of political science, public policy, and public health (Fafard & Cassola, 2020). Specifically, within the public health literature, only a limited number of theories of the policy process have been cited (Breton & de Leeuw, 2011; Cairney, 2016; Cairney et al., 2016, 2022; Moloughney, 2012), and the application of these theories tends to be superficial or descriptive (Breton & de Leeuw, 2011; Clarke et al., 2016; Moloughney, 2012).
Public health scholarship often endorses (implicitly or explicitly) a linear evidence-to-policy model of policy decision-making, where scientific evidence flows directly from knowledge producer (i.e., researchers) to users (i.e., policymakers) (Cairney, 2016; Fafard & Hoffman, 2020; Fafard et al., 2022). Evidence-informed public health (EIPH) is an example of this model (National Collaborating Centre for Methods and Tools, 2018). In contrast to the evidence-to-policy model, important scholarship has identified that the production and dissemination of scientific evidence alone does not have substantive impact on public policymaking (Cairney, 2016; Cairney & Oliver, 2017; Fafard & Cassola, 2020). Although scientific evidence can help to reduce uncertainty (i.e., lacking information on a policy problem), it does little to reduce ambiguity (i.e., lacking agreement on how to define/frame a policy problem) (Cairney, 2016; Cairney et al., 2022). To resolve ambiguity, policymakers draw upon different forms of “evidence” (e.g., value judgements, public opinion, “expert” consultation, emotions) to legitimize how policy problems are framed or prioritized (Cairney, 2016; Cairney & Oliver, 2017; Cairney et al., 2016; Oliver, 2022). Moreover, although often perceived as apolitical, the production, interpretation, and use of scientific evidence are value-based, contested, and influenced by structures of politics and power (Cassola et al., 2022; Parkhurst, 2017).
The learning mechanism of policy diffusion explicitly focuses on identifying indicators of policy success and effectiveness, which can include (but is not limited to) scientific evidence (Cairney, 2016; Olive & Boyd, 2021; Shipan & Volden, 2008). However, measures of success or effectiveness are rarely clear, can vary between governments, and are often based on limited scientific evidence (Cairney, 2016; Shipan & Volden, 2012). Overall, policy diffusion is not a technocratic process, but instead involves varied measures of policy success, value judgements, assessments of policy compatibility, and political considerations (Cairney, 2016; Olive & Boyd, 2021). We therefore seek to identify the role of scientific evidence in the policy diffusion process, and whether this differs across the diffusion mechanisms.
Overall, our aim is to identify and synthesize published, peer-reviewed scholarship that applies policy diffusion theory to public health policy (defined as a subset of public policies that aim to improve the health of populations), with particular attention to the role of scientific evidence in the diffusion process. We also aimed to identify challenges and gaps for research at the intersection of political science, public policy, and public health. To do so, we posed four research questions of the peer-reviewed literature:In what geographic settings and public health policy domains has policy diffusion theory been used or applied?How common are the five mechanisms identified in policy diffusion theory in the diffusion of public health policy?What role does scientific evidence play in policy diffusion, and how does this relate to the five mechanisms, if at all?To what extent is there cross-disciplinary engagement with diffusion theory in public health policy, particularly between public health, public policy, and political science?
## Methods
We undertook a scoping review, following methods described by Arksey and O’Malley [2005] and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis extension for Scoping Reviews (Tricco et al., 2018). The aim of a scoping review is to identify what is known about a particular concept (i.e., the application of diffusion theory to public health policies) and types of available evidence (Arksey & O’Malley, 2005).
## Data sources and search strategy
We systematically searched 17 electronic academic databases for peer-reviewed, English-language articles (see Fig. 1 (PRISMA) for the full list of academic databases). We used the search terms “policy diffusion” and (“population health” or “health promotion” or “public health”) in the article’s subject heading, title, abstract, keyword, or full text. As noted above, scholarship in the 1990s highlighted significant flaws in traditional methodologies of diffusion and innovation, and with the introduction of new empirical techniques, newer approaches have emerged and strengthened (Berry & Berry, 2014, 2018; Karch, 2022). We considered articles published between January 1, 2000, and June 20, 2021, to focus primarily on this contemporary era of diffusion theory. Fig. 1PRISMA flow diagram of included articles Two authors (KF and CSP) independently screened citation abstracts and titles using Covidence reference management software (Covidence, 2021). Full-text versions of all potentially relevant citations were independently reviewed by the same two authors, through which a final list of articles was compiled for extraction and analysis.
## Inclusion and exclusion criteria
We only considered primary research; reviews and commentaries were excluded. We also excluded books, book chapters, conference papers, abstracts, and student journals. Articles had to go beyond description of policy diffusion to integrate key concepts into a framework and/or to guide data collection and analysis (adapted from Breton & de Leeuw, 2011). In other words, articles that described diffusion but lacked an explicit application of diffusion theory, were excluded.
As per our definition of public policy (above), we only considered articles that focused on policy diffusion across a discrete political system (i.e., local, subnational, national, international) as opposed to a smaller, organizational level of governance such as schools, workplaces, or hospitals. “ Public health policy” is different from “health (care) policy” and we excluded articles that focused on a healthcare-oriented policy (Dalla Lana School of Public Health, n.d.). Articles had to describe the implications of the public policy for population health outcomes. This permitted us to embrace a broad definition of “public health policy”, which included (for example) infectious disease prevention (e.g., vaccination) and tobacco control, as well as broader social policies, such as gun control or animal regulation.
## Analysis
Guided by our research questions, we collated key information from each article using a coding template developed iteratively by KF and LM throughout the full-text article extraction phase. The following information was recorded for each article: publication year, general study design (i.e., quantitative, qualitative, mixed methods), journal title, key objectives and/or hypotheses, results in relation to policy diffusion theory, policy setting (i.e., primary geographic location), level of government (i.e., local, subnational, national, or international), policy domain (i.e., policy area), and diffusion direction (i.e., horizontal, bottom-up, top-down, replication). We also recorded information on the mechanisms of diffusion (i.e., learning, emulation, coercion, competition, and social contagion), including how these were defined and operationalized.
Finally, against the backdrop of evidence-informed public health, and a lack of integrated research partnerships between public health, political science, and public policy scholars, we recorded information on [1] the role of scientific evidence in policy diffusion, and in relation to the diffusion mechanisms specifically (if discussed), and [2] each author’s scholarly discipline (based on their formal academic training and academic department appointment) to gauge the extent of interdisciplinary research teams.
## Results
From an initial set of 628 articles, of which 349 were deemed potentially relevant based on title/abstract, we ultimately analyzed 39 peer-reviewed research articles that applied policy diffusion theory to a public health policy (see Fig. 1 (PRISMA), and Table 1 for descriptive study characteristics).Table 1Summary table of descriptive study characteristicsStudyAuthor research discipline(s)Study designGeographic settingGovernment-level; diffusion directionPolicy domainDiffusion mechanism(s)Did policy diffusion occur?Referenced scientific evidenceAgostinis, 2019Political scienceQualitativeVariousInternational; horizontalVarious (cancer control and public health education)Learning*YesNoAnderson et al., 2016Public administration, public health policy, engineering, community health sciencesQuantitativeUSASubnational; horizontalVarious (alcohol, driver safety, impaired driving)N/AYesNoBessho, S. & Ibuka, Y., 2018EconomicsQuantitativeJapanLocal; horizontalVaccinationN/AYesNoBoehmke, F., 2009Political scienceQuantitativeUSASubnational; horizontalObesityN/AMixed; methods paper outlining different approaches to model policy diffusionNoBoyle et al., 2015SociologyQuantitativeVariousInternational; horizontalAbortion liberalizationN/AYes, typical and atypicalNoChorev, 2012SociologyQualitativeVariousInternational; horizontalHIV/AIDSLearning*YesNoClark, 2013Public policy, public administrationQuantitativeVariousInternational; horizontalHIV/AIDSLearningMixed; role of diffusion is not clearNoClark, 2009Public policy, public administrationQuantitativeVariousInternational; horizontalHIV/AIDSN/AYes, atypicalNoClouser-McCann et al., 2015Political science, public health, public policyQuantitativeUSANational, subnational, local; horizontal, vertical (bottom-up), vertical (top-down)Anti-smoking/tobaccoN/AYesNoCuriel et al., 2020Political science, public health dentistry, dental ecologyQuantitativeUSALocal; horizontalCommunity water fluoridationEmulation*YesNoFix & Mitchell, 2017Political scienceQuantitativeUSASubnational; horizontalLocal breed legislationLearning*YesNoGivens & Mistur, 2021Political science, public administration, public policyQuantitativeVariousInternational; horizontalCOVID-19Emulation*YesYesGodwin, M. & Schroedel, JR, 2000Political science, public administrationMixed methodsUSALocal; horizontalGun controlN/AYesNoJohns, 2015Sociology, political scienceMixed methodsUSALocal; replicationMarijuanaLearning*YesNoJohnson, B. & White, S., 2010Political science, urban planning, land resourcesQualitativeUSALocal; horizontalSustainable transportation infrastructureLearning*YesNoKadowaki et al., 2015SociologyQuantitativeUSASubnational, local; horizontal, vertical (bottom-up)e-cigarette restrictionsN/AInconclusiveYesKavanagh et al., 2021Political science, international healthMixed methodsVariousInternational; horizontalHIV/AIDSLearningNoYesMacinko & Silver, 2015Health policy, public health policy, public administrationQuantitativeUSASubnational; horizontal, replicationImpaired drivingLearningYesYesMallinson, 2016Political science, public policyQuantitativeUSASubnational; horizontalAnti-bullying measuresLearning*Yes, atypicalNoMichael, 2016Political scienceMixed methodsVariousInternational; horizontalPharmacuticalsCoercion*CompetitionEmulationLearningYesNoMitchell & Stewart, 2014Political science, public policy, public administrationQuantitativeUSALocal; horizontalAnti-smoking/tobaccoCompetition*EmulationLearning*YesNoMoreland-Russell, S. et al., 2013Public health, public policy, community health, social ecologyQuantitativeUSALocal; horizontalComplete streets policiesN/AYesNoNykiforuk, C. et al., 2018Public health, health promotionQuantitativeCanadaLocal; horizontalFast food bylawLearningYesNoNykiforuk, C. et al., 2018Public health, health sciences, geographyQuantitativeCanadaSubnational, local; horizontal, vertical (top-down)Anti-smoking/tobaccoN/AYesNoOlstad et al., 2015Public health, nutritionQualitativeCanadaSubnational; horizontalDaily physical activityN/AYesYesPacheco, 2017Political science, public policyQuantitativeUSASubnational; horizontalAnti-smoking/tobaccoCompetition*Yes, typical and atypicalNoPacheco, 2012Political science, public policyQuantitativeUSASubnational; horizontalAnti-smoking/tobaccoCompetitionLearningSocial contagion*YesNoPacheco & Boushey, 2014Political science, public policyQuantitativeUSANational, subnational; horizontal, vertical (top-down)Various (anti-smoking/tobacco, vaccination)N/AYesNoSebhatu et al., 2020Sociology, business, political scienceQuantitativeVariousInternational; horizontalCOVID-19Emulation*YesNoSeptiono, W. et al., 2019Public healthQuantitativeIndonesiaSubnational, local; horizontal, vertical (top-down)Anti-smoking/tobaccoN/AYesNoShen, 2014Public health, health policyQuantitativeVariousInternational; horizontalMental healthCoercionEmulation*YesNoShipan & Volden, 2008Political science, public policyQuantitativeUSASubnational, local; horizontal, vertical (top-down)Anti-smoking/tobaccoCoercion*Competition*Emulation*Learning*YesNoShipan & Volden, 2006Political science, public policyQuantitativeUSANational, subnational, local; horizontal, vertical (bottom-up), vertical (top-down)Anti-smoking/tobaccoN/AYesNoShipan & Volden, 2014Political science, public policyQuantitativeUSASubnational; horizontalAnti-smoking/tobaccoLearning*YesNoSieger, M. & Rebbe, R., 2020Social workQuantitativeUSASubnational; horizontalChild abuse prevention and treatment actN/AYesNoTrain & Snow, 2019Political scienceQualitativeCanadaNational, subnational; replication, horizontal, vertical (top-down)MarijuanaCoercion*CompetitionEmulationLearning*YesNoTrein, 2017Political scienceQuantitativeSwitzerlandSubnational; horizontalAnti-smoking/tobaccoSocial contagion*YesNoValente, T. et al., 2015Public health, communication, sociology, communication and information sciences, international relationsQuantitativeVariousInternational; horizontalAnti-smoking/tobaccoN/AYesNoWinder & LaPlant, 2000Political scienceQuantitativeUSASubnational; horizontalAnti-smoking/tobaccoN/AMixed; report diffusion but no consistent regional patternNo*Mechanism significant**Atypical = slowed/decreased government action
## Public health policy geographic settings, government level, and diffusion direction
Most articles focused on policies in the United States ($$n = 21$$/39) or, to a much lesser extent, Canada ($$n = 4$$/39). Other primary settings included Japan, Indonesia, and Switzerland. In terms of government level, the most common were subnational (e.g., province, canton, state) ($$n = 20$$/39) and local level (e.g., county, municipality) ($$n = 14$$/39); international policy diffusion (i.e., country-to-country) ($$n = 11$$/39) was also common. For diffusion direction,1 nearly all articles examined horizontal diffusion ($$n = 38$$/39), with notably fewer examining top-down ($$n = 7$$/39), bottom-up ($$n = 3$$/39), or replication ($$n = 3$$/39) (Table 1).
## Public health policy domains and evidence of diffusion
Policy diffusion was applied to several public health domains, most commonly anti-smoking- and tobacco-related policies ($$n = 13$$/39) (e.g., Shipan & Volden, 2006) and HIV/AIDS-related policies ($$n = 4$$/39) (Chorev, 2012; Clark, 2013; Clarke et al., 2016; Kavanagh et al., 2021). Other policy domains included COVID-19 ($$n = 2$$/39) (Givens & Mistur, 2021; Sebhatu et al., 2020), marijuana ($$n = 2$$/39) (Johns, 2015; Train & Snow, 2019), vaccinations ($$n = 2$$/39) (Pacheco & Boushey, 2014), and impaired driving ($$n = 2$$/39) (Anderson et al., 2016; Macinko & Silver, 2015) (see Table 1 for full list of policy domains).
Although assessment of whether diffusion occurred or not is complicated by different research questions and methods, we ultimately identified that most ($$n = 34$$/39) articles showed evidence of policy diffusion. For example, in the USA, Shipan and Volden [2006] found that the likelihood of state-level governments adopting an anti-smoking policy increased as neighbouring states passed such policies. In the Canadian context, all four articles2 demonstrated the role of policy diffusion in the adoption and spread of school-based daily physical activity policies (provincial) (Olstad et al., 2015), fast food drive-through and smoking restriction bylaws (local) (Nykiforuk et al., 2008, 2018), and recreational marijuana regulation (provincial) (Train & Snow, 2019).
Four articles found an atypical pattern of diffusion, where neighbouring policy adoption slowed or decreased the likelihood of local policy adoption, in the policy domains of tobacco control (Pacheco, 2017), anti-bullying (Mallinson, 2016), abortion liberalization (Boyle et al., 2015), and HIV/AIDs (Clark, 2009). For example, Clark [2009] identified that as the proportion of AIDS program adoption in geographically neighbouring countries increased, the time leading to local adoption also increased. In contrast, several articles ($$n = 5$$/39) found mixed, inconclusive, or nonsignificant evidence of diffusion. For example, Kavanagh et al. [ 2021] identified formal government structures and racial stratification as better predictors of HIV treatment policy adoption compared to the policy choices of neighbouring governments.
## The mechanisms of public health policy diffusion: learning, emulation, competition, coercion, and social contagion
Just over half of the articles ($$n = 22$$/39) referenced at least one mechanism of diffusion. The most common was learning ($$n = 16$$/39), then emulation ($$n = 8$$/39), competition ($$n = 6$$/39), coercion ($$n = 4$$/39), and social contagion ($\frac{2}{39}$).3 There was heterogeneity in terms of how the diffusion mechanisms were measured or conceptualized, with different indicators used for the same mechanism across the included articles.4 For example, in the case of policy learning, measurements ranged from broad-level indicators, such as the number of bordering governments that adopted a policy the previous year (Mitchell & Stewart, 2014), to more specific indicators, such as demonstrated success of a policy adopted by a government elsewhere (Shipan & Volden, 2014) or explicit reference to another government as a source of information and legitimacy (Chorev, 2012).
Notwithstanding these different ways of measuring each mechanism, there were examples of each occurring, which varied by geographic context and policy domain. Policy learning was evident in ten articles, including the adoption of cancer control policies and public health training in South America (Agostinis, 2019), youth tobacco restriction policy adoption in the USA (Shipan & Volden, 2014), intellectual property rights of AIDS drugs (Chorev, 2012), and dog breed specific legislation in the USA (Fix & Mitchell, 2017). Two articles identified the role of the learning mechanism via replication diffusion in marijuana regulation (Johns, 2015; Train & Snow, 2019); for example, a greater number of American cities in the state of Colorado permitted the sale of recreational marijuana if they had previously implemented a medical marijuana-use policy (Johns, 2015).
Emulation was significant in the adoption of COVID-19 policies (Givens & Mistur, 2021; Sebhatu et al., 2020) and mental health policy (Shen, 2014) internationally, and local-level anti-smoking (Shipan & Volden, 2008) and community water fluoridation policies5 (Curiel et al., 2020) in the USA. For example, one study identified that “nationalist”6 countries were more likely to implement a policy change the day after a country with a similar nationalist regime changed its respective COVID-19 policies (Givens & Mistur, 2021). At the local level of government in the USA, Shipan and Volden [2008] found American cities more likely to adopt an anti-smoking law when the nearest, largest neighbouring city had previously adopted such a law.
Competition was evident in anti-smoking and tobacco control policies in the USA at the local (e.g., clean indoor air laws, youth access policies) (Mitchell & Stewart, 2014; Shipan & Volden, 2008) and state levels (e.g., tobacco sale and consumption) (Pacheco, 2017). Pacheco [2017] identified two ways that competition can influence tobacco and anti-smoking policy at the state-level in the USA: competitive races (i.e., policy changes in one jurisdiction encourage others to adopt similar policies to gain economic or other benefits) and free-rider dynamics (i.e., positive spillover effects of a policy in one jurisdiction incentivize others not to adopt).
Coercive pressures contributed to policy adoption in the domains of marijuana regulation (Train & Snow, 2019), intellectual property right laws (Michael, 2016), and anti-smoking and tobacco (Shipan & Volden, 2008). One study examined the global diffusion of intellectual property right agreement laws for pharmaceutical clinical trial data; it identified that powerful countries can dictate the terms of these laws to other countries by threatening to withhold benefits during trade negotiations (Michael, 2016). In Canada, coercive pressures from the federal government influenced the diffusion of marijuana legalization at the provincial level in Ontario and New Brunswick by placing heavy constraints on provincial autonomy to regulate the production, distribution, sale, and consumption of cannabis (Train & Snow, 2019).
Finally, two articles, both in the anti-smoking and tobacco domain, reported evidence of social contagion (Pacheco, 2012; Trein, 2017). Pacheco [2012] identified that public opinion of restaurant smoking bans is influenced by the policy decisions in neighbouring states; if state-wide opinion becomes supportive of these bans, officials then respond by enacting similar policies locally.
## Internal government characteristics and policy diffusion mechanisms
Diffusion mechanisms sometimes overlapped in the same policy domain or geographic setting (Mitchell & Stewart, 2014; Shipan & Volden, 2008; Train & Snow, 2019). Moreover, they were sometimes contingent on internal government characteristics, such as government regime (Givens & Mistur, 2021; Sebhatu et al., 2020), policy expertise (Shipan & Volden, 2014), legislative professionalism (Pacheco & Boushey, 2014; Shipan & Volden, 2014), and policy problem severity (Fix & Mitchell, 2017). For example, in the USA, states with a higher number of dog fight cases or fatalities from dog bites (i.e., high problem severity) were more likely to adopt breed-specific legislation, compared to states with lower numbers (Fix & Mitchell, 2017). Conversely, Givens and Mistur [2021] did not find a consistent significant relationship between policy problem severity (in the form of COVID-19 cases per capita) and the adoption of COVID-19 policies by “nationalist” countries.
## Scientific evidence in public health policy diffusion and the policy diffusion mechanisms
The role of scientific evidence in the policy diffusion process was not frequently examined. Five studies discussed scientific evidence in some capacity ($$n = 5$$/39) (Givens & Mistur, 2021; Kadowaki et al., 2015; Kavanagh et al., 2021; Macinko & Silver, 2015; Olstad et al., 2015). Only three articles ($$n = 3$$/5) referenced at least one policy diffusion mechanism and scientific evidence; however, none of these articles empirically examined the role of scientific evidence in relation to the diffusion mechanisms.
In one article, Givens and Mistur [2021] interpreted the observed pattern of COVID-19 policy adoption by “nationalist” countries (see above) as suggesting that these governments “emulate” the policies of other countries with similar nationalist regimes, instead of following scientific evidence. In another article, Macinko and Silver [2015] examined the role of policy learning (via replication)7 and other determinants in evidence-based impaired driving law adoption in the USA; although the authors assert more generally that patterns of state-level health policy adoption ought to be understood as more than a direct response to emerging evidence, this was not explicitly examined in their analysis. Finally, one article empirically considered the role of scientific evidence and policy learning in global HIV treatment policy decision-making, but neither were found to be strong or consistent indicators of policy adoption (Kavanagh et al., 2021).
Two articles discussed scientific evidence more broadly but did not examine any diffusion mechanisms in their analysis (Kadowaki et al., 2015; Olstad et al., 2015). Kadowaki et al. [ 2015] identified a spatially uneven pattern of adoption of state- and local-level e-cigarette clean air policies in the USA, and partially attributed this to policy needs outpacing available scientific evidence, and a general lack of consistent scientific evidence creating confusion among policymakers. In the Canadian context, Olstad et al. [ 2015] identified that provincial governments (Alberta, British Columbia, Manitoba, and Saskatchewan) cited an international body of evidence as a rationale for adopting daily physical activity policies for children. However, it was not clear whether or the extent to which this evidence informed the specific provisions of each province’s policy; provincial policies varied across the country, and in some cases, did not coincide with the established national guidelines.
## Cross-disciplinary engagement with diffusion theory: public health, public policy, and political science
The majority of authorship teams on studies included in our review consisted of scholars from the political science, public policy, and public administration research domains only ($$n = 19$$/39). There were fewer cross-disciplinary research teams consisting of both public health and political science or public policy scholars ($$n = 6$$/39), and even fewer consisting of public health scholars only ($$n = 3$$/39). Other research disciplines included sociology ($$n = 3$$/39), social work ($$n = 1$$/39), and economics ($$n = 1$$/39) (see Table 1).
## Discussion
Policy diffusion theory highlights the importance of considering the interdependence of public health policy decisions. We found that application of the theory is particularly developed in the domain of anti-smoking and tobacco policy in the USA. Comparatively, there were relatively fewer articles in the Canadian context, which examined a range of policy domains and levels of government (Nykiforuk et al., 2008, 2018; Olstad et al., 2015; Train & Snow, 2019).
Despite recognition of the importance and relevance of policy diffusion research by Canadian researchers (Place Research Lab, n.d.; Politis et al., 2014), we found few examples of public health policy diffusion scholarship in the Canadian context, consistent with findings elsewhere (Olive & Boyd, 2021). Our findings build on existing public health policy diffusion scholarship in Canada (Campbell et al., 2020; Nykiforuk et al., 2008, 2018; Olstad et al., 2015; Place Research Lab, n.d.), which primarily adapts Roger’s diffusion of innovations theory to explain adoption patterns (Rogers, 2003). Although Roger’s theory is widespread in health sciences and healthcare innovation research, our review captures contemporary policy diffusion scholarship to include (for example) Berry and Berry [1990], Maggetti and Gilardi [2016], Shipan and Volden [2008], and Volden [2006].
We found evidence of five mechanisms of diffusion (i.e., learning, emulation, competition, coercion, and social contagion), which vary depending on policy domain, geographic context, and internal government characteristics. Our findings show that local public health problem severity (e.g., motor vehicle fatalities, COVID-19 cases) is not a reliable predictor of policy action (Givens & Mistur, 2021; Kavanagh et al., 2021; Sebhatu et al., 2020; Winder & LaPlant, 2000). From the perspective of public health practice, this finding confirms tacit understanding that public health surveillance, while important and necessary, is not sufficient to prompt public policy action (Chambers et al., 2006). Governments may vary in their capacity to obtain, analyze, and use this information (Clouser-McCann et al., 2015; Shipan & Volden, 2014), or governments may be aware of public health threats, but privilege other factors in decision-making, such as non-health measures of policy success (Shipan & Volden, 2008), or pressures from other government jurisdictions via one or more diffusion mechanisms.
Evidence-to-policy models in public health often assert that improved knowledge translation efforts (i.e., researchers more effectively providing policy decisionmakers with scientific evidence) will increase the likelihood that scientific evidence will inform policy decisions (Fafard, 2008). However, this is not well supported by our findings. Scientific evidence was either absent or did not play a significant role in policy diffusion more generally, or across the five diffusion mechanisms. Even when policymakers are aware of and able to articulate pertinent scientific evidence (Kavanagh et al., 2021), they may privilege other factors in policy decisions. In the case of policy learning, for example, instead of engaging directly with scientific evidence, governments may look for other indicators of policy success, such as widespread policy adoption without subsequent abandonment across other jurisdictions (Shipan & Volden, 2008). Thus, there is a need for public health research to consider what constitutes appropriate and relevant evidence in the policy diffusion process, and in relation to each of the diffusion mechanisms, as opposed to what “should” inform policymaking based on established hierarchies that favour certain types of scientific evidence (e.g., systematic reviews, randomized controlled trials) and their accompanying epistemological perspectives (Oliver, 2022; Parkhurst, 2016).
Finally, despite the complementary nature of political science, public policy, and public health disciplines, we found little evidence of interdisciplinary research partnerships ($$n = 6$$/39), with most article authors having formal academic training in political science or public policy studies. To address this challenge, scholars have emphasized the need for a more collaborative approach to public health policy analysis, termed “public health political science” (Fafard & Cassola, 2020; Greer et al., 2017). Public health political science seeks to incorporate insights from public health, public policy, and political science to provide a more robust approach to address politics, political systems, and the public health policy process (Fafard & Cassola, 2020; Greer et al., 2017). Based on the relatively low number of cross-disciplinary research teams in our sample, we see this as an important area of growth in public health policy scholarship.
This scoping review has several limitations. First, we only considered articles that used the term “policy diffusion” and did not include related terms such as policy transfer or convergence in our database search; these terms—though related and complementary—are distinct research areas, and we therefore maintained our conceptual focus on policy diffusion (Gilardi & Wasserfallen, 2019; Graham et al., 2013; Petridou, 2014; Shipan & Volden, 2012). Nonetheless, our omission of these related subfields may underrepresent the number of articles that examine government-level public health policy interdependence, as well as the extent of interdisciplinary engagement with this literature.
Second, as this is a scoping review, we did not assess the quality or rigour of the included studies. In terms of strengths, we highlight our systematic approach to identify relevant peer-reviewed articles, and in particular, our comprehensive search of 17 electronic databases, and the use of two authors to screen abstracts and full-text articles. Moreover, this is the first review to examine the application of policy diffusion theory to government-level public health policy specifically; historically, policy diffusion theory has not been included in reviews on the application of policy process theories in public health research (Breton & de Leeuw, 2011; Moloughney, 2012). Directions for future research could include (but are not limited to) examining the specific role(s) of scientific evidence and other “types” of evidence in relation to the five mechanisms of policy diffusion, and across different public health policy domains and geographic contexts.
## References
1. Agostinis G. **Regional intergovernmental organizations as catalysts for transnational policy diffusion: The case of UNASUR Health**. *Journal of Common Market Studies* (2019.0) **57** 1111-1129. DOI: 10.1111/jcms.12875
2. Anderson R, Jimenez G, Bae JY, Silver D, Macinko J, Porfiri M. **Understanding policy diffusion in the U.S.: An information-theoretical approach to unveil connectivity structures in slowly evolving complex systems**. *SIAM Journal on Applied Dynamical Systems* (2016.0) **15** 1384-1409. DOI: 10.1137/15M1041584
3. Arksey H, O’Malley L. **Scoping studies: Towards a methodological framework**. *International Journal of Social Research Methodology* (2005.0) **8** 19-32. DOI: 10.1080/1364557032000119616
4. Berry, F. S., & Berry, W. D. (2014). Innovation and diffusion models in policy research. In P. Sabatier & C. Weible (Eds.), Theories of the policy process, 307–360. Westview Press.
5. Berry, F. S., & Berry, W. D. (2018). Innovation and diffusion models in policy research. In P. Sabatier & C. Weible (Eds.), Theories of the policy process, 253–297. Routledge. 10.4324/9780429494284-8
6. Berry FS, Berry WD. **State lottery adoptions as policy innovations: An event history analysis**. *American Political Science Review* (1990.0) **84** 395-415. DOI: 10.2307/1963526
7. Bessho, S., & Ibuka, Y. (2018). Interdependency in vaccination policies among Japanese municipalities. Health Economics, 28(2):299–310.10.1002/hec.3845
8. Boehmke, F. (2009). Approaches to modeling the adoption and diffusion of policies with multiple components. State Politics & Policy Quarterly, 9(2), 229–252. 10.1177/153244000900900205
9. Boyle EH, Kim M, Longhofer W. **Abortion liberalization in world society, 1960–2009**. *American Journal of Sociology* (2015.0) **121** 882-913. DOI: 10.1086/682827
10. Breton E, de Leeuw E. **Theories of the policy process in health promotion research: A review**. *Health Promotion International* (2011.0) **26** 82-90. DOI: 10.1093/heapro/daq051
11. Cairney P, Oliver K. **Evidence-based policymaking is not like evidence-based medicine, so how far should you go to bridge the divide between evidence and policy?**. *Health Research Policy and Systems* (2017.0) **15** 35. DOI: 10.1186/s12961-017-0192-x
12. Cairney P, Oliver K, Wellstead A. **To bridge the divide between evidence and policy: Reduce ambiguity as much as uncertainty**. *Public Administration Review* (2016.0) **76** 399-402. DOI: 10.1111/puar.12555
13. Cairney, P., St. Denny, E., & Mitchell, H. (2022). How can policy theory help to address the expectations gap in preventive public health and ‘Health in All Policies’? Fafard P., Cassola A., & de Leeuw E. (Eds.), Integrating science and politics for public health (pp. 239–265). Palgrave Springer.
14. Cairney, P. (2016). The role of evidence in theories of the policy process. In P. Cairney, The politics of evidence-based policy making (pp. 13–50). Palgrave Macmillan UK. 10.1057/978-1-137-51781-4_2
15. Campbell, E., Olstad, D., Spence, J., Storey, K., & Nykiforuk, C. (2020). Policy-influencer perspectives on the development, adoption, and implementation of provincial school-based daily physical activity policies across Canada: A national case study. SSM - Population Health, 11. 10.1016/j.ssmph.2020.100612
16. Cassola, A., Fafard, P., Palkovits, M., & Hoffman, S. (2022). Mechanisms to bridge the gap between science and politics in evidence-informed policymaking: Mapping the landscape. P. Fafard, A. Cassola, & E. de Leeuw (Eds.), Integrating science and politics for public health (pp. 293–328). Palgrave Springer.
17. Chambers L, Ehrlich A, Steel O’Connor K, Edwards P, Hockin J. **Health surveillance: An essential tool to protect and promote the health of the public**. *Canadian Journal of Public Health* (2006.0) **97** 1-8
18. Chorev N. **Changing global norms through reactive diffusion: The case of intellectual property protection of AIDS drugs**. *American Sociological Review* (2012.0) **77** 831-853. DOI: 10.1177/0003122412457156
19. Clark B. **Policy adoption in dynamic international environments: Evidence from national AIDS programs**. *Public Administration and Development* (2009.0) **29** 362-373. DOI: 10.1002/pad.555
20. Clark B. **Multilateral, regional, and national determinants of policy adoption: The case of HIV/AIDS legislative action**. *International Journal of Public Health* (2013.0) **58** 285-293. DOI: 10.1007/s00038-012-0393-6
21. Clarke B, Swinburn B, Sacks G. **The application of theories of the policy process to obesity prevention: A systematic review and meta-synthesis**. *BMC Public Health* (2016.0) **16** 1084. DOI: 10.1186/s12889-016-3639-z
22. Clouser-McCann P, Shipan C, Volden C. **Top-down federalism: State policy responses to national government discussions**. *Publius: The Journal of Federalism* (2015.0) **45** 495-525. DOI: 10.1093/publius/pjv013
23. Covidence. (2021). Covidence systematic review software. Retrieved May 17, 2021 from www.covidence.org
24. Curiel JA, Sanders AE, Slade GD. **Emulation of community water fluoridation coverage across US counties**. *JDR Clinical & Translational Research* (2020.0) **5** 376-384. DOI: 10.1177/2380084419887696
25. Dalla Lana School of Public Health. (n.d.). Public health policy. University of Toronto. https://www.dlsph.utoronto.ca/program/collaborative-specialization-in-public-health-policy/
26. Fafard P. **Beyond the usual suspects: Using political science to enhance public health policy making**. *Journal of Epidemiology and Community Health* (2015.0) **69** 1129-1132. DOI: 10.1136/jech-2014-204608
27. Fafard P, Cassola A. **Public health and political science: Challenges and opportunities for a productive partnership**. *Public Health* (2020.0) **186** 107-109. DOI: 10.1016/j.puhe.2020.07.004
28. Fafard P, Hoffman SJ. **Rethinking knowledge translation for public health policy**. *Evidence & Policy: A Journal of Research, Debate and Practice* (2020.0) **16** 165-175. DOI: 10.1332/174426418X15212871808802
29. Fafard P, de Leeuw E, Cassola A, Fafard P, Cassola A, de Leeuw E. **Introduction: Virchow revisited on the importance of public health political science**. *Integrating science and politics for public health* (2022.0) 3-14
30. Fafard, P. (2008). Evidence and healthy public policy: Insights from health and political sciences. National Collaborating Centre for Healthy Public Policy. https://www.ncchpp.ca/docs/FafardEvidence08June.pdf
31. Fix MP, Mitchell JL. **Examining the policy learning dynamics of atypical policies with an application to state preemption of local dog laws**. *Statistics, Politics and Policy* (2017.0) **8** 223-247. DOI: 10.1515/spp-2017-0009
32. Gilardi F, Wasserfallen F. **The politics of policy diffusion**. *European Journal of Political Research* (2019.0) **58** 1245-1256. DOI: 10.1111/1475-6765.12326
33. Givens JW, Mistur E. **The sincerest form of flattery: Nationalist emulation during the COVID-19 pandemic**. *Journal of Chinese Political Science* (2021.0) **26** 213-234. DOI: 10.1007/s11366-020-09702-7
34. Godwin, M., & Schroedel, J. (2000). Policy diffusion and strategies for promoting policy change: Evidence From California local gun control ordinances. Policy Studies Journal, 28(4):760–776. 10.1111/j.1541-0072.2000.tb02061.x
35. Graham ER, Shipan CR, Volden C. **The diffusion of policy diffusion research in political science**. *British Journal of Political Science* (2013.0) **43** 673-701. DOI: 10.1017/S0007123412000415
36. Greer SL, Bekker M, de Leeuw E, Wismar M, Helderman J-K, Ribeiro S, Stuckler D. **Policy, politics and public health**. *European Journal of Public Health* (2017.0) **27** 40-43. DOI: 10.1093/eurpub/ckx152
37. Hancock T. **Beyond health care: From public health policy to healthy public policy**. *Canadian Journal of Public Health* (1985.0) **76** 9-11. PMID: 4027852
38. Johns T. **Managing a policy experiment: Adopting and implementing recreational marijuana policies in Colorado**. *State and Local Government Review* (2015.0) **47** 193-204. DOI: 10.1177/0160323X15612149
39. Johnson, B., & Swearingen White, S. (2010). Promoting sustainability through transportation infrastructure? Innovation and inertia in the Kansas City Metropolitan Area. Journal of Urban Planning and Development, 136(4), 303–313. 10.1061/(ASCE)UP.1943-5444.0000027
40. Kadowaki J, Vuolo M, Kelly B. **A review of the current geographic distribution of and debate surrounding electronic cigarette clean air regulations in the United States**. *Health & Place* (2015.0) **31** 75-82. DOI: 10.1016/j.healthplace.2014.11.003
41. Karch, A. (2022). Innovation and diffusion: Connecting theory and method. In C. Weible & S. Workman (Eds.), Methods of the Policy Process, 181–207. 10.4324/9781003269083-7
42. Kavanagh MM, Parish K, Gupta S. **Drivers of health policy adoption: A political economy of HIV treatment policy**. *Policy & Politics* (2021.0). DOI: 10.1332/030557321X16108172938639
43. Macinko J, Silver D. **Diffusion of impaired driving laws among US states**. *American Journal of Public Health* (2015.0) **105** 1893-1900. DOI: 10.2105/AJPH.2015.302670
44. Maggetti M, Gilardi F. **Problems (and solutions) in the measurement of policy diffusion mechanisms**. *Journal of Public Policy* (2016.0) **36** 87-107. DOI: 10.1017/S0143814X1400035X
45. Mallinson DJ. **Schoolyard politics: Measuring and explaining variation in state antibullying policy comprehensiveness**. *State and Local Government Review* (2016.0) **48** 100-113. DOI: 10.1177/0160323X16655383
46. Michael GJ. **International coercion and the diffusion of regulatory data protection: International coercion and data protection**. *The Journal of World Intellectual Property* (2016.0) **19** 2-27. DOI: 10.1111/jwip.12051
47. Mitchell J, Stewart LS. **Emulation, learning, or competition? Examining inter-county anti-smoking laws in the state of Missouri**. *Public Administration Quarterly* (2014.0) **31** 317-346
48. Moloughney, B. (2012). The use of policy frameworks to understand public health-related public policy processes: A literature review: Final report. Peel Public Health. https://www.peelregion.ca/health/library/pdf/Policy_Frameworks.PDF
49. Moreland-Russell, S., Eyler, A., Barbero, C., Hipp, J. & Walsh, H. (2013). Diffusion of Complete Streets policies across US communities. Journal of Public Health Management and Practice, 19(3 Suppl 1), S89–S96. 10.1097/PHH.0b013e3182849ec2
50. National Collaborating Centre for Methods and Tools. (2018). Evidence-informed decision making: A model for evidence-informed decision making in public health. https://www.nccmt.ca/uploads/media/media/0001/02/5da8cf329a940bdd81a956a1984f05456c4a7910.pdf. Accessed 17 Mar 2022.
51. Nykiforuk, C., Eyles, J., & Campbell, H. (2008). Smoke-free spaces over time: A policy diffusion study of bylaw development in Alberta and Ontario, Canada. Health & Social Care in the Community,16(1), 64–74. 10.1111/j.1365-2524.2007.00727.x
52. Nykiforuk C, Campbell E, Macridis S, McKennitt D, Atkey K, Raine K. **Adoption and diffusion of zoning bylaws banning fast food drive-through services across Canadian municipalities**. *BMC Public Health* (2018.0) **18** 137. DOI: 10.1186/s12889-018-5061-1
53. Olive, A., & Boyd, B. (2021). Conclusion: (re)theorizing diffusion and transfer in Canada’s Federal System. In B. Boyd & A. Olive (Eds.), Provincial policy laboratories: Policy diffusion and transfer in Canada’s federal system, 154–67. Palgrave Macmillan UK. 10.1057/978-1-137-51781-4_2
54. Oliver, K. (2022). How policy appetites shape, and are shaped by evidence production and use. In P. Fafard, A. Cassola, & E. de Leeuw (Eds.), Integrating science and politics for public health (pp. 77–101). Palgrave Springer. 10.1007/978-3-030-98985-9_5
55. Olstad D, Campbell E, Raine K, Nykiforuk C. **A multiple case history and systematic review of adoption, diffusion, implementation and impact of provincial daily physical activity policies in Canadian schools**. *BMC Public Health* (2015.0) **15** 385. DOI: 10.1186/s12889-015-1669-6
56. Pacheco J. **The social contagion model: Exploring the role of public opinion on the diffusion of antismoking legislation across the American States**. *The Journal of Politics* (2012.0) **74** 187-202. DOI: 10.1017/S0022381611001241
57. Pacheco J. **Free-riders or competitive races? Strategic interaction across the American states on tobacco policy making**. *State Politics & Policy Quarterly* (2017.0) **17** 299-318. DOI: 10.1177/1532440017705150
58. Pacheco J, Boushey G. **Public health and agenda setting: Determinants of state attention to tobacco and vaccines**. *Journal of Health Politics, Policy and Law* (2014.0) **39** 565-589. DOI: 10.1215/03616878-2682612
59. Parkhurst, J. (2016). The politics of evidence: From evidence-based policy to the good governance of evidence (1st edition). Routledge, Taylor & Francis Group. 10.4324/9781315675008
60. Petridou E. **Theories of the policy process: Contemporary scholarship and future directions**. *Policy Studies Journal* (2014.0) **42** S12-S32. DOI: 10.1111/psj.12054
61. Place Research Lab. (n.d.). Diffusion of healthy public policies. https://placeresearchlab.com/diffusion-of-healthy-public-policies/. Accessed 22 Dec 2021.
62. Politis, C. E., Halligan, M. H., Keen, D., & Kerner, J. F. (2014). Supporting the diffusion of healthy public policy in Canada: The Prevention Policies Directory. Online Journal of Public Health Informatics, 6(2). 10.5210/ojphi.v6i2.5372
63. Raphael, D., Bryant, T., Mikkonen J., Raphael, A.(2020). Social determinants of health: The Canadian facts, 2nd edition. Oshawa: Ontario Tech University Faculty of Health Sciences and Toronto: York University School of
Health Policy and Management. https://thecanadianfacts.org/The_Canadian_Facts-2nd_ed.pdf
64. Rogers E. *Diffusion of innovations* (1962.0)
65. Rogers E. *Diffusion of innovations* (2003.0)
66. Sebhatu A, Wennberg K, Arora-Jonsson S, Lindberg SI. **Explaining the homogeneous diffusion of COVID-19 nonpharmaceutical interventions across heterogeneous countries**. *Proceedings of the National Academy of Sciences* (2020.0) **117** 21201-21208. DOI: 10.1073/pnas.2010625117
67. Septiono, W., Kuipers, M., Ng, N., & Kunst, A. (2019). Progress of smoke-free policy adoption at district level in Indonesia: A policy diffusion study. The International Journal of Drug Policy, 71, 93–102. 10.1016/j.drugpo.2019.06.015
68. Sieger, M., & Rebbe, R. (2020). Variation in States' implementation of CAPTA's substance-exposed infants mandates: A policy diffusion analysis. Child Maltreatment, 25(4), 457–467. 10.1177/1077559520922313
69. Shen GC. **Cross-national diffusion of mental health policy**. *International Journal of Health Policy and Management* (2014.0) **3** 269-282. DOI: 10.15171/ijhpm.2014.96
70. Shipan CR, Volden C. **Bottom-Up Federalism: The diffusion of antismoking policies from U.S. cities to states**. *American Journal of Political Science* (2006.0) **50** 825-843. DOI: 10.1111/j.1540-5907.2006.00218.x
71. Shipan CR, Volden C. **The mechanisms of policy diffusion**. *American Journal of Political Science* (2008.0) **52** 840-857. DOI: 10.1111/j.1540-5907.2008.00346.x
72. Shipan CR, Volden C. **Policy diffusion: Seven lessons for scholars and practitioners**. *Public Administration Review* (2012.0) **72** 788-796. DOI: 10.1111/j.1540-6210.2012.02610.x
73. Shipan CR, Volden C. **When the smoke clears: Expertise, learning and policy diffusion**. *Journal of Public Policy* (2014.0) **34** 357-387. DOI: 10.1017/S0143814X14000142
74. Train A, Snow D. **Cannabis policy diffusion in Ontario and New Brunswick: Coercion, learning, and replication**. *Canadian Public Administration* (2019.0) **62** 549-572. DOI: 10.1111/capa.12346
75. Trein P. **Europeanisation beyond the European Union: Tobacco advertisement restrictions in Swiss cantons**. *Journal of Public Policy* (2017.0) **37** 113-142. DOI: 10.1017/S0143814X16000167
76. Tricco, A. C., Lillie, E., Zarin, W., O’Brien, K. K., Colquhoun, H., Levac, D., Moher, D., Peters, M. D. J., Horsley, T., Weeks, L., Hempel, S., Akl, E. A., Chang, C., McGowan, J., Stewart, L., Hartling, L., Aldcroft, A., Wilson, M. G., Garritty, C., … Straus, S. E. (2018). PRISMA extension for scoping reviews (PRISMA-ScR): Checklist and explanation. Annals of Internal Medicine, 169(7), 467–473. 10.7326/M18-0850
77. Valente, T., Dyal, S., Chu, K., Wipfli, H., & Fujimoto, K. (2015). Diffusion of Innovations theory applied to global tobacco control treaty ratification. Social Science & Medicine, 145, 89–97. 10.1016/j.socscimed.2015.10.0
78. Volden C. **States as policy laboratories: Emulating success in the Children’s Health Insurance Program**. *American Journal of Political Science* (2006.0) **50** 294-312. DOI: 10.1111/j.1540-5907.2006.00185.x
79. Weible, C. M. (2014). Introducing the scope and focus of policy process research and theory. In P. Sabatier & C. Weible (Eds.), Theories of the Policy Process, 3–22. Westview Press.
80. Winder DW, LaPlant JT. **State lawsuits against “Big Tobacco”: A test of diffusion theory**. *State and Local Government Review* (2000.0) **32** 132-141. DOI: 10.1177/0160323X0003200205
81. World Health Organization. (2010). A conceptual framework for action on the social determinants of health: Debates, policy & practice, case studies.http://apps.who.int/iris/bitstream/10665/44489/1/9789241500852_eng.pdf
|
---
title: Optimization of energy and time predicts dynamic speeds for human walking
authors:
- Rebecca Elizabeth Carlisle
- Arthur D Kuo
journal: eLife
year: 2023
pmcid: PMC10030114
doi: 10.7554/eLife.81939
license: CC BY 4.0
---
# Optimization of energy and time predicts dynamic speeds for human walking
## Abstract
Humans make a number of choices when they walk, such as how fast and for how long. The preferred steady walking speed seems chosen to minimize energy expenditure per distance traveled. But the speed of actual walking bouts is not only steady, but rather a time-varying trajectory, which can also be modulated by task urgency or an individual’s movement vigor. Here we show that speed trajectories and durations of human walking bouts are explained better by an objective to minimize Energy and Time, meaning the total work or energy to reach destination, plus a cost proportional to bout duration. Applied to a computational model of walking dynamics, this objective predicts dynamic speed vs. time trajectories with inverted U shapes. Model and human experiment ($$n = 10$$) show that shorter bouts are unsteady and dominated by the time and effort of accelerating, and longer ones are steadier and faster and dominated by steady-state time and effort. Individual-dependent vigor may be characterized by the energy one is willing to spend to save a unit of time, which explains why some may walk faster than others, but everyone may have similar-shaped trajectories due to similar walking dynamics. Tradeoffs between energy and time costs can predict transient, steady, and vigor-related aspects of walking.
## Introduction
Many aspects of human walking are determined by minimization of metabolic energy expenditure. For example, the preferred step length (Atzler and Herbst, 1927) and step width (Donelan et al., 2001) minimize energy expenditure for a given steady speed, and the preferred steady speed approximately coincides with minimum energy expenditure per distance traveled (Figure 1A, Ralston, 1958). This speed, as well as the economy of walking, both decline with age, disability, or poor health. As such, preferred speed is widely employed as a clinically useful indicator of overall mobility (Afilalo et al., 2010; Studenski et al., 2011). However, there are naturally many other factors that also influence walking. All walking tasks have a beginning and end, and some may spend little or no time at steady speed. Some tasks may also occur with a degree of urgency, and some individuals may habitually walk faster than others, for reasons not obviously explained by economy. Energy economy is a powerful and objective explanation for steady walking speed, but it does not readily accommodate these everyday observations. Realistic walking tasks must therefore be governed by more than energy economy alone.
**Figure 1.:** *Humans prefer an economical speed for steady walking, but not all walking is steady.(A) The preferred steady walking speed v* coincides with minimum metabolic cost of transport (‘min COT’), which has a convex dependence on speed (after Ralston, 1958). (B) The distribution of human walking bouts during daily living, plotted as percentage of observed bouts vs. number of steps (in bins of ±1), as reported by Orendurff et al., 2008. About 50% of bouts were less than 16 steps (shaded area), observed from ten adults over fourteen days. (C) A typical walking task is to walk a given distance D, starting and ending at rest. (D) Walking speed is therefore expected to be a trajectory that starts and ends at zero, potentially differs from steady v*, and has a finite duration T. Hypothetical trajectories are shown as dashed lines.*
The specific energy measure thought to govern steady walking speed is the gross metabolic cost of transport (COT). Defined as energy expended per distance travelled and body weight (or mass), it has a convex dependency on speed. Its minimum (termed min-COT here) seems to predict the steady preferred speed of about 1.25 ms-1, as widely reported for laboratory settings (Ralston, 1958; Martin et al., 1992; Willis et al., 2005; Browning and Kram, 2005; Browning et al., 2006; Rose et al., 2006; Entin et al., 2010), and observed of some other animals (Hoyt and Taylor, 1981; Langman et al., 2012; Watson et al., 2011). However, much of actual daily living also involves relatively short bouts of walking (Figure 1B), with about half of daily bouts taking less than 16 steps as reported by Orendurff et al., 2008. Such bouts, say of distance D (Figure 1C), may spend substantial time and energy on starting from and stopping at rest, and relatively little time at steady speed. For example, in short bouts of walking up to about a dozen steps, peak speed is slower than the steady optimum, and only attains that value with more steps (Seethapathi and Srinivasan, 2015). There is a substantial energetic cost to changing speeds that could account for 4–$8\%$ of daily walking energy budget (Seethapathi and Srinivasan, 2015). If energy economy is important for walking, it should apply to an entire walking bout or task, and not only to steady speed.
Another important factor for walking is time. Time is valuable in part because energy is always expended even when one is at rest (Jetté et al., 1990), and because walking faster can save time to reach destination, but at greater energy cost (Ralston, 1958). Time is also subjectively valuable, because the urgency of a task, or even of an individual’s personality, surroundings, or culture, could influence their speed. It has long been observed that people walk faster in big cities than in small towns, by a factor of more than twofold (about 0.75–1.75 ms-1), or about ±$40\%$ of 1.25 ms-1 (Bornstein and Bornstein, 1976). Perhaps population density affects a person’s valuation of time (Bornstein, 1979; Levine and Bartlett, 1984; Li and Cao, 2019). Time is certainly a factor in deciding whether to walk or run (Summerside et al., 2018), and is considered an important factor in the general vigor of movements, beyond walking alone (Labaune et al., 2020). It is clearly worthwhile to expend more energy if time is of essence.
There are, however, challenges to incorporating time into walking. One method is to factor time into the equivalent of temporally discounted reward (Shadmehr et al., 2010), which refers to offering a reduced reward for longer durations, typically employed in fields such as movement vigor, foraging theory (Green and Myerson, 1996), and reinforcement learning (Sutton and Barto, 2018). Another is to express time as a cost that increases for longer movement durations, trading off against greater energy cost for shorter durations. Both the energy cost for an entire walking bout, plus a cost for time duration, could thus be combined into a single objective function to be minimized (Wong et al., 2021). This presents a second challenge, which is how to determine the optimum. Unlike the case of steady walking at a single speed (Figure 1A), an entire walking bout requires a time-varying trajectory of walking speed. This cannot be determined from the cost of transport curve, but can potentially be predicted by a quantitative, dynamical model. Simple models of walking (Figure 7), based on the pendulum-like dynamics of walking, can predict aspects such as optimal step length and step width (Kuo et al., 2005) for a steady speed, and optimal speed fluctuations for uneven steps (Darici and Kuo, 2022a; Darici and Kuo, 2022b). It remains to be determined whether they can predict the energetics and timing of walking bouts with transient conditions.
The purpose of the present study was to test whether the combined costs of energy and time can predict dynamic variation in walking speed. We propose a basic quantitative objective function called the Energy-Time hypothesis, which includes a cost for total energy expenditure or mechanical work for a walking bout, plus a penalty increasing with the bout’s time duration. We apply this objective to a computational walking model, using dynamic optimization to predict dynamic speed profiles for walking bouts of varying distance (Figure 1D). For relatively short walking bouts, this hypothesis predicts speeds that vary within a bout, and speed profiles that vary across bout distances. For longer distances, it predicts a steady walking speed, not as an explicit outcome but rather as an emergent behavior. To test these predictions, we performed a human subjects experiment, comparing empirical speed profiles against model predictions. If the model is able to predict human speed profiles, it may suggest that a valuation of time and energy can influence walking, and thus be compatible with walking bouts of any distance and any degree of urgency.
## Model predictions
A simple model of walking dynamics predicts theoretically optimal speed trajectories and walking bout durations. The Energy-*Time hypothesis* is that humans perform walking bouts that minimize an objective including the total energy and time expended for the bout. The dynamic optimization problem may be summarized asminimize(Energyexpenditure)+CT(Timeduration)subjectto:startingandendingatrestwithNstepsofpendulum−likewalkingdynamicsathuman−likesteplength where the total metabolic energy expenditure is evaluated for the entire walking task, and the time duration is weighted by a metabolic energy coefficient CT (in units of energy per time). In the model, positive mechanical work is used as a proportional indicator of human energy expenditure, with (lower-case) work coefficient cT. This coefficient is a valuation of time, and may be interpreted as the energy or work one is willing to expend to save a unit of time. The overall objective is to be minimized with an appropriate trajectory of the model’s speed, which in the human is the outcome of active control actions. The optimal control actions are subject to constraints, namely the specified distance of a walking bout and the governing walking dynamics (see Methods for details). Walking dynamics refers to the dynamics of the body, where the stance leg behaves like an inverted pendulum and the swing leg like a swinging pendulum. These dynamics also describe the mechanical work and energy associated with a speed trajectory, and how long each step takes. Step length was nominally kept fixed, and then varied in parameter sensitivity studies below. The time duration T of a bout is the outcome of the optimization, where greater valuation of time CT favors shorter duration.
The optimization predicts the speed profiles for a representative, ten-step task (Figure 2A). To focus on Energy first, the duration is kept fixed here. The Energy-Time objective predicts a gradual increase in speed, with a gently rounded profile that peaks mid-way through the bout. For this relatively short distance, little or no time is spent at steady speed. This contrasts with two other possibilities, to maintain steady speed at min-COT (Ralston, 1958), or to maintain steady acceleration. The steady min-COT objective produces a speed profile resembling a trapezoid, accelerating immediately to attain a fixed steady speed, maintained throughout the bout, before terminating just as quickly. Steady acceleration causes speed to increase linearly over time until peaking mid-bout, followed by a linear decrease back to rest. Here, all three alternatives are directed to walk the same distance in the same time, but at different costs.
**Figure 2.:** *A computational model of walking predicts that a rounded speed profile is most economical for a short walking bout of fixed time and distance.(A) Predicted speed profiles for a walking bout of ten steps, comparing minimization of Energy and Time (Energy-Time, solid line) against maintaining steady speed (min-COT, dotted line) or steady acceleration and deceleration (steady accel, dashed line). Energy-Time minimizes the total push-off (PO) work plus time expended for a walking bout, for a model with pendulum-like legs (inset). Steady min-COT walks at the steady speed that minimizes cost of transport, by accelerating immediately to that speed. Steady acceleration walks with linearly increasing speed until mid-point, then decelerates linearly back to rest. Energy-Time predicts a gently rounded speed profile, min-COT a trapezoidal profile (always at min-COT speed), and steady acceleration a triangular profile. Speeds are discretely sampled as the average forward speed over each step (filled dots), starting with an initiation impulse to accelerate from standing and a termination impulse to decelerate at the end (gray dots). (B) Positive work inputs for each hypothesis, including initiation work (gray dots) and push-off work (colored dots, one per step). Energy-time hypothesis predicts the least total work (inset bar graph compares Energy-Time "E-T", min-"COT", steady-"accel" costs). Predictions are for a dynamic walking model with pendulum-like legs (inset, see Methods). All predictions are designed for the same duration based on steady min-COT speed as a reference, resulting in cost of Time cT=0.020. Predictions are plotted in terms of normalized units based on body mass M, leg length L, and gravitational acceleration g; scale for typical human also shown, mass 70 kg, leg length 1 m.*
Examination of the positive work inputs reveals why Energy-*Time is* least costly (Figure 2B). Its gentle acceleration requires moderate push-offs, which trail off over time as the model nearly coasts to a stop at destination, taking advantage of each step’s collision loss to reduce speed at little cost. In contrast, the steady min-COT objective pays a high cost to initiate gait, and then a moderate and constant amount of work for all push-offs. It also does not take advantage of coasting to a stop. Steady acceleration pays a high cost to peak at a high speed, which is not made up for by greatly reduced push-offs as it comes to a stop. Some intuition may be gained by considering the analogous situation of a vehicle driving a short fixed distance between two stop signs, in fixed time. It is generally economical to accelerate and decelerate gradually, and not necessarily maintain steady speed except beyond a certain distance. A trapezoidal (min-COT) speed profile is not economical, because considerable energy is spent in fast acceleration, and braking maximally at the end is more wasteful than lifting off the accelerator early and coasting. A triangular (steady acceleration) profile is also not favorable, due to the work needed to briefly attain a high speed. Of course, walking and driving have different dynamics, but both have similar energetic loss rates that increase approximately with the cube of speed. The higher losses incurred at greater speeds is an important reason for the Energy-time optimality of a rounded speed profile. For this task, the min-COT hypothesis ultimately costs $11\%$ more total work, and the steady acceleration hypothesis $31\%$ more, than minimizing Energy-Time.
Having established the energetic advantages of the Energy-Time hypothesis, we next examine how the optimal speed profiles vary with Time and other model parameters (Figure 3). Here, there are three parameters of interest: the value of time cT, step length, and walking bout distance. We considered step lengths s fixed at nominal (0.68 m), at slightly shorter and longer lengths (0.59 m and 0.78 m), and increasing with speed according to the human preferred step length relationship (see Methods for details). We also considered bouts of one to twenty steps, or about 0.68 m–13.7 m, as well as time valuations cT ranging ten-fold, 0.006–0.06 (dimensionless). Regardless of the combination of these parameters, a few characteristics emerge. The speed profiles generally retain a gently rounded profile (Figure 3A), smoothly accelerating from rest and leveling off at a peak speed before decelerating back to rest. Unlike the trapezoidal, min-COT prediction, the human speed profiles are always peaked, particularly for short bouts. The longer the distance, the greater the peak speed (unlike min-COT), and the more sustained that peak, contrast to more rounded speed profiles for shorter distances. The acceleration and deceleration slopes increase slightly with longer bouts, and only for distances of about 10 m or more is there a steady gait near peak speed. The peak speed also initially increases sharply with walking distance (Figure 3B), but then approaches an asymptote for greater distances, as the cost of acceleration and deceleration becomes inconsequential to overall cost J. In fact, the asymptotic peak speed for long walks is a steady speed, not unlike the minimum-COT speed. But for finite walk distances, the speed profile generally does not agree with the steady min-COT hypothesis, because it varies with bout distance, and dynamically within each bout.
**Figure 3.:** *Energy-Time hypothesis predicts a family of speed profiles.(A) Predicted speed profiles vs. time for a range of walking distances, with longer bouts reaching higher and steadier peak speeds. In the main plot, multiple predictions for different time valuations and step lengths are scaled and superimposed on each other to emphasize self-similarity. Original, unscaled traces are shown in surrounding insets. (Horizontal insets, bottom:) Three different step lengths including shorter and longer steps than nominal, and human preferred step length relationship (dashed lines); main plot also includes nominal step length. (Vertical insets, right:) Varying valuation of time cT results in two-fold variation in peak speeds (labeled) and walk durations. The time cost and step length therefore affect only how quickly the task is completed, and not the shape of the family of speed profiles. Shown are trajectories of discrete speed, defined as the average forward speed over each step. (B) Peak speeds are predicted to increase sharply with distance, approaching an asymptote for distances of about 12 m or more. Again, despite different peak speeds, the curves are self-similar and can be scaled to a single shape (thick lines). (C) Walking durations increase with distance, with slightly curvilinear relationship (also scaled to a single shape, thick lines). In (A), time cost cT is varied between 0.006 and 0.06 (in units of Mg1.5L0.5), and distances range from 2 to 20 steps. Model predictions are plotted in dimensionless units, using body mass M, leg length L, and gravitational acceleration g as base units; scale for typical human also shown, mass 70 kg, leg length 1 m.*
Another feature of the Energy-Time prediction is consistency with respect to parameter values (Figure 3A). The main free parameter is the time valuation cT, for which higher values call for higher peak speeds, and therefore shorter walking durations. But with peak speeds ranging more than two-fold (Figure 3A, inset), the speed profiles all had similar shape. In fact, scaling each of the profiles in time and amplitude yielded a very similar family of trajectories regardless of parameter values (Figure 3A). This is also the case for variation in step length, with nominal, long, and preferred human step lengths all producing similar trajectories. Similarly, the peak speed vs. distance curves resembled a saturating exponential regardless of parameter values (Figure 3B), and these were also scalable in amplitude to yield a single family of curves. Walking durations vs. distance (Figure 3C), also had similar, scalable and curvilinear shape for all parameters. Similar profiles are produced regardless of whether the model takes step lengths that are fixed, or that scale according to the empirical step length vs. walking speed relationship for steady walking (insets, Figure 3A). We therefore subsequently keep step length fixed (equivalent of 0.68 m for human) for simplicity. As a result, the time cost coefficient cT is effectively the model’s sole free parameter, and the predicted speed profile shapes scale very consistently with respect to that parameter.
There are thus three main predictions from the model that can be tested in human. First, the speed profiles should fall within a single consistent family, which includes more rounded shapes for short walks, and flatter for longer walks (Figure 3A). These profiles should exhibit self-similarity, and be scalable in peak speed and time to resemble a single, relatively uniform family of profiles. Second, the peak speed should increase with distance, with an approximately exponential saturation toward an asymptote (Figure 3B). Again, that relationship is expected to be scalable by peak speed, and testable by a single saturating exponential. And third, walking durations should increase with distance, in a slightly curvilinear relationship (Figure 3C) approaching a straight-line asymptote for longer distances. For shorter distances, much of the time should be spent accelerating and decelerating, compared to relatively brief cruising periods that become proportionately greater with distance. We thus treat the time valuation cT as an empirical parameter that mainly affects the scale, but not the shape of the speed profiles and dependency on distance.
## Experimental results
The human speed profiles for all trials and all distances were found to exhibit consistent profiles between subjects and between individual trials (Figure 4). These profiles resembled predictions from the Energy-Time hypothesis. Qualitatively, humans produced inverted U profiles similar to model, with sharper and lower peak speeds for shorter bouts. Longer bouts had higher and flatter peaks, where a steady speed could be discerned. Each individual subject walked at a somewhat different speed and for a somewhat different time (Figure 4A). For example, the range of peak speeds across subjects, observed for the longest (12.7 m) bout, was 1.21 ms–1.78 ms–1, and the corresponding range of durations was 8.51–11.86 s. Nevertheless, the profile shapes were all quite similar across subjects. In contrast to the min-COT hypothesis, the human peak speeds increased with distance, many well below the min-COT speed of about 1.25 ms–1. In addition, the human speed trajectories did not resemble the trapezoidal profiles of the steady min-COT hypothesis for all distances, nor the triangular profiles of steady acceleration.
**Figure 4.:** *Human speed profiles vs.time for (A) all subjects (N=10), and (B) all subjects normalized to the average.Body speeds are plotted for all ten walking bout distances (colored lines). In (A), one representative subject is highlighted (thicker lines) to show a typical person’s variability between trials. In (B), all traces for each subject are normalized by that person’s average peak speed (‘Speed norm’) for the longest distance, and by their average time for that bout (‘Timing norm’). Also shown are the mean walking bouts across subjects (thick black lines, N=10) for each distance, to illustrate how different subjects resemble each other despite varying in how fast they walk. Average bouts were computed by resampling each trial to the most common step count for each distance, averaging across such profiles for each distance, and then rescaling time to reflect the average duration for each distance. Body speed is defined (in Methods) as an average for each step, dividing step length by the time between between mid-stance instances.*
The experimental speed trajectories were scalable in speed and time, to yield a self-similar family of trajectories (Figure 4B). Each individual’s trajectories were normalized in time by the duration for that subject’s longest bout, and in speed by the maximum speed of their longest bout. These were then re-scaled to match the average duration and peak seed across subjects, to yield a normalized set of speed profiles for all subjects (Figure 4B). The resulting normalized trajectories reveals considerable similarity between individuals, with a single, relatively uniform family of profiles for all subjects. Thus, the peak speed and duration of a walking bout of 2 m was consistently related to one of 12 m, and vice versa.
This scalability may be quantified in terms of peak speeds and durations. Examining the peak speed for each distance reveals a consistent pattern (Figure 5A). Peak speeds increased with distance, sharply for short distances and then saturating for longer distances. The overall pattern resembled a saturating exponential, similar to model predictions. The overall maximum speed was 1.52 ± 0.14 ms–1 (mean ± s.d. across subjects), almost always for the longest distance. We normalized each individual’s peak speed by their own maximum, and found the resulting peak speed vs. distance curves to be scalable into a single normalized curve across subjects. With normalization, the variability (s.d. across individuals) of peak speeds was reduced significantly ($$P \leq 1.6$$×10-6), by 0.07±0.02ms−1 (mean ± s.d.) across all bout distances or about $54\%$ compared to un-normalized. Thus, even though each individual walked at their own pace, that tendency was consistent across all distances. Much of the inter-subject variability was reduced by normalizing the peak speeds, revealing a common relationship between peak speed and bout distance.
**Figure 5.:** *Human walking bouts show increases in (A) peak speed and (B) walking duration vs. distance.(A) Peak speeds are shown for each walking distance, averaged across subjects’ normalized data (filled symbols, N=10), along with variability (standard deviation, thin black error bars). These are accompanied by a saturating exponential fit (thick solid line, R2=0.86). Also shown are fits for each subject’s unnormalized data (thin colored lines; dimensional vertical axis at left), and unnormalized variability between all trials (standard deviation, light gray error bars). (B) Walking durations are shown for each walking distance, averaged across subjects’ normalized data (filled symbols), along with variability (standard deviation, error bars) and a saturating exponential fit (R2=0.98). Shaded areas denote rise time (0%–90% of peak speed), cruise time (90% of peak and greater), and fall time (90%–0%). Rise and fall times appear to dominate shorter walking bouts, and cruise time for longer walking bouts. Filled black dots denote mean data, error bars denote s.d. The entire range of unnormalized peak speeds and durations for all subjects is shown in Figure 4, along with normalization definitions (Speed norm, Time norm). Dimensional vertical axis (left) is based on the average normalization constant relative to normalized data (right vertical axis). Normalization yielded reduced variability in peak speeds and durations.*
There was a similarly consistent pattern for walking durations across distances (Figure 5B). Walking durations increased with distance in a slightly curvilinear fashion. Again, we normalized each individual’s durations by the duration for the longest bout (9.86±0.75s), and found the duration vs. distances to be scalable into a single normalized curve across subjects. With normalization, the variability of durations was also reduced significantly ($$P \leq 0.03$$), by 0.10±0.13s across all bout distances, or about $18\%$ compared to un-normalized. Similar to peak speeds, much of the inter-subject variability was reduced by normalizing. There was a common and consistent relationship between different walking bouts, similar to model predictions.
The change in peakiness or flatness of speed profiles was indicated by the time spent accelerating, decelerating, or at approximately constant speed (Figure 5B). This was described by rise time, defined as the time to accelerate from $0\%$ to $90\%$ of peak speed, cruise time as the time spent at $90\%$ of peak speed or more, and fall time as the time to decelerate between $90\%$ and $0\%$ of peak speed (Figure 5B). These measures of time increased with bout distance. As a fraction of each bout’s duration, the rise and fall times appeared to take up a greater proportion for shorter bouts, and only a very small proportion was spent at steady speed. Conversely, cruise time took up a greater proportion of the time for longer bouts. These behaviors were consistent with predictions from the Energy-Time hypothesis.
The peak speed was described reasonably well by a saturating exponential (Figure 5A). An ad hoc, least-squares nonlinear fit to the normalized data yielded a saturating exponential curve[1]v(D)=cv(1−e−D/dv) where v(D) is the peak speed as a function of total walking distance D, and fitted values were cv=1.516ms−1 (1.496, 1.536 CI, $95\%$ confidence interval) and dv=1.877m (1.798, 1.955 CI), with a goodness-of-fit of R2=0.86. The curve fit shows that there was considerable consistency in maximum speed; even short walking bouts of slow peak speed were still consistent with longer bouts of higher speed.
Similarly, walking duration increased with walking distance (Figure 5B), with a slightly curvilinear relationship. The total walk duration T(D) may be treated as a saturating exponential approaching a straight asymptote, defined ad hoc as distance D divided by preferred steady walking speed plus an offset T0. The curve was of the form[2]T(D)=DvT+T0(1−e−D/dT).
where fitted coefficients were vT=1.494ms−1 (1.466, 1.521 CI), T0=1.470s (1.375, 1.565 CI), and dT=0.790m (0.610, 0.970 CI).
We also performed similar analyses on a grass walking surface to test for sensitivity to slightly uneven terrain. An identical set of conditions was collected on short grass outdoors. The fit of peak speed vs. bout distance yielded cv= 1.446 ms–1, dv= 1.822 ms–1 (R2=0.85), and for duration versus distance vT= 1.426 ms–1, T0= 1.336 s, and dT= 0.503 m (R2=0.98). These relationships were quite similar to those obtained on sidewalk.
We next estimated the relationship between human valuation of time and steady walking speed (Figure 6). Here we used an empirical human metabolic power curve (Figure 6A) from literature (Elftman, 1966) to predict how steady walking speed should increase with metabolic value of time CT (Figure 6B), and how the energetic cost of transport vs. steady walking speed (Figure 6C) may be regarded in terms of competing costs for Energy and Time. This human power curve was fitted to the model, to facilitate scaling the model’s mechanical energy into human metabolic energy. The optimal steady walking speed emerges from that curve (Srinivasan, 2009), as a function of CT (Figure 6A). Based on these crude assumptions, a time valuation of zero yields the same optimal speed v* of 1.25 ms–1 as min COT, and close to the minimum steady speed (among subjects) of 1.21 ms–1 observed here. It is instructive to increment CT by multiples of the metabolic equivalent (MET), a standard physiological resting rate of about 1.23 Wkg–1 [Watts per kilogram] (Jetté et al., 1990). An increment of 1 or 2 MET yields optimal speeds of 1.54 ms–1 or 1.75 ms–1, respectively, quite close to the observed mean and maximum steady speeds (among subjects), respectively (of 1.52 ms–1 and 1.75 ms–1). Thus, if the same metabolic power curve were applicable to all, the slowest subject would have valued time at about zero MET, the mean subject at +1 MET, and the fastest at +2 MET. This also suggests that most subjects preferred faster steady speeds than min-COT. Incrementing CT yields diminishing returns in speed (Figure 6B), because it is increasingly costly to walk faster. One interpretation afforded by the Energy-*Time hypothesis* is that there is an effective cost of transport that may be separated into two terms (Figure 6C), Walking and Time: one for the net metabolic cost for walking alone (due to push-off work), the other a cost of time that lumps the resting rate with (or within) an individual’s CT. This reveals a trade-off, where the cost of walking increases with speed, and the cost of time decreases (hyperbolically with speed), such that the two opposing curves (or rather their opposing slopes) determine an optimum. A greater valuation of time adds to this effective cost of transport, equal to the actual metabolic energy plus a subjectively scalable cost of time, per distance traveled. This shows how the effective Energy-Time cost per distance is minimized at higher speeds for greater CT.
**Figure 6.:** *Prediction of steady walking speed emerges from Energy-Time hypothesis.(A) Human metabolic power vs. speed for steady walking (adapted from Fig. 11 of Elftman, 1966), along with a model-based curve fit (R2=0.999; see Equation 15). Faster walking can be produced by valuing time more, with metabolic CT=1MET yielding 1.54 ms–1 and 2MET yielding 1.75 ms–1. (MET is metabolic resting rate, serving as a standard reference value.) (B) Model steady speed vs. value of time CT increases such that each increment of CT in model yields a diminishing increase in speed, due to the increasingly high energetic cost of walking faster. Walking speeds observed in present experiment (‘Exp’ range) may be interpreted as human CT ranging from about 0–2 MET above resting. (C) Model energetic cost of transport (COT) may be regarded as the sum of two competing costs: a net physiological cost for Walking and a scalable cost for the Time expended. Steady walking speed is optimized where the two costs have equal and opposite slope. As the valuation of time CT increases, preferred steady speed increases. The valuation of time includes the resting metabolic rate plus a subjective component that does not literally cost energy. The valuation of time represents how much metabolic energy an individual is willing to spend to save a unit of time. The three marked speeds (asterisks ‘*’) roughly denote the minimum, mean, and maximum steady speeds observed here; they correspond with a gross valuation of time starting at resting rate, and incremented by one or two MET.*
## Discussion
We had sought to test whether humans optimize not only metabolic energy but also a valuation of time spent walking. Although the prevailing theory of minimizing the energetic COT explains steady walking, it does not explain shorter walks that lack a steady speed, nor does it readily accommodate individual tendencies toward faster or slower speeds. We found that humans walk bouts of finite distance with a trajectory of speeds varying with distance. These bouts fall within a consistent family of trajectories across subjects, despite individual differences in overall speed or duration. These results are in agreement with a simple mechanistic model of walking, governed by optimization. The findings suggest that humans optimize a combined objective that trades off the energy to arrive at destination against the time it takes to get there.
Each human walking bout consisted of a dynamically varying trajectory of speed with an inverted U shape. Many of these bouts included a period of steady walking, at speed similar to (but usually faster than) the supposed min-COT (Ralston, 1958), but mainly for the longer distances (Figure 4). Shorter bouts of say 10 m or less exhibited a relatively brief peak slower than the typical min-COT speed (Figure 5B). All such bouts also spent substantial time and energy in acceleration and deceleration (Figure 5B). Moreover, short distances such as this are quite ecological, accounting for about half of the daily living walking bouts reported by Orendurff et al., 2008, with acceleration and deceleration potentially accounting for 4–$8\%$ of daily walking energy budget (Seethapathi and Srinivasan, 2015). This contrasts with the min-COT hypothesis, which predicts only a single steady speed, and cannot explain how to start or end a bout. People usually walk to a destination of known and fixed distance, for which it is more sensible to minimize the total energy for that distance, rather than energy per distance (min-COT). This is not to dismiss the energy spent for steady walking, which has been well-characterized in laboratory settings where the vast majority of published studies have been performed. But in daily living, humans walk a variety of speeds and distances, many too short to be steady. To our knowledge, this is the first study to consider the time and energy spent in brief walking bouts.
Even though there were considerable differences between individuals, each subject was quite consistent within their own walking bouts. Those with a slower or faster peak speed during longer bouts were also consistently so during shorter, non-steady bouts (Figure 5), as evidenced by the $54\%$ reduced variability after normalizing peak speeds by the longest bout. Moreover, the bouts across all subjects were scalable to a single, self-similar family of trajectories (Figure 4). These trajectories were not consistent with a fixed acceleration or deceleration profile (Figure 2A), and instead exhibited a greater peak speed and longer time to that peak with greater distance (Figure 4B). This pattern suggests that there are systematic criteria or principles that govern walking bouts of finite distance. Even though some individuals are faster than others (Figure 4A), they all seem to follow similar principles.
These observations agree with the primary hypothesis that humans optimize for energy and time. The trade-off between the two, described by the valuation of time CT (for human metabolic cost, or cT in terms of model work), is readily explained for steady walking. A valuation of zero corresponds to minimizing the gross metabolic cost of transport (Figure 6B), yielding the min-COT speed. ( Or equivalently, resting metabolic rate could be treated as part of a running time cost, trading off against the net metabolic rate for walking, as in Figure 6C.) Of course, humans do not always walk the same speed, and faster speeds save time but cost more energy to cover a given distance. One’s actual steady speed must therefore be observed empirically, from which their valuation of time may also be estimated (ranging roughly 0–2 MET, Figure 6B). But what sets time valuation apart is its predictive value for non-steady walking. The speed trajectory for a fixed bout distance need not even contain a steady portion, and is often dominated by acceleration and deceleration. The Energy-Time model can nevertheless predict an entire family of trajectories across a variety of distances (Figure 3) despite individual-specific step lengths and time valuations. The model also suggests how peak speeds, and time to peak, and durations should increase with distance (Figure 3), similar to human data. All this is based on one free parameter, the individual-specific valuation of time. That valuation may depend on complex physiological and socio-psychological traits, but it nonetheless appears to have predictive value for a given context. Not tested here is the presumption that different contexts, for example changing the saliency of a task or adding time pressure, will also lead to systematic changes in walking bouts. If an individual’s valuation of time can be estimated empirically, our hypothesis provides an operational means of integrating it into a quantitative model.
These predictions are produced by a mechanistic model governed almost entirely by dynamics. The timing comes from the dynamics of pendulum-like walking, and the energetics from the step-to-step transition between pendulum-like steps. The step-to-step transition requires mechanical work to accelerate and to restore collision losses, such that for short walks it is uneconomical to accelerate quickly to min-COT speed (Figure 2, steady min-COT). An alternative is to accelerate more gradually, but that is costly because of the high peak speed attained (Figure 2, steady accel). The model thus favors an intermediate and smooth acceleration to a slower and continuously varying speed with an inverted U shape. Separate studies have found step-to-step transition work to predict human metabolic energy expenditure as a function of step length (Donelan et al., 2002) and changing speed (Seethapathi and Srinivasan, 2015). Here, we have constrained the pendulum-like dynamics so that there is only one physical parameter, step length, which in any case has very little effect on the characteristic shape of speed trajectories (Figure 3). As such, this was set to nominal value and not fitted to data, making the model as predictive as possible. Indeed, the very same model also predicts human compensation strategies for walking on surfaces of uneven height (Darici and Kuo, 2022a; Darici and Kuo, 2022b). Of course, the human body has many degrees of freedom capable of far different motions, but model analysis suggests that pendulum-like walking is the most economical means to move the COM at slow to moderate speeds (Srinivasan and Ruina, 2006), and that push-off during the step-to-step transition is the most economical means of powering such pendulum-like walking (Kuo, 2001). These models are predicated on mechanical work as the major cost, and the COM as the major inertia in the system.
It is instructive to consider what other models might explain or predict our experimental results. We did not explore more complex models here, but would expect similar predictions to result from any model based on pendulum-like walking and step-to-step transitions. This includes those cited in the previous paragraph, as well as a family of such models including three-dimensional motion (Donelan et al., 2001), knees and bi-articular actuators (Dean and Kuo, 2009), and plantarflexing ankles (Zelik et al., 2014). It is also possible that more complex, musculoskeletal models also perform substantial work and expend energy for step-to-step transitions, and might therefore agree with the present model. But here, relatively simple principles account for a fairly wide array of predictions (Figure 3), which are unlikely to result from happenstance. We are also unaware of any current hypothesis that could plausibly substitute for the present one. We therefore doubt if alternative models not based on pendulum-like principles could predict or reproduce these results, except with numerous fitted parameters.
This model is optimized with an additional control parameter, for the valuation of time. Time has long been recognized as a factor in the pace of life (Levine and Bartlett, 1984), and in reward and vigor in motor control (Shadmehr et al., 2010). It is typically expressed as a temporal discounting of reward, which appears key to human decision making and the theory of reinforcement learning. Here we expressed it as a trade-off equivalency between energy and time. This was mainly due to the need for compatibility with our energetics model, but also because neither model nor experiment included an explicit reward to be discounted. We used a simple, linear valuation of time in terms of energy, rather than a nonlinear, exponential or hyperbolic temporal discounting factor (Green and Myerson, 1996). Energy is a physiological cost endemic to life, that is not obviously more or less valuable at different points in time. It is sufficient to predict and explain the present results, and there is currently insufficient evidence to favor a nonlinear cost over our linear valuation. But regardless of the particular formulation, a default valuation of time may be an individualistic trait, generalizable to other tasks such as hand and eye movements (Labaune et al., 2020). Indeed, we have found a similar valuation of time to explain how reaching durations and speed trajectories vary with reaching distance (Wong et al., 2021). Another implication of our model is that humans may incorporate prediction of time within central nervous system internal models. Such models have long been proposed to explain humans predict and adapt their movement trajectories, for example to novel dynamics (Todorov, 2004). If movement duration is also part of human planning, it suggests the ability to predict not only movement trajectories and energetics, but also time. Here such prediction is made operational within a quantitative model.
Valuation of time offers another perspective on minimizing the gross energetic cost of transport. Actual walking tasks are not purely steady, and are probably planned with consideration of what happens at the destination. Long and Srinivasan, 2013 proposed a task to minimize the total energy expended to walk to destination within a more than ample allocation of time. They showed that total energy should be optimized by mixing resting and walking (and running if necessary). Suppose the task is extended to an indefinite duration, where a considerable amount of time is spent resting. The optimal total energy and walking duration may be found by applying our Energy-Time objective with time valuation (CT) equal to zero (Figure 6). Walking faster than optimal would yield more time to rest, but at a greater total energy cost for walking. Walking slower would cost less energy for the walking motion alone, but at a greater total cost due to less time available to rest. After all, CT is the energy one is willing to expend to save a unit of time, and the resting rate is the energy expended to rest for a unit of time. This may seem like a trivial restatement of the min-COT hypothesis, but it differs in two important ways. First, it can predict both the duration of walking and the entire speed trajectory, even for short bouts where there is no steady portion. Second, it considers how valuable time is at the destination. Minimizing the gross cost of transport is most sensible for maximizing the survivable range distance (Srinivasan, 2009), which may not be a concern in modern life where survival rates are high, walks short, and calories plentiful. Rather, it may be a sensible default to value time at close to the resting rate (particularly for long walks), and then to vary the valuation dependingon context. One might thus rush toward a long-lost friend or hurry in a big city, if the time spent at destination is far more valuable than resting (Bornstein, 1979; Levine and Bartlett, 1984; Li and Cao, 2019). Similarly, we do not consider walking slowly to be a waste of energy per distance, but rather a waste of time. Even then, there are cases when humans might wish to waste time, for example to avoid an odious task, according to the expression ‘the slow march to the gallows’.
The consistency of individual walking trajectories may have practical implications. Although walking speed is used as a clinical indicator of mobility, it is difficult to standardize (Middleton et al., 2015), because evaluations may be confined to the length of the available walkway, which may be too short (e.g. less than 10 m) for a steady speed to be reached. But given the time to walk a fixed distance, it may be possible to predict the duration and steady speed for another distance, referenced from a universal family of walking trajectories. We have identified one such family that applies to healthy young adults with pendulum-like gait. We do not know whether that family also applies to older adults, who prefer slower steady speeds and expend more energy to walk the same speed (Malatesta et al., 2003). Perhaps an age-related valuation of time might explain some of the differences in speed. Of course, some clinical conditions might be manifested by a deviance from that family, perhaps in the acceleration or deceleration phases, or in how the trajectories vary with distance. If quantified, such deviance might prove clinically useful. The methodology employed here does not require specialized equipment beyond inertial measurement units, and the characterization of distance-dependent speed trajectories can potentially provide more information than available from steady speed alone.
The Energy-*Time hypothesis* could be tested by further inquiries. We have thus far regarded the valuation of time as a relatively fixed parameter for each subject. That valuation is likely influenced, and therefore testable, by many contextual factors, including physiological and socio-psychological variables and task constraints. For example, caffeine intake, feeding status (e.g. Taylor and Faisal, 2018), or monetary reward could be used make time more valuable as a trade-off against energy. Conversely, energy may be helpful for assessing the valuation of time (or temporally discounted reward), which is not easy to measure other than indirectly. ( *Energy is* also a universal currency, because all animals use energy, whereas only some use money.) Walking has a well-characterized physiological energy cost, and could serve as a useful trade-off against time or reward. The hypothesized optimal gait is the point at which the costs of energy and time have equal and opposite slopes (i.e., partial derivatives) with respect to an independent variable such as speed (e.g. Figure 6C), carried load, or incline. There are thus a variety of opportunities to manipulate the energetic cost of walking, as a means to assess the proposed valuation of time.
There are a number of limitations to this study. Although we tested model predictions in terms of speed trajectories, we did not measure mechanical work or metabolic energy expenditure in human subjects, which would provide greater insight regarding the proposed trade-offs against time. We also did not evaluate each individual’s metabolic cost of transport vs. speed, which would reveal more precise differences between the min-COT speed and the actual self-selected speed. Nor did we evaluate gait kinematics or kinetics, which may be helpful for detailing other ways that walking bouts vary with distance. The simple walking model also only includes a crude representation of step-to-step transitions, which we have crudely estimated to account for as much as $70\%$ of net metabolic cost in steady walking (Donelan et al., 2002). We did not include other factors such as forced leg motion and step length modulation (Doke et al., 2005, up to say $33\%$ of cost) that likely also affect energetic cost, and could therefore be used to test the valuation of time. Nor did we include factors such stability (Bauby and Kuo, 2000; Donelan et al., 2004; Rebula et al., 2017) and three-dimensional motion (Donelan et al., 2001), despite being part of our previous models, because we believe them to contribute little to the present task. In fact, because the optimal control model successfully completes the walking task, its feedforward motion attains implicit stability, which could reduce (but not eliminate) the need for feedback stabilization (Darici and Kuo, 2022a; Darici and Kuo, 2022b). We also did not include an explicit reward, which could facilitate assessment of energy and time in terms of other trade-offs such as money or food. In fact, the Energy-*Time hypothesis* should be regarded as a subset of the many factors that govern human actions, rendered here in a simple but quantitative form.
## Conclusion
Humans appear to select walking speed dynamically to minimize a combination of energy and time expenditure. This is both compatible with and extends the traditional hypothesis that humans minimize gross energy expenditure per unit distance. We found it more accurate to minimize the total cost of a walking bout, due to the ability to predict an entire speed trajectory, with the optimal steady speed as an emergent property. By including a cost for time expenditure, we introduce a quantitative and operational means to make walking models compatible with the study of movement vigor. Tasks may also be broadened beyond walking, to include consideration of the reward to be gained or further energy to be expended once the destination is reached. Walking may thus be integrated into broader questions of how and why humans take the actions they do. As a modification to the traditional adage about money, we suggest that ‘*Time is* energy’.
## Methods
We experimentally tested how human walking speed varies with walking distance. The speed trajectories observed from human subjects were compared against predictions from the Energy-*Time hypothesis* and against the minimum-COT speed. To formulate the hypothesis and make quantitative predictions, we expressed it as an optimal control problem including both energy and time. We first state the hypothesis for human walking, and describe how it is adapted for a simple walking model to yield predicted speed trajectories. This is then followed by description of the experiment regarding human walking speed, and finally an analysis of steady speed as a property of the model.
## Walking model
We use the ‘simplest walking model’ (Kuo, 2002) to operationalize this optimization problem (Figure 7A). The model treats the stance leg as an inverted pendulum and requires mechanical work to power the gait. The body center of mass (COM) is modeled as a point mass supported by the stance leg, so that each pendulum-like step follows an arc, which itself requires no energy input. Work is performed during the step-to-step transition (Figure 7B), to redirect the COM velocity from forward-and-downward the end of one arc, and forward-and-upward at the beginning of the next. This is accomplished most economically with an active, impulsive push-off along the axis of the trailing leg, immediately followed by an impulsive, dissipative collision between the rigid leading leg and ground. In steady gait, the optimal push-off restores the collision losses, with mutually canceling impulses of equal magnitude. Speeding up is a matter of a greater push-off than collision, and a net increase in COM velocity during the step-to-step transition (Figure 7C). Positive and negative work are proportional to the square of the push-off and collision impulses, respectively (Kuo, 2002), so that speeding up also dissipates less collision energy than steady gait. Slowing down is the same in reverse, with collisions exceeding push-offs. This model predicts how step-to-step transition work for steady walking should increase as a function of step length and step width (Donelan et al., 2002). The model mainly predicts mechanical work for push-off, which appears to be a proportional predictor of the majority of human metabolic energy during steady walking (Donelan et al., 2002). That work also yields a mechanical cost of transport that varies curvilinearly with steady speed, similar to the empirical metabolic curve (Figure 1a; Ralston, 1958). There are of course other contributions to the metabolic cost of walking such as to move the swing leg (Kuo, 2001), but of smaller magnitude than step-to-step transitions, which are to be tested alone for their predictive value. Details of this model have been described in greater detail previously (Darici et al., 2020; Kuo, 2002), and are recounted only briefly here.
**Figure 7.:** *Simple optimization model of walking.(A) Walking dynamics modeled as a point center-of-mass (COM, mass M), supported by an inverted-pendulum stance leg (length L). (B) The inverted pendulum stance phase is punctuated by a step-to-step transition, modeled with an impulsive push-off (PO) from the trailing leg, followed by impulsive, inelastic collision (CO) with leading leg and ground. The COM velocity is v- at end of stance, then is redirected by PO and CO to yield velocity v+ at end of step-to-step transition, beginning the next stance phase. (C) For the model to speed up, the magnitude of PO must exceed that of CO, and v+ must have greater magnitude than v-. (D) The walking bout is initiated by a forward impulse applied at the pelvis, described by positive work u0.*
A walking bout consists of a sequence of N steps, starting and ending at rest. It may be described by the discrete sequence of body speeds vi ($i = 1$,2,…,N), each equal to the distance traveled for step i divided by that step’s time duration τi. The model begins at rest in an upright position (Figure 7D), and is set into motion by a forward initiation impulse acting on the pelvis. In humans, the torso can serve as an inertia that the hip muscles can act against, but for simplicity this action is represented as a translational impulse at the pelvis, summarized by the associated positive work u0. The total positive work performed by the model consists of the work from initiation and the successive push-offs, a sequence ui ($i = 0$,1,…,N). There is also a corresponding sequence of dissipative collision impulses by the leading leg, and a dissipative gait termination to end at upright.
The step-to-step transition starts just before leading leg ground contact contact, and consists of a perfectly impulsive push-off from the trailing leg, followed in immediate succession by a perfectly inelastic and impulsive collision of the leading leg with ground. The COM velocity at the end of one stance phase is vi-, directed forward and downward according to the pendulum arc. Mechanical work is performed only during the step-to-step transition, with a succession of ideal impulses. First is positive push-off work from the trailing leg, directed from its foot to the COM, and second is a perfectly inelastic heel-strike collision of the leading leg with ground, directed from the leading foot to the COM. For brevity, the equations presented here use dimensionless versions of quantities, with body mass M, gravitational acceleration g, and leg length L as base units. The push-off work is denoted ui (in units of mass-normalized work), and the push-off and collision sequence act to redirect the COM velocity to vi+ at the beginning of the next stance phase, directed forward and upward according to the next pendulum arc. Using impulse-momentum, the step-to-step transition is described by[3]vi+=vi−cos2α+2uisin2α where 2α is the inter-leg angle (Figure 7A). There is no work performed during the passive, inverted pendulum phases, and so the step-to-step transition is responsible for all energy inputs (ui) and energy losses (from collisions).
The dynamics of an inverted pendulum describe all of the other motion in the system, consisting of the falling of one inverted pendulum toward the step-to-step transition, and the rising of the next inverted pendulum toward mid-stance. These dynamics determine the respective velocities and timing of these respective instances. The velocities may be found through conservation of energy:[4]vi−=2(1−cosα)+vi2[5]vi+1=(vi+)2+2(cosα−1) The step time τi is defined as the time for the stance leg angle θ to move between successive mid-stance instants, and the corresponding velocities from vi to vi+1. It may be regarded as the sum of a time τi- from mid-stance to the step-to-step transition, and then the time τi+ from the step-to-step transition until next mid-stance. Using the linearized dynamics, the dimensionless time τi- of step i is[6]τi−=logα+vi2+α2vi.
The other time τi+ is[7]τi+=logvi++αvi+−α For comparison with experiment, we also defined an average (as opposed to mid-stance) speed for each step i as the step length divided by the step time between mid-stance instances,[8]Bodyspeedi=2Lsinατi−+τi+ The trajectory of this body speed is plotted for different walking bouts, for both model and experiment. The equations for body speed and step time are summarized as constraints f and g below.
We chose nominal parameters to correspond to typical human walking. A person with body mass M 70 kg and leg length L of 1 m may typically walk at 1.25 ms–1, with step length of 0.68 m and step time of 0.58 s, and corresponding fixed constant value α=0.35. The step length was also varied in parameter sensitivity studies. Using dynamic similarity, parameters and results are reported here either in SI units, or in normalized units with body mass M, gravitational acceleration g, and L as base units.
## Optimal control formulation
We applied optimal control to the model for short walk bouts of varying distance (Figure 7). In humans, both positive and negative work appear to cost positive metabolic energy with different proportionalities (Margaria, 1976). In the model, we assess a cost only for positive work, because the net work of a level walking bout is zero. Minimizing positive work thus also implicitly minimizes the negative work, as well as metabolic cost of any proportionality. The push-offs have a one-to-one relationship with the speeds, and so either push-offs or speeds can can describe the trajectory. For the model, the goal is to minimize an objective function Jmodel comprising the total positive work for the walking bout, plus the cost for the time duration:[9]Jmodel=(Positivework)+cT(Timeduration).
where the coefficient cT is the model’s valuation of time in terms of work, and equal to the mechanical work the model is willing to spend to save a unit of time. It is treated as proportional to the human’s valuation CT for metabolic energy per time.
This objective is applied as follows. The total distance D of a walking bout may be achieved by taking an appropriate number of steps N. The walking trajectory is described by a discrete sequence of speeds vi (step $i = 1$,2,…,N), starting and ending from standing at rest, given a standard step length. The corresponding control actions include the initiation impulse and the push-off impulses, for a total of N+1 actions ui ($i = 0$,1,2,…,N). Using these variables, the model’s objective is thus[10]Jmodel=∑$i = 0$Nui+cT∑$i = 1$Nτi for the optimization problem[11]minimizevi ($i = 1$,...,N)Jmodel(vi)subjectto[12]restconstraints:v0=0,vN=0[13]walkingdynamics:vi+1=f(vi,ui),τi+1=g(vi,ui).
where the model begins and ends at rest, and walking dynamics constrain how the speed and duration of the next step depend on the current step’s speed and push-off (functions f and g detailed above). Note there are actually N+1 controls, consisting of the initiation input u0 and the N step-to-step transition push-offs (u1,u2,…uN).
The time valuation cT is treated as an unknown but constant coefficient. Greater cT is expected to yield faster walking bouts, with experimental data used to determine an appropriate range of values. Within a fixed experimental context, we expect cT to be constant. We found values of cT ranging 0.006–0.06 Mg1.5L0.5 to yield speeds approximately similar to subjects.
Step lengths were examined with a nominal fixed step length of 0.68 m, and sensitivity analyses performed with fixed lengths of 0.59 m and 0.78 m, and varying lengths following the human preferred step length relationship s=v0.42(Grieve, 1968). We have previously proposed that step-to-step transitions account for most of the metabolic cost of human walking (say, up to $70\%$ Donelan et al., 2002), and that forced swing leg motions to modulate step length also contribute a non-negligible (Kuo, 2001), but lesser cost (say, up to $33\%$ Doke et al., 2005) that is neglected here. This is not to dismiss this and other costs of locomotion, but merely to hypothesize that step-to-step transitions will still dominate in transient walking bouts. The failure to include other costs, if sufficiently critical, should cause the model to make poor predictions.
We also considered two alternative hypotheses. One was that walking occurs almost entirely at the optimal steady speed. Termed the steady min-COT hypothesis, the goal is to walk at the min-COT speed v*, or close to it, as much as possible. This is accomplished by minimizing deviations from v* throughout the bout, with objective[14]Jsteady=∑$i = 1$N(vi−v∗)2 subject to the same constraints as the Energy-Time hypothesis. This objective is expected to cause the model to accelerate immediately from rest to v*, then remain at that steady speed, and then finally decelerate immediately back to rest. The second alternative was a steady acceleration hypothesis, which contrasts with min-COT’s immediate high acceleration. Here, the acceleration is made as gradual as possible, albeit at the expense of a higher peak speed needed to travel the same distance and duration. Both of these alternatives help illustrate how different speed trajectories requires differing amounts of mechanical work, to be compared against the work produced by the Energy-Time hypothesis.
Model predictions were produced using computational optimization. Optimal control was computed using the JuMP optimization package for the Julia language (Dunning et al., 2017), formulated as a discrete collocation problem, minimized by nonlinear programming (Ipopt). Walking bouts were conducted for N ranging 1–20 steps. The resulting trajectories were condensed into a scalable, self-similar family of speed trajectories.
## Experimental methods
We tested the model predictions by experimentally measuring the speed profiles of healthy adults walking a series of short distances, ranging about 2–20 steps. Subjects ($$n = 10$$, 6 male and 4 female, 24–38 yr) were instructed to walk at a comfortable speed in ten distance conditions, starting from standing at one numbered marker on the ground, and ending at another as requested by the experimenter. After each trial, there was a brief waiting interval of about 15–30 s, to reduce interference between successive trials and to avoid any incentive to rush through trials. The walking surface was a level sidewalk. The numbered markers were separated by distances of 1.1, 1.7, 2.2, 2.8, 3.3, 3.8, 5.1, 7, 9.1, and 12.7 m. Subjects were provided with a simple task upon reaching the target: They were provided a pointer stick and instructed to walk to and touch the pointer to the target marker. This was intended to provide a context for the task, reflecting the fact that humans often walk to a particular destination to accomplish a task. Each distance condition was conducted a total of four times in two pairs of out-and-back trials, with the distances in random order. There were therefore a total of 400 trials, from 10 subjects walking 10 distances, each four times.
Walking speeds were measured from foot-mounted inertial measurement units (IMUs). These were used to compute the spatiotemporal trajectory of each foot in 3D, which was then processed to yield forward walking speed for the body per step. Each IMU (Opal sensors, APDM Inc, Portland, Oregon) was placed on the top of each foot, taped to the outside of the shoe. The recorded data of linear acceleration and angular velocity data were integrated using a previously described algorithm (Rebula et al., 2013) to yield foot trajectories. Briefly, the algorithm detects footfalls as instances in time when the foot is momentarily at rest on the ground, as defined by thresholds for acceleration and angular velocity. The footfall instance was defined as the mid-point of the below-threshold interval, and used to correct the integrated foot velocity (from gravity-corrected inertial accelerations) to zero, thus reducing IMU integration drift. The footfalls were also used to segment data into discrete strides, from which speed and length of each stride was calculated. ( Subjects also wore another IMU on a waist belt, the data from which was used to demarcate the trials, but not for any further quantitative analysis.) There were a few other analysis adjustments required to produce forward walking data. The absolute position and compass heading of the IMUs were unknown, yielding independent foot trajectories with no relation to each other. However, the experimental conditions called for forward walking for a known distance, so we rotated each foot path to align them into a single forward direction. We also assumed that both feet travelled approximately the same distance for each walk, and translated and rescaled the start and end points to match each other, to yield a processed position-time graph of the two feet (see representative data in Figure 8). We also devised a definition for the starting and ending times for each trial based on IMU data. Humans initiate their gait by shifting their weight before moving the feet (Mann et al., 1979), so that the footfall threshold defined above may not detect the actual gait initiation. We therefore defined a rough approximation to gait initiation and termination, starting before and ending after threshold crossing, by an amount equal to half the average below-threshold time during walking. This adjustment may be incorrect compared to actual weight shift by several tens of milliseconds. The experiment is mainly concerned with speed profiles over time on the order of several seconds. The accuracy of the experiment can thus tolerate small errors in detecting gait initiation or termination.
**Figure 8.:** *Experimental estimation of walking speed from inertial measurement units (IMUs).Forward position vs. time are shown for both feet (black and red lines) for a single walking bout of eight steps. Forward position is determined from foot trajectories, computed by integrating gravity-corrected inertial data (top inset). Each foot moves one stride length at a time, and the crossing points of the two feet define mid-stance instances that separate individual steps (black dots). Body speed is defined as the step length divided by step duration (slope of dotted line) for each step. Walking speed trajectories are plotted as discrete body speed vs. time. There are three durations defined within a walking bout (right inset): rise time, cruise time, and fall time. Rise and fall times are to accelerate from rest to 90% of peak speed and the converse. Cruise time is the time spent between 90% and peak speed.*
Finally, the body’s walking speed and length of each step were calculated as follows (Figure 8). The trajectory of each foot’s strides were found to cross each other, approximating the time in mid-stance when one foot passes by the other. These points of intersection were used to define step length as the spatial distance between intersections, and step time as the temporal difference between intersections. These were used to determine the average speed for each step (‘body speed’), defined as the step length divided by step time ending at each intersection. This assumes that the body moves as much as the feet between mid-stance instances. These discrete, step-averaged data were used to produce trajectories of body speed for each bout (Figure 8, inset), without regard to continuous-time undulations in velocity for the body center of mass. For comparison with these data, similar discrete body speeds and times were computed from model predictions.
We used these data to test the model predictions. We examined how human speed profiles varied with bout distance, and exhibited more rounded peaks for shorter bouts and flatter ones for longer ones. We tested for self-similarity by scaling the profiles by speed and time and performing statistical tests regarding peak speeds and walking durations. We tested whether a saturating exponential describes the increase in peak speed with bout distance (R2; test $95\%$ confidence interval of parameters not including zero). Expecting a self-similar shape for the peak speed vs. distance relationship, we scaled the curves by peak speed and tested for a single exponential. We tested self-similarity in terms of a reduction of variability in peak speed (standard deviation across subjects) for each condition, comparing non-normalized to normalized peak speeds (rescaled to mean overall peak speed) with paired t-test. We examined the walking durations as a function of bout distance, and also tested self-similarity by significant reduction in standard deviations across subjects, comparing non-normalized to normalized data (rescaled to mean longest duration) with paired t-test. We also described walking durations in terms of rise and fall times (between $10\%$ and $90\%$ of peak speed).
Prior to the experiment, subjects provided informed consent as approved by the University of Calgary Conjoint Health Research Ethics Board (REB21-1497). Pre-established exclusion criteria included significant health or other conditions that preclude ability to walk on uneven terrain or moderate hiking trails; no prospective participants were excluded. Subjects were recruited from the community surrounding the University of Calgary; the city has a moderately affluent population of about 1.4 M, with a developed Western culture. The experiment was performed once.
## Effect of valuation of time on steady walking speed
We performed an additional analysis to consider how the hypothesized energetic value of time may affect human steady walking speeds (Figure 6). This requires a valuation of time in terms of human metabolic energy rather than the model’s mechanical work, and a consideration of longer walking bouts where steady walking dominates. To empirically quantify human cost as a function of speed, we fitted the model’s steady mechanical work rate to human net metabolic power reported by Elftman, 1966, with a resting power adjusted to agree with the optimal steady speed of 1.25 ms–1 reported by Ralston, 1958. The model was of the form[15]E˙(v)=a(v+bgL)n+d where a, b and d are empirical coefficients, and n is a model constraint. The exponent n is not critical, and values ranging 2–4 are sufficient to describe the increase. However, we used a value of $$n = 3$.42$ as predicted by the simple model for human-like walking (Kuo, 2002). For metabolic power in, the empirical coefficients are $a = 4.90$Wkg−1, $b = 1.16$ms−1, and $d = 1.56$Wkg−1 (R2=0.99). The y-intercept may be regarded as a resting rate, at 1.73 Wkg–1 (Figure 6A). The resulting cost is therefore proportional to the model’s mechanical work, while matching well with human metabolic power and optimal steady speed data. The curve may be expressed as cost of transport by dividing power by speed, E˙/v.
We then used our own walking data to estimate the human valuation of time. We used the peak walking speeds from the longest walking bout as indicator of steady speed. These were compared to the steady speed predicted by the metabolic cost curve with an added variable, the metabolic valuation of time CT. The Energy-Time curve was converted to cost of transport, and then minimized to yield optimal speed. This is equivalent to taking the limit of the Energy-Time objective as function of increasing distance, thus making the costs of starting and ending a walking bout small. The result predicts that steady speed will increase approximately with the cube root of CT (Figure 6B). This curve was thus used to estimate CT for experimentally observed range of steady speeds. It was also used to estimate the effective cost of transport, including the valuation of time, as a function of speed (Figure 6C). This cost of transport may further be regarded as the sum of separate costs for Walking and Time (Figure 6C), where Walking refers to the cost of transport due to push-off work alone, and Time refers to the cost of transport due to the CT term alone.
## Funding Information
This paper was supported by the following grants:
## Data availability
Data (Carlisle and Kuo, 2023, https://github.com/kuo-lab/short_walk_experiment) and code (https://github.com/kuo-lab/simplelocomotionmodel, copy archived at Kuo, 2023) for this study are in publicly-accessible archives.
## References
1. Afilalo J, Eisenberg MJ, Morin JF, Bergman H, Monette J, Noiseux N, Perrault LP, Alexander KP, Langlois Y, Dendukuri N, Chamoun P, Kasparian G, Robichaud S, Gharacholou SM, Boivin JF. **Gait speed as an incremental predictor of mortality and major morbidity in elderly patients undergoing cardiac surgery**. *Journal of the American College of Cardiology* (2010) **56** 1668-1676. DOI: 10.1016/j.jacc.2010.06.039
2. Atzler E, Herbst R. **Arbeitsphysiologische studien**. *Pflüg Arch Ges Physiol* (1927) **215** 291-328. DOI: 10.1007/BF01731318
3. Bauby CE, Kuo AD. **Active control of lateral balance in human walking**. *Journal of Biomechanics* (2000) **33** 1433-1440. DOI: 10.1016/s0021-9290(00)00101-9
4. Bornstein MH, Bornstein HG. **The pace of life**. *Nature* (1976) **259** 557-559. DOI: 10.1038/259557a0
5. Bornstein MH. **The pace of life: Revisited**. *International Journal of Psychology* (1979) **14** 83-90. DOI: 10.1080/00207597908246715
6. Browning RC, Kram R. **Energetic cost and preferred speed of walking in obese vs. normal weight women**. *Obesity Research* (2005) **13** 891-899. DOI: 10.1038/oby.2005.103
7. Browning RC, Baker EA, Herron JA, Kram R. **Effects of obesity and sex on the energetic cost and preferred speed of walking**. *Journal of Applied Physiology* (2006) **100** 390-398. DOI: 10.1152/japplphysiol.00767.2005
8. Carlisle RE, Kuo AD. *GitHub* (2023)
9. Darici O, Temeltas H, Kuo AD. **Anticipatory control of momentum for bipedal walking on uneven terrain**. *Scientific Reports* (2020) **10**. DOI: 10.1038/s41598-019-57156-6
10. Darici O, Kuo AD. **Humans optimally anticipate and compensate for an uneven step during walking**. *eLife* (2022a) **11**. DOI: 10.7554/eLife.65402
11. Darici O, Kuo AD. **Humans plan for the near future to walk economically on uneven terrain**. *arXiv* (2022b)
12. Dean JC, Kuo AD. **Elastic coupling of limb joints enables faster bipedal walking**. *Journal of the Royal Society, Interface* (2009) **6** 561-573. DOI: 10.1098/rsif.2008.0415
13. Doke J, Donelan JM, Kuo AD. **Mechanics and energetics of swinging the human leg**. *The Journal of Experimental Biology* (2005) **208** 439-445. DOI: 10.1242/jeb.01408
14. Donelan JM, Kram R, Kuo AD. **Mechanical and metabolic determinants of the preferred step width in human walking**. *Proceedings. Biological Sciences* (2001) **268** 1985-1992. DOI: 10.1098/rspb.2001.1761
15. Donelan JM, Kram R, Kuo AD. **Mechanical work for step-to-step transitions is a major determinant of the metabolic cost of human walking**. *The Journal of Experimental Biology* (2002) **205** 3717-3727. DOI: 10.1242/jeb.205.23.3717
16. Donelan JMJM, Shipman DWDW, Kram R, Kuo ADAD. **Mechanical and metabolic requirements for active lateral stabilization in human walking**. *Journal of Biomechanics* (2004) **37** 827-835. DOI: 10.1016/j.jbiomech.2003.06.002
17. Dunning I, Huchette J, Lubin M. **JuMP: A modeling language for mathematical optimization**. *SIAM Review* (2017) **59** 295-320. DOI: 10.1137/15M1020575
18. Elftman H. **Biomechanics of muscle**. *The Journal of Bone & Joint Surgery* (1966) **48** 363-377. DOI: 10.2106/00004623-196648020-00017
19. Entin PL, Gest C, Trancik S, Richard Coast J. **Fuel oxidation in relation to walking speed: Influence of gradient and external load**. *European Journal of Applied Physiology* (2010) **110** 515-521. DOI: 10.1007/s00421-010-1523-6
20. Green L, Myerson J. **Exponential versus hyperbolic discounting of delayed outcomes: Risk and waiting time**. *American Zoologist* (1996) **36** 496-505. DOI: 10.1093/icb/36.4.496
21. Grieve DW. **Gait patterns and the speed of walking**. *Bio-Med Engng* (1968) **3** 119-122. DOI: 10.1007/978-1-349-04835-9_32
22. Hoyt DF, Taylor CR. **Gait and the energetics of locomotion in horses**. *Nature* (1981) **292** 239-240. DOI: 10.1038/292239a0
23. Jetté M, Sidney K, Blümchen G. **Metabolic equivalents (mets) in exercise testing, exercise prescription, and evaluation of functional capacity**. *Clinical Cardiology* (1990) **13** 555-565. DOI: 10.1002/clc.4960130809
24. Kuo AD. **A simple model of bipedal walking predicts the preferred speed-step length relationship**. *Journal of Biomechanical Engineering* (2001) **123** 264-269. DOI: 10.1115/1.1372322
25. Kuo AD. **Energetics of actively powered locomotion using the simplest walking model**. *Journal of Biomechanical Engineering* (2002) **124** 113-120. DOI: 10.1115/1.1427703
26. Kuo AD, Donelan JM, Ruina A. **Energetic consequences of walking like an inverted pendulum: Step-to-step transitions**. *Exercise and Sport Sciences Reviews* (2005) **33** 88-97. DOI: 10.1097/00003677-200504000-00006
27. Kuo AD. *Software Heritage* (2023)
28. Labaune O, Deroche T, Teulier C, Berret B. **Vigor of reaching, walking, and gazing movements: On the consistency of interindividual differences**. *Journal of Neurophysiology* (2020) **123** 234-242. DOI: 10.1152/jn.00344.2019
29. Langman VA, Rowe MF, Roberts TJ, Langman NV, Taylor CR. **Minimum cost of transport in Asian elephants: Do we really need a bigger elephant?**. *The Journal of Experimental Biology* (2012) **215** 1509-1514. DOI: 10.1242/jeb.063032
30. Levine RV, Bartlett K. **Pace of life, punctuality, and coronary heart disease in six countries**. *Journal of Cross-Cultural Psychology* (1984) **15** 233-255. DOI: 10.1177/0022002184015002009
31. Li H, Cao Y. **Moving at the speed of life: How life PACE influences temporal reasoning**. *Metaphor and Symbol* (2019) **34** 158-166. DOI: 10.1080/10926488.2019.1649839
32. Long LL, Srinivasan M. **Walking, running, and resting under time, distance, and average speed constraints: Optimality of walk-run-rest mixtures**. *Journal of the Royal Society, Interface* (2013) **10**. DOI: 10.1098/rsif.2012.0980
33. Malatesta D, Simar D, Dauvilliers Y, Candau R, Borrani F, Prefaut C, Caillaud C. **Energy cost of walking and gait instability in healthy 65- and 80-yr-olds**. *Journal of Applied Physiology* (2003) **95** 2248-2256. DOI: 10.1152/japplphysiol.01106.2002
34. Mann RA, Hagy JL, White V, Liddell D. **The initiation of gait**. *The Journal of Bone and Joint Surgery. American Volume* (1979) **61** 232-239. DOI: 10.2106/00004623-197961020-00011
35. Margaria R. *Biomechanics and Energetics of Muscular Exercise* (1976)
36. Martin PE, Rothstein DE, Larish DD. **Effects of age and physical activity status on the speed-aerobic demand relationship of walking**. *Journal of Applied Physiology* (1992) **73** 200-206. DOI: 10.1152/jappl.1992.73.1.200
37. Middleton A, Fritz SL, Lusardi M. **Walking speed: The functional vital sign**. *Journal of Aging and Physical Activity* (2015) **23** 314-322. DOI: 10.1123/japa.2013-0236
38. Orendurff MS, Schoen JA, Bernatz GC, Segal AD, Klute GK. **How humans walk: Bout duration, steps per bout, and rest duration**. *Journal of Rehabilitation Research and Development* (2008) **45** 1077-1089. DOI: 10.1682/jrrd.2007.11.0197
39. Ralston HJ. **Energy-speed relation and optimal speed during level walking**. *Internationale Zeitschrift Fur Angewandte Physiologie, Einschliesslich Arbeitsphysiologie* (1958) **17** 277-283. DOI: 10.1007/BF00698754
40. Rebula JR, Ojeda LV, Adamczyk PG, Kuo AD. **Measurement of foot placement and its variability with inertial sensors**. *Gait & Posture* (2013) **38** 974-980. DOI: 10.1016/j.gaitpost.2013.05.012
41. Rebula JR, Ojeda LV, Adamczyk PG, Kuo AD. **The stabilizing properties of foot yaw in human walking**. *Journal of Biomechanics* (2017) **53** 1-8. DOI: 10.1016/j.jbiomech.2016.11.059
42. Rose J, Morgan DW, Gamble JG. *Energetics of walking, Ch. 5* (2006)
43. Seethapathi N, Srinivasan M. **The metabolic cost of changing walking speeds is significant, implies lower optimal speeds for shorter distances, and increases daily energy estimates**. *Biology Letters* (2015) **11**. DOI: 10.1098/rsbl.2015.0486
44. Shadmehr R, Orban de Xivry JJ, Xu-Wilson M, Shih T-Y. **Temporal discounting of reward and the cost of time in motor control**. *The Journal of Neuroscience* (2010) **30** 10507-10516. DOI: 10.1523/JNEUROSCI.1343-10.2010
45. Srinivasan M, Ruina A. **Computer optimization of a minimal biped model discovers walking and running**. *Nature* (2006) **439** 72-75. DOI: 10.1038/nature04113
46. Srinivasan M. **Optimal speeds for walking and running, and walking on a moving walkway**. *Chaos* (2009) **19**. DOI: 10.1063/1.3141428
47. Studenski S, Perera S, Patel K, Rosano C, Faulkner K, Inzitari M, Brach J, Chandler J, Cawthon P, Connor EB, Nevitt M, Visser M, Kritchevsky S, Badinelli S, Harris T, Newman AB, Cauley J, Ferrucci L, Guralnik J. **Gait speed and survival in older adults**. *JAMA* (2011) **305** 50-58. DOI: 10.1001/jama.2010.1923
48. Summerside EM, Kram R, Ahmed AA. **Contributions of metabolic and temporal costs to human gait selection**. *Journal of the Royal Society, Interface* (2018) **15**. DOI: 10.1098/rsif.2018.0197
49. Sutton RS, Barto AG. *Reinforcement Learning, Second Edition: An Introduction* (2018)
50. Taylor SV, Faisal AA. **Does Internal Metabolic State Determine Our Motor Coordination Strategy?**. *bioRxiv* (2018). DOI: 10.1101/312454
51. Todorov E. **Optimality principles in sensorimotor control**. *Nature Neuroscience* (2004) **7** 907-915. DOI: 10.1038/nn1309
52. Watson RR, Rubenson J, Coder L, Hoyt DF, Propert MWG, Marsh RL. **Gait-specific energetics contributes to economical walking and running in emus and ostriches**. *Proceedings. Biological Sciences* (2011) **278** 2040-2046. DOI: 10.1098/rspb.2010.2022
53. Willis WT, Ganley KJ, Herman RM. **Fuel oxidation during human walking**. *Metabolism* (2005) **54** 793-799. DOI: 10.1016/j.metabol.2005.01.024
54. Wong JD, Cluff T, Kuo AD. **The energetic basis for smooth human arm movements**. *eLife* (2021) **10**. DOI: 10.7554/eLife.68013
55. Zelik KE, Huang TWP, Adamczyk PG, Kuo AD. **The role of series ankle elasticity in bipedal walking**. *Journal of Theoretical Biology* (2014) **346** 75-85. DOI: 10.1016/j.jtbi.2013.12.014
|
---
title: Clinical Profile of COVID-19 Patients Admitted at a Private Hospital During
Three Surges in Mandalay, Myanmar
journal: Cureus
year: 2023
pmcid: PMC10030156
doi: 10.7759/cureus.35167
license: CC BY 3.0
---
# Clinical Profile of COVID-19 Patients Admitted at a Private Hospital During Three Surges in Mandalay, Myanmar
## Abstract
Introduction During the coronavirus disease 2019 (COVID-19) pandemic, private hospitals in Mandalay started to manage COVID-19 infections according to national treatment guidelines since February 2021. Variations of clinical characteristics and their outcomes in different surges could be evaluated in the private hospital. This study aimed to assess the clinical profile and outcomes of COVID-19 patients admitted at a private hospital during three surges in Mandalay.
Methods This study is a retrospective record review of the case series of COVID-19 patients admitted at City Hospital, Mandalay. The study was conducted from January to December 2022. All of the hospital records of COVID-19 patients admitted during the second wave from February 2020 to 26 May 2021, the third wave from 27 May 2021 to 27 January 2022, and the fourth wave from 28 January to April 2022 were included in the study.
Results A total of 1606 admitted cases were included in the study. The mean with standard deviation (SD) of age was 55.7±18.5, and males were 778 ($48.4\%$). The mean duration of hospital stay in days was 10.8±5.94, 10.6±6.11, and 7.3±2.88 in second, third, and fourth waves, respectively. The mean duration of hospital stay was shortened in the fourth wave. Comorbid conditions with hypertension and/or diabetes diseases were mostly observed in three waves of COVID-19 infection. Fever was the most presented symptom in three waves. Cough, sore throat, and rhinorrhea were observed more in the fourth wave compared with previous waves. Complication with pneumonia ($71.3\%$), liver dysfunction ($21.0\%$), acute respiratory distress syndrome ($10.0\%$), thrombocytopenia ($6.2\%$), acute kidney injury ($5.5\%$), bleeding ($3.9\%$), and pulmonary embolism ($2.9\%$) were investigated. Antiviral treatment such as remdesivir or molnupiravir was used more in the patients of third and fourth waves than those of the second wave. Oxygen therapy ($59.9\%$), prone position ($35.5\%$), non-invasive ventilation ($9.5\%$), invasive ventilation ($0.5\%$), inotropes ($4.6\%$), and renal replacement therapy ($1.1\%$) were recorded in serious cases. Only $7.9\%$ and $9.4\%$ died in the hospital in second and third waves. No mortality was observed in the fourth wave.
Conclusions The study recommended that COVID-19 patients with comorbid conditions of hypertension or diabetes and ages 65 and older should be taken with intensive care support at the hospital. This study also concluded that a private hospital in Mandalay could tackle with COVID-19 severe cases in line with national treatment guidelines since the second wave and could provide better management in the fourth wave. Antiviral treatment should be used in severe COVID-19 cases for further emergency management. In conclusion, private hospital involvement in the COVID-19 pandemic is supportive of the healthcare provision in Myanmar in an emergency situation.
## Introduction
The coronavirus disease 2019 (COVID-19) pandemic transmission started at Wuhan, China, in late 2019. It is an infectious disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). COVID-19 patients may present with many symptoms such as fever, cough, anorexia, fatigue, shortness of breath, and myalgia. Moreover, flu-like symptoms are also presented. Clinical presentations may vary in accordance with regional variation, pandemic waves, and pathogen alteration. COVID-19 pandemic waves occur due to the rising of infection transmission with new variant appearances in countries [1]. Regarding the diagnosis, clinical presentation is highly supportive. Nucleic acid amplifying test of respiratory specimens such as reverse transcription-polymerase chain reaction (RT-PCR) for COVID-19 infection is used as laboratory confirmatory diagnosis. However, rapid antigen testing is useful if RT-PCR is not available in some conditions with limited resources [2]. In the World Health Organization’s (WHO) living guidance paper, other differential diagnoses such as influenza, dengue, malaria, typhoid, and respiratory tract infections are not negligible in hospital management [3].
Critical COVID-19 cases such as acute respiratory distress syndrome (ARDS) and septic shock are cared aligned with standard management guideline [4]. Treatment with antimicrobial therapy for co-infections and secondary infections coinciding with COVID-19 has to be started as soon as possible according to laboratory investigations, clinical judgment, local epidemiology, and patient host factor [5]. COVID-19 patients are manifested with neurological, neuropsychiatric, and mental problems. Those should be treated appropriately at the hospital [6]. If the disease becomes very severe, intubation with mechanical ventilation is a lifesaving management in advanced hospitals. Antiviral as a specific treatment for severe disease is also practicable. Remdesivir, a ribonucleic acid (RNA)-dependent polymerase inhibitor, is the most promising broad-spectrum antiviral agent that resulted from clinical studies [7,8].
Severe COVID-19 patients may be cured and die. The mortality of severe and critically ill patients may be different substantially in different case series all over the world. The World Health Organization (WHO) recommended collecting clinical data of admitted COVID-19 patients from countries all over the world and contributing clinical characterization data at the WHO Global COVID-19 Clinical Data Platform [1].
In Myanmar, the first case of COVID-19 infection was detected on 23 March 2020. Then, an outbreak of infection started, and new cases reached up to 124,630 in 2020 and 530,834 in 2021. The Ministry of Health (MoH) had response actions on the COVID-19 pandemic through the Coronavirus Disease 2019 (COVID-19) Containment and Emergency Response Committee. The MoH laid down the national management guidelines in line with local contexts. The Department of Public Health, Department of Medical Services, and Department of Medical Research conduct the preparedness and response strategic actions on COVID-19 transmission. In 2020, clinical treatment for hospitalized patients was focused by public hospitals such as secondary-level (township/district) hospitals and tertiary-level (state/regional/general) hospitals. However, because of increasing transmission and human resource shortage that resulted from an unstable country situation, the support from private sector involvement was needed to manage the surge of infection. Therefore, in early 2021, some specialized private hospitals were allowed to treat COVID-19 patients in line with national treatment guidelines [9].
During the COVID-19 pandemic in Myanmar, there were four surges called as waves: the first wave was from 23 March to 15 August 2020, the second wave from 16 August 2020 to 26 May 2021, the third wave from 27 May 2021 to 27 January 2022, and the fourth wave from 28 January 2022 up to the present. These waves were determined by the Central Epidemiological Unit of the Department of Public Health and approved by the Ministry of Health, Myanmar. Therefore, variations of clinical characteristics in different waves have been practiced in the private hospitals. These clinical characteristics and their outcomes among the patients are evaluated by waves. Therefore, this study aimed to assess the clinical profile and outcomes of COVID-19 patients admitted at a private hospital in Mandalay.
## Materials and methods
This study is a retrospective record review of the case series of COVID-19 patients admitted at a private, city hospital at Mandalay. It is 300-bedded specialist hospital equipped with high-technology medical care units, intensive care units, high-standard medical laboratories, and operation theaters. This study was conducted from January to December 2022. All of the hospital records of COVID-19 patients admitted at City Hospital from February 2021 to April 2022 were included in this study. During the COVID-19 pandemic in Myanmar, there were four waves: the first wave was from 23 March to 15 August 2020, the second wave from 16 August 2020 to 26 May 2021, the third wave from 27 May 2021 to 27 January 2022, and the fourth wave from 28 January 2022 up to the present. According to wave distribution, the studied cases were distributed in three waves: second, third, and fourth waves. The patients who presented to the hospital with no symptoms but needed investigation, the patients with clinical signs of pneumonia, and the suspected patients with comorbid conditions and severely ill cases were confirmed with either rapid diagnostic testing or PCR testing.
All of the COVID-19-confirmed cases were admitted to the hospital except the mild cases without risk or comorbid conditions. Those mild cases were transferred to the regional-level clinical management committee. Then, the admitted cases were categorized and treated according to national management guidelines for COVID-19 of Myanmar. The symptomatic patients without evidence of viral pneumonia or hypoxia were defined as mild disease; adolescents or adults with clinical signs of pneumonia (fever, cough, dyspnea, and fast breathing) as moderate disease (pneumonia); adolescents or adults with clinical signs of pneumonia (fever, cough, dyspnea, and fast breathing) plus either respiratory rate of >30 breaths/minute, severe respiratory distress, or saturation of peripheral oxygen (SpO2) of <$93\%$ on room air as severe disease (severe pneumonia); and complications such as ARDS, organ failure, sepsis, and shock as critically ill. This study included all COVID-19-confirmed cases. Hospital records of patients who had a history of COVID-19 infection and were admitted for the second time for another disease were not enrolled in this study. The patients who were vaccinated and re-admitted for COVID-19 infections were included in the study. Both the hospital patient chart and the electronic medical record (EMR) system of the hospital were used as a source of data document. The clinical data were collected in case record form (CRF) modified from the standard case record form (CRF) developed by the WHO Guideline Development Group for the WHO Global COVID-19 Clinical Data Platform [1].
Data analysis *The data* on clinical characteristics, laboratory investigations, treatment, and final outcome status were reviewed using the electronic medical record (EMR). Individual clinical data were abstracted after discharge from hospital. The data on the clinical profile of COVID-19 patients were collected with CRF developed using the WHO Global COVID-19 Clinical Data Platform. Data entry and analysis were done using the Statistical Package for Social Sciences (SPSS) software (version 18.0, PASW Statistics, Chicago, IL). Descriptive analysis was carried out presenting with mean, medium, standard deviation (SD), and interquartile range (IQR) for continuous numerical data and frequency and percentage for categorical data. The variables among the waves were analyzed using chi-square (χ2) test for comparing categorical data, analysis of variance (ANOVA) for comparing means, and nonparametric test (median test) for comparing medians. Statistical significance was set at p-value of <0.05.
Ethical approval The study was conducted after obtaining the ethical approval from Research Ethics Committee of University of Medicine, Mandalay (approval number: 1535/UMM/Research, dated 14 June 2022). The waiver of informed consent taking in conduct of this retrospective record review was approved by the Research Ethics Committee. This research was registered at Myanmar Health Research Registry (http://www.mhrr-mohs.com) (PLRID-00192_V14).
## Results
Between 8 February 2021 and 14 April 2022, a total of 1606 cases were admitted at City Hospital, Mandalay, for COVID-19 infections. Among them, 67 ($4.2\%$) were re-admitted as post-COVID-19 complications, and 306 ($19.1\%$) were vaccinated. The patients were distributed in the second, third, and fourth waves. A total of 139 cases in the second wave, 1330 cases in the third wave, and 137 cases in the fourth wave were reviewed. The following Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) diagram shows the distribution of cases by waves and their outcomes (Figure 1).
**Figure 1:** *STROBE diagram showing the distribution of cases and their outcomes by waves.STROBE: Strengthening the Reporting of Observational Studies in Epidemiology*
In this study, the mean (±SD) age of a total of 1606 cases was 55.7±18.5. The mean age by wave distribution showed 58.2±16.8, 54.35±18.34, and 65.6±18.1 in the second, third, and fourth waves, respectively. More than half of them were within 51 and 75 years old age group. Males were 778 ($48.4\%$), and females were 828 ($51.6\%$). Among them, healthcare workers were 128 ($8.0\%$), and laboratory workers were six ($0.4\%$). Pregnant mothers were 24 ($1.5\%$) (Table 1).
**Table 1**
| Variables | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total (n=1606) | P-value |
| --- | --- | --- | --- | --- | --- |
| Age (mean±SD) | 58.2±16.8 | 54.35±18.34 | 65.6±18.1 | 55.7±18.5 | <0.001 |
| Age group | Age group | Age group | Age group | Age group | Age group |
| <25 years | 7 (5.0) | 81 (6.1) | 4 (2.9) | 92 (5.7) | <0.001 |
| 25-50 years | 29 (20.9) | 420 (31.6) | 18 (13.1) | 467 (29.1) | <0.001 |
| 51-75 years | 89 (64.0) | 667 (50.2) | 71 (51.8) | 827 (51.5) | <0.001 |
| >75 years | 14 (10.1) | 162 (12.2) | 44 (32.1) | 220 (13.7) | <0.001 |
| Gender | Gender | Gender | Gender | Gender | Gender |
| Male | 79 (56.8) | 629 (47.3) | 70 (51.1) | 778 (48.4) | 0.082 |
| Female | 60 (43.2) | 701 (52.7) | 67 (48.9) | 828 (51.6) | 0.082 |
| Healthcare worker | 12 (8.6) | 114 (8.6) | 2 (1.5) | 128 (8.0) | 0.013 |
| Laboratory worker | 0 (0) | 5 (0.4) | 1 (0.7) | 6 (0.4) | 0.495 |
| Pregnant | 1 (0.7) | 21 (1.6) | 2 (1.5) | 24 (1.5) | 0.682 |
Table 2 shows the baseline characteristics and comorbid conditions of the COVID-19 patients. Among the hospital patients, comorbid conditions with hypertension or diabetes diseases were mostly observed in three waves of COVID-19 infection in City Hospital, Mandalay, with $48.2\%$, $39.8\%$, and $54.0\%$ for hypertension and $41.0\%$, $34.3\%$, and $39.4\%$ for diabetes in the second, third, and fourth waves, respectively. The vaccination program in Myanmar had been initiated since January 2021, so almost all of the patients had not received vaccination in the second wave. Cumulatively, about $80.9\%$ of the patients had not received any dose of vaccination. Among the 1606 cases, there were 67 ($4.2\%$) cases that were re-admitted as post-COVID-19 complications.
**Table 2**
| Variables | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total (n=1606) | P-value |
| --- | --- | --- | --- | --- | --- |
| Temperature (degree Fahrenheit) (mean±SD) | 99.0±0.9 | 98.7±0.7 | 98.9±1.1 | 98.8±0.7 | <0.001 |
| Heart rate (beat per minute) (mean±SD) | 92.9±16.3 | 93.6±17.1 | 93.5±21.4 | 93.6±17.4 | 0.900 |
| Systolic blood pressure (mmHg) (mean±SD) | 131.8±18.9 | 131.3±19.9 | 138.2±26.8 | 131.7±20.6 | 0.001 |
| Diastolic blood pressure (mmHg) (mean±SD) | 81.2±10.8 | 82.7±12.0 | 84.2±15.2 | 82.7±12.2 | 0.124 |
| Respiratory rate (mean±SD) | 23.4±4.9 | 23.3±10.1 | 22.8±7.5 | 23.3±9.6 | 0.857 |
| SpO2 (%) (mean±SD) | 92.4±9.4 | 93.7±7.7 | 95.9±5.2 | 93.80±7.7 | 0.001 |
| O2 supply on admission | O2 supply on admission | O2 supply on admission | O2 supply on admission | O2 supply on admission | O2 supply on admission |
| Room air | 97 (69.8) | 986 (74.1) | 108 (78.8) | 1191 (74.2) | 0.086 |
| Oxygen therapy | 42 (30.2) | 344 (25.9) | 29 (21.2) | 415 (25.8) | 0.086 |
| Comorbidity | Comorbidity | Comorbidity | Comorbidity | Comorbidity | Comorbidity |
| Chronic cardiac disorder | 17 (12.2) | 168 (12.6) | 40 (29.2) | 225 (14.0) | <0.001 |
| Hypertension | 67 (48.2) | 529 (39.8) | 74 (54.0) | 670 (41.7) | <0.002 |
| Chronic pulmonary disease | 4 (2.9) | 33 (2.5) | 5 (3.6) | 42 (2.6) | 0.692 |
| Asthma | 6 (4.3) | 28 (2.1) | 7 (5.1) | 41 (2.6) | 0.041 |
| Chronic kidney disease | 8 (5.8) | 78 (5.9) | 19 (13.9) | 105 (6.5) | 0.007 |
| Chronic liver disease | 3 (2.2) | 34 (2.6) | 9 (6.6) | 46 (2.9) | 0.024 |
| Chronic neurological disorder | 5 (3.6) | 43 (3.2) | 20 (14.6) | 68 (4.2) | <0.001 |
| HIV on ART | 0 (0) | 7 (0.5) | 0 (0) | 7 (0.4) | 0.409 |
| HIV not on ART | 0 (0) | 2 (0.2) | 1 (0.7) | 3 (0.2) | 0.409 |
| Diabetes | 57 (41.0) | 456 (34.3) | 54 (39.4) | 567 (35.3) | 0.166 |
| Current smoking | 10 (7.2) | 76 (5.7) | 8 (5.8) | 94 (5.9) | 0.629 |
| Tuberculosis | 0 (0) | 2 (0.2) | 0 (0) | 2 (0.1) | 0.812 |
| Malignancy | 2 (1.4) | 17 (1.3) | 3 (2.2) | 22 (1.4) | 0.681 |
| Anemia | 11 (7.9) | 95 (7.1) | 10 (7.3) | 116 (7.2) | 0.945 |
| Hypothyroid | 4 (2.9) | 28 (2.1) | 3 (2.2) | 35 (2.2) | 0.838 |
| Obesity | 1 (0.7) | 36 (2.7) | 4 (2.9) | 41 (2.6) | 0.354 |
| Prediabetes | 1 (0.7) | 42 (3.2) | 6 (4.4) | 49 (3.1) | 0.077 |
| Arthritis | 6 (4.3) | 42 (3.2) | 5 (3.6) | 53 (3.3) | 0.746 |
| BPH | 2 (1.4) | 17 (1.3) | 11 (8.0) | 30 (1.9) | <0.001 |
| Chronic medication history | Chronic medication history | Chronic medication history | Chronic medication history | Chronic medication history | Chronic medication history |
| ACEI | 5 (3.6) | 51 (3.8) | 12 (8.8) | 68 (4.2) | 0.023 |
| ARB | 19 (13.7) | 149 (11.2) | 18 (13.1) | 186 (11.6) | 0.576 |
| NSAID | 10 (7.2) | 82 (6.2) | 20 (14.6) | 112 (7.0) | 0.017 |
| Vaccination status | Vaccination status | Vaccination status | Vaccination status | Vaccination status | Vaccination status |
| No vaccination | 137 (98.6) | 1123 (84.4) | 40 (29.2) | 1300 (80.9) | <0.001 |
| First dose completed | 2 (1.4) | 207 (15.6) | 97 (70.8) | 306 (19.1) | <0.001 |
| Second dose completed | 0 (0) | 126 (9.5) | 97 (70.8) | 223 (13.9) | <0.001 |
| Third dose completed | 0 (0) | 6 (0.5) | 33 (24.1) | 39 (2.4) | <0.001 |
| Multiple admission | Multiple admission | Multiple admission | Multiple admission | Multiple admission | Multiple admission |
| Re-admitted cases | 9 (6.5) | 52 (3.9) | 6 (4.4) | 67 (4.2) | 0.352 |
Table 3 reveals the clinical signs and symptoms of COVID-19 patients in three waves. Fever, cough, and headache were the most presenting symptoms in three waves. Clinical presentation with sore throat and rhinorrhea was observed more among the patients in the fourth wave of COVID-19 infection compared with previous waves.
**Table 3**
| Clinical presentations | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total (n=1606) | P-value (χ2 test) |
| --- | --- | --- | --- | --- | --- |
| Fever | 111 (79.9) | 1076 (80.9) | 95 (69.3) | 1282 (79.8) | 0.006 |
| Cough | 98 (70.5) | 1008 (75.8) | 105 (76.6) | 1211 (75.4) | 0.364 |
| Shortness of breath | 85 (61.2) | 683 (51.4) | 53 (38.7) | 821 (51.1) | <0.001 |
| Cough with sputum | 56 (40.3) | 547 (41.1) | 63 (46.0) | 666 (41.5) | <0.523 |
| Myalgia | 23 (16.5) | 456 (34.3) | 41 (29.9) | 520 (32.4) | <0.001 |
| Malaise | 36 (25.9) | 336 (25.3) | 37 (27.0) | 409 (25.5) | 0.899 |
| Loss of appetite | 18 (12.9) | 318 (23.9) | 47 (34.3) | 383 (23.8) | <0.001 |
| Diarrhea | 20 (14.4) | 333 (25.0) | 13 (61.2) | 366 (22.8) | <0.001 |
| Anosmia | 10 (7.2) | 225 (16.9) | 3 (2.2) | 238 (14.8) | <0.001 |
| Sore throat | 8 (5.8) | 186 (14.0) | 32 (23.4) | 226 (14.1) | <0.001 |
| Headache | 7 (5.0) | 169 (12.7) | 13 (9.5) | 189 (11.8) | 0.019 |
| Rhinorrhea | 6 (4.3) | 105 (7.9) | 23 (16.8) | 134 (8.3) | <0.001 |
| Nausea/vomiting | 10 (7.2) | 62 (4.7) | 17 (12.4) | 89 (5.5) | 0.001 |
| Arthralgia | 5 (3.6) | 74 (5.6) | 4 (2.9) | 83 (5.2) | 0.281 |
| Inability to walk | 8 (5.8) | 46 (3.5) | 13 (9.5) | 67 (4.2) | 0.002 |
| Ageusia | 0 (0) | 59 (4.4) | 0 (0) | 59 (3.7) | 0.002 |
| Abdominal pain | 5 (3.6) | 35 (2.6) | 8 (5.8) | 48 (3.0) | 0.101 |
| Chest pain | 8 (5.8) | 31 (2.3) | 6 (4.4) | 45 (2.8) | 0.034 |
| Cough with hemoptysis | 0 (0) | 41 (3.1) | 2 (1.5) | 43 (2.7) | 0.066 |
| Altered consciousness | 3 (2.2) | 23 (1.7) | 8 (5.8) | 34 (2.1) | 0.006 |
| Hemorrhage | 2 (1.4) | 16 (1.2) | 5 (3.6) | 23 (1.4) | 0.072 |
| Insomnia | 2 (1.4) | 15 (1.1) | 4 (2.9) | 21 (1.3) | 0.211 |
| Wheezing | 6 (4.3) | 9 (0.7) | 5 (3.6) | 20 (1.2) | <0.001 |
| Nasal stuffiness | 2 (1.4) | 8 (0.6) | 4 (2.9) | 14 (0.9) | 0.016 |
| Skin rash | 0 (0) | 6 (0.5) | 1 (0.7) | 7 (0.4) | 0.641 |
| Conjunctivitis | 0 (0) | 3 (0.2) | 2 (1.5) | 5 (0.3) | 0.037 |
| Seizures | 0 (0) | 4 (0.3) | 0 (0) | 4 (0.2) | 0.660 |
| Syncope | 0 (0) | 3 (0.2) | 0 (0) | 3 (0.2) | 0.732 |
| Skin ulcers | 1 (0.7) | 1 (0.1) | 0 (0) | 2 (0.1) | 0.112 |
| Lower chest wall indrawing | 2 (1.4) | 0 (0) | 0 (0) | 2 (0.1) | <0.001 |
The distribution of the complications of COVID-19 patients in three waves are shown in Figure 2. Among all patients, as a cumulative incidence during three waves, complication with pneumonia ($71.3\%$), liver dysfunction ($21.0\%$), ARDS ($10.0\%$), thrombocytopenia ($6.2\%$), acute kidney injury ($5.5\%$), bleeding ($3.9\%$), and pulmonary embolism ($2.9\%$) were investigated. Respiratory complications such as ARDS and pneumonia were observed more in the second and third waves compared to the fourth wave. Liver dysfunction was also recorded less in the fourth wave compared to previous waves.
**Figure 2:** *Complications of the COVID-19 patients in percentage by waves.ARDS, acute respiratory distress syndrome; DKA, diabetic ketoacidosis; DVT, deep vein thrombosis; DIC, disseminated intravascular coagulation; COVID-19, coronavirus disease 2019*
Table 4 shows the laboratory findings on admission, mid-time, and discharge in three waves of the COVID-19 patients. The reduction of mean lymphocyte count was observed among the patients during three waves especially in the second wave. Similarly, the increased median value of procalcitonin, hyper-sensitive C-reactive protein (hsCRP), and D-dimer were also recorded throughout three waves especially in second and third waves. High levels of the mean of ferritin were also reviewed among the patients in three waves especially in third and fourth.
**Table 4**
| Variables | Variables.1 | COVID-19 waves (mean±SD), median (IQR) | COVID-19 waves (mean±SD), median (IQR).1 | COVID-19 waves (mean±SD), median (IQR).2 | Total | Unnamed: 6 |
| --- | --- | --- | --- | --- | --- | --- |
| Variables | Variables | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total | P-value |
| Hemoglobin, g/dL (mean±SD) | Adm | 12.85±1.94 (n=139) | 12.76±2.03 (n=1251) | 12.39±2.47 (n=135) | 12.73±2.07 (n=1525) | 0.122 |
| Hemoglobin, g/dL (mean±SD) | Mid | 12.76±2.32 (n=71) | 12.45±1.99 (n=877) | 12.07±1.98 (n=955) | 12.44±2.02 (n=1043) | 0.086 |
| Hemoglobin, g/dL (mean±SD) | DC | 12.67±1.75 (n=106) | 12.70±1.91 (n=1017) | 12.33±2.33 (n=119) | 12.67±1.94 (n=1242) | 0.141 |
| WBC count, 103/µL (mean±SD) | Adm | 8.69±4.44 (n=139) | 8.49±5.54 (n=1251) | 9.28±7.05 (n=135) | 8.58±5.60 (n=1525) | 0.294 |
| WBC count, 103/µL (mean±SD) | Mid | 12.78±5.53 (n=71) | 11.86±8.00 (n=876) | 8.83±4.36 (n=95) | 11.64±7.65 (n=1042) | <0.001 |
| WBC count, 103/µL (mean±SD) | DC | 11.83±6.36 (n=106) | 12.10±7.15 (n=1015) | 9.42±5.11 (n=119) | 11.82±6.95 (n=1240) | <0.001 |
| Lymphocyte count, 103/µL (mean±SD) | Adm | 1.32±0.75 (n=138) | 1.27±0.74 (n=1243) | 1.29±0.77 (n=135) | 1.27±0.74 (n=1516) | 0.765 |
| Lymphocyte count, 103/µL (mean±SD) | Mid | 1.29±1.22 (n=71) | 1.42±0.95 (n=877) | 1.68±0.88 (n=95) | 1.43±0.97 (n=1043) | 0.018 |
| Lymphocyte count, 103/µL (mean±SD) | DC | 1.82±1.17 (n=106) | 1.90±1.12 (n=1014) | 1.96±1.18 (n=119) | 1.90±1.13 (n=1239) | 0.688 |
| Hematocrit, % (mean±SD) | Adm | 37.57±5.64 (n=138) | 37.82±5.64 (n=1221) | 36.95±8.48 (n=132) | 37.72±5.95 (n=1491) | 0.265 |
| Hematocrit, % (mean±SD) | Mid | 37.45±6.49 (n=71) | 37.06±5.65 (n=878) | 36.64±5.98 (n=95) | 36.96±5.75 (n=1044) | 0.055 |
| Hematocrit, % (mean±SD) | DC | 37.35±5.08 (n=106) | 37.72±5.43 (n=1014) | 35.98±6.41 (n=119) | 37.52±5.53 (n=1239) | 0.005 |
| Platelet count, 103/µL (mean±SD) | Adm | 247.81±87.26 (n=139) | 236.69±96.18 (n=1250) | 236.86±96.99 (n=136) | 237.72±95.47 (n=1525) | 0.426 |
| Platelet count, 103/µL (mean±SD) | Mid | 317.49±131.05 (n=71) | 268.25±119.91 (n=877) | 237.88±80.19 (n=95) | 268.84±118.62 (n=1043) | <0.001 |
| Platelet count, 103/µL (mean±SD) | DC | 346.36±330.71 (n=106) | 297.23±123.74 (n=1016) | 261.69±96.39 (n=119) | 298.02±151.68 (n=1241) | <0.001 |
| INR, median (IQR) | Adm | 1.72±0.68 (n=12) | 1.62±0.84 (n=62) | 1.78±1.93 (n=16) | 1.66±1.08 (n=90) | 0.853 |
| INR, median (IQR) | Mid | | 2.52±1.94 (n=13) | 1.47±0.66 (n=2) | 2.37±1.84 (n=15) | 0.476 |
| INR, median (IQR) | DC | 1.39 (n=1) | 1.96±1.29 (n=13) | 1.21±0.78 (n=2) | 1.83±1.19 (n=16) | 0.687 |
| ALT, U/L, median (IQR) | Adm | 26 (17.25-44) (n=128) | 28 (17-47) (n=1193) | 18 (13.5-27.5) (n=133) | 26 (17-45) (n=1454) | <0.001 |
| ALT, U/L, median (IQR) | Mid | 44 (25.25-76.5) (n=60) | 33 (21-56) (n=751) | 19 (13-25) (n=87) | 31 (20-54.25) (n=898) | <0.001 |
| ALT, U/L, median (IQR) | DC | 37 (24-59.5) (n=77) | 35 (22-55) (n=835) | 22 (14-31) (n=108) | 33 (21-53.75) (n=1020) | <0.001 |
| AST, U/L, median (IQR) | Adm | 35.5 (25-54.5) (n=125) | 34 (23-53) (n=1189) | 26 (19-38) (n=134) | 33 (23-51) (n=1448) | <0.001 |
| AST, U/L, median (IQR) | Mid | 37 (27.25-53) (n=60) | 31 (22-48) (n=748) | 26 (21-33) (n=87) | 30 (22-46) (n=895) | <0.001 |
| AST, U/L, median (IQR) | DC | 26 (19-40) (n=76) | 26 (19-38) (n=826) | 24 (17.2-32.5) (n=108) | 26 (19-37) (n=1010) | 0.171 |
| Total bilirubin, mg/dL, median (IQR) | Adm | 0.5 (0.4-0.8) (n=127) | 0.5 (0.3-0.7) (n=1192) | 0.5 (0.3-0.7) (n=134) | 0.5 (0.3-0.7) (n=1453) | 0.007 |
| Total bilirubin, mg/dL, median (IQR) | Mid | 0.5 (0.4-0.67) (n=60) | 0.5 (0.3-0.7) (n=749) | 0.4 (0.3-0.6) (n=87) | 0.5 (0.3-0.7) (n=896) | 0.035 |
| Total bilirubin, mg/dL, median (IQR) | DC | 0.5 (0.4-0.8) (n=78) | 0.5 (0.4-0.7) (n=835) | 0.5 (0.3-0.7) (n=108) | 0.5 (0.4-0.7) (n=1021) | 0.380 |
| Urea, mg/dL (mean±SD) | Adm | 44.95±37.91 (n=106) | 45.90±47.70 (n=509) | 45.25±32.21 (n=42) | 45.71±45.35 (n=657) | 0.979 |
| Urea, mg/dL (mean±SD) | Mid | 75.56±73.37 (n=27) | 70.03±58.25 (n=58) | | 71.78±63.05 (n=85) | 0.709 |
| Urea, mg/dL (mean±SD) | DC | 53.89±35.84 (n=48) | 56.68±40.37 (n=120) | 41.0±18.19 (n=3) | 55.62±38.79 (n=171) | 0.739 |
| Creatinine, mg/dL, median (IQR) | Adm | 1.0 (0.8-1.2) (n=136) | 0.9 (0.8-1.2) (n=1225) | 1.0 (0.8-1.3) (n=135) | 0.9 (0.8-1.2) (n=1496) | 0.053 |
| Creatinine, mg/dL, median (IQR) | Mid | 1.0 (0.8-1.3) (n=73) | 0.9 (0.7-1.1) (n=813) | 0.9 (0.7-1.2) (n=91) | 0.9 (0.7-1.1) (n=977) | 0.003 |
| Creatinine, mg/dL, median (IQR) | DC | 0.9 (0.7-1.1) (n=94) | 0.8 (0.7-1.0) (n=938) | 0.8 (0.7-1.1) (n=114) | 0.8 (0.7-1.0) (n=1496) | 0.012 |
| Sodium, mmol/L (mean±SD) | Adm | 136.35±5.25 (n=135) | 136.36±4.52 (n=1221) | 135.2±5.86 (n=134) | 136.26±4.73 (n=1490) | 0.025 |
| Sodium, mmol/L (mean±SD) | Mid | 136.03±5.50 (n=71) | 137.08±4.69 (n=818) | 136.93±4.45 (n=96) | 136.99±4.73 (n=985) | 0.197 |
| Sodium, mmol/L (mean±SD) | DC | 137.5±3.35 (n=93) | 137.47±4.14 (n=919) | 137.15±3.81 (n=113) | 137.45±4.05 (n=1125) | 0.726 |
| Potassium, mmol/L (mean±SD) | Adm | 4.03±0.47 (n=135) | 4.03±0.61 (n=1221) | 4.03±0.63 (n=134) | 4.02±0.60 (n=1490) | 0.962 |
| Potassium, mmol/L (mean±SD) | Mid | 4.37±0.57 (n=70) | 4.17±0.65 (n=819) | 4.01±0.63 (n=96) | 4.17±0.64 (n=985) | 0.002 |
| Potassium, mmol/L (mean±SD) | DC | 4.19±0.53 (n=93) | 4.16±0.57 (n=917) | 4.06±0.60 (n=113) | 4.16±0.57 (n=1123) | 0.177 |
| Procalcitonin, ng/mL, median (IQR) | Adm | 0.24 (0.18-0.32) (n=93) | 0.12 (0.06-0.26) (n=373) | 0.20 (0.11-0.45) (n=22) | 0.17 (0.7-0.28) (n=488) | <0.001 |
| Procalcitonin, ng/mL, median (IQR) | Mid | 0.34 (0.22-0.40) (n=67) | 0.21 (0.08-0.41) (n=195) | 0.39 (0.21-6.45) (n=5) | 0.27 (0.11-0.41) (n=267) | 0.001 |
| Procalcitonin, ng/mL, median (IQR) | DC | 0.30 (0.22-0.37) (n=83) | 0.10 (0.04-0.30) (n=249) | 0.15 (0.06-0.33) (n=14) | 0.19 (0.6-0.33) (n=346) | <0.001 |
| hsCRP, mg/L, median (IQR) | Adm | 61.1 (13.57-189.62) (n=138) | 27.3 (7.57-105.9) (n=1242) | 13.6 (4.5-39.5) (n=136) | 26.57 (7.35-105.46) (n=1516) | <0.001 |
| hsCRP, mg/L, median (IQR) | Mid | 44.6 (12.11-108.47) (n=68) | 15.3 (6.3-38.6) (n=832) | 15.6 (4.5-42.4) (n=89) | 15.93 (6.3-40.79) (n=989) | 0.010 |
| hsCRP, mg/L, median (IQR) | DC | 5.22 (2.38-13.13) (n=101) | 4.3 (1.48-12.15) (n=997) | 4.5 (1.8-13.1) (n=116) | 4.55 (1.57-12.21) (n=1214) | 0.140 |
| LDH, U/L (mean±SD) | Adm | 341.29±173.0 (n=37) | 372.87±205.12 (n=157) | 463.4±626.11 (n=5) | 369.28±216.55 (n=199) | 0.450 |
| LDH, U/L (mean±SD) | Mid | 421.83±333.37 (n=6) | 437.38±233.31 (n=32) | | 434.92±246.29 (n=38) | 0.889 |
| LDH, U/L (mean±SD) | DC | 315.42±212.09 (n=12) | 308.92±157.78 (n=52) | 257.0±130.09 (n=3) | 307.76±165.45 (n=67) | 0.860 |
| D-dimer, µg/mL, median (IQR) | Adm | 0.39 (0.20-0.96) (n=122) | 0.41 (0.22-0.88) (n=1185) | 0.49 (0.21-1.17) (n=133) | 0.41 (0.22-0.93) (n=1440) | 0.554 |
| D-dimer, µg/mL, median (IQR) | Mid | 1.02 (0.56-6.90) (n=51) | 0.63 (0.30-1.64) (n=812) | 0.44 (0.21-0.91) (n=86) | 0.63 (0.30-1.62) (n=949) | 0.004 |
| D-dimer, µg/mL, median (IQR) | DC | 0.77 (0.41-2.02) (n=85) | 0.44 (0.25-1.01) (n=967) | 0.42 (0.19-0.89) (n=112) | 0.46 (0.25-1.05) (n=1164) | <0.001 |
| Ferritin, ng/mL (mean±SD) | Adm | 407.6±378.95 (n=6) | 710.6±547.795 (n=248) | 878.05±813.41 (n=11) | 710.7±558.38 (n=265) | 0.253 |
| Ferritin, ng/mL (mean±SD) | Mid | 366.5±308.9 (n=4) | 859.95±521.38 (n=50) | 1581 (n=1) | 837.17±528.3 (n=55) | 0.070 |
| Ferritin, ng/mL (mean±SD) | DC | 462.13±472.63 (n=4) | 777.24±564.82 (n=86) | 962.22±765.94 (n=6) | 775.68±574.21 (n=96) | 0.405 |
Treatment with intravenous fluids, oral fluids, corticosteroid, antiviral such as remdesivir, antibiotics such as fluoroquinolone, cephalosporin, and carbapenem was mostly used for the COVID-19 patients admitted at the hospital in three waves of the study period. Antiviral treatment had been used since the second wave, and antiviral drug such as remdesivir or molnupiravir was used more in the patients of the third and fourth waves than those of the second wave. Most of the patients were supplied with pipeline oxygen therapy. Non-invasive ventilation, invasive ventilation, and inotropes were observed in serious cases. Renal replacement therapy was also done in $1.1\%$ of the cases during three waves. Disease severity during three waves was cumulatively showed as mild, moderate, severe, and critical cases with 56 ($3.5\%$), 344($21.4\%$), 871 ($54.2\%$), and 335 ($20.9\%$), respectively. Critical cases were observed mostly in the second wave compared to later waves (Table 5).
**Table 5**
| Treatments | Treatments.1 | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total (n=1606) | P-value (χ2 test) |
| --- | --- | --- | --- | --- | --- | --- |
| Oral/orogastric fluids | Oral/orogastric fluids | 28 (20.3) | 150 (11.3) | 5 (3.6) | 183 (11.4) | <0.001 |
| IV fluids | IV fluids | 89 (64.5) | 967 (72.7) | 111 (81.0) | 1167 (72.7) | 0.009 |
| Corticosteroid | Corticosteroid | 108 (78.3) | 979 (73.6) | 84 (61.3) | 1171 (73.0) | 0.003 |
| IL-6 inhibitor | IL-6 inhibitor | 9 (6.5) | 106 (8.0) | 0 (0) | 115 (7.2) | 0.003 |
| JAK inhibitor | JAK inhibitor | 0 (0) | 105 (7.9) | 15 (10.9) | 120 (7.5) | 0.001 |
| Monoclonal antibodies | Monoclonal antibodies | 0 (0) | 1 (0.1) | 0 (0) | 1 (0.1) | 0.901 |
| Antiviral: remdesivir | Antiviral: remdesivir | 52 (37.4) | 805 (60.5) | 104 (75.9) | 961 (59.8) | <0.001 |
| Antiviral: molnupiravir | Antiviral: molnupiravir | 0 (0) | 2 (0.2) | 12 (8.8) | 14 (0.9) | <0.001 |
| Antiviral: favipiravir | Antiviral: favipiravir | 0 (0) | 13 (1.0) | 0 (0) | 13 (0.1) | 0.257 |
| NSAID | NSAID | 13 (8.1) | 125 (9.4) | 22 (16.1) | 160 (10.0) | 0.046 |
| Antibiotics | Antibiotics | Antibiotics | Antibiotics | Antibiotics | Antibiotics | Antibiotics |
| Macrolides | Macrolides | 11 (7.9) | 43 (3.2) | 7 (5.1) | 61 (3.8) | 0.016 |
| Fluoroquinolones | Fluoroquinolones | 84 (60.4) | 495 (37.2) | 63 (46.0) | 642 (40.0) | <0.001 |
| Third/fourth-generation cephalosporin | Third/fourth-generation cephalosporin | 57 (41.0) | 645 (48.5) | 53 (38.7) | 755 (47.0) | 0.030 |
| Carbapenem | Carbapenem | 41 (29.5) | 422 (31.7) | 32 (23.4) | 495 (30.8) | 0.122 |
| Amoxicillin-clavulanate | Amoxicillin-clavulanate | 50 (36.0) | 246 (18.5) | 13 (9.5) | 309 (19.2) | <0.001 |
| Cefoperazone-sulbactam | Cefoperazone-sulbactam | 6 (4.3) | 125 (9.4) | 10 (7.3) | 141 (8.8) | 0.107 |
| Piperacillin-tazobactam | Piperacillin-tazobactam | 22 (15.8) | 48 (3.6) | 8 (5.8) | 78 (4.9) | <0.001 |
| Gentamycin/amikacin | Gentamycin/amikacin | 0 (0) | 12 (0.9) | 0 (0) | 12 (0.7) | 0.285 |
| Linezolid/tedizolid | Linezolid/tedizolid | 9 (6.5) | 32 (2.4) | 0 (0) | 41 (2.6) | 0.002 |
| Metronidazole/ornidazole | Metronidazole/ornidazole | 4 (2.9) | 25 (1.9) | 6 (4.4) | 35 (2.2) | 0.136 |
| Clindamycin/lincosamide | Clindamycin/lincosamide | 3 (2.2) | 45 (3.4) | 1 (0.7) | 49 (3.1) | 0.186 |
| Vancomycin/teicoplanin | Vancomycin/teicoplanin | 2 (1.4) | 4 (0.3) | 0 (0) | 6 (1.8) | 0.085 |
| Supportive care | Supportive care | Supportive care | Supportive care | Supportive care | Supportive care | Supportive care |
| ICU/HDU admission | ICU/HDU admission | 40 (28.8) | 228 (17.1) | 13 (9.5) | 281 (17.5) | <0.001 |
| Oxygen therapy | Oxygen therapy | 99 (71.2) | 798 (60.0) | 65 (47.4) | 962 (59.9) | <0.001 |
| Interface | Nasal prongs | 31 (22.3) | 296 (22.3) | 39 (10.7) | 366 (22.8) | <0.001 |
| Interface | Masks | 15 (10.8) | 91 (6.8) | 5 (3.6) | 111 (6.9) | <0.001 |
| Interface | HFNC | 31 (22.3) | 61 (4.6) | 0 (0) | 92 (5.7) | <0.001 |
| Interface | Masks with reservoir | 16 (11.5) | 204 (15.3) | 17 (12.4) | 237 (14.8) | <0.001 |
| Interface | NIV mask | 6 (4.3) | 145 (10.9) | 4 (2.9) | 155 (9.7) | <0.001 |
| Oxygen flow | 1-5 L/minute | 31 (22.3) | 309 (23.2) | 39 (28.5) | 379 (23.6) | <0.001 |
| Oxygen flow | 6-10 L/minute | 18 (12.9) | 90 (6.8) | 8 (5.8) | 116 (7.2) | <0.001 |
| Oxygen flow | 11-15 L/minute | 13 (9.4) | 78 (5.9) | 6 (4.4) | 97 (6.0) | <0.001 |
| Oxygen flow | >15 L/minute | 37 (26.6) | 322 (24.2) | 12 (8.8) | 371 (23.1) | <0.001 |
| Source of oxygen | Piped | 98 (70.5) | 787 (59.2) | 65 (47.4) | 950 (59.2) | <0.009 |
| Source of oxygen | Cylinder | 1 (0.7) | 10 (0.8) | 0 (0) | 11 (0.7) | <0.009 |
| Source of oxygen | Concentrator | 0 (0) | 1 (0.1) | 0 (0) | 1 (0.1) | <0.009 |
| NIV | NIV | 6 (4.3) | 141 (10.6) | 4 (2.9) | 151 (9.4) | 0.001 |
| Invasive ventilation | Invasive ventilation | 0 (0) | 8 (0.6) | 0 (0) | 8 (0.5) | 0.436 |
| Inotropes/vasopressors | Inotropes/vasopressors | 4 (2.9) | 64 (4.8) | 6 (4.4) | 74 (4.6) | 0.580 |
| Prone position | Prone position | 43 (30.9) | 490 (36.8) | 37 (27.0) | 570 (35.5) | 0.036 |
| RRT | RRT | 1 (0.7) | 14 (1.1) | 3 (2.2) | 18 (1.1) | 0.434 |
| Severity of the diseases | Mild | 5 (3.6) | 48 (3.6) | 3 (2.2) | 56 (3.5) | 0.064 |
| Severity of the diseases | Moderate | 27 (19.4) | 285 (21.4) | 32 (23.4) | 344 (21.4) | 0.064 |
| Severity of the diseases | Severe | 64 (46.0) | 727 (54.7) | 80 (58.4) | 871 (54.2) | 0.064 |
| Severity of the diseases | Critical | 43 (30.9) | 270 (20.3) | 22 (16.1) | 335 (20.9) | 0.064 |
Treatment outcomes of the COVID-19 patients admitted at the hospital were described in Table 6. The mean duration of hospital stay in days was 10.8±5.94, 10.6±6.11, and 7.3±2.88 in the second, third, and fourth waves, respectively. About $90\%$ of the hospitalized COVID-19 patients were discharged alive, and only $7.9\%$ and $9.4\%$ died in the hospital in the second and third waves. No mortality was observed in the fourth wave. There were 67 cases re-admitted as post-COVID-19 complications. Among them, only 13 ($19.4\%$) had been vaccinated, and four ($6\%$) died with post-COVID-19 complications.
**Table 6**
| Outcomes | Second wave (n=139) | Third wave (n=1330) | Fourth wave (n=137) | Total (n=1606) | P-value |
| --- | --- | --- | --- | --- | --- |
| Hospital stay in days (mean±SD) | 10.8±5.94 | 10.6±6.11 | 7.3±2.88 | 10.3±5.96 | <0.001 (t test) |
| Discharged alive | 123 (88.5) | 1162 (87.4) | 124 (90.5) | 1409 (87.7) | <0.001 (χ2 test) |
| Palliative discharge | 5 (3.6) | 37 (2.8) | 10 (7.3) | 52 (3.2) | <0.001 (χ2 test) |
| Transferred to other facilities | 0 (0) | 6 (0.5) | 3 (2.2) | 9 (0.6) | <0.001 (χ2 test) |
| Death | 11 (7.9) | 125 (9.4) | 0 (0) | 136 (8.5) | <0.001 (χ2 test) |
## Discussion
At the beginning of the COVID-19 pandemic in 2020, transmission was controlled by the Department of Public Health, and serious cases were treated at public hospitals of the Department of Medical Services under national treatment guidelines developed by the Ministry of Health. Later, a shortage of human resources resulted because of both the pandemic severity and the unstable country situation. Moreover, a surge in infection also led to the need for support by private sector involvement. Therefore, private hospitals started to manage COVID-19 infections in early 2021. This study evaluated the management of hospitalized patients by a private hospital in three different waves, the second, third, and fourth waves, in Myanmar. This study describes the clinical profile and outcomes of COVID-19 patients admitted at a private hospital in Mandalay, Myanmar. According to the findings, the private hospital in Mandalay could tackle COVID-19 severe cases in line with national treatment guidelines since the second wave and could provide better management in the fourth wave.
The mean age of the patients in the third wave was younger than those admitted in the second wave. COVID-19 immunization in Myanmar started in January 2021 with old-age people and healthcare workers as the first priority. Because of the immunization schedule mentioned earlier, older people were prioritized, and younger people who were going out for work were infected more than older people. This is similar to the studies of other countries in Asia [10-12]. However, the mean age of the patients in the fourth wave was older than those admitted in the second and third waves. In the later wave, people became familiar with mild-to-moderate infection and took self-care at home, and old people with severe infection took hospital care at a private hospital. Another reason may be because old people suffered more severe effects of the omicron variant than young people. Therefore, old age with severe cases should be closely monitored at the hospital. This finding is consistent with other studies [13-16]. Infection among healthcare workers was involved at $8.0\%$. The proportion was higher in the second and third waves with $8.6\%$ each. However, it was lower in the fourth wave with $1.2\%$. It might be due to the vaccination effect among healthcare workers in the fourth wave, and the finding is similar to a retrospective study in India [17].
Comorbid conditions with hypertension or diabetes diseases were most commonly observed in three waves of COVID-19 infection in City Hospital, Mandalay. These were recorded with $48.2\%$, $39.8\%$, and $54.0\%$ for hypertension and $41.0\%$, $34.3\%$, and $39.4\%$ for diabetes in the second, third, and fourth waves, respectively. Therefore, patients with comorbid conditions of hypertension or diabetes and who are old should be taken with close care at the hospital. Although fever, cough, and headache were the most presenting symptoms in all waves, influenza-like symptoms such as sore throat and rhinorrhea were observed in high proportion in the fourth wave compared to previous waves. These findings were in contrast with the findings of the studies in Qatar and Iran [10,18].
Complications such as pneumonia, liver dysfunction, anemia, ARDS, cardiac injury, coagulopathy, acute kidney injury, and shock were observed in three waves. Respiratory complications such as ARDS and pneumonia were observed more in the second and third waves compared to fourth wave. Those are similar with a study in Wuhan and Qatar [14,18]. This led to higher proportion of oxygen therapy and higher proportion of ICU admission in the second and third waves compared to the fourth wave. This finding is also similar with the study done in 41 hospitals in India [19].
In this study, increased level of procalcitonin, C-reactive protein, and D-dimer were recorded among the patients throughout three waves especially in the second and third waves. Therefore, oxygen requirement with assisted ventilation was high among the patients who had high CRP and D-dimer in the second and third waves like other studies [10,20].
The hospital could manage the severe cases according to national treatment guidelines in all waves. Antiviral treatment was observed more in patients of the third and fourth wave than those of the second wave because of the less proportion of severe cases needed for antiviral treatment in the second wave, or locally recommended antiviral treatment were not available in the area or hospital in that period. Nevertheless, antiviral treatment might lead to better outcomes in the fourth wave. Similarly, treatment with non-invasive ventilation and invasive ventilation was used in more patients of the third wave compared to those of other waves. These findings are similar to those found in Africa [21].
The mean duration of hospital stay was shortened in the fourth wave compared to previous waves. This finding is also consistent with other studies [18,21]. No mortality was observed in the fourth wave. This successful result may be reflected from experienced practices of the hospital achieved in previous waves. Therefore, it showed that hospital management became better in the fourth wave compared to previous waves [22,23].
Limitations of the study This study is a retrospective record review, so it is not a prospective study. This study describes the clinical profile of COVID-19 patients by assessment with CRF developed using the WHO Global COVID-19 Clinical Data Platform. The CRF contains three modules: module 1, to be completed on the first day of admission to the health center; module 2, to be completed daily during hospital stay for as many days as resources allow and to continue to follow-up patients who transfer between wards; and module 3, to be completed at discharge or death. However, the current study reviewed the patients’ records using WHO module that was modified. It could not be completed during hospital stay and continued to follow-up. The record review was done after the patients were discharged. Therefore, the assessment might not be complete enough. Another limitation is that although there were four waves in Myanmar, the clinical profile of the patients of the first wave was not studied in this research. In this connection, the emergency response and control activities were sufficient with available and efficient workforce in 2020. Therefore, private hospital involvement was not needed in 2020, resulting in no studies relating private hospital management in the first wave of 2020. In addition, laboratory results were reviewed on available findings in EMR. Many laboratory investigations of the patients were not available, and those were not included in the analysis. Moreover, clinical outcomes of the admitted patients were determined by reviewing the EMR, and those who were discharged or referred to other facilities were not followed. This study was conducted in one private hospital only; it was not done as a multicenter study.
## Conclusions
At the beginning of the COVID-19 pandemic in 2020, cases were contained and treated at public hospitals in Myanmar. However, because of an unstable country situation, the private hospitals supported the management of the COVID-19 crisis in the subsequent waves: Delta in 2021 and Omicron in 2022. This study recommended that COVID-19 patients with comorbid conditions of hypertension or diabetes and who are 65 years and older should be taken with intensive care support at the hospital and should be encouraged for vaccination. This study also concluded that a private hospital in Mandalay could tackle COVID-19 severe cases in line with national treatment guidelines since the second wave and could provide better management in the fourth wave. Antiviral treatment should be used in severe COVID-19 cases for further emergency management. In conclusion, private hospital involvement in the COVID-19 pandemic is supportive of the healthcare provision in Myanmar in an emergency situation.
## References
1. **World Health Organization: the WHO global clinical platform for COVID-19**. (2020)
2. World Health Organization (WHO). (2020. **World Health Organization: antigen-detection in the diagnosis of SARS-CoV-2 infection**. (2020)
3. **World Health Organization: therapeutics and COVID-19: living guideline**. (2021)
4. Rhodes A, Evans LE, Alhazzani W. **Surviving sepsis campaign: international guidelines for management of sepsis and septic shock: 2016**. *Intensive Care Med* (2017) **43** 304-377. PMID: 28101605
5. Rawson TM, Moore LS, Zhu N. **Bacterial and fungal coinfection in individuals with coronavirus: a rapid review to support COVID-19 antimicrobial prescribing**. *Clin Infect Dis* (2020) **71** 2459-2468. PMID: 32358954
6. Poyiadji N, Shahin G, Noujaim D, Stone M, Patel S, Griffith B. **COVID-19-associated acute hemorrhagic necrotizing encephalopathy: imaging features**. *Radiology* (2020) **296** 0-20
7. Sheahan TP, Sims AC, Graham RL. **Broad-spectrum antiviral GS-5734 inhibits both epidemic and zoonotic coronaviruses**. *Sci Transl Med* (2017) **9** 0
8. Martinez MA. **Compounds with therapeutic potential against novel respiratory 2019 coronavirus**. *Antimicrob Agents Chemother* (2020) **64** 0-20
9. Department of Medical Services. **Clinical management guidelines for COVID-19 acute respiratory disease**. (2020) **8** 2020
10. Jalali SF, Ghassemzadeh M, Mouodi S, Javanian M, Akbari Kani M, Ghadimi R, Bijani A. **Epidemiologic comparison of the first and second waves of coronavirus disease in Babol, north of Iran**. *Caspian J Intern Med* (2020) **11** 544-550. PMID: 33425273
11. Seong H, Hyun HJ, Yun JG, Noh JY, Cheong HJ, Kim WJ, Song JY. **Comparison of the second and third waves of the COVID-19 pandemic in South Korea: importance of early public health intervention**. *Int J Infect Dis* (2021) **104** 742-745. PMID: 33556610
12. Saito S, Asai Y, Matsunaga N. **First and second COVID-19 waves in Japan: a comparison of disease severity and characteristics**. *J Infect* (2021) **82** 84-123
13. Liu Y, Mao B, Liang S. **Association between age and clinical characteristics and outcomes of COVID-19**. *Eur Respir J* (2020) **55** 2001112. PMID: 32312864
14. Wang D, Hu B, Hu C. **Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan, China**. *JAMA* (2020) **323** 1061-1069. PMID: 32031570
15. Guan WJ, Ni ZY, Hu Y. **Clinical characteristics of coronavirus disease 2019 in China**. *N Engl J Med* (2020) **382** 1708-1720. PMID: 32109013
16. Ito N, Kitahara Y, Miwata K, Okimoto M, Takafuta T. **Comparison of COVID-19 pneumonia during the SARS-CoV-2 Omicron wave and the previous non-Omicron wave in a single facility**. *Respir Investig* (2022) **60** 772-778
17. Kundavaram AP, Selvan S, Raja V. **Retrospective study of comparison of clinical severity and outcome of hospitalised COVID-19 patients during the first and second waves of the pandemic in India**. *BMJ Open* (2022) **12** 0
18. Naushad VA, Purayil NK, Chandra P. **Comparison of demographic, clinical and laboratory characteristics between first and second COVID-19 waves in a secondary care hospital in Qatar: a retrospective study**. *BMJ Open* (2022) **12** 0
19. Kumar G, Mukherjee A, Sharma RK. **Clinical profile of hospitalized COVID-19 patients in first & second wave of the pandemic: insights from an Indian registry based observational study**. *Indian J Med Res* (2021) **153** 619-628. PMID: 34259194
20. Iftimie S, López-Azcona AF, Vallverdú I. **First and second waves of coronavirus disease-19: a comparative study in hospitalized patients in Reus, Spain**. *PLoS One* (2021) **16** 0
21. Maslo C, Friedland R, Toubkin M, Laubscher A, Akaloo T, Kama B. **Characteristics and outcomes of hospitalized patients in South Africa during the COVID-19 Omicron wave compared with previous waves**. *JAMA* (2022) **327** 583-584. PMID: 34967859
22. Chhabra S, Sethi S, Kaur S. **Clinical profile of coronavirus disease 2019 comparing the first and second waves: a single-center study from North India**. *Int J Appl Basic Med Res* (2022) **12** 95-102. PMID: 35754672
23. Chams N, Chams S, Badran R. **COVID-19: a multidisciplinary review**. *Front Public Health* (2020) **8** 383. PMID: 32850602
|
---
title: 'COVID-19 and Pre-existing Type 2 Diabetes Mellitus: A Retrospective Observational
Study From Eastern India on the Association Between Glycaemic Control and Treatment
Outcomes'
journal: Cureus
year: 2023
pmcid: PMC10030157
doi: 10.7759/cureus.35165
license: CC BY 3.0
---
# COVID-19 and Pre-existing Type 2 Diabetes Mellitus: A Retrospective Observational Study From Eastern India on the Association Between Glycaemic Control and Treatment Outcomes
## Abstract
Background: Diabetes has emerged as an important risk factor for causing severe illness and death from COVID-19. There is a paucity of structured data from the Indian subcontinent on the impact that glycaemic control (both immediate and remote) has on the degree of required medical intervention and mortality among hospitalized COVID-19 patients with type 2 diabetes mellitus (T2DM).
Objectives: To evaluate the differences in clinical characteristics and treatment outcomes between well-controlled and poorly controlled patients with T2DM and COVID-19.
Methods: This was a retrospective observational study. Data on 177 patients who were hospitalized between February 2021 and July 2021 were categorized into four groups using a cut-off admission plasma glucose of <200 mg/dL and glycated hemoglobin (HbA1c) <$7.5\%$.
Results: Patients with poorly controlled diabetes presented at a significantly older age than the other groups. Radiological findings suggested severe lung involvement in them. As a combined group patients with HbA1c ≥$7.5\%$ required more ventilatory requirement as compared with the group having HbA1c <$7.5\%$ irrespective of admission glucose. They also required prolonged hospitalization and intensive care unit (ICU) stays as compared with the well-controlled diabetes group. In this study, within similar ranges of HbA1c admission glucose seemed to have a numerical impact on mortality without being able to achieve statistical significance.
Conclusion: From the current study, it can be concluded that poor glycaemic control, particularly HbA1c ≥$7.5\%$, is an important risk factor for the development of severe COVID-19 and a predictor for the requirement of more intensive treatment and adverse treatment outcomes leading to increased hospital and ICU stay.
## Introduction
Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the novel coronavirus that causes coronavirus disease 2019 (COVID-19), was first reported in Wuhan, China, in December 2019 and has spread worldwide. As of 11 April 2022, 497,057,239 globally confirmed cases of COVID-19 have been reported on the World Health Organization COVID-19 dashboard, including 6,179,104 deaths [1].
Studies have suggested that most people affected by COVID-19 have comorbidities, the most prevalent of which are hypertension, diabetes mellitus, and cardiovascular disease [2-5]. Generally, about 10-$20\%$ of patients with COVID-19 had diabetes. Research suggests that diabetic patients were more susceptive to SARS-CoV-2 and in the long run, had poor disease-related outcomes [6-8]. In diabetic patients glucotoxicity coupled with endothelial damage by inflammation, oxidative stress, and cytokine production contribute to an increased risk of thromboembolic complications and damage to vital organs [9]. Moreover, drugs commonly used in the treatment of patients with COVID-19, such as systemic corticosteroids or antiviral agents, might aggravate hyperglycemia.
The cornerstone for diabetes management is glucose control. Some researchers have clarified the importance and provided insights for glucose control in patients with diabetes and COVID-19 [10-12]. For individuals with COVID-19 and pre-existing diabetes, a key challenge for clinicians is to improve outcomes in the face of uncertainty regarding the degree of glycaemic management that should be maintained and any effects this might have on the benefits and risks of the overall treatment. Thus, detailed analysis of data from such patients is needed that links glycaemic control with treatment outcomes, including mortality. The prevalence of type 2 diabetes mellitus (T2DM) in *India is* high, and the paucity of data on its association with COVID-19 warrants the identification of factors responsible for severe outcomes in such patients. The present study included COVID-19 patients with pre-existing T2DM admitted to a tertiary care hospital in Kolkata, West Bengal focusing on the association between their glycaemic control and treatment outcomes.
## Materials and methods
Study design This retrospective observational study evaluated T2DM patients hospitalized at Kali Prasad Chowdhury Medical College and Hospital in Kolkata, West Bengal during the second wave of the COVID-19 pandemic between February 2021 and July 2021 with laboratory-confirmed COVID-19 illness. The objective of the study was to find out whether good glycaemic control at the time of admission is associated with better treatment outcomes for COVID-19. We used a cut-off admission plasma glucose of <200 mg/dL and HbA1c <$7.5\%$ as a cut-off for grouping the patients' data. So there were four groups based on their respective glycaemic control: Group (I) - T2DM patients with HbA1c <$7.5\%$ and admission glucose <200 mg/dL Group (II) - T2DM patients with HbA1c <$7.5\%$ and admission glucose ≥200 mg/dL
Group (III) - T2DM patients with HbA1c ≥$7.5\%$ and admission glucose <200 mg/dL Group (IV) - T2DM patients with HbA1c ≥$7.5\%$ and admission glucose ≥200 mg/dL The patients were initially admitted to the Covid ward of the hospital and afterward, patients with warning signs were transferred to the intensive care unit (ICU). The management protocol of the patients followed in the study was as per the guideline approved by the department of health and family welfare, the government of West Bengal [13]. All patients were given insulin to control blood sugar.
Ethical considerations The study being a retrospective observational study ethical approval was not sought. Moreover, due to the retrospective nature of the study, participants were de-identified and could not be contacted hence informed consent was not obtained. Confidentiality and anonymity were maintained. This study is solely intended for academic purposes. All tenets of Helsinki's declaration on bioethics policy were adhered to.
Definitions As per WHO interim guidelines reverse transcriptase polymerase chain reaction (RT-PCR) positive patients (sample source nasal and pharyngeal swab) were considered to be confirmed COVID-19-positive cases. Patients who required a hospital stay greater than 24 hours were taken as hospitalized patients [14]. The first glucose measurement within a time window of 4 hours before and up to 4 hours after admission was used as admission glucose and categorized as well-controlled (< 200 mg/dL) and poorly controlled (≥200mg/dL) [15,16]. Moreover, according to the National Institute for Health and Care Excellence (NICE) guidelines for the management of type 2 diabetes in adults, we selected HbA1c value <$7.5\%$ (58 mmol/mol) as a criterion for good glycaemic control [17].
Outcome measures Data for four groups of patients were compared against each other to assess differences in [1] demographic and clinical characteristics like age, gender, and radiological findings using an average visual score from CT severity score (CTSS) from high-resolution computed tomography (HRCT) chest (assigned out of 25 based upon percentage area involved in each of the five lobes); [2] treatment outcome including ICU stay, length of hospital stay, the requirement of ventilator support and death.
Statistical analysis In the present study, we carried out descriptive and inferential statistical analysis. Results on continuous variables were presented as mean (SD) for parametric data. The normality of data was tested by the Anderson-Darling test, Shapiro-Wilk, Kolmogorov-Smirnoff test, and visually by QQ plot. The variance was tested by Levene’s test. For data measured in ratio scales like CTSS, ICU stay, and hospital stays, the analysis of covariance (ANCOVA) test was used to find the difference between the groups, and post hoc Dunnett’s test or Bonferroni test was carried out to find the inter-group differences. For categorical variables, the difference between groups was tested by the Chi-square test. Fischer's exact approach was used for post hoc analysis of a Chi-squared test. Statistical evaluations carried out were two-sided and the cut-off for statistical significance was taken as $p \leq 0.05.$
Statistical software The statistical software namely SAS (Statistical Analysis System) version 9.2 for windows, SAS Institute Inc. Cary, NC, USA and Statistical Package for Social Sciences (SPSS Complex Samples) version 21.0 for windows, SPSS, Inc., Chicago, IL, USA, has been used for data analysis.
## Results
In total, there were 177 patients with pre-existing T2DM and COVID-19 included in this study. Thirty-two patients ($18.1\%$) had HbA1c <$7.5\%$ and admission glucose <200 mg/dL (Group-I), 36 patients ($20.34\%$) had HbA1c <$7.5\%$, and admission glucose ≥200 mg/dL (Group-II). Forty-seven patients ($26.6\%$) had HbA1c ≥$7.5\%$ and admission glucose <200 mg/dL (Group-III) and 62 patients ($35.03\%$) had HbA1c ≥$7.5\%$ and admission glucose ≥200 mg/dL(Group-IV).
The age distribution of the participants is shown in Table 1. Overall the mean age of the patients was 53.75 years (standard deviation of 14.02 years). There was a significant difference in mean age across four groups of patients (ANOVA, $$p \leq 0.004$$). Inter-group analysis was done by the Bonferroni test. Patients with HbA1c ≥$7.5\%$ + admission glucose ≥200 mg/dL - (Group-IV i.e poor glycaemic control) tended to be significantly older compared to patients of Group-I having HbA1c <7.5 either with admission glucose <200 mg/dL ($$p \leq 0.008$$) or of Group-II with admission glucose ≥200 mg/dL ($$p \leq 0.031$$). Besides significant age difference was found between patients having HbA1c ≥$7.5\%$ and patients having HbA1c <$7.5\%$ both with admission glucose <200 mg/dL ($$p \leq 0.031$$) viz Group-III versus Group-I. A total of 110 patients ($62.15\%$) were male and 67 patients ($37.85\%$) were female, however, with respect to gender there was no statistically significant difference between the groups (Chi-square, $$p \leq 0.0529$$).
**Table 1**
| Group | N | Mean | Std. deviation | P (ANOVA) | Post hoc test |
| --- | --- | --- | --- | --- | --- |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 45.66 | 15.924 | 0.004* | a-0.203 |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 45.66 | 15.924 | 0.004* | b-0.038* |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 45.66 | 15.924 | 0.004* | c-0.008* |
| Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 36 | 48.54 | 13.336 | 0.004* | d-0.118 |
| Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 47 | 53.83 | 11.781 | 0.004* | d-0.118 |
| Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 60.81 | 11.437 | 0.004* | e-0.031* |
| Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 60.81 | 11.437 | 0.004* | f-0.082 |
| Total | 177 | 53.75 | 14.021 | 0.004* | |
Table 2 presents the effects of glycaemic control on the severity of COVID-19 infection of patients with T2DM and COVID-19 after controlling for the effects of baseline age and body weight. Three parameters were assessed viz chest CTSS, ICU stay, and length of hospital stay. The mean CTSS was 12.47 (standard deviation 2.76). The mean duration of ICU stay and hospital stay of the patients was 5.36 days (standard deviation of 3.34 days) and 9.94 days (standard deviation of 3.26 days), respectively. There was a statistically significant difference in mean CTSS, mean duration of ICU stay, and mean duration of hospital stay between the four groups of patients (ANCOVA, $$p \leq 0.011$$, <0.001, 0.002, respectively). Post hoc Dunnett’s test showed that compared to the well-controlled group - Group I (HbA1c <$7.5\%$ and admission glucose <200 mg/dL), CTSS was significantly higher ($$p \leq 0.005$$) in the poor-controlled group - Group IV (HbA1c ≥$7.5\%$ + admission glucose ≥200 mg/dL). A total of 133 patients ($75.14\%$) needed admission to ICU. The length of ICU stay and hospital stay was significantly less in the well-controlled group compared to the groups with poor long-term glycaemic control i.e. HbA1c ≥$7.5\%$. However, within the poorly controlled HbA1c groups, viz Group III and Group IV (HbA1c >$7.5\%$) admission glucose did influence the ICU stay, with higher admission glucose leading to a significantly higher ICU stay. Hospital stay on the other hand was influenced by HbA1c with higher HbA1c requiring significantly more days in hospital management.
**Table 2**
| Clinical characteristics | Groups | N | Mean | Std. deviation | Std. error | 95% Confidence interval for mean | 95% Confidence interval for mean.1 | P value (ANCOVA) | Post hoc test |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Clinical characteristics | Groups | N | Mean | Std. deviation | Std. error | Lower bound | Upper bound | P value (ANCOVA) | Post hoc test |
| CTSS | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 11.41 | 1.829 | .323 | 10.75 | 12.07 | 0.011* | a-0.583 |
| CTSS | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 11.41 | 1.829 | .323 | 10.75 | 12.07 | 0.011* | b-0.230 |
| CTSS | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 11.41 | 1.829 | .323 | 10.75 | 12.07 | 0.011* | c-0.005* |
| CTSS | Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 36 | 12.03 | 1.543 | .261 | 11.50 | 12.56 | 0.011* | d-0.929 |
| CTSS | Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 47 | 12.47 | 2.765 | .403 | 11.66 | 13.28 | 0.011* | d-0.929 |
| CTSS | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 13.27 | 3.422 | .435 | 12.41 | 14.14 | 0.011* | e-0.091 |
| CTSS | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 13.27 | 3.422 | .435 | 12.41 | 14.14 | 0.011* | f-0.683 |
| CTSS | Total | 177 | 12.47 | 2.761 | .208 | 12.06 | 12.88 | 0.011* | |
| ICU stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 22 | 3.50 | 2.559 | .546 | 2.37 | 4.63 | <0.001* | a-1.000 |
| ICU stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 22 | 3.50 | 2.559 | .546 | 2.37 | 4.63 | <0.001* | b-0.038* |
| ICU stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 22 | 3.50 | 2.559 | .546 | 2.37 | 4.63 | <0.001* | c<0.001* |
| ICU stay | Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 24 | 3.50 | 1.560 | .319 | 2.84 | 4.16 | <0.001* | d-0.006* |
| ICU stay | Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 40 | 5.70 | 3.502 | .554 | 4.58 | 6.82 | <0.001* | d-0.006* |
| ICU stay | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 47 | 6.89 | 3.383 | .493 | 5.90 | 7.89 | <0.001* | e-0.501 |
| ICU stay | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 47 | 6.89 | 3.383 | .493 | 5.90 | 7.89 | <0.001* | f<0.001* |
| ICU stay | Total | 133 | 5.36 | 3.340 | .290 | 4.79 | 5.93 | <0.001* | |
| Hospital stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 8.59 | 2.525 | .446 | 7.68 | 9.50 | 0.002* | a-0.990 |
| Hospital stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 8.59 | 2.525 | .446 | 7.68 | 9.50 | 0.002* | b-0.008* |
| Hospital stay | Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 32 | 8.59 | 2.525 | .446 | 7.68 | 9.50 | 0.002* | c-0.023* |
| Hospital stay | Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 36 | 8.91 | 1.669 | .282 | 8.34 | 9.49 | 0.002* | d-0.009* |
| Hospital stay | Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 47 | 10.87 | 3.555 | .518 | 9.83 | 11.92 | 0.002* | d-0.009* |
| Hospital stay | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 10.52 | 3.686 | .472 | 9.58 | 11.47 | 0.002* | e-0.997 |
| Hospital stay | Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 62 | 10.52 | 3.686 | .472 | 9.58 | 11.47 | 0.002* | f-0.701 |
| Hospital stay | Total | 177 | 9.94 | 3.257 | .246 | 9.46 | 10.43 | 0.002* | |
Table 3 presents the results of the associations of glycaemic control and prognostic outcome of the patients. A total of 18 patients ($10.17\%$) died out of which 11 patients ($17.7\%$) were in the poorly controlled group. However, there was no significant difference in mortality among the four groups of patients (Chi-square, $$p \leq 0.090$$). Within the two groups of HbA1c greater than $7.5\%$ and less than $7.5\%$, however, those with admission glucose less than 200mg/dL had numerical supremacy in terms of less mortality which failed to reach statistical significance.
**Table 3**
| Group | Outcome | Outcome.1 | Total N (%) | P |
| --- | --- | --- | --- | --- |
| Group | Alive N (%) | Death N (%) | Total N (%) | P |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 31 (96.9) | 1 (3.1) | 32 (100) | 0.09 |
| Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 33 (91.7) | 3 (8.3) | 36 (100) | 0.09 |
| Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 44 (93.6) | 3 (6.4) | 47 (100) | 0.09 |
| Known T2DM + HbA1c≥7.5% + Admission glucose≥200 mg/dL (IV) | 51 (82.3) | 11 (17.7) | 62 (100) | 0.09 |
| Total | 159 (89.8) | 18 (10.2) | 177 (100) | 0.09 |
Glycaemic control was found to have a significant effect on the requirement of more intensive in-hospital treatment like mechanical ventilator support in patients with T2DM and COVID-19 (Chi-square, p=<0.001) (Table 4). A total of $54.8\%$ of patients with poor glycaemic control (HbA1c ≥$7.5\%$ + admission glucose ≥200 mg/dL) required ventilation. In a within-group analysis it was seen that ventilator support was applied significantly more frequently to the individuals with poor long-term glycaemic control viz Group III and Group IV (i.e HbA1c ≥$7.5\%$) compared to the well-controlled group viz Group 1 and Group II (i.e HbA1c <$7.5\%$) irrespective of the admission glucose values. Within similar tertiles of HbA1c, however, more patients with admission glucose greater than 200mg/dL required ventilatory assistance than those with admission glucose less than 200mg/dL without the numbers achieving statistical significance.
**Table 4**
| Group | Ventilatory support | Ventilatory support.1 | Total N (%) | P (Chi-square test) | Post hoc test |
| --- | --- | --- | --- | --- | --- |
| Group | No ventilation N (%) | Ventilation N (%) | Total N (%) | P (Chi-square test) | Post hoc test |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 29 (90.6) | 3 (9.4) | 32 (100) | <0.001* | a-0.870 |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 29 (90.6) | 3 (9.4) | 32 (100) | <0.001* | b-0.042* |
| Known T2DM + HbA1c<7.5% + Admission glucose<200 mg/dL (I) | 29 (90.6) | 3 (9.4) | 32 (100) | <0.001* | c<0.001* |
| Known T2DM + HbA1c<7.5% + Admission glucose≥200 mg/dL (II) | 30 (83.3) | 6 (16.7) | 36 (100) | <0.001* | d-0.016* |
| Known T2DM + HbA1c≥7.5% + Admission glucose<200 mg/dL (III) | 26 (55.3) | 21 (44.7) | 47 (100) | <0.001* | d-0.016* |
| Known T2DM + HbA1c≥7.5% + Admission glucose ≥200 mg/dL (IV) | 28 (45.2) | 34 (54.8) | 62 (100) | <0.001* | e-0.001* |
| Known T2DM + HbA1c≥7.5% + Admission glucose ≥200 mg/dL (IV) | 28 (45.2) | 34 (54.8) | 62 (100) | <0.001* | f-0.687 |
| Total | 113 (63.8) | 64 (36.2) | 177 (100) | <0.001* | |
## Discussion
In this retrospective study, we examined the effect of glycaemic control on the prognosis of COVID-19 in patients with T2DM. We considered both random admission glucose <200 mg/dL and HbA1c <$7.5\%$ as a criterion of good glycaemic control over the past three months and immediately, respectively. Few other observational studies concentrated on blood glucose control for the prognosis of COVID-19 patients with diabetes however they used only blood glucose level as a surrogate of glycaemic control [7,14,18-20]. Bhandari S et al. conducted a similar study but they categorized patients into two groups based on an HbA1c cut-off of less than $8\%$ [21]. Our results show that as a combined group, patients with poorly controlled diabetes presented to the hospital at an older age and progressed to severe COVID-19 which is in line with the report of risk factors for severe COVID-19, with the strongest associations found for age, obesity, and diabetes [14,18].
The presence of diabetes mellitus and the individual degree of hyperglycemia seems to be independently associated with COVID-19 severity and increased mortality [22-24]. The present study also confirms this finding. Patients with poor glycaemic control were found to have higher chest CTSS indicating extensive lung involvement and subsequently higher requirement for ventilatory support. The association of glycaemic control and disease severity remained significant even after adjustment for age and body weight. However HbA1c level ≥$7.5\%$ was a better predictor for a worse prognosis than admission plasma glucose in terms of length of stay in the hospital and need for supportive therapies like ICU care and need of mechanical ventilation.
Last, the primary endpoint of our study was in-hospital death. The overall mortality in our study was $10.17\%$. However, no significant difference was observed in mortality rates between the well-controlled and poorly controlled patients. Raoufi M et al. collected clinical characteristics of 117 patients with coexistent COVID-19 and diabetes using HbA1c as an index of glucose management and reported similar findings [25]. Li Y et al. included 132 patients with COVID-19 and diabetes and suggested patients with admission glucose >11mmol/L had an increased risk of mortality and in-hospital complications [26].
Diabetes mellitus has already been a leading cause of morbidity worldwide and is capable of affecting almost every system of the body. Consequently, a dysregulated immune system might develop; predisposing to various infections in patients with T2DM [27].
Although this study generated a significant suggestion for glucose management for COVID-19 patients with type 2 diabetes, several limitations should be addressed. First, the study was conducted during a pandemic setting when the simultaneous influx of a large number of patients jeopardized the healthcare system. Since conducting a randomized controlled in this situation could have been unethical hence the retrospective study was carried out. Our study protocol was designed for descriptive analysis and it was beyond the scope of this assessment to quantify the magnitude of the association between diabetes and glycaemic control and treatment outcomes. Second, as hospitalized patients were only included in the study these results cannot be directly extrapolated to patients with milder diseases. Third, our dataset did not include recognized comorbidities for death from COVID-19, such as hypertension or cardiovascular disease. Fourth, given the retrospective nature of the study, it was not possible for us to determine if active management of blood glucose levels to a more normal range could ameliorate COVID-19 severity or adverse outcomes. Another limitation of this study includes the lack of differentiation between the duration of the disease. Lastly, data collected from a single center and thus not very large in number exposes the statistical analysis to selection bias and the results may be less efficacious from a statistical standpoint. Therefore, large-scale prospective cohort studies will be required in ethnically and geographically diverse cohorts to better understand the association and importance of glycaemic control in the disease progression of COVID-19.
## Conclusions
The present study showed that neither HbA1c nor admission glucose values could predict mortality in patients with T2DM suffering from COVID-19 but within the same range of HbA1c, those with admission glucose less than 200mg/dL had numerical supremacy in terms of less mortality which failed to reach statistical significance. CTSS and hence lung involvement was significantly worse in those with high HbA1c and high admission glucose in comparison to those having controlled HbA1c and admission glucose values. High HbA1c was significantly associated with a longer duration of hospital stay and ICU stays, however, amongst those with high HbA1c higher admission glucose significantly increased the duration of ICU stay. Keeping in mind the high prevalence of diabetes in India, it predisposes a large proportion of the population to COVID-19 and its complications. Our results provided a simple and practical way to risk stratify COVID-19 in-patients with T2DM for hierarchical management, particularly where medical resources are in severe shortage during the pandemic provided a larger number of studies in future revalidate the findings.
## References
1. **World Health Organization. WHO Coronavirus (COVID-19) Dashboard. [Online]**. (2022)
2. Yang X, Yu Y, Xu J. **Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study**. *Lancet Respir Med* (2020) **8** 475-481. PMID: 32105632
3. Zhang JJ, Dong X, Cao YY. **Clinical characteristics of 140 patients infected with SARS-CoV-2 in Wuhan, China**. *Allergy* (2020) **75** 1730-1741. PMID: 32077115
4. Guan WJ, Ni ZY, Hu Y. **Clinical characteristics of coronavirus disease 2019 in China**. *N Engl J Med* (2020) **382** 1708-1720. PMID: 32109013
5. Huang C, Wang Y, Li X. **Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China**. *Lancet* (2020) **395** 497-506. PMID: 31986264
6. Fang L, Karakiulakis G, Roth M. **Are patients with hypertension and diabetes mellitus at increased risk for COVID-19 infection?**. *Lancet Respir Med* (2020) **8** 0
7. Zhu L, She ZG, Cheng X. **Association of blood glucose control and outcomes in patients with COVID-19 and pre-existing type 2 diabetes**. *Cell Metab* (2020) **31** 1068-1077. PMID: 32369736
8. **Covid-19 and long term conditions: what if you have cancer, diabetes, or chronic kidney disease?**. *BMJ* (2020) **368** 0
9. Arachchillage DRJ, Laffan M. **Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia**. *J Thromb Haemost* (2020) **18** 1233-1234. PMID: 32291954
10. Moghissi ES, Korytkowski MT, DiNardo M. **American Association of Clinical Endocrinologists and American Diabetes Association consensus statement on inpatient glycemic control**. *Diabetes Care* (2009) **32** 1119-1131. PMID: 19429873
11. Wang S, Ma P, Zhang S. **Fasting blood glucose at admission is an independent predictor for 28-day mortality in patients with COVID-19 without previous diagnosis of diabetes: a multi-centre retrospective study**. *Diabetologia* (2020) **63** 2102-2111. PMID: 32647915
12. Sardu C, D'Onofrio N, Balestrieri ML. **Outcomes in patients with hyperglycemia affected by COVID-19: can we do more on glycemic control?**. *Diabetes Care* (2020) **43** 1408-1415. PMID: 32430456
13. **Management Protocol for COVID-19**. (2023)
14. Bode B, Garrett V, Messler J, McFarland R, Crowe J, Booth R, Klonoff DC. **Glycemic characteristics and clinical outcomes of COVID-19 patients hospitalized in the United States**. *J Diabetes Sci Technol* (2020) **14** 813-821. PMID: 32389027
15. van Vught LA, Wiewel MA, Klein Klouwenberg PMC. **Admission hyperglycemia in critically ill sepsis patients: association with outcome and host response**. *Crit Care Med* (2016) **44** 1338-1346. PMID: 26958752
16. Robba C, Bilotta F. **Admission hyperglycemia and outcome in ICU patients with sepsis**. *J Thorac Dis* (2016) **8** 0-3
17. McGuire H, Longson D, Adler A, Farmer A, Lewin I. **Management of type 2 diabetes in adults: summary of updated NICE guidance**. *BMJ* (2016) **353** 0
18. Shauly-Aharonov M, Shafrir A, Paltiel O. **Both high and low pre-infection glucose levels associated with increased risk for severe COVID-19: new insights from a population-based study**. *PLoS One* (2021) **16** 0
19. Wu J, Huang J, Zhu G. **Elevation of blood glucose level predicts worse outcomes in hospitalized patients with COVID-19: a retrospective cohort study**. *BMJ Open Diabetes Res Care* (2020) **8** 0
20. Duan W, Li L, Li X. **Association of blood glucose level and prognosis of inpatients with coexistent diabetes and COVID-19**. *Endocrine* (2022) **75** 1-9. PMID: 34716852
21. Bhandari S, Rankawat G, Singh A, Gupta V, Kakkar S. **Impact of glycemic control in diabetes mellitus on management of COVID-19 infection**. *Int J Diabetes Dev Ctries* (2020) **40** 340-345. PMID: 32905072
22. Holman N, Knighton P, Kar P. **Risk factors for COVID-19-related mortality in people with type 1 and type 2 diabetes in England: a population-based cohort study**. *Lancet Diabetes Endocrinol* (2020) **8** 823-833. PMID: 32798471
23. Grasselli G, Zangrillo A, Zanella A. **Baseline characteristics and outcomes of 1591 patients infected with SARS-CoV-2 admitted to ICUs of the Lombardy region, Italy**. *JAMA* (2020) **323** 1574-1581. PMID: 32250385
24. Yang J, Zheng Y, Gou X. **Prevalence of comorbidities and its effects in patients infected with SARS-CoV-2: a systematic review and meta-analysis**. *Int J Infect Dis* (2020) **94** 91-95. PMID: 32173574
25. Raoufi M, Khalili S, Mansouri M, Mahdavi A, Khalili N. **Well-controlled vs poorly-controlled diabetes in patients with COVID-19: are there any differences in outcomes and imaging findings?**. *Diabetes Res Clin Pract* (2020) **166** 108286. PMID: 32592836
26. Li Y, Han X, Alwalid O. **Baseline characteristics and risk factors for short-term outcomes in 132 COVID-19 patients with diabetes in Wuhan China: a retrospective study**. *Diabetes Res Clin Pract* (2020) **166** 108299. PMID: 32623030
27. Knapp S. **Diabetes and infection: is there a link? - a mini-review**. *Gerontology* (2013) **59** 99-104. PMID: 23182884
|
---
title: Electrocardiographic Findings in Children With Growth Hormone Deficiency
journal: Cureus
year: 2023
pmcid: PMC10030162
doi: 10.7759/cureus.36385
license: CC BY 3.0
---
# Electrocardiographic Findings in Children With Growth Hormone Deficiency
## Abstract
Introduction It has been shown that cardiac functions begin to deteriorate in growth hormone (GH) deficiency even in childhood. However, little is known about how GH deficiency affects arrhythmogenesis. The aim of this study was to evaluate the parameters of P wave dispersion (Pd), QT dispersion (QTd), corrected QT (QTc) dispersion (QTcd), T wave peak-to-end (Tp-e) interval, Tp-e/QT ratio, and Tp-e/QTc ratio in children with GH deficiency. This study also aimed to evaluate the relationship of these parameters with insulin-like growth factor 1 (IGF-1) and insulin-like growth factor binding protein 3 (IGFBP-3).
Method In the study, records of children diagnosed with GH deficiency in Adana City Training and Research Hospital Pediatric Endocrine Outpatient Clinic between September 2021 and December 2022 were retrospectively reviewed. The control group consisted of children in the same age group who applied to the Emergency Outpatient Clinic with a complaint of chest pain and no pathological finding was detected. The electrocardiograms (ECGs) of all patients were retrospectively evaluated.
Results There were a total of 82 children in the study, 41 of whom were diagnosed with GH deficiency and 41 in the healthy control group. The age and male/female ratio of children with GH deficiency were similar to those in the control group ($p \leq 0.05$). There were 27 ($66\%$) children with complete GH deficiency and 14 ($34\%$) children with partial GH deficiency. P wave dispersion was similar in both GH-deficient children and control group children. It was also similar in children with complete and partial GH deficiency ($p \leq 0.05$). QT and QTc dispersions were found to be increased in children with GH deficiency, although not statistically significant, compared to the control group ($p \leq 0.05$). Tp-e interval, Tp-e/QTmax (longest QT interval), and Tp-e/QTcmax (longest QTc interval) ratios were increased in children with GH deficiency compared to the control group ($$p \leq 0.001$$, $$p \leq 0.003$$, and $$p \leq 0.001$$, respectively). QT and QTc dispersion, Tp-e interval, Tp-e/QTmax, and Tp-e/QTcmax ratios were found to be increased in children with complete GH deficiency compared to children with partial GH deficiency, but the difference was not significant ($p \leq 0.05$). No correlation was found between these ECG parameters and IGF-1, IGFBP-3, and peak GH levels after stimulation tests ($p \leq 0.05$).
Conclusion We found in our study that the Tp-e interval was longer and Tp-e/QT and Tp-e/QTc ratios were increased in children with GH deficiency. These results suggest that the risk of ventricular arrhythmias in children with GH deficiency may start to increase from childhood. However, further prospective studies are needed to confirm our results.
## Introduction
Growth hormone (GH) has effects on many systems as well as stimulating growth in children. In addition to cardiac growth in the cardiovascular system, it plays a role in the regulation of cardiac structure and function [1]. Both GH and insulin-like growth factor 1 (IGF-1) receptors are expressed in the myocardium, and IGF-1 has been shown to stimulate myofibril development and increase isometric strength [2,3]. Less than $1\%$ of total serum IGF-1 is freely circulating, most of it is bound to insulin-like growth factor binding proteins (IGFBPs). IGFBPs have IGF-dependent and IGF-independent functions [4,5]. Of the six different circulating IGFBPs, IGFBP-3 is the most abundant, and it has been shown that low IGFBP-3 levels are associated with an increased risk of cardiovascular disease [4,6].
Growth hormone deficiency is characterized by significant growth retardation, slowing of growth rate, retardation in bone age, and low GH level despite spontaneous and pharmacological stimulation, when there is no other reason to explain the short stature [7,8]. It is known that GH deficiency affects sympathetic activity and has negative effects on the structure and function of the cardiovascular system. GH deficiency can cause decreased left ventricular mass and ejection fraction, resulting in decreased cardiac performance in adults. It has been shown that cardiac functions are affected even in childhood in GH deficiency and begin to improve after 12 months of GH treatment [1,7,9]. However, little is known about how GH deficiency affects arrhythmogenesis [1,10].
In experimental studies, it has been reported that GH exerts cytoprotective effects after coronary artery occlusion and reduces local myocardial norepinephrine release and arrhythmogenesis [11]. In studies with adults, heart rate variability was evaluated in patients with GH deficiency, and in studies using both frequency domain and time domain parameters, impaired response to sympathetic activation has been demonstrated [12,13]. There is only one study evaluating heart rate variability in children with GH deficiency, and global heart rate variability was found to be reduced in this study [14].
A surface electrocardiogram (ECG) is an easily accessible test that provides a rapid assessment of cardiac electrophysiology and can be used to assess arrhythmic risk [15]. In the assessment of arrhythmia risk, QT dispersion (QTd) for myocardial conduction disturbances, heart rate corrected QT (QTc) dispersion (QTcd), and P wave dispersion (Pd) for atrial conduction variability are used. In addition, in recent years, the interval from the peak to the end of the T wave (Tp-e interval), Tp-e/QT ratio, and Tp-e/QTc ratio have also been defined as predictive electrocardiographic markers for ventricular arrhythmias [16-20]. There are only two studies in the literature evaluating ECG in children with GH deficiency. In the study of Alkan et al. [ 21], in which they evaluated ECG parameters in children with complete and partial GH deficiency and in the control group, changes were found in the P wave and QTc dispersion on the ECG between the groups. On the other hand, Nygren et al. [ 22] showed that the QTc interval did not change before and after GH treatment.
The aim of this study was to evaluate the parameters of P wave dispersion, QT dispersion, QTc dispersion, Tp-e interval, Tp-e/QT ratio, and Tp-e/QTc ratio in children with GH deficiency. In addition, the study is also aimed at evaluating the relationship of these parameters with IGF-1 and IGFBP-3.
## Materials and methods
In the study, records of children diagnosed with GH deficiency in Adana City Training and Research Hospital Pediatric Endocrine Outpatient Clinic between September 2021 and December 2022 were retrospectively reviewed. GH deficiency was diagnosed according to clinical and auxology criteria. The diagnosis of GH deficiency was considered in a child with short stature (height below 3 standard deviation (SD) or less than 2 SD and annual growth rate below 2 SD) after excluding other causes of growth retardation (hypothyroidism, chronic systemic disease, Turner syndrome, or skeletal system disorders). GH provocation tests (L-dopa and insulin tolerance test) were performed. A peak GH value of <10 µg/L after two separate GH provocation tests was defined as GH deficiency [8]. Children with GH deficiency were grouped after two provocation tests as children with a peak GH level of <7 µg/L (complete GH deficiency) and those with a peak GH level of 7-10 µg/L (partial GH deficiency). Patients diagnosed with GH deficiency and whose ECG was taken after provocation tests were included in the study. Those with pituitary hormone deficiency other than GH and those with pathological findings in pituitary magnetic resonance imaging were excluded from the study. Children in the same age group, who presented to the Pediatric Emergency Outpatient Clinic with the complaint of chest pain, whose ECG was recorded and no pathological cause was found, were included in the study. This group consisted of healthy children with idiopathic chest pain whose history, physical examination, laboratory, and imaging revealed no pathology. First-degree relatives who would cause changes in ECG parameters in the family and those with a history of drug use that could cause changes in ECG parameters were excluded from the study, both in the GH deficiency group and in the control group. The ECG recordings of the patients in the control group and GH deficiency group were reviewed retrospectively by the same pediatric cardiologist.
ECG recordings were made at a rate of 25 mm/second, with an amplitude of 1 mV, and with 12 standard leads containing at least six QRS complexes for each lead. Measurements in all leads were performed manually. The difference between the longest P wave (Pmax) and the shortest P wave (Pmin) was considered as the P dispersion (Pd=Pmax-Pmin). The time between the onset of the QRS complex and the point where the descending branch of the T wave cut the isoelectric segment was taken as the QT interval. Leads in which T wave could not be detected were excluded. QT dispersion was defined as the difference between the longest (QTmax) and the shortest (QTmin) QT interval (QTd=QTmax-QTmin). Bazett’s formula (QTc=QT/√RR) was used for QTc. QTc dispersion was similarly determined as the difference between the longest QTc (QTcmax) interval and the shortest QTc (QTcmin) interval. The Tp-e interval was defined as the interval from the peak of the T wave to the end of the T wave. Tp-e interval measurements were made from lead V5. In addition, Tp-e/QTmax and Tp-e/QTcmax ratios were calculated.
Approval was obtained from the Non-Invasive Clinical Research Ethics Committee of Pamukkale University ($\frac{10.01.2023}{01}$) before conducting the study.
The Statistical Package for the Social Sciences (SPSS) for Windows version 20 (IBM SPSS Statistics, Armonk, NY, USA) was used for the statistical evaluation of data. Continuous variables were given as mean±standard deviation (minimum-maximum), and categorical variables were given as number (percentage). Kolmogorov-Smirnov and Shapiro-Wilk tests were used to determine the normal distribution. In the comparison of independent data, Student’s t-test was used for parametric data and the chi-square test was used for the comparison of categorical variables. The correlation of continuous variables was evaluated using Pearson correlation analysis, and $p \leq 0.05$ was accepted as statistically significant.
## Results
A total of 82 children were included in the study, 41 of whom were diagnosed with GH deficiency and 41 in the healthy control group. When demographic findings are evaluated, the age and male/female ratio of the children with GH deficiency were similar to the children in the control group ($p \leq 0.05$) (Table 1). The mean height, body weight, and body mass index were 123±18.4 (86.5-153.8) cm, 26.1±9.7 (11.2-59) kg, and 16.6±2.2 (12.8-24.9) kg/m2, respectively, in children with GH deficiency. On the other hand, these values were 140.5±21.7 [100-184] cm, 37.9±16.7 [15-74] kg, and 17.9±3.5 (13.2-25.51) kg/m2, respectively, in the control group. The height and body weight of children with GH deficiency were significantly lower than the control group ($$p \leq 0.001$$), and the body mass index was at the low threshold ($$p \leq 0.05$$). While height standard deviation score (SDS) and body weight SDS were significantly lower in children with GH deficiency ($p \leq 0.001$), no significant difference was found in body mass index SDS ($$p \leq 0.074$$) (Table 1). The peak GH levels of children with GH deficiency after IGF-1, IGFBP-3, and provocation tests are shown in Table 1.
**Table 1**
| Mean±SD (min-max) | Patients with growth hormone deficiency (n=41) | Control group (n=41) | p value |
| --- | --- | --- | --- |
| Sex (male/female) | 26/15 | 27/14 | 0.817 |
| Age (years) | 9.75±3.2 (4-15) | 9.78±3.2 (4-15) | 0.972 |
| Height SDS | -2.67±0.86 (-5.77-(-2)) | 0.29±1.2 (-1.9-1.92) | 0.001 |
| Weight SDS | -1.89±0.85 (-3.31-(-0.6)) | 0.13±1.1 (1.8-1.94) | 0.001 |
| Body mass index SDS | -0.56±1 (-2.63-1.13) | -0.11±1.14 (-1.8-1.57) | 0.074 |
| IGF-1 (µg/mL) | 116.8±64 (20.3-334) | - | |
| IGFBP-3 (ng/mL) | 20.3±44.9 (1.86-140) | - | |
| PGH with L-dopa stimulation test (µg/L) | 5.2±3 (0.1-9.9) | - | |
| PGH with insulin stimulation test (µg/L) | 3.1±2.6 (0.1-9.9) | - | |
| HR (bpm) | 84.7±16.4 (56-115) | 92.1±21.4 (58-154) | 0.082 |
| Pmax (ms) | 80.8±14.6 (44-120) | 84.1±8.6 (66-110) | 0.208 |
| Pmin (ms) | 59.5±14.4 (36-90) | 63.2±9.1 (42-90) | 0.161 |
| Pd (ms) | 21.3±7.3 (8-40) | 20.9±4.5 (12-30) | 0.798 |
| QTmax (ms) | 344.8±32.5 (280-420) | 342.1±32.5 (275-400) | 0.71 |
| QTmin (ms) | 314.6±36.5 (240-380) | 313.4±30.7 (240-360) | 0.87 |
| QTd (ms) | 30.1±12.7 (10-60) | 28.7±9.7 (10-40) | 0.559 |
| QTcmax (ms) | 408.2±13 (360-425) | 410±13.7 (365-423) | 0.474 |
| QTcmin (ms) | 360±21.3 (329-406) | 369±24.5 (320-423) | 0.065 |
| QTcd (ms) | 48.3±21.5 (5-88) | 41±24.5 (5-90) | 0.153 |
| Tp-e (ms) | 71.8±10.3 (52-90) | 64.7±8.3 (48-80) | 0.001 |
| Tp-e/QTmax | 0.21±0.03 (0.15-0.28) | 0.19±0.02 (0.15-0.25) | 0.003 |
| Tp-e/QTcmax | 0.18±0.02 (0.13-0.22) | 0.16±0.02 (0.12-0.20) | 0.001 |
When ECG parameters are compared, no significant difference was found in heart rates between children with GH deficiency and in the control group ($p \leq 0.05$). The maximum and minimum P wave intervals and P dispersion were similar in children with GH deficiency and in the control group ($p \leq 0.05$) (Table 1). There was no significant difference between the groups in terms of maximum and minimum QT and QTc durations ($p \leq 0.05$). Both QT and QTc dispersions were increased in children with GH deficiency compared to the control group, but the difference was not statistically significant ($p \leq 0.05$) (Table 1). When we evaluate the Tp-e interval, both Tp-e/QTmax and Tp-e/QTcmax ratios with Tp-e interval were found to be significantly increased in children with GH deficiency compared to the control group ($$p \leq 0.001$$, $$p \leq 0.003$$, and $$p \leq 0.001$$, respectively).
When the children with GH deficiency were divided into two groups according to the results of the provocation tests, there were 27 ($66\%$) children with complete GH deficiency and 14 ($34\%$) children with partial GH deficiency. There was no significant difference between children with complete and partial GH deficiency in terms of age, gender, height SDS, body weight SDS, and body mass index SDS ($p \leq 0.05$), and heart rates were also found to be similar ($p \leq 0.05$) (Table 2). There was no significant difference between children with complete and partial GH deficiency in terms of maximum P wave duration, minimum P wave duration, P wave dispersion, and maximum and minimum QT and QTc intervals ($p \leq 0.05$). QT dispersion, QTc dispersion, Tp-e interval, Tp-e/QTmax ratio, and Tp-e/QTcmax ratio were increased in children with complete GH deficiency compared to children with partial GH deficiency. However, this difference between groups was not statistically significant ($p \leq 0.05$) (Table 2). In addition, P wave, QT and QTc dispersion, Tp-e interval, Tp-e/QTmax, and Tp-e/QTcmax ratios were not significantly related to peak GH levels after IGF-1, IGFBP, and GH peak GH levels after stimulation tests ($p \leq 0.05$) (Table 3).
## Discussion
In this study, we evaluated ECG findings in children with GH deficiency, and we found that the Tp-e interval, Tp-e/QTmax, and Tp-e/QTcmax ratios were increased in children with GH deficiency compared to the healthy control group.
In adults, both GH excess and deficiency have been associated with an increased risk of cardiac arrhythmias, including atrial fibrillation (AF), and structural remodeling of the left atrium has been demonstrated that may mediate the increased risk of AF in patients with GH deficiency [1,23,24]. Pd is a strong marker of the heterogeneous spread of anatomical remodeling and activation in the atrium. It has been suggested that Pd, believed to reflect non-homogeneous atrial conduction, is useful in determining the risk of supraventricular arrhythmias, particularly paroxysmal AF [18,19,25]. In the study of Alkan et al. [ 21], in which they evaluated 47 children with GH deficiency (30 complete GH deficiency and 17 partial GH deficiency), the P wave interval was found to be shorter in children with partial GH deficiency. On the other hand, Pd was found to be similar in children with complete and partial GH deficiency and in the healthy control group. In our study, there was no difference in terms of Pd in children with GH deficiency and in the healthy control group. In addition, Pd was similar in children with complete and partial GH deficiency. No correlation was found between Pd and peak GH levels after IGF-1, IGFBP-3, and provocation tests.
Experimental studies have shown that IGF-1 regulates sarcolemmal potassium channel activity and late sodium current in rat cardiomyocytes, thereby affecting cardiac repolarization and QTc [26,27]. In addition, in a sample of elderly people aged 60-64 years, decreased levels of IGF-1 were associated with prolonged QTc interval [4]. Long QT and QTc intervals and increased QT and QTc dispersions are well-known risk factors for all-cause mortality and morbidity [15,28]. In the study of Alkan et al. [ 21], although QTc and QT dispersion were longer in children with GH deficiency compared to the control group, no statistically significant difference was found. Nygren et al. [ 22], in their study evaluating 89 children with GH deficiency, reported that the QTc interval did not change before and after GH treatment. In our study, QT and QTc dispersions were increased in children with GH deficiency and in the control group, although not statistically significant. When patients with complete and partial GH deficiency are evaluated, QT and QTc dispersions were again increased in children with complete GH deficiency, although not statistically significant. QTd and QTcd were found to be not correlated with the peak level of GH after IGF-1, IGFBP-3, and provocation tests.
IGF-1 is thought to protect cardiac myocytes from arrhythmogenesis and apoptosis by activating PI3K/Akt intracellular signal transduction [26]. Changes in T wave shape or duration reflect the heterogeneity of ventricular repolarization, and prolonged Tp-e interval on ECG indicates repolarization heterogeneity [17]. It is argued that an abnormally prolonged Tp-e interval on ECG is a risk factor for ventricular arrhythmic mortality and all-cause mortality, independent of age, gender, comorbidities, QRS interval, and corrected QT interval [16]. It has been suggested that the Tp-e/QTc ratio is a better predictor of ventricular repolarization because of the interaction between QT and heart rate and body weight [17,20]. Tp-e interval and Tp-e/QT ratios have been investigated in many diseases that affect the cardiovascular system, such as diabetes and hypothyroidism, and have been reported to be risk factors for ventricular arrhythmias [29,30]. Alkan et al. [ 21] found a longer Tp-e interval in children with complete GH deficiency compared to children with partial GH deficiency and the control group. In the same study, although the Tp-e/QTc ratio was higher in children with complete GH deficiency, no statistically significant difference was found. In our study, Tp-e interval, Tp-e/QTmax, and Tp-e/QTcmax ratios were increased in children with GH deficiency compared to the control group. Although the same parameters were found to be increased in children with complete GH deficiency compared to children with partial GH deficiency, the difference was not significant. No correlation was found between these ECG parameters and peak GH levels after IGF-1, IGFBP-3, and peak GH levels after stimulation tests. The increase in Tp-e interval, Tp-e/QTmax, and Tp-e/QTcmax ratios suggests that there may be an increased risk of arrhythmia in children with GH deficiency. Therefore, we think that ECG monitoring is important in these children.
The main limitation of this study is that it is a retrospective study with a relatively small sample size. Other limitations are the lack of echocardiographic evaluation and evaluation of children with GH deficiency after treatment. Therefore, further prospective studies with larger populations are needed to confirm our results.
## Conclusions
We found in our study that the Tp-e interval is longer and the Tp-e/QT and Tp-e/QTc ratios are increased in children with GH deficiency. These results suggest that the risk of ventricular arrhythmias in children with GH deficiency may start to increase from childhood. However, further prospective studies are needed to confirm our results.
## References
1. Climent VE, Picó A, Sogorb F, Aznar S, Lip GY, Marín F. **Growth hormone therapy and the heart**. *Am J Cardiol* (2006) **97** 1097-1102. PMID: 16563925
2. Minczykowski A, Gryczynska M, Ziemnicka K, Czepczynski R, Sowinski J, Wysocki H. **The influence of growth hormone (GH) therapy on cardiac performance in patients with childhood onset GH deficiency**. *Growth Horm IGF Res* (2005) **15** 156-164. PMID: 15809020
3. Guse AH, Kiess W, Funk B, Kessler U, Berg I, Gercken G. **Identification and characterization of insulin-like growth factor receptors on adult rat cardiac myocytes: linkage to inositol 1,4,5-trisphosphate formation**. *Endocrinology* (1992) **130** 145-151. PMID: 1309323
4. Charalambous C, Moon JC, Holly JM, Chaturvedi N, Hughes AD, Captur G. **Declining levels and bioavailability of IGF-I in cardiovascular aging associate with QT prolongation-results from the 1946 British birth cohort**. *Front Cardiovasc Med* (2022) **9** 863988. PMID: 35528832
5. Ranke MB. **Insulin-like growth factor binding-protein-3 (IGFBP-3)**. *Best Pract Res Clin Endocrinol Metab* (2015) **29** 701-711. PMID: 26522455
6. Kaplan RC, McGinn AP, Pollak MN. **Association of total insulin-like growth factor-I, insulin-like growth factor binding protein-1 (IGFBP-1), and IGFBP-3 levels with incident coronary events and ischemic stroke**. *J Clin Endocrinol Metab* (2007) **92** 1319-1325. PMID: 17264182
7. Alkan F, Ersoy B, Kızılay DO, Coskun S. **Cardiac functions in children with growth hormone deficiency: effects of one year of GH replacement therapy**. *Growth Horm IGF Res* (2021) **60-61** 101432. PMID: 34715474
8. Growth Hormone Research Society. **Consensus guidelines for the diagnosis and treatment of growth hormone (GH) deficiency in childhood and adolescence: summary statement of the GH Research Society. GH Research Society**. *J Clin Endocrinol Metab* (2000) **85** 3990-3993. PMID: 11095419
9. Zhang S, Li Z, Lv Y, Sun L, Xiao X, Gang X, Wang G. **Cardiovascular effects of growth hormone (GH) treatment on GH-deficient adults: a meta-analysis update**. *Pituitary* (2020) **23** 467-475. PMID: 32166618
10. Isgaard J, Arcopinto M, Karason K, Cittadini A. **GH and the cardiovascular system: an update on a topic at heart**. *Endocrine* (2015) **48** 25-35. PMID: 24972804
11. Stamatis KV, Kontonika M, Daskalopoulos EP, Kolettis TM. **Electrophysiologic effects of growth hormone post-myocardial infarction**. *Int J Mol Sci* (2020) **21**
12. Leong KS, Mann P, Wallymahmed M, MacFarlane IA, Wilding JP. **The influence of growth hormone replacement on heart rate variability in adults with growth hormone deficiency**. *Clin Endocrinol (Oxf)* (2001) **54** 819-826. PMID: 11422118
13. Tanriverdi F, Eryol NK, Atmaca H. **The effects of 12 months of growth hormone replacement therapy on cardiac autonomic tone in adults with growth hormone deficiency**. *Clin Endocrinol (Oxf)* (2005) **62** 706-712. PMID: 15943833
14. Buchhorn R, Willaschek C. **Growth hormone treatment, cardiovascular risk and autonomic maturation in children and adolescents with growth hormone deficiency or born small for gestational age**. *Open J Pediatr* (2020) **10** 12-29
15. O'Neal WT, Singleton MJ, Roberts JD. **Association between QT-interval components and sudden cardiac death: the ARIC study (Atherosclerosis Risk in Communities)**. *Circ Arrhythm Electrophysiol* (2017) **10**
16. Ramírez J, van Duijvenboden S, Young WJ, Orini M, Lambiase PD, Munroe PB, Tinker A. **Common genetic variants modulate the electrocardiographic Tpeak-to-Tend interval**. *Am J Hum Genet* (2020) **106** 764-778. PMID: 32386560
17. Arteyeva NV, Goshka SL, Sedova KA, Bernikova OG, Azarov JE. **What does the T(peak)-T(end) interval reflect? An experimental and model study**. *J Electrocardiol* (2013) **46** 296-298
18. Okutucu S, Aytemir K, Oto A. **P-wave dispersion: what we know till now?**. *JRSM Cardiovasc Dis* (2016) **5** 2048004016639443. PMID: 27081484
19. Pérez-Riera AR, de Abreu LC, Barbosa-Barros R, Grindler J, Fernandes-Cardoso A, Baranchuk A. **P-wave dispersion: an update**. *Indian Pacing Electrophysiol J* (2016) **16** 126-133. PMID: 27924760
20. Gupta P, Patel C, Patel H, Narayanaswamy S, Malhotra B, Green JT, Yan GX. **T(p-e)/QT ratio as an index of arrhythmogenesis**. *J Electrocardiol* (2008) **41** 567-574. PMID: 18790499
21. Alkan F, Ersoy B, Kızılay DO, Ozyurt BC, Coskun S. **Evaluation of cardiac structure, exercise capacity and electrocardiography parameters in children with partial and complete growth hormone deficiency and their changes with short term growth hormone replacement therapy**. *Pituitary* (2023) **26** 115-123. PMID: 36463549
22. Nygren A, Sunnegårdh J, Teien D. **Rapid cardiovascular effects of growth hormone treatment in short prepubertal children: impact of treatment duration**. *Clin Endocrinol (Oxf)* (2012) **77** 877-884. PMID: 22651572
23. Higham CE, Johannsson G, Shalet SM. **Hypopituitarism**. *Lancet* (2016) **388** 2403-2415. PMID: 27041067
24. Aguilar M, Rose RA, Takawale A, Nattel S, Reilly S. **New aspects of endocrine control of atrial fibrillation and possibilities for clinical translation**. *Cardiovasc Res* (2021) **117** 1645-1661. PMID: 33723575
25. Aytemir K, Ozer N, Atalar E. **P wave dispersion on 12-lead electrocardiography in patients with paroxysmal atrial fibrillation**. *Pacing Clin Electrophysiol* (2000) **23** 1109-1112. PMID: 10914366
26. Millis RM, Alvin ZV, Zhao A, Haddad GE. **Effects of IGF-1 on I(K) and I(K1) channels via PI3K/Akt signaling in neonatal cardiac myocytes**. *Int J Cell Biol* (2012) **2012** 712153. PMID: 22761619
27. Ballou LM, Lin RZ, Cohen IS. **Control of cardiac repolarization by phosphoinositide 3-kinase signaling to ion channels**. *Circ Res* (2015) **116** 127-137. PMID: 25552692
28. Okin PM, Devereux RB, Howard BV, Fabsitz RR, Lee ET, Welty TK. **Assessment of QT interval and QT dispersion for prediction of all-cause and cardiovascular mortality in American Indians: the Strong Heart study**. *Circulation* (2000) **101** 61-66. PMID: 10618305
29. Gürdal A, Eroğlu H, Helvaci F. **Evaluation of Tp-e interval, Tp-e/QT ratio and Tp-e/QTc ratio in patients with subclinical hypothyroidism**. *Ther Adv Endocrinol Metab* (2017) **8** 25-32. PMID: 28377800
30. Tokatli A, Kiliçaslan F, Alis M, Yiginer O, Uzun M. **Prolonged Tp-e interval, Tp-e/QT ratio and Tp-e/QTc ratio in patients with type 2 diabetes mellitus**. *Endocrinol Metab (Seoul)* (2016) **31** 105-112. PMID: 26676332
|
---
title: Circulating Adiponectin Levels Are Inversely Associated with Mortality and
Respiratory Failure in Patients Hospitalized with COVID-19
authors:
- Bettina Hindsberger
- Birgitte Lindegaard
- Liv Rabøl Andersen
- Simone Bastrup Israelsen
- Lise Pedersen
- Pal Bela Szecsi
- Thomas Benfield
journal: International Journal of Endocrinology
year: 2023
pmcid: PMC10030212
doi: 10.1155/2023/4427873
license: CC BY 4.0
---
# Circulating Adiponectin Levels Are Inversely Associated with Mortality and Respiratory Failure in Patients Hospitalized with COVID-19
## Abstract
### Background
Chronic low-grade inflammation associated with a dysregulated adipose tissue might contribute to amplifying the inflammatory response in severe COVID-19. The aim of this study was to examine the association between levels of circulating leptin and adiponectin and the severity and mortality of COVID-19.
### Methods
Serum levels of leptin and adiponectin were determined at admission in 123 individuals with confirmed COVID-19 and their association with 90-day mortality and respiratory failure was analyzed by logistic regression analysis and expressed as odds ratios (ORs) with $95\%$ confidence intervals (CIs).
### Results
The median values of circulating leptin and adiponectin were 7.2 ng/mL (IQR 3.8–13.4) and 9.0 μg/mL (IQR 5.7–14.6), respectively. After adjustment for age, sex, body mass index, hypertension, diabetes, chronic obstructive pulmonary disease, and oxygen saturation at admission, a doubling of circulating adiponectin was associated with a $38\%$ reduction in odds of 90-day mortality (OR 0.62, CI 0.43–0.89) and a $40\%$ reduction in odds of respiratory failure (OR 0.60, CI 0.42–0.86). The association tended to be strongest in individuals below the median age of 72 years. Circulating leptin was not associated with outcomes.
### Conclusions
Circulating adiponectin at admission was inversely associated with mortality and respiratory failure in SARS-CoV-2 infection. Further studies are needed to elucidate how exactly adipokines, especially adiponectin, are linked to the progression and prognosis of COVID-19.
## 1. Background
While most cases of COVID-19 are mild, some individuals suffer from more severe forms of the disease. They are at risk of developing pneumonia, acute respiratory distress syndrome (ARDS), and viral sepsis, which may require intensive care and mechanical ventilation [1]. The mechanisms underlying the pathogenesis of severe COVID-19 are still not fully understood, but hyperinflammation plays an important part [2]. Furthermore, visceral adiposity, obesity, and obesity-associated conditions such as diabetes and cardiovascular disease have been found to increase the risk of severe disease and mortality in COVID-19 [3, 4]. It has been suggested that the chronic low-grade inflammation associated with a dysregulated adipose tissue might contribute by amplifying the inflammatory response [5].
In the past decades, it has become evident that adipose tissue is not merely an energy store, but rather a physiologically active tissue involved in the regulation of both endocrine and immune processes. These functions are mediated by the production of soluble factors termed adipokines, of which leptin and adiponectin are the most abundant [6]. Leptin functions as a key regulator of body weight and appetite, and circulating leptin levels are proportional to the amount of white adipose tissue [7]. It also exerts proinflammatory effects, and thus high leptin levels have been linked to autoimmune diseases, while low levels might increase susceptibility to infections [8]. Adiponectin regulates glucose and lipid metabolism and exhibits anti-inflammatory features. Low levels have been reported in unhealthy obesity, diabetes, coronary heart disease, and nonalcoholic fatty liver disease [9–12]. The role of these adipokines in acute illness is incompletely understood, but they might provide a link between dysfunctional adipose tissue and the exaggerated inflammatory response seen in severe COVID-19. Hence, this study aims to examine the association between circulating leptin and adiponectin and the severity and mortality of SARS-CoV-2 infection in a cohort of patients hospitalized with COVID-19.
## 2.1. Study Design
This retrospective study included adults aged 18 years or older with COVID-19 admitted to Copenhagen University Hospital–Amager and Hvidovre, Denmark, between March 10 and May 31, 2020. Details of the cohort have been described previously [13]. In brief, all consecutive individuals admitted were included. All cases were confirmed by SARS-CoV-2 reverse-transcriptase-polymerase-chain-reaction on an oropharyngeal swab or lower respiratory tract specimen. At the time of the study, the Wuhan COVID-19 variant was dominant. Remdesivir, dexamethasone, and anti-interleukin-6 treatment was not yet available. Data including patient characteristics (age, sex, body mass index (BMI), comorbidities, and treatment limits), vital parameters, and laboratory measurements were transferred from electronic health records and managed using Research Electronic Data Capture browser-based software (REDCap, Vanderbilt, TN, USA). Only patients with an available blood sample, drawn within four days from admission, were included in our study. Serum was separated by centrifugation and stored at minus 80°C. Individuals were followed for 90 days from sampling or until death, whichever occurred first.
## 2.2. Study Approval
The study was approved by the Danish Patient Safety Authority (record no. 31-1521-309) and the Regional Data Protection Center (record no. P-2020-492). Measurements of biomarkers in stored samples from the biobank were approved by the Ethical Committee of the Capital Region of Denmark (record no. H-20047597). A requirement of individual informed consent was exempted by the committee.
## 2.3. Outcomes
The primary outcome was 90-day mortality from the time of blood sampling. The secondary outcome was respiratory failure defined as receiving mechanical ventilation during admission.
## 2.4. Adiponectin and Leptin Measurements
Adiponectin and leptin levels in serum were determined by immunoassay with minor modifications according to the manufacturer's instructions using magnetic fluorescently labelled microsphere beads with suspension array system ProcartaPlex (EPX01A-12032-901 and EPX01A-12039-901) (ThermoFisher, Vienna, Austria) and analyzed on a BioPlex 200 (Bio-Rad, Hercules, CA, USA). For each of the analytes, the lower and upper limits of quantification were defined as the lowest/highest measurable standard ±three times the standard deviation. Values below or above these limits were assigned a value of $10\%$ lower or higher than the limit of quantification.
## 2.5. Statistics
Descriptive statistics are presented as medians with interquartile ranges (IQRs) for continuous variables and numbers with percentages for categorical variables. The Mann–Whitney U test, Kruskal–Wallis test, χ2-test, and Fisher's exact test were used to compare groups, as appropriate. Correlations were estimated by Spearman's coefficient. Odds ratios (ORs) and $95\%$ confidence intervals (CIs) associated with outcomes were estimated by logistic regression. ORs and corresponding $95\%$ CIs were displayed in forest plots. Models were adjusted for age, sex, and BMI, as these variables were considered potential confounders. Further adjustment for confounding was performed after evaluating if any other variables were associated with circulating adipokine levels or outcome. Only variables available for >$90\%$ of the subjects were added to this fully adjusted model. An additional analysis on respiratory failure was performed, in which subjects with a do-not-intubate order were omitted. The non-normally distributed variables leptin, adiponectin, adiponectin/leptin ratio, and BMI were log2-transformed prior to analysis. Thus, the estimated OR corresponds to the OR associated with a doubling of circulating adiponectin and leptin levels. Age was modelled as categorical (≤60 years, 61–80 years, ≥81 years) as the relationship between age and mortality is nonlinear [4]. A two-tailed value of $p \leq 0.05$ was considered statistically significant. Statistics were performed in R version 4.0.3 [14].
## 3.1. Study Population
We obtained adipokine measurements from 123 patients out of 324 patients hospitalized with COVID-19. The median time from admission to blood sampling was two days (IQR 2–3). Subjects had a median age of 72 years, slightly more were males, and they were slightly overweight (Table 1). The most common comorbidity was hypertension ($48\%$) followed by diabetes ($29\%$). At admission, most subjects presented with infiltration on chest X-ray, half received supplemental oxygen, and the majority had elevated levels of plasma lactate dehydrogenase (LDH) and C-reactive protein (CRP) (Table 1). When comparing our cohort with the overall cohort, subjects with or without available adipokine measurements were similar on many parameters, including age, BMI, and comorbidities (Supplementary Table 1). However, some differences were present as well. In the group with measured circulating adipokine levels, more were male, and they were more ill at admission, with a slightly higher respiratory rate, a slightly lower saturation, more need of supplemental oxygen, and higher levels of plasma alanine aminotransferase (ALT), LDH, and CRP (Supplementary Table 1).
## 3.2. Associations between Circulating Adipokine Levels and Clinical Characteristics
The median values of circulating leptin and adiponectin were 7.2 ng/mL (IQR 3.8–13.4) and 9.0 μg/mL (IQR 5.7–14.6), respectively (Table 1). Levels of circulating leptin did not differ across the three age groups ($$p \leq 0.37$$), while levels of circulating adiponectin were found to be higher with increasing age ($p \leq 0.001$) (Table 2). Women had higher levels of circulating adipokines than men (leptin (12.2 ng/mL vs. 5.7 ng/mL) and adiponectin (12.5 µg/mL vs. 8.0 µg/mL)). In addition, levels of circulating adiponectin were affected in patients with certain comorbidities compared to patients without these comorbidities, namely, hypertension (10.6 µg/mL vs. 7.6 µg/mL), diabetes (7.2 µg/mL vs. 9.4 µg/mL), and chronic obstructive pulmonary disease (COPD) (14.8 µg/mL vs. 8.9 µg/mL) (Table 2). Circulating leptin was not affected by any comorbidity. In addition to the associations listed in Table 2, correlations between levels of circulating adipokines and BMI were assessed. Circulating leptin was positively correlated with BMI ($r = 0.46$, $p \leq 0.0001$), while there was only a weak inverse correlation between BMI and circulating adiponectin (r = −0.20, $$p \leq 0.04$$).
## 3.3. Association of Circulating Leptin and Adiponectin with 90-Day Mortality
At the 90-day followup, 37 patients ($30\%$) had died. Generally, subjects who had died at followup were older and more likely to suffer from hypertension. At admission, they had lower saturation and blood lymphocyte counts as well as higher levels of plasma creatinine and LDH than patients alive at followup (Table 1). No differences in survival were found by sex, BMI, or any of the other included comorbidities, vital parameters, or laboratory findings. Nor were there any differences in levels of circulating leptin or adiponectin between survivors and nonsurvivors (Table 1).
Circulating adipokine levels were not significantly associated with 90-day mortality in unadjusted logistic regression models (Figure 1(a)). However, adjusted for age groups, sex, and BMI, circulating adiponectin was associated with 90-day mortality (OR 0.68, CI 0.49–0.93). Thus, a doubling of adiponectin corresponded to a $32\%$ decrease in the odds of 90-day mortality. Hypertension, diabetes, COPD, and oxygen saturation at admission were found to be potential confounders (Tables 1 and 2), and consequently, these comorbidities were included in a third analysis (Figure 1(a)). In this model, the association between circulating adiponectin and 90-day mortality remained (OR 0.62, CI 0.43–0.89). When evaluating the impact of the individual variables on the model, adjustment for age groups had the biggest effect (OR 0.72, CI 0.56–0.94), followed by adjustment for hypertension (OR 0.79, CI 0.63–0.99) (not shown).
Because age correlated with adiponectin levels and was the strongest modifier of the association between circulating adiponectin and mortality, exploratory stratified survival curve analysis was performed by median age (72 years) and median level of circulating adiponectin (9.0 μg/mL). The survival curve analysis illustrates that older age increases the risk of 90-day mortality independently of adiponectin levels (Figure 2). Among 19 individuals aged 72 years or younger, all were alive at the 90-day followup for individuals with circulating adiponectin above the median, while in the group with below median circulating adiponectin, 10 of 44 ($23\%$) had died at follow-up (Figure 2). For individuals older than 72 years, levels of circulating adiponectin did not alter survival status at day 90.
Circulating leptin at admission was not associated with 90-day mortality in any of the adjusted models (Figure 1(a)).
## 3.4. Association of Circulating Leptin and Adiponectin with Respiratory Failure
Respiratory failure leading to mechanical ventilation had developed in 20 patients ($16\%$) at the 90-day followup. Median levels of circulating adiponectin in subjects receiving mechanical ventilation compared to subjects not receiving mechanical ventilation were 6.4 μg/mL (IQR 2.8–9.5) and 9.3 μg/mL (IQR 6.1–16.3), respectively, ($p \leq 0.01$). For circulating leptin, the median in these two groups were 10.7 ng/mL (IQR 6.3–13.1) and 6.2 ng/mL (IQR 3.6–13.4) ($$p \leq 0.28$$). Accordingly, circulating adiponectin was associated with respiratory failure in an unadjusted model (OR 0.69, CI 0.55–0.88) (Figure 1(b)). The association was stronger in a model adjusted for age groups, sex, and BMI (OR 0.63, CI 0.46–0.87), and when including comorbidities and oxygen saturation at admission (OR 0.60, CI 0.42–0.86). The analysis performed exclusively on subjects without a do-not-intubate order provided similar results for both the unadjusted model (OR 0.74, CI 0.57–0.96) and the fully adjusted model (OR 0.59, CI 0.40–0.87). Levels of circulating leptin at admission were not associated with respiratory failure in any of the models (Figure 1(b)).
## 3.5. Association of Adiponectin/Leptin Ratio with Mortality and Respiratory Failure
The adiponectin/leptin ratio was associated with respiratory failure in unadjusted analysis (OR 0.75, CI 0.61–0.91) and after adjustment for age, sex, and BMI (OR 0.69, CI 0.52–0.91). The adiponectin/leptin ratio was not associated with 90-day mortality in unadjusted analysis (OR 0.95, CI 0.80–1.11) or after adjustment for age, sex, and BMI (OR 0.81, CI 0.64–1.02).
## 4. Discussion
In our study, we found that higher circulating adiponectin at admission was associated with reduced odds of 90-day mortality and respiratory failure in patients hospitalized with COVID-19. Circulating leptin was not associated with 90-day mortality or respiratory failure in any of the models, while the adiponectin/leptin ratio was associated with respiratory failure but not mortality. Generally, levels of circulating adipokines varied between males and females, while only circulating adiponectin differed between age groups and between subjects with or without hypertension, diabetes, and COPD. Both circulating leptin and circulating adiponectin correlated with BMI, though the correlation between circulating leptin and BMI was stronger.
The associations between higher circulating adiponectin and reduced odds of 90-day mortality and respiratory failure were strengthened when adjusting for age, sex, BMI, hypertension, diabetes, COPD, and oxygen saturation at admission. Adjustment for age had the biggest individual effect, probably due to the strong association between age and circulating adiponectin. We observed higher levels of circulating adiponectin in older patients, consistent with previous reports on healthy subjects [15]. In their study, Obata and colleagues found that adiponectin was increased with age, independently of body fat status, glucose metabolism, and lipid profiles. This relationship might also explain, why subjects with hypertension were found to have significantly higher levels of adiponectin even though hypertension has been linked to low levels previously [16]. Furthermore, survival curve analysis performed by median age and median circulating adiponectin level showed that the effect of adiponectin levels on the risk of 90-day mortality appears to be more profound among younger individuals than older individuals. Likewise, for a range of comorbidities including cardiovascular disease, hypertension, and diabetes, it has been shown that the effect of the comorbidity on COVID-19 severity increases with young age [17]. Perhaps, the effect of one condition is less pronounced in the elderly population, as the occurrence of multiple chronic conditions, disability, and frailty is higher in this group than in the general population, suggesting an overall increased vulnerability to disease [18].
Our results indicate that adiponectin could be involved in the pathogenesis of severe COVID-19. Other studies on adiponectin and COVID-19 were mostly smaller and results were sometimes discrepant with ours. A single study of similar size but with a younger population with lower mortality assessed circulating adiponectin in relation to mortality or respiratory failure and was unable to show an association between circulating adiponectin and admission to intensive care units or in-hospital death in patients hospitalized with COVID-19 [19]. Two studies, however, showed an association between adiponectin and the severity of COVID-19 but did not assess outcomes in relation to adiponectin [20, 21]. Reiterer et al. compared levels of circulating adiponectin between groups of critically ill patients and found that circulating adiponectin in patients with COVID-19 ARDS was decreased by 50–$60\%$ compared to non-COVID-19 controls with and without ARDS [20]. Similarly, Kearns et al. found that circulating adiponectin was significantly lower in patients with COVID-19 ARDS compared to non-COVID-19 ARDS [21]. Other studies of limited size were unable to show any association between adiponectin levels and COVID-19 according to disease severity (mild, moderate, or severe), Sequential Organ Failure Assessment (SOFA) score, and type of respiratory support required [22–25]. Further studies on adiponectin's role in COVID-19 are warranted in order to confirm and refute our findings and to study adiponectin's role in populations treated with antivirals and anti-inflammatory agents.
As is the case for adiponectin, reports on the role of circulating leptin in SARS-CoV-2 infection are somewhat conflicting. Wang et al. showed that levels of circulating leptin were higher in patients with severe COVID-19 compared to milder cases, while Singh et al. found nonsurvivors to have higher levels than survivors [26, 27]. In addition, the study by Singh et al. showed a correlation between circulating leptin and SOFA score at baseline. Tonon et al. found higher circulating leptin levels in patients requiring noninvasive mechanical ventilation, but this association did not persist in multivariate regression models [25]. Other studies report no association between circulating leptin and severity or outcomes in SARS-CoV-2 infection [19, 22, 24, 28], in agreement with the results from our study. This suggests that leptin plays only a limited role in the pathogenesis of COVID-19.
Our study was strengthened by the consecutive enrollment of patients, the fairly large study size, and the complete followup on subjects. Given that blood samples were drawn within four days of admission, and since the study was carried out before any treatment regimens for COVID-19 were established, levels of circulating adipokines were unlikely to be affected by the administration of dexamethasone, tocilizumab, or other immunomodulatory therapy. Some limitations of the study should be considered. Association studies do not show causality and we cannot account for unmeasured residual confounding. Also, we only included one sample from each patient, and therefore the dynamics in adipokine levels could not be examined.
In conclusion, we found that higher circulating adiponectin was associated with reduced odds of mortality and respiratory failure in patients hospitalized with COVID-19. Further studies are needed to elucidate how adipokines, especially adiponectin, affect the progression and prognosis of COVID-19, also in the context of immunomodulatory treatment with corticosteroids and IL-6 inhibitors.
## Data Availability
The data used to support the findings of this study are available from the corresponding author upon reasonable request.
## Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper. TB reports grants from Pfizer, Novo Nordisk Foundation, Simonsen Foundation, Kai foundation, Erik and Susanne Olesen's Charitable Fund, and Lundbeck Foundation, grants and personal fees from GSK and Pfizer, and personal fees from Boehringer Ingelheim, Astra Zeneca, and Janssen, outside the submitted work.
## References
1. Guan W. J., Ni Z. Y., Hu Y.. **Clinical characteristics of coronavirus disease 2019 in China**. (2020) **382** 1708-1720. DOI: 10.1056/NEJMoa2002032
2. Lucas C., Wong P., Klein J.. **Longitudinal analyses reveal immunological misfiring in severe COVID-19**. (2020) **584** 463-469. DOI: 10.1038/s41586-020-2588-y
3. Yang Y., Ding L., Zou X.. **Visceral adiposity and high intramuscular fat deposition independently predict critical illness in patients with sars‐COV‐2**. (2020) **28** 2040-2048. DOI: 10.1002/oby.22971
4. Williamson E. J., Walker A. J., Bhaskaran K.. **Factors associated with COVID-19-related death using OpenSAFELY**. (2020) **584** 430-436. DOI: 10.1038/s41586-020-2521-4
5. Kim J., Nam J.-H.. **Insight into the relationship between obesity-inducedlow-level chronic inflammation and COVID-19 infection**. (2020) **44** 1541-1542. DOI: 10.1038/s41366-020-0602-y
6. Tilg H., Moschen A. R.. **Adipocytokines: mediators linking adipose tissue, inflammation and immunity**. (2006) **6** 772-783. DOI: 10.1038/nri1937
7. Zhang Y., Proenca R., Maffei M., Barone M., Leopold L., Friedman J. M.. **Positional cloning of the mouse obese gene and its human homologue**. (1994) **372** 425-432. DOI: 10.1038/372425a0
8. Matarese G., Moschos S., Mantzoros C. S.. **Leptin in immunology**. (2005) **174** 3137-3142. DOI: 10.4049/jimmunol.174.6.3137
9. Yamauchi T., Kamon J., Waki H.. **The fat-derived hormone adiponectin reverses insulin resistance associated with both lipoatrophy and obesity**. (2001) **7** 941-946. DOI: 10.1038/90984
10. Li S., Shin H. J., Ding E. L., van Dam R. M.. **Adiponectin levels and risk of type 2 diabetes: a systematic review and meta-analysis**. (2009) **302** p. 179. DOI: 10.1001/jama.2009.976
11. Pischon T.. **Plasma adiponectin levels and risk of myocardial infarction in men**. (2004) **291** p. 1730. DOI: 10.1001/jama.291.14.1730
12. Xu A., Wang Y., Keshaw H., Xu L. Y., Lam K. S. L., Cooper G. J. S.. **The fat-derived hormone adiponectin alleviates alcoholic and nonalcoholic fatty liver diseases in mice**. (2003) **112** 91-100. DOI: 10.1172/jci200317797
13. Israelsen S. B., Kristiansen K. T., Hindsberger B.. **Characteristics of patients with COVID-19 pneumonia at Hvidovre hospital, march-april 2020**. (2020) **67**
14. 14R Core Team
R: A Language and Environment for Statistical Computing
2020Vienna, AustriaR Foundation for Statistical Computing. (2020)
15. Obata Y., Yamada Y., Takahi Y.. **Relationship between serum adiponectin levels and age in healthy subjects and patients with type 2 diabetes**. (2013) **79** 204-210. DOI: 10.1111/cen.12041
16. Adamczak M., Więcek A., Funahashi T., Chudek J., Kokot F., Matsuzawa Y.. **Decreased plasma adiponectin concentration in patients with essential hypertension**. (2003) **16** 72-75. DOI: 10.1016/s0895-7061(02)03197-7
17. Fernández Villalobos N. V., Ott J. J., Klett-Tammen C. J.. **Effect modification of the association between comorbidities and severe course of COVID-19 disease by age of study participants: a systematic review and meta-analysis**. (2021) **10** p. 194. DOI: 10.1186/s13643-021-01732-3
18. Fried L. P., Ferrucci L., Darer J., Williamson J. D., Anderson G.. **Untangling the concepts of disability, frailty, and comorbidity: implications for improved targeting and care**. (2004) **59** M255-M263. DOI: 10.1093/gerona/59.3.m255
19. Minuzzi L. G., Seelaender M., Silva B. S. D. A.. **COVID-19 outcome relates with circulating BDNF, according to patient adiposity and age**. (2021) **8**. DOI: 10.3389/fnut.2021.784429
20. Reiterer M., Rajan M., Gómez-Banoy N.. **Hyperglycemia in acute COVID-19 is characterized by insulin resistance and adipose tissue infectivity by SARS-CoV-2**. (2021) **33** 2174-2188.e5. DOI: 10.1016/j.cmet.2021.09.009
21. Kearns S. M., Ahern K. W., Patrie J. T., Horton W. B., Harris T. E., Kadl A.. **Reduced adiponectin levels in patients with COVID‐19 acute respiratory failure: a case‐control study**. (2021) **9**. DOI: 10.14814/phy2.14843
22. Di Filippo L., De Lorenzo R., Sciorati C.. **Adiponectin to leptin ratio reflects inflammatory burden and survival in COVID-19**. (2021) **47**. DOI: 10.1016/j.diabet.2021.101268
23. Caterino M., Gelzo M., Sol S.. **Dysregulation of lipid metabolism and pathological inflammation in patients with COVID-19**. (2021) **11** p. 2941. DOI: 10.1038/s41598-021-82426-7
24. Blot M., Masson D., Nguyen M., Bourredjem A., Binquet C., Piroth L.. **Are adipokines the missing link between obesity, immune response, and outcomes in severe COVID-19?**. (2005) **45**
25. Tonon F., Di Bella S., Giudici F.. **Discriminatory value of adiponectin to leptin ratio for COVID-19 pneumonia**. (2022) **2022** 9. DOI: 10.1155/2022/9908450
26. Wang J., Xu Y., Zhang X.. **Leptin correlates with monocytes activation and severe condition in COVID-19 patients**. (2021) **110** 9-20. DOI: 10.1002/jlb.5hi1020-704r
27. Singh R., Hemati H., Bajpai M.. **Sustained expression of inflammatory monocytes and activated T cells in COVID‐19 patients and recovered convalescent plasma donors**. (2021) **9** 1279-1290. DOI: 10.1002/iid3.476
28. Larsson A., Lipcsey M., Hultström M., Frithiof R., Eriksson M.. **Plasma leptin is increased in intensive care patients with COVID-19—an investigation performed in the PronMed-cohort**. (2021) **10** p. 4. DOI: 10.3390/biomedicines10010004
|
---
title: Attenuation of Strychnine-Induced Epilepsy Employing Amaranthus viridis L.
Leaves Extract in Experimental Rats
authors:
- Aashish Bharadwaj
- Ashwani Sharma
- Talever Singh
- Devender Pathak
- Tarun Virmani
- Girish Kumar
- Anjali Sharma
- Abdulsalam Alhalmi
journal: Behavioural Neurology
year: 2023
pmcid: PMC10030215
doi: 10.1155/2023/6684781
license: CC BY 4.0
---
# Attenuation of Strychnine-Induced Epilepsy Employing Amaranthus viridis L. Leaves Extract in Experimental Rats
## Abstract
### Objective
Epilepsy is one of the most prevalent neurological illnesses defined by periodic seizures with or without loss of consciousness caused by aberrant neural activity. There are many allopathic medications available for the treatment of epilepsy such as phenytoin (PHY), but the side effects are a major concern. Therefore, the present study involved the evaluation of the pharmacological significance of *Amaranthus viridis* L. extract (EAV) in the management of strychnine (STR)-induced epilepsy.
### Method
STR (3.5 mg/kg, i.p.) was injected into male rats 30 minutes after the pre-treatment of a standard drug (PHY: 20 mg/kg) and the two doses of EAV (EAV-200 and EAV-400 mg/kg, p.o.) to the respective groups to cause the convulsions. The anti-convulsant effect of EAV-200 and EAV-400 against STR-induced convulsion in rats was investigated in terms of convulsion onset, duration of convulsions, number of convulsions, and convulsion score. Furthermore, the mitochondrial function and integrity in the brain's prefrontal cortex (PFC) were also estimated.
### Results
EAV-400 significantly increased the onset of convulsion from 61.67 ± 3.051 to 119.2 ± 2.738 and reduced the STR-induced duration of convulsions from 144.8 ± 3.582 to 69.17 ± 3.736, number of convulsions from 4.000 ± 0.1592 to 1.533 ± 0.1542, and convulsion score from 5.000 ± 0.3651 to 2.833 ± 0.3073 in rats. EAV-400 significantly attenuated the STR-induced decrease in the mitochondrial function and integrity of the rat PFC. In rats, EAV-400 significantly accelerated the onset of convulsions while decreasing the STR-induced duration, frequency, and score.
### Conclusion
Based on investigational findings, EAV-400 could be inferred to be a possible anti-epileptic option for the treatment of epilepsy of this plan in preclinical research.
## 1. Introduction
Epilepsy is a condition in which the balance of neurotransmitters and neuromodulators is disturbed, and it is one of the most often occurring neurological disorders related to unusual, hypersynchronous neuronal activity, characterised by recurrent seizures both with and without unconsciousness [1–3]. In India, it is believed that over 10 million people suffer from epilepsy [4]. As per the World Health Organization (WHO), there are approximately 50 million persons with epilepsy globally, with $80\%$ of them living in developing countries such as India [5].
“Epileptic seizure” is known as a seizure caused by abnormal neural activity as opposed to a non-epileptic event, for example, a psychogenic seizure. The disorder known as “epilepsy” is characterised by recurring and unprovoked seizures. Each of the several epilepsy causes reveals a fundamental brain malfunction [6]. A classification system was first published by the International League against Epilepsy (ILAE) in the year 1960, and the last official update was in the year 1989. The classification is as follows.
## 1.1. Generalised Seizures
This type of seizure affects both sides of the brain [7].
## 1.1.1. Absence Seizures
Previously, this sort of seizure was referred to as a petit mal seizure. This type of seizures typically ends in 2–15 seconds and might happen only a very less times per day or over 100 times each day. These manifest as dull gazing that might be misinterpreted as daydreaming, somatic automatisms such as twitching of the facial or body muscles, lip biting, tripping, or plucking at clothing. The individual will not remember what occurred during the seizure. It primarily affects youngsters between the ages of 4 and 12.
## 1.1.2. Generalised Tonic–Clonic Seizure
Grand mal seizure was the old name for this type of seizure. During this seizure, arms and legs stiffen initially; this stage is known as tonicity. This phase is followed by the clonic phase when limbs and heads begin to twitch. Generalised seizures, like other seizures, can vary, with most people having only the tonic–clonic phase. Throughout the seizures, the individual may lessen or stop breathing, chew their tongue or the interior of their mouth, or exhibit incontinence. After the seizure, the individual will most likely be disoriented, will forget what occurred, and may need to sleep and have a headache. Depending on the individual, recovery time might range from minutes to hours.
The events that make up this seizure often start with bilateral myoclonic jerks; then, there is a tonic contraction of the muscles in the extremities as well as the axial trunk, which causes the extremities and neck to extend.
## 1.1.3. Myoclonic Seizures
The patient's body jerks as a result of these seizures, such as arm/leg twitching. Myoclonic seizures normally do not require first aid, but if it is the first seizure, then there is a need to consult a doctor to find out the cause.
## 1.1.4. Atonic Seizures
These types of seizures are also referred to as astatic seizures. These seizures cause muscles to relax abruptly and lead to collapse without warning. These seizures cause a portion of the entire body to shuffle. This implies that a person's head may abruptly droop, or may sag or perhaps collapse, falling on the floor.
## 1.1.5. Partial Seizures
Seizures of this sort are the most prevalent. When only one side of the brain is affected, it occurs. These seizures can cause activity to start in one area of the brain and then move to other, or it may remain in the same location. The symptoms vary according to the part of the brain affected. For instance, the capacity to speak will be impaired if a seizure occurs in the speech centre of the brain. If a seizure begins as a partial seizure and subsequently extends to involve the whole brain, it is called a partial seizure secondary [7].
The procedure of transforming a non-epileptic mind into any capable of causing recurring, spontaneous seizures is known as epileptogenesis. The process is believed to be brought on by an imbalance in inhibitory and excitatory actions inside a neuronal network, which causes it to function exceedingly, hypersynchronously, and oscillatory. If this imbalance persists, it can disrupt other neural circuits as well as normal neuronal processing [8, 9]. Epileptogenic networks are broadly dispersed in generalised epilepsies, involving bilateral thalamocortical regions. Neural circuits in one hemisphere, most frequently limbic or neocortical, are involved in networks for focal epilepsies [8, 10]. Epileptogenic networks are caused by an imbalance between excitation and inhibition, which is not always only an increase in excitation or a decrease in inhibition; an aberrant rise in inhibition can also be pro-epileptogenic in specific circumstances, such as the absence of seizures or limbic epilepsies in the developing brain. The vast majority of generalised epilepsies are assumed to have a hereditary origin. In contrast, structural brain abnormalities were assumed the most prominent feature of localised epilepsies, particularly in drug-resistant epilepsy. However, growing cases of hereditary and de-novo genetic alterations have been discovered in non-lesional focal epilepsy [11].
There are many allopathic medications available for the treatment of epilepsy such as phenytoin (PHY), topiramate, tiagabine, oxcarbazepine, zonisamide, vigabatrin, clobazam, felbamate, lamotrigine, and gabapentin, but as we all know, they all show many side effects as given in Table 1. In contrast to synthetics, which are viewed as being hazardous to use for treatment, herbal products now stand for safety. Based on the worldwide problem and epidemiology of epilepsy, it is reasonable to expect that alternate and supplementary medications for epilepsy management are required.
Natural products are a source of bioactive compounds and are utilised as traditional medical treatments in the treatment of a wide range of disorders, including epilepsy, all over the globe [13]. According to a literature review, the plant *Amaranthus viridis* L. extract (EAV) is recommended for the treatment of several disorders, including diabetes, high cholesterol, pyretic and nociceptive pain, hepatoprotective conditions, and anti-oxidant conditions. Although herbs have been valued for their therapeutic, flavouring, and aromatic properties for millennia, the modern era's synthetic goods temporarily overshadowed their significance. However, the slavish reliance on synthetics has ended, and people are going back to nature in search of stability and safety. For the current study, A. viridis L. is selected because its traditional uses show its activity against inflammations, abscesses, gonorrhoea, orchitis, and haemorrhoids; its infusion is used to purify the blood; the pounded root is applied against dysentery and eye infection and also showed its neuroprotective activity in neurological disorders [14] but not in strychnine (STR)-induced epilepsy in rats.
## 2.1. Chemicals and Reagents
PHY was obtained from Sigma Aldrich (St. Louis, MO, USA). All other chemicals and reagents of High-performance liquid chromatography (HPLC) and analytical grade were procured from Merck Pvt. Ltd., New Delhi, and Himedia Laboratories Pvt. Ltd., Mumbai, India. Other necessary chemicals were issued from the chemical store of Rajiv Academy for Pharmacy such as dichloromethane, carboxymethylcellulose (CMC) sodium, and sodium chloride.
## 2.2. Collection of Plant Material and Identification
Fresh leaves of A. viridis L. were collected from the Mathura region, Uttar Pradesh, India, in 2022. Leaves were identified and submitted on 31 March 2022 as a specimen in Council of Scientific & Industrial Research (CSIR)-National Institute of Science Communication and Policy Research, New Delhi (authentication number of A. viridis L. leaves—NIScPR/RHMD/Consult/$\frac{2022}{4055}$-56). The identification was made through macroscopic examinations of the sample, an in-depth examination of the literature, and a comparison of the sample with real samples kept in the Raw Material Herbarium and Museum in Delhi (RHMD).
## 2.3. Preparation of Extract of Leaves of A. viridis
The plant leaf was collected and dried in the shade. The shade-dried leaves were then roughly pulverised. To get the coarse powder, use a mixer grinder and sieve number 60. The weighted, coarsely powdered ingredients were then used for soxhlet extraction, phytochemical studies, and pharmacological studies after being sealed in an airtight container.
A 1000 ml soxhlet apparatus was filled with 350 g of coarsely crushed leaves, and dichloromethane was then extracted for 72 hours with continuous hot percolation. The solvent was removed after extraction, and the extract was then concentrated at room temperature.
## 2.4. Soxhlet Extraction
The soxhlet extraction method is a constant extraction technique that involves repeatedly cycling the same solvent through the extractor. The steps in this process include solvent extraction and evaporation. The drug is subjected to continuous extraction, whereas the solvent vapours are routed to a condenser, as well as the distillate is then returned. The extractor body of a soxhlet apparatus designed for continuous extraction is coupled to a side tube as well as a syphon tube. Conventional couplings are used to connect the extractor's mouth to a condenser and its lower end to a distillation flask. A thimble made of filter paper or thin muslin cloth or the soxhlet device itself can be used to load the powdered, crude medicament. The inner diameter of the soxhlet apparatus corresponds to the diameter of the thimble. The extraction assembly is finished by installing a distillation flask and a condenser. Before heating, the solvent is initially given a chance to fix the powder. Fresh activated porcelain pieces are placed next to the flask to prevent solvent bumping. The level of fluid in the collector and syphon tube is gradually raised by condensed liquid and vapours that pass through the side tube. A syphon is set up, and the extraction chamber's contents are moved to the flask when the liquid nears the point of return. The cycle of evaporation of the solvent and syphoning back could be repeated as often as is practical to ensure successful extraction. Despite being a continuous extraction process, it is only a series of quick macerations.
## 2.5.1. Tests for Carbohydrate
Separately, a little portion of the extract was mixed in 4 ml of distilled water, and then, it was filtered. The following tests were performed on the filtrate to check for the absence of carbohydrates and glycosides.
## 2.5.2. Molisch's Test
Two millilitres of concentrated sulphuric acid was then put into the test tube after the filtrate had been treated with a few drops of a $1\%$ alcoholic–naphthol solution. When two liquids come together, there should not be a brown ring since that means there are carbohydrates.
## 2.5.3. Fehling's Test
After being treated with 1 ml of each of Fehling's solutions A and B, the filtrate was boiled in the water bath. A red precipitate confirmed the presence of the carbohydrate.
## 2.5.4. Test for Pentose Sugar
Treat test solution with HCL, and then heat and add the phloroglucinol crystal which produce a colored compound with high molar absorptivity which indicate the presence of pentose.
## 2.5.5. Test for Fixed Oil and Fats
[1] Spot Test [16]. The absence of an oil stain on the paper, which indicated the absence of fixed oil, was determined by pressing a little quantity of the extract between two sheets of filter paper.
[2] Saponification Test [17]. Add a few drops of 0.5 N alcoholic potassium hydroxide and a drop of phenolphthalein individually to a tiny amount of different extracts and boil in a hot water bath for 1–2 hours. The absence of changes implies the absence of oil and fats.
[3] Test for Glycerine [18]. Add $10\%$ sodium hydroxide solution after 5 drops of the sample have been treated with $1\%$ sulphate solution. Glycerine is found in the sample, which is confirmed by the creation of a clear blue solution. The cupric hydroxide produced during the process dissolves in glycerine without precipitating.
## 2.6.1. Biuret Test [19]
The violet colour shows the presence of protein in the extract solution (2 ml) and biuret reagents.
## 2.6.2. Xanthoprotein Test [20]
A yellow precipitate was created when 5 ml of extract solution and 1 ml of nitric acid were heated. A $40\%$ solution of sodium hydroxide was then added after cooling, and an orange colour resulted.
## 2.6.3. Test for Steroids [21]
[1] Salkowski's Test. A few drops of strong sulphuric acid were added to 1 ml of chloroform solution. The presence of phytosterols results in a brown colour.
[2] Libermann–Burchard's Test. Acetic anhydride was diluted with a few drops and added to the extract. After heating, strong sulphuric acid was then poured into the test tube from the side, which caused the upper layer to become green and a brown ring to form at the junction of two layers, confirming the presence of steroids.
## 2.6.4. Test for Glycosides [22]
[1] Test A. A total of 200 mg of the drug was extracted with 5 ml of diluted H2SO4 by warming on a water bath. After filtering, a $5\%$ solution of sodium hydroxide was used to neutralise the acid that was extracted. To make it alkaline, 0.1 ml of Fehling's solutions A and B was added. The mixture was then cooked in a water bath for 2 minutes. A comparison was made between the quantity of red precipitate produced and that produced in test B.
[2] Test B. 5 ml of water was added to drug sample and heated on water bath for 2 minutes after being added to 0.1 ml of Fehling's solutions A and B until it turned alkaline. After boiling, the same amount of water was added. It was heated in a water bath for 2 minutes after being added to 0.1 ml of Fehling's solutions A and B until it turned alkaline. A measurement was made of the amount of red precipitate produced. A comparison was made between the quantity of precipitate produced in tests A and B. Additionally, it shows that glycosides exist.
## 2.7.1. Baljet Test
Picric acid or sodium picrate was added to the extract. The colour orange was developed, and the presence of glycosides is shown.
## 2.7.2. Legal's Test
The existence of cardiac glycosides was confirmed by the observation of a blood-red colour in the alcoholic solution of extract, 1 ml pyridine, and 1 ml sodium nitroprusside solution.
## 2.7.3. Keller–Killiani's Test
A mixture of 2 ml of the extract, 3 ml of glacial acetic acid, and 1 drop of $5\%$ ferric chloride was added. The findings were noted after this solution was carefully applied to the surface of 2 ml of concentrated H2SO4.
## 2.8.1. Borntrager's Test
In a test tube, the test sample was heated for 5 minutes with 1 cc H2SO4. It was hot-filtered, then chilled, and shaken with an equivalent amount of dichloromethane/chloroform. Separating the lowest layer of dichloromethane or chloroform and shaking it with $\frac{1}{2}$ its volume of diluted ammonia in the ammonical layer, no rose pink to red colour was created.
## 2.8.2. Modified Borntrager's Test
A total of 200 mg of the substance was cooked in 2 ml of diluted H2SO4. It was then treated for 5 minutes with 2 ml of $5\%$ aqueous ferric chloride. It was mixed with equal parts chloroform and water. The layer of organic solvent was separated, and an equivalent amount of dilute ammonia was added, resulting in a pinkish-red ammonia layer, which indicates the presence of glycosides.
## 2.9.1. Shinoda's Test
Drop by drop, strong HCl was added to the extract solution along with a few magnesium turns. After a few minutes, the colours pink, crimson, and red emerged, revealing the presence of flavonoids. After adding a small amount of leftover lead acetate solution, the colour changed.
## 2.10.1. Solubility Test
(i) A few ml of chloroform was added to 2–3 ml of the extract, and solubility was observed.
(ii) A few ml of $90\%$ methanol was added to 2–3 ml of the extract, and solubility was checked.
## 2.11.1. Ferric Chloride Test
When ferric chloride solution was added to the test solution, a green colour developed, indicating the presence of condensed tannins.
## 2.11.2. Phenazone Test
A total of 0.5 g of sodium phosphate was added to the test solution, warmed, and filtered. When $2\%$ phenazone solution was added to the filtrate, a huge precipitate was produced, which was frequently coloured, suggesting that tannins are present.
## 2.11.3. Test for Alkaloids
Separately, the extract was evaporated. Dilute HCL was added to the residue, which was thoroughly shaken and filtered. The following experiments were carried out.
## 2.11.4. Dragendroff's Reagents
A few drops of Dragendroff's reagents were applied to 2–3 ml of filtrate, and a precipitate was seen. It denotes the existence of an alkaloid.
## 2.11.5. Mayer's Test
A few drops of Mayer's reagents were applied to 2–3 ml of filtrate, and a precipitate was seen. It shows that alkaloids are present.
## 2.12.1. Ninhydrin Test
3 ml of the test solution was treated in a boiling water bath for 10 minutes with 3 drops of $5\%$ ninhydrin solution. Dark blue colour appearance confirms the presence of amino acids.
## 2.12.2. Experimental Animals and Their Housing
Adult male Albino Wistar rats (180–220 g) of 10–12 weeks were received and used for the study from Animal House, Rajiv Academy for Pharmacy, Mathura. Under regular conditions (25 ± 2°C temperature, 45–$55\%$ humid environment, and a 12 h light–12 h dark cycle), the animals were divided into groups and kept in poly-acrylic houses (cages) lined with husks. The animals were given unrestricted access to their usual pellet food and unrestricted water. Food was withheld from the animals for 16–18 hours before the experiments, but they were permitted to drink as much as they wanted. The animal ethical approval was received from Rajiv Academy for Pharmacy, Mathura under the Institutional Animal Ethical Committee (IAEC) of Rajiv Academy in the meeting held on 08 April 2022, and 30 male Albino Wistar rats have been sanctioned under this proposal for a duration of the next 1.5 months with registration no. is 882/PO/Re/S05/CPCSEA.
## 2.12.3. Experimental Design
The entire experimental regimen was planned to last 14 days. To summarise, all animals were placed into five groups of six, namely control, STR, STR + EAV-200, STR + EAV-400, and STR + PHY. All animals in the STR + EAV-200, STR + EAV-400, and STR + PHY groups received EAV (200 mg/kg, p.o.) [ 24], EAV (400 mg/kg, p.o.) [ 24], and PHY (20 mg/kg, i.p.) [ 25] for 14 days. Except for the control group rats, all other animals were given STR (3.5 mg/kg, i.p.) [ 26] on the 14th day of the experiment following 30 minutes of drug pre-treatment. For the duration of the experiment, all control group rats received once daily CMC ($0.5\%$ w/v) as a vehicle. In the STR, EAV-200, EAV-400, and PHY groups, death rates were $70\%$, $20\%$, $30\%$, and $20\%$, respectively. All behavioural characteristics were recorded, and the animals were killed via cervical derangement. The cortical area of the rat brain was separated and promptly kept at –80°C for biochemical analysis [27, 28]. For repeatability, all biochemical studies were done twice.
## 2.13. Grouping of Animals
Animal grouping was done in the following manner: Group 1: normal control: normal saline $0.9\%$ (10 ml/kg) orally for 14 days.
Group 2: disease control: STR (3.5 mg/kg i.p.).
Group 3: standard: PHY (20 mg/kg, p.o.) ( STR + PHY 20 mg/kg).
Group 4: extract of A. viridis L. 200 mg/kg, p.o. ( STR + EAV-200 mg/kg).
Group 5: extract of A. viridis L. 400 mg/kg, p.o. ( STR + EAV-400 mg/kg).
## 2.14. STR-Induced Convulsion
STR (3.5 mg/kg, i.p.) was injected into rats 30 minutes after the drug administration [26]. The onset, duration, and number of convulsions were observed and recorded. The convulsion signs were observed, and the severity was graded as per the following scale; moreover, the percentage of mortality rate and protection at 10 minutes was recorded.
## 2.15.1. Isolation of Mitochondria from Rat PFC
To separate mitochondria from rat prefrontal cortex (PFC), the conventional procedure of Pedersen et al. [ 1978] [29] was utilised. The mitochondrial protein concentration in tissue fractions was determined using Lowry et al. [ 30] standard technique [1951].
## 2.15.2. Estimation of Mitochondrial Function in Rat PFC
The (3-(4, 5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide [MTT]) reduction test was used to determine mitochondrial activity in tissue fractions by measuring the amount of formazan produced at 595 nm using a spectrophotometric technique [31]. The results were represented in milligrammes of formazan produced each minute per milligramme of protein.
## 2.15.3. Evaluation of MMP in Rat PFC
The mitochondrial membrane potential (MMP) was evaluated by measuring the amount of rhodamine dye taken up by mitochondria in a spectrofluorometer (Hitachi, F-2500) at 535 ± 10 nm excitation and 580 ± 10 nm emission [32]. The results were reported in terms of fluorescence intensity per milligramme of protein.
## 2.16. Histological Analysis of Brain
Samples provided were fixed in $10\%$ neutral buffered formalin. Fixed tissues were dehydrated in graded series of $30\%$, $50\%$, $70\%$, $90\%$, and $100\%$ alcohol to remove water and crystals of picric acid present in Bouins' solution used as fixative. Fixed specimens were dehydrated. Tissues were cleared in xylene and infiltrated with paraffin wax, and sections of uniform thickness (4–6 μ) were made by using a microtome. Casting or blocking specimens are embedded in paraffin using embedding rings and orienting tissue to the area of interest. Blocks were placed at 4°C for 15 minutes to solidify. 5 μm sections were cut using a spencer-type rotary microtome. Cut sections were placed in a 45°C water bath and put on sialinated slides. Slides were allowed to dry in a 37°C oven overnight before staining takes place. Hematoxylin and eosin (H&E) dyes were used to stain the sections that were further visualized in a microscope equipped with an Amscope MU1000 camera [33].
## 3.1. Phytochemical Analysis
The results of the phytochemical analysis are depicted in Table 2.
## 3.2. Mortality Rate and Percent Protection
The effect of EAV-200, EAV-400, and PHY on the mortality rate and percentage protection is shown in Table 3.
From the above table, it can be observed that the mortality rate decreased after the administration of STR + EAV-200 and STR + EAV-400, and the protection increased.
The highest protection was provided by PHY, which is the standard drug followed by EAV-200 and 400 mg/kg.
## 3.3.1. Onset of Convulsion
Rats showed convulsion after administration of PHY. Figure 1 illustrates the impact of STR + EAV-200, EAV-400, and PHY on the onset of convulsion in STR-challenged rats. The results of the statistical analysis showed in Table 4, which shows an onset of convulsion [F[4, 25] = 312, $p \leq 0.001$]. EAV-400 in the onset of convulsion significantly showed a difference compared to control, STR, and STR + PHY, which states that EAV-400 has the potential to attenuate the convulsion onset.
## 3.3.2. Duration of Convulsion
Figure 2 illustrates the impact of STR+EAV-200, EAV-400, and PHY on the duration of convulsions in STR-challenged rats. The results of the statistical analysis showed that the duration of the convulsion has a significant difference [F[4, 25] = 246.5, $p \leq 0.001$]. EAV-400 in the duration of convulsion showed a significant difference when compared to control, STR, and STR + EAV-200, which states that EAV-400 has the potential to attenuate the duration of convulsion. The figure is showing the relationship between STR, EAV-200, EAV-400, and PHY in the duration of convulsions. The duration of convulsion was decreased by the administration of EAV-400 in comparison to STR. EAV-400 has provided a much better effect in comparison to EAV-200 and decreased the duration of convulsions.
## 3.3.3. Number of Convulsions
Figure 3 illustrates the impact of STR+EAV-200, EAV-400, and PHY in the number of convulsions in STR-challenged rats. The results of the statistical analysis showed in Table 5, and the number of convulsions has a significant difference [F[4, 25] = 91.95, $p \leq 0.001$]. EAV-400 in several convulsion significantly showed a difference compared to control, STR, and STR + EAV-200 but not showed a significant difference compared to STR + PHY, which states that EAV-400 have the potential to attenuate the duration of convulsion. The number of convulsions per minute was also recorded, which is shown in Figure 3. As it is evident from the figure, the rats treated with STR suffered an average of 4 convulsions per minute, which was decreased to an average of 1–2 convulsions per minute with the help of EAV-200, EAV-400, and PHY.
## 3.3.4. Convulsion Score
Figure 4 illustrates the impact of STR + EAV-200, EAV-400, and PHY on convulsions score in STR-challenged rats. The results of the statistical analysis are shown in Table 6, in that the number of convulsions has a significant difference [F[4, 25] = 38.06, $p \leq 0.001$]. EAV-400 in convulsion score significantly showed a difference compared to control, STR, and STR + EAV-200 but not showed a significant difference compared to STR + PHY which states that EAV-400 have the potential to attenuate the severity of convulsion.
As discussed earlier, scores were given to the rats based on the severity of the convulsions they are getting. In Figure 4, it is demonstrated that the convulsion scores were highest in the case of STR which gradually decreased after the administration of EAV-400. PHY has demonstrated the least convulsion score.
## 3.4. EAV-400 Attenuates STR-Induced Decrease in Mitochondrial Function and Integrity in Rat PFC
The effect of EAV-400 on the STR-induced changes in the mitochondrial function in terms of the level of formazan produced in MTT assay (A) and integrity in terms of the fluorescence intensity of tetramethylrhodamine, methyl ester (TMRM) (B) in rat PFC is illustrated in Figure 5, and results are given in Table 7. The statistical analysis is shown in the table, which revealed that there were significant differences in mitochondrial function and integrity in rat PFC ([F[4, 25] = 16.21, $P \leq 0.001$] and [F[4, 25] = 77.92, $P \leq 0.001$], respectively). Post hoc analysis showed that EAV-400 and PHY significantly increased the STR-induced decrease in mitochondrial function and integrity in rat PFC. Moreover, there were no significant differences in mitochondrial function and integrity of rats PFC among STR + EAV-400 and STR + PHY groups.
MTT is reduced to formazan in the mitochondria, an increase in the level of formazan increases the mitochondrial mass thereby increasing metabolic feasibility. As it is evident from the graph, the amount of formazan produced was highest in the case of PHY and lowest in the case of STR. Whereas, EAV-400 helped to produce more formazan in comparison to STR and EAV-200. Further, PHY increased the production of formazan in the highest amount.
TMRM is localized in mitochondria and used to detect mitochondrial membrane depolarization. MMP is very high in the case of the normal control group, which is decreased by the STR treatment. EAV-400 has gradually increased the MMP and is highest in the case of PHY.
## 3.5. Histological Analysis of the Brain with H&E Staining
Figure 6 shows the differences anatomically in STR-induced duration of convulsion, STR + EAV-200, STR + EAV-400, control, and PHY.
## 4. Discussion
The current work indicates for the first time that EAV-400 is neuroprotective against STR-induced convulsions in experimental rats. Furthermore, the level of neuroprotection provided by EAV-400 was comparable in the experimental rats. EAV-400 is found to reduce the STR-induced mitochondrial dysfunction in rats' PFC. As a result, EAV might be regarded as an alternate approach to epilepsy management.
STR caused convulsions with a $70\%$ fatality rate in rats in the current investigation, which was similar to previous studies [34]. EAV-400 considerably delayed the start of STR-induced convulsions in the rats. Furthermore, EAV-400 considerably decreased the length, frequency, and severity (score) of STR-induced convulsions in rats. Furthermore, compared to the previous trial, the anti-convulsant action of the EAV was found statistically equivalent to that of the conventional medication PHY [35].
These observations clearly demarcate the fact that EAV possesses glycine receptor agonist activity in the brain.
It is widely established that mitochondria play a vital role in the development of neurons at the subcellular level [36]. As a result, the current investigation of the mitochondrial basis of EAV-400's anti-convulsant action against STR-induced convulsion in experimental rats. EAV-400 significantly reduced STR-induced decrements in mitochondrial function and integrity in the rat PFC in the current investigation.
## 5. Conclusion
Finally, EAV-400 demonstrated anti-convulsant efficacy in a STR-induced animal model of epilepsy. Furthermore, EAV-400 reduced STR-induced mitochondrial damage in rat PFC. As a result, EAV might be regarded as an additional preventative strategy in the treatment of epilepsy. Furthermore, including EAV in the dietary diet can help lower-income people avoid malnutrition-induced epilepsy. Based on investigational findings, EAV-400 could be inferred to be a possible anti-epileptic option for the treatment of epilepsy of this plan in preclinical research.
## Data Availability
The supporting literature, data, and other necessary information used to support the findings of this study are available from the corresponding author upon request.
## Conflicts of Interest
The authors declare that they have no conflicts of interest.
## References
1. Sander J. W., Shorvon S. D.. **Incidence and prevalence studies in epilepsy and their methodological problems: a review**. (1987) **50** 829-839. PMID: 3305790
2. Bertini G., Bramanti P., Constantin G.. **New players in the neurovascular unit: insights from experimental and clinical epilepsy**. (2013) **63** 652-659. PMID: 23962437
3. Fisher R. S., Acevedo C., Arzimanoglou A.. **ILAE official report: a practical clinical definition of epilepsy**. (2014) **55** 475-482. PMID: 24730690
4. Sridharan R., Murthy B. N.. **Prevalence and pattern of epilepsy in India**. (1999) **40** 631-636. PMID: 10386533
5. Haneef Z., Skrehot H. C.. **Neurostimulation in generalized epilepsy: a systematic review and meta-analysis**. (2023)
6. Stafstrom C. E., Carmant L.. **Seizures and epilepsy: an overview for neuroscientists**. (2015) **5** p. a022426
7. Ganeshkumar J.. **Evaluation of anti-epileptic activity of ethanolic extract of leaves of Cassia alata Linn. by maximal electroshock (MES) and isoniazid induced convulsions on mice**. (2016)
8. Fisher R. S., Van Emde W., Boas W. B.. **Epileptic seizures and epilepsy: definitions proposed by the International League Against Epilepsy (ILAE) and the International Bureau for Epilepsy (IBE)**. (2005) **46** 470-472. PMID: 15816939
9. Pitkänen A., Engel J.. **Past and present definitions of epileptogenesis and its biomarkers**. (2014) **11** 231-241. PMID: 24492975
10. Crunelli V., Leresche N.. **Childhood absence epilepsy: genes, channels, neurons and networks**. (2002) **3** 371-382. PMID: 11988776
11. Thijs R. D., Surges R., O’Brien T. J., Sander J. W.. **Epilepsy in adults**. (2019) **393** 689-701
12. Dichter M. A., Brodie M. J.. **New antiepileptic drugs**. (1996) **334** 1583-1590. PMID: 8628341
13. Ayyanar M., Ignacimuthu S.. **Ethnobotanical survey of medicinal plants commonly used by Kani tribals in Tirunelveli hills of Western Ghats, India**. (2011) **134** 851-864. PMID: 21291981
14. Alegbejo J. O.. **Nutritional value and utilization of Amaranthus (Amaranthus spp.)–a review**. (2013) **6** 136-143
15. Ragupathi G.. **Neuropharmacological effects of methanolic extract of leaves of Amaranthus viridis. L in rats & mice**. (2017)
16. Haskel R., Leona G.. **A colorimetric spot test for heated oils**. (1981) **14** 150-152
17. Marcali K., Rieman, III W.. **Microdetermination of saponification number of fats and oils. Decigram, centigram and milligram procedures**. (1946) **18** 144-145
18. Hehner O.. **On the estimation of glycerine in fats**. (1887) **12** 44-46
19. Pinckney A. J.. **The biuret test as applied to the estimation of wheat protein**. (1961) **38** 501-506
20. Subroto E., Lembong E., Filianty F.. **The analysis techniques of amino acid and protein in food and agricultural products**. (2020) **9** 29-36
21. Sharma V., Agarwal A., Chaudhary U., Singh M.. **Phytochemical investigation of various extracts of leaves and stems of Achyranthes aspera Linn**. (2013) **5** 317-320
22. Deb N., Majumdar P., Ghosh A. K.. **Pharmacognostic and phytochemical evaluation of the rhizomes of Curcuma longa Linn**. (2013) **2** 81-86
23. Victor O. N., Chidi O. B. I.. **Phytochemical constituents of some selected medicinal plants**. (2009) **3** 228-233
24. Krishnamurthy G., Lakshman K., Pruthvi N., Chandrika P. U.. **Antihyperglycemic and hypolipidemic activity of methanolic extract of Amaranthus viridis leaves in experimental diabetes**. (2011) **43** p. 450
25. Sefil F., Faruk Bagirici M., Acar D., Marangoz C.. **Influence of carbenoxolone on the anticonvulsant efficacy of phenytoin in pentylenetetrazole kindled rats**. (2012) **72** 177-184. PMID: 22810219
26. Sadek B., Oz M., Nurulain S. M.. **Phenylalanine derivatives with modulating effects on human**. (2017) **138** 124-131. PMID: 28554717
27. Serdiuk S. E., Gmiro V. E., Veselkina O. S.. **Stimulation of gastric mucosa afferents by phenylephrine potentiates anticonvulsive and eliminates sedative action of sodium valproate in the pentylenetetrazol kindling model in rats**. (2014) **100** 96-104
28. Lee W. H., Sonntag W. E., Mitschelen M., Yan H., Lee Y. W.. **Irradiation induces regionally specific alterations in pro-inflammatory environments in rat brain**. (2010) **86** 132-144. PMID: 20148699
29. Pedersen P. L., Greenawalt J. W., Reynafarje B.. **Preparation and characterization of mitochondria and submitochondrial particles of rat liver and liver-derived tissues**. (1978) **20** 411-481. PMID: 151184
30. Lowry O. H.. **Protein measurement with the Folin phenol reagent**. (1951) **193** 265-275. PMID: 14907713
31. Kamboj S. S., Kumar V., Kamboj A., Sandhir R.. **Mitochondrial oxidative stress and dysfunction in rat brain induced by carbofuran exposure**. (2008) **28** 961-969. DOI: 10.1007/s10571-008-9270-5
32. Sharma A., Singh T., Pathak D., Virmani T., Kumar G., Alhalmi A.. **Antidepressive-like effect of Aegle marmelos leaf extract in chronic unpredictable mild stress-induced depression-like behaviour in rats**. (2022) **24**
33. Fischer A. H., Jacobson K. A., Rose J., Zeller R.. **Hematoxylin and eosin staining of tissue and cell sections**. (2008) **2008** p. pdb-prot4986
34. Sanya E. O., Soladoye A. O., Desalu O. O., Kolo P. M., Olatunji L. A., Olarinoye J. K.. **Antiseizure effects of ketogenic diet on seizures induced with pentylenetetrazole, 4-aminopyridine and strychnine in Wistar rats**. (2016) **31** 115-119
35. Kehne J. H., Kane J. M., Miller F. P.. **MDL 27,531 selectively reverses strychnine-induced seizures in mice**. (1992) **106** 910-916. PMID: 1327393
36. Rivero-Segura N. A., Coronado-Mares M. I., Rincón-Heredia R.. **Prolactin prevents mitochondrial dysfunction induced by glutamate excitotoxicity in hippocampal neurons**. (2019) **701** 58-64. PMID: 30790645
|
---
title: Ethnicity Disparities in the Prevalence, Awareness, Treatment, and Control
Rates of Hypertension in China
authors:
- Yanan Yang
- Kunlin He
- Yuewen Zhang
- Xiuming Wu
- Weizhong Chen
- Dongqing Gu
- Ziqian Zeng
journal: International Journal of Hypertension
year: 2023
pmcid: PMC10030218
doi: 10.1155/2023/1432727
license: CC BY 4.0
---
# Ethnicity Disparities in the Prevalence, Awareness, Treatment, and Control Rates of Hypertension in China
## Abstract
### Objectives
Previous studies reported that there were disparities in hypertension management among different ethnic groups, and this study aimed to systematically determine the prevalence, awareness, treatment, and control rates of hypertension in multiple Chinese ethnic groups.
### Methods
We searched Embase, PubMed, and Web of Science for articles up to 25 October, 2022. The pooled prevalence, awareness, treatment, and control rates of hypertension were estimated with $95\%$ confidence intervals (CI). The heterogeneity of estimates among studies was assessed by the Cochran Q test and I2 statistic. Meta-regression analyses were conducted to identify the factors influencing the heterogeneity of the pooled prevalence, awareness, treatment, and control rate of hypertension.
### Results
In total, 45 publications including 193,788 cases and 587,826 subjects were eligible for the analyses. The lowest prevalence was found in the Han group ($27.0\%$), and the highest prevalence was in the Mongolian population ($39.8\%$). The awareness rates ranged from $24.4\%$ to $58.0\%$ in the four ethnic groups. Both the highest treatment and control rates were found in the Mongolian population ($50.6\%$ and $16.0\%$, respectively), whereas the Yi group had the lowest control rate ($8.0\%$). In addition, the study year, the mean age of subjects, mean body mass index of subjects, tobacco use (%), alcohol use (%), residence (urban%), and education (primary school%) had varied effects on heterogeneity.
### Conclusions
These findings highlight the disparities in prevalence, awareness, treatment, and control rates of hypertension in a different ethnic population of China, which could provide suggestions for making targeted prevention measures.
## 1. Introduction
Hypertension is an important risk factor for cardiovascular diseases and other chronic diseases which impose a huge burden on economic and human development globally [1, 2]. Nevertheless, low awareness, treatment, and control of hypertension will handle the prevention progress of cardiovascular diseases [3–5]. In China, the median prevalence of hypertension was $32\%$ in females and $37\%$ in males [6], which were higher than the average level of $20\%$ of women and $24\%$ of men around the world [1]. The awareness, treatment, and control rates reported by a systematic review were only $53\%$, $44\%$, and $17\%$, respectively, among the female population, and these rates were even lower in men [6].
Some studies reported that the prevalence rate of hypertension may be different in each ethnic group, which could be attributed to the different genetic and environmental conditions [7–10]. China is a unified, multinational country, with 56 ethnic groups in all. It is evident that the status and management of hypertension are varied in these groups [11]. Understanding the status of the ethnic groups was beneficial for hypertension control.
Most of the individual studies were conducted for one ethnic group, however, systematically collected data to compare the prevalence, awareness, treatment, and control rates among varied ethnic populations were insufficient. Therefore, this study investigated the disparities in the management of hypertension among ethnic groups to improve hypertension prevention in order to achieve the global noncommunicable disease targets.
## 2.1. Data Sources
We searched the three databases (Embase, PubMed, and Web of Science) up to 25 October, 2022. The medical subject heading (MeSH) terms were used for the search strategies. The search strategy was as follows: (blood pressure OR hypertension) AND (epidemiology OR prevalence OR incidence OR awareness OR treatment OR control) AND (China). To maximize the yield of articles, we further screened the references of all included studies or meta-analyses in the database to select additional eligible articles.
## 2.2. Inclusion and Exclusion Criteria
The eligible articles were under the following criteria: [1] articles that were published in peer-review journals in both English and Chinese; [2] articles that defined hypertension as SBP ≥ 140 mmHg or/and DBP ≥ 90 mmHg; or self-reported pharmacological treatment for hypertension; [3] articles that reported the prevalence or awareness or treatment or control rate; [4] articles that provided the cases and sample size or sufficient data to calculate these rates. Some exclusion criteria were set before data extraction: [1] articles that provided limited information; [2] articles that were not the original research articles, such as review, comment, or letter; [3] articles that focused on the treatment or medication of hypertension. Two investigators (G.DQ and Z.ZQ) evaluated each study and checked the eligibility independently.
## 2.3. Data Extraction
Data were extracted by two researchers (Y.YN and Z.ZQ) using the standard information forms, including first author name, publishing year, study year, study design type, ethnicity, case number, sample size, prevalence rate, awareness rate, treatment rate, control rate, mean age of the population, age range of population, mean body mass index (BMI), tobacco use (%), alcohol use (%), gender (female%), residence (residence%), and education (education%). When the data from the same survey were published in more than one publication, we used the largest one.
## 2.4. Quality Assessment
The quality of included studies was independently assessed by two investigators (H. KL and Y. YN) using the adapted Newcastle-Ottawa Scale (NOS) [12]. If there were any disagreements, it was resolved by consultation with another author (G. DQ). Nine items were applied to assess the included studies, and each study received a total score (TS) for methodological quality. The score ranged from 0–10: TS ≥ 7 ranks as “high quality,” 7<TS ≤ 6 ranks as “moderate quality,” and TS < 6 ranks as “low quality.”
## 2.5. Statistical Analysis
A pooled prevalence, awareness, treatment, and control rates of hypertension were estimated with $95\%$ confidence intervals (CI). The heterogeneity of estimates among studies was assessed by the Cochran Q test and I2 statistic. Meta-regression analyses were conducted to identify the factors influencing the heterogeneity. In regression models, the study year (<2010, ≥2010), mean age (years), mean BMI, tobacco use (%), alcohol use (%), gender (female%), residence(urban%), and education (primary school%) were included as covariates. The sensitivity analysis was also conducted to explore the potential heterogeneity. The Egger test and Begg test were performed for publication bias. Statistical analyses were conducted using Stata 14 software (https://www.stata.com/) p ≤ 0.05 was defined as statistical significance.
## 3.1. Characteristics of Included Studies
A PRISMA flowchart was used to illustrate the selection of studies (Figure 1) [13]. We screened a total of 3,618 articles, and finally, 45 publications of 193,788 cases and 587,826 subjects were included in the analyses. According to the number of datasets, a total of four ethnic groups (Han, Tibetan, Yi, and Mongolian) were considered to conduct the meta-analyses. There were 14 datasets used for the Han population, 23 datasets used for the Tibetan population, 7 datasets used for the Yi population, and 10 datasets used for the Mongolian population. There were nine datasets conducted in 1996–2005, 37 datasets investigated in 2006–2015, eight datasets conducted in 2016–2019, and one dataset conducted in 1979. More characteristics of included studies were displayed in Supplementary Table 1.
## 3.2. Quality of the Studies
The quality of included studies was relatively high. According to the results of the quality assessment, 31 ($68.9\%$) studies were categorized as “high quality” and 14 ($31.1\%$) studies were classified as “moderate quality.” The details of the quality assessment are shown in Supplementary Table 2.
## 3.3. Prevalence, Awareness, Treatment, and Control of Hypertension
All prevalence rates were calculated using the cut-off value ($\frac{140}{90}$ mmHg), and the awareness, treatment, and control rates were calculated among cases with hypertension. The pooled prevalence, awareness, treatment, and control rates of the total population were $31.7\%$, $42.5\%$, $33.4\%$, and $12.2\%$, respectively. The lowest prevalence rate was found in the Han group ($27.0\%$), and the highest prevalence was in the Mongolian population ($39.8\%$). The awareness rates ranged from $24.4\%$ to $58.0\%$% in four ethnic groups. The highest treatment and control rates were found in the Mongolian population ($50.6\%$ and $16.0\%$, respectively). The Yi group had the lowest control rate ($8.0\%$), see Table 1.
## 3.4. Results from Meta-Regression Analyses
The meta-regression analyses were conducted for the four ethnic groups, respectively. In the meta-regression model for the Han group, the results of mean age and mean BMI were statistically significant, explaining $40.96\%$ ($$p \leq 0.015$$) and $35.16\%$ ($$p \leq 0.032$$) of the heterogeneity in prevalence between studies, respectively. The mean BMI, alcohol use, and study year explained $54.21\%$ ($$p \leq 0.006$$), $66.83\%$ ($$p \leq 0.001$$), and $54.20\%$ ($$p \leq 0.003$$) of the heterogeneity in awareness between studies, respectively. And the mean BMI, alcohol use, study year, and tobacco use were statistically significant, explaining $48.44\%$ ($$p \leq 0.011$$), $53.19\%$ (ws), $40.19\%$ ($$p \leq 0.009$$), and $26.99\%$ ($$p \leq 0.041$$) of the heterogeneity in treatment or control between studies (See Table 2). In the meta-regression model for the Mongolian group, the source of heterogeneity for prevalence was mean age (Adj. R2 = $69.12\%$, $$p \leq 0.004$$), and education (Adj. R2 = $46.36\%$, $$p \leq 0.039$$) for control rate (see Table 3). In addition, the residence (Adj. R2 = $100.00\%$, $$p \leq 0.048$$) affected the awareness rate of the Tibetan group (see Supplementary Table 3). The mean BMI (Adj. R2 = $76.68\%$, $$p \leq 0.034$$) and study year (Adj. R2 = $57.40\%$, $$p \leq 0.036$$) also had effects on the prevalence of the Yi group, respectively (see Supplementary Table 4).
## 3.5. Bias and Sensitivity Analysis
31 ($68.9\%$) studies were judged to be of high quality with a low risk of bias, while 14 ($31.1\%$) studies were at moderate risk of bias. In the sensitivity analyses, no individual study was found to have any major impact on the pooled prevalence, awareness, treatment, and control rate of hypertension (see supplementary Figures 1–4). The Egger test was not statistically significant for testing the publication bias, except in analyses of pooled awareness of the Han population ($$p \leq 0.0002$$) and the pooled treatment of the Tibetan population ($$p \leq 0.0005$$). The Begg test showed no small-study bias in these meta-analyses, all the p values were greater than 0.05.
## 4. Discussion
To our knowledge, this study is the first to systematically summarize the rates of hypertension management in multiple Chinese ethnic groups. In this study, we finally selected four main ethnic groups (Han, Tibetan, Yi, and Mongolian) in our analyses according to the number of databases. Our results indicated that the status varied in these four groups, in which Mongolian had the highest prevalence, treatment, and control rates. However, Yi people had a different story of increasing the prevalence rate and relatively lower the control rate. The results denoted that understanding the differences in varied ethnic groups could inform targeted prevention measures to improve hypertension management and reduce the disparities.
Based on previous findings, the prevalence rates of hypertension would vary in different ethnic populations due to genetic and environmental factors [14, 15]. In our study, the combined prevalence among four ethnic groups was $31.7\%$, which was slightly higher than the results from the Chinese Hypertension Survey in 2012–2015 which found that the average rate of the Chinese population was $27.9\%$ [16]. In these ethnic groups, the rate of Han nationality was the lowest ($27.0\%$), while the rates of Mongolian and Tibetan were more than $30\%$. These results were consistent with previous studies [17]. Most of the Tibetan population in China are living in the highlands, and some publications indicated that chronic hypoxia exposure may cause a higher prevalence of hypertension [18]. In addition, according to our previous studies, hypoxia exposure may also increase the risk of other diseases, such as sleep disorders, which were regarded as the risk factors for hypertension [19, 20]. For the Mongolian population, the prevalence was high in the recent thirty years in some large-scale surveys [21]. This phenomenon may be attributable to the changes in their diet culture because Mongolian people prefer to consume meat and milk as the principal food, especially among people living in pastoral regions [22]. Yi population was found to have a low prevalence of both overweight and obesity [23], which may contribute to a lower rate of hypertension ($28\%$). However, the population attribution fraction (PAF) of overweight or obesity in this population was also increasing in recent years. For example, in 1996, the PAF was $27.66\%$ and the rate of hypertension was $5.33\%$ in 1996, while the PAF increased to $33.26\%$ in 2015 with a prevalence of hypertension of $17.2\%$ [24].
Furthermore, among people with hypertension, the awareness, treatment, and control rates were also analyzed in this study. In China, the average rates were $51.5\%$, $46.1\%$, and $16.9\%$ reported by a nationwide survey in 2015 [25], which were higher than the rates reported in other developing countries, such as Ghana [26]. In this study, only the rates from the Mongolian population were close to this level, and the Tibetan and Han population slightly lower than that level, while these rates in the Yi group were the lowest. Two possible reasons might account for these results. Firstly, compared to the other groups, Yi people had a relatively lower prevalence of hypertension over a long-term period, which may make them pay less attention to the increase of risk factors at the population level. For example, some studies had reported that Yi people's dietary patterns changed largely in the past two decades, which caused the gradual growth of the prevalence of hypertension [27, 28]. Secondly, the previous studies showed that numerous Yi people migrated from rural to urban areas, which imposed heavy pressure on the prevention of these two distinct groups. Yi migrants usually had a higher prevalence of hypertension than Yi farmers, and it was useless to implement the same measures to improve their awareness, treatment, and control situation among Yi people. More targeted strategies and policies are needed in the Yi population.
There are some strengths in our study. The pooled estimates of the prevalence, awareness, treatment, and control of hypertension in four main Chinese ethnic groups were reported comprehensively. The search strategy involved plenty of studies and subjects based on the standard of PRISMA [29]. Our results were robust and corroborated by further sensitivity analyses and the tests of small-study bias and publication bias. Nonetheless, there are also some limitations. Firstly, high heterogeneity was found in our meta-analyses, which were familiar with other meta-analytic studies focusing on the pooled rates because of the differences in the study population, inclusion and exclusion criteria, and assessment [30, 31]. In this systematic review, we also included some common covariates to explore the source of heterogeneity, such as BMI, gender, tobacco use, alcohol use, age, study year, and education. Of which, mean age, mean BMI, study year, alcohol use, and tobacco use were significantly associated with the rates of hypertension management. Secondly, some other Chinese ethnic groups, such as Zhuang, Uygur, and Hui, were not included in the meta-analyses because of insufficient datasets.
## 5. Conclusions
The systematic review included four main Chinese ethnic groups and conducted meta-analyses for prevalence, awareness, treatment, and control rates of hypertension. The findings suggest that hypertension management varied in these Chinese ethnic groups which Mongolian and Tibetan had higher prevalence rates. Notably, the Yi population was regarded as the low-risk population for hypertension, and it should pay more attention to prevention because the prevalence rates were rising in recent years, while their awareness, treatment, and control rates were still low. We call for further studies to better understand the specific risk factors and mechanisms of hypertension in the Tibetan and Mongolian populations to make more targeted preventive measures. In addition, for Yi people, it is urgent to implement interventions to improve their awareness and treatment, and finally achieve the goal of hypertension control.
## Data Availability
The data used to support the findings of this research are available from major databases, including Embase, PubMed, and Web of Science.
## Conflicts of Interest
The authors declare that they have no conflicts of interest.
## Authors' Contributions
Y.YN and H.KL contributed equally. They extracted the data, performed the analysis, and wrote the first draft. Z.YW, W.XM, and C.WZ were involved in the data analysis. Z.ZQ conducted the research and revised the draft. G.DQ and Z.ZQ contributed equally and supervised the data extraction and interpretation of the statistics.
## References
1. **Worldwide trends in blood pressure from 1975 to 2015: a pooled analysis of 1479 population-based measurement studies with 19·1 million participants**. (2017) **389** 37-55. DOI: 10.1016/S0140-6736(16)31919-5
2. **Cardiovascular disease, chronic kidney disease, and diabetes mortality burden of cardiometabolic risk factors from 1980 to 2010: a comparative risk assessment**. (2014) **2** 634-647. DOI: 10.1016/S2213-8587(14)70102-0
3. Dhungana R. R., Pandey A. R., Shrestha N.. **Trends in the prevalence, awareness, treatment, and control of hypertension in Nepal between 2000 and 2025: a systematic review and meta-analysis**. (2021) **2021** 8. DOI: 10.1155/2021/6610649
4. Kuate Defo B., Mbanya J. C., Tardif J. C.. **Diagnosis, prevalence, awareness, treatment, prevention, and control of hypertension in Cameroon: protocol for a systematic review and meta-analysis of clinic-based and community-based studies**. (2017) **6** p. e102. DOI: 10.2196/resprot.7807
5. Mohammadian Khonsari N., Shahrestanaki E., Ejtahed H. S.. **Long-term trends in hypertension prevalence, awareness, treatment, and control rate in the Middle East and north africa: a systematic review and meta-analysis of 178 population-based studies**. (2021) **23** p. 41. DOI: 10.1007/s11906-021-01159-0
6. Redfern A., Peters S. A. E., Luo R.. **Sex differences in the awareness, treatment, and control of hypertension in China: a systematic review with meta-analyses**. (2019) **42** 273-283. DOI: 10.1038/s41440-018-0154-x
7. Miranda R., Meeks K. A. C., Snijder M. B.. **Health literacy and hypertension outcomes in a multi-ethnic population: the HELIUS study**. (2020) **30** 516-521. DOI: 10.1093/eurpub/ckz174
8. Hill L. K., Thayer J. F.. **The autonomic nervous system and hypertension: ethnic differences and psychosocial factors**. (2019) **21** p. 15. DOI: 10.1007/s11886-019-1100-5
9. Chan G. C., Teo B. W., Tay J. C.. **Hypertension in a multi-ethnic Asian population of Singapore**. (2021) **23** 522-528. DOI: 10.1111/jch.14140
10. van der Linden E. L., Couwenhoven B. N., Beune E. J. A. J., Daams J. G., van den Born B. J. H., Agyemang C.. **Hypertension awareness, treatment and control among ethnic minority populations in Europe: a systematic review and meta-analysis**. (2021) **39** 202-213. DOI: 10.1097/hjh.0000000000002651
11. Heizhati M., Wang L., Yao X.. **Prevalence, awareness, treatment and control of hypertension in various ethnic groups (Hui, Kazakh, Kyrgyz, Mongolian, Tajik) in Xinjiang, Northwest China**. (2020) **29** 276-284. DOI: 10.1080/08037051.2020.1745055
12. Wells G., Shea B., O’connell D.. (2000)
13. Moher D., Liberati A., Tetzlaff J., Altman D. G.. **Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement**. (2009) **6**. DOI: 10.1371/journal.pmed.1000097
14. Dubroff J., Melendres L., Lin Y., Beene D. R., Ketai L.. **High geographic prevalence of pulmonary artery hypertension: associations with ethnicity, drug use, and altitude**. (2020) **10** 1-10. DOI: 10.1177/2045894019894534
15. Liu F., Adi D., Xie X.. **Prevalence of isolated diastolic hypertension and associated risk factors among different ethnicity groups in xinjiang, China**. (2015) **10**. DOI: 10.1371/journal.pone.0145325
16. **Chinese guidelines for prevention and treatment of hypertension-A report of the revision committee of Chinese guidelines for prevention and treatment of hypertension**. (2019) **16** 182-241. DOI: 10.11909/j.issn.1671-5411.2019.03.014
17. Yu P., Ning Y., Gao Y.. **Hypertension among Mongolian adults in China: a cross-sectional study of prevalence, awareness, treatment, control, and related factors: hypertension among Mongolian adults in China**. (2021) **23** 1786-1801. DOI: 10.1111/jch.14348
18. Brito J., Siques P., Pena E.. **Long-term chronic intermittent hypoxia: a particular form of chronic high-altitude pulmonary hypertension**. (2020) **10** 5-12. DOI: 10.1177/2045894020934625
19. Han B., Chen W. Z., Li Y. C., Chen J., Zeng Z. Q.. **Sleep and hypertension**. (2020) **24** 351-356. DOI: 10.1007/s11325-019-01907-2
20. Wang L., Hu Y., Wang X., Yang S., Chen W., Zeng Z.. **The association between sleep duration and hypertension: a meta and study sequential analysis**. (2021) **35** 621-626. DOI: 10.1038/s41371-020-0372-y
21. Li W. J., Tian J., Zhang L. P.. **Epidemiological characteristics of hypertension in Mongolian population**. (2019) **19** 37-38
22. Xiu L.. **A study on the changes of Mongolian culinary culture with the dissemination of agricultural knowledge to the northern part of China**. (2020) **52** 10-17
23. Chen L., Zong Y., Wei T.. **Prevalence, awareness, medication, control, and risk factors associated with hypertension in Yi ethnic group aged 50 years and over in rural China: the Yunnan minority eye study**. (2015) **15** p. 383. DOI: 10.1186/s12889-015-1687-4
24. Zhang J., Wan S., Dong F.. **Secular trends of the impact of overweight and obesity on hypertension in Yi people: Yi migrant study**. (2020) **2020** 9. DOI: 10.1155/2020/5368357
25. Wang Z., Chen Z., Zhang L.. **Status of hypertension in China: results from the China hypertension survey, 2012-2015**. (2018) **137** 2344-2356. DOI: 10.1161/circulationaha.117.032380
26. Bosu W. K., Bosu D. K.. **Prevalence, awareness and control of hypertension in Ghana: a systematic review and meta-analysis**. (2021) **16**. DOI: 10.1371/journal.pone.0248137
27. Shan G., Wei D., Wang C.. **Trends of overweight and obesity in Yi people between 1996 and 2007: an Yi migrant study**. (2011) **24** 467-474. DOI: 10.3967/0895-3988.2011.05.003
28. Li M., Shi Z.. **Dietary pattern during 1991-2011 and its association with cardio metabolic risks in Chinese adults: the China health and nutrition survey**. (2017) **9** p. 1218. DOI: 10.3390/nu9111218
29. Arya S., Kaji A. H., Boermeester M. A.. **PRISMA reporting guidelines for meta-analyses and systematic reviews**. (2021) **156** 789-790. DOI: 10.1001/jamasurg.2021.0546
30. Sun L., Feng Y.. **Can results of meta-analysis with high heterogeneity provide any predictive values?**. (2019) **40** p. 3205. DOI: 10.1093/eurheartj/ehz530
31. Alba A. C., Alexander P. E., Chang J., MacIsaac J., DeFry S., Guyatt G. H.. **High statistical heterogeneity is more frequent in meta-analysis of continuous than binary outcomes**. (2016) **70** 129-135. DOI: 10.1016/j.jclinepi.2015.09.005
|
---
title: Ginsenoside Rg1 Improves Inflammation and Autophagy of the Pancreas and Spleen
in Streptozotocin-Induced Type 1 Diabetic Mice
authors:
- Yi Zong
- Weihua Yu
- Hanghang Hong
- Zhiqiang Zhu
- Wenbo Xiao
- Kewu Wang
- Guoqiang Xu
journal: International Journal of Endocrinology
year: 2023
pmcid: PMC10030220
doi: 10.1155/2023/3595992
license: CC BY 4.0
---
# Ginsenoside Rg1 Improves Inflammation and Autophagy of the Pancreas and Spleen in Streptozotocin-Induced Type 1 Diabetic Mice
## Abstract
### Background
Ginsenoside Rg1 (Rg1) is one of the key bioactive components of the precious Traditional Chinese Medicine that has been used to treat diabetes in China. Ginsenosides have been reported to protect diabetics from tissue damage, inflammation, and insulin resistance. Type 1 diabetes (T1D) is an organ-specific autoimmune disease that occurred frequently among adolescents over the world, its development was related to inflammation and β-cells immunodeficiency. The aim of this study is to explore the biological mechanism of Rg1 on inflammation and autophagy of β-cells in T1D and its therapeutic potential.
### Methods
The model of T1D mice was established by injecting Streptozotocin (STZ) (55 mg/kg) or citric acids once a day for 5 days and from the fourth day of injection, mice were administered with Rg1 (20 mg/kg) or saline by gavage once a day for 12 days. Hematoxylin-eosin staining, immunofluorescence, ELISA, quantitative real-time PCR, and Western blot were used to observe the histopathological changes, inflammatory factor levels, and autophagy markers after administration of ginsenoside Rg1.
### Results
Compared to the T1D mice, Rg1 improved the weight ($p \leq 0.05$) and blood glucose ($p \leq 0.01$) of mice, advanced the injury and apoptosis of β-cells in islets ($p \leq 0.01$), and markedly inhibited the protein expression degrees of CD45, CXCL16, ox-LDL, and TF in the pancreas and spleens ($p \leq 0.01$), also activated the degrees of insulin in serum ($p \leq 0.01$). Besides, in T1D mice' pancreas and spleen, Rg1 markedly repressed the IL-1β, TNF-α, and NOS2 mRNA levels ($p \leq 0.05$ or $p \leq 0.01$), inhibited the CXCL16, NF-κB, and TF proteins ($p \leq 0.05$ or $p \leq 0.01$), while elevating the ratio of LC3 II/I ($p \leq 0.01$) and P62 ($p \leq 0.05$) protein level.
### Conclusions
This study proved that Rg1 protected mice against T1D possibly by improving islet injury and tissue inflammation, raising serum insulin, and tissue autophagy marker.
## 1. Introduction
Type 1 diabetes (T1D) is an organ-specific autoimmune disease with selective destruction of β-cells in islets and dysfunction of insulin secretion, affecting more than 490 thousand of the world's children [1]. Epidemiological studies suggest that childhood obesity, eating habits, virus infection, and other environmental factors are related to the onset of T1D [2]. Additionally, the destruction of the β-cell is one of the reasons to result in the reduction of insulin, the uncontrolled production of glucose, and hyperglycemia [3]. Inhibition of hyperglycemia has always been the focus of T1D treatment [4]. Additionally, modern science focused on preventing or postponing β-cell loss in T1D [5, 6]. Scientists reported that the autoimmune response in T1D resulted in an inflammatory state in β-cells [7]. Clinical evaluation of children with high genetic risk found that inflammation, cytotoxicity, angiogenesis, and antigen-presenting cell activity increased in children progressing towards islet autoimmunity [8]. Also, Park's team noticed that Alpha-1 antitrypsin with anti-inflammatory properties favorably impacted the development of T1D in mice [9]. It suggests that the anti-inflammatory treatment of T1D may be effective.
Scientists found that many plant extracts have the potential to improve diabetes [10–12]. Extracts of *Olea europaea* treatment reduced fasting blood glucose in diabetes animals [13]. Scientists reported that the extract of *Malva neglecta* Wallr inhibited the level of oxidative stress in diabetes animals [14, 15]. Ginsenoside Rg1 (Rg1) is one of the key bioactive components of Panax ginseng C. A. Mey (family Araliaceae), a precious Traditional Chinese Medicine that has been used to treat diabetes in China [16, 17]. Rg1 is a triterpenoid saponin containing a protopanaxatriol structure [18]. In recent years, scientists have discovered that ginsenosides can protect tissues such as the heart, pancreas, and spleen, antagonize inflammation, and improve insulin resistance in diabetes [19–21]. Research pointed out that Rg1 inhibited the IL-1β and IL-18 levels in T1D mice [22]. Similarly, Yu and colleagues studied that Rg1 could prevent the high glucose-/palmitate-induced damage in H9C2 cells via the AKT-GSK-3beta-Nrf2 pathway [23]. And Luo and colleagues reported that Rg1 blocked the pro-inflammatory effects of lipopolysaccharide on neonatal rat cardiomyocytes [24]. In addition, the regulation of autophagy in β-cells is important to maintain the stability of insulin [25]. A study reported that inhibiting autophagy promoted the IL-1β level [26]. And extracts of ginsenoside can induce protective autophagy in HepG2 cells [27] and regulate autophagy in acute liver injury [28]. Modern research argues that inflammation and autophagy play crucial roles in the treatment of diabetes [29, 30]. However, there was still a lack of research on the biological effect of Rg1 on T1D. To explore the biological mechanism of Rg1 on inflammation in T1D and study its therapeutic potential, this study constructed the T1D mouse model by injecting STZ and proposed the possibility that Rg1 ameliorates inflammation in T1D mice via elevating autophagy.
## 2.1. Animals and Treatments
Male C57BL/6 mice (25–30 g) aged seven to eight weeks were provided by Shanghai Ling Chang Biotech Co., Ltd. (Shanghai, CHN). All animal tests were performed according to the guidelines of the Institutional Animal Care and Use Committee and approved by the Animal Experimentation Ethics Committee of the Zhejiang Eyong Pharmaceutical Research and Development Center (Certificate No. SYXK (Zhe)2021-0033). A temperature-, humidity-, and light-controlled animal room (20°C, $60\%$ humidity, and a 12 h light/dark cycle) was used to house all mice with free access to food and water for 7 days. The mice were divided into 4 groups with 6 mice in each group: the mice in the DM group and the DM + Rg1 group were injected intraperitoneally with Streptozotocin (STZ) (55 mg/kg) ($98\%$, S817944, Mackin, Shanghai, CHN) that was dissolved in citric acids (CA, 50 mM, pH4.5) once a day for 5 days; the mice in the control group and Rg1 group were injected intraperitoneally with CA for 5 days, whose volume was equal to the amount of STZ in the DM group. Additionally, the mice in the Rg1 group and DM + Rg1 group were further administered with Rg1 (20 mg/kg) (≥$98\%$, G909436, Mackin, Shanghai, CHN) by gavage once a day for 12 days since the fourth day of STZ/CA injection. The time axis of animal treatments is shown in Figure 1(a). The dose of Rg1 was set according to previous studies on Rg1 [22, 31].
## 2.2. Evaluation of Type 1 Diabetic Mice Modeling
On the 7th day after the last STZ was injected, the tail vein blood of mice was randomly taken to measure blood glucose for 3 days, and the weight of the mice was recorded. The level of blood glucose was greater than or equal to 16.7 mmol/L for 3 consecutive days, indicating the success of the T1D model.
## 2.3. Collection of Serum and Tissue Samples
On the 10th day after the last injection of STZ, the mice were anesthetized with ether and euthanasia with carbon dioxide. The pancreas and spleen were collected and kept in $10\%$ saline formalin. The serum was separated by centrifugation at 4000 rpm for 15 min from the blood that was collected from the retroorbital venous plexus by small capillary tubes before euthanasia. Tissue and serum were stored at −80°C for subsequent experiments.
## 2.4. Detection of ox-LDL and Insulin (INS) in Serum
According to the instructions of the ox-LDL (MM-0908M1, Meimian, Jiangsu, CHN) and INS (MM-0579M1, Meimian, Jiangsu, CHN) ELISA kits, serum samples were added to the enzyme label plate to incubate at 37°C for 60 min. After washing the plate, chromogenic agents A and B that were added to the plate were well-mixed and stood in the dark for 15 min, before being finally added to be termination solution to terminate the reaction. The absorbance (expressed as an OD value) was detected at a wavelength of 450 nm, and there were 6 parallel setups in each group.
## 2.5. Hematoxylin-Eosin (HE) Staining of the Pancreas
Pancreases have been fixed, dehydrated with an alcohol gradient, made transparent with xylene and alcohol, and then embedded in paraffin. The tissues with paraffin wrapping were made into 5 μm sections and stained by HE staining kits (G1003, ServiceBio, Wuhan, CHN). An optical microscope was used to observe the tissues. The histology of the pancreas was evaluated according to the islet volume and the regular edges of the islet, and the score was proportional to the degree of injury.
## 2.6. Immunofluorescence of CXCL16, CD45, Insulin Receptor (INSR), ox-LDL, and Tissue Factor (TF) in the Pancreas and Spleen
The sections were repaired with an EDTA antigen repair buffer (pH = 8, G1206, Servicebio, CHN) in a microwave, and the $3\%$ BSA was added dropwise to block for 30 min. Then the CD45R (1: 50, sc19597, Santa, US), INSR-β antibody (1: 50, sc57342, Santa, US), CXCL16 (1: 100, DF13312, Affinity, CHN), anti-LDL receptor antibody (1: 100, ab30532, Abcam, US) and anti-TF antibody (F-8) (1: 50, sc-373785, SCBS, US) were used for incubating with the sections overnight at 4°C, and then after washing, the tissue that was incubated with the goat antirabbit IgG H&L (ab150078, Abcam, US) that was incubated in dark for 50 min. Following PBS washing, the nuclei were counterstained with DAPI (G1012, ServiceBio, CHN). Thirdly, after being washed, the sections were quenched by the antifluorescence quenching sealing reagent (G1401, Servicebio, CHN) for 5 min and rinsed with running water for 10 min. Between every two steps, the sections were washed 3 times with PBS (pH = 7.4) on the shaking table. Finally, a NIKON eclipse upright microscope was used to observe and image the fluorescence (the nucleus is blue, and the positive expression is red or green).
## 2.7. Quantitative Real-Time PCR Analysis (qRT-PCR) of Inflammatory Factors in the Pancreas and Spleen
The tissue was lysed with the Total RNA Extractor (B511311, Sangon Biotech, CHN) and centrifuged to remove the precipitation, then the supernatant was dissolved in chloroform and centrifuged again to separate the supernatant. The supernatant was soaked in isopropanol for 15 minutes to obtain precipitation, and then the precipitation was soaked in $75\%$ absolute ethanol. Finally, the dried precipitate was dissolved in DEPC water and subjected to reverse a transcription reaction by HiFiScript cDNA Synthesis Kit (CW2569, CWBIO, CHN). The qRT-PCR was carried out according to the following procedure: denaturation at 95°C for 10 min and 95°C for 15 s, then at 60°C for 60 s, performing 40 cycles. Primer information is shown in Table 1.
## 2.8. Western Blot of CXCL16, NF-κB/P65, TF, LC3, and P62 in the Pancreas and Spleen
The tissue was soaked in RIPA lysate and homogenized on ice; then it was centrifuged. The total protein concentration was measured by the BCA Protein Assay Kit (PC0020, Solarbio, CHN) and mixed with a loading buffer (1: 4) in a 100°C water bath for 5 min. The SDS-PAGE and PVDF membranes were used to separate the protein by vertical electrophoresis and wet transfer. Following blocking by $5\%$ BSA, the sample was incubated with the primary antibody for one night. Secondarily, the sample was incubated with HRP-conjugated goat IgG secondary antibodies at 20°C for 1 h. After wet transfer, the sample was cleaned 3 times/10 min with TBST between every two steps. Finally, ECL luminescent reagent was used to present the photo in the chemiluminescence imager. The antibodies for CXCL16 (1: 2000, DF13312) and GAPDH (1: 5000, AF7021) were purchased from Affinity Biosciences, Ltd. (Jiangsu, CHN). And NF-κB (1: 1000, 6956T), LC3 (1: 1000, 4599T), and P62 (1: 1000, 5114T) antibodies were provided by Cell Signaling Technology, Inc. (MA, US). The TF antibody (1: 100, sc-373785) was bought from Santa Cruz Biotechnology, CA (MO, US).
## 2.9. Statistical Analyses
Data analysis by SPSS 16.0. The comparisons of multiple groups were analyzed by one-way ANOVA followed by Tukey, and the t-test was used to analyze the differences between every two groups. If it is a multigroup comparison with a nonnormal distribution or uneven variance, the Kruskal−Wallis H-Test was used. All data were expressed as mean ± standard deviation (mean ± SD), $p \leq 0.05$ meaning that the difference was statistically significant.
## 3.1. Rg1 Improved the Blood Glucose and Body Weight in T1D Mice
On day 7 after the first intraperitoneal injection of STZ, the blood glucose of mice ≥16.7 lasted for 3 days, which means that the T1D models in the DM group and the DM + Rg1 group were successfully constructed (Figure 1(b)). At the end of the 15th day of Rg1 gavage, the blood glucose in the DM group was remarkably higher than that in the control group. While the blood glucose in the DM + Rg1 group was remarkably lower than that in the DM group (Figures 1(b) and 1(d)). The weight of mice in the DM group was lower than that in the control group, while the weight of mice in the DM + Rg1 group was higher than that in the DM group on day 15 (Figures 1(c) and 1(e)).
## 3.2. Rg1 Ameliorated the Pathological Damage to the Pancreas in T1D Mice
The type 1 diabetic mouse model was built successfully, as shown in Figure 2, the results of pancreatic HE staining. In the control and Rg1 groups, the islet was plump, and the edges of islet cells were evenly regular (Figures 2(a) and 2(b)). While in the DM group, the size and edges of the islets were uneven (Figure 2(c)). And Rg1 treatment could improve the histology of the injury (Figure 2(d)). Compared with the control group, the score of the DM group was notably higher ($p \leq 0.01$). And the score of the DM + Rg1 group was immensely lower (Figure 2(e)).
## 3.3. Rg1 Regulated the Expressions of CXCL16, CD45, and INSR in the Pancreas
The colocalization of CD45 or INSR protein with CXCL16 in the pancreas was observed by immunofluorescence (Figures 3(a)–3(f)). The CD45 positive expression was obviously detected in the DM group, which was notably higher than that in the control group ($p \leq 0.01$), and the expression of the colocated CXCL16 was also higher in the DM group too ($p \leq 0.01$). Then it was obvious that the expressions of CD45 and CXCL16 proteins decreased enormously by Rg1 treatment in DM mice ($p \leq 0.01$) (Figures 3(b) and 3(c)). By analyzing the level of INSR expression, it was found that there were no statistical differences in it among the four groups ($p \leq 0.05$). Then the expression of CXCL16 was significantly upregulated in the DM group compared to the control group, while the CXCL16 expression was significantly reduced by Rg1 treatment ($p \leq 0.01$) (Figures 3(d)–3(f)).
## 3.4. Rg1 Regulates the Expression of CXCL16 and CD45 Proteins in the Spleen
Through an immunofluorescence assay, we observed the colocalization of CXCL16 and CD45 in the spleen (Figures 4(a)–4(c)). The levels of CD45 and CXCL16 proteins were markedly increased in the DM group compared to the control group ($p \leq 0.01$). While in the DM + Rg1 group, the levels of CD45 and CXCL16 were both lower than that in the DM group ($p \leq 0.01$).
## 3.5. Rg1 Regulates the Expression of ox-LDL an TF Proteins in the Pancreas and Spleen
The degrees of ox-LDL and TF proteins in the pancreas and spleen of T1D mice were observed by immunofluorescence (Figures 5 and 6). In the pancreas and spleen, compared with the control group, the degrees of ox-LDL and TF proteins in DM mice in the Rg1 group were not significant ($p \leq 0.05$), the expression of these two proteins in the DM group was significantly upregulated ($p \leq 0.01$). The ox-LDL and TF proteins levels of the pancreas and spleen in the DM + Rg1 group compared with the DM group was markedly reduced ($p \leq 0.01$ and $p \leq 0.05$).
## 3.6. Rg1 Inhibited the Level of ox-LDL and Advanced the Degree of INS in Serum
As Figures 7(a) and 7(b) show, according to the results of the ELISA, we noticed that Rg1 treatment inhibited the levels of ox-LDL and advanced the expression degree of INS in the serum of T1DM. In the DM group, the level of ox-LDL was immensely upregulated ($p \leq 0.01$), and the INS level was repressed markedly more than that in the control group ($p \leq 0.01$). While in the DM + Rg1 group, compared with the DM group, the degree of ox-LDL was immensely inhibited ($p \leq 0.05$), and the INS level was significantly increased ($p \leq 0.01$).
## 3.7. Rg1 Restrained the Transcription Level of Inflammatory Factors in the Pancreas and Spleens
The transcription degrees of IL-1β, TNF-α, and NOS2 were semiquantified by qRT-PCR. The results are shown in Figures 7(c)–7(h), in the DM group, the degrees of IL-1β, TNF-α, and NOS2 mRNA in the pancreas and spleen were notably superior to those in the control group ($p \leq 0.01$). And in the DM + Rg1 group, the transcription degrees of IL-1β, TNF-α, and NOS2 in the pancreas and spleens were inhibited compared with those in the DM group ($p \leq 0.05$) (Figures 7(c) and 7(h)).
## 3.8. Rg1 Suppresses the Expression of CXCL16, NF-κB, TF, and Activated LC3 and P62 Proteins in the Pancreas and Spleen of DM Mice
By WB analyzing, in the DM group, the expression of CXCL16, NF-κB, and TF proteins in the pancreas and spleen were raised prominently compared to the control group ($p \leq 0.01$), and the degrees of CXCL16, NF-κB, and TF proteins were also huge down-regulated in the Rg1 treatment T1DM mice ($p \leq 0.05$) (Figures 8(a)–8(c) and 8(g)–8(i)).
On the contrary, the LC3 and P62 protein levels in the DM group were inhibited significantly compared with the control group ($p \leq 0.01$) (Figures 8(d), 8(e), 8(j) and 8(k)). And compared with the DM group, the expression degrees of LC3 protein in the pancreas and spleen were elevated enormously in the Rg1 treatment T1DM mice ($p \leq 0.01$) (Figures 8(d) and 8(j)). As well as the level of P62 protein in the pancreas and spleen was raised markedly in the DM + Rg1 group ($p \leq 0.05$) (Figures 8(e) and 8(k)).
## 4. Discussion
The β-cells in islets are the main sites to regulate INS. Islet atrophy and apoptosis of β-cells in T1D rats is increased. Research reported that improved the apoptosis of β-cells and the morphology of islets were able to reduce blood glucose levels in the model rats [32]. This report discovered that Rg1 treatment on DM mice advanced the mice's weight, improved the number and morphology of islets, and inhibited the blood glucose level. Also, Rg1 treatment raised the serum INS level. It reminded that Rg1 can improve the function of the islet in T1D mice. Besides, a piece of evidence displayed an increased ox-LDL level in T1DM patients [33]. The increase of ox-LDL induced lipid accumulation and an inflammatory response, which caused damage to tissues [34, 35]. And activation of ox-LDL could induce the inflammatory response and promote the secretion of inflammatory factors, such as TF, the inflammatory biomarker, which would be activated through the CXCL16/ox-LDL pathway in β-cell [36, 37]. CXCL16, a special chemokine, can combine with immune cells to drive into the inflammatory part, and some can internalize ox-LDL [38, 39]. This study observed the colocalized expression of INSR or CD45 with CXCL16 by immunofluorescence and found there was inflammatory infiltration in the pancreas of T1D mice, and Rg1 treatment could significantly improve this situation. The above experimental results indicated that Rg1 had the potential to improve T1D mice by adjusting ox-LDL.
Furthermore, this study found IL-1β and TNF-α levels were repressed significantly by Rg1 treatment in the pancreas of DM model mice. A study reported that in macrophages of diabetics, the IL-1β and TNF-α levels increased significantly [40–42]. Additionally, scientists have proved that β-cell injury in T1D was accompanied by the up-regulation of inflammatory factors, NF-κB and iNOS, while the expression levels of them were decreased after improving the injury [43]. Dampening NF-κB-iNOS-NO pathway negatively regulated the inflammation and apoptosis of β-cells [44]. Through qRT-PCR and WB, this study found the expression of NF-κB and iNOS was antagonized by Rg1 treatment. In short, it is suggested that Rg1 can antagonize the inflammation of the pancreas in T1D mice via inhibiting the NF-κB-iNOS signaling pathways.
Additionally, scientists have proven that immune deficiency in diabetes is associated with the damage of the spleen [45]. Moreover, splenocytes from a diabetic animal can transfer the T1D to healthy recipients [46]. Therefore, this study focused on the effect of Rg1 on the spleen of T1D mice and found it could improve the inflammation in the spleen. Studies have reported IL-1β, TNF-α, and NOS2 were closely related to macrophages polarization and macrophages played an important role in the development of inflammation [47]. The results of this study indicated that the levels of macrophage polarization-related cytokines were enormously upregulated in the spleen of T1D mice, and Rg1 treatment was able to reduce them in the spleen. Rg1 treatment could significantly inhibit macrophage polarization and improve inflammation in the spleen of T1D mice.
Moreover, this study also analyzed the levels of autophagy markers, LC3 and P62, in the pancreas and spleen [44]. Additionally, the immune-related signaling molecules can regulate autophagy, which is essential to improving immune disorders and the inflammatory response [48]. In recent years, scientists have proposed that autophagy disorder of islets of T1D patients, and autophagy can reduce the harmful effects of ER stress and DNA damage by delaying apoptosis of β-cells [49, 50]. A study about metformin reported that increasing autophagy can improve Th17 inflammation [51]. This study discovered that Rg1 stimulated the expression of LC3 and P62 proteins suggesting that Rg1 has a promoting effect on the autophagy of β-cells and spleen tissues. This report found that the effect of Rg1 on autophagy in T1D mice may be related to CXCL16 and NF-κB. A study reported that the incretion of ox-LDL sensitized adipocytes to the lower insulin-induced glucose uptake and increased the levels of NF-κB and the marker of apoptosis and autophagy (Bnip3), suggesting the ox-LDL was related to inflammation, apoptosis, and autophagy [52]. The effect of Rg1 on autophagy in T1D deserves further study. Additionally, this study has some shortcomings. If multiple dose groups of Rg1 can be set, the results would be more convincing in this study, and this study will further explore the effects of multi-dose Rg1 in the future. In summary, this paper studied the pharmacology efficacy of the mechanism of Rg1 improving T1D, which indicated that the protective impact of Rg1 on T1D might be associated with antagonizing inflammation and improving autophagy disorder. This team has planned to further explorate the relevant targets in the follow-up research.
## 5. Conclusion
The study successfully constructed T1D mice and revealed the biological action of Rg1 on improving T1D by antagonizing inflammation and improving autophagy disorders. It proved that the Rg1 ameliorated the blood glucose, body weight, pancreas histological damage. Also, this study proved that Rg1 decreased the INS level in serum and pancreas, and reduced the levels of CD45, CXCL16, INS, ox-LDL, and TF in the spleen and pancreas. And the inhibition of Rg1 on the mRNA levels of IL-1β, NOS2, and TNF-α in the spleen and pancreas suggested that Rg1 ameliorates T1D of mice by inhibiting inflammation. Interestingly, this study proved that the Rg1 treatment on T1D mice could raise the ratio of LC3 II protein to LC3 I protein which suggesting that Rg1 has potential to advance the autophagy to treatment of T1D. This study provides a scientific basis for the Rg1 treatment for T1D and a new direction for investigation of its biological mechanism.
## Data Availability
The data supporting the findings of this study are available from the corresponding author, G. X., upon reasonable request.
## Ethical Approval
All animal tests were performed according to the guidelines of the Institutional Animal Care and Use Committee and approved by the Animal Experimentation Ethics Committee of Zhejiang Eyong Pharmaceutical Research and Development Center (Certificate No. SYXK (Zhe) 2021-0033).
## Disclosure
Yi Zong and Weihua Yu contributed equally to this study and should be considered co-first authors.
## Conflicts of Interest
All authors declare that they have no conflicts of interest.
## Authors' Contributions
Conception and design of the research by Kewu Wang and Guoqiang Xu; data collection by Yi Zong, Weihua Yu, Wenbo Xiao; statistical analysis by Yi Zong, Weihua Yu, Hanghang Hong and Zhiqiang Zhu, manuscript preparation by Yi Zong, Weihua Yu, Hanghang Hong, Zhiqiang Zhu and Wenbo Xiao; revision of manuscript by Kewu Wang and Guoqiang Xu. Funds were acquired by Kewu Wang and Guoqiang Xu. Yi Zong and Weihua Yu contribute equally.pt was done by Kewu Wang and Guoqiang Xu. Funds were acquired by Kewu Wang and Guoqiang Xu. Yi Zong and Weihua Yu contribute equally.
## References
1. Desai S., Deshmukh A.. **Mapping of type 1 diabetes mellitus**. (2020) **16** 438-441. DOI: 10.2174/1573399815666191004112647
2. Norris J. M., Johnson R. K., Stene L. C.. **Type 1 diabetes-early life origins and changing epidemiology**. (2020) **8** 226-238. DOI: 10.1016/s2213-8587(19)30412-7
3. Lehuen A., Diana J., Zaccone P., Cooke A.. **Immune cell crosstalk in type 1 diabetes**. (2010) **10** 501-513. DOI: 10.1038/nri2787
4. Greenbaum C., VanBuecken D., Lord S.. **Disease-modifying therapies in type 1 diabetes: a look into the future of diabetes practice**. (2019) **79** 43-61. DOI: 10.1007/s40265-018-1035-y
5. Maahs D. M., West N. A., Lawrence J. M., Mayer-Davis E. J.. **Epidemiology of type 1 diabetes**. (2010) **39** 481-497. DOI: 10.1016/j.ecl.2010.05.011
6. Sabir S., Saleem A., Akhtar M., Saleem M., Raza M.. **Increasing beta cell mass to treat diabetes mellitus**. (2018) **27** 1309-1315. DOI: 10.17219/acem/74452
7. Engin F.. **ER stress and development of type 1 diabetes**. (2016) **64** 2-6. DOI: 10.1097/jim.0000000000000229
8. Balzano-Nogueira L., Ramirez R., Zamkovaya T.. **Integrative analyses of TEDDY Omics data reveal lipid metabolism abnormalities, increased intracellular ROS and heightened inflammation prior to autoimmunity for type 1 diabetes**. (2021) **22**. DOI: 10.1186/s13059-021-02262-w
9. Park S. S., Rodriguez Ortega R., Agudelo C. W.. **Therapeutic potential of alpha-1 antitrypsin in type 1 and type 2 diabetes mellitus**. (2021) **57**. DOI: 10.3390/medicina57040397
10. Zulfqar F., Akhtar M. F., Saleem A., Akhtar B., Sharif A., Saleem U.. **Chemical characterization, antioxidant evaluation, and antidiabetic potential of Pinus gerardiana (Pine nuts) extracts**. (2020) **44**. DOI: 10.1111/jfbc.13199
11. Saleem M., Tanvir M., Akhtar M. F., Iqbal M., Saleem A.. **Antidiabetic potential of mangifera indica L. Cv. Anwar ratol leaves: medicinal application of food wastes**. (2019) **55**. DOI: 10.3390/medicina55070353
12. Saleem A., Muhammad F. A., Ayesha L., Mohamed M. A.-D.. **Chemical characterisation and appraisal of antidiabetic potential of Terminalia citrina extract in streptozotocin induced hyperglycaemia in Wistar rats**. (2021) **17**
13. Akhtar M. F., Ashraf K. M., Saleem A., Sharif A., Zubair H. M., Anwar F.. **Antidiabetic potential and antioxidant activity of Olea europaea subsp. cuspidata (Indian olive) seed extracts**. (2022) **2022** 12. DOI: 10.1155/2022/5164985
14. Akhtar M. F., Farooq U., Saleem A., Saleem M., Rahman M. H., Ashraf G. M.. **Ameliorating effect of Malva neglecta wallr on obesity and diabetes in wistar rats: a mechanistic study**. (2022) **2022** 15. DOI: 10.1155/2022/2614599
15. Saleem M., Hussain A., Akhtar M. F., Saleem A., Sadeeqa S., Naheed S.. **Ameliorating effect of Malva Neglecta on hyperglycemia and hyperlipidemia in diabetic rats**. (2021) **57**. DOI: 10.1590/s2175-97902020000418901
16. Shi Y., Wan X., Shao N., Ye R., Zhang N., Zhang Y.. **Protective and anti-angiopathy effects of ginsenoside Re against diabetes mellitus via the activation of p38 MAPK, ERK1/2 and JNK signaling**. (2016) **14** 4849-4856. DOI: 10.3892/mmr.2016.5821
17. Bai L., Gao J., Wei F., Zhao J., Wang D., Wei J.. **Therapeutic potential of ginsenosides as an adjuvant treatment for diabetes**. (2018) **9**. DOI: 10.3389/fphar.2018.00423
18. Tian M., Li L. N., Zheng R. R., Yang L., Wang Z. T.. **Advances on hormone-like activity of Panax ginseng and ginsenosides**. (2020) **18** 526-535. DOI: 10.1016/s1875-5364(20)30063-7
19. Lo S. H., Hsu C. T., Niu H. S., Niu C. S., Cheng J. T., Chen Z. C.. **Ginsenoside Rh2 improves cardiac fibrosis via ppar**. (2017) **18**. DOI: 10.3390/ijms18071364
20. Kim S. S., Jang H., Oh M.. **Ginsenoside Rg3 enhances islet cell function and attenuates apoptosis in mouse islets**. (2014) **46** 1150-1155. DOI: 10.1016/j.transproceed.2013.12.028
21. Yuan L.. **Effect of ginsenoside Rg1 on transplanted leukemia mouse model**. (2021) **44** 2571-2577
22. Gao Y., Li J., Chu S.. **Ginsenoside Rg1 protects mice against streptozotocin-induced type 1 diabetic by modulating the NLRP3 and Keap1/Nrf2/HO-1 pathways**. (2020) **866**. DOI: 10.1016/j.ejphar.2019.172801
23. Yu H., Zhen J., Yang Y., Du J., Leng J., Tong Q.. **Rg1 protects H9C2 cells from high glucose-/palmitate-induced injury via activation of AKT/GSK-3**. (2020) **24** 8194-8205. DOI: 10.1111/jcmm.15486
24. Luo M., Yan D., Sun Q.. **Ginsenoside Rg1 attenuates cardiomyocyte apoptosis and inflammation via the TLR4/NF-kB/NLRP3 pathway**. (2020) **121** 2994-3004. DOI: 10.1002/jcb.29556
25. Goginashvili A., Zhang Z., Erbs E.. **Insulin secretory granules control autophagy in pancreatic**. (2015) **347** 878-882. DOI: 10.1126/science.aaa2628
26. Crişan T. O., Plantinga T. S., van de Veerdonk F. L.. **Inflammasome-independent modulation of cytokine response by autophagy in human cells**. (2011) **6**. DOI: 10.1371/journal.pone.0018666
27. Zhang F., Xu H., Xia R.. **Pseudo-ginsenoside Rh2 induces protective autophagy in hepatocellular carcinoma HepG2 cells**. (2021) **16** 521-532. DOI: 10.2174/1574892816666210607100239
28. Zhao J., He B., Zhang S., Huang W., Li X.. **Ginsenoside Rg1 alleviates acute liver injury through the induction of autophagy and suppressing NF-**. (2021) **18** 1382-1389. DOI: 10.7150/ijms.50919
29. Turkmen K.. **Inflammation, oxidative stress, apoptosis, and autophagy in diabetes mellitus and diabetic kidney disease: the Four Horsemen of the Apocalypse**. (2017) **49** 837-844. DOI: 10.1007/s11255-016-1488-4
30. Tao T., Xu H.. **Autophagy and obesity and diabetes**. (2020) **1207** 445-461. DOI: 10.1007/978-981-15-4272-5_32
31. Yu H. T., Zhen J., Pang B., Gu J., Wu S.. **Ginsenoside Rg1 ameliorates oxidative stress and myocardial apoptosis in streptozotocin-induced diabetic rats**. (2015) **16** 344-354. DOI: 10.1631/jzus.b1400204
32. Gonçalves M. G., Medeiros M. A., de Lemos L. I. C.. **Effects of creatine supplementation on histopathological and biochemical parameters in the kidney and pancreas of streptozotocin-induced diabetic rats**. (2022) **14**. DOI: 10.3390/nu14030431
33. Rathsman B., Haas J., Persson M.. **LDL cholesterol level as a risk factor for retinopathy and nephropathy in children and adults with type 1 diabetes mellitus: a nationwide cohort study**. (2021) **289** 873-886. DOI: 10.1111/joim.13212
34. Wang Y., Zhang R., Zhang J., Liu F.. **MicroRNA-326-3p ameliorates high glucose and ox-LDL-IC- induced fibrotic injury in renal mesangial cells by targeting Fc**. (2018) **23** 1031-1038. DOI: 10.1111/nep.13168
35. Chen Y. P., Hsu H., Baskaran R.. **Short-term hypoxia reverses ox-LDL-induced CD36 and GLUT4 switching metabolic pathways in H9c2 cardiomyoblast cells**. (2017) **118** 3785-3795. DOI: 10.1002/jcb.26027
36. Liao Y., Zhu E., Zhou W.. **Ox-LDL aggravates the oxidative stress and inflammatory responses of THP-1 macrophages by reducing the inhibition effect of miR-491-5p on MMP-9**. (2021) **8**. DOI: 10.3389/fcvm.2021.697236
37. Darwish M. A., Abdel-Bakky M. S., Messiha B. A., Abo-Saif A. A., Abo-Youssef A. M.. **Resveratrol mitigates pancreatic TF activation and autophagy-mediated beta cell death via inhibition of CXCL16/ox-LDL pathway: a novel protective mechanism against type 1 diabetes mellitus in mice**. (2021) **901**. DOI: 10.1016/j.ejphar.2021.174059
38. Sandor A. M., Lindsay R. S., Dyjack N.. **CD11c(+) cells are gatekeepers for lymphocyte trafficking to infiltrated islets during type 1 diabetes**. (2019) **10**. DOI: 10.3389/fimmu.2019.00099
39. Abdel-Bakky M. S., Alqasoumi A., Altowayan W. M., Amin E., Darwish M. A.. **Resveratrol inhibited ADAM10 mediated CXCL16-cleavage and T-cells recruitment to pancreatic**. (2022) **14**. DOI: 10.3390/pharmaceutics14030594
40. Ding S., Xu S., Ma Y., Liu G., Jang H., Fang J.. **Modulatory mechanisms of the NLRP3 inflammasomes in diabetes**. (2019) **9**. DOI: 10.3390/biom9120850
41. Davanso M. R., Crisma A., Braga T.. **Macrophage inflammatory state in Type 1 diabetes: triggered by NLRP3/iNOS pathway and attenuated by docosahexaenoic acid**. (2021) **135** 19-34. DOI: 10.1042/cs20201348
42. S Rosa J., Heydari S., Oliver S. R.. **Inflammatory cytokine profiles during exercise in obese, diabetic, and healthy children**. (2011) **3** 115-121. DOI: 10.4274/jcrpe.v3i3.23
43. Zhang Z., Ding Y., Dai X., Wang J., Li Y.. **Epigallocatechin-3-gallate protects pro-inflammatory cytokine induced injuries in insulin-producing cells through the mitochondrial pathway**. (2011) **670** 311-316. DOI: 10.1016/j.ejphar.2011.08.033
44. Jiang X., Zhou Y., Wu K. K. L., Chen Z., Xu A., Cheng K. K. Y.. **APPL1 prevents pancreatic beta cell death and inflammation by dampening NF**. (2017) **60** 464-474. DOI: 10.1007/s00125-016-4185-z
45. Wang M., Xiong Y., Zhu W.. **Sleeve gastrectomy ameliorates diabetes-related spleen damage by improving oxidative stress status in diabetic obese rats**. (2021) **31** 1183-1195. DOI: 10.1007/s11695-020-05073-3
46. Wicker L. S., Miller B. J., Mullen Y.. **Transfer of autoimmune diabetes mellitus with splenocytes from nonobese diabetic (NOD) mice**. (1986) **35** 855-860. DOI: 10.2337/diabetes.35.8.855
47. Funes S. C., Rios M., Escobar-Vera J., Kalergis A. M.. **Implications of macrophage polarization in autoimmunity**. (2018) **154** 186-195. DOI: 10.1111/imm.12910
48. Levine B., Mizushima N., Virgin H. W.. **Autophagy in immunity and inflammation**. (2011) **469** 323-335. DOI: 10.1038/nature09782
49. Hayes H. L., Peterson B. S., Haldeman J. M., Newgard C. B., Hohmeier H. E., Stephens S. B.. **Delayed apoptosis allows islet**. (2017) **12**. DOI: 10.1371/journal.pone.0172567
50. Muralidharan C., Conteh A. M., Marasco M. R.. **Pancreatic beta cell autophagy is impaired in type 1 diabetes**. (2021) **64** 865-877. DOI: 10.1007/s00125-021-05387-6
51. Bharath L. P., Agrawal M., McCambridge G.. **Metformin enhances autophagy and normalizes mitochondrial function to alleviate aging-associated inflammation**. (2020) **32** 44-55.e6. DOI: 10.1016/j.cmet.2020.04.015
52. Santiago-Fernández C., Martin-Reyes F., Tome M.. **Oxidized LDL modify the human adipocyte phenotype to an insulin resistant, proinflamatory and proapoptotic profile**. (2020) **10**. DOI: 10.3390/biom10040534
|
---
title: Image Classification Based on Light Convolutional Neural Network Using Pulse
Couple Neural Network
authors:
- Maminiaina Alphonse Rafidison
- Hajasoa Malalatiana Ramafiarisona
- Paul Auguste Randriamitantsoa
- Sabine Harisoa Jacques Rafanantenana
- Faniriharisoa Maxime Rajaonarison Toky
- Lovasoa Patrick Rakotondrazaka
- Andry Harivony Rakotomihamina
journal: Computational Intelligence and Neuroscience
year: 2023
pmcid: PMC10030221
doi: 10.1155/2023/7371907
license: CC BY 4.0
---
# Image Classification Based on Light Convolutional Neural Network Using Pulse Couple Neural Network
## Abstract
Recently, most image classification studies solicit the intervention of convolutional neural networks because these DL-based classification methods generally outperform other methodologies with higher accuracy. However, this type of deep learning networks require many parameters and have a complex structure with multiple convolutional and pooling layers depending on the objective. These layers compute a large volume of data and it may impact the processing time and the performance. Therefore, this paper proposes a new method of image classification based on the light convolutional neural network. It consists of replacing the feature extraction layers of standard convolutional neural network with a single pulse coupled neural network by introducing the notion of foveation. This module provides the feature map of input image and the data compression using Discrete Wavelet Transform which is an optional step depending on the information quantity of this signature. The fully connected neural network, which has six hidden layers, classifies the image. With this technique, the computation time is reduced, and the network architecture is identical and simple independent of the type of dataset. The number of parameter is less than that in current research. The proposed method was validated with different dataset such as Caltech-101, Caltech-256, CIFAR-10, CIFAR-100, and ImageNet, and the accuracy reaches $92\%$, $90\%$, $99\%$, $94\%$, and $91\%$, respectively, which are better than the previous related works.
## 1. Introduction
For a software developer, it is a big challenge to search an image in database based on keyword, and the appropriate solution is to associate a label to all existing image. Finding a labelled image in database with table indexed facilitates the task. This operation of labeling is mainly called image classification which refers to a process in computer vision that can classify an image according To its visual content. Human visual is a perfect solution of image recognition; however, we cannot allocate a human resource to accomplish this task, and then automation is required.
The CNN or convolution neural network is categorized as a deep learning model, which is inspired by the organization of animal visual cortex used for processing data that has a grid pattern, such as images [1–3], and designed to automatically and adaptively learn spatial hierarchies of features from low- to high-level patterns. Convolution, pooling, and fully connected layers are the three types of layersthat constitute the CNN neural network. The feature extraction is ensured by convolution and pooling layers (first two layers), whereas the third, a fully connected layer, maps the extracted features into the final output for classification. The major recent works related to image classification use CNN to have a good result.
In 2014, GoogLeNet-19 developed by Google [4] was placed in first rank using 4 million parameters with a $6.67\%$ of top-5 error rate, and in the second place, VGGNet-16, created by Simonyan, Zisserman [4] with 138 million parameters, and the top-5 error rate is $7.3\%$. It is evident that managing these parameters is difficult with a high number of layers. So, in this paper, we will propose an efficient approach with minimum computation time, minimum parameters, and minimum number of layers to classify images based on the light convolutional neural network (LCNN). To accomplish this, we suggest swapping the convolution and pooling layers of CNN with a single layer of pulse coupled neural network (PCNN) plus foveation contribution (when we visualize an image, we do not stare for longtime but we focus only on the pertinent information. It is a human cortex visual behavior called “foveation”) and an optional feature representation by the discrete wavelet transform (DWT). The fully connected layer remains the same but with minimum of neurons and hidden layers. To validate our method, we applied it to three databases with different classes and compare the result with several recent state-of-the-art methods. The main contributions of this work are cited as follows:The proposed image classification system has a simple architecture, and the topology remains unchanged, which is independent of image input, and due to this simplicity, the quantity of data to process is reduced compared with CNN, and it allows us to have an optimal computation time. Such kind of solution may be supported by embedded systems. Related to the first contribution, the approach works with minimum number of parameters, that is, less than 20.Foveation intervenes to collect the pertinent information to facilitate the construction of the image signature. It is a simple process compared with the succession of convolution and spooling operations used by CNN.DWT reduces the size (row × column) of image map in the aim to have a minimum number of neurons for the deep learning network. The approach provides high accuracy greater than or equal to the technique based on CNN, and even the proposed architecture is very simple.
The rest of the paper is organized as follows: Section 2 summarizes the recent works related to our proposed approach. The Section 3 describes the mathematic model of PCNN. The proposed method is the purpose of Section 4 followed by experimental results in Section 5 and discussion in Section 6. Finally, Section 7 concludes the paper with motivation. To ensure a good understanding of this paper, Table 1 presents the list of abbreviations and definitions.
## 2. Literature Review
Ferraz and Gonzaga [5] introduced a study focused on object classification based on local texture descriptor and a support vector machine. Recently, two new texture descriptors are proposed for object detection based on the Local Mapped Pattern (LPM) approach. The Center-Symmetric Local Mapped Pattern (CS-LMP) and Mean-Local Mapped Pattern (MLMP) exhibit better performance than SIFT and CS-LBP, but prior results have proven that the size of descriptors could be decreased without loss of sensitivity. In their research, they investigated the decreasing size of the M‐LMP descriptor, and the performance measurement was done by using the support vector machine (SVM) classifier for object classification. In those experiments, they applied an object recognition system based on the M-LMP reduced descriptor and compared those effects with the CS-LMP, Local Intensity Order Pattern (LIOP), and SIFT descriptors. The object classification outcomes analyzed the use of a Bag of Features (BoF) model and an SVM classifier, with the end result that overall performance using the reduced descriptor is higher than the other three well-known techniques tested and additionally requires less processing time. The experience was done with Caltech-101 and ImageNet dataset and the performance was good except with background Google class because the extraction feature drops some sensitive information and leads to the wrong deduction. This research can be compared with study done by Srivastava et al. [ 6] because both have the same objective and use a common Caltech-101 dataset to validate their experience. The last is a new concept of image classification using bag of LBP features constructed by clustering with fixed centers and SURF. This study presents a known approach for the variety of datasets having specific types of images. Hindi Signature, Bangla Signature, ORL Face, and Caltech-101 are the four datasets that are employed to validate the proposed classification method. The algorithm is spat into three steps as follows: the identification of Region of Interest (ROI) is the first step using SURF (Speed Up Robust Transform) Points, then LBP (Local Binary Pattern) extracts the features present in ROI as the second step, and the last step consists the clustering of LBP features which are done with a new proposed approach as CFC (Clustering with Fixed Centers) to construct Bag of LBP Features. Through proposed CFC technique, each image is tagged/annotated with a fixed Bag of Features to avoid the training of machine again and again. For image classification task, SVM intervenes because it has been experimentally found to give the best performance when compared with Random Forest, Decision Tree, Linear, K Nearest Neighbor, and Linear Method. The accuracy obtained for Signature (Bangla and Hindi), ORL, Face, and Caltech-101 is $87.0\%$, $81.6\%$, $75.0\%$, and $79.0\%$, respectively. Thus, the average accuracy obtained through the proposed approach is $81.7\%$ in contrast to other state of art approaches having average accuracy as $64.15\%$, $76.47\%$, and $77.65\%$.
Han et al. [ 7] proposed a new CNN technique which could classify the images without difficulty compared to the other traditional models and gain better overall performance. With this method, the useful characteristic presentation of pretrained network can be efficaciously transferred to target task, and the original dataset can be augmented with the most treasured Internet images for classification. The method not only greatly reduces the requirement of a large training data but additionally effectively increases the training dataset. Both methods' capabilities make contributions to the considerable over-fitting reduction of deep CNNs on a small dataset. In addition, they successfully apply Bayesian optimization to remedy the tuff problem, hyper-parameter tuning, in network fine-tuning. The approach is applied to six public small datasets. Extensive experiments show that compared to conventional methods, the solution can help the famous deep-learning CNNs to achieve better performance. Specially, ResNet can outperform all the state-of-the-art models on six small datasets. The experiment results prove that the proposed solution can be a remarkable tool for dealing with practice problems that might be related to using deep CNNs on a small dataset; however, the accuracy decreases once the approach is applied to the large dataset or the dataset has many classes.
Çalik and Demirci [8] presented an image classification approach on embedded systems. The challenge was to apply CNN with device having a limited memory, and the result gives $85.9\%$ accuracy using CIFAR-10 dataset with memory allocation of 2 GB. The limitation of this method is same as Srivastava et al. [ 6] research which has a difficulty to train through a big dataset. Dhouibi [9] published a paper-entitled optimization of the CNN model for image classification. It is talking about topology optimization of CNN in terms of number of layers and the number of neurons per layer. This optimal solution allows to reduce the model and enable to deploy it in embedded platforms. This research was experimented with the same previous dataset, and they obtained $82.43\%$ accuracy. A third experience with the CIFAR-10 dataset is presented by Sharma and Phonsa [10]. They used the sequential method for the CNN and implemented the program in Jupiter notebook. They took 3 classes and classify them using CNN. The classes were airplane, bird, and car. They present the classification by using CNN, and they took batch size as 64. They got $94\%$ accuracy for the 3 classes.
Wang and Sun [11] present a new method of image classification using CNN with wavelet domain inputs. The idea is to replace the first several convolutional layers part of feature extraction of standard CNN with wavelet packet transform or dual-tree complex wavelet transform. These wavelets transform allows to have a higher resolution of the image in preprocessing step. The advantage is to keep the essential information present in image to ensure a correct classification because with CNN, some important information may loss during convolution calculation. During the experience, Caltech-256 dataset and DTD dataset with ResNet-50 are used, and there is a maximum improvement of $2.15\%$ and $10.26\%$, respectively, as accuracy.
Now, we are interested on the methods using ImageNet dataset qualified as largest image database on this area. Xception [12] or Extreme *Inception is* an improved version of the CNN inception model. Two levels are present on this conception as follows: the first level is composed by a single layer which slices the output into 3 segments and sent it to next filters. 1 ∗ 1, 3 ∗ 3 are, respectively, the convolution level of each filter. The depth-wise separable convolution [13–15] is the component which defines the Xception model. This technique intervenes in image classification with wide range of image having hundreds of classes ($79\%$ of accuracy for ImageNet dataset).VGG16 [12], which is inspired from AlexNet, has 16 layers and 3 fully connected layers. In the middle, there is 5 max pooling, and the *Softmax is* the output activation function [16–18] and ReLU for hidden layers. VGG19 [19] has a same concept as VGG16; however, this CNN contains 19 layers with 3 fully connected layers for classification and 16 convolution layers for feature extraction. The accuracy top-1 score for both is $71.3\%$.ResNet152V2 and MobileNetV2 [20] are well-known as CNNs for pretrained deep learning. They are specialized on feature extraction, prediction, and classification. A fully convolution layer through 32 filters and 19 residual bottleneck layers forms the architecture model of MobileNetV2. Concerning the ResNet152V2, it has thousands or hundreds of convolution layers, and the particularity compared with the previous version is that it employs a normalization batch before each weight layer. $78.0\%$ and $71.3\%$ are the recognition rate got with ImageNet dataset. NASNetLarge is a generation of CNN having a capacity to train more than a million pictures from ImageNet dataset and classify more than thousand objects. An input image of this network has 331 × 311 size and the strong point of this concept is that it has learned rich feature representations for a wide range of images. The experience is showing that the final accuracy rate reaches $82.5\%$. On the other hand, $84.3\%$ is the performance using EfficientNetB7 [21]. EfficientNetB7 is a release of EfficientNet which is a lightweight NAS-based network created by Google in 2019.
The common point of these studies is the ambition to optimize the standard CNN. Each research has its own methodology to extract image feature to reach the goal. Concerning the classification layer, some stay with one or more fully connected neural networks and the other tries to intervene SVM. They are selected as part of state of the art in this paper because the objective is similar even the experimental dataset then we have a possibility to compare the performance.
## 3. Pulse Coupled Neural Network
According to Srinivasan et al. [ 22] presentation, PCNN is inspired from behaviors of cat visual cortex phenomena. The modelling architecture is composed of three parts, namely, the dendritic tree, the linking modulation, and the pulse generator. The first part has two types of entries, namely, feeding and linking. The feeding receives the local and external stimulus; however, the linking captures the local only. The second part, which is the linking modulation, combines the outputs from two channels by adding a bias to the linking and multiplying it with feeding. Internal state of neuron *Uj is* the result of such combination, and this internal state and the threshold help the last part pulse generator to generate the pulse.
Lo et al. [ 23] introduce PCNN in image processing area and the mathematics modelling is defined below. The Table 2 explains the meaning of different parameters in PCNN.(i)First part (dendritic tree):[1]Fijn=exp −αFδnFijn−1+Sij+VF∑klMijklYkln−1,Lijn=exp −αLδnLijn−1+VL∑klWijklYkln−1.(ii)Second part (linking modulation):[2]Uijn=Fijn1+β. Lijn.(iii)Last part (pulse generator):The internal state of the neuron is compared to a dynamic threshold, Θ, to produce the output, Y, by[3]Yijn=1,if Uijn>Θijn,0,Otherwise.
The threshold is dynamic in that when the neuron fires (Y > Θ) the threshold then significantly increases its value [23]. This value then decays until the neuron fires again. This process is described by[4]Θijn=Θijn−1exp −αΘ+VΘYijn−1.
According to equation [3], the output is binary and then there is a lot of candidates for the foveation points because with standard PCNN, a threshold function having output 0 or 1 is used by the pulse generator module. This issue can be solved by adapting the sigmoid pulse generator as defined in equation [5] [24, 25] as given as follows:[5]Yijn=11+exp −ΥUijn−Θijn−1.
Figure 1 represents the described model and the output varies from 0 to 1 [26].
## 4. Proposed Method
Now, we have more visibility about PCNN which is an element involved in the image classification method. The wavelet transforms and fully connected neural network (FCNN) will be explained briefly during these interventions in the approach. The proposed system has two modules, namely, feature extraction and deep learning module, and a clear presentation of the approach is shown in Figure 2.
## 4.1. Feature Extraction
First step is to choose the image dataset and split it in two parts, namely, training and validation. All existing image in database must be converted to grayscale and resized (optional) because PCNN can process only a matrix with one dimension instead of three like an RGB image. Image resizing is applicable only when the image has a large dimension. A part of color conversion, preprocessing module, has two filters, namely, Canny and blurring filter. The reason of this choice is to reduce the quantity of information to be processed. Canny filter is an edge detection operator that uses a multistage algorithm to detect a wide range of edges in images. It was developed by John F. Canny [27] in 1986. Blurring filter [27, 28] is a low pass filter, because it allows low frequency to enter and stop high frequency. Here, frequency means the change of pixel value. Around edge pixel, value changes rapidly as blur image is smooth; so high frequency should be filtered out. The Figure 3 represents such details.
PCNN extracts the essential part from blurring image and eliminates the noise background. High number of iterations is required to ensure that PCNN accomplishes his task. Before starting the iteration, we should initiate the neural network parameters as follows:(i)Weights matrix[6]M=$W = 0.70710.7071110.70710.707.$(ii)Initial values of matrixThe preliminary values of linking L, feeding F matrix, and stimulus S are similar to the enter image. The convolution among null matrix which has the same length as the enter image R × C and weights matrix initiates the output value Y of PCNN. The initial value of dynamic threshold Θ is an R-by-C matrix of two.(iii)Constants delay[7]αF=0.1,αθ=1,αL=1.2.(iv)Constants normalization[8]VF=0.5,VΘ=20,VL=0.2,Υ=0.9,β=0.1.
The maximum number of iterations is fixed to 40 and the calculation of the percentage of misclassified pixel [29] indicates the image to be selected. The first minimum rate corresponds to excellent image segmentation and the second to edge detection, so we are interested in the second result shown in Figure 4. Its gray level varies between 0 and 1 due to the sigmoidal pulse generator used by the PCNN neural network.
PCNN task is completed by extracting the relevant information. Currently, we solicit the foveation method to collect the data sensitive to human eyes. For this, we apply an image threshold and we have the result shown in Figure 5(a).
Now, we should reduce the dimension of the image (this step is optional if the image has a small size like 32 × 32), and it can be done by Haar Wavelet Transform (HWT). HWT operates simultaneously in spatial and frequency domain information in image processing. It is a transform for which the wavelets are sampled at discrete intervals [30, 31]. Haar wavelet operates on data by calculating the sums and differences of adjacent elements. To apply HWT on images, a simple explanation is shown in Figure 6. Four subbands, namely, LL, HL, LH, and HH subbands (L = Low, H = High) compose the resulting image where LL-subband contains an approximation of the authentic image while the other subbands comprise the missing details. The LL-subband output from any stage can be decomposed similarly [32].
We apply HWT transform three times to the foveation image, and we are interested on the second LL-subband (in Figures 5(b)–5(d)). The resulting image will be reshaped to vector to constitute the value of input layer of FCNN.
## 4.2. Classification
FCNN has three parts, namely, input, hidden, and output layers. As the name is called fully connected, it means that each neuron connects to all neurons existing in the next layer. Before going to the activation function, the computation of input, weight, and bias must be done beforehand. We focus only on two activation functions, namely, the nonlinear ReLU function and softmax function. They are defined in equations [9] and [10].[9]fReLUxi=0,if xi<0,xi,if xi≥0,[10]fsoftmaxxi=exi∑$j = 1$Nexj,where xi is the sum inputs improved by means of weights plus bias and N the number of neurons in the output layer. The value of the ReLU function is 0 or xi, and for softmax function, it is between 0 and 1 because it is indicating the probability that in which class the image belongs. The feature map of the input image constitutes the input layer (size of image signature × 1), and the image class membership forms the output layer [33].
Six hidden layers are required at least and, in this paper, we fix it to 6. The activation characteristic for them is the ReLU function, and all weights are initialized randomly. It means that there are six weights, namely, w1(h1, 1024), w2(h2, h1), w3(h3, h2), w4(h4, h3), w5(h5, h4), w6(class number, h5) where (hj × hk) is the size of weight wi. For experience purpose, the value of h is a square root of size of image signature and number of classes. Concerning output layer (number of classes x 1), the number of neurons is the same as the number of classes present in dataset. The neuron which has a high probability value determinates the belonging class. The activation function softmax ensures this probability format. Evidently, the number of neurons in input layer is equivalent to the length of image signature vector. The percentage of image allocated for testing depends on the searcher choice but it is important to have a percentage training dataset more than testing images. During training phase, the output neuron corresponding to input image signature is 1 and 0 for leftovers.
## 5. Experiments
To evaluate the performance of the proposed method, we introduce three datasets that are used by different research cited in literature review Section 2 in the aim to compare these performances with ours. They are publicly available. The Section 4.1 describes the content of each dataset and Section 4.2 details the performance using image classification measurement like accuracy [34], loss [35], precision, recall, and F1 score [36, 37].
## 5.1. Dataset Description
Caltech-101 (The dataset is available at https://www.kaggle.com/datasets/862ae86edba271c39f76d0b530edeb55076b4b82b971160637210900747c44b1) is the first image dataset that we use to test our conception. It includes photos of gadgets belonging to 101 classes plus one background clutter class. Every photo is labelled with single item and every class carries kind of forty to 800 pics, totaling to 9146 photos. We are not able to show here all content of this dataset; however, a sample of images is presented in Figure 7 [24].
The second dataset is Caltech-256 (The dataset is available at https://www.kaggle.com/datasets/jessicali9530/caltech256) dataset [38] having 30607 natural photographs, consisting of 256 object categories and 1 random background class. The common variety of photos in every class is 119 (variety from eighty to 827) and the average photo dimension is 371 × 326. A sample snapshot is presented in Figure 8.
The third dataset that we use for testing is CIFAR-10 (The dataset is available at https://www.cs.toronto.edu/~kriz/cifar.html) (Canadian Institute for Advanced Research, 10 classes). This dataset contains 60000 32 × 32 color images divided into 10 classes (airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck), each with 6000 images [39]. Figure 9 shows a few sample images from the CIFAR-10 dataset.
The fourth dataset is CIFAR-100 (The dataset is available at https://www.cs.toronto.edu/~kriz/cifar.html) which is similar to the CIFAR-10, except it has 600 images for each class (100 classes in total). In CIFAR-100, there are 20 super classes subgrouped into 100 classes. The dataset comes with two labels for each image such as a “fine” label (class) and a “coarse” label (superclass). A sample of images present in this dataset is shown in Figure 10.
The last dataset for experiments is ImageNet (The dataset is available at https://www.image-net.org/download.php). It is a wide database having more than one million images and spans 1000 object classes. ImageNet dataset is publicly available and a snap shot is shown in Figure 11.
## 5.2. Performance Measurement
We fix the number of epochs to 2500, it does not depend on dataset but it can be increased to improve the accuracy. The first experience was done with Caltech-101 dataset that $75\%$ of image will be processed for training purpose and $25\%$ (2279 images) of remaining dataset will pass through our network for validation. It means that we test $25\%$ for each class. The dataset split must be the same as used by previous studies; otherwise, we cannot compare the result. The accuracy average is around $91\%$, and the sample of performance is the object of Table 3. The precision is excellent when the number of images belonging to a class is not high. We remark also that the accuracy commences acceptable when reaching 1500th epoch according to the Figure 12. Concerning the loss, it converges to null once the epoch is near to 1700.
The Caltech-256 is considered an improvement to its predecessor, the Caltech 101 dataset, with new features such as larger category sizes, new and larger clutter categories, and overall increased difficulty. The accuracy is reduced $2\%$ compared with Caltech-101 (Figure 12) because the number of class is increased; however, the performance is better if the number of images in one class is large. We can observe it for motorbikes experience (Table 4). The loss value is considerable until the end of experience (Figure 13). To fix this issue, it is possible to augment the number of epochs but it will have an impact on the other parameter. For precision, the loss function used is the cross entropy as defined in [11][11]LCE=−∑$i = 1$Ntilogpi,where ti is the truth label, pi the softmax probability for the ith class and N, the number of image class present in dataset [40].
The experience with CIFAR-10 is rapid because the number of classes is less which is why the accuracy rate is high from 1000th epoch. Resizing image and HWT is not required because the image has a small dimension (32 × 32). We select 50000 images ($90\%$) for training and 10000 images ($10\%$) for testing. This partition is the common partition used by previous researchers' works. Same as proceed with Caltech-256, the full result is presented in Table 5 which provides the accuracy details for each class. Regarding the epoch, it is shown in Figure 14. The proposed method by Sharma and Phonsa [10] was tested with 3 classes, namely, aeroplane, bird, and cat. If we limit only our test with these classes, we got an accuracy of $99\%$.
Now, we test the technique with largest image dataset like CIFAR-100 and ImageNet. The performance is reduced because the dataset has many classes and the number of images for testing is less too (Figures 14 and 15). It can be improved by increasing the number of epochs; however, it may have an impact in computational time. To support such suggestion with the embedded system, a device having a good configuration is necessary. As we see in Figure 13, the loss function starts with highest value and it becomes negligible at the end of the epoch. The cross-entropy trend for both datasets is different comparing with three previous ones. The experience metrics are presented in Tables 6 and 7, and we notice that our accuracy is still competitive.
Most of image classification research studies based on CNN use ImageNet as dataset, and we will compare these performances with ours using the same device configuration which is described as follows:CPU: AMD EPYC Processor (with IBPB) (92 core)RAM: 1.7TGPU: Tesla A100Batch size: 32 *As a* part of top-1 accuracy, we compare also the top-5 accuracy, number of parameters, and computation time per each method in Table 8. We remark that our proposed method has a good performance. With another device having a limited memory like embedded systems 2 GB, the computation time augments but is still tolerable. The research done by Çalik and Demirci [8] is dedicated for small dataset (CIFAR-10); however, we have high rate of recognition $85.9\%$ vs $99.11\%$.
Before closing this paragraph, we confront our result with some research studies using a smallest dataset such as Caltech-101, Caltech-256, CIFAR-10, and CIFAR-100 (Table 9). We see that the proposed approach leads the performance except for Caltech-256 experience in which we are on the second position. The symbol «−» in tables means that that the authors did not provide the information in these paper publications and «∗», the maximum value.
## 6. Discussion
Most of recent research in image classification choose CNN as a neural network to accomplish the task. It collects the relevant information in feature layer which is the estate of convolution and pooling. Both operations reduce the volume of information to be processed and the final important information jugged essential called image map or signature is going through fully connected neural network for image classification purpose. This technique required from thousand to million parameters, and the architecture changes according to the dataset to be treated. It means that the solution is complex and may have an impact on the performance. For this reason, we propose this approach with 11000 parameters maximum and simple/static architecture, and the accuracy is improved. Such result was due to the foveation produced by PCNN. The methods that are CNN-based have a facility to classify an image containing a background because they give an importance on such information; however, ours has a weakness which is why the accuracy for the background class in Caltech-101 dataset is less ($85\%$) because the PCNN ignores this information. Here, we are talking about top-1 accuracy but the top-5 accuracy is at $90\%$.
Regarding the test with CIFAR-10 image dataset, the approach proposed by Sharma and Phonsa [10] has an accuracy less than ours, and even the number of classes is less because the dataset has only ten classes and the image inside does not have a large dimension. Different type of image like dog and cat may not have similar signature due to foveation which is why the performance is always high. So, we advise people to choose our approach if the number of image's class is less, and with no much background, even image dimension is considerable. Otherwise, a large number of epochs is recommended for a large dataset like Caltech-256, CIFAR-100, and ImageNet. MCNN and CNN for image classification are complementary to ensure an excellent result. A module of preprocessing should be added in chain of processing to decide in which case the system uses convolution/pooling or PCNN/foveation as the feature extraction layer.
Before concluding this paper, we resume in Table 10 the advantage and disadvantage for each algorithm.
## 7. Research Motivation and Conclusion
Applying CNN for image classification demands high number of parameters and the feature extraction layers require a big computing resource for getting an image map, and this step may cause a delay in processing. So, the first motivation of this research is to propose a simple architecture and a simple static model independent of input image or dataset with minimum computation time. The second motivation is to have a neural network more efficient with an accuracy more than existing image recognition algorithms. To attend on these objectives, we resize if the image has a large dimension and converts to gray level before PCNN and foveation processing. The resulting image goes through wavelet transformation in the three level by keeping the final approximation matrix for the FCNN input layer. This transformation reduced the information with minimum loss. For validation, we choose five datasets, namely, Caltech-101, Caltech-256, CIFAR-10, CIFAR-100, and ImageNet and comparing the existing methods with same dataset, the proposed method has a good performance especially with small dataset like CIFAR-10.
PCNN always keeps an unmissable step in image processing area, and foveation is an application of this intelligent neural network. Aside searching picture in database, we are able to apply this approach in face recognition and finger print recognition, for example. The axe improvement of this study may be oriented to replace the PCNN model with modified pulse coupled neural networks (MPCNN) or intersecting cortical version (ICM). Three works [42–44] are published recently, and they can be a source of inspiration to improve this research. In future work, we focus only in two class of images, namely, person with and without facemask, and in case of image with facemask, we will proceed to check whether the mask is worn correctly.
## Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
## Conflicts of Interest
The authors declare that they have no conflicts of interest.
## References
1. Yamashita R., Nishio M., Do R. K. G., Togashi K.. **Convolutional neural networks: an overview and application in radiology**. (2018.0) **9** 611-629. DOI: 10.1007/s13244-018-0639-9
2. Hubel D. H., Wiesel T. N.. **Receptive fields and functional architecture of monkey striate cortex**. (1968.0) **195** 215-243. DOI: 10.1113/jphysiol.1968.sp008455
3. Fukushima K.. **Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position**. (1980.0) **36** 193-202. DOI: 10.1007/BF00344251
4. **The Development of Visual Priors Across the Lifespan**. (2020.0)
5. Ferraz C. T., Gonzaga A.. **Object classification using a local texture descriptor and a support vector machine**. (2017.0) **76** 20609-20641. DOI: 10.1007/s11042-016-4003-0
6. Srivastava D., Bakthula R., Agarwal S.. **Image classification using SURF and bag of LBP features constructed by clustering with fixed centers**. (2019.0) **78** 14129-14153. DOI: 10.1007/s11042-018-6793-8
7. Han D., Liu Q., Fan W.. **A new image classification method using CNN transfer learning and web data augmentation**. (2018.0) **95** 43-56. DOI: 10.1016/j.eswa.2017.11.028
8. Çalik R. C., Demirci M. F.. **Cifar-10 image classification with convolutional neural networks for embedded systems**. 1-2. DOI: 10.1109/AICCSA.2018.8612873
9. Dhouibi M.. **Optimization of CNN model for image classification**. 1-6. DOI: 10.1109/DTS52014.2021.9497988
10. Sharma A., Phonsa G.. **Image classification using CNN**. DOI: 10.2139/ssrn.3833453
11. Wang L., Sun Y.. **Image classification using convolutional neural network with wavelet domain inputs**. (2022.0) **16** 2037-2048. DOI: 10.1049/ipr2.12466
12. Scanlon M.. **Semantic annotation of aerial images using deep learning, transfer learning, and synthetic training data**. (2018.0)
13. Aldoski J.. (2022.0)
14. Asgher U., Khalil K., Khan M. J.. **Enhanced accuracy for multiclass mental workload detection using long short-term memory for brain–computer interface**. (2020.0) **14** p. 584. DOI: 10.3389/fnins.2020.00584
15. Goutte C., Gaussier E.. **A probabilistic interpretation of precision, recall and F-score, with implication for evaluation**. (2005.0) **3408** 345-359. DOI: 10.1007/978-3-540-31865-1_25
16. Mahum R., Irtaza A., Nawaz M., Nazir T., Masood M., Mehmood A.. (2021.0)
17. Griffin G., Holub A., Perona P.. **Caltech-256 Object Category Dataset**. (2007.0)
18. Krizhevsky A.. **Learning multiple layers of features from tiny images**. (2009.0)
19. Toledo Y., Almeida T., Bernardini F., Andrade E.. (2019.0)
20. Cai Q., Li F., Chen Y., Li H., Cao J., Li S.. **Label rectification learning through kernel Extreme learning machine**. (2021.0) **59** 1-6. DOI: 10.1155/2021/6669081
21. Kurniawan I., Tarwidi D.. **QSAR modeling of PTP1B inhibitor by using Genetic algorithm-Neural network methods**. (2019.0) **1192**. DOI: 10.1088/1742-6596/1192/1/012059
22. Srinivasan K., Garg L., Datta D.. **Performance Comparison of deep CNN models for detecting driver’s distraction**. (2021.0) **68** 4109-4124. DOI: 10.32604/cmc.2021.016736
23. Lo W., Yang X., Wang Y.. **An Xception convolutional neural network for malware classification with transfer learning**. 1-5
24. Liu H., Liu M., Li D., Zheng W., Yin L., Wang R.. **Recent advances in pulse-coupled neural networks with applications in image processing**. (2022.0) **11** p. 3264. DOI: 10.3390/electronics11203264
25. Yang Z., Lian J., Guo Y.. **An overview of PCNN model’s development and its application in image processing**. (2019.0) **26** 491-505. DOI: 10.1007/s11831-018-9253-8
26. Duan X., Gou M., Liu N., Wang W., Qin C.. **High-capacity image steganography based on improved Xception**. (2020.0) **20** p. 7253. DOI: 10.3390/s20247253
27. Kassani S. H., Kassani P., Khazaeinezhad H. R., Wesolowski M. J., Schneider K., Liu L.. **Diabetic retinopathy classification using a modified Xception architecture**. 1-6. DOI: 10.1109/ISSPIT47144.2019.9001846
28. Datta D., Mittal D., Mathew N. P., Sairabanu J.. **Comparison of performance of parallel computation of CPU cores on CNN model**. 24-25. DOI: 10.1109/ic-ETITE47903.2020.142
29. Qassim H., Verma A., Feinzimer D.. **Compressed residual-VGG16 CNN model for big data places image recognition**. 8-10. DOI: 10.1109/CCWC.2018.8301729
30. Bansal M., Kumar M., Sachdeva M., Mittal A.. **Transfer learning for image classification using VGG19: Caltech-101 image data set**. (2021.0) **18** 1-12. DOI: 10.1007/s12652-021-03488-z
31. Elshennawy N. M., Ibrahim D. M.. **Deep-pneumonia framework using deep learning models based on chest X-ray images**. (2020.0) **10** p. 649. DOI: 10.3390/diagnostics10090649
32. Bhakti B., Shubham Gajre S., Talbar S.. **Eff-UNet: a novel architecture for semantic segmentation in unstructured environment**. 1473-1481. DOI: 10.1109/CVPRW50498.2020.00187
33. Eckhorn R., Reitboeck H. J., Arndt M., Dicke P.. **Feature linking via synchronization among distributed assemblies: simulations of results from cat visual cortex**. (1990.0) **2** 293-307. DOI: 10.1162/neco.1990.2.3.293
34. Lindblad T., Kinser J. M.. (2005.0)
35. Tanaka M., Watanabe T., Baba Y., Kurita T., Mishima T.. **Autonomous foveating system and integration of the foveated images**. 559-564. DOI: 10.1109/ICSMC.1999.814153
36. Canny J.. **A computational approach to edge detection**. (1986.0) **8** 679-698. DOI: 10.1109/TPAMI.1986.4767851
37. Gedraite E. S., Hadad M.. **Investigation on the effect of a Gaussian Blur in image filtering and segmentation**. (2011.0) **1** 393-396
38. Hayashi T., Tsubouchi T.. **Estimation and sharpening of blur in degraded images captured by a camera on a moving object**. (2022.0) **22** p. 1635. DOI: 10.3390/s22041635
39. Budde L. E., Bulatov D., Iwaszczuk D.. **Identification of misclassified pixels in semantic segmentation with uncertainty evaluation**. (2021.0) **XLIII-B2-2021** 441-448. DOI: 10.5194/isprs-archives-XLIII-B2-2021-441-2021
40. Ibrahem W. N.. **Wavelets and multiresolution processing,**. (2016.0) **8** 1-12
41. Houssein E. H., Ali M. A. S., Hassanien A. E.. **An image steganography algorithm using haar discrete wavelet transform with advanced encryption system**. 641-644
42. Zhang W., Yao G., Yang B., Zheng W., Liu C.. **Motion prediction of beating heart using spatio-temporal LSTM**. (2022.0) **29** 787-791. DOI: 10.1109/LSP.2022.3154317
43. Guo F., Yang B., Zheng W., Liu S.. **Power frequency estimation using sine filtering of optimal initial phase**. (2021.0) **186**. DOI: 10.1016/j.measurement.2021.110165
44. Xu C., Yang B., Guo F., Zheng W., Poignet P.. **Sparse-view CBCT reconstruction via weighted Schatten p-norm minimization**. (2020.0) **28** 35469-35482. DOI: 10.1364/OE.404471
45. Rezende E., Ruppert G., Carvalho T., Theophilo A., Ramos F., Latifi S.. **Malicious software classification using VGG16 deep neural network’s bottleneck features**. (2018.0)
46. Rinky B. P., Mondal P., Manikantan K., Ramachandran S.. **DWT based feature extraction using edge tracked scale normalization for enhanced face recognition**. (2012.0) **6** 344-353. DOI: 10.1016/j.protcy.2012.10.041
|
---
title: 'Development of the Systems Thinking for Health Actions framework: a literature
review and a case study'
authors:
- Jenna Thelen
- Carmen Sant Fruchtman
- Muhammad Bilal
- Kebabonye Gabaake
- Shahid Iqbal
- Tshiamo Keakabetse
- Aku Kwamie
- Ellen Mokalake
- Lucia Mungapeyi Mupara
- Onalenna Seitio-Kgokgwe
- Shamsa Zafar
- Daniel Cobos Muñoz
journal: BMJ Global Health
year: 2023
pmcid: PMC10030275
doi: 10.1136/bmjgh-2022-010191
license: CC BY 3.0
---
# Development of the Systems Thinking for Health Actions framework: a literature review and a case study
## Abstract
### Background
Systems thinking is an approach that views systems with a holistic lens, focusing on how components of systems are interconnected. Specifically, the application of systems thinking has proven to be beneficial when applied to health systems. Although there is plenty of theory surrounding systems thinking, there is a gap between the theoretical use of systems thinking and its actual application to tackle health challenges. This study aimed to create a framework to expose systems thinking characteristics in the design and implementation of actions to improve health.
### Methods
A systematised literature review was conducted and a Taxonomy of Systems Thinking Objectives was adapted to develop the new ‘Systems Thinking for Health Actions’ (STHA) framework. The applicability of the framework was tested using the COVID-19 response in Pakistan as a case study.
### Results
The framework identifies six key characteristics of systems thinking: [1] recognising and understanding interconnections and system structure, [2] identifying and understanding feedback, [3] identifying leverage points, [4] understanding dynamic behaviour, [5] using mental models to suggest possible solutions to a problem and [6] creating simulation models to test policies. The STHA framework proved beneficial in identifying systems thinking characteristics in the COVID-19 national health response in Pakistan.
### Conclusion
The proposed framework can provide support for those aiming to applying systems thinking while developing and implementing health actions. We also envision this framework as a retrospective tool that can help assess if systems thinking was applied in health actions.
## Introduction
Complex adaptive systems (CAS) are systems that contain a myriad of intricately interconnected components. They are dynamic, open systems that change and evolve due to multiple interactions within and across the system, including positive and negative feedback, time delays and tipping points. CAS are self-organising and holistic.1 Health systems can be identified as CAS as they have many interconnected components (ie, agents, such as providers, patients, community, policy makers, and insurance agencies, and structures such as policies, norms, values, histories and capacities) that are constantly changing and adapting to changes.2 In the past 15 years, there have been increasing recommendations to use systems thinking (ST) in health systems because of their complex nature.3 4 ST is a discipline that can support us in making sense of CAS, it focuses on how components of a system are interconnected and how the system behaves.5 6 ST comprised theories, methods and tools that assist with addressing complex problems. It began in the 20th century and has been applied in countless disciplines, including biology, psychology, computer science and anthropology.6 It first emerged as a method for scientific investigation, but in the 1940s, it gained traction as a way to solve real-world problems related to World War II.7 Despite its long history, there is still no single agreed-upon definition of what ST is.8 Forrester and Richmond were among the first to define ST7. In Richmond’s article, we find the first complete definition of ST as ‘the art and science of making reliable inferences about behavior by developing an increasingly deep understanding of underlying structure’.9 Many definitions have followed and they all contain two common attributes: seeing the system holistically beyond just its components, and seeing the components in the context of the whole system.7 In other words, ST focuses on the holistic perspective of a system, and the observed behaviours that emerge from the interactions between the parts of the system.7
Using an ST approach has claimed to be beneficial for understanding and intervening in a health system.3 Trochim et al10 have suggested that ST can be used in health systems to create a more holistic view of financing, broaden non-traditional collaborations among disciplines, address the impact of social and political factors and identify barriers to implementing systems approaches.10 ST enables a change in mindset which allows individuals to solve complex problems through a holistic lens.11 Although there is a wealth of theoretical applications of ST in health systems, there is a gap between the conceptual use of ST and its actual application in the real world.12–14 Kwamie et al15 suggest that the application of ST needs to be documented better to build a stronger evidence base.15 One practical application of ST in health has been the Systems Thinking for District Health Systems (ST-DHS) initiative, which supported countries and health districts, to apply ST tools and practices to understand and intervene in their local health systems.16 As part of this initiative and to address the gaps in the application of ST in health systems, we have developed the ‘Systems Thinking for Health Actions (STHA)’ framework. This framework provides a structured approach to assessing the extent of application of ST in health actions, where ST terminology may not have been explicitly used. The framework aims to explore the application of ST principles and attributes in the formulation and evaluation of health actions. Furthermore, this new framework intends to be used in health actions as an operational checklist, with ST tools and methods that can be directly applied to the actions.
## Methods
The STHA framework was developed using a combination of a systematised literature review and expert inputs. The developed framework was then applied to a case study within the ST-DHS initiative to explore the ST characteristics in the COVID-19 response in Pakistan.
## Study setting
This research was conducted as part of a bigger project, the ST-DHS initiative, which was implemented in three countries: Botswana, Pakistan and Timor-Leste. The initiative provided two districts per country with ST tools and methods, aiming to improve local health systems with the new knowledge on how to apply ST.16
## Systematised literature review
A systematised literature review was conducted to create the STHA framework. This type of review contains elements of a systematic review but is missing elements that a systematic review would have, such as having two reviewers and registration of the review.17 A systematised review was chosen as time and resources were limited.
We developed a search strategy that included keywords to identify practical and theoretical uses of ST, as seen in online supplemental appendix 1. Using PubMed and Google Scholar, the first 50 results in each search term, based on the best match filter, were assessed for inclusion in the systematised review. The titles and abstracts were independently screened for relevance. If the abstract was pertinent, the full text was read and determined if it was to be included based on the content. Data were extracted and the ST tools and methods used in the manuscripts were compiled into a table (online supplemental appendix 2). They were then categorised into an Xmind map based on their intended uses as stated in the manuscripts.
The inclusion criteria were studies that explicitly mentioned and described ST concepts, tools or methods in theory or practice, were available in English and were accessible through the University of Basel. In addition, papers were included if they were published between 1 January 2009 and 31 December 2021. The year 2009 was chosen, as that is the year that ST in health systems gained traction with the publication of Systems Thinking for Health Systems Strengthening.3
## Framework development
The aim of the framework was to close the gap between ST theory and application.12–14 Stave and Hopper’s Taxonomy of Systems Thinking Objectives was used as the starting point of the framework. JT, DCM and CSF adapted the taxonomy based on the results of the systematised literature review and developed the first draft of the STHA framework.8 Once the ST characteristics were developed, we included definitions for each of them, as well as categorised the mapped ST tools by characteristic.
The first draft of the STHA framework was presented and discussed at two different virtual participatory workshops consisting of 14 health system researchers with extensive ST experience from Botswana, Pakistan, Switzerland and Timor-Leste. The first workshop consisted of individuals who had participated in the ST-DHS initiative, including the funder. The framework was presented and feedback collected regarding the categories, definitions and categorisation of the tools. The framework was then adapted and a second draft was presented in a workshop with health system researchers based at Swiss Tropical and Public Health Institute (Swiss TPH). The same approach was taken and the feedback was incorporated into a third draft of the framework.
JT, DCM and CSF made the final decision of what to include in the framework. The final draft was shared via email with all participants and those with interest were invited to participate in the write-up of this manuscript (see list of coauthors).
## Pakistan COVID-19 case study
Once we had the final STHA framework, we conducted a case study in Pakistan to validate and test if the framework adequately identified ST in Pakistan’s COVID-19 response. The case study was conducted with key informant interviews and a document review of the National Action Plan for Coronavirus Disease (COVID-19) Pakistan (National Action Plan). The key informant interviews were part of the rapid realist evaluation that was conducted for the ST-DHS initiative evaluation.18 19 The National Action *Plan is* a document that was developed by the Ministry of National Health Services, Regulation, and Coordination and Government of Pakistan to guarantee that the COVID-19 procedures for outbreak preparedness, containment and mitigation were followed.20 The Pakistan case study was chosen based on convenience as we were able to use the same data that were collected for the ST-DHS initiative.
## Key informant interviews
The interviewees were two male and one female district health managers from the Islamabad district in Pakistan, representing $38\%$ of total district health managers from the district.
The participants were purposively selected from the ST-DHS initiative and all participants provided consent to the interview. Saturation was discussed and due to the nature of the research, it was determined that three interviews were adequate to test the STHA framework.21 JT interviewed the health managers with whom she had no prior relationship. In addition to JT and the interviewee, a male researcher; MB, from Child Advocacy International, the local research partner, who had been involved in the implementation of the ST-DHS initiative, participated in all three interviews. The interviews lasted between 30 and 45 min and were conducted in English. The interviews were completed over Zoom, recorded with live transcription and stored on the Swiss TPH drive. The interviews were conducted using a semistructured interview guide that was created to evaluate the ST-DHS initiative. The guide was tested before implementation by conducting two practice interviews. Field notes were taken during and after the interviews. The participants did not review the interview quotation table before submitting the manuscript. No interviews were repeated.
## Patient and public involvement
This research did not contain any patient or public involvement.
## Identification of studies
The initial search of the literature yielded 454 articles, which included four articles from expert input. After removing duplicates and screening the abstracts, 71 articles remained. Following a full-text review, 38 articles met the inclusion criteria and were included in developing the framework (figure 1). The 38 articles comprised systematic reviews ($$n = 2$$), literature reviews ($$n = 3$$), qualitative and/or quantitative studies ($$n = 31$$) and commentaries ($$n = 2$$). The research was conducted in a multitude of countries, including the USA ($$n = 7$$), Australia ($$n = 7$$), Ghana ($$n = 1$$), Uganda ($$n = 2$$), Zambia ($$n = 1$$), Canada ($$n = 1$$), India ($$n = 1$$), Pakistan ($$n = 1$$), Singapore ($$n = 1$$) and Thailand ($$n = 1$$). Additionally, there were nine studies with multiple countries and six did not have a specific country of research. The disciplines included in the review were health ($$n = 36$$), food production ($$n = 1$$), producing research ($$n = 2$$), policymaking ($$n = 1$$) and road traffic safety ($$n = 1$$). The most commonly used ST tools were causal loop diagrams ($$n = 27$$), systems dynamics modelling ($$n = 12$$), agent-based modelling ($$n = 8$$) and concept mapping ($$n = 6$$) (table 1).
**Figure 1:** *Systematised literature review results.* TABLE_PLACEHOLDER:Table 1
## STHA framework
An iceberg model, shown in figure 2, was chosen to represent this framework, as it allows a perspective shift from the visible health system performance and actions to how the application of ST can unveil underlying structures, patterns and behaviours of the system.22 The ST characteristics are not intended to be considered in order, rather the hierarchy of them only represents the varying levels of technical complexity required to apply them.22 The framework proposes a number of considerations that policy makers, managers, researchers or health practitioners can use to apply ST principles to the design, implementation and evaluation of health actions.
**Figure 2:** *An iceberg model representing the varying complexity of the six systems thinking characteristics of the STHA framework.*
One of the first steps in moving from traditional linear thinking to ST is recognising and understanding interconnections and system structure. This characteristic comprised the recognising and understanding that health systems are composed of different interconnected parts. By recognising and understanding interconnections, we are able to gain insight of the key actors within a health system (including the agency of self within the system), as well as the overall system structure. It enables the stakeholders within the system to create shared goals and acknowledge how the health system can work together to implement successful health actions.23 The second ST characteristic is identifying and understanding feedback in the system since CAS (as health systems) are governed by feedback.3 *Feedback is* the cause and effect relationships that occur among the different elements in a system.8 *It is* critical to identify these relationships, both positive and negative, that occur among the parts of the health systems by recognising feedback loops and determining chains of causality within the system.8 This characteristic builds on recognising interconnections, as it recognises the connections and the directionality between them, and how indirectly an intervention can have a balancing effect on the desired outcome. In Uganda, a causal loop diagram demonstrated how feedback from government restrictions and policies influenced how the dual practice policy developed over time.24 Having an adequate feedback system in health actions allows the changing needs in the system to be identified and allows for interaction and seamless communication among all stakeholders, as well as preventing unintended outcomes.24 25 The third characteristic is identifying leverage points, which is a vital characteristic of ST, as these are areas where small changes can have a large impact.26 Identifying leverage points can help determine where to allocate scarce resources in most efficient ways or what small changes in health systems will yield substantial improvements in performance.27 In health systems, identifying leverage points systematically illuminates key areas to intervene, allowing for more targeted health actions.28 Glenn et al29 used qualitative data to develop a model of the neglected tropical disease system to identify potential leverage points for eliminating neglected tropical diseases.29 Approaching the more technically complex categories of the STHA framework is the characteristic understanding dynamic behaviour. As health systems are CAS, they are non-linear and dynamic over time.29 *It is* imperative to recognise the feedback loops8 and the interactions between the components of the health system that are responsible for generating patterns of behaviour that can change over time.30 Therefore, recognising dynamic behaviour can help determine the effect that behaviours from components of a health system have on the entire system.8
Using models to suggest possible solutions to a problem refers to the use of visualisations to display causality, feedback loops and variables to achieve the purpose of a health system. The visualised models do not depict a real-world system, rather how actors view the system. Developing systems models also include multistakeholder dialogue, which involves the process of defining the problem and delineating the boundaries of the system as well. By depicting the system, stakeholders can gain common understanding of the health system or health actions, as well as each other’s mental models, which facilitates finding potential solutions to the problem.29 Finally, creating simulation models for testing policies would arguably be the most complex analysis to integrate ST is creating simulation models for testing policies. Using ST simulation models helps translate multifaceted scientific findings into easy-to-understand outcomes.31 Simulation models should combine all previous characteristics and use qualitative and quantitative data to create a comprehensive model of the overall system.8 In health systems, simulation models can be used for assessing vulnerability, economic impact, measuring performance, emergency preparedness and how health systems are interdependent on other systems.32 Using simulation models is an important part of ST as it helps predict the impact a change will have and compare possible solutions to a problem.8 To transform the six ST characteristics into an operational framework for use in health actions, a checklist was created to explore the use of each characteristic in the formulation, design and evaluation of health actions.
The checklist, presented in table 2, shows each of the six framework characteristics in separate categories, as well as the corresponding checklist components and ST tools for each characteristic. The checklist was created as a guide to provoke ideas of how to apply an ST approach. The outlined characteristics and items do not have to be completely checked off to adequately incorporate the ST characteristic in the health action. Additionally, the ST characteristics are not intended to be considered in order.
**Table 2**
| Systems Thinking for Health Actions checklist | Relevant systems thinking tools |
| --- | --- |
| Recognising and understanding interconnections and system structure. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Identified components of the health system. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Visually or textually showed the connections between components of the health system. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Conducted focus groups and/or interviews of key stakeholders to understand the health system better. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Invited other relevant sectors to participate in the design of the intervention. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Recognised the need for stakeholder involvement. | Stakeholder mapping/analysis.Social network analysis.Analysis of industry documents, tactics and strategies.Stakeholder interviews.Sociogram.Process mapping.Causal loop diagram.Logic models.Reflective practice. |
| Identifying and understanding feedback. | Causal loop diagramming.Markov modelling.Stakeholder interviews.Agent-based modelling.Stock and flow diagrams.Systemic policy analysis.Logic models.Sociogram. |
| Visually or textually addressed the feedback loops that exist in the health system. | Causal loop diagramming.Markov modelling.Stakeholder interviews.Agent-based modelling.Stock and flow diagrams.Systemic policy analysis.Logic models.Sociogram. |
| Identified the positive and negative effects one component of the health system has on other components. | Causal loop diagramming.Markov modelling.Stakeholder interviews.Agent-based modelling.Stock and flow diagrams.Systemic policy analysis.Logic models.Sociogram. |
| Identifying leverage points. | Iceberg tool.Scenario planning.Decision tree modelling.Logic models.Group model building.Systems dynamics modelling.Focus groups and stakeholder interviews.Business process mapping/discrete event modelling. |
| Determined the root causes of a problem through pictorial or written mapping. | Iceberg tool.Scenario planning.Decision tree modelling.Logic models.Group model building.Systems dynamics modelling.Focus groups and stakeholder interviews.Business process mapping/discrete event modelling. |
| Attempted to identify gaps. | Iceberg tool.Scenario planning.Decision tree modelling.Logic models.Group model building.Systems dynamics modelling.Focus groups and stakeholder interviews.Business process mapping/discrete event modelling. |
| Determined the key actions for leverage points. | Iceberg tool.Scenario planning.Decision tree modelling.Logic models.Group model building.Systems dynamics modelling.Focus groups and stakeholder interviews.Business process mapping/discrete event modelling. |
| Understanding dynamic behaviour. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Showed how a problem changes over time. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Addressed problems between components of the health system. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Predicted the impact a change to one component of the health system has on the rest of the system. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Identified how components of the health system change over time. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Addressed path dependence. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Developed a mechanism to identify emerging behaviours in the health system. | Causal loop diagram.Behaviour over time graphs.Dynamic thinking.Innovation/change management history.Systems archetypes.Stock and flow diagram.Causal loop diagram with variable distinction.Table differentiating the variables. |
| Using models to suggest possible solutions to a problem. | Conceptual model.Theory of change. |
| Explained the expected outcome of and action on the health system. | Conceptual model.Theory of change. |
| Explained why the expected outcome is anticipated. | Conceptual model.Theory of change. |
| Used a diagram, descriptive text or a pictorial model to represent the system. | Conceptual model.Theory of change. |
| Creating simulation models for testing policies. | Agent-based models.Systems dynamics models.Scenario planning models.Simulation models. |
| Used qualitative and quantitative data to create models. | Agent-based models.Systems dynamics models.Scenario planning models.Simulation models. |
| Used identified leverage points to test a change. | Agent-based models.Systems dynamics models.Scenario planning models.Simulation models. |
| Interpreted model outcomes. | Agent-based models.Systems dynamics models.Scenario planning models.Simulation models. |
| Compared solutions from different leverage points. | Agent-based models.Systems dynamics models.Scenario planning models.Simulation models. |
## Data analysis
The National Action Plan and key informant interviews from the COVID-19 response in Pakistan were analysed using thematic analysis with a deductive approach.33 We applied the STHA framework to analyse the data. First, the interview transcripts and National Action Plan were read and a codebook was created using the STHA framework as a guide. The text was highlighted using the corresponding themes from the codebook. With the relevant areas of the transcript selected, the highlighted text was applied to the six attributes of the framework to determine if and where the selected text fit best. Microsoft Excel was used to manage the data extracted from the transcripts and the data were coded by JT.
The Consolidated Criteria for Reporting Qualitative Research checklist was consulted for reporting this qualitative research.34 The participants gave verbal consent to participate in the interviews, which were recorded.
The National Action Plan identified the system structure by listing the key stakeholders and sectors (pp 18–20) involved in the COVID-19 response (including those outside the health system) and the actions expected to be taken by each one of them (p 104). Furthermore, the National Action Plan also textually showed the connections and emergent behaviour between the components of the system through ‘rapidly establishing and strengthening coordination to deliver strategic, technical and operational support through existing mechanisms and country partnerships’ (p 11) and creating a ‘policy framework for federal, provincial and regional stakeholders for building capacity to prevent, detect and respond to any events due to COVID-2019 or other novel pathogens with pandemic potential in Pakistan’ (p 9). The action plan also recognised the need for stakeholder involvement and included the relevant stakeholders in the COVID-19 surveillance system (p 14) and the development of the Risk Communication and Community Engagement initiative (p 17).
Respondents 2 and 3 identified the use of process mapping in the district, which assisted with identifying the stakeholders involved in the system. Respondent 3 mentioned the value of mapping stakeholder connections and roles to see how they impact each other. Additionally, respondents 2 and 3 also described identifying emergent behaviour in the system using reflective practice.
To identify and understand feedback the COVID-19 action plan implemented a monitoring and evaluation plan for constant improvement of the COVID-19 response (p 20). This included a parallel evaluation to continually point out areas for improvement. Additionally, the action plan identified the effect that one component of the system (specifically funding) can have on other components by acknowledging the roles that funding has on surveillance structures, data, laboratory diagnostic capacity, case management, stockpiling and logistics, infection prevention and control, burial policy and risk communication (p 17).
Respondent 3 explained how understanding the connections between the components of the process map assisted in realising the effect one component of the system has on another: ‘[process mapping] gave a very clear pattern of how things were how many stakeholders were involved in everything, and the reflective processes, has had a lot as well you know, to be honest.’ The COVID-19 action plan in Pakistan identified the gaps and leverage points in their COVID-19 system response. ‘ Assessments of risks and capacities to determine priorities for emergency preparedness’ were conducted (p 12). Among the gaps mentioned in the action plan were the capacities of case management, risk communication and infection prevention and control at health facilities (p 13), and the disease outbreak management system needing to be strengthened (p 14). The action plan also identified 19 key action areas or leverage points where actions could help minimise the spread of COVID-19 (pp 20–31).
The action plan also addressed the system’s dynamic behaviour by predicting how preparedness in the initial phase and strict containment in the second phase would determine the impact of the virus (p 8). It also mentioned ‘strengthening and reforms of the organizational, structural and coordination mechanisms to ensure the maximum level of preparedness over time’ (p 10).
In summary, four out of the six characteristics of the STHA framework were identified in Pakistan’s COVID-19 response.
## Discussion
ST has been a commonly used approach in various disciplines to address multifaceted problems in CAS.6 Adopting an ST approach is an attractive method in the field of health systems, but there is still a lack of understanding of the practical uses of ST.35 This framework aimed to bridge the gap between theoretical ST and practical ST.13 33 Six key ST characteristics were identified in the framework: [1] recognising and understanding interconnections and system structure, [2] identifying and understanding feedback, [3] identifying leverage points, [4] understanding dynamic behaviour, [5] using mental models to suggest possible solutions to a problem and [6] creating simulation models to test policies. We identified two potential applications for the framework and checklist: [1] Prospectively, to support in the design or implementation of health actions. Applying the framework prospectively can be done as a guide to translate ST concepts into practical steps that can be integrated in the design or implementation of a health action. [ 2] Retrospectively, to investigate where ST was applied and where it can be further applied the next time. The case study provided in this article is a retrospective example of applying the STHA framework.
We developed the STHA checklist to act as a guide to explore the application of ST prospectively or retrospectively in health actions. Checklists ease work in demanding or tense situations and have been increasingly used in healthcare.36 Checklists help promote active cooperation and communication among stakeholders.37 Therefore, using our checklist to assist in translating the application of ST concepts into practical steps can be beneficial. The checklist, which should not be taken as a strict and linear document, can help relieve some of the barriers to applying ST, such as many stakeholders understanding that ST requires sophisticated and resource-intensive interventions, as well as the lack of knowledge on how to start using ST.
The application of the STHA framework to the COVID-19 response in Pakistan revealed that the less complex characteristics were applied throughout the response despite the exact ST terminology not necessarily being mentioned. In our interviews, we identified several ST tools being used by district health officials, such as reflective practice or process mapping.38 The district officials and the research team used these tools as part of the ST-DHS initiative. In addition to the use of ST tools, we also identified other ST characteristics in the COVID-19 response (such as the understanding of dynamic behaviour in the National Action Plan) that were not influenced by the ST-DHS initiative. This highlights how ST is often used without being explicitly mentioned. Given the high complexity and changing environment of the pandemic, the health officials had to apply holistic responses.
The COVID-19 pandemic presented a crisis where policy makers had to quickly respond to a threat without knowing the exact extent of it.35 Therefore, the lack of application of the other three categories in the framework could be due to the time-sensitive nature of the pandemic and the lack of time between creating policies and implementing them. In previous literature, several barriers have been described to the application of ST approaches such as assumed costliness, lack of understanding, competing political interests across health and non-health actors, which often lead to prioritisation of vertical programmes, work in silos and difficulty ensuring meaningful multistakeholder involvement.39 Although the framework cannot mend all of these perceived barriers, it can assist in informing health action creators with more knowledge of using ST tools. Additionally, by revealing that ST exists in many health actions, it can show that the use of ST does not have to be costly, as it is already being applied to health actions without further costs.
In the Pakistan case study, we were able to validate the framework and determine that the framework was effective in identifying ST characteristics in the action plan and key informant interviews. However, sometimes it was difficult to determine if a section of text adequately included an ST characteristic. To lessen these uncertainties, further defining the characteristics will ease the use of the checklist in the future. Additionally, the checklist was only applied retrospectively to assure its use in assisting with the application of ST in impending health actions, the framework should also be validated in prospective cases.
To create the framework, the literature review was not limited to ST in health systems, rather other disciplines were also included (ie, road traffic safety, policymaking and food production). The limitations of this method were that not all disciplines were included in the search, rather disciplines that were previously well known for ST and expert inputs. In addition, the search was limited to English language, manuscripts that could be accessed through the University of Basel, manuscripts that were found on PubMed and Google Scholar and there was only one reviewer. Searching only two academic research databases may have limited the number of results retrieved for the literature review, thereby missing manuscripts that should have been included.40 Therefore, when developing the STHA framework further, a more extensive literature review could be conducted, with the number of disciplines and databases expanded. In addition, expert inputs from other systems thinkers would be helpful to expand the checklist developed for each characteristic of ST. The interviews conducted for testing the framework were specific to the ST-DHS initiative, meaning that the respondents already had knowledge of ST, which could have been a bias in their answers. The framework was also applied to only three interviews, with the respondents having similar positions in the district health system, which may have limited the ideas that interviewing other positions may have added. Expanding the application of the STHA framework to more interviews and with a wider range of positions could assist in further developing it. This was a first pilot of the STHA framework. In order for it to be relevant for a wider set of topics, actions and stakeholders, as well as context, further research is needed to identify how to improve it.
## Conclusion
This paper aimed to bridge the gap between theoretical and practical ST in health actions. Stave and Hopper8 created an excellent starting point for bridging this gap by identifying the level of ST in individuals in their Taxonomy of Systems Thinking Objectives. The STHA framework has made additional progress in closing this gap by creating a tangible checklist for designing, implementing and evaluating health actions. The STHA framework is a new, innovative way to apply ST. This framework can be used retrospectively, and it can be a guide in the development of health actions to explore where ST can still be applied. Further research is needed to ensure the STHA framework reaches its full potential.
## Data availability statement
No data are available.
## Patient consent for publication
Not applicable.
## Ethics approval
This study involves human participants and was approved by the Swiss Ethics Committee in Basel and the Health Services Academy in Islamabad, Pakistan. Participants gave informed consent to participate in the study before taking part.
## References
1. Pype P, Mertens F, Helewaut F. **Healthcare teams as complex adaptive systems: understanding team behaviour through team members’ perception of interpersonal interaction**. *BMC Health Serv Res* (2018) **18**. DOI: 10.1186/s12913-018-3392-3
2. Swanson RC, Cattaneo A, Bradley E. **Rethinking health systems strengthening: key systems thinking tools and strategies for transformational change**. *Health Policy Plan* (2012) **27 Suppl 4** iv54-61. DOI: 10.1093/heapol/czs090
3. Savigny D, Adam T. *Alliance for Health Policy and Systems Research* (2009)
4. **Healthy development: the world bank strategy for health, nutrition, and population results**. (2007)
5. Hassan I, Obaid F, Ahmed R. **A systems thinking approach for responding to the COVID-19 pandemic**. *East Mediterr Health J* (2020) **26** 872-6. DOI: 10.26719/emhj.20.090
6. Peters DH. **The application of systems thinking in health: why use systems thinking?**. *Health Res Policy Syst* (2014) **12**. DOI: 10.1186/1478-4505-12-51
7. Shaked H, Schechter C. *Systems Thinking for School Leaders* (2017). DOI: 10.1007/978-3-319-53571-5_25
8. Stave KA, Hopper M. **What constitutes systems thinking: A proposed taxonomy**. (2007)
9. Richmond B. **Systems thinking/system dynamics: let’s just get on with it**. *Syst Dyn Rev* (1994) **10** 135-57. DOI: 10.1002/sdr.4260100204
10. Trochim WM, Cabrera DA, Milstein B. **Practical challenges of systems thinking and modeling in public health**. *Am J Public Health* (2006) **96** 538-46. DOI: 10.2105/AJPH.2005.066001
11. Adam T, de Savigny D. **Systems thinking for strengthening health systems in lmics: need for a paradigm shift**. *Health Policy Plan* (2012) **27 Suppl 4** iv1-3. DOI: 10.1093/heapol/czs084
12. **Systems thinking for noncommunicable disease prevention policy: guidance to bring systems approaches into practice**. (2022)
13. de Savigny D, Blanchet K, Adam T. *Applied systems thinking for health systems research: A methodological handbook* (2017)
14. Rusoja E, Haynie D, Sievers J. **Thinking about complexity in health: a systematic review of the key systems thinking and complexity ideas in health**. *J Eval Clin Pract* (2018) **24** 600-6. DOI: 10.1111/jep.12856
15. Kwamie A, Ha S, Ghaffar A. **Applied systems thinking: unlocking theory, evidence and practice for health policy and systems research**. *Health Policy Plan* (2021) **36** 1715-7. DOI: 10.1093/heapol/czab062
16. Sant Fruchtman C, Bilal Khalid M, Keakabetse T. **Digital communities of practice: one step towards decolonising global health partnerships**. *BMJ Glob Health* (2022) **7**. DOI: 10.1136/bmjgh-2021-008174
17. Grant MJ, Booth A. **A typology of reviews: an analysis of 14 review types and associated methodologies**. *Health Info Libr J* (2009) **26** 91-108. DOI: 10.1111/j.1471-1842.2009.00848.x
18. Westhorp G. **Realist impact evaluation: an introduction**. *Overseas Development Institute* (2014)
19. Vindrola-Padros C, Johnson GA. **Rapid techniques in qualitative research: a critical review of the literature**. *Qual Health Res* (2020) **30** 1596-604. DOI: 10.1177/1049732320921835
20. 20Ministry of National Health Services. National action plan for coronavirus disease (COVID-19) pakistan. 2020: 136.. *National action plan for coronavirus disease (COVID-19) pakistan* (2020) 136
21. Braun V, Clarke V. **To saturate or not to saturate? Questioning data saturation as a useful concept for thematic analysis and sample-size rationales**. *Qualitative Research in Sport, Exercise and Health* (2021) **13** 201-16. DOI: 10.1080/2159676X.2019.1704846
22. Amran A. **Iceberg model**. (2022)
23. Wilkinson J, Goff M, Rusoja E. **The application of systems thinking concepts, methods, and tools to global health practices: an analysis of case studies**. *J Eval Clin Pract* (2018) **24** 607-18. DOI: 10.1111/jep.12842
24. Paina L, Bennett S, Ssengooba F. **Advancing the application of systems thinking in health: exploring dual practice and its management in Kampala, Uganda**. *Health Res Policy Syst* (2014) **12** 41. DOI: 10.1186/1478-4505-12-41
25. Gowda NR, Wankar A, Arya SK. **Feedback system in healthcare: the why, what and how**. *IJMS* (2020) **12** 52. DOI: 10.5539/ijms.v12n1p52
26. Rwashana AS, Nakubulwa S, Nakakeeto-Kijjambu M. **Advancing the application of systems thinking in health: understanding the dynamics of neonatal mortality in Uganda**. *Health Res Policy Syst* (2014) **12** 36. DOI: 10.1186/1478-4505-12-36
27. Zukowski N, Davidson S, Yates MJ. **Systems approaches to population health in Canada: how have they been applied, and what are the insights and future implications for practice?**. *Can J Public Health* (2019) **110** 741-51. DOI: 10.17269/s41997-019-00230-3
28. Hodges S, Ferreira K, Israel N. **Strategies for system of care development: making change in complex systems**. (2006) **15**
29. Glenn J, Kamara K, Umar ZA. **Applied systems thinking: a viable approach to identify leverage points for accelerating progress towards ending neglected tropical diseases**. *Health Res Policy Syst* (2020) **18**. DOI: 10.1186/s12961-020-00570-4
30. Assaraf OB-Z, Orion N. **Development of system thinking skills in the context of earth system education**. *J Res Sci Teach* (2005) **42** 518-60. DOI: 10.1002/tea.20061
31. Powell KE, Kibbe DL, Ferencik R. **Systems thinking and simulation modeling to inform childhood obesity policy and practice**. *Public Health Rep* (2017) **132** 33S-38S. DOI: 10.1177/0033354917723601
32. McLean C, Lee YT, Jain S. **Modeling and simulation of healthcare systems for homeland security applications NIST IR 7784**. (2011)
33. Burnard P, Gill P, Stewart K. **Analysing and presenting qualitative data**. *Br Dent J* (2008) **204** 429-32. DOI: 10.1038/sj.bdj.2008.292
34. Tong A, Sainsbury P, Craig J. **Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups**. *Int J Qual Health Care* (2007) **19** 349-57. DOI: 10.1093/intqhc/mzm042
35. Bradley DT, Mansouri MA, Kee F. **A systems approach to preventing and responding to COVID-19**. *EClinicalMedicine* (2020) **21**. DOI: 10.1016/j.eclinm.2020.100325
36. Greig P, Maloney A, Higham H. **Emergencies in general practice: could checklists support teams in stressful situations?**. *Br J Gen Pract* (2020) **70** 304-5. DOI: 10.3399/bjgp20X709373
37. Grzybowski A, Patryn R, Sak J. **Surgical safety checklist. procedure for the safety of procedures**. *Int J Surg* (2017) **41** 174-5. DOI: 10.1016/j.ijsu.2017.03.087
38. Zafar S, Fruchtman CS, Bilal Khalid M. **Lessons learnt of the COVID-19 contact tracing strategy in islamabad capital territory, pakistan using systems thinking processes**. *Front Public Health* (2022) **10**. DOI: 10.3389/fpubh.2022.909931
39. El-Jardali F, Adam T, Ataya N. **Constraints to applying systems thinking concepts in health systems: a regional perspective from surveying stakeholders in eastern Mediterranean countries**. *Int J Health Policy Manag* (2014) **3** 399-407. DOI: 10.15171/ijhpm.2014.124
40. Bramer WM, Rethlefsen ML, Kleijnen J. **Optimal database combinations for literature searches in systematic reviews: a prospective exploratory study**. *Syst Rev* (2017) **6** 245. DOI: 10.1186/s13643-017-0644-y
41. McGill E, Er V, Penney T. **Evaluation of public health interventions from a complex systems perspective: a research methods review**. *Soc Sci Med* (2021) **272** 113697. DOI: 10.1016/j.socscimed.2021.113697
42. Karim A, Cobos Munoz D, Mäusezahl D. **A systems approach to assessing complexity in health interventions: an effectiveness decay model for integrated community case management**. *Glob Health Action* (2020) **13**. DOI: 10.1080/16549716.2020.1794106
43. Owen B, Brown AD, Kuhlberg J. **Understanding a successful obesity prevention initiative in children under 5 from a systems perspective**. *PLoS ONE* (2018) **13**. DOI: 10.1371/journal.pone.0195141
44. Kwamie A, van Dijk H, Agyepong IA. **Advancing the application of systems thinking in health: realist evaluation of the leadership development programme for district manager decision-making in Ghana**. *Health Res Policy Syst* (2014) **12** 29. DOI: 10.1186/1478-4505-12-29
45. Varghese J, Kutty VR, Paina L. **Advancing the application of systems thinking in health: understanding the growing complexity governing immunization services in Kerala, India**. *Health Res Policy Syst* (2014) **12**. DOI: 10.1186/1478-4505-12-47
46. Bensberg M, Allender S, Sacks G. **Building a systems thinking prevention workforce**. *Health Promot J Austr* (2020) **31** 436-46. DOI: 10.1002/hpja.325
47. Agyepong IA, Aryeetey GC, Nonvignon J. **Advancing the application of systems thinking in health: provider payment and service supply behaviour and incentives in the Ghana National health insurance scheme -- a systems approach**. *Health Res Policy Syst* (2014) **12** 35. DOI: 10.1186/1478-4505-12-35
48. Hamdani Y, Jetha A, Norman C. **Systems thinking perspectives applied to healthcare transition for youth with disabilities: a paradigm shift for practice, policy and research**. *Child Care Health Dev* (2011) **37** 806-14. DOI: 10.1111/j.1365-2214.2011.01313.x
49. Schoenenberger LK, Bayer S, Ansah JP. **Emergency department crowding in singapore: insights from a systems thinking approach**. *SAGE Open Med* (2016) **4**. DOI: 10.1177/2050312116671953
50. Clarke B, Kwon J, Swinburn B. **Understanding the dynamics of obesity prevention policy decision-making using a systems perspective: a case study of healthy together Victoria**. *PLoS One* (2021) **16**. DOI: 10.1371/journal.pone.0245535
51. Friel S, Pescud M, Malbon E. **Using systems science to understand the determinants of inequities in healthy eating**. *PLOS ONE* (2017) **12**. DOI: 10.1371/journal.pone.0188872
52. Knai C, Petticrew M, Mays N. **Systems thinking as a framework for analyzing commercial determinants of health**. *Milbank Q* (2018) **96** 472-98. DOI: 10.1111/1468-0009.12339
53. Mutale W, Ayles H, Bond V. **Application of systems thinking: 12-month postintervention evaluation of a complex health system intervention in Zambia: the case of the BHOMA**. *J Eval Clin Pract* (2017) **23** 439-52. DOI: 10.1111/jep.12354
54. Kunpeuk W, Teekasap P, Kosiyaporn H. **Understanding the problem of access to public health insurance schemes among Cross-border migrants in Thailand through systems thinking**. *Int J Environ Res Public Health* (2020) **17**. DOI: 10.3390/ijerph17145113
55. Pocock NS, Chan Z, Loganathan T. **Moving towards culturally competent health systems for migrants? applying systems thinking in a qualitative study in malaysia and thailand**. *PLoS ONE* (2020) **15**. DOI: 10.1371/journal.pone.0231154
56. Brereton CF, Jagals P. **Applications of systems science to understand and manage multiple influences within children’s environmental health in least developed countries: a causal loop diagram approach**. *Int J Environ Res Public Health* (2021) **18**. DOI: 10.3390/ijerph18063010
57. BeLue R, Carmack C, Myers KR. **Systems thinking tools as applied to community-based participatory research: a case study**. *Health Educ Behav* (2012) **39** 745-51. DOI: 10.1177/1090198111430708
58. Rosas SR. **Systems thinking and complexity: considerations for health promoting schools**. *Health Promot Int* (2017) **32** 301-11. DOI: 10.1093/heapro/dav109
59. Allender S, Brown AD, Bolton KA. **Translating systems thinking into practice for community action on childhood obesity**. *Obes Rev* (2019) **20 Suppl 2** 179-84. DOI: 10.1111/obr.12865
60. Riley T, Hopkins L, Gomez M. **A systems thinking methodology for studying prevention efforts in communities**. *Syst Pract Action Res* (2021) **34** 555-73. DOI: 10.1007/s11213-020-09544-7
61. Naumann RB, Sandt L, Kumfer W. **Systems thinking in the context of road safety: can systems tools help us realize a true “ safe systems ” approach?**. *Curr Epidemiol Rep* (2020) **7** 343-51. DOI: 10.1007/s40471-020-00248-z
62. Xue H, Slivka L, Igusa T. **Applications of systems modelling in obesity research: systems modelling review**. *Obes Rev* (2018) **19** 1293-308. DOI: 10.1111/obr.12695
63. Minyard KJ, Ferencik R, Ann Phillips M. **Using systems thinking in state health policymaking: an educational initiative**. *Health Syst (Basingstoke)* (2014) **3** 117-23. DOI: 10.1057/hs.2013.17
64. Jetha A, Pransky G, Hettinger LJ. **Capturing complexity in work disability research: application of system dynamics modeling methodology**. *Disabil Rehabil* (2016) **38** 189-94. DOI: 10.3109/09638288.2015.1031291
65. Diez Roux AV. **Complex systems thinking and current impasses in health disparities research**. *Am J Public Health* (2011) **101** 1627-34. DOI: 10.2105/AJPH.2011.300149
66. Broomhead T. *Systems science and oral health: implications for dental public health* (2019) 55-62
67. Carey G, Malbon E, Carey N. **Systems science and systems thinking for public health: a systematic review of the field**. *BMJ Open* (2015) **5**. DOI: 10.1136/bmjopen-2015-009002
68. Tracy M, Cerdá M, Keyes KM. **Agent-based modeling in public health: current applications and future directions**. *Annu Rev Public Health* (2018) **39** 77-94. DOI: 10.1146/annurev-publhealth-040617-014317
69. Malik AU, Willis CD, Hamid S. **Advancing the application of systems thinking in health: advice seeking behavior among primary health care physicians in Pakistan**. *Health Res Policy Syst* (2014) **12**. DOI: 10.1186/1478-4505-12-43
70. Calancie L, Anderson S, Branscomb J. **Using behavior over time graphs to Spur systems thinking among public health practitioners**. *Prev Chronic Dis* (2018) **15**. DOI: 10.5888/pcd15.170254
71. Durski KN, Naidoo D, Singaravelu S. **Systems thinking for health emergencies: use of process mapping during outbreak response**. *BMJ Glob Health* (2020) **5**. DOI: 10.1136/bmjgh-2020-003901
72. Conte KP, Davidson S. **Using a'‘ich picture'’to facilitate systems thinking in research coproduction**. *Health Res Policy Sys* (2020) **18**. DOI: 10.1186/s12961-019-0514-2
|
---
title: 'The long-term effects of Kerala Diabetes Prevention Program on diabetes incidence
and cardiometabolic risk: a study protocol'
authors:
- Tilahun Haregu
- T. R. Lekha
- Smitha Jasper
- Nitin Kapoor
- Thirunavukkarasu Sathish
- Jeemon Panniyammakal
- Robyn Tapp
- Kavumpurathu Raman Thankappan
- Ajay Mahal
- Pilvikki Absetz
- Edwin B. Fisher
- Brian Oldenburg
journal: BMC Public Health
year: 2023
pmcid: PMC10030347
doi: 10.1186/s12889-023-15392-6
license: CC BY 4.0
---
# The long-term effects of Kerala Diabetes Prevention Program on diabetes incidence and cardiometabolic risk: a study protocol
## Abstract
### Introduction
India currently has more than 74.2 million people with Type 2 Diabetes Mellitus (T2DM). This is predicted to increase to 124.9 million by 2045. In combination with controlling blood glucose levels among those with T2DM, preventing the onset of diabetes among those at high risk of developing it is essential. Although many diabetes prevention interventions have been implemented in resource-limited settings in recent years, there is limited evidence about their long-term effectiveness, cost-effectiveness, and sustainability. Moreover, evidence on the impact of a diabetes prevention program on cardiovascular risk over time is limited.
### Objectives
The overall aim of this study is to evaluate the long-term cardiometabolic effects of the Kerala Diabetes Prevention Program (K-DPP). Specific aims are 1) to measure the long-term effectiveness of K-DPP on diabetes incidence and cardiometabolic risk after nine years from participant recruitment; 2) to assess retinal microvasculature, microalbuminuria, and ECG abnormalities and their association with cardiometabolic risk factors over nine years of the intervention; 3) to evaluate the long-term cost-effectiveness and return on investment of the K-DPP; and 4) to assess the sustainability of community engagement, peer-support, and other related community activities after nine years.
### Methods
The nine-year follow-up study aims to reach all 1007 study participants (500 intervention and 507 control) from 60 randomized polling areas recruited to the original trial. Data are being collected in two phases. In phase 1 (Survey), we are admintsering a structured questionnaire, undertake physical measurements, and collect blood and urine samples for biochemical analysis. In phase II, we are inviting participants to undergo retinal imaging, body composition measurements, and ECG. All data collection is being conducted by trained Nurses. The primary outcome is the incidence of T2DM. Secondary outcomes include behavioral, psychosocial, clinical, biochemical, and retinal vasculature measures. Data analysis strategies include a comparison of outcome indicators with baseline, and follow-up measurements conducted at 12 and 24 months. Analysis of the long-term cost-effectiveness of the intervention is planned.
### Discussion
Findings from this follow-up study will contribute to improved policy and practice regarding the long-term effects of lifestyle interventions for diabetes prevention in India and other resource-limited settings.
### Trial registration
Australia and New Zealand Clinical Trials Registry–(updated from the original trial)ACTRN12611000262909; India: CTRI/$\frac{2021}{10}$/037191.
## Introduction
Based on recent estimates, there are 537 million people with diabetes globally with 433 million ($80.6\%$) living in low and middle-income countries such as India [1]. The global number of people with diabetes is projected to reach 783 million by 2045. It is estimated that $94\%$ of this increase will occur in low and middle-income countries. Almost $90\%$ of people with undiagnosed diabetes live in low- and middle-income countries [1].
India has the second highest number of people with diabetes (74.2 million), accounting for 1 in 7 of all adults living with diabetes worldwide, and this is expected to increase to 124.9 million by 2045. About $57\%$ of the estimated number of people with diabetes in India are undiagnosed [2]. Partly due to its large population size, India has the second-highest annual number of deaths from diabetes, approximately 0.6 million [1]. In this regard, the prevention and management of diabetes, through community-based approaches need urgent attention [3].
The efficacy, effectiveness, and implementation of lifestyle interventions aimed at preventing the onset of type 2 diabetes have been well-established at least in high-income countries. A review of 38 implementation trials has demonstrated a reduction in type 2 diabetes incidence of between 40–$60\%$ [4]. The lifestyle interventions used in these implementation trials have been shown to be more effective when delivered via groups, such as peer support groups. However, most of the trials have been conducted in high-income countries. In addition, there is limited evidence about their effects and sustainability after the active intervention phase.
In this study protocol paper, we briefly discuss the findings of a community-based diabetes prevention program conducted in the Indian state of Kerala, the Kerala Diabetes Prevention Program (K-DPP), from 2013–2016. We also present the rationale to conduct a nine-year follow-up of the original K-DPP participants; the aims, measures to be undertaken, and the analysis plan.
## Summary of Kerala Diabetes Prevention Program (K-DPP) findings at 12 and 24 months
The Kerala Diabetes Prevention Program (K-DPP) is a cluster randomized trial of a group-based and peer-led lifestyle intervention among adults at high-risk for type 2 diabetes in Kerala, India [5]. The intervention was adapted from evidence-based lifestyle interventions implemented in high-income countries, namely, Finland, the United States, and Australia [6]. A total of 60 polling areas (1007 participants) were randomized to the intervention arm (500 participants) or control arm (507 participants) in rural Kerala, India. The intervention was administered over 12 months period. Data on sociodemographic factors, lifestyle or behavioral factors, psychosocial, anthropometric, and biochemical measures were collected at baseline, 12 months, and 24 months. Among K-DPP participants, diabetes had developed in $17.1\%$ of control participants and $14.9\%$ of intervention participants (RR = 0.88; $95\%$ CI: 0.66 -1.16, $$P \leq 0.36$$) at 24 months [7]. There was a significant reduction in diabetes incidence among a sub-group of participants with impaired glucose tolerance (IGT) (RR = 0.61, $95\%$ CI: 0.41–0.92) in keeping with findings from the landmark clinical trials conducted on people with IGT [8–10]. The K-DPP program was also associated with a $32\%$ reduction in 10-year cardiovascular disease (CVD) risk in the intervention group compared to the control group (RR = 0.68; $95\%$ CI: 0.47–0.99; $$P \leq 0.047$$) among those aged 40 years or more [11]. Effectiveness evaluation has demonstrated significant improvements in certain key CVD risk factors and the physical functioning score of the health-related Quality of life (HRQoL) scale [7]. Moreover, this community-based peer-support lifestyle intervention was found to be cost-effective in individuals at high risk of developing diabetes in India over a 2-year period [11, 12]. The implementation evaluation has shown that the K-DPP program was feasible and acceptable in changing lifestyle behaviors in high-risk individuals [13]. Measures of program reach, adoption, and implementation were excellent for K-DPP. Twenty-nine of the 30 intervention communities delivered all intervention components. Two-thirds of intervention participants attended more than $50\%$ of all peer-led group sessions and program recognition was high among local leaders and champions. There was also very strong community engagement and program sustainability in intervention communities at 24 months with $\frac{29}{30}$ groups having conducted other diabetes prevention and health promotion activities in their communities [13].
## Diabetes incidence and cardiometablic risk
There is limited long-term data on the effect of lifestyle interventions on diabetes incidence and cardiomentabolic risk, especially in low- and middle-income countries. Extended follow-up studies of previous lifestyle intervention studies, such as the Da Qing Diabetes Prevention Study and the U.S. Diabetes Prevention Program, showed benefits in reducing cardiovascular events after many years of follow-up [14, 15]. However, the current evidence is limited and more follow-up studies, particularly from resource-constrained settings, are needed to address this gap. The additional nine years of follow-up in K-DPP enable an assessment of the long-term benefit of lifestyle interventions in terms of reducing diabetes incidence and other cardiometabolic risk. Current and more detailed analyses, including measures of glycemic levels, serum lipids, heart function (ECG), renal function, and the retinal microvaculature, can supplement previous findings. Given the strong study findings and trends at 24 months, it is now important to undertake longterm follow-up to evaluate the program effectiveness, cost-effectiveness, and sustainability of K-DPP.
## Multimorbidity
Diabetes mellitus is one of the main drivers of multimorbidity as it is associated with a large number of risk factors and complications [16]. Consequently, patients with type 2 diabetes mellitus often live with and develop multiple co-occurring conditions with negative impacts on disease management and quality of life [17]. Given the rich baseline and short-term follow-up data from people at high risk of diabetes at enrolment, K-DPP would enable an analysis of progressive development of various multimorbidities, including heart disease and kidney disease, over time in the study population. It also enables evaluation of the association between multimorbidity and glycemic levels in this population.
## Retinal vasculature
Changes in the retinal microvasculature, such as retinal arteriolar narrowing, venular dilatation, and retinopathy, have been associated with various systemic conditions, including impaired fasting glucose, obesity, blood pressure, incident hypertension, incident diabetes, and cardiovascular disease [18]. Retinal vascular calibre has been shown, in prospective cohorts, to be a predictor of type 2 diabetes and impaired fasting glucose. More recent studies and analyses implicate that a wider retinal venular calibre is a marker of chronic hyperglycemia, prediabetes, and microvascular complications of diabetic retinopathy and nephropathy [19]. Wider retinal venular calibre has also been associated changes in body composition, carotid artery disease, incident coronary heart disease, and incident stroke [20]. As compared to retinal venular changes, retinal arteriolar changes are more predictive of hypertension-related comorbidities [21]. Retinal vascular geometry alterations are a novel marker to predict the progression of retinopathy [22]. A follow-up study of K-DPP enables evaluation of the differences in the retinal vasculature between the intervention group and control group who were at high risk for diabetes at enrolment. It will also enable an analysis of anthropological and metabolic parameters that influence retinal vasculature measurements and their differences between groups.
## Study aims
The overall aim of this study is to evaluate the long-term effects of the K-DPP on diabetes incidence and cardiometabolic risk at nine years following participant recruitment.
The specific objectives are:To estimate the effects of K-DPP on diabetes incidence and cardiometabolic risk at nine yearsTo assess retinal microvasculature, microalbuminuria, and electrocardiogram (ECG) abnormalities and their association with incident diabetes and cardiometabolic risk factorsTo evaluate the cost-effectiveness and return on investment of the K-DPP at nine yearsTo assess the sustainability of community engagement, peer support, and other related community activities after nine years
## Study design and setting/context
The current study is the nineth-year follow-up assessment of the K-DPP. The protocol paper for the original trial has been published elsewhere [5]. K-DPP was a cluster RCT conducted in 60 polling areas (clusters, electoral divisions with geographical boundaries) of Neyyattinkara taluk (subdistrict) in Trivandrum district, Kerala state, India. Participants were screened and recruited in 2013 and the 12-month intervention was completed in 2013–2014 [23, 24]. The study area has been affected by floods during monsoon season in more recent years [25]. particularly the study areas of Kattakada and Neyyatinkara. The continuum of care for chronic diseases during this period was affected adversely due to the impact on access to care and increased risk of other medical conditions. The recent COVID-19 pandemic has also severely affected the study area with cases peaking in October 2020, May 2021, and August 2021 [9]. Additionally, travel restrictions during lockdowns during COVID pandemic posed discontinuities in NCD care [26].
## Study population
The study population for K-DPP is adults 30–60 years of age who were at high risk for diabetes based on the Indian Diabetes Risk Score (IDRS ≥ 60) [27] at the time of enrolment. Of the total 1007 participants who were enrolled in K-DPP, $47\%$ were females. Each peer group selected two peer leaders (one male and one female) from among the participants with the assistance of the intervention team. The 60 peer groups each with 10 to 23 participants had an approximately equal number of males and females. The peer leaders were trained by the intervention team consisted of an intervention manager and an intervention assistant to deliver the intervention [7].
## Primary outcomes
Based on the available data at 12 and 24 months, we anticipate more than $90\%$ of K-DPP participants at baseline can provide complete and valid data for all key measures underlying diabetes incidence and 10-year CVD risk predicted using the Framingham Risk Score [28] at nine years. The proportion of participants (≥ 40 years of age at baseline) at high-rosk of CVD (with a predicted 10-year CVD risk greater than $20\%$ [29]) is expected to be $34.5\%$ in the control arm ($$n = 348$$) versus $17.9\%$ in the K-DPP arm ($$n = 337$$) after nine years, assuming a linear increasing trend overtime based on our data so far. The power to detect this difference is more than $97\%$ when accounting for an intra-cluster correlation of 0.02 at a two-sided $5\%$ level of significance.
## Secondary outcomes
Loss to follow-up has been very low so far, i.e., < $5\%$ at 24 months and a 7-year cohort study conducted in the same region had a follow-up rate of $92\%$ [30]. Assuming missing data (i.e., no lost outcome data due to reasons such as loss-to-follow-up or no blood sample) reaches $18\%$ in the control arm and $22\%$ in the intervention arm, 415 and 390 participants in the control and intervention arm, respectively are predicted to provide data at 9-years of follow-up. With the sample sizes of 415 and 390 participants in the control and intervention arms respectively, we shall be able to detect an effect size (i.e., the absolute difference in means divided by standard deviation) in other continuous outcomes (e.g., anthropometric, BP, microvascular and cholesterol measurements) of 0.23 with $80\%$ power at the two-sided $5\%$ significance level, when accounting for an intra-cluster correlation of 0.02. Power was derived using power twoprop and clustersampsi in Stata software (version 17 StataCorp LP, College Station, TX, USA).
## Intervention
Detailed description of the cultural adaptation and development of the K-DPP intervention has been published previously [6, 31]. Briefly, K-DPP involved a 12-month peer-support program that consisted of 15 group sessions: an introductory session delivered by the K-DPP team; two education sessions conducted by local experts; and 12 sessions delivered by trained lay peer leaders. Group sessions were held in the local community in a convenient neighbourhood facility and at convenient time during weekends. The peer group sessions aimed to increase physical activity, promote healthy eating habits, maintain appropriate body weight by balancing calorie intake and physical activity, tobacco cessation, reduce alcohol consumption, and ensure adequate sleep. In the control communities, an educational booklet concerning information about diabetes and its risk factors, as well as standard advice about lifestyle change were provided. After the implementation of interventions, a 12-month and 24-month follow-up assessment was conducted [5, 6].
## Data collection instruments and measurements
Most of the measurements undertaken in previous rounds of assessment are also undertaken in this follow-up assessment (Table 1). A detailed description of these measurements has been published [5]. Newly added measurements include an assessment of the retinal microvasculature, ECG, multimorbidity questionnaire, urine albumin, serum creatinine, sedentary behavior, and community engagement/sustainability questions. The measurements used in all rounds of assessment are summarized in Table 1 below. Table 1Measurements included in the four time points of assessmentVariableComponentMeasurement tools/questionsBaseline12 months24 months9-YearsSocio-demographyAge, sex, marital status, education, religion, occupation, household size, and monthly household expenditure✓✓✓✓Behavioral measuresTobacco useWHO STEPS questionnaire [32]✓✓✓✓Alcohol useWHO STEPS questionnaire [32]✓✓✓✓Physical activityGlobal Physical Activity Questionnaire [33]✓✓✓✓Sedentary behaviorSitting and screen timeXXX✓Sedentary behaviorTime spent in front of a screen [34]✓✓✓✓DietFood Frequency Questionnaire [35]✓✓✓XDiabetes knowledgeBarriers to healthy eatingScale designed for trial✓✓✓✓Barriers to PAAdapted from Booth et al. [ 36]✓✓✓✓Self-efficacyAdapted Schwarzer and Renner [37]✓✓✓✓Psychosocial measuresDepressionPatient Health Questionnaire-9 [38]✓✓✓✓StressChronic stress scale used in MESA study [39]✓✓✓✓AnxietyGeneral anxiety disorder scale [40]✓✓✓✓HRQoLShort Form-36 [41]✓✓✓✓Social supportENRICHD social support scale [42]✓✓✓✓Life satisfactionSingle question: Satisfaction with life✓✓✓✓Medical historyMedical historyCVD and diabetes-related medications✓✓✓✓Family historyUpdated family history of disease✓✓✓✓MultimorbidityAdapted from the WHO SAGE questionnaire [43]XXX✓Clinical measuresAnthropometryWaist circumference (Seca measuring tape)✓✓✓✓Height (Seca stadiometer)✓✓✓XWeight and body composition (TANITA)✓✓✓✓Bioimpedance (TANITA)✓✓✓✓Blood pressureOmron automatic blood pressure monitor✓✓✓✓Biochemical measuresPathologyGlycaemic control (fasting plasma glucose)✓✓✓✓Glycaemic control -OGTT✓✓✓XGlycaemic control (HbA1c)✓✓✓✓Lipid profile (TBC, HDL, LDL, triglycerides)✓✓✓✓Serum creatinineXXX✓Urine albuminXXX✓Assessment of heart conditionsECGXXX✓Retinal microvasculatureRetinal imaging measures(Retinal architecture, arteriolar and venular diameters, and tortuosity) [44]XXX✓MortalityStatus and self-reported causes of deathXXX✓Economic EvaluationHealthcare utilizationDirect and indirect costs associated with outpatient and inpatient healthcare services, sources of financing, and time away from work due to ill health + SF-36✓✓✓✓Community engagementSustainabilitySustainability – continued engagement in peer groups/activities; contacts with peer leaders; intention to continue engagementXXX✓WHO World Health Organization, SF-36 Short Form 36, MESA Multi-Ethnic Study of Atherosclerosis, OGTT Oral Glucose Tolerance Test, TBC Total Blood Cholesterol, HDL High Density Lipoprotein, LDL Low Density Lipopotein
## The research team
The research team consist of three research nurses, a research associate, and a project manager. The team attended a five-day training on the study objectives, data collection tools, and data collection procedures. They have also receive additional training on data entry using RedCap, Retinal Imaging, ECG, and TANITA measurements. Each of these training has practical sessions.
Phase I (Survey) data collection: *In this* phase, the research team collects Questionnaire-based data including weight and BP measurements and biochemical (blood and urine) samples. Members of the research team visit the houses of participants to collect data and samples. The samples are be transferred to A laboratory accredited by the National Accreditation Board for Laboratories(NABL) for testing.
Phase II (community-based clinics) data collection: *In this* phase, we identify fixed facilities in which all the equipment can be placed and organize community-based clinics at these community facilities on specific dates and invite participants to have their measurements (height, body composition analyzer (TANITA), refractometry, retinal measurements, and ECG)undertaken.
## Strategy to reach out to participants after nine years
During the preparation phase of the K-DPP follow-up study, participants were contacted via phone to check their availability and willingness to participate in the follow-up study. Unique K-DPP identification numbers and contact details of the study participants and local resource persons (LRP) from the previous assessments were used to reach to participants. For the survey data collection, the study team visited the homes of participants. Repeat visits are being organized if participants are not available on the first visit. The participants who denied consent are not part of the current evaluation. Concerted efforts to reach out to the missing participants (The participants who are not available at the time of data collection and whose contact details could not be traced) are being made. For community-based clinics, participants are invited to undertake the measurements.
## Long-term effectiveness outcomes
The key study outcomes are the incidence of diabetes and CVD risk at nine years after the intervention. Other primary outcomes include Blood pressure, weight, HbA1c, triglycerides, SF-36, and tobacco. Additionally, body mass index, fat percent, muscle mass, waist circumference, diastolic blood pressure, Fasting Plasma Glucose, lipid profile (total cholesterol, LDL cholesterol, and HDL cholesterol), current alcohol use, diet, and physical activity will be assessed. New measures of pre-clinical retinal microvasculature, heart function and renal disease are also being assessed. We are also considering multimorbidity as a secondary outcome in this follow-up study.
## Anthropometry and behavioral measures
Anthropometry (weight, total body fat percent, muscle mass, waist circumference) and Blood Pressure are measured using standard protocols. Self-reported physical activity and sedentary behavior are estimated using the Global physical activity questionnaire (GPAQ). Fruit and vegetable consumption, alcohol use, and tobacco use are assessed using the WHO STEPs Questionnaire.
Two TANITA machines (model SC330) used in the previous rounds of assessment were sent to the designated national TANITA service center for recalibration. Weight and body composition parameters will be measured using these TANITA body composition analyzers while the participant is standing still without footwear, with one foot on each side of the scale, facing forward, and arms at their side.
## Retinal vasculature measurement
A CANON CR-2/Plus/AF RX non-mydriatic retinal camera is being used to obtain retinal images. Two images of each eye are being taken (macular centred and disc centred), following a standard protocol. Image quality will be continuously monitored throughout the study. Images will be graded using fully automated software providing measures of arteriolar and venular diameter and tortuosity. The refractive status of the eyes are being assessed using aCanon RK-F2 Autorefractometer.
## ECG
We are using ECG equipment (KardioScreen-1612) to acquire the report that is being asssed to determine whether the ECG waves and intervals are normal or pathological. A standard operating procedure adapted from the American College of Cardiology guidelines [45]. KardioScreen records 12-lead ECG and works along with an application on Android Tablet/Mobile. ECGs are being recorded in a standardized way with the patient in the supine position, and leads will be placed. The patient’s clinical history is recorded along with the ECG. The application records ECG readings and stores them on the Cloud. A real-time automatic AI-based software (iMedrix) is fed by the recorded ECGs, and data classification and analytical monitoring of heart health parameters is being done. This includes heart rate [14], PR, QT/QTc, QRS duration, and the specific ECG patterns that occur based on electrophysiological changes in a diseased heart.
## Biochemistry
All biochemical measurements are following standard procedures. Samples are being centrifuged within 30 min of collection, and transported with dry ice to a nationally accredited laboratory. Plasma glucose are being measured with the Hexokinase method on a COBAS 6000 analyzer, with kits supplied by Roche Diagnostics. The quality of plasma glucose measurements from the proposed laboratory is good for the K-DPP trial; the intra-class correlation coefficient approached 1.0. HbA1c is being measured by High-Performance Liquid Chromatography on a D-10 BIORAD analyzer and lipids by enzymatic methods on COBAS 6000, kits supplied by Roche.
## Microalbuminuria
The random urine samples are being processed using Roche/Hitachi Cobas 501 analyzer system by Immunoturbidimetric assay with Roche reagent cassettes. Before running the samples, quality control procedures are being followed using Roche precinorm PUC and Precipath PUC. Besides, a monthly Biorad EQAS urine chemistry programme including microalbumin and creatinine is also being followed. Cobas 501 system automatically calculates microalbumin 24-h, random and microalbumin and creatinine ratio.
## Economic evaluation
The economic evaluation involves cost-effectiveness analysis of the intervention from both the health system and societal perspectives. The key outcome measure will be Quality Adjusted Life Years (QALY), which will be estimated using the utility values (SF-6D) derived from the SF-36 survey form. Costs being assessed include both direct medical and non-medical costs and indirect (lost days of productivity) costs associated with the intervention. These are be based on information on staff inputs by duration and type and resource use collected during the study. Any additional “knock-on” effect on hospital admissions or outpatient service use is being quantified and costed for each study participant. We are calculating direct costs to the health service and patients from health facility records, project financial accounts, and participant responses to a survey including questions on health service use and spending. These analyses will be based on updated estimates for health facility costs and personal costs Personal costs to individuals will be collected as part of the questionnaire administered to study participants at nine years. These include transportation costs and out-of-pocket spending for health services use, especially for medications and will capture how the individual financed these costs (e.g., borrowing, asset sales, dis-savings). Any indirect costs to patients arising from being absent from work due to the intervention or illness are also being estimated. Savings from lowered health service use associated with the intervention are subtracted from intervention costs to derive incremental costs at the 9-year mark.
## Program sustainability
We are assessing three key features of K-DPP program sustainability at nine years from intervention using a structured interview with K-DPP participants. This assessment focuses on the past 12 months preceding the assessment. The five main areas of assessment for program sustainability are:Participation in group/educational sessions and/or peer group activitiesContact with peer leaders and/or group membersCurrent and future interest to participate in group sessions/activitiesRecent efforts made in relation to improving lifestyle factorsFactors that would encourage or hinder participation in peer group activities
## The survey, biomedical and clinical data
Data are being entered by the research team into the RedCap database at SCTIMST servers. The database has validation checks for the values. Skip patterns are included using branching logic functions. Each team member double-checks the data s/he entered before synchronization. The project manager checks the entered data on weekly basis. Any data entry issues are rectified against the questionnaire. At $25\%$, $50\%$, $75\%$, and $100\%$ of the data entry, the project manager re-checks $10\%$ of the data for quality. After data cleaning is complete, a master copy of the dataset will be created and backed up. All data transformations (derivation of new variables) will be documented in the data dictionary. All hardcopy and electronic data will be properly secured at SCTIMST offices and servers. Data sharing will be based on the guidelines stipulated by the Data Management and Publications committee of the project.
## Retinal data
Image quality is being continuously monitored throughout the study. Initially, the day’s images were reviewed on site and feedback was provided to the photographer(s) immediately and appropriate measures have been taken to ensure that the images are captured accurately. During subsequent camps / data acquisition during the study, a few images per day are being sent for evaluation and any concerns while taking the photograph will be sent along with the images and a feedback is sent with comments and suggestions within 48 h. In case of any emergency, the contact persons can be contacted for problem resolutions.
Images are being stored on the laptop computer and backed up at the end of each day onto the external hard drive and institutional repository. Each back-up is being labelled by the date of back-up. Apart from quality control images, image transfer is be performed at the end of the study by physical transfer of an external USB hard drive. The best image for each eye will be uploaded into the software for gradeability for each view and then all measurement indices will be obtained.
The data from grading of retinal images will be entered into the RedCap database of the K-DPP Follow-up study. The Participant ID will be used to link the retinal imaging data and the rest of the data from the K-DPP Follow-up study. All data cleaning will be conducted in the same database.
## Statistical analysis
The two co-primary outcomes are diabetes incidence and CVD risk. The analyses will observe the intention-to-treat principle. A generalized estimating equation [46] model with appropriate link function and robust standard errors to account for clustering at polling areas will be fitted to estimate the relative risk (and $95\%$ CI) for the co-primary outcomes at nine years. The sensitivity of the results to assumptions on the missing outcome data underlying this model will be examined using the method of multiple imputations to handle missing data and a pattern-mixture model.
To evaluate the treatment effect on continuous outcomes, a repeated measures mixed-effects linear regression model, adjusting for the baseline measures, will be fitted. Any skewed continuous outcome variable may be transformed before fitting this model. The model will specify treatment, time-point, and treatment by time-point interaction as fixed effects. Random effects will be specified for polling areas, to account for the clustered study design, and for participants, to account for correlation between the repeated measurements on the same participant. The estimated treatment effect (e.g., the absolute difference in means between the treatment arms) at nine years and two-sided $95\%$ CI will be obtained. Binary other outcomes (e.g., diabetes) will be analyzed using GEE models similar to that of the co-primary outcomes.
The retinal image measurements will be compared between the treatment and control groups on the impact of a lifestyle modification program on pre-clinical changes in the retinal microvasculature and its correlation with cardiometabolic risk factors. Retinal imaging measures will include measures retinal architecture such as arteriolar and venular diameters and tortuosity.
The economic evaluation will consist of a cost-effectiveness analysis that will compare the incremental costs and 9-year effects (QALYs) between the study groups. Generalized linear models (GLM) with gamma family and log link components will be used to estimate the incremental costs and QALYs and the results will be presented as incremental cost-effectiveness ratios (ICERs) [47]. QALY models will be adjusted for the baseline utility values (SF-6D). Uncertainty in the cost and QALY estimates will be estimated by non-parametric bootstrapping method [48] and the results will be graphically presented as cost-effectiveness planes and cost-effectiveness acceptability curves. Several sensitivity analyses will be carried out based on different assumptions about the costs, using different discount rates ($3\%$, $5\%$ and $10\%$ per annum), and multiple imputation to assess the sensitivity of the main results due missing cost and QALY data.
Quantitative descriptive analyses will use key indicators—participation in group activities, maintenance of contact with peer leaders, interest to participate in group activities and recent effors made to improve lifestyle factors—to examine the extent of program sustainability from program participants perspectives. Statistical analyses will characterize the relationships among these indicators and the patterns of maintenance of other outcome measures over time.
## Ethics approvals
For the K-DPP Follow-up study, we obtained Ethics approval from the Institutional Review Board of SCTIMST (ID: SCT/IEC/1349/APRIL-2019) and the Alfred Human Research Ethics Committee (ID: $\frac{463}{21}$). Local approvals from HMSC and State Health Departments have been obtained before the start of data collection. A participant information sheet that describes the objectives of the study and the study procedures are being provided to participants along with the informed consent form. Participants provided written informed consent for each phase of the study.
## Discussion
This paper describes the protocol for the nine-year follow-up of a cluster randomized controlled trial of a peer-led lifestyle intervention program – the Kerala Diabetes Prevention Program—to reduce the incidence of type 2 diabetes among individuals at high risk of developing type 2 diabetes at baseline. The study will generate evidence that could establish the long-term effectiveness, cost-effectiveness, and sustainability of lifestyle intervention programs to prevent diabetes in India and other resource-constrained settings. Moreover, evidence on the progressive development of various multimorbidities over time among people with or at high risk of diabetes would inform policies and practice in the prevention and management of diabetes in India and other low-resource settings.
## Trial status
Currently following-up participants.
## References
1. 1.International Diabetes FederationIDF Diabetes Atlas202110BrusselsIDF. *IDF Diabetes Atlas* (2021.0)
2. Pradeepa R, Mohan V. **Epidemiology of type 2 diabetes in India**. *Indian J Ophthalmol* (2021.0) **69** 2932-2938. DOI: 10.4103/ijo.IJO_1627_21
3. Chan JCN, Lim L-L, Wareham NJ, Shaw JE, Orchard TJ, Zhang P. **The <em>Lancet</em> commission on diabetes: using data to transform diabetes care and patient lives**. *Lancet* (2020.0) **396** 2019-2082. DOI: 10.1016/S0140-6736(20)32374-6
4. Aziz Z, Absetz P, Oldroyd J, Pronk NP, Oldenburg B. **A systematic review of real-world diabetes prevention programs: learnings from the last 15 years**. *Implement Sci* (2015.0) **10** 172. DOI: 10.1186/s13012-015-0354-6
5. Sathish T, Williams ED, Pasricha N, Absetz P, Lorgelly P, Wolfe R. **Cluster randomised controlled trial of a peer-led lifestyle intervention program: study protocol for the Kerala diabetes prevention program**. *BMC Public Health* (2013.0) **13** 1035. DOI: 10.1186/1471-2458-13-1035
6. 6.Mathews E, Thomas E, Absetz P, D’Esposito F, Aziz Z, Balachandran S, et al. Cultural adaptation of a peer-led lifestyle intervention program for diabetes prevention in India: the Kerala diabetes prevention program (K-DPP). BMC Public Health. 2018;17(1):974.
7. Thankappan KR, Sathish T, Tapp RJ, Shaw JE, Lotfaliany M, Wolfe R. **A peer-support lifestyle intervention for preventing type 2 diabetes in India: a cluster-randomized controlled trial of the Kerala Diabetes Prevention Program**. *PLoS Med* (2018.0) **15** e1002575. DOI: 10.1371/journal.pmed.1002575
8. **The Diabetes Prevention Program: baseline characteristics of the randomized cohort. The Diabetes Prevention Program Research Group**. *Diabetes Care* (2000.0) **23** 1619-1629. DOI: 10.2337/diacare.23.11.1619
9. Uusitupa M, Louheranta A, Lindström J, Valle T, Sundvall J, Eriksson J. **The Finnish Diabetes Prevention Study**. *Br J Nutr* (2000.0) **83** S137-S142. DOI: 10.1017/S0007114500001070
10. Ramachandran A, Snehalatha C, Mary S, Mukesh B, Bhaskar AD, Vijay V. **The Indian Diabetes Prevention Programme shows that lifestyle modification and metformin prevent type 2 diabetes in Asian Indian subjects with impaired glucose tolerance (IDPP-1)**. *Diabetologia* (2006.0) **49** 289-297. DOI: 10.1007/s00125-005-0097-z
11. Lotfaliany M, Sathish T, Shaw J, Thomas E, Tapp RJ, Kapoor N. **Effects of a lifestyle intervention on cardiovascular risk among high-risk individuals for diabetes in a low- and middle-income setting: Secondary analysis of the Kerala Diabetes Prevention Program**. *Prev Med* (2020.0) **139** 106068. DOI: 10.1016/j.ypmed.2020.106068
12. Sathish T, Oldenburg B, Thankappan KR, Absetz P, Shaw JE, Tapp RJ. **Cost-effectiveness of a lifestyle intervention in high-risk individuals for diabetes in a low- and middle-income setting: trial-based analysis of the Kerala Diabetes Prevention Program**. *BMC Med* (2020.0) **18** 251. DOI: 10.1186/s12916-020-01704-9
13. Aziz Z, Mathews E, Absetz P, Sathish T, Oldroyd J, Balachandran S. **A group-based lifestyle intervention for diabetes prevention in low- and middle-income country: implementation evaluation of the Kerala Diabetes Prevention Program**. *Implement Sci* (2018.0) **13** 97. DOI: 10.1186/s13012-018-0791-0
14. Lee CG, Heckman-Stoddard B, Dabelea D, Gadde KM, Ehrmann D, Ford L. **Effect of metformin and lifestyle interventions on mortality in the Diabetes Prevention Program and Diabetes Prevention Program Outcomes Study**. *Diabetes Care* (2021.0) **44** 2775-2782. DOI: 10.2337/dc21-1046
15. Orchard TJ, Temprosa M, Barrett-Connor E, Fowler SE, Goldberg RB, Mather KJ. **Long-term effects of the Diabetes Prevention Program interventions on cardiovascular risk factors: a report from the DPP Outcomes Study**. *Diabetic Med* (2013.0) **30** 46-55. DOI: 10.1111/j.1464-5491.2012.03750.x
16. Haug N, Sorger J, Gisinger T, Gyimesi M, Kautzky-Willer A, Thurner S. **Decompression of multimorbidity along the disease trajectories of diabetes mellitus patients**. *Front Physiol* (2020.0) **11** 612604. DOI: 10.3389/fphys.2020.612604
17. Cicek M, Buckley J, Pearson-Stuttard J, Gregg EW. **Characterizing multimorbidity from type 2 diabetes: insights from clustering approaches**. *Endocrinol Metab Clin North Am* (2021.0) **50** 531-558. DOI: 10.1016/j.ecl.2021.05.012
18. Wong TY, Klein R, Klein BE, Tielsch JM, Hubbard L, Nieto FJ. **Retinal microvascular abnormalities and their relationship with hypertension, cardiovascular disease, and mortality**. *Surv Ophthalmol* (2001.0) **46** 59-80. DOI: 10.1016/S0039-6257(01)00234-X
19. Ikram MK, Cheung CY, Lorenzi M, Klein R, Jones TL, Wong TY. **Retinal vascular caliber as a biomarker for diabetes microvascular complications**. *Diabetes Care* (2013.0) **36** 750-759. DOI: 10.2337/dc12-1554
20. Saito K, Tanabe Y, Kawasaki R, Daimon M, Oizumi T, Kato T. **Is retinal vasculature change associated with risk of obesity? Longitudinal cohort study in Japanese adults: the Funagata study**. *J Diabetes Investig* (2011.0) **2** 225-232. DOI: 10.1111/j.2040-1124.2010.00086.x
21. 21.Nguyen TT WJ, Wong TY. Retinal vascular changes in pre-diabetes and prehypertension: new findings and their research and clinical implications. Diabetes Care. 2007;10:2708–15.
22. 22.Maged S Habib BA-D, Andrew Hunter and David HW Steel. The association between retinal vascular geometry changes and diabetic retinopathy and their role in prediction of progression – an exploratory study. 2014;14:89
23. Sathish T, Aziz Z, Absetz P, Thankappan KR, Tapp RJ, Balachandran S. **Participant recruitment into a community-based diabetes prevention trial in india: learnings from the Kerala Diabetes Prevention Program**. *Contemp Clin Trials Commun* (2019.0) **15** 100382. DOI: 10.1016/j.conctc.2019.100382
24. Sathish T, Shaw JE, Tapp RJ, Wolfe R, Thankappan KR, Balachandran S. **Targeted screening for prediabetes and undiagnosed diabetes in a community setting in India**. *Diabetes Metab Syndr* (2019.0) **13** 1785-1790. DOI: 10.1016/j.dsx.2019.03.042
25. Ganeshkumar P, Saigal R, Gopal B, Shankar H, Kaur P. **Provision of the continuum of care to noncommunicable diseases post-floods in Kerala, India 2018**. *Disaster Med Public Health Prep* (2022.0) **16** 825-828. DOI: 10.1017/dmp.2020.461
26. 26.Jaya AM, Harries AD, Rahman A, Khogali M, Chinnakali P, Gopalakrishnan LG, et al. Epidemiology and response to the COVID-19 pandemic in Kerala, India, 2020–2021: a cross-sectional study. Trop Med Infect Dis. 2022;7(6):105.
27. Mohan V, Deepa R, Deepa M, Somannavar S, Datta M. **A simplified Indian diabetes risk score for screening for undiagnosed diabetic subjects**. *J Assoc Physicians India* (2005.0) **53** 759-763. PMID: 16334618
28. 28.D’Agostino RB Sr, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, et al. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation. 2008;117(6):743–53.
29. Force UPST. **Risk Assessment for cardiovascular disease with nontraditional risk factors: US preventive services task force recommendation statement**. *JAMA* (2018.0) **320** 272-280. DOI: 10.1001/jama.2018.8359
30. Sathish T, Kannan S, Sarma SP, Razum O, Sauzet O, Thankappan KR. **Seven-year longitudinal change in risk factors for non-communicable diseases in rural Kerala, India: The WHO STEPS approach**. *PLoS ONE* (2017.0) **12** e0178949. DOI: 10.1371/journal.pone.0178949
31. Daivadanam M, Absetz P, Sathish T, Thankappan KR, Fisher EB, Philip NE. **Lifestyle change in Kerala, India: needs assessment and planning for a community-based diabetes prevention trial**. *BMC Public Health* (2013.0) **13** 95. DOI: 10.1186/1471-2458-13-95
32. Riley L, Guthold R, Cowan M, Savin S, Bhatti L, Armstrong T. **The World Health Organization STEPwise Approach to noncommunicable disease risk-factor surveillance: methods, challenges, and opportunities**. *Am J Public Health* (2016.0) **106** 74-78. DOI: 10.2105/AJPH.2015.302962
33. Bull FC, Maslin TS, Armstrong T. **Global physical activity questionnaire (GPAQ): nine country reliability and validity study**. *J Phys Act Health* (2009.0) **6** 790-804. DOI: 10.1123/jpah.6.6.790
34. Hu FB, Li TY, Colditz GA, Willett WC, Manson JE. **Television watching and other sedentary behaviors in relation to risk of obesity and type 2 diabetes mellitus in women**. *JAMA* (2003.0) **289** 1785-1791. DOI: 10.1001/jama.289.14.1785
35. Soman CR, Shahulhameed S, Ramankutty V, Vijayakumar K, Kunjukrishnapillai R, Ajayan K. **Cohort profile: the PROLIFE study in Kerala**. *India Int J Epidemiol* (2011.0) **40** 10-14. DOI: 10.1093/ije/dyp275
36. Booth ML, Bauman A, Owen N, Gore CJ. **Physical activity preferences, preferred sources of assistance, and perceived barriers to increased activity among physically inactive Australians**. *Prev Med* (1997.0) **26** 131-137. DOI: 10.1006/pmed.1996.9982
37. 37.Renner B, Schwarzer R. Risk and health behaviors. Documentation of the scales of the research project:“risk Appraisal consequences in Korea”(RACK). 2005;1–55.
38. Poongothai S, Pradeepa R, Ganesan A, Mohan V. **Reliability and validity of a modified PHQ-9 item inventory (PHQ-12) as a screening instrument for assessing depression in Asian Indians (CURES-65)**. *J Assoc Physicians India* (2009.0) **57** 147-152. PMID: 19582982
39. Bild DE, Bluemke DA, Burke GL, Detrano R, Diez Roux AV, Folsom AR. **Multi-Ethnic Study of Atherosclerosis: objectives and design**. *Am J Epidemiol* (2002.0) **156** 871-881. DOI: 10.1093/aje/kwf113
40. Spitzer RL, Kroenke K, Williams JB, 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
41. Ware JE, Sherbourne CD. **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
42. Mitchell PH, Powell L, Blumenthal J, Norten J, Ironson G, Pitula CR. **A short social support measure for patients recovering from myocardial infarction: the ENRICHD Social Support Inventory**. *J Cardiopulm Rehabil* (2003.0) **23** 398-403. DOI: 10.1097/00008483-200311000-00001
43. 43Chatterji S. World Health Organisation’s (WHO) Study on Global Ageing and Adult Health (SAGE). BMC Proc. 2013;7(Suppl 4 European Workshop on Health and Disability Surveilla):S1.
44. Forster RB, Garcia ES, Sluiman AJ, Grecian SM, McLachlan S, MacGillivray TJ. **Retinal venular tortuosity and fractal dimension predict incident retinopathy in adults with type 2 diabetes: the Edinburgh Type 2 Diabetes Study**. *Diabetologia* (2021.0) **64** 1103-1112. DOI: 10.1007/s00125-021-05388-5
45. Kligfield P, Gettes LS, Bailey JJ, Childers R, Deal BJ, Hancock EW. **Recommendations for the standardization and interpretation of the electrocardiogram. Part I: the electrocardiogram and its technology. A scientific statement from the American Heart Association Electrocardiography and Arrhythmias Committee, Council on Clinical Cardiology; the American College of Cardiology Foundation; and the Heart Rhythm Society**. *Heart Rhythm* (2007.0) **4** 394-412. DOI: 10.1016/j.hrthm.2007.01.027
46. Lund C, Alem A, Schneider M, Hanlon C, Ahrens J, Bandawe C. **Generating evidence to narrow the treatment gap for mental disorders in sub-Saharan Africa: rationale, overview and methods of AFFIRM**. *Epidemiol Psychiatr Sci* (2015.0) **24** 233-240. DOI: 10.1017/S2045796015000281
47. Ng ES, Diaz-Ordaz K, Grieve R, Nixon RM, Thompson SG, Carpenter JR. **Multilevel models for cost-effectiveness analyses that use cluster randomised trial data: an approach to model choice**. *Stat Methods Med Res* (2016.0) **25** 2036-2052. DOI: 10.1177/0962280213511719
48. Nixon RM, Wonderling D, Grieve RD. **Non-parametric methods for cost-effectiveness analysis: the central limit theorem and the bootstrap compared**. *Health Econ* (2010.0) **19** 316-333. DOI: 10.1002/hec.1477
|
---
title: 'Mortality patterns among COVID‐19 patients in two Saudi hospitals: Demographics,
etiology, and treatment'
authors:
- Fatimah S. AlGhawi
- Sami S. AlMudarra
- Abdullah M. Assiri
journal: Influenza and Other Respiratory Viruses
year: 2023
pmcid: PMC10030359
doi: 10.1111/irv.13127
license: CC BY 4.0
---
# Mortality patterns among COVID‐19 patients in two Saudi hospitals: Demographics, etiology, and treatment
## Abstract
### Background
Saudi Arabia (SA) reported its first case of COVID‐19 on 2 March 2020. Mortality varied nationwide; by April 14, 2020, Medina had $16\%$ of SA's total COVID‐19 cases and $40\%$ of all COVID‐19 deaths. A team of epidemiologists investigated to identify factors impacting survival.
### Methods
We reviewed medical records from two hospitals: Hospital A in Medina and Hospital B in Dammam. All patients with a registered COVID‐related death between March and May 1, 2020, were included. We collected data on demographics, chronic health conditions, clinical presentation, and treatment. We analyzed data using SPSS.
### Results
We identified 76 cases: 38 cases from each hospital. More fatalities were among non‐Saudis at Hospital A ($89\%$) versus Hospital B ($82\%$, $p \leq 0.001$). Hypertension prevalence was higher among cases at Hospital B ($42\%$) versus Hospital A ($21\%$) ($p \leq 0.05$). We found statistically significant differences ($p \leq 0.05$) in symptoms at initial presentation among cases at Hospital B versus Hospital A, including body temperature (38°C vs. 37°C), heart rate (104 bpm vs. 89 bpm), and regular breathing rhythms ($61\%$ vs. $55\%$). Fewer cases ($50\%$) at Hospital A received heparin versus Hospital B ($97\%$, p‐value < 0.001).
### Conclusion
Patients who died typically presented with more severe illnesses and were more likely to have underlying health conditions. Migrant workers may be at increased risk due to poorer baseline health and reluctance to seek care. This highlights the importance of cross‐cultural outreach to prevent deaths. Health education efforts should be multilingual and accommodate all literacy levels.
## INTRODUCTION
COVID‐19, a global pandemic that first appeared in China in December 2019, has quickly spread across the globe since the first case was reported. 1 KSA had its first case on March 2, 2020; by March 23, 2022, case had been reported nationwide. On April 14, the Saudi Field Epidemiology Training Program (FETP) investigated a cluster of COVID‐19 deaths in Medina. The majority of deaths occurred in Hospital A, which was Medina's designated COVID‐19 hospital. At that time, Medina had $16\%$ of KSA's total COVID‐19 cases and $40\%$ of all COVID‐19 deaths. Subsequently, Hospital B in Dammam was chosen for comparison as it was Dammam's designated COVID‐19 hospital.
Clark et al. 2 developed a prediction model that estimated a potential occurrence of 1.0 to 2.4 billion severe COVID‐19 infections among people with severe clinical conditions such as cardiovascular disease, chronic kidney disease, chronic respiratory disease, and diabetes.
The severity of the disease varies significantly, ranging from asymptomatic infection to the development of severe complications and death. 3 Age, gender, and the presence of co‐morbidities have been reported to be contributing factors to COVID‐19 severity. 4, 5, 6, 7 Patients with diabetes or chronic obstructive pulmonary disease (COPD) are more likely to have longer hospitalizations, be admitted to intensive care units (ICU), and require mechanical ventilation. 5, 8 Conversely, mild prognosis has been reported among pediatric cases. However, the severity of the infection has also been reported among children affecting up to $5\%$ of the infected patients although compared with adults, children and/or adolescents tend to have a mild COVID‐19 course with a good prognosis. 9, 10, 11 Evidence on the pathology behind the development of the disease has also been variable. At first, it was thought that the virus affects the respiratory tract only; however, reports showed that it could affect many organs, including the blood, heart, brain, kidneys, pancreas, and eyes. 12, 13 Moreover, the severity of the infection has also been related to several laboratory variables. Prothrombin time, C‐reactive protein, D‐dimer, procalcitonin, and fibrinogen levels have reportedly been associated with the deterioration of the disease. 14, 15, 16, 17, 18 Some of these biomarkers have helped in building prediction models to decrease mortality among critically‐ill COVID‐19 patients. 19, 20 Other investigations have reported an association between patients' blood type and the prognosis of the infection. 21, 22 This indicates the fact that mortality because of COVID‐19 is different due to the different epidemiology among the affected populations. 23 For that, we aim to assess and compare the different factors related to COVID‐19 mortality, including patients' demographics, clinical characteristics, and treatment regimens used among patients of two Saudi hospitals.
## Study design
We conducted a retrospective study by reviewing the patients' records at Hospital A in Medina and Hospital B in Dammam. The study was conducted between March 2020 to May 1, 2020.
## Subject selection and procedure
All confirmed COVID‐19 patients with a registered COVID‐related death were included in the current study. No restrictions were made regarding age, gender, nationality, or the admitting department. We excluded deaths not certified as a direct result of COVID‐19 or patients with suspected COVID‐19 status.
Patients' demographics, documented deaths, cause of death, treatments provided, admission details, and hospital stay details were all collected. We reviewed records for cases between March 2020 and May 1, 2020.
## Data analysis
Data entry and analyses were conducted using SPSS v.26 (IBM, NY). Nominal variables were presented as frequencies (n) and percentages (%). The Chi2 test (or Fisher's exact test, as appropriate) was used for identifying differences between hospitals. The continuous variables were presented as means and standard deviations (SDs). We used a t‐test or Mann–Whitney test based on the distribution of the data (normally distributed or not).
A binary logistic regression model was constructed to control any potential confounders and determine the significantly associated factors with the mortality outcome. The odds ratio and $95\%$ confidence interval ([$95\%$ CI]) are presented. Statistical significance was set at a P‐value < 0.05 for all analyses.
## Informed consent and ethical considerations
No identifying information on any patient was collected, and all collected data were exclusively used for statistical analysis. All data were kept confidential. Before commencement, the study protocol was cleared by the institutional review board and the ethics committee at King Fahad Medical City, Riyadh, Saudi Arabia.
## Baseline characteristics
We identified 76 patients that met the inclusion criteria; 38 were patients at Hospital A, and 38 were patients at Hospital B. The mean age of the included patients was 51.7 years, and $93\%$ of the patients were male. We found no statistically significant differences among the included patients between both hospitals in terms of age (P‐value = 0.322), gender (P‐value = 0.500) or the percentage of overweight (P‐value = 0.911). In contrast, we found a statically significant difference in the nationality distribution among the aforementioned two hospitals (P‐value < 0.001) (Table 1).
**TABLE 1**
| Variable | Variable.1 | Hospital B | Hospital B.1 | Hospital A | Hospital A.1 | Total | Total.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variable | Variable | n | % | n | % | N | % | P‐value |
| Age; Mean ± SD | Age; Mean ± SD | 53.1 ± 11.2 | 53.1 ± 11.2 | 50.3 ± 13.3 | 50.3 ± 13.3 | 51.7 ± 12.3 | 51.7 ± 12.3 | 0.322 |
| Gender | Female | 3 | 7.9 | 2 | 5.3 | 5 | 6.6 | 0.500 |
| | Male | 35 | 92.1 | 36 | 94.7 | 71 | 93.4 | |
| BMI > 25 kg/m2 | No | 16 | 42.1 | 7 | 43.8 | 23 | 42.6 | 0.911 |
| | Yes | 22 | 57.9 | 9 | 56.3 | 31 | 57.4 | |
| Healthcare worker | No | 37 | 97.4 | 36 | 94.7 | 73 | 96.1 | 0.556 |
| | Yes | 1 | 2.6 | 2 | 5.3 | 3 | 3.9 | |
| Nationality | Saudi | 7 | 18.4 | 4 | 10.5 | 11 | 14.5 | <0.001* |
| | Non‐Saudi | 31 | 81.6 | 31 | 89.5 | 65 | 85.5 | |
| Smoker | Yes | 3 | 7.9 | 6 | 15.8 | 9 | 11.8 | 0.578 |
| | No | 21 | 55.3 | 21 | 55.3 | 42 | 55.3 | |
| | Unknown | 14 | 36.8 | 11 | 28.9 | 25 | 32.9 | |
| Comorbidities | No | 18 | 47.4 | 22 | 57.9 | 40 | 52.6 | 0.358 |
| | Yes | 20 | 52.6 | 16 | 42.1 | 36 | 47.4 | |
| Diabetes mellitus | Yes | 13 | 34.2 | 14 | 36.8 | 27 | 35.5 | 0.811 |
| | No | 25 | 65.7 | 24 | 63.1 | 49 | 64.4 | |
| Hypertension | Yes | 16 | 42.1 | 8 | 21.1 | 24 | 31.6 | 0.048* |
| | No | 22 | 57.8 | 30 | 78.9 | 52 | 68.4 | |
| Ischemic deart disease | Yes | 6 | 15.8 | 2 | 5.3 | 8 | 10.5 | 0.135 |
| | No | 32 | 84.2 | 36 | 94.7 | 68 | 89.4 | |
Only $12\%$ of the patients reported smoking; $47\%$ of the patients had one or more documented comorbidities. The most prevalent comorbidity was diabetes mellitus (DM), being present in $36\%$ of the patients, followed by hypertension ($32\%$) and ischemic heart disease (IHD) in $11\%$. We also found a statistically significant difference in the prevalence rates of hypertension between patients between the two hospitals (P‐value = 0.048) (Table 1).
## Patients' admission details and baseline clinical data
The mean admission body temperature of all patients was 37.7°C, whereas their mean initial respiratory rate was 26.6 breaths per minute, and the mean initial heart rate was 98.6 ± 19.3 beats per minute. The mean visual triage score was 6.7 ± 1.9, and the mean Glasgow Coma Scale was 10.2 ± 5.6. For Hospital B, the mean visual triage score was 6.3 ± 2.5, while the mean Glasgow Coma Scale was 10.8 ± 5.4. For those patients with available data, $76\%$ of the patients were registered as an emergency, $17\%$ were directly admitted to the ICU, and $7\%$ were registered from the outpatient department. In the same context, the admission source was variable among the included patients; $30.3\%$ of the patients were admitted from the emergency room, $26\%$ were referred from another hospital, $22\%$ were admitted from the clinic, and $21.1\%$ were in the hospital ward. Most of the patients ($76\%$) did not sign a “Do not resuscitate” form, while only $24\%$ did sign it. Nevertheless, there was a statistically significant difference between Hospital B and Hospital A in terms of initial body temperature (P‐value = 0.001), initial heart rate (P‐value = 0.002), and registry type (P‐value< 0001) (Tables 2 and 3).
On admission, $79\%$ of the included patients presented with fever, $76\%$ with shortness of breath, $10\%$ with a sore throat, and $75\%$ with a cough. Regarding the cough type at presentation, $16\%$ of the included patients presented with productive cough, and $15\%$ presented with non‐productive cough, whereas the remaining portion either did not have a cough or did not have a documented cough type. We did not find any statistically significant differences regarding fever (P‐value = 1.000), shortness of breath (P‐value = 1.000), sore throat (P‐value = 0.711), cough (P‐value = 0.791), or type of cough among the included patients (P‐value = 0.250) among the included patients (Table 4).
**TABLE 4**
| Variables | Variables.1 | Hospital B | Hospital B.1 | Hospital A | Hospital A.1 | Total | Total.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | Variables | n | % | n | % | N | % | P‐value |
| Fever | No | 8 | 21.1 | 8 | 21.1 | 16 | 21.1 | 1.000 |
| | Yes | 30 | 78.9 | 30 | 78.9 | 60 | 78.9 | |
| Shortness of breath | No | 9 | 23.7 | 9 | 23.7 | 18 | 23.7 | 1.000 |
| | Yes | 29 | 76.3 | 29 | 76.3 | 58 | 76.3 | |
| Sore throat | No | 35 | 92.1 | 33 | 86.8 | 68 | 89.5 | 0.711 |
| | Yes | 3 | 7.9 | 5 | 13.2 | 8 | 10.5 | |
| Cough | No | 10 | 26.3 | 9 | 23.7 | 19 | 25.0 | 0.791 |
| | Yes | 28 | 73.7 | 29 | 76.3 | 57 | 75.0 | |
| Cough type | Non‐productive | 8 | 21.1 | 3 | 7.9 | 11 | 14.5 | 0.250 |
| | Productive | 5 | 13.2 | 7 | 18.4 | 12 | 15.8 | |
| | None/NA | 25 | 65.8 | 28 | 73.7 | 53 | 69.7 | |
| Breathing rhythm | Regular | 23 | 60.5 | 21 | 55.3 | 44 | 57.9 | 0.017* |
| | Irregular | 1 | 2.6 | 9 | 23.7 | 10 | 13.2 | |
| | Not available | 14 | 36.8 | 8 | 21.1 | 22 | 28.9 | |
| Breathing quality | Normal | 23 | 60.5 | 21 | 55.5 | 44 | 57.9 | 0.029* |
| | Labored | 1 | 2.6 | 7 | 18.4 | 8 | 10.5 | |
| | Not available | 14 | 36.8 | 10 | 26.3 | 24 | 31.1 | |
## Comparison of patients' findings and examination results
The mean O2 saturation at the admission was 83.9 ± 9.2, with a mean of 85.4 ± 8.5 and 82.5 ± 9.8 at Hospital B and Hospital A, respectively. Regarding the radiological findings, bilateral infiltrates were present in $29\%$ of the included patients at admission, with a prevalence of $26\%$ at Hospital B and $32\%$ in Hospital A (Table 5).
**TABLE 5**
| Variables | Variables.1 | Hospital B | Hospital B.1 | Hospital A | Hospital A.1 | Total | Total.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | Variables | N | % | N | % | N | % | P‐value |
| Initial O2 saturation (%); Mean ± SD | Initial O2 saturation (%); Mean ± SD | 85.4 ± 8.5 | 85.4 ± 8.5 | 82.5 ± 9.8 | 82.5 ± 9.8 | 83.9 ± 9.2 | 83.9 ± 9.2 | 0.173 |
| Bilateral infiltrates | No | 28 | 73.7 | 26 | 68.4 | 54 | 71.1 | 0.613 |
| | Yes | 10 | 26.3 | 12 | 31.6 | 22 | 28.9 | |
The local chest examination findings were variable among patients. The examination is usually carried out and documented by the attending physician according to the standard definitions during the initial assessment. Breathing rhythm was regular in $58\%$ of the patients, irregular in $13\%$ of the patients, and the data about the remaining $29\%$ were not documented. For the breathing depth, $53\%$ of the patients showed a normal depth of respiration, $10\%$ showed shallow breathing, $7\%$ showed deep breathing, and $30\%$ did have a documented breathing depth. For the breathing quality, $57.9\%$ of the patients showed a normal breathing quality, $10\%$ showed labored breathing, and the remaining $32\%$ did have a documented breathing depth. In the same context, $65\%$ of the included patients did not have any added sounds; however, there was a high variability of the breathing added sounds among the remaining ones. Bilateral crepitations were found in $17\%$ of the patients. Eight percent had wheezes, $7\%$ had bilateral crackles, $3\%$ had rhonchi, and $1\%$ had scattered crepitations. There was a statistically significant difference between the two hospitals in the patterns of breathing rhythm (P‐value = 0.017), breathing quality (P‐value = 0.029), and added sounds (P‐value = 0.029) among the included patients (Tables 4 and 5).
## Comparison of interventions/treatments used for patients in both hospitals
The majority of the patients ($93\%$) were admitted to the ICU at some point, and most of the patients ($89\%$) required ventilation during their treatment course. There was no statistically significant difference between hospitals in the ICU admission (P‐value = 0.644) or ventilation rates (P‐value = 0.262) among the included patients. Regarding drugs administered, about two‐thirds ($64\%$) of the patients were treated with hydroxychloroquine, and most patients ($74\%$) were treated with heparin. There was a statistically significant difference between the two hospitals in heparin usage rates (P‐value < 0.001), whereas hydroxychloroquine usage rates were comparable (P‐value = 0.811) (Table 6).
**TABLE 6**
| Variables | Variables.1 | Hospital B | Hospital B.1 | Hospital A | Hospital A.1 | Total | Total.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | Variables | n | % | n | % | N | % | P‐value |
| ICU admission | No | 3 | 7.9 | 2 | 5.3 | 5 | 6.6 | 0.644 |
| | Yes | 35 | 92.1 | 36 | 94.7 | 71 | 93.4 | |
| Ventilated | No | 2 | 5.3 | 6 | 15.8 | 8 | 10.5 | 0.262 |
| | Yes | 36 | 94.7 | 32 | 84.2 | 68 | 89.5 | |
| Hydroxychloroquine use | No | 14 | 36.8 | 13 | 34.2 | 27 | 35.5 | 0.811 |
| | Yes | 24 | 63.2 | 25 | 65.8 | 49 | 64.5 | |
| Heparin use | No | 1 | 2.6 | 19 | 50.0 | 20 | 26.3 | <0.001* |
| | Yes | 37 | 97.4 | 19 | 50.0 | 56 | 73.7 | |
| Continuous renal replacement therapy | No | 34 | 89.5 | 32 | 84.2 | 66 | 86.8 | 0.736 |
| | Yes | 4 | 10.5 | 6 | 15.8 | 10 | 13.2 | |
## Comparison of patients' outcomes
The mean hospital time of all included patients was 6.4 ± 4.5 days, with a mean of 7.1 ± 4.3 days and 5.6 ± 4.7 days in Hospital B and Hospital A, respectively. For the time span from admission to ICU admission, the meantime in days was 0.8 ± 1.4 days, with a mean of 0.9 ± 1.2 days and 0.8 ± 1.6 days in Hospital B and Hospital A, respectively. For the time span from admission to ventilation, the meantime in days was 1.6 ± 2.3 days, with a mean of 1.6 ± 2.3 days and 1.5 ± 2.2 days in Hospital B and Hospital A, respectively. For the time span from ICU admission to death, the meantime in days was 5.6 ± 4.3 days, with a mean of 6.2 ± 4.5 days and 5.1 ± 4.1 days in Hospital B and Hospital A, respectively. For the time span from ventilation to death, the meantime in days was 5.5 ± 4.3 days, with a mean of 6.0 ± 4.3 days and 4.9 ± 4.3 days in Hospital B and Hospital A, respectively. There were no statistically significant differences between the two hospitals in all recorded outcomes (Table 7).
**TABLE 7**
| Variables | Hospital B | Hospital B.1 | Hospital A | Hospital A.1 | Total | Total.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | Mean | SD | Mean | SD | Mean | SD | P‐value |
| Hospital time (Days) | 7.1 | 4.3 | 5.6 | 4.7 | 6.4 | 4.5 | 0.158 |
| Time to ICU (Days) | 0.9 | 1.2 | 0.8 | 1.6 | 0.8 | 1.4 | 0.753 |
| Time to vent (Days) | 1.6 | 2.3 | 1.5 | 2.2 | 1.6 | 2.3 | 0.945 |
| ICU to death time (Days) | 6.2 | 4.5 | 5.1 | 4.1 | 5.6 | 4.3 | 0.263 |
| Ventilation to death (Days) | 6.0 | 4.3 | 4.9 | 4.3 | 5.5 | 4.3 | 0.310 |
## Effect of hospital choice on patients' outcomes
Logistic regression was performed to test whether the hospital choice has an effect on the patients' outcomes or not. This was tested using Hospital A as a reference and doing the test for choosing Hospital B. Accordingly, there was a reduction in all outcomes in Hospital B when compared with Hospital A. Length of hospitalization (OR = 0.93; $95\%$ CI = 0.83–1.03), time from admission to ICU (OR = 0.95; $95\%$ CI = 0.68–1.32), time from admission to ventilation (OR = 0.99; $95\%$ CI = 0.79–1.25), time from ICU admission to death (OR = 0.94; $95\%$ CI = 0.84–1.05), and time from ventilation to death (OR = 0.94; $95\%$ CI = 0.84–1.06). These reductions were not statistically significant (Table 8).
**TABLE 8**
| Predictor | Estimate | SE | Odds ratio | 95% confidence interval | 95% confidence interval.1 | P‐value |
| --- | --- | --- | --- | --- | --- | --- |
| Predictor | Estimate | SE | Odds ratio | Lower | Upper | P‐value |
| Hospital time (Days) | −0.08 | 0.05 | 0.93 | 0.83 | 1.03 | 0.161 |
| Time to ICU (Days) | −0.05 | 0.17 | 0.95 | 0.68 | 1.32 | 0.749 |
| Time to vent (Days) | −0.01 | 0.12 | 0.99 | 0.79 | 1.25 | 0.944 |
| ICU to death time (Days) | −0.07 | 0.06 | 0.94 | 0.84 | 1.05 | 0.263 |
| Ventilation to death (Days) | −0.06 | 0.06 | 0.94 | 0.84 | 1.06 | 0.308 |
## DISCUSSION
In our study, we compared a hospital in the Saudi Western Province (Medina) to a hospital in the Saudi Eastern Province (Dammam). The COVID‐related mortalities during the observed duration were similar between the two hospitals/provinces. This is consistent with the Saudi official records where the total cases in the Eastern Province were 82,072, with overall deaths of 557 (mortality rate of $0.68\%$). 24 In the same context, the total cases in Medina were 23,272, with overall deaths of 132 (mortality rate of $0.57\%$). 24 Overall, Saudi Arabia's case fatality rate is also among the lowest fatality rates in the world, ranging from $0\%$ to $28.9\%$. 25 According to our results, it is consistent in different regions of Saudi Arabia, which supports that the quality of healthcare is relatively homogenous and of adequate quality.
Our results showed a relatively consistent presentation of clinical symptoms/signs among the included patients in comparing the two hospitals. Nevertheless, there were some differences in the initial presentation, including the initial body temperature, initial heart rate, breathing rhythms, breathing quality, and added sounds. Many of the previously reported MERS‐CoV 26 and SARS‐CoV 27 patients also showed similar comorbidities, which predisposed to increasing the risk of infection with MERS‐CoV and increasing the case fatality rates. 28 Regarding clinical presentation, the predominant presentations among COVID‐19 patients were low‐grade high fever (mean temperature 37.7) and cough, which seems to be consistent with the initial reports from different countries. 3, 12, 29, 30, 31 According to our results, treatments used were homogenous among the two hospitals, except for heparin use. Others have reported that many COVID‐19 patients suffer from a hypercoagulability state. 32, 33 To our knowledge, this is the first study to compare COVID‐patients in two Saudi hospitals in two different provinces. However, the study has some limitations. The relatively small number of included patients may affect the magnitude of differences and the statistical significance. Moreover, some patients' data were missing, which may also affect our results.
## CONCLUSION
Throughout the COVID‐19 outbreak in Saudi Arabia, the Kingdom has maintained a robust healthcare system and minimized case fatalities. we found a relatively consistent presentation of clinical symptoms/signs among the included patients in comparing the two hospitals. The majority of COVID‐19 deaths occur mainly among men, which is consistent with global reports of COVID‐19 fatalities and Saudi Arabia's COVID‐19 case distribution. Patients who died typically presented with more severe illnesses and were more likely to have underlying health conditions. Migrant workers may be at increased risk due to poorer baseline health and reluctance to seek care. This highlights the importance of cross‐cultural outreach to prevent deaths. Health education efforts should be multilingual and accommodate all literacy levels. Following further validation, heparin should be considered on a wider scale in the Western region (Medina) since we noticed a major gap in usage.
## AUTHOR CONTRIBUTIONS
Fatimah Saeed AlGhawi: Conceptualization; formal analysis; investigation; methodology; project administration; validation; writing—original draft; writing—review and editing. Abdullah M. Assiri: Conceptualization; formal analysis; investigation; methodology; project administration; supervision; validation; writing—review and editing. Sami AlMudarra: Conceptualization; investigation; methodology; project administration; supervision; validation; writing—review and editing.
## CONFLICT OF INTEREST STATEMENT
The authors report no conflict of interest.
## ETHICS STATEMENT
The study was approved by the institutional review boards at King Fahad Medical City, Riyadh, Saudi Arabia.
## PEER REVIEW
The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/irv.13127.
## DATA AVAILABILITY STATEMENT
The additional data that support the findings of this study are available from the corresponding author upon reasonable request.
## References
1. Riou J, Althaus CL. **Pattern of early human‐to‐human transmission of Wuhan 2019 novel coronavirus (2019‐Ncov), December 2019 to January 2020**. *Euro Surveill* (2020.0) **25**. DOI: 10.2807/1560-7917.es.2020.25.4.2000058
2. Clark A, Jit M, Warren‐Gash C. **Global, regional, and national estimates of the population at increased risk of severe Covid‐19 due to underlying health conditions in 2020: a modelling study**. *Lancet Glob Health* (2020.0) **8** e1003-e1017. DOI: 10.1016/s2214-109x(20)30264-3
3. Chen N, Zhou M, Dong X. **Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study**. *Lancet* (2020.0) **395** 507-513. DOI: 10.1016/s0140-6736(20)30211-7
4. Bhopal SS, Bhopal R. **Sex differential in Covid‐19 mortality varies markedly by age**. *Lancet* (2020.0) **396** 532-533. DOI: 10.1016/S0140-6736(20)31748-7
5. Singh AK, Gupta R, Ghosh A, Misra A. **Diabetes in Covid‐19: prevalence, pathophysiology, prognosis and practical considerations**. *Diabetes Metab Syndr Clin Res Rev* (2020.0) **14** 303-310. DOI: 10.1016/j.dsx.2020.04.004
6. Liu K, Chen Y, Lin R, Han K. **Clinical features of Covid‐19 in elderly patients: a comparison with young and middle‐aged patients**. *J Infect* (2020.0) **80** e14-e18. DOI: 10.1016/j.jinf.2020.03.005
7. Zhou F, Yu T, du R. **Clinical course and risk factors for mortality of adult inpatients with Covid‐19 in Wuhan, China: a retrospective cohort study**. *Lancet* (2020.0) **395** 1054-1062. DOI: 10.1016/s0140-6736(20)30566-3
8. Zhao Q, Meng M, Kumar R. **The impact of Copd and smoking history on the severity of Covid‐19: a systemic review and meta‐analysis**. *J Med Virol* (2020.0) **92** 1915-1921. DOI: 10.1002/jmv.25889
9. Mantovani A, Rinaldi E, Zusi C, Beatrice G, Saccomani MD, Dalbeni A. **Coronavirus disease 2019 (Covid‐19) in children and/or adolescents: a meta‐analysis**. *Pediatr Res* (2020.0). DOI: 10.1038/s41390-020-1015-2
10. Dong Y, Mo X, Hu Y. **Epidemiology of Covid‐19 among children in China**. *Pediatrics* (2020.0) **145**. DOI: 10.1542/peds.2020-0702
11. Meena J, Yadav J, Saini L, Yadav A, Kumar J. **Clinical features and outcome of Sars‐Cov‐2 infection in children: a systematic review and meta‐analysis**. *Indian Pediatr* (2020.0) **57** 820-826. DOI: 10.1007/s13312-020-1961-0
12. Scalinci SZ, Battagliola ET. **Conjunctivitis can be the only presenting sign and symptom of Covid‐19**. *IDCases* (2020.0) **20**. DOI: 10.1016/j.idcr.2020.e00774
13. Gupta A, Madhavan MV, Sehgal K. **Extrapulmonary manifestations of Covid‐19**. *Nat Med* (2020.0) **26** 1017-1032. DOI: 10.1038/s41591-020-0968-3
14. Vidali S, Morosetti D, Cossu E. **D‐dimer as an indicator of prognosis in Sars‐Cov‐2 infection: a systematic review**. *ERJ Open Res* (2020.0) **6**. DOI: 10.1183/23120541.00260-2020
15. Helms J, Tacquard C, Severac F. **High risk of thrombosis in patients with severe Sars‐Cov‐2 infection: a multicenter prospective cohort study**. *Intensive Care Med* (2020.0) **46** 1089-1098. DOI: 10.1007/s00134-020-06062-x
16. Qiu P, Zhou Y, Wang F. **Clinical characteristics, laboratory outcome characteristics, comorbidities, and complications of related Covid‐19 deceased: a systematic review and meta‐analysis**. *Aging Clin Exp Res* (2020.0) **32** 1-10. DOI: 10.1007/s40520-020-01664-3
17. Huang I, Pranata R, Lim MA, Oehadian A, Alisjahbana B. **C‐reactive protein, procalcitonin, D‐dimer, and ferritin in severe coronavirus disease‐2019: a meta‐analysis**. *Ther Adv Respir Dis* (2020.0) **14** 1-4. DOI: 10.1177/1753466620937175
18. Shah S, Shah K, Patel SB. **Elevated D‐dimer levels are associated with increased risk of mortality in Covid‐19: a systematic review and meta‐analysis**. *Cardiol Rev* (2020.0) **28** 295-302. DOI: 10.1097/CRD.0000000000000330
19. Wynants L, Van Calster B, Collins GS. **Prediction models for diagnosis and prognosis of Covid‐19: systematic review and critical appraisal**. *BMJ* (2020.0) **369**. DOI: 10.1136/bmj.m1328
20. Wang F, Hou H, Wang T. **Establishing a model for predicting the outcome of Covid‐19 based on combination of laboratory tests**. *Travel Med Infect Dis* (2020.0) **36**. DOI: 10.1016/j.tmaid.2020.101782
21. Latz CA, DeCarlo C, Boitano L. **Blood type and outcomes in patients with Covid‐19**. *Ann Hematol* (2020.0) **99** 2113-2118. DOI: 10.1007/s00277-020-04169-1
22. Wu Y, Feng Z, Li P, Yu Q. **Relationship between abo blood group distribution and clinical characteristics in patients with Covid‐19**. *Clin Chim Acta* (2020.0) **509** 220-223. DOI: 10.1016/j.cca.2020.06.026
23. Lipsitch M, Swerdlow DL, Finelli L. **Defining the epidemiology of Covid‐19—studies needed**. *N Engl J Med* (2020.0) **382** 1194-1196. DOI: 10.1056/NEJMp2002125
24. **Covid 19 Dashboard: Saudi Arabia**
25. **Mortality Analyses**
26. Memish ZA, Al‐Tawfiq JA, Alhakeem RF. **Middle East Respiratory Syndrome Coronavirus (Mers‐Cov): a cluster analysis with implications for global management of suspected cases**
27. Moni MA, Liò P. **Network‐based analysis of comorbidities risk during an infection: SARS and HIV case studies**. *BMC Bioinf* (2014.0) **15**. DOI: 10.1186/1471-2105-15-333
28. Assiri A, Al‐Tawfiq JA, Al‐Rabeeah AA. **Epidemiological, demographic, and clinical characteristics of 47 cases of Middle East Respiratory Syndrome Coronavirus Disease from Saudi Arabia: a descriptive study**
29. Wang D, Hu B, Hu C. **Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus‐infected pneumonia in Wuhan, China**. *Jama* (2020.0) **323** 1061-1069. DOI: 10.1001/jama.2020.1585
30. Huang C, Wang Y, Li X. **Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China**. *Lancet* (2020.0) **395** 497-506. DOI: 10.1016/S0140-6736(20)30183-5
31. Xu X‐W, Wu XX, Jiang XG. **Clinical findings in a group of patients infected with the 2019 novel coronavirus (Sars‐Cov‐2) outside of Wuhan, China: retrospective case series**. *BMJ* **368**. DOI: 10.1136/bmj.m606
32. Barrett CD, Moore HB, Yaffe MB, Moore EE. **Isth interim guidance on recognition and management of coagulopathy in Covid‐19: a comment**. *J Thromb Haemost* (2020.0) **18** 2060-2063. DOI: 10.1111/jth.14860
33. Barth H, Schnober EK, Zhang F. **Viral and cellular determinants of the hepatitis C virus envelope‐heparan sulfate interaction**. *J Virol* (2006.0) **80** 10579-10590. DOI: 10.1128/jvi.00941-06
|
---
title: Clusters of lifestyle behaviours and their associations with socio-demographic
characteristics in Dutch toddlers
authors:
- Anne Krijger
- Elly Steenbergen
- Lieke Schiphof-Godart
- Caroline van Rossum
- Janneke Verkaik-Kloosterman
- Liset Elstgeest
- Sovianne ter Borg
- Hein Raat
- Koen Joosten
journal: European Journal of Nutrition
year: 2022
pmcid: PMC10030397
doi: 10.1007/s00394-022-03056-x
license: CC BY 4.0
---
# Clusters of lifestyle behaviours and their associations with socio-demographic characteristics in Dutch toddlers
## Abstract
### Purpose
This study aimed to identify clusters of lifestyle behaviours in toddlers and assess associations with socio-demographic characteristics.
### Methods
We used data from the Dutch National Food Consumption Survey 2012–2016 and included 646 children aged 1–3 years. Based on 24-h dietary recalls and a questionnaire, a two-step cluster analysis was conducted to identify clusters in the intake of fruit, vegetables, sugar-sweetened beverages and unhealthy snacks, physical activity and screen time. Logistic regression models assessed associations between socio-demographic characteristics and cluster allocation.
### Results
Three clusters emerged from the data. The ‘relatively healthy cluster’ demonstrated a high intake of fruit and vegetables, low sugar-sweetened beverage and unhealthy snack intake and low screen time. The ‘active snacking cluster’ was characterised by high unhealthy snack intake and high physical activity, and the ‘sedentary sweet beverage cluster’ by high intake of sugar-sweetened beverages and high screen time. Children aged 1 year were most likely to be allocated to the ‘relatively healthy cluster’. Compared to children of parents with a high education level, children of parents with a low or middle education level were less likely to be in the ‘relatively healthy cluster’, but more likely to be in the ‘sedentary sweet beverage cluster’.
### Conclusion
Clusters of lifestyle behaviours can be distinguished already in children aged 1–3 years. To promote healthy lifestyle behaviour, efforts may focus on maintaining healthy behaviour in 1-year-olds and more on switching towards healthy behaviour in 2- and 3-year-olds.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s00394-022-03056-x.
## Introduction
Overweight and obesity can occur as early as toddlerhood. Globally, $5.7\%$ of children under 5 years were overweight or obese in 2020 [1]. This is a major public health concern as childhood obesity increases the risk of other (chronic) diseases, affecting both physical and mental health [2]. Moreover, childhood obesity often tracks into adulthood [3]. The main underlying cause of overweight and obesity lies in lifestyle behaviour, which may be established at a young age and likely persists as the child ages [2, 4, 5]. Unfavourable lifestyle behaviours, such as the intake of energy-dense, nutrient-poor foods, including sugar-sweetened beverages and snacks, as well as high levels of sedentary behaviour, are positively associated with obesity [6, 7]. Contrarily, diets characterised by high amounts of fruits and vegetables, and regular physical activity are associated with lower obesity risk [8, 9].
Many children do not meet the daily recommendations for dietary intake, physical activity and sedentary behaviour [10, 11]. However, children’s lifestyles can comprise both healthy and unhealthy behaviours simultaneously. Characterising lifestyle behaviour patterns in children can support the understanding of interrelationships (i.e. co-occurrence and interaction) between multiple lifestyle behaviours. Ultimately, this can contribute to developing guidelines and interventions that simultaneously address multiple unfavourable lifestyle behaviours in children.
Exploratory, data-driven techniques, such as cluster analysis and principal component analysis, can be used to gain insight into behaviour patterns [12]. Reviews of studies applying these methods to identify lifestyle behaviour clusters in children found that diet, physical activity and sedentary behaviour cluster in complex ways [13, 14]. In addition to clusters entirely characterised by healthy or unhealthy diets, physical activity and sedentary behaviours, clusters with a mixture of healthy and unhealthy behaviours have been commonly distinguished. To reach children most at risk of adverse health effects, it is essential to identify shared determinants of lifestyle behaviour clusters. As to determinants of lifestyle behavioural patterns in children, it has been shown that age, sex and socio-economic status (SES) are associated with lifestyle behaviour patterns [13, 14]. Lower SES, mostly indicated by parental education level, was found to be associated with unhealthier lifestyle patterns [13–15]. How other socio-demographic factors are associated with lifestyle behaviour patterns in children remains unclear.
To our knowledge, most studies on the clustering of lifestyle behaviours in children have been conducted in older children (≥ 5 years). Nevertheless, lifestyle habits develop early in life, and early identification of patterns and associated socio-demographic determinants might help to initiate timely interventions for modifying lifestyle behaviours when needed. Therefore, our study aims to identify clusters of co-occurring lifestyle behaviours, including intake of fruit, vegetables, sugar-sweetened beverages and unhealthy snacks, physical activity and screen time, and analyse their associations with socio-demographic characteristics in children aged 1–3 years who participated in the Dutch National Food Consumption Survey (DNFCS) 2012–2016.
## Study population and data collection
We used data from the most recent DNFCS (2012–2016). The DNFCS is a recurrent survey on food and drinks consumption among the general Dutch population and specific subgroups. A detailed description of the DNFCS 2012–2016 has been published elsewhere [16]. Between November 2012 and January 2017, 6733 people aged 1–79 years were invited to participate in the study. Participants were drawn from market research consumer panels, representative for the Dutch population with regard to age, sex, education level (of the parents or caretakers for children up to 18 years), household region and household location urbanisation level. Data collection was completed for a set of 4313 participants, comprising 672 children aged 1–3 years. For the current study, we included children with complete data on all lifestyle behaviours of interest ($$n = 646$$). A flowchart of the study population selection is presented in Supplementary File 1.
An age-specific, general questionnaire completed by the parent(s) or caregiver(s) provided socio-demographic characteristics and information on lifestyle (e.g. amount of physical activity and electronic screen time) of the participating children. Dietary assessment was performed according to European Food Safety Authority (EFSA) guidelines [17]. Trained dieticians carried out two non-consecutive 24-h dietary recalls [19], equally spread across days of the week and seasons. The first 24-h dietary recall was conducted with a parent or caregiver during a home visit. The second 24-h dietary recall was completed by telephone about 4 weeks later. To adequately capture nutritional intake outside the home, for example at day care, both dietary recalls were combined with a food diary concerning the same day.
The Medical Ethical Committee of the University Medical Centre Utrecht approved the protocol and declared that the Dutch Medical Research Involving Human Subjects Act (WMO) was not applicable to the DNFCS 2012–2016 (reference number 12–359/C). Written informed consent was obtained from all parents/caregivers of participating children during the home visit.
## Diet
The foods and drinks consumed as obtained by the 24-h dietary recalls were classified according to the food groups of the Dutch food-based dietary guidelines (‘Wheel of Five’ guidelines) [20]. Foods and drinks are categorised ‘within the Wheel of Five’ when consumption is advised by the Dutch food-based dietary guidelines and ‘outside the Wheel of Five’ when it is recommended to limit consumption of that particular food or drink. For the drinks category, for example, water and tea are categorised within the Wheel of Five, whereas sugar-sweetened beverages are not part of it. All sweet and savoury snacks, such as cookies, ice cream, and crisps, are categorised outside the Wheel of Five. We used the average intake of the two recall days per participant of the food groups fruit, vegetables, drinks outside the Wheel of Five (mainly sugar-sweetened beverages, therefore referred to as sugar-sweetened beverages in this paper) and snacks outside the Wheel of five (in this paper referred to as unhealthy snacks) in our analyses (g/day).
## Physical activity
Time spent playing outside and participation in organised physical activity, such as swimming, toddler sports classes and dancing, was obtained from the general questionnaire. Parents or caregivers reported frequency of both activities on response categories ranging from ‘never/less than 1 day per week’ to ‘every day’. Response categories for average duration of playing outside ranged from ‘less than half an hour per day’ to ‘more than 3 h per day’. Average duration was converted from hours to minutes. Regarding organised physical activity, we translated one session as 60 min. We calculated the amount of physical activity (min/day) by the following equation: ((days playing outside * average duration of playing outside) + (days participating in organised physical activity * 60))/7.
## Screen time
Time spent watching television or videos and using the computer or other types of electronic screens (such as a handheld game console or tablet) was also obtained from the general questionnaire. Frequency and average duration per session were reported by the parents on scales ranging from ‘never/less than 1 day per week’ to ‘every day’ and ‘less than half an hour per day’ to ‘more than 3 h per day’, respectively. Duration values were converted from hours to minutes. We calculated total screen time (min/day) by adding the amount of watching television/videos to the amount of computer/other screen use: ((days watching television * average duration of watching television) + (days using the computer * average duration of using the computer))/7.
## Socio-demographic characteristics
Information on age, sex, migration background, parental education level, and household size were obtained from the general questionnaire. Children’s migration background (Dutch, Western migration, non-Western migration) was defined based on the parents’ or caregivers’ country of birth. Children were assigned to the latter two categories when at least one parent or caregiver was born abroad [21]. Parental education level was divided into three categories (low, primary education, lower vocational education, advanced elementary education; middle, intermediate vocational education, higher secondary education; high, higher vocational education and university). The market research agency held data on household location region based on the Nielsen CBS division (west, north, east, south (of the Netherlands) and urbanisation level (strongly urbanised, > 1.500 addresses/km2; moderately urbanised, 1.000–1.500 addresses/km2; hardly urbanised, < 1.000 addresses/km2).
## Statistical analyses
All analyses were performed by using SPSS Statistics software (IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.). Characteristics of the children were described in percentages and medians. After standardisation (by calculating Z-scores) of the lifestyle behaviour data, we performed a cluster analysis procedure comprising a hierarchical and consecutive non-hierarchical step. This cluster analysis approach was previously used by Fernández-Alvira et al. [ 22] and Yang et al. [ 23]. First, Ward’s method using squared Euclidean distance was applied to create initial cluster centres, with solutions ranging from two to six clusters. Thereafter, non-hierarchical k-means cluster analysis based on these cluster centres was conducted. The stability of the generated cluster solutions was examined by repeating the clustering procedure in a random sample of $50\%$ of the study population and testing cluster allocation agreement by Cohen’s kappa. Mean values of lifestyle behaviours per cluster were described. Logistic regression models (univariable and multivariable) were used to calculate odds ratios (OR) for allocation to the generated clusters based on the socio-demographic determinants. We applied Bonferroni correction to adjust for multiple testing [$$p \leq 0.05$$/(number of clusters * number of socio-demographic characteristics)] [23].
## Non-response analysis
Of the 672 children aged 1–3 years that participated in the DNFCS, children with missing data on the lifestyle behaviours of interest ($$n = 26$$) were compared (on lifestyle behaviours and socio-demographic characteristics) with children with complete data ($$n = 646$$) by using independent t tests and Chi-square tests.
Children with missing data on the lifestyle behaviours of interest ($$n = 26$$) all lacked data on physical activity only. These children did not differ with regard to the other lifestyle behaviours, nor in socio-demographic characteristics (for all, $p \leq 0.05$) with the children that had complete data ($$n = 646$$, data not shown).
## Population characteristics
The study sample included 646 children aged 1 ($34.2\%$), 2 ($31.0\%$) or 3 ($34.8\%$) years, of which $49.7\%$ were boys (Table 1). The majority of them were of Dutch origin ($92.6\%$) and had parents with a high education level ($66.7\%$). The most common household size consisted of four persons ($43.5\%$). Participating children most often lived in the western part of the Netherlands ($45.5\%$), which is analogous to a strongly urbanised household location ($45.7\%$). The children consumed a median of 140 g (IQR 114) of fruit, 49 g (IQR 60) of vegetables, 362 g [340] of sugar-sweetened beverages, and 32 g (IQR 44) of unhealthy snacks per day. Further, they spent 54 (IQR 62) min/day on physical activity and used electronic screens for 39 min/day (IQR 78) (median values).Table 1Characteristics of children aged 1–3 years in the DNFCS 2012–2016 ($$n = 646$$)CharacteristicValueAge 1 year221 (34.2) 2 years200 (31.0) 3 years225 (34.8)Sex (boys)321 (49.7)Migration background Dutch598 (92.6) Western migration17 (2.6) Non-Western migration31 (4.8)Parental education Low27 (4.2) Middle188 (29.1) High431 (66.7)Size of household Two or three persons186 (28.8) Four persons281 (43.5) Five or more persons179 (27.7)Region of household location West294 (45.5) North75 (11.6) East146 (22.6) South131 (20.3)Household location urbanisation level Strongly urbanised295 (45.7) Moderately urbanised141 (21.8) Hardly urbanised210 (32.5)Fruit intake (g/d)140 [114]Vegetable intake (g/d)49 [60]Sugar-sweetened beverage intake (g/d)362 [340]Unhealthy snack intake (g/d)32 [44]Duration of physical activity (min/d)54 [62]Duration of screen time (min/d)39 [78]Values are frequencies with percentages for categorical variables and medians with interquartile ranges for continuous variables
## Cluster description
Based on the dendrogram and highest Cohen’s kappa coefficient, a three-cluster solution based on the six lifestyle behaviours appeared to be the most accurate (κ = 0.937). Cluster 1 (comprising $49.7\%$ of all children) was labelled the ‘relatively healthy cluster’ because compared to children in the other clusters, children in this cluster complied with guidelines relatively most [20, 24]. It was characterised by healthy dietary factors and low screen time as the Z-score was 0.14 (SE 0.05) for fruit intake, 0.25 (SE 0.06) for vegetable intake, – 0.54 (SE 0.03) for sugar-sweetened beverage intake, – 0.48 (SE 0.03) for unhealthy snack intake, and – 0.49 (SE 0.03) for screen time. High unhealthy snack intake (Z-score = 0.89, SE 0.11) and high physical activity (Z-score = 1.23, SE 0.09) were the main features of cluster 2, which was therefore labelled the ‘active snacking cluster’. Cluster 3 was mainly characterised by high intake of sugar-sweetened beverages (Z-score = 0.93, SE 0.07) and high screen time (Z-score = 0.83, SE 0.08) and was labelled ‘sedentary sweet beverage cluster’. The ‘relatively healthy cluster’ comprised $76\%$ of the 1-year-olds. The mean age for the ‘relatively healthy cluster’ was 1.7 (SD 0.8) years and 2.3 (SD 0.7) years for the other two clusters (Table 2). Figure 1 demonstrates the lifestyle behaviour Z-scores of the various clusters in a radar chart. Table 2Lifestyle behaviours by clusters of children aged 1–3 years in the DNFCS 2012–2016Cluster 1 ‘relatively healthy cluster’aCluster 2 ‘active snacking cluster’bCluster 3 ‘sedentary sweet beverage cluster’bN = 321 ($49.7\%$)$$n = 135$$ ($20.9\%$)$$n = 190$$ ($29.4\%$)Age, y, mean (SD)1.7 (0.8)2.3 (0.7)2.3 (0.7)Fruit consumption, mean (SD)c160 [81]147 [103]129 [83]Z-score (SE)0.14 (0.05) –0.01 (0.10) –0.22 (0.07)Vegetable consumption, mean (SD)c69 [51]53 [44]40 [34]Z-score (SE)0.25 (0.06) –0.09 (0.08) – 0.36 (0.05)Sugar-sweetened beverage consumption, mean (SD)c242 [174]398 [225]676 [298]Z-score (SE) –0.54 (0.03) –0.02 (0.07)0.93 (0.07)Unhealthy snack consumption, mean (SD)c24 [20]72 [45]47 [29]Z-score (SE) –0.48 (0.03)0.89 (0.11)0.19 (0.06)Physical activity, mean (SD)d44 [35]133 [52]63 [39]Z-score (SE) –0.45 (0.04)1.23 (0.09) –0.11 (0.05)Screen time, mean (SD)d24 [26]48 [43]90 [57]Z-score (SE) –0.49 (0.03) –0.01 (0.07)0.83 (0.08)aOverall most consistent with national guidelinesbNamed after most distinguishing lifestyle behaviourscg/daydMin/dayFig. 1Z-scores of lifestyle behaviours in clusters of children aged 1–3 years in the DNFCS 2012–2016
## Association between socio-demographic characteristics and cluster allocation
The ORs for cluster allocation based on the socio-demographic characteristics are presented in Table 3. Based on the three cluster solution, we used a Bonferroni adjusted p value of 0.003 [$$p \leq 0.05$$/(3*6)]. Children aged 1 year had higher odds for allocation to the ‘relatively healthy cluster’ than children aged 3 years old, with an OR of 7.48 ($95\%$ CI 4.91, 11.39; $p \leq 0.001$). Moreover, children aged 1 year had lower odds for allocation to the ‘active snacking cluster’ and ‘sedentary sweet beverage cluster’ compared to children aged 3 years, with ORs of 0.27 ($95\%$ CI 0.16, 0.46; $p \leq 0.001$) and 0.23 ($95\%$ CI 0.15, 0.37; $p \leq 0.001$), respectively. Compared to children of parents with a high education level, children of parents with a low education level had an OR of 0.06 ($95\%$ CI 0.01, 0.26; $p \leq 0.001$) for allocation to the ‘relatively healthy cluster’, and children of parents with a middle education level of 0.48 ($95\%$ CI 0.34, 0.68; $p \leq 0.001$). Contrarily, children of parents with a low education level had an OR of 6.71 ($95\%$ CI 2.92, 15.40; $p \leq 0.001$) for allocation to the ‘sedentary sweet beverage cluster’, and children of parents with a middle education level of 2.13 ($95\%$ CI 1.47, 3.08; $p \leq 0.001$), compared to children of parents with a high education level. We found no associations between parental education level and the ‘active snacking cluster’. Children from households of two or three persons had higher odds for the ‘relatively healthy cluster’ than children from four-person households, OR 1.87 ($95\%$ CI 1.28, 2.73, $$p \leq 0.001$$). This association disappeared in the multivariable model. Sex, migration background, region of household location, and household location urbanisation level were not associated with allocation to any cluster. Table 3Association of socio-demographic characteristics with clusters of children aged 1–3 years in the DNFCSUnivariable modelsaMultivariable models‘Relatively healthy cluster’cN = 321OR ($95\%$ CI)‘Active snacking cluster’dN = 135OR ($95\%$ CI)‘Sedentary sweet beverage cluster’dN = 190OR ($95\%$ CI)‘Relatively healthy cluster’cN = 321OR ($95\%$ CI)‘Active snacking cluster’dN = 135OR ($95\%$ CI)‘Sedentary sweet beverage cluster’dN = 190OR ($95\%$ CI)Age 1 year7.48 (4.91,11.39)**0.27 (0.16, 0.46)**0.23 (0.15, 0.37)**7.78 (4.92, 12.31)**0.30 (0.17, 0.52)**0.22 (0.14, 0.37)** 2 years1.78 (1.19, 2.65)*0.78 (0.50, 1.20)0.70 (0.47, 1.04)1.81 (1.18, 2.76)*0.82 (0.53, 1.28)0.67 (0.44, 1.02) 3 yearsRefRefRefRefRefRefSex GirlRefRefRefRefRefRef Boy1.01 (0.74, 1.38)1.16 (0.79, 1.69)0.88 (0.63, 1.23)1.00 (0.70, 1.43)1.23 (0.83, 1.84)0.82 (0.57, 1.18)Migration background DutchRefRefRefRefRefRef Western migration2.57 (0.89, 7.37)0.22 (0.03, 1.66)0.74 (0.24, 2.29)3.42 (1.13, 10.39)*0.19 (0.02, 1.44)0.67 (0.21, 2.18) Non-Western migration1.94 (0.92, 4.13)0.12 (0.02, 0.86)*1.14 (0.53, 2.47)1.50 (0.63, 3.60)0.16 (0.02, 1.18)1.62 (0.69, 3.79)Parental education Low0.06 (0.01, 0.26)**1.40 (0.58, 3.43)6.71 (2.92, 15.40)**0.06 (0.01, 0.27)**1.03 (0.40, 2.64)5.91 (2.47, 14.11)** Middle0.48 (0.34, 0.68)**1.15 (0.76, 1.75)2.13 (1.47, 3.08)**0.41 (0.28, 0.61)**1.18 (0.76, 1.82)2.21 (1.50, 3.26)** HighRefRefRefRefRefRefSize of household Two or three persons1.87 (1.28, 2.73)**0.55 (0.33, 0.91)*0.71 (0.47, 1.08)1.18 (0.76, 1.83)0.73 (0.42, 1.24)1.03 (0.65, 1.65) Four personsRefRefRefRefRefRef Five or more persons0.78 (0.53, 1.14)1.29 (0.84, 2.00)1.07 (0.72, 1.60)0.65 (0.42, 1.00)1.32 (0.84, 2.08)1.21 (0.78, 1.87)Region of household location WestRefRefRefRefRefRef North0.64 (0.38, 1.07)1.13 (0.60, 2.10)1.51 (0.89, 2.58)0.70 (0.37, 1.29)0.88 (0.44, 1.75)1.55 (0.84, 2.86) East0.91 (0.61, 1.35)1.26 (0.78, 2.04)0.93 (0.59, 1.45)1.05 (0.66, 1.68)1.11 (0.66, 1.88)0.87 (0.53, 1.44) South0.84 (0.56, 1.27)1.13 (0.68, 1.88)1.12 (0.71, 1.75)1.04 (0.64, 1.70)0.96 (0.56, 1.64)1.00 (0.61, 1.64)Household location urbanisation level Strongly urbanisedRefRefRefRefRefRef Moderately urbanised0.87 (0.58, 1.30)1.03 (0.62, 1.72)1.16 (0.75, 1.79)1.12 (0.70, 1.78)0.88 (0.51, 1.50)1.05 (0.65, 1.69) Hardly urbanised0.71 (0.50, 1.02)1.47 (0.96, 2.26)1.10 (0.74, 1.62)0.90 (0.58, 1.40)1.23 (0.76, 2.00)0.96 (0.60, 1.51)Values are ORs with $95\%$ CI, calculated by using logistic regressionThe * and ** indicate the significance levelStatistically significant values are highlighted in boldaIn the univariable models, each independent variable was entered separatelybIn the multivariable models, all independent variables were entered simultaneouslycOverall most consistent with national guidelinesdNamed after most distinguishing lifestyle behaviours *$p \leq 0.05$, **$p \leq 0.003$ (Bonferroni-corrected p value)
## Discussion
We aimed to identify clusters of lifestyle behaviours in Dutch children aged 1–3 years and assess associations with socio-demographic characteristics. Three distinct lifestyle clusters emerged from the data: the ‘relatively healthy cluster’, ‘active snacking cluster’ and ‘sedentary sweet beverage cluster’. The socio-demographic factors age, parental education level and household size were associated with cluster allocation. We found no associations with sex, migration background, region of household location and household location urbanisation level.
In accordance with our findings, previous studies demonstrated healthy, unhealthy and mixed clusters in children [13, 14]. However, precise results differ, partly due to differences in the behaviours considered and in behavioural assessment and clustering techniques. Gubbels et al. and Wang et al. also examined clustering of lifestyle behaviours in Dutch toddlers and identified two and three clusters, respectively [25, 26]. Among 2-year-olds, a ‘sedentary snacking cluster’, characterised by high screen time and high intake of unhealthy snacks and drinks, and a ‘fibre cluster’, mainly depicted by high intakes of fruit, vegetables and brown bread, and low white bread intakes, emerged [25]. Clusters labelled as ‘unhealthy lifestyle pattern’, ‘low snacking and low screen time pattern’, and ‘active, high fruit and vegetable, high snacking and high screen time pattern’ were distinguished among 3-year-olds [26]. Similar to these Dutch studies [25, 26] and to results from other countries [4, 27, 28], we demonstrated that high screen time levels often cluster with high consumption of energy-dense products. Studies in children 5 years and older have suggested that screen time activities, such as watching TV, act as a conditioned cue to drink or eat and distract from feelings of satiety, which might be the two most important underlying mechanisms [29]. In addition, unhealthy food advertisements on TV, computer or other electronic screens may enhance this consumption behaviour [30]. Our other cluster demonstrated high physical activity co-occurring with high intake of unhealthy snacks. This was previously also found in Dutch children of 6 years old [23]. One could argue that parents offer their child a snack as a reward or energy replenishment after physical activity; however, possible explanations need to be further elucidated.
Children aged 1 year were most likely to be allocated to the ‘relatively healthy cluster’. As 1-year-olds have not been included in previous cluster analyses, this is a novel finding. Nevertheless, there are several reasons why lifestyle behaviour in this age group might differ from those of 2- and 3-year-olds. Children aged 1 year have just transitioned from breast or bottle feeding and complementary foods to the family meal time routine. One could argue that parents are, therefore, still conscious of their child’s diet, which is reflected in a relatively higher intake of fruit and vegetables and lower intake of sugar-sweetened beverages and unhealthy snacks. This reason, more focus and consciousness, may also be underlying the fact that children from a household with two or three persons—and therefore most likely one child—had higher odds for allocation to the ‘relatively healthy cluster’. The absence of an association with household size in the multivariable model argues that another factor, possibly age, plays an underlying role. Children aged 1 year might also be more accepting of the (healthy) food their parents offer and most likely will not ask for unhealthy snacks, sugar-sweetened beverages or screen time themselves. They might also consume less of those unhealthier foods because of their lower nutritional needs and longer sleep duration than children aged 2 and 3 years. We presume that the low amount of physical activity in the ‘relatively healthy cluster’ is an underestimation attributable to the physical activity items in the questionnaire. As forms of movement for children aged 1 year (e.g. creeping, crawling, floor play) had not been assessed in this questionnaire, the total amount of physical activity would probably have been greater. Nonetheless, as our results indicate that lifestyle behaviours are healthier in 1-year-olds than in 2- and 3-year-olds, preventive efforts should focus on preserving healthy behaviours in 1-year-old children, i.e. before unhealthy behaviours have rooted.
Although we have to be careful with strong statements given the small group of parents with a low education level, our results support previous studies that have shown that a lower parental education level is associated with clusters comprising less healthy behaviours in young children [4, 23, 25–28]. It seems possible that lower-educated parents possess less knowledge about healthy lifestyle habits for their children or that parenting practices and food environment mediate this association [31–33]. Howbeit, as parents play a crucial role in providing and controlling food and activity habits of children aged 1–3 years, interventions aimed at improving these habits should be tailored to the needs of parents with lower education levels.
## Strengths and limitations
Dietary assessment through 24-h dietary recalls is a major strength of our study, as it does not alter food consumption and has an infinite degree of specificity of the foods consumed. In addition, 24-h dietary recalls are sensitive to culture-specific differences and, when repeatedly conducted, can capture habitual dietary habits. The young age of the study participants, especially 1-year-olds, is another asset and adds new evidence to the importance of early preventive health care.
The young age of the participants might also be a limitation, as age might have been the most important factor in distinguishing lifestyle clusters. Furthermore, it was technically impossible to calculate the exact habitual intake for every individual separately. Therefore, we used the average intake of the two recall days per participant as a reflection of habitual intake, but we are aware that this method might be less accurate. Data on physical activity and screen time were obtained by means of categorical questions. Although included as continuous variables in our analyses, the results of physical activity and screen time, therefore, have limited precision, i.e. are accurate to half an hour. We also acknowledge the sample size as a limitation that may have hampered the robustness of the clusters identified and may have led to selection bias. The low number of participants of non-Dutch origin and from parents with a low education level is another limitation that possibly affected the reliability and generalisability of our results. Due to the cross-sectional design of the DNFCS, we could not draw causal conclusions on the association between cluster allocation and weight status. Besides, data was obtained between 2012 and 2017 and new ‘Wheel of Five’ guidelines have been published in the meantime, which may affect current dietary intake.
## Conclusions
We distinguished three clusters of lifestyle behaviours in children as young as 1–3 years of age. Children aged 1 year were more likely to be in the cluster that portrayed healthy behaviour than children aged 2 and 3 years, which suggests that maintaining healthy behaviour and changing towards more healthy behaviour should be promoted in these age groups, respectively. These preventive efforts should take parental education level into consideration. Future longitudinal research should assess cluster allocation evolution and its association with weight status.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 48 KB)
## References
1. 1.United Nations Children's Fund (UNICEF), World Health Organization, & International Bank for Reconstruction and Development/The World BankLevels and trends in child malnutrition: key findings of the 2021 edition of the joint child malnutrition estimates2021GenevaWHO. *Levels and trends in child malnutrition: key findings of the 2021 edition of the joint child malnutrition estimates* (2021.0)
2. Kumar S, Kelly AS. **Review of childhood obesity: from epidemiology, etiology, and comorbidities to clinical assessment and treatment**. *Mayo Clin Proc* (2017.0) **92** 251-265. DOI: 10.1016/j.mayocp.2016.09.017
3. Singh AS, Mulder C, Twisk JW, van Mechelen W, Chinapaw MJ. **Tracking of childhood overweight into adulthood: a systematic review of the literature**. *Obes Rev* (2008.0) **9** 474-488. DOI: 10.1111/j.1467-789X.2008.00475.x
4. Lioret S, Campbell KJ, McNaughton SA, Cameron AJ, Salmon J, Abbott G. **Lifestyle patterns begin in early childhood, persist and are socioeconomically patterned, confirming the importance of early life interventions**. *Nutrients* (2020.0). DOI: 10.3390/nu12030724
5. Kunaratnam K, Halaki M, Wen LM, Baur LA, Flood VM. **Tracking preschoolers’ lifestyle behaviors and testing maternal sociodemographics and BMI in predicting child obesity risk**. *J Nutr* (2020.0) **150** 3068-3074. DOI: 10.1093/jn/nxaa292
6. Keller A, Torre BD. **Sugar-sweetened beverages and obesity among children and adolescents: a review of systematic literature reviews**. *Child Obes* (2015.0) **11** 338-346. DOI: 10.1089/chi.2014.0117
7. Poitras VJ, Gray CE, Janssen X, Aubert S, Carson V, Faulkner G. **Systematic review of the relationships between sedentary behaviour and health indicators in the early years (0–4 years)**. *BMC Public Health* (2017.0) **17** 868. DOI: 10.1186/s12889-017-4849-8
8. Ledoux TA, Hingle MD, Baranowski T. **Relationship of fruit and vegetable intake with adiposity: a systematic review**. *Obes Rev* (2011.0) **12** e143-150. DOI: 10.1111/j.1467-789X.2010.00786.x
9. Carson V, Lee EY, Hewitt L, Jennings C, Hunter S, Kuzik N. **Systematic review of the relationships between physical activity and health indicators in the early years (0–4 years)**. *BMC Public Health* (2017.0) **17** 854. DOI: 10.1186/s12889-017-4860-0
10. Steenbergen E, Krijger A, Verkaik-Kloosterman J, Elstgeest LEM, Ter Borg S, Joosten KFM. **Evaluation of nutrient intake and food consumption among Dutch toddlers**. *Nutrients* (2021.0). DOI: 10.3390/nu13051531
11. Friel CP, Duran AT, Shechter A, Diaz KM. **U.S. children meeting physical activity, screen time, and sleep guidelines**. *Am J Prev Med* (2020.0) **59** 513-521. DOI: 10.1016/j.amepre.2020.05.007
12. Newby PK, Tucker KL. **Empirically derived eating patterns using factor or cluster analysis: a review**. *Nutr Rev* (2004.0) **62** 177-203. DOI: 10.1301/nr.2004.may.177-203
13. Gubbels JS, van Assema P, Kremers SP. **Physical activity, sedentary behavior, and dietary patterns among children**. *Curr Nutr Rep* (2013.0) **2** 105-112. DOI: 10.1007/s13668-013-0042-6
14. Leech RM, McNaughton SA, Timperio A. **The clustering of diet, physical activity and sedentary behavior in children and adolescents: a review**. *Int J Behav Nutr Phys Act* (2014.0) **11** 4. DOI: 10.1186/1479-5868-11-4
15. D'Souza NJ, Kuswara K, Zheng M, Leech R, Downing KL, Lioret S. **A systematic review of lifestyle patterns and their association with adiposity in children aged 5–12 years**. *Obes Rev* (2020.0) **21** e13029. DOI: 10.1111/obr.13029
16. Van Rossum C, Buurma-Rethans E, Dinnissen C, Beukers M, Brants H, Dekkers A. *The diet of the Dutch: results of the Dutch national food consumption survey 2012–2016* (2020.0)
17. **Guidance on the EU menu methodology**. *EFSA J* (2014.0) **12** 3944. DOI: 10.2903/j.efsa.2014.3944
18. de Boer EJ, Slimani N, van t’Veer P, Boeing H, Feinberg M, Leclercq C. **The European food consumption validation project: conclusions and recommendations**. *Eur J Clin Nutr* (2011.0) **65** S102-107. DOI: 10.1038/ejcn.2011.94
19. Slimani N, Ferrari P, Ocké M, Welch A, Boeing H, Liere M. **Standardization of the 24-hour diet recall calibration method used in the european prospective investigation into cancer and nutrition (EPIC): general concepts and preliminary results**. *Eur J Clin Nutr* (2000.0) **54** 900-917. DOI: 10.1038/sj.ejcn.1601107
20. Brink E, van Rossum C, Postma-Smeets A, Stafleu A, Wolvers D, van Dooren C. **Development of healthy and sustainable food-based dietary guidelines for the Netherlands**. *Public Health Nutr* (2019.0) **22** 2419-2435. DOI: 10.1017/S1368980019001435
21. 21.Definition of migration background. Retrieved from https://www.cbs.nl/nl-nl/onze-diensten/methoden/begrippen/migratieachtergrond (accessed on Jul 28, 2021)
22. Fernández-Alvira JM, De Bourdeaudhuij I, Singh AS, Vik FN, Manios Y, Kovacs E. **Clustering of energy balance-related behaviors and parental education in European children: the ENERGY-project**. *Int J Behav Nutr Phys Act* (2013.0) **10** 5. DOI: 10.1186/1479-5868-10-5
23. Yang-Huang J, van Grieken A, Wang L, Jansen W, Raat H. **Clustering of sedentary behaviours, physical activity, and energy-dense food intake in six-year-old children: associations with family socioeconomic status**. *Nutrients* (2020.0). DOI: 10.3390/nu12061722
24. 24.World Health OrganizationGuidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age2019GenevaWorld Health Organization. *Guidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age* (2019.0)
25. Gubbels JS, Kremers SP, Stafleu A, Dagnelie PC, de Vries SI, de Vries NK. **Clustering of dietary intake and sedentary behavior in 2-year-old children**. *J Pediatr* (2009.0) **155** 194-198. DOI: 10.1016/j.jpeds.2009.02.027
26. Wang L, Jansen W, van Grieken A, Vlasblom E, Boere-Boonekamp MM, L'Hoir MP. **Identifying patterns of lifestyle behaviours among children of 3 years old**. *Eur J Public Health* (2020.0) **30** 1115-1121. DOI: 10.1093/eurpub/ckaa109
27. Saldanha-Gomes C, Marbac M, Sedki M, Cornet M, Plancoulaine S, Charles MA. **Clusters of diet, physical activity, television exposure and sleep habits and their association with adiposity in preschool children: the EDEN mother-child cohort**. *Int J Behav Nutr Phys Act* (2020.0) **17** 20. DOI: 10.1186/s12966-020-00927-6
28. Miguel-Berges ML, Zachari K, Santaliestra-Pasias AM, Mouratidou T, Androutsos O, Iotova V. **Clustering of energy balance-related behaviours and parental education in European preschool children: the ToyBox study**. *Br J Nutr* (2017.0) **118** 1089-1096. DOI: 10.1017/S0007114517003129
29. Marsh S, Ni Mhurchu C, Maddison R. **The non-advertising effects of screen-based sedentary activities on acute eating behaviours in children, adolescents, and young adults. A systematic review**. *Appetite* (2013.0) **71** 259-273. DOI: 10.1016/j.appet.2013.08.017
30. Andreyeva T, Kelly IR, Harris JL. **Exposure to food advertising on television: associations with children's fast food and soft drink consumption and obesity**. *Econ Hum Biol* (2011.0) **9** 221-233. DOI: 10.1016/j.ehb.2011.02.004
31. Gibson EL, Androutsos O, Moreno L, Flores-Barrantes P, Socha P, Iotova V. **Influences of parental snacking-related attitudes, behaviours and nutritional knowledge on young children’s healthy and unhealthy snacking: the ToyBox study**. *Nutrients* (2020.0). DOI: 10.3390/nu12020432
32. Rodenburg G, Kremers SP, Oenema A, van de Mheen D. **Associations of parental feeding styles with child snacking behaviour and weight in the context of general parenting**. *Public Health Nutr* (2014.0) **17** 960-969. DOI: 10.1017/S1368980013000712
33. Wijtzes AI, Jansen W, Jansen PW, Jaddoe VW, Hofman A, Raat H. **Maternal educational level and preschool children's consumption of high-calorie snacks and sugar-containing beverages: mediation by the family food environment**. *Prev Med* (2013.0) **57** 607-612. DOI: 10.1016/j.ypmed.2013.08.014
|
---
title: 'Dietary fat intake is associated with insulin resistance and an adverse vascular
profile in patients with T1D: a pooled analysis'
authors:
- Noppadol Kietsiriroje
- Hanya Shah
- Marios Zare
- Lauren L. O’Mahoney
- Daniel J. West
- Sam M. Pearson
- Ramzi A. Ajjan
- Matthew D. Campbell
journal: European Journal of Nutrition
year: 2022
pmcid: PMC10030402
doi: 10.1007/s00394-022-03070-z
license: CC BY 4.0
---
# Dietary fat intake is associated with insulin resistance and an adverse vascular profile in patients with T1D: a pooled analysis
## Abstract
### Background
Insulin resistance (IR) increases vascular risk in individuals with Type 1 Diabetes (T1D). We aimed to investigate the relationship between dietary intake and IR, as well as vascular biomarkers in T1D.
### Methods
Baseline data from three randomised controlled trials were pooled. Estimated glucose disposal rate (eGDR) was used as an IR marker. Employing multivariate nutrient density substitution models, we examined the association between macronutrient composition and IR/vascular biomarkers (tumour necrosis factor-α, fibrinogen, tissue factor activity, and plasminogen activator inhibitor-1).
### Results
Of the 107 patients, $50.5\%$ were male with mean age of 29 ± 6 years. Those with lower eGDR were older with a longer diabetes duration, higher insulin requirements, and an adverse vascular profile ($p \leq 0.05$). Patients with higher degrees of IR had higher total energy intake (3192 ± 566 vs. 2772 ± 268 vs. 2626 ± 395 kcal/d for eGDR < 5.1 vs. 5.1–8.6 vs. ≥ 8.7 mg/kg/min, $p \leq 0.001$) and consumed a higher absolute and proportional amount of fat (47.6 ± 18.6 vs. 30.4 ± 8.1 vs. 25.8 ± $10.4\%$, $p \leq 0.001$). After adjusting for total energy intake, age, sex, and diabetes duration, increased carbohydrate intake offset by an isoenergetic decrease in fat was associated with higher eGDR (β = 0.103, $95\%$ CI 0.044–0.163). In contrast, increased dietary fat at the expense of dietary protein intake was associated with lower eGDR (β = − 0.119, $95\%$ CI − 0.199 to − 0.040). Replacing fat with $5\%$ isoenergetic amount of carbohydrate resulted in decreased vascular biomarkers ($p \leq 0.05$).
### Conclusion
Higher fat, but not carbohydrate, intake is associated with increased IR and an adverse vascular profile in patients with T1D.
## Introduction
Weight gain and insulin resistance (IR) in type 1 diabetes (T1D) are prevalent and a significant source of morbidity and mortality [1]. In the most recent UK National Diabetes Audit, $63.5\%$ of individuals with T1D were classified as overweight or obese [2]—a phenotype which expresses greater IR and a concomitant increased risk of vascular complications irrespective of glycaemic control [1, 3]. Mechanisms contributing to the increased risk of vascular complications partly due to interaction between insulin resistance and inflammation creating prothrombotic environment [4]. Elevation of vascular biomarkers, such as tumour necrotic factor-alpha (TNF-α), fibrinogen, tissue factor (TF) activity, and plasminogen activator inhibitor-1 (PAI-1) [5], has been linked to endothelial dysfunction, prothrombotic clot formation and hypofibrinolysis, thus resulting in increased vascular risk for atherosclerosis [5, 6].
Excessive energy intake leads to weight gain and predisposes to IR. However, it is widely acknowledged that individual dietary macronutrients consumed in differing isoenergetic quantities have differential effects on IR and vascular risk factors [7]. Recently, it has been shown that in T1D, the relative distribution of dietary macronutrients was associated with the presence of the metabolic syndrome components [8], in men but not women. Favouring carbohydrate intake over fat was associated with lower waist circumference and that favouring either carbohydrates or fat over protein was associated with a lower prevalence of blood pressure [8]. To the best of our knowledge, the association between macronutrient intake and IR—a key mechanistic driver of increased vascular complications—has never been evaluated in people with T1D.
In T1D, the focus on carbohydrate intake is often emphasised whereby structured education provided to patients for managing mealtime insulin dose is centred on total carbohydrate amount [9]. However, preference for high fat and protein over carbohydrate has been previously reported [10] and anecdotally, within the T1D community, there is often a concern that increased carbohydrate intake increases IR and worsens glucose management and that carbohydrate restriction should be promoted. In the present study, we pooled pre-treatment data from three randomised controlled trials (RCTs) and employed a multivariable nutrient density substitution model to assess the association between relative macronutrient components with IR and vascular biomarkers in a population of well-defined T1D patients.
## Study population
We pooled data from three randomised controlled trials (RCTs; Clinical trial registration: ISRCTN40811115; ISRCTN13641847, NCT05231642) each of which received ethical approval from local National Health Service Research Ethics Committees (REC reference: 17/NE/0244, 20/LO/0650, 21/WA/0381.) Written informed consent was obtained from all participants. In the present analysis, we included 107 participants that met inclusion criteria as described previously [11, 12] including classical presentation of T1D, aged 18–50 years, diabetes duration of ≥ 5 years, treated on a stable (> 12-months) basal-bolus insulin regimen delivered through multiple daily injections or continuous subcutaneous insulin infusion and no established diabetes-related complications.
## Data collection and study procedures
We performed cross-sectional-analyses using baseline pre-treatment data across each RCT. Overnight fasting venous blood samples were obtained and analysed for plasma levels of vascular biomarkers, including TNF-α (Human TNF-α Quantikine ELISA; R&D Systems, Roche Diagnostics, UK), fibrinogen (ab108842, Fibrinogen Human ELISA Kit; Abcam, Japan), TF activity (Human Tissue Factor activity ab108906; Abcam, UK), and PAI-1 activity (Human PAI-1/serpin ELISA Kit DSE100; R&D systems, UK). Estimated glucose disposal rate (eGDR) was calculated using a validated formula: eGDR = 19.02−(0.22 × BMI [kg/m2])−(3.26 × HTN)−(0.61 × HbA1c [%]), whereby HTN is hypertension (1 = yes, 0 = no) [4]. Participants were defined as hypertensive if blood pressure ≥ $\frac{140}{90}$ mmHg, they had a pre-existing diagnosis of hypertension or were prescribed antihypertensive drugs.
To estimate dietary intake, participants completed two independent dietary assessments; a 48-h weighed food diary and a validated DINE Food Frequency Questionnaire (FFQ) [13]. We employed both assessment techniques to facilitate cross-validation and improve accuracy of reporting [14]. Analysis of the 48-h weighed food diary was performed using the validated MyFood24 tool [15]. Using the DINE method, fat intake [13] was categorised into three groups whereby frequencies of fat consumption reported by patients were translated into a score. A DINE fat score < 30 (equivalent to ≤ 83 g/day) indicates low fat intake (DINE1), whereas 30–40 (equivalent to 84–122 g/day) and score > 40 (equivalent to > 122 g/day) indicate moderate (DINE2) and high-fat intake (DINE3), respectively.
## Statistical analysis
Baseline characteristics were presented according to eGDR tertiles. Continuous variables are reported as mean ± SD and categorical variables are reported as frequency (percentage). Conditional differences assessed used one-way ANOVA with post hoc Bonferroni for continuous variables and Chi-square for categorical variables. Associations between macronutrient intake and IR were investigated using a generalised linear regression analysis whereby eGDR was entered as a dependent variable and one macronutrient (from the 48-h weighed food diary) at a time was entered as an independent variable with total energy intake, age, sex, and diabetes duration used as covariates.
We employed a series of multivariate nutrient density substitution models to examine the effect of increasing an isoenergetic amount of one macronutrient at the expense of another on IR and vascular biomarkers. This technique has been described in detail elsewhere [8] but in brief, a series of sequential generalised linear regression analyses were performed featuring either eGDR or vascular biomarkers as a dependent variable; Using macronutrient variables assessed by the 48-h weighed food diary, we included all but one macronutrient (per 5 E% presented in parentheses) and total energy intake as covariates; in a second adjusted model, we included age, sex, and diabetes duration as additional covariates. For example, in a model replacing fat with a $5\%$ isoenergetic amount of carbohydrate [↑CHO (↓FAT)], the %E of carbohydrate after $5\%$ isoenergetic substitution was entered as an independent variable, whereas fat was excluded from the model. Protein intake, total energy intake, and other selected variables were used as covariates. The results can be interpreted as the increase or decrease in the dependent outcome related to isoenergetic (5E%) substitution of a given macronutrient in the model with the macronutrient omitted from the model. For instance, in an equation: eGDR = β0 + β1 (5E% increase from carbohydrates) + β2 (E% from protein) + β3 (total energy intake), β1 would be interpreted as the change in the eGDR value when dietary carbohydrate intake is increased by 5E% at the expense of fat.
Data were analysed using SPSS (IBM SPSS Statistics 25, IBM Corporation, USA). Statistical significance was set at $p \leq 0.05$ for all analyses.
## Results
Our study population consisted of $$n = 107$$ patients with T1D. We stratified this cohort by eGDR tertiles, with lower eGDR values conferring higher degrees of IR. Baseline demographic and clinical characteristics are presented. Of the 107 patients, $50.5\%$ were male with mean age of 29 ± 6 years. Those with lower eGDR were typically older, (Table 1) had a longer diabetes duration, required higher insulin doses, and had an adverse vascular profile ($p \leq 0.05$).Table 1Clinical characteristics, vascular biomarker levels, and nutritional intake of the study population categorised by eGDR tertiles ($$n = 107$$)All patientseGDR tertiles (mg/kg/min)p value < 5.15.1–8.6≥ 8.7N107363833Clinical characteristics Sex (%male)$50.5\%$$52.8\%$$50.0\%$$48.5\%$0.936 Age (years)29 ± 632 ± 628 ± 5*27 ± 5* < 0.001 BMI26.3 ± 3.428.7 ± 3.526.6 ± 2.4*23.3 ± 1.4* < 0.001 Hypertension (%)$44.9\%$$100\%$$31.6\%$$0\%$ < 0.001 HbA1c (%) [mmol/mol]8.0 ± 1.2 [63.8 ± 12.8]9.0 ± 1.0 [75.1 ± 11.4]7.8 ± 1.0* [62.3 ± 10.9]*7.0 ± 0.3* [53.4 ± 2.8]* < 0.001 eGDR (mg/kg/min)6.9 ± 2.53.9 ± 1.07.3 ± 1.2*9.6 ± 0.4* < 0.001 Diabetes duration (years)16.5 ± 7.019.5 ± 8.417.6 ± 4.311.9 ± 5.5* < 0.001 Daily insulin dose (units)47 ± 1555 ± 2047 ± 10*39 ± 5* < 0.001Vascular biomarker levels TNF-α (pg/mL)4.23 ± 1.655.99 ± 1.313.83 ± 0.81*2.77 ± 0.74* < 0.001 Fibrinogen (mg/mL)2.19 ± 1.143.38 ± 0.891.87 ± 0.69*1.27 ± 0.54* < 0.001 TF activity (units/mL)70.5 ± 30.4102.4 ± 19.563.3 ± 19.6*44.0 ± 16.3* < 0.001 PAI-1 (pg/mL)1274 ± 6571995 ± 4871059 ± 366*735 ± 282* < 0.001Nutritional intake from 48-h weighed food diary Energy intake (Kcal/d)2869 ± 4853192 ± 5662772 ± 268*2626 ± 395* < 0.001 Carbohydrate (% energy)48.3 ± 17.135.8 ± 18.753.0 ± 12.2*56.6 ± 12.0* < 0.001 Fat (% energy)34.8 ± 16.147.6 ± 18.630.4 ± 8.1*25.8 ± 10.4* < 0.001 Protein (% energy)16.9 ± 7.416.5 ± 8.816.6 ± 6.617.6 ± 6.70.798DINE assessment N93353325 Total fat rating30.4 ± 13.142.9 ± 11.227.2 ± 5.4*17.1 ± 3.8* < 0.001 Unsaturated fat rating8.6 ± 2.88.4 ± 2.38.2 ± 2.89.4 ± 3.20.272 Fibre rating29.6 ± 14.330.5 ± 16.630.8 ± 12.026.6 ± 13.80.475DINE fat group (%) 1 Low fat$54.8\%$$0\%$$78.8\%$$100\%$ < 0.001 2 Medium fat$26.9\%$$54.3\%$$18.2\%$$0\%$ 3 High fat$18.3\%$$45.7\%$$3.0\%$$0\%$Continuous variables are reported as mean ± SD; categorical variables are reported as frequency (percentage). Conditional differences assessed used one-way ANOVA for continuous variables and Chi-square for categorical variablesTNF-α tumour necrotic factor-alpha, TF tissue factor, PAI-1 plasminogen activator inhibitor-10*Post hoc Bonferroni $p \leq 0.05$ compared with 1st tertile When evaluating self-reported 48-h dietary intake in our patients, daily energy intake was inversely correlated with eGDR, with patients in the lowest eGDR tertile (i.e., highest degree of IR), consuming the greatest energy intakes. When exploring differences in macronutrient intake, patients with lowest eGDR reported higher-fat and lower-carbohydrate intakes. Protein intake was similar across eGDR tertiles. When categorising patients based on the DINE method, a similar pattern was evident, with fat intake increasing in a stepwise fashion with decreasing eGDR. Unsaturated fat and fibre levels were similar across eGDR tertiles (Table 2).Table 2Association between macronutrient intake from the 48-h weighed food diary and eGDR in patients with T1DModel 1Model 2β (CI)p valueβ (CI)p valueCarbohydrate0.046 (0.005 to 0.087)0.029*0.031 (− 0.005 to 0.067)0.094Fat− 0.129 (− 0.196 to − 0.062) < 0.001**− 0.101 (− 0.161 to − 0.042)0.001*Protein− 0.002 (− 0.059 to 0.054)0.9380.008 (− 0.042 to 0.057)0.764Model 1 is adjusted for energy intake, and Model 2 was fit to estimate associations with adjustment for age, sex, and diabetes duration and energy intake*and bold text denotes significant association at $p \leq 0.05$**and bold text denotes a significant association at $p \leq 0.001$ To investigate the relationship between IR and macronutrient intake further, we employed a series of multivariate nutrient density models to examine the associations between dietary macronutrient amounts from the 48-h food diary and IR. The percentage of energy from carbohydrate was positively associated with eGDR (greater insulin sensitivity), and fat inversely associated with eGDR (greater IR) in unadjusted models; the association between fat and eGDR remained robust following adjustment. We then studied the association of the relative proportions of dietary macronutrients with IR in adjusted energy-controlled substitution models. These models indicate that increased carbohydrate intake offset by an isoenergetic decrease in fat is associated with higher eGDR (decreased IR) and that increased dietary fat at the expense of dietary carbohydrate or protein intake is associated with lower eGDR (increased IR) (Table 3)Table 3The changes in eGDR levels after isoenergetic substitution of one macronutrient (from 48-h weighed food diary) to another (in parenthesis) by $5\%$ of total energy ($$n = 107$$)Model 1Model 2β (CI)p valueβ (CI)p value↑CHO (↓Fat)0.130 (0.063 to 0.198) < 0.001**0.103 (0.044 to 0.163)0.001*↑CHO (↓Protein)− 0.010 (− 0.063 to 0.044)0.720− 0.016 (− 0.063 to 0.031)0.504↑Fat (↓Protein)− 0.014 (− 0.231 to − 0.049)0.003*− 0.119 (− 0.199 to − 0.040)0.003*Model 1 is adjusted for energy intake; Model 2 is fit to estimate associations with adjustment for age, sex, and diabetes duration and energy intake. In each model, a given macronutrient is included as an independent variable and one of the macronutrients (in parentheses) is excluded from the model. The remaining macronutrients and other covariates (total energy intake, age, sex, and diabetes duration) are included as covariates. The β represents the increase or decrease in the eGDR variable when increasing the intake of the independent macronutrient by $5\%$ of total energy, while simultaneously reducing an isoenergetic amount of the excluded macronutrient [8]*and bold text denotes significant association at $p \leq 0.05$**and bold text denotes a significant association at $p \leq 0.001$ When evaluating the relationship between macronutrients and vascular biomarkers, substituting fat with a $5\%$ isoenergetic amount of carbohydrate resulted in a decrease across all chosen vascular biomarkers in unadjusted analyses. Following adjustment for age, sex, and diabetes duration, the associations remained robust for PAI-1, (Table 4) TF activity, and Fibrinogen, but not TNF-α. Conversely, substituting protein with fat resulted in a concomitant increase in biomarkers in unadjusted analyses, but the association was no longer significant for TNF-α and TF activity following adjustment. Table 4The changes in vascular biomarker levels after isoenergetic substitution of one macronutrient from 48-h weighed food diary to another (in parenthesis) by $5\%$ of total energy ($$n = 107$$)TNF-α (pg/mL)Fibrinogen (mg/mL)TF activity (units/mL)PAI-1 (pg/mL)β ($95\%$ CI)p valueβ ($95\%$ CI)p valueβ ($95\%$ CI)p valueβ ($95\%$ CI)p valueModel 1 ↑CHO (↓Fat)− 0.055 (− 0.101, − 0.009)0.020*− 0.041 (− 0.073, − 0.009)0.012*− 1.408 (− 2.253, − 0.562)0.001**− 31.22 (− 49.39, − 13.05)0.001* ↑CHO (↓Protein)0.011 (− 0.025, 0.048)0.5510.007 (− 0.018, 0.033)0.576− 0.173 (− 0.843, 0.497)0.6130.415 (− 13.99, 14.82)0.955 ↑Fat (↓Protein)0.066 (0.004, 0.128)0.037*0.048 (0.005, 0.091)0.028*1.235 (0.097, 2.373)0.03331.64 (7.18, 56.11)0.011*Model 2 ↑CHO (↓Fat)− 0.037 (− 0.077, 0.002)0.064− 0.029 (− 0.058, − 0.001)0.042*− 1.064 (− 1.804, − 0.324)0.005**− 24.75 (− 40.91, − 8.60)0.003* ↑CHO (↓Protein)0.013 (− 0.018, 0.045)0.3980.009 (− 0.013, 0.032)0.408− 0.082 (− 0.664, 0.501)0.7831.47 (− 11.25, 14.19)0.821 ↑Fat (↓Protein)0.051 (− 0.002, 0.104)0.0590.039 (0.001, 0.077)0.045*0.982 (− 0.005, 1.970)0.05126.22 (4.65, 47.80)0.017*Model 1 is adjusted for energy intake; Model 2 is Model 1 with further adjustment for age, sex, and diabetes duration and energy intake. In each model, a given macronutrient is included as an independent variable and one of the macronutrients (in parentheses) is excluded from the model. The remaining macronutrient and other confounders (total energy, age, sex, and diabetes duration) are included as covariates. The β represents the increase or decrease in the vascular biomarkers when increasing the intake of the independent macronutrient by $5\%$ of total energy, while simultaneously reducing an isoenergetic amount of the excluded macronutrient [8]*and bold text denotes significant association at $p \leq 0.05$** and bold text denotes a significant association at $p \leq 0.001$
## Discussion
To the best of our knowledge, this is the first report showing that dietary intake is associated with IR and an adverse vascular profile in patients with T1D. Patients with a high degree of IR tended to have higher overall energy intakes resulting largely from a relatively higher-fat intake, as compared to those with lower degrees of IR. Utilizing our nutrient substitutional model, we reveal that increasing amounts of carbohydrate offset by an isoenergetic decrease in fat results in greater insulin sensitivity levels, whereas increased dietary fat at the expense of dietary carbohydrate or protein intake is associated with increased IR. We also show that increased fat consumption is associated with an adverse vascular biomarker profile, whereas replacing fat intake with carbohydrate is associated with a favourable vascular biomarker profile.
The link between IR (assessed by eGDR) and vascular health has been demonstrated recently by our group [1, 16] and others [17, 18], whereby IR increases risk of vascular complications in T1D. Increased total energy intake was associated with IR, which, in this study, was mainly driven by increased fat consumption. In our patients categorised in the lowest eGDR tertile, energy from fat accounted for ~ $48\%$ of total energy intake, as compared to ~ 26–$30\%$ in other eGDR tertiles. Importantly, the association between fat intake and IR remained robust in our nutrient substitution model whereby one macronutrient is substituted for an isoenergetic amount of another. Given the cross-sectional nature of the present study with self-report diet data captured at a single time-point, conclusions should be interpreted with caution. However, a plausible explanation for our findings is that a higher proportion of fat at a given energy intake, on average, induces IR to a greater degree than an equivalent calorific amount of carbohydrate, thus favouring a lower-fat and higher-carbohydrate diet. Data from short-term preliminary clinical studies are equivocal with regards to the metabolic advantages of lower-fat higher-carbohydrate verses lower-carbohydrate higher-fat diets [19]. However, in animal models, high-fat high calorie feeding has been shown to dramatically induce IR [20, 21], and in our acute feeding studies in humans with T1D, we have shown that a high-fat feeding challenge promotes adverse glucose and inflammatory profiles and increases insulin requirements [11].
In T1D, preference for high fat and protein over carbohydrate has been previously reported [10]. Within the T1D community, there is often a concern that increased carbohydrate intake increases IR and worsens glucose management and that carbohydrate intake should be restricted [22]. However, our cross-sectional data do not support this notion, whereby increased carbohydrate intake was associated with lower IR, improved glucose management (HbA1c), and a more favourable vascular profile. These findings, albeit preliminary, support previously published research in which high-fat intake was associated with increased coronary heart disease risk and coronary artery calcium in a cohort of 571 individuals with T1D [10].
*In* general, most dietary guidelines focus predominately on single nutrients, recommending to reduce saturated fat consumption and advocating whole foods over those foods which are heavily processed [23–25]. Beyond macronutrients, diet quality and food processing are important considerations. For example, we have previously shown that T1D individuals express differential fatty acid profiles with regards to IR status and vascular biomarkers [26] and that postprandial vascular-inflammatory and thrombotic responses to high-fat feeding are influenced differentially not only be total fat amount, but also food processing [27].
The impact of high-fat intake (particularly saturated fatty acids, SFAs) on IR is heavily mediated by inflammatory processes [28] directly inducing multiple pleiotropic proinflammatory pathways. Namely, activation of Toll-like receptor-4 pathway which further activates secondary cascades such as c-Jun N-terminal kinase, nuclear factor-kappa B, and protein C kinase signalling pathways which are implicated in the desensitisation of insulin receptors.[28] A cross-sectional analysis in 555 patients with T2D from the Insulin Resistance Atherosclerosis Study has also demonstrated associations between serum total SFA and various vascular-inflammatory markers, including PAI-1, TNF-α, and fibrinogen [29]. In the present study, those individuals with the greatest fat intake also consumed the largest absolute and relative amounts of saturated fat. Therefore, we cannot exclude the possibility that the associations between fat intake and IR/vascular biomarkers were driven not only be fat amount, but also fat type. Due to data structure and sample size, it was not possible to test the hypothesis that substituting an isoenergetic amount of SFA’s at the expense of unsaturated fat increases IR and worsens vascular biomarkers. However, an extensive review of studies in non-T1D individuals shows that substitution of SFAs by isocaloric exchange with monosaturated fatty acids (MUFAs) or polysaturated fatty acids (PUFAs) improves lipid metabolism (including lower levels of LDL-C, triglycerides, ApoB, and ApoA-I, as well as total cholesterol:HDL-C ratio), and glucose homeostasis (including lower HbA1c and IR measured as HOMA-IR), although findings were less conclusive regarding the impact of this on cardiovascular disease risk [30]. Further, results from the OmniHeart trial demonstrated that replacing carbohydrate with unsaturated but not saturated fat improves insulin sensitivity in individuals with pre-hypertension or hypertension stage I without diabetes [31]. A meta-analysis has shown that substituting carbohydrate or PUFA-enriched diets with an MUFA-enriched diet, improved body weight, fasting glucose, lipid profiles, and blood pressure in 1,460 people with T2D [32].
While our pooled retrospective analysis is the first to explore and offers valuable insight into the association between dietary fat intake with IR and vascular health, this study is not without limitations and include [1] our cross-sectional design featuring self-reported dietary intake at a single time-point. Whereas it is possible that increasing fat intake may increase IR, it is also possible that those patients presenting with IR may have previously transitioned to a lower-carbohydrate diet. [ 2] Self-report dietary assessments have inherent limitations, although our results were consistent between the self-reported weighed food diary and validated FFQ. We used a weighted food diary to obtain accurate estimates of food intake; however, there are known limitations of this including increased participant burden, participant biases, and issues regarding the representative nature of acute versus longer-term dietary patterns. Therefore, we also employed a brief (to minimise participant burden) DINE FFQ which captures generalised long-term dietary patterns. [ 3] From our current analysis, it was not possible to assess diet quality and food processing, or individual nutrient sub-groups which have previously been shown to impact metabolic health [27, 33] beyond dietary macronutrient distribution. [ 4] The association of dietary components with outcomes of interest are likely to be non-linear [34]. To address these limitations, a longitudinal observation in a larger representative sample assessing diet in more detail, specifically the threshold at which dietary components increase risk of IR, is warranted.
## Conclusion and future direction
This is the first study to demonstrate dietary macronutrient intake, specifically higher-fat intake, is associated with IR and an adverse vascular profile in patients with T1D. Patients with higher degrees of IR presented with higher total energy intakes and consumed a higher absolute and proportional amount of dietary fat. In the present study, patients with IR and high dietary fat intakes presented with an adverse vascular profile. Future research is required to explore the impact of diet in greater detail with a specific focus on individual dietary components, including diet quality, processing, and timing, thus enabling more accurate and personalized individually dietary management. A different dietary assessment tool such as 7-day food diary may also be required to better display dietary patterns in this group.
## References
1. Helliwell R, Warnes H, Kietsiriroje N, Campbell M, Birch R, Pearson SM. **Body mass index, estimated glucose disposal rate and vascular complications in type 1 diabetes: beyond glycated haemoglobin**. *Diabetic Med* (2021.0) **38** e14529. DOI: 10.1111/dme.14529
2. 2.National Diabetes Audit, 2019–2020, Type 1 Diabetes 2021 [Available from: https://digital.nhs.uk/data-and-information/publications/statistical/national-diabetes-audit/national-diabetes-audit-2019-20-type-1-diabetes. Accessed 03 Dec 2021
3. Merger SR, Kerner W, Stadler M, Zeyfang A, Jehle P, Muller-Korbsch M. **Prevalence and comorbidities of double diabetes**. *Diabetes Res Clin Pract* (2016.0) **119** 48-56. DOI: 10.1016/j.diabres.2016.06.003
4. Kietsiriroje N, Pearson S, Campbell M, Ariens RAS, Ajjan RA. **Double diabetes: a distinct high-risk group?**. *Diabetes Obes Metab* (2019.0) **21** 2609-2618. DOI: 10.1111/dom.13848
5. Alzahrani SH, Ajjan RA. **Coagulation and fibrinolysis in diabetes**. *Diab Vasc Dis Res* (2010.0) **7** 260-273. DOI: 10.1177/1479164110383723
6. Kietsiriroje N, Ariens RAS, Ajjan RA. **Fibrinolysis in acute and chronic cardiovascular disease**. *Semin Thromb Hemost* (2021.0) **47** 490-505. DOI: 10.1055/s-0040-1718923
7. Wali JA, Solon-Biet SM, Freire T, Brandon AE. **Macronutrient determinants of obesity, insulin resistance and metabolic health**. *Biology* (2021.0) **10** 1-27. DOI: 10.3390/biology10040336
8. Ahola AJ, Harjutsalo V, Thorn LM, Freese R, Forsblom C, Makimattila S. **The association between macronutrient intake and the metabolic syndrome and its components in type 1 diabetes**. *Br J Nutr* (2017.0) **117** 450-456. DOI: 10.1017/S0007114517000198
9. Bell KJ, Barclay AW, Petocz P, Colagiuri S, Brand-Miller JC. **Efficacy of carbohydrate counting in type 1 diabetes: a systematic review and meta-analysis**. *Lancet Diabetes Endocrinol* (2014.0) **2** 133-140. DOI: 10.1016/S2213-8587(13)70144-X
10. Snell-Bergeon JK, Chartier-Logan C, Maahs DM, Ogden LG, Hokanson JE, Kinney GL. **Adults with type 1 diabetes eat a high-fat atherogenic diet that is associated with coronary artery calcium**. *Diabetologia* (2009.0) **52** 801-809. DOI: 10.1007/s00125-009-1280-4
11. Campbell MD, Walker M, Ajjan RA, Birch KM, Gonzalez JT, West DJ. **An additional bolus of rapid-acting insulin to normalise postprandial cardiovascular risk factors following a high-carbohydrate high-fat meal in patients with type 1 diabetes: a randomised controlled trial**. *Diab Vasc Dis Res* (2017.0) **14** 336-344. DOI: 10.1177/1479164117698918
12. O'Mahoney LL, Dunseath G, Churm R, Holmes M, Boesch C, Stavropoulos-Kalinoglou A. **Omega-3 polyunsaturated fatty acid supplementation versus placebo on vascular health, glycaemic control, and metabolic parameters in people with type 1 diabetes: a randomised controlled preliminary trial**. *Cardiovasc Diabetol* (2020.0) **19** 127. DOI: 10.1186/s12933-020-01094-5
13. Roe L, Strong C, Whiteside C, Neil A, Mant D. **Dietary intervention in primary care: validity of the DINE method for diet assessment**. *Fam Pract* (1994.0) **11** 375-381. DOI: 10.1093/fampra/11.4.375
14. Bingham SA, Gill C, Welch A, Day K, Cassidy A, Khaw KT. **Comparison of dietary assessment methods in nutritional epidemiology: weighed records v. 24 h recalls, food-frequency questionnaires and estimated-diet records**. *Br J Nutr* (1994.0) **72** 619-643. DOI: 10.1079/bjn19940064
15. Carter MC, Hancock N, Albar SA, Brown H, Greenwood DC, Hardie LJ. **Development of a new branded UK Food composition database for an online dietary assessment tool**. *Nutrients* (2016.0) **8** 480. DOI: 10.3390/nu8080480
16. O'Mahoney LL, Kietsiriroje N, Pearson S, West DJ, Holmes M, Ajjan RA. **Estimated glucose disposal rate as a candidate biomarker for thrombotic biomarkers in T1D: a pooled analysis**. *J Endocrinol Invest* (2021.0) **44** 2417-2426. DOI: 10.1007/s40618-021-01550-3
17. Nystrom T, Holzmann MJ, Eliasson B, Svensson AM, Sartipy U. **Estimated glucose disposal rate predicts mortality in adults with type 1 diabetes**. *Diabetes Obes Metab* (2018.0) **20** 556-563. DOI: 10.1111/dom.13110
18. Helmink MAG, de Vries M, Visseren FLJ, de Ranitz WL, de Valk HW, Westerink J. **Insulin resistance and risk of vascular events, interventions and mortality in type 1 diabetes**. *Eur J Endocrinol* (2021.0) **185** 831-840. DOI: 10.1530/EJE-21-0636
19. Yang Q, Lang X, Li W, Liang Y. **The effects of low-fat, high-carbohydrate diets vs. low-carbohydrate, high-fat diets on weight, blood pressure, serum liquids and blood glucose: a systematic review and meta-analysis**. *Eur J Clin Nutr* (2022.0) **76** 16-27. DOI: 10.1038/s41430-021-00927-0
20. de Wit N, Derrien M, Bosch-Vermeulen H, Oosterink E, Keshtkar S, Duval C. **Saturated fat stimulates obesity and hepatic steatosis and affects gut microbiota composition by an enhanced overflow of dietary fat to the distal intestine**. *Am J Physiol Gastrointest Liver Physiol* (2012.0) **303** G589-G599. DOI: 10.1152/ajpgi.00488.2011
21. Small L, Brandon AE, Turner N, Cooney GJ. **Modeling insulin resistance in rodents by alterations in diet: what have high-fat and high-calorie diets revealed?**. *Am J Physiol Endocrinol Metab* (2018.0) **314** E251-E265. DOI: 10.1152/ajpendo.00337.2017
22. Hart M, Pursey K, Smart C. **Low carbohydrate diets in eating disorders and type 1 diabetes**. *Clin Child Psychol Psychiatry* (2021.0) **26** 643-655. DOI: 10.1177/1359104520980778
23. Millen BE, Abrams S, Adams-Campbell L, Anderson CA, Brenna JT, Campbell WW. **The 2015 dietary guidelines advisory committee scientific report: development and major conclusions**. *Adv Nutr* (2016.0) **7** 438-444. DOI: 10.3945/an.116.012120
24. Dyson PA, Twenefour D, Breen C, Duncan A, Elvin E, Goff L. **Diabetes UK evidence-based nutrition guidelines for the prevention and management of diabetes**. *Diabetic Med* (2018.0) **35** 541-547. DOI: 10.1111/dme.13603
25. Astrup A, Magkos F, Bier DM, Brenna JT, de Oliveira Otto MC, Hill JO. **Saturated fats and health: a reassessment and proposal for food-based recommendations: JACC state-of-the-art review**. *J Am Coll Cardiol* (2020.0) **76** 844-857. DOI: 10.1016/j.jacc.2020.05.077
26. O'Mahoney LL, Churm R, Stavropoulos-Kalinoglou A, Ajjan RA, Orsi NM, Mappa G. **Associations between erythrocyte membrane fatty acid compositions and biomarkers of vascular health in adults with type 1 diabetes with and without insulin resistance: a cross-sectional analysis**. *Can J Diabetes* (2021.0) **46** 111-117. DOI: 10.1016/j.jcjd.2021.06.005
27. Howard E, Attenbourgh A, O'Mahoney LL, Sakar A, Ke L, Campbell MD. **Postprandial vascular-inflammatory and thrombotic responses to high-fat feeding are augmented by manipulating the lipid droplet size distribution**. *Nutr Metab Cardiovasc Dis* (2021.0) **31** 2716-2723. DOI: 10.1016/j.numecd.2021.05.021
28. Johnson AM, Olefsky JM. **The origins and drivers of insulin resistance**. *Cell* (2013.0) **152** 673-684. DOI: 10.1016/j.cell.2013.01.041
29. Santaren ID, Watkins SM, Liese AD, Wagenknecht LE, Rewers MJ, Haffner SM. **Individual serum saturated fatty acids and markers of chronic subclinical inflammation: the insulin resistance atherosclerosis study**. *J Lipid Res* (2017.0) **58** 2171-2179. DOI: 10.1194/jlr.P076836
30. Wu JHY, Micha R, Mozaffarian D. **Dietary fats and cardiometabolic disease: mechanisms and effects on risk factors and outcomes**. *Nat Rev Cardiol* (2019.0) **16** 581-601. DOI: 10.1038/s41569-019-0206-1
31. Gadgil MD, Appel LJ, Yeung E, Anderson CA, Sacks FM, Miller ER. **The effects of carbohydrate, unsaturated fat, and protein intake on measures of insulin sensitivity: results from the OmniHeart trial**. *Diabetes Care* (2013.0) **36** 1132-1137. DOI: 10.2337/dc12-0869
32. Qian F, Korat AA, Malik V, Hu FB. **Metabolic effects of monounsaturated fatty acid-enriched diets compared with carbohydrate or polyunsaturated fatty acid-enriched diets in patients with type 2 diabetes: a systematic review and meta-analysis of randomized controlled trials**. *Diabetes Care* (2016.0) **39** 1448-1457. DOI: 10.2337/dc16-0513
33. Wang Z, Adair LS, Cai J, Gordon-Larsen P, Siega-Riz AM, Zhang B. **Diet quality is linked to insulin resistance among adults in China**. *J Nutr* (2017.0) **147** 2102-2108. DOI: 10.3945/jn.117.256180
34. Ho FK, Gray SR, Welsh P, Petermann-Rocha F, Foster H, Waddell H. **Associations of fat and carbohydrate intake with cardiovascular disease and mortality: prospective cohort study of UK Biobank participants**. *BMJ* (2020.0) **368** m688. DOI: 10.1136/bmj.m688
|
---
title: 'An investigation of low-protein diets’ qualification rates and an analysis
of their short-term effects for patients with CKD stages 3–5: a single-center retrospective
cohort study from China'
authors:
- Xian-long Zhang
- Min Zhang
- Nuo Lei
- Wen-wei Ouyang
- Hui-fen Chen
- Bei-ni Lao
- Yan-min Xu
- Fang Tang
- Li-zhe Fu
- Xu-sheng Liu
- Yi-fan Wu
journal: International Urology and Nephrology
year: 2022
pmcid: PMC10030416
doi: 10.1007/s11255-022-03390-3
license: CC BY 4.0
---
# An investigation of low-protein diets’ qualification rates and an analysis of their short-term effects for patients with CKD stages 3–5: a single-center retrospective cohort study from China
## Abstract
### Background
The feasibility and efficacy of low-protein diets (LPD) treatment in chronic kidney disease (CKD) is controversial. Based on the characteristics of the Chinese diet, we observe the qualification rates and short-term clinical effects of LPD for CKD patients in our center.
### Methods
This is a retrospective cohort study. CKD stages 3–5 patients who were regularly followed up 5 times (over 2 years) and treated with LPD were included. We collected clinical data to observe the changes in LPD qualification rates and divided patients into LPD and non-LPD group according to the average dietary protein intake (DPI) of 5 follow-up time points and compared the changes in primary and secondary outcome measures between the two groups.
### Results
We analyzed data from 161 eligible CKD stages 3–5 patients. From baseline to the 5th follow-up time point, the LPD qualification rates of all patients were $11.80\%$, $35.40\%$, $47.82\%$, $53.43\%$ and $54.04\%$, respectively. For primary outcome measures, the urine protein/creatinine ratio (UPCR) decreased more in the LPD group than in the non-LPD group [Median (interquartile range, IQR) of the difference between the 5th follow-up time point and baseline: 0.19 (− 0.01–0.73) vs. 0.10 (− 0.08–0.27), $P \leq 0.001$]. We constructed three classes of mixed linear models (model I, II, III). The UPCR slopes were all negative in the LPD group and positive in the non-LPD group ($P \leq 0.001$). Meanwhile, in model I, the estimate glomerular filtration rate(eGFR) decline slope in the LPD group was lower than that in the non-LPD group [slope (standard error): − 1.32 (0.37) vs. − 2.35 (0.33), $$P \leq 0.036$$]. For secondary outcome measures, body mass index (BMI) triglycerides (TG), body weight, and fat free mass (FFM) showed stable statistical differences in the comparison of LPD and non-LPD groups, with greater declines in the former.
### Conclusion
The results of this study suggest that LPD treatment can reduce UPCR in patients with CKD stages 3–5, and may also delay the decline in eGFR. Meanwhile, it also reduces BMI, TG, body weight, and FFM, thus the need to prevent malnutrition in clinical implementation.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s11255-022-03390-3.
## Background
Chronic kidney disease (CKD) has a global prevalence of around 8–$16\%$. It can diminish quality of life, and increase the risk of cardiovascular disease and mortality. As such, it has been recognized as a global public health problem [1, 2]. In recent years, nutrition management has been considered an important approach for conservative treatment for CKD patients, and accordingly, it is gaining attention [3, 4]. Effective nutritional treatment not only can ensure that CKD patients’ nutritional requirements are met, but also can reduce metabolite accumulation, therefore contributing to the control of uremic symptoms and other complications, such as electrolyte disturbances and acid–base imbalances, sodium and water retention, and mineral and bone disorder (CKD-MBD) syndrome [3].
Low-protein diets (LPD) have long been controversial in nutritional management. The initial results of the modification of diet in renal disease (MDRD) study showed that LPD had little benefit in the short-term, and did not delay CKD progression [5]. However, other scholars have arrived at different conclusions after reanalyzing the MDRD study, arguing that inconclusive evidence should not be mistaken for evidence supporting null hypotheses [6]. In 2011, the Clinical Practice Guidelines issued by the British Renal Society (BRS) recommended a protein intake of 0.75 g/kg ideal body weight (IBW)/day for CKD stage 4–5 non-dialysis patients [7]. In 2019, it was changed to 0.8–1.0 g/kg IBW/day, and the BRS declared that they believed there was insufficient evidence to recommend LPD therapy (1C) [8]. In contrast, the 2020 update of the National Kidney Foundation's Kidney Disease Outcomes Quality Initiative (KDOQI) Clinical Practice Guidelines for Nutrition in CKD pointed out that a protein-restricted diet could reduce the risk of end-stage renal disease (ESRD) or death (1A), and improve quality of life (2C). For patients with CKD stages 3–5 and without diabetes, LPD should be limited to 0.55–0.60 g dietary protein/kg body weight (BW)/day, and for patients with CKD stage 3–5 and diabetes, LPD should be limited to 0.60–0.80 g dietary protein/kg BW/day [9].
However, perhaps due to differences in dietary habits, LPD has been recommended for nutritional therapy for CKD in China since the 2005 Consensus on Protein Nutritional Therapy for Chronic Kidney Disease [10]. The Health Industry Standard of the People's Republic of China—Dietary Guide for Chronic Kidney Disease Patients WS/557–2017 promulgated in 2017 also provides detailed guidance on LPD [11]. The latest Clinical Practice Guidelines for Nutritional Therapy of Chronic Kidney Disease in China (2021 Edition) also still recommends protein diet restriction [12].
Another problem pertaining to LPD is that its implementation in clinical settings is difficult. Results from a US survey showed that average daily protein intakes were 1.3 g/kg/day [13]. Thus, to meet LPD treatment standards, it would be necessary to reduce the total protein intake by more than half, which in turn would affect patients' dietary satisfaction and motivation [14]. Coupled with patients’ varying education levels and diet patterns, the lack of effective monitoring and feedback mechanisms in medical institutions, as well as patients’ minimal understanding of nephropathy, insufficient doctor-patient communication, insufficient family support, and other reasons, LPD treatment compliance is often compromised [15]. Moreover, many nephrologists lack training and experience in LPD; they fear malnutrition and are unable to develop detailed LPD regimens. As such, they do not implement the LPD treatment [16].
Our center emphasizes LPD therapy for CKD, and has established a professional nutrition management team to oversee patients’ dietary intake, and to collect clinical data. The Chinese Nutrition Society established the current protein dietary reference intakes (DRIs) in China in 2013, which include the estimated averaged requirement (EAR) and recommended nutrient intake (RNI). The EAR for protein is 60 g/day for adult males and 50 g/day for adult females (0.9 g/kg/day). The RNI for protein is 65 g/day for adult males and 55 g/day for adult females (1.0 g/kg/day). A study showed that the total dietary protein intake of the Chinese population would vary with the seasons, with an average estimate of 68.48 ± 22.07 g/day [17, 18]. However, due to the specificity of the Chinese diet, which consists of primarily mixed foods, there are few international reports on LPD in Chinese CKD patients. With this in mind, how is LPD adherence among CKD patients with Chinese dietary habits? What effect does LPD have on CKD prognosis? Can restricting dietary protein affect nutritional status? We conducted a retrospective study to observe the LPD qualification rates among patients with CKD stages 3–5, based on our data. We then analyzed the short-term effects of LPD on CKD progression.
## Study setting and study design
This is a retrospective cohort study. It was conducted in the Chronic Disease Management Center at the Nephrology Department of Guangdong Provincial Hospital of Chinese Medicine, Guangzhou, Guangdong Province, China. This study has been registered on the Chinese Clinical Trial Registry, registration number: ChiCTR1900024633.
## Diagnostic and staging criteria
CKD diagnosis and clinical staging was based on the KDOQI clinical practice guidelines [19]. We estimated GFR using the Chronic Kidney Disease Epidemiology Collaboration equation (CKD-EPI 2009 Serum creatinine) [20].
## Inclusion and exclusion criteria
The inclusion criteria were: (I) patients aged 18 years and older; (II) estimated (e)GFR < 60 mL/min/1.73 m2; (III) patients who had signed written informed consent to receive self-management and were willing to share clinical data; (IV) results of routine blood tests, routine urine tests, biochemical tests, urinary protein/creatinine ratio (UPCR), 24-h urine analysis, and body composition analysis needed to have been obtained every 6 months over the previous 2 years; The exclusion criteria were: (I) patients with psychiatric illness or other reasons which prevented their cooperation; (II) patients undergoing renal replacement therapy; (III) Patients with acute, subacute or chronic inflammatory conditions; (IV) patients with cancer; (V) patients who are ingesting steroids.
## Dietary restriction of protein intake
This cohort study’s nutrition management team consisted of 2 nephrologists, 3 nurses, 1 registered dietitian, and several clinical graduate students. After patients had established a complete record file at the chronic disease management center, physicians would prescribe nutritional prescriptions, and nurses would provide dietary education according to the patients’ disease conditions.
For detailed diet management steps, see WS/557–2017 [11]: Step 1: Calculate energy intake and food exchange portion. WS/557–2017 uses the Broca-*Katsura formula* (Katsura method) to calculate the standard body weight (SBW) for Asians and recommends daily caloric and protein intakes for patients with CKD based on the SBW [21, 22], (male) SBW = (height cm-100) *0.9 (kg); (female) SBW = (height cm-100) *0.9 (kg) -2.5 (kg); for patients with CKD stages 3–5, energy intake needed to be maintained at 35 kcal/kg SBW/day (age ≤ 60 years) or 30 kcal to 35 kcal/kg SBW/day (age > 60 years).
Step 2: Calculate protein intake based on the nephropathy food exchange portion in China. This method, based on the renal exchange list, is a practical tool for dietary planning and facilitates nutritional management of kidney disease in developing countries or regions [23, 24]. It classifies foods into 3 levels according to their protein content: 0–1 g, 4 g, and 7 g. 0–1 g foods included oils, starches, melon vegetables, and some fruits; 4 g foods included grains and yams, green leafy vegetables; 7 g foods included beans, meat, eggs and dairy. Every food portion had a unique weight, but all could provide calories in 90 kcal or multiples of 90 kcal (with the exception of vegetables). The total calories that a patient needed to consume per day divided by 90 kcal was how many food portions that patient needed per day. Not only was this method conducive to rapid LPD calculation, but it also ensured sufficient energy intake.
Step 3: Allocate food. At least $50\%$ of total protein intake needed to be high-quality protein. First, allocate the high-quality protein foods (7 g foods) to guarantee the proportion of high-quality protein, and then allocate the non-high-quality protein food (4 g foods) and consider the rationality of food pairings at the same time. The remaining calories could be provided by the low-protein foods (0–1 g foods) since they contained little or no protein.
Step 4: Develop specific recipes. Pair specific foods according to the results of the above food exchange portion, and according to taste.
## Personalized adjustment
In actual application, the above steps would be fine-tuned based on age, weight, physical activity type, comorbidities, stress conditions, and the specific protein content of the food. As for the intake of minerals and vitamins, we would select appropriate foods according to patient needs. We measured the 24-h urinary urea excretion to assess the patients’ actual protein intake. A professional dietitian from the team would handle any cases with serious conditions or complex nutritional management needs, and nurses would give ordinary patients routine nutritional guidance.
## Assessing nutritional status
We followed up with patients recruited for the study at least every 6 months. At each follow-up, patients would be required to fill out the Subjective Global Assessment (SGA) scale and perform laboratory tests including serum albumin (ALB), triglycerides (TG), total cholesterol (TC), hemoglobin (Hb), and body composition analysis based on bioelectrical impedance.
## Grouping
We measured the 24-h urinary urea excretion to calculate the normalized protein equivalent of nitrogen appearance rate (nPNA) according to the Maroni-Mitch formula. In this way, we could evaluate the patients’ actual protein intake [25]. Next, we calculated the patients’ dietary protein intake (DPI), and then we averaged the DPI of the 5 follow-up time points within 2 years. According to WS/557–2017, we considered patients with DPI < 0.8 g/kg SBW/day qualified, and included them in the LDP group. Otherwise, they were included in the non-LDP group. The specific indicators we used to observe LPD’s effect on kidney-related indicators were as follows.
## Primary and secondary outcome measures
This study’s primary outcome measures were UPCR and estimated glomerular filtration rate (eGFR), where UPCR is a spot (random) urine protein creatinine ratio (P/C ratio), an alternative, rapid and simple method for detecting and estimating quantitative proteinuria assessment with good reliability in a wide range of disease states [26, 27]. The secondary outcome measures were serum creatinine (SCR), blood urea (UREA), serum uric acid (UA), carbon dioxide combining power (CO2CP), TG, TC, ALB, Hb and the related indicators of nutrition assessment as measured by a Body Composition Analyzer (Ver. LookinBody120, InBody, South Korea). This included body mass index (BMI), total body water (TBW), extracellular water ratio (EWR), waist-hip ratio (WHR), fat free mass (FFM), arm circumference (AC) and arm muscle circumference (AMC).
We performed all laboratory testing in Guangdong Provincial Hospital of Chinese Medicine’s laboratory, and collected the data from the hospital’s Hospital Information System (HIS).
## Statistical analysis
Continuous variables conforming to a normal distribution are presented as mean ± standard deviation (SD), and we analyzed them with a t-test. Continuous variables not conforming to the normal distribution are presented as medians (interquartile range, IQR), and we compared them using a Mann–Whitney U-test. Categorical variables are presented as frequencies (percentage), and we compared them using either a chi-square test or Fisher’s exact test. We used mixed linear model (MLM) for repeated measures to calculate the slope changes in the two groups’ indicators. We constructed three classes of random intercepts and random slope models (I, II, III). Then, we entered group, time, and the interaction between group and time into Model I. Model II adjusted the demographic data based on Model I [including age, sex, height, weight, SBW, BMI, education level, retirement status, marital status, comorbidities, systolic blood pressure (SBP) and diastolic blood pressure (DBP)]. We assessed comorbidities using the Charlson comorbidity index, which could predict the risk of death in patients with CKD [28]. Model III adjusted for all covariates in Model II and for body composition analysis indicators (including TBW, EWR, BMI, WHR, FFM, AC and AMC), DPI, dietary energy intake (based on a three-day diet diary) and laboratory test indicators (including SCR, UREA, UA, CO2CP, eGFR, TG, TC, ALB, Hb and UPCR). We used Little's test to validate whether the missing data was a random sample of the total data. We determined that the missing data in this study were indeed randomly missing, and that the missing rate was below $10\%$. The missing values can be supplemented automatically when using mixed linear model analysis [29]. The study’s significance level was set at P ≤ 0.05, and all data was analyzed using SPSS 22.0 (IBM Corp., Armonk, NY, USA).
## Ethical issues
All patients included in the study signed an informed consent that allowed their clinical data to be used for medical research. Additionally, this study was approved by the Ethics Committee at Guangdong Provincial Hospital of Chinese Medicine (Approval notice: AF/04-$\frac{06.1}{10.0}$, 4, July 2019, ZF2019-153-01).
## Patient selection and grouping results
A total of 218 patients with CKD stages 3–5 were enrolled from October 01, 2019 to October 31, 2021, and each patient was followed up at least every 6 months. We retrospectively collected clinical data for these 218 patients at 5 follow-up points (2 years). 204 patients completed CKD routine laboratory tests and 189 patients had the results of 24-h urine analysis; 161 patients received regular nutritional status assessment via body composition analyzers. Therefore, 161 patients were eligible for inclusion in the final analysis. Finally, according to the grouping criteria, we included 69 patients from the study in the LPD group, and 92 patients in the non-LPD group.
## Baseline data after grouping
We compared the two groups’ baseline characteristics. Most baseline characteristics showed no significant differences in baseline characteristics. This included demographics, laboratory tests, and body composition analysis between the two groups. However, there were statistically significant differences in height, SBW, primary disease, and DPI between the 2 groups (Tables 1 and 2). The LPD group had higher height, higher SBW, and lower DPI than the non-LPD group. Additionally, patients in the LPD group had a greater proportion of primary glomerulopathy, while patients in the non-LPD group had a greater proportion of diabetic kidney disease and unknown primary disease. In the MLM, we adjusted any of the above variables which had statistically significant differences. Table 1Baseline characteristics of the demographic dataBaseline characteristicsLPD groupn = 69Non-LPD groupn = 92P valueAge (year)59.00 (45.50–57.00)60.00 (51.00–68.00)0.270*Sex0.408† Male42 (60.90)50 (54.30) Female29 (39.10)42 (45.70)Height (cm)164.14 ± 7.85160.57 ± 8.790.008**Weight (kg)60.22 ± 10.2459.82 ± 10.700.809**SBW (kg)56.75 ± 7.9853.37 ± 8.910.014**BMI (kg/m2)22.85 (20.62–24.22)23.35 (20.55–25.08)0.219*SBP (mm/Hg)127.00 (116.50–136.00)126.00 (118.00–134.00)0.856*DBP (mm/Hg)74.00 (68.00–81.50)73.00 (66.00–80.00)0.607*CCI4.00 (2.00–5.00)4.00 (3.00–5.00)0.333*CKD stage0.321† Stage 339 (56.50)60 (65.20) Stage 419 (27.50)24 (24.60) Stage 511 (16.00)8 (8.70)Education level0.691‡ Primary or below8 (11.60)12 (13.00) Junior high school15 (21.70)26 (28.30) High school or polytechnic school21 (30.40)28 (30.40) University or junior college25 (36.20)25 (27.20) Postgraduate01 (1.10)Employment status0.646† Retired31 (44.90)38 (41.30) Not retired38 (55.10)54 (58.70)Marital status0.499‡ Married64 (92.80)88 (95.70) Unmarried5 (7.20)4 (4.30)Primary disease0.010‡ Primary glomerulopathy29 (42.00)22 (23.90) Interstitial nephritis02 (2.20) Autoimmune disease4 (5.80)1 (1.10) Diabetic kidney disease5 (7.20)18 (19.60) Polycystic kidney disease6 (8.70)4 (4.30) Nephrosclerosis19 (27.50)27 (19.30) Obstructive nephropathy2 (2.90)3 (3.30) Unknown4 (5.80)15 (16.30)Values are given as n (%), mean ± standard deviation (SD), or median (interquartile range, IQR). P ≤ 0.05 was considered statistically significantSBW standard body weight, BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, CCI Charlson Comorbidity Index, CKD chronic kidney disease*Mann–Whitney U-test; **t-test; †Chi-square test; ‡Fisher’s exact testTable 2Baseline characteristics of laboratory tests, dietary protein intake and body compositionBaseline characteristicsLPD groupn = 69Non-LPD groupn = 92P valueSCR (μmol/L)171.00 (127.00–282.75)158.50 (120.25–216.00)0.217*Urea (μmol/L)10.56 (7.63–13.58)9.57 (7.91–12.00)0.296*UA (μmol/L)423.54 ± 81.96437.92 ± 80.670.270**CO2CP (mmol/L)22.84 ± 3.2023.36 ± 2.720.272**TG (mmol/L)1.48 (1.10, 2.14)1.49 (1.05, 1.93)0.538*TC (mmol/L)4.88 ± 1.144.98 ± 1.240.598**eGFR (ml/min/1.73m2)35.49 (19.78–48.08)37.61 (24.58–46.93)0.483*Hb (g/L)125.13 ± 19.17124.80 ± 18.160.912**ALB (g/L)44.91 ± 3.7144.71 ± 3.470.726**UPCR (g/g)1.02 (0.22–2.02)0.56 (0.15–1.52)0.082*DPI (g/kg SBW/day)0.95 (0.83–1.13)1.08 (0.91–1.23)0.009*TBW (kg)33.10 (27.30–37.85)31.60 (26.65–36.45)0.228*WHR0.85 ± 0.050.85 ± 0.050.965**FFM (kg)44.85 (37.10–50.98)43.00 (36.30–49.40)0.233*EWR0.39 (0.38–0.39)0.39 (0.38–0.39)0.150*AC (cm)28.16 ± 2.6328.89 ± 2.900.109**AMC (cm)24.94 ± 2.4125.21 ± 2.330.474**Values are given as n (%), mean ± standard deviation (SD), or median (interquartile range, IQR). P ≤ 0.05 was considered statistically significant*Mann–Whitney U-test; **t-testSCR serum creatinine, UA serum uric acid, CO2CP carbon dioxide combining power, TG triglycerides, TC serum total cholesterol, eGFR estimated glomerular filtration rate, Hb hemoglobin, ALB albumin, UPCR urinary protein/creatinine ratio, DPI dietary protein intake, TBW total body water, WHR waist-hip ratio, FFM fat free mass, EWR extracellular water ratio, AC arm circumference, AMC arm muscle circumference
## DPI and its qualification rate
Compared with the baseline, most patients’ DPI declined gradually from the second follow-up time point onward, and the qualification rate also gradually increased (Figs. 1 and 2). The median DPI for all patients decreased from 0.99 (0.90–1.21) g/kg SBW/day to 0.77 (0.67–0.95) g/kg SBW/day from baseline to the fourth follow-up time point, with only a slight rebound at the fifth follow-up time point. The median DPI for the fifth follow-up time point was 0.79 (0.66–0.96) g/kg SBW/day. The DPI qualification rate showed a trend of gradual improvement. The DPI qualification rate was $11.80\%$ at baseline and $54.40\%$ at the fifth follow-up point. Meanwhile, DPI in the two patient groups gradually decreased along ensuing follow-up time points. However, DPI decreased more in the LPD group than in the non-LPD group [0.95 (0.83–1.13) to 0.66 (0.61, 0.73) vs. 1.08 (0.93–1.23) to 0.92 (0.82–1.13)]. Among patients with qualified DPI intake at each follow-up time point, the LPD group accounted for a greater proportion than the non-LPD group (Table 3 and Supplementary material, Table 1).Fig. 1Dietary protein intake for 161 patients at 5 follow-up time points. DPI dietary protein intake, SBW standard body weight. Each boxplot contains 5 nodes for each dataset, from top to bottom are the maximum value, the 75th percentile, the median, the 25th percentile, and the minimum value. All 161 patients received continuous nutrition education. Most patients had their highest DPI at baseline, and gradually decreased on the 2nd–4th follow-up time points. Compared with the 4th follow-up time point, some patients’ DPI at the 5th follow-up time point increased slightlyFig. 2Qualified dietary protein intake rate among 161 patients at 5 follow-up time points. DPI dietary protein intake. From baseline to the 5th follow-up time point, the 161 patients’ DPI qualification rate gradually increased and the disqualification rate gradually decreased. The highest increase in qualification rate was at the second follow-up time point, while the lowest increase in qualification rate was at the 5th follow-up time pointTable 3DPI at 5 follow-up time pointsFollow-up time pointDPIn = 161DPI (LPD group)$$n = 69$$DPI (Non-LPD group)$$n = 92$$Baseline time point0.99 (0.90–1.21)0.95 (0.83–1.13)1.08 (0.93–1.23)Second follow-up time point0.87 (0.75–1.05)0.74 (0.63–0.80)1.03 (0.89–1.22)Third follow-up time point0.81 (0.71–1.01)0.69 (0.62–0.77)0.97 (0.83–1.10)Fourth follow-up time point0.77 (0.67–0.95)0.66 (0.62–0.74)0.93 (0.80–1.12)Fifth follow-up time point0.79 (0.66–0.96)0.66 (0.61–0.73)0.92 (0.82–1.13)Values are given as medians (interquartile range, IQR)DPI dietary protein intake
## Primary outcomes
There was no occurrence of renal replacement therapy in this study, from baseline to the 5th follow-up time point. For the primary outcome measures, at the 5th follow-up point, the UPCR of the LPD group was lower than that of the non-LPD group [0.52 (0.23–0.97) vs. 0.95 (0.22–2.00), $$P \leq 0.050$$], and from baseline to the 5th follow-up time point, UPCR decline in the LPD group exceeded that of the non-LPD group [0.19 (− 0.01–0.73) vs. − 0.11 (− 0.05–0.03), $P \leq 0.001$]. Meanwhile, there was no statistically significant difference in eGFR between the two groups at the fifth follow-up time point. There was also no statistically significant difference between the two groups in the difference in eGFR between the baseline and the 5th follow-up time point (Supplementary material, Table 2).
The data obtained in this study were had been measured repeatedly. We used a mixed linear model to compare the differences with regards to various indicators’ trends between the two groups. In this study, we set various indicators (BMI, SBP, DBP, Hb, PCR, ALB, UREA, SCR, CO2CP, UA, TG, TC, eGFR, TBW, WHR, FFM, EWR, AC and AMC) as outcome variables, and estimated the varied intercept differences and varied slope differences of various outcome variables over time. Therefore, according to the factors and covariates, 3 classes of models (I, II, and III) were fitted, and models II and III adjusted for factors that had statistically significant differences at baseline. The models’ accuracy could be determined by comparing the − 2log-likelihood. The smaller the value, the more errors were explained, the more reliable the estimated results, and the more accurate the models; the − 2log-likelihoods of most indicators in Model III were superior to those of Model II; Model II also was superior to Model I (supplementary materials). The differences in the variation trend for each indicator could be determined by comparing the slope of the curve. The positive and negative values of the slope indicated the upward and downward varied trend for each indicator, while the magnitude of the slope indicated the degree of the varied trend.
In all three MLMs, the UPCR slopes in the LPD group were negative, presenting an overall downward trend; the UPCR slopes in the non-LPD group were all positive, presenting an overall upward trend [Model I: Slope (standard error, SE): − 0.21 (0.05) vs. 0.17 (0.04), $P \leq 0.001.$ Model II: − 0.21 (0.05) vs. 0.14 (0.05), $P \leq 0.001.$ Model III: − 0.32 (0.06) vs. 0.003 (0.05), $P \leq 0.001$]. In Model I, there was a statistically significant difference in slope between the two groups, and the decrease in eGFR in the LPD group was less than that in the non-LPD group [− 1.32 (0.37) vs. − 2.35 (0.33), $$P \leq 0.036$$]. In Models II and III, there was no statistically significant difference in slope between the two groups (Supplementary material, Table 2). The data distribution trends of the primary outcome measures for the 5 follow-up time points are shown in Fig. 3.Fig. 3Boxplot of primary outcome measures at 5 follow-up time points. In the boxplot, some data are reflected as outliers. The circles in the middle of the box with the gray fill color are the averages, while the short red underlined lines connect the medians. The $75\%$th and $25\%$th quartiles of the five box data are connected by gray line segments, and they form a trend graph over time with the median connecting line. In the primary outcome measures, the urinary protein/creatinine ratio gradually decreased in the LDP group, while the opposite was true for the Non-LPD group. Estimated glomerular filtration rate also showed an opposite trend in both groups, but it was not significant enough
## Secondary outcomes
We compared secondary outcome measures in the LPD and Non-LPD groups, and the trends in the data at the 5 follow-up time points are shown in Supplemental Figs. 1–4. Specific data comparing the two groups are shown in Supplemental Tables 3–6. In terms of renal function, there were no significant differences in SCR, urea, UA or CO2CP between the two groups. In terms of lipids, TG and TC were lower in the LPD group than in the non-LPD group, but only TG had a statistically difference between the two groups, as well as in the comparison of Model I and Model II [difference comparison: 0.11 (− 0.29–0.50) vs. − 0.09 (− 0.55–0.28), $$P \leq 0.035.$$ Model I: − 0.12 (0.05) vs. 0.09 (0.05), $$P \leq 0.006.$$ Model II: − 0.09 (0.06) vs. 0.08 (0.05), $$P \leq 0.024$$]. Comparison of body composition analysis between the two groups showed that in terms of body weight, the LPD group had lower body weight than the non-LPD group, and the LPD group lost more weight compared to the baseline [difference comparison: 1.69 (− 0.41–4.26) vs. 0.69 (− 0.96–2.55), $$P \leq 0.050$$]. The BMI of the two patient groups showed a downward trend during the follow-up period, while BMI showed an even greater decrease in the LPD group. [ 5th follow-up time point: 21.65 (19.58–23.30) vs. 22.40 (20.45–24.35)), $$P \leq 0.035.$$ Difference comparison: 0.75 (− 0.07–1.57) vs. 0.27 (− 0.39–0.95), $$P \leq 0.041.$$ Model I: − 0.32 (0.06) vs. − 0.15 (0.05), $$P \leq 0.032.$$ Model II: − 0.17 (0.04) vs. − 0.04 (0.03), $$P \leq 0.013.$$ Model III: − 0.17 (0.05) vs. − 0.04 (0.04), $$P \leq 0.032$$].
The TBW loss in the LPD group was greater than that in the non-LPD group, but the observed change trend was unstable when we used the MLM to evaluate it [Difference comparison: 0.55 (0–1.28) vs. 0 (− 0.60–0.90), $$P \leq 0.026.$$ Model I: − 0.25 (0.07) vs. − 0.06 (0.06), $$P \leq 0.039.$$ Model II: − 0.14 (0.07) vs. 0.06 (0.06), $$P \leq 0.013.$$ Model III: 0.02 (0.005) vs. − 0.001 (0.004) $$P \leq 0.003$$]. During the follow-up period, there was a statistically significant difference between the two patient groups’ WHR slopes, but because the slope value was too small, this might not have had clinical significance [Model I: 0.001 (0.002) vs. − 0.004 (0.001), $$P \leq 0.026.$$ Model II: 0.003 (0.002) vs. − 0.003 (0.002), $$P \leq 0.007$$]. The decrease in FFM in the LPD group was greater than that in the non-LPD group, and all three classes models suggested a downward trend in the LPD group during the follow-up period, while in Models II and III, the non-LPD group showed an upward trend [difference: 0.75 (0–1.78) vs. 0.10 (− 0.80–1.20), $$P \leq 0.018.$$ Model I: − 0.36 (0.10) vs. − 0.07 (0.08), $$P \leq 0.025.$$ Model II: − 0.20 (0.09) vs. 0.10 (0.08), $$P \leq 0.008.$$ Model III: − 0.02 (0.007) vs. 0.003 (0.006), $$P \leq 0.003$$]. In addition, the output from Model III suggested that the changing Hb trend in the LPD group might have been an improvement over that of the non-LPD group [2.15 (0.70) vs. 0.43 (0.59), $$P \leq 0.035$$].
## Discussion
In this study, we observed the qualification rate and short-term clinical benefits of LPD for patients at our center with stages 3–5 CKD.
The results showed that the LPD qualification rate was only about $50\%$ at the 5th follow-up time point (the 24th month), which is similar to the results from studies in other countries. An Italian cross-sectional study showed that adherence to dietary prescriptions in children and adolescents was $56\%$ [30]. Another retrospective study from Brazil included 321 non-dialysis CKD patients, 158 of whom had adhered to LPD, and the adherence was $49\%$ [31]. There are many reasons for the low LPD qualification rate. In this study, the proportion of diabetic kidney disease patients in the non-LPD group ($19.6\%$) exceeded that in the LPD group ($7.2\%$) (Table 2). Previous studies have shown that implementing LPD is more complicated when patients have diabetic nephropathy. Blood glucose management is more difficult for them as well, and their adherence to management is lower. Therefore, this may be one of the reasons for the low LPD qualification rate in the non-LPD group [32, 33].
Meanwhile, according to the characteristics of the patients in the center, we considered that the reasons for the low qualification LPD rate may have included: [1] The LPD implementation process was indeed cumbersome; [2] The patients in our study were middle-aged or elderly, and the dietary structure was difficult to change; [3] cooking methods and food-pairing for Chinese cuisine is rich and varied, and it is not easy for patients to follow an LPD alone when they are accustomed to eating with family. Therefore, clinicians could improve CKD patients’ low LPD qualification rates in two ways. First, for patients of means, low-protein or protein-free products could be recommended, and energy supplements and keto acid preparations should be taken as needed [34, 35]. Second, personalized nutrition prescriptions could be formulated based on patients' dietary habits and disease conditions. In China, low-protein foods are recommended for CKD patients, including wheat starches, sweet potato starches, cassava starches, mung beans, pea starches and products derived from them; recipes could also be recommended according to patients’ unique situations, to diversify diets. This would not only meet patients’ tastes, but also increase the diversity of patient choices, thereby improving their compliance with LPD [11]. However, the ideal situation would be for patients need to keep in regular communication with the clinical team. Throughout chronic disease management, patients’ dietary habits and dietary structure would gradually approach the LPD standards due to continuous precise nutritional guidance and dietary education. This could improve the compliance of patients with CKD stages 3–5 to LPD [36]. In addition, there is still a lack of reports on whether the nutrient intake qualification rate for patients receiving dietary guidance has continued to improve, or has reversed [37]. The results of this study showed that the DPI qualification rate of 161 patients increased from $11.80\%$ at baseline to $54.40\%$ at the 5th follow-up time point. Therefore, we believe that although the dietary structure in *China is* complex, LPD is still feasible, and also would optimize the diet of CKD patients through better management.
The results of this study also indicate that LPD could offer short-term benefits for patients with CKD stages 3–5. We performed statistical analysis using a parametric test, a non-parametric test, and a mixed linear model. The primary outcome measure results showed that LPD could reduce urinary protein excretion, reduce UPCR in CKD stages 3–5 patients, and may also slow eGFR decline. The secondary outcome measures showed that LPD reduced TG and decreased body weight, BMI and FFM; it may also have an effect on WHR, TBW and Hb.
CKD patients in Japan, another country with East Asian cultural roots, have also been treated with LPD under comprehensive management. Based on the pooled results of meta-analyses from several clinical studies, Japanese scholars also believed that LPD could reduce urinary protein excretion, protect renal function and alleviate subjective symptoms [38]. The mechanism through which LPD reduces urine protein excretion may be a factor in inhibiting the renin-angiotensin system (RAS) [39]. In clinical practice, the application of angiotensin-converting enzyme inhibitors (ACE-Is) or angiotensin receptor blockers (ARBs) reduces proteinuria, and delays CKD progression in patients with large amounts of proteinuria. Studies have also shown that the overexpression of transforming growth factor-β (TGF-β) influences CKD progression and proteinuria occurrence, and that the combined treatment of ACEI/ARB and LPD reduces TGF-βexpression [40, 41].
The results of this study also suggest that LPD both causes weight loss, and lowers BMI and TG. Obesity and hyperlipidemia are the most prevalent independent risk factors for CKD [42]. Meanwhile, high BMI and lipid metabolism disorders were both risk factors for CKD co-occurring with cardiovascular disease [43]. A previous study has revealed that kidney disease progression is correlated with obesity and high BMI (> 30 kg/m2), and has a U-shaped relationship with mortality [44]. In addition to body weight, a cross-sectional study in China has also shown that hypertriglyceridemia is correlated with increased urinary albumin-to-creatinine ratios (ACR) [45]. The results of the meta-analysis also showed that LPD could reduce proteinuria, cause weight loss, adjust lipid metabolism, and reduce the onset of azotemia, thereby delaying the patient’s time to end-stage renal disease [46, 47]. It is important to note the “obesity paradox” in CKD patients: a study has shown that obesity is correlated with a lower risk of death in patients with CKD, especially ESRD [48]. This is because obesity may represent the body's nutritional reserves, and studies have suggested that adipose tissue may help chelate uremia toxins in CKD patients [49]. Therefore, in nutrient management, we need to distinguish between the pathological state of abnormal lipid accumulation and the good nutritional state of body fullness, according to patients’ varying conditions, and differentiate between the effects of lipid packing, dyslipidemia, and obesity on patients during different periods. We also must avoid abnormal lipid distribution, lipid loss, and muscle consumption, and provide the most beneficial nutritional guidance for patients [50, 51].
LPD is a safe treatment for CKD patients; it causes few adverse reactions, and few studies have reported protein-energy waste (PEW) during LPD [46, 52]. There were no cases of PEW in this study, which may be related to the absence of patients included in hemodialysis or peritoneal dialysis. In addition, we effectively avoided PEW by several methods, including (I) prescribing nutritional supplements such as keto acids to appropriate patients; (II) recommending low-protein staple foods to increase dietary energy intake; (III) providing ongoing nutritional counseling to optimize dietary nutritional intake; (IV) enhancing management of chronic kidney disease comorbidities; and (V) instructing patients to perform appropriate exercise. Such LPD treatment also does not increase the phosphorus burden, and the data at the beginning and end of this study showed that the median serum phosphorus of patients was about 1.2 mmol/L. At the 5th follow-up time point, we observed no statistically significant differences between ALB and other nutritional monitoring indicators such as AMC and AC, between the two groups. However, the results of the comparison of the difference between the two patient groups and the MLM in this study suggested that during the LPD implementation, patients tended to decrease their body weight and FFM. Although the decline was not large, it still reminds us that we should regularly monitor patient nutrition, follow patients’ nutritional status, and adjust nutrition programs in real-time according to actual situations, so as to prevent lipid loss and muscle waste.
This study has its weaknesses. In addition to the common limitations of non-randomized methods, the follow-up period was not long enough. Moreover, this was a single-center retrospective study, and we did not investigate the specific reasons for LPD non-compliance. We were also unable to perform Kaplan-Mayer and *Cox analysis* of the primary outcome indicators because no patients had entered the endpoint event by the time data collection was complete. To ensure the integrity of the 24-h urinary urea excretion results and other data, the patients included in the study were screened. Therefore, the sample size of this cohort is small, the patients are younger, and there are fewer comorbidities, and their follow-up adherence may have been better than patients who were not included. The results and limitations of this study highlight the need for further study, and we hope to further verify the therapeutic effects of LPD with prospective and/or randomized studies in the future.
## Conclusion
The results of this study suggest that LPD treatment can reduce UPCR in patients with CKD stage 3–5, and may delay the decline in eGFR. Meanwhile, it also reduces BMI, TG, body weight, and FFM. We found no adverse reactions during LPD implementation, yet we still recommend regular nutritional status assessment to prevent PEW. However, with Chinese dietary habits, the LPD qualification rate for patients with CKD stages 3–5 is still low. Yet guidance and education offer room for additional improvement, and thus, further studies are needed to improve LPD qualification rates according to the characteristics of the Chinese diet.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 597 KB)
## References
1. Drawz P, Rahman M. **Chronic kidney disease**. *Ann Intern Med* (2015.0) **162** ITC1-ITC16. DOI: 10.7326/AITC201506020
2. Jha V, Garcia-Garcia G, Iseki K, Li Z, Naicker S, Plattner B. **Chronic kidney disease: global dimension and perspectives**. *Lancet* (2013.0) **382** 260-272. DOI: 10.1016/S0140-6736(13)60687-X
3. Kalantar-Zadeh K, Fouque D. **Nutritional management of chronic kidney disease**. *N Engl J Med* (2017.0) **377** 1765-1776. DOI: 10.1056/NEJMra1700312
4. Downer S, Berkowitz SA, Harlan TS, Olstad DL, Mozaffarian D. **Food is medicine: actions to integrate food and nutrition into healthcare**. *BMJ* (2020.0) **369** m2482. DOI: 10.1136/bmj.m2482
5. Klahr S, Levey AS, Beck GJ, Caggiula AW, Hunsicker L, Kusek JW. **The effects of dietary protein restriction and blood-pressure control on the progression of chronic renal disease. Modification of Diet in Renal Disease Study Group**. *N Engl J Med* (1994.0) **330** 877-884. DOI: 10.1056/NEJM199403313301301
6. Levey AS, Greene T, Beck GJ, Caggiula AW, Kusek JW, Hunsicker LG. **Dietary protein restriction and the progression of chronic renal disease: what have all of the results of the MDRD study shown? Modification of Diet in Renal Disease Study group**. *J Am Soc Nephrol* (1999.0) **10** 2426-2439. DOI: 10.1681/ASN.V10112426
7. Wright M, Jones C. **Renal Association Clinical Practice Guideline on nutrition in CKD**. *Nephron Clin Pract* (2011.0) **118** c153-c164. DOI: 10.1159/000328067
8. Wright M, Southcott E, MacLaughlin H, Wineberg S. **Clinical practice guideline on undernutrition in chronic kidney disease**. *BMC Nephrol* (2019.0) **20** 370. DOI: 10.1186/s12882-019-1530-8
9. Ikizler TA, Burrowes JD, Byham-Gray LD, Campbell KL, Carrero JJ, Chan W. **KDOQI Clinical Practice Guideline for Nutrition in CKD: 2020 Update**. *Am J Kidney Dis* (2020.0) **76** S1-S107. DOI: 10.1053/j.ajkd.2020.05.006
10. 10.Nephrology ECGotAoA-KAFi (2005) Consensus on protein nutrition therapy for chronic kidney disease. Chinese J Nephrol 421–424
11. 11.National Health and Family Planning Commission (2017) Hygiene standard. Dietary guide for chronic kidney disease patients. WS/557-2017[S]
12. 12.Chinese Nephrologist Association ECGoNTGotnCotCAoIM (2021) Clinical practice guidelines for the nutritional treatment of chronic kidney disease in China, 2021 edn. Natl Med J China. 101:539–559
13. Moore LW, Byham-Gray LD, Scott Parrott J, Rigassio-Radler D, Mandayam S, Jones SL. **The mean dietary protein intake at different stages of chronic kidney disease is higher than current guidelines**. *Kidney Int* (2013.0) **83** 724-732. DOI: 10.1038/ki.2012.420
14. Piccoli GB, Di Iorio BR, Chatrenet A, D'Alessandro C, Nazha M, Capizzi I. **Dietary satisfaction and quality of life in chronic kidney disease patients on low-protein diets: a multicentre study with long-term outcome data (TOrino-Pisa study)**. *Nephrol Dial Transplant* (2020.0) **35** 790-802. DOI: 10.1093/ndt/gfz147
15. Milas NC, Nowalk MP, Akpele L, Castaldo L, Coyne T, Doroshenko L. **Factors associated with adherence to the dietary protein intervention in the Modification of Diet in Renal Disease Study**. *J Am Diet Assoc* (1995.0) **95** 1295-1300. DOI: 10.1016/S0002-8223(95)00340-1
16. Kalantar-Zadeh K, Moore LW, Tortorici AR, Chou JA, St-Jules DE, Aoun A. **North American experience with Low protein diet for Non-dialysis-dependent chronic kidney disease**. *BMC Nephrol* (2016.0) **17** 90. DOI: 10.1186/s12882-016-0304-9
17. 17.Ouyang Y, Tan T, Song X, Huang F, Zhang B, Ding G et al (2021) Dietary protein intake dynamics in elderly Chinese from 1991 to 2018. Nutrients 13
18. Huang K, Yu D, Guo Q, Yang Y, Wei X, Zhao L. **Validation of the MSM and NCI method for estimating the usual intake of nutrients and food according to four seasons of seven consecutive daily 24 hour dietary recalls in Chinese adults**. *Nutrients* (2022.0) **14** 445. DOI: 10.3390/nu14030445
19. National KF. **K/DOQI clinical practice guidelines for chronic kidney disease: evaluation, classification, and stratification**. *Am J Kidney Dis* (2002.0) **39** S1-266. PMID: 11904577
20. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF, Feldman HI. **A new equation to estimate glomerular filtration rate**. *Ann Intern Med* (2009.0) **150** 604-612. DOI: 10.7326/0003-4819-150-9-200905050-00006
21. Urata H, Tahara Y, Nishiyama K, Fukuyama Y, Tsunawake N, Moji K. **Validity of various indices of obesity calculated from height and weight data for adult males use of the underwater-weighing method as a reference**. *Nihon Koshu Eisei Zasshi* (2001.0) **48** 560-567. PMID: 11524831
22. Nishida M, Funahashi T. **Validity of indices (BMI, Rohrer index, Broca method) for assessment of obesity**. *Nihon Rinsho* (2009.0) **67** 301-306. PMID: 19202903
23. Ziegler VS, Sucher KP, Downes NJ. **Southeast Asian renal exchange list**. *J Am Diet Assoc* (1989.0) **89** 85-92. DOI: 10.1016/S0002-8223(21)02067-8
24. Ameh OI, Cilliers L, Okpechi IG. **A practical approach to the nutritional management of chronic kidney disease patients in Cape Town, South Africa**. *BMC Nephrol* (2016.0) **17** 68. DOI: 10.1186/s12882-016-0297-4
25. Maroni BJ, Steinman TI, Mitch WE. **A method for estimating nitrogen intake of patients with chronic renal failure**. *Kidney Int* (1985.0) **27** 58-65. DOI: 10.1038/ki.1985.10
26. Chen YT, Hsu HJ, Hsu CK, Lee CC, Hsu KH, Sun CY. **Correlation between spot and 24h proteinuria: derivation and validation of equation to estimate daily proteinuria**. *PLoS ONE* (2019.0) **14** e0214614. DOI: 10.1371/journal.pone.0214614
27. Kaminska J, Dymicka-Piekarska V, Tomaszewska J, Matowicka-Karna J, Koper-Lenkiewicz OM. **Diagnostic utility of protein to creatinine ratio (P/C ratio) in spot urine sample within routine clinical practice**. *Crit Rev Clin Lab Sci* (2020.0) **57** 345-364. DOI: 10.1080/10408363.2020.1723487
28. Hemmelgarn BR, Manns BJ, Quan H, Ghali WA. **Adapting the Charlson Comorbidity Index for use in patients with ESRD**. *Am J Kidney Dis* (2003.0) **42** 125-132. DOI: 10.1016/S0272-6386(03)00415-3
29. Peters SA, Bots ML, den Ruijter HM, Palmer MK, Grobbee DE, Crouse JR. **Multiple imputation of missing repeated outcome measurements did not add to linear mixed-effects models**. *J Clin Epidemiol* (2012.0) **65** 686-695. DOI: 10.1016/j.jclinepi.2011.11.012
30. Cotugno G, Nicolo R, Cappelletti S, Goffredo BM, Dionisi Vici C, Di Ciommo V. **Adherence to diet and quality of life in patients with phenylketonuria**. *Acta Paediatr* (2011.0) **100** 1144-1149. DOI: 10.1111/j.1651-2227.2011.02227.x
31. Rizzetto F, Leal VO, Bastos LS, Fouque D, Mafra D. **Chronic kidney disease progression: a retrospective analysis of 3-year adherence to a low protein diet**. *Ren Fail* (2017.0) **39** 357-362. DOI: 10.1080/0886022X.2017.1282374
32. Betonico CC, Titan SM, Correa-Giannella ML, Nery M, Queiroz M. **Management of diabetes mellitus in individuals with chronic kidney disease: therapeutic perspectives and glycemic control**. *Clinics (Sao Paulo)* (2016.0) **71** 47-53. DOI: 10.6061/clinics/2016(01)08
33. Williams AF, Manias E, Walker R. **Adherence to multiple, prescribed medications in diabetic kidney disease: a qualitative study of consumers' and health professionals' perspectives**. *Int J Nurs Stud* (2008.0) **45** 1742-1756. DOI: 10.1016/j.ijnurstu.2008.07.002
34. D'Alessandro C, Rossi A, Innocenti M, Ricchiuti G, Bozzoli L, Sbragia G. **Dietary protein restriction for renal patients: don't forget protein-free foods**. *J Ren Nutr* (2013.0) **23** 367-371. DOI: 10.1053/j.jrn.2012.12.006
35. Cupisti A, Bolasco P. **Keto-analogues and essential aminoacids and other supplements in the conservative management of chronic kidney disease**. *Panminerva Med* (2017.0) **59** 149-156. DOI: 10.23736/S0031-0808.16.03288-2
36. Laddu D, Hauser M. **Addressing the nutritional phenotype through personalized nutrition for chronic disease prevention and management**. *Prog Cardiovasc Dis* (2019.0) **62** 9-14. DOI: 10.1016/j.pcad.2018.12.004
37. Palmer SC, Maggo JK, Campbell KL, Craig JC, Johnson DW, Sutanto B. **Dietary interventions for adults with chronic kidney disease**. *Cochrane Database Syst Rev* (2017.0) **4** CD011998. PMID: 28434208
38. Watanabe S. **Low-protein diet for the prevention of renal failure**. *Proc Jpn Acad Ser B Phys Biol Sci* (2017.0) **93** 1-9. DOI: 10.2183/pjab.93.001
39. Zhang JY, Yin Y, Ni L, Long Q, You L, Zhang Q. **Low-protein diet supplemented with ketoacids ameliorates proteinuria in 3/4 nephrectomised rats by directly inhibiting the intrarenal renin-angiotensin system**. *Br J Nutr* (2016.0) **116** 1491-1501. DOI: 10.1017/S0007114516003536
40. Peters H, Border WA, Noble NA. **Angiotensin II blockade and low-protein diet produce additive therapeutic effects in experimental glomerulonephritis**. *Kidney Int* (2000.0) **57** 1493-1501. DOI: 10.1046/j.1523-1755.2000.00994.x
41. Koppe L, Fouque D. **The role for protein restriction in addition to renin-angiotensin-aldosterone system inhibitors in the management of CKD**. *Am J Kidney Dis* (2019.0) **73** 248-257. DOI: 10.1053/j.ajkd.2018.06.016
42. Gai Z, Wang T, Visentin M, Kullak-Ublick GA, Fu X, Wang Z. **Lipid accumulation and chronic kidney disease**. *Nutrients* (2019.0) **11** 722. DOI: 10.3390/nu11040722
43. 43.Global Burden of Metabolic Risk Factors for Chronic Diseases C (2014) Cardiovascular disease, chronic kidney disease, and diabetes mortality burden of cardiometabolic risk factors from 1980 to 2010: a comparative risk assessment. Lancet Diabetes Endocrinol 2:634-647
44. Lu JL, Molnar MZ, Naseer A, Mikkelsen MK, Kalantar-Zadeh K, Kovesdy CP. **Association of age and BMI with kidney function and mortality: a cohort study**. *Lancet Diabetes Endocrinol* (2015.0) **3** 704-714. DOI: 10.1016/S2213-8587(15)00128-X
45. Su W, Wang J, Mu Y. **Association between hypertriglyceridemic waist phenotype and increased urinary albumin-creatinine ratio in Chinese adults: the REACTION study**. *Diabetes Metab Syndr Obes* (2020.0) **13** 2965-2974. DOI: 10.2147/DMSO.S257736
46. Rhee CM, Ahmadi SF, Kovesdy CP, Kalantar-Zadeh K. **Low-protein diet for conservative management of chronic kidney disease: a systematic review and meta-analysis of controlled trials**. *J Cachexia Sarcopenia Muscle* (2018.0) **9** 235-245. DOI: 10.1002/jcsm.12264
47. Li XF, Xu J, Liu LJ, Wang F, He SL, Su Y. **Efficacy of low-protein diet in diabetic nephropathy: a meta-analysis of randomized controlled trials**. *Lipids Health Dis* (2019.0) **18** 82. DOI: 10.1186/s12944-019-1007-6
48. Vashistha T, Mehrotra R, Park J, Streja E, Dukkipati R, Nissenson AR. **Effect of age and dialysis vintage on obesity paradox in long-term hemodialysis patients**. *Am J Kidney Dis* (2014.0) **63** 612-622. DOI: 10.1053/j.ajkd.2013.07.021
49. Kovesdy CP, Anderson JE, Kalantar-Zadeh K. **Paradoxical association between body mass index and mortality in men with CKD not yet on dialysis**. *Am J Kidney Dis* (2007.0) **49** 581-591. DOI: 10.1053/j.ajkd.2007.02.277
50. Guebre-Egziabher F, Alix PM, Koppe L, Pelletier CC, Kalbacher E, Fouque D. **Ectopic lipid accumulation: a potential cause for metabolic disturbances and a contributor to the alteration of kidney function**. *Biochimie* (2013.0) **95** 1971-1979. DOI: 10.1016/j.biochi.2013.07.017
51. Wahl P, Ducasa GM, Fornoni A. **Systemic and renal lipids in kidney disease development and progression**. *Am J Physiol Renal Physiol* (2016.0) **310** F433-F445. DOI: 10.1152/ajprenal.00375.2015
52. Garneata L, Stancu A, Dragomir D, Stefan G, Mircescu G. **Ketoanalogue-supplemented vegetarian very low-protein diet and CKD progression**. *J Am Soc Nephrol* (2016.0) **27** 2164-2176. DOI: 10.1681/ASN.2015040369
|
---
title: 'Substituting meat for mycoprotein reduces genotoxicity and increases the abundance
of beneficial microbes in the gut: Mycomeat, a randomised crossover control trial'
authors:
- Dominic N. Farsi
- Jose Lara Gallegos
- Georgios Koutsidis
- Andrew Nelson
- Tim J. A. Finnigan
- William Cheung
- Jose L. Muñoz-Muñoz
- Daniel M. Commane
journal: European Journal of Nutrition
year: 2023
pmcid: PMC10030420
doi: 10.1007/s00394-023-03088-x
license: CC BY 4.0
---
# Substituting meat for mycoprotein reduces genotoxicity and increases the abundance of beneficial microbes in the gut: Mycomeat, a randomised crossover control trial
## Abstract
### Purpose
The high-meat, low-fibre Western diet is strongly associated with colorectal cancer risk. Mycoprotein, produced from Fusarium venanatum, has been sold as a high-fibre alternative to meat for decades. Hitherto, the effects of mycoprotein in the human bowel have not been well considered. Here, we explored the effects of replacing a high red and processed meat intake with mycoprotein on markers of intestinal genotoxicity and gut health.
### Methods
Mycomeat (clinicaltrials.gov NCT03944421) was an investigator-blind, randomised, crossover dietary intervention trial. Twenty healthy male adults were randomised to consume 240 g day−1 red and processed meat for 2 weeks, with crossover to 2 weeks 240 g day−1 mycoprotein, separated by a 4-week washout period. Primary end points were faecal genotoxicity and genotoxins, while secondary end points comprised changes in gut microbiome composition and activity.
### Results
The meat diet increased faecal genotoxicity and nitroso compound excretion, whereas the weight-matched consumption of mycoprotein decreased faecal genotoxicity and nitroso compounds. In addition, meat intake increased the abundance of Oscillobacter and Alistipes, whereas mycoprotein consumption increased Lactobacilli, Roseburia and Akkermansia, as well as the excretion of short chain fatty acids.
### Conclusion
Replacing red and processed meat with the Fusarium-based meat alternative, mycoprotein, significantly reduces faecal genotoxicity and genotoxin excretion and increases the abundance of microbial genera with putative health benefits in the gut. This work demonstrates that mycoprotein may be a beneficial alternative to meat within the context of gut health and colorectal cancer prevention.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s00394-023-03088-x.
## Introduction
Epidemiological data consistently associates red and processed meat consumption with an increased risk of colorectal cancer (CRC) [1–4]. Mechanistic studies reveal that meat consumption increases both faecal water genotoxicity and the excretion of faecal genotoxins, such as nitroso compounds (NOC) [5, 6]. Further, meat may displace plant foods, and thus fibre, from the diet, leading to a reduced production of anti-carcinogenic gut microbial metabolites such as butyrate [7].
This evidence is reflected in recommendations from both EatLancet and IARC to reduce meat consumption [8, 9]. However, reducing meat intake is challenging due to food preferences, social norms, and strongly held culinary traditions [10]. Meat alternatives may facilitate meat reduction without requiring drastic changes to culinary practice [11, 12]. Mycoprotein is a meat alternative produced from Fusarium venenatum, which is high in both protein and fibre [13]. From a gut health perspective, the fibre fraction is an interesting combination of 1,$\frac{3}{1}$,6 β-glucan, chitin and lesser studied mannoproteins [14]. In mixed-culture fermentations with faecal inoculate, the fibre fraction induces a significant increase in short chain fatty acids (SCFA) [15]. When compared to myofibrillar protein, the fermentation of mycoprotein yields lower concentrations of ammonia [16]. Previous work has shown cardiometabolic benefits when replacing meat with mycoprotein that may be attributable to its fibre composition [17, 18]. However, to date, there have been no human intervention studies of the effect of mycoprotein on CRC risk markers or other facets of gut health.
The objective of the present study (Mycomeat) was to compare the effects of substituting a high intake of red and processed meat with mycoprotein on biomarkers of CRC risk and gut health, namely, faecal water genotoxicity, faecal genotoxins, gut microbial composition, and the excreted metabolome.
## Methods
Procedures for this study were followed in accordance with the Declaration of Helsinki and were approved by the Northumbria University Ethics Committee (reference number 15274). All study participants provided written informed consent.
## Study design
Mycomeat was a single-site, investigator-blind, randomised crossover trial (NCT03944421) among male adults randomised to consume either 240 g/day (uncooked weight) of red and processed meat (Meat) or mycoprotein, as Quorn™ products (Mycoprotein) for a 2-week period. Participants entered a 4-week washout period before swapping over to complete the alternative 2-week study arm (Fig. 1). The meat dose was selected to represent high intakes in the UK population and as a reflection of previous intervention trials assessing changes in faecal water genotoxicity in response to meat [6, 19]. The chronic feeding period of 2 weeks was selected based on this previous research, as well as a reflection of the average time taken to reach a new steady state for the microbiome in models of gut fermentation following a change in substrate [20]. A sample size of 20 to reach a statistical power of $80\%$ at an α-level of 0.05 was established based on a repeated measures analysis of the primary end point, faecal water genotoxicity. This power calculation was informed by a previous report that found 300 g/day of red meat for 7 days increased faecal water genotoxicity compared to participants’ habitual diets (before intervention 9.6 ± 13.8, after intervention 16.7 ± 18.2) [21]. Participant recruitment was stopped at the completion of $$n = 20$$ participants. Fig. 1Mycomeat study design. Mycomeat was an investigator-blind randomised crossover design comprising two study phases (2 weeks) separated by a washout period (4 weeks). During the study phases, participants supplemented habitual diet with either 240 g/day of red and processed meat (Meat), or mycoprotein-based foods (Mycoprotein)
## Participants
Participants were recruited from the North-East of England, UK, via on-campus poster advertisement and using a database of previous study participants. Inclusion criteria were: age 18–50 yr; BMI 18–30 kg/m2; willingness to refrain from pre- and probiotics, vitamin supplements as well as alcoholic beverages during the study. Exclusion criteria included: gastrointestinal disease; use of medications that affect gastrointestinal motility; use of antibiotic, prebiotic, or probiotics in the previous 3 months; use of tobacco or recreational drugs and history of coronary artery disease, diabetes, or other chronic disorders. We also excluded study participants who had been enrolled in dietary trials in the previous 3 months. All participants took part in a screening visit where they were characterised by blood parameters within standard clinical cutoff values to confirm study eligibility.
Participant enrolment began on 1 June 2019 and continued through to 1 December 2019. The date of final follow-up data collection was 29 January 2020. The Consort diagram in Supplementary Material: Fig. 1 depicts the flow of participants through the Mycomeat Study. Baseline characteristics of the participants who completed the study are shown in Table 1Table 1Participants’ baseline characteristics ($$n = 20$$)aAge, y30.4 ± 7.92Anthropometric parameters Weight, kg80.6 ± 10.90 BMI, kg/m224.0 ± 2.87 Waist circumference, cm86.9 ± 8.16 Body fat, %15.5 ± 5.56 Trunk fat, %16.3 ± 6.90 Systolic blood pressure, mmHg126 ± 12.2 Diastolic blood pressure, mmHg71.5 ± 9.27Blood parameters Total cholesterol, mmol/L4.32 ± 0.83 HDL cholesterol, mmol/L1.57 ± 0.43 LDL cholesterol, mmol/L2.29 ± 0.88 Triglycerides, mmol/L0.89 ± 0.46 Glucose, mmol/L4.86 ± 0.45Dietary intake Energy, kcal/d2515.59 ± 754.85 Protein, g/d126.44 ± 45.42 Protein, % of energy20.10 ± 7.22 Carbohydrate, g/d281.20 ± 81.41 Carbohydrate, % of energy44.71 ± 12.94 Fat, g/d98.37 ± 46.75 Fat, % of energy35.19 ± 16.73 Saturated fat, g/d35.94 ± 21.22 Saturated fat, % of energy12.86 ± 7.59 Fibre, g/d26.16 ± 7.79 Sodium, mg/d2853.02 ± 1243.93 Total meat intake, g/d221.60 ± 176.87 Red and processed meat intake g/d80.25 ± 83.47 Mycoprotein intake, g/d0 ± 0aData are presented as means ± SDs
## Intervention
At the beginning of each study phase, pre-frozen study foods (14 × 240 g packs; equating to 240 g/day for 14 days) were provided to participants, with instructions for appropriate storage, preparation, and cooking. The study foods and instructions were packaged externally and transported in concealed boxes. The boxes were assigned A or B, depending on the study arm, with the participants randomly assigned to receive either A or B in the first study phase, followed by the alternative in the second phase. In this way, the investigators were blinded to diet order. Randomisation was carried out using an online sequence randomisation generator, with $$n = 12$$ randomised to Mycoprotein phase first and $$n = 8$$ randomised to Meat phase first. All mycoprotein products were supplied by Quorn™, and all meat products were sourced from Asda supermarket chain (UK), and both mycoprotein and meat products were distributed on-site at the research facility.
To promote compliance, products were provided based on a 7-day rotation and included the following: beef steak, pork sausages, ham slices, gammon steak, bacon, beef mince and hot dogs. For the Mycoprotein phase, the equivalent Quorn™ products were included: peppered steak, sausages, deli ham slices, gammon steak, bacon, mince, and hot dogs. During intervention periods, participants were asked to maintain their usual diet, but to avoid consuming any other meat or mycoprotein products other than the supplied study foods as well as additional high protein, fibre or probiotic supplements. As stated, the study foods were matched for weight (240 g uncooked); however, the total energy content of meat and mycoprotein products were closely matched to avoid introducing additional effects from energy intake between the diets. The nutritional composition of the study foods and differences in nutrients are included in Supplementary Material, Table1i and 1ii. 1-day food records were collected at baseline and during each study phase, from which energy and macronutrient intake were calculated using Nutritics nutrition analysis software (version 5.66 Education) [22].
## Sample collection
At baseline and conclusion of each study phase, participants visited the Brain, Performance and Nutrition Research Centre (BPNRC/NUTRAN, Northumbria University, Newcastle, UK) having fasted overnight (Fig. 1). At each visit, a first spot urine sample and stool sample (collected ≤ 10 h prior to study visit) were collected from participants. Urine samples were immediately put on ice, then aliquoted into sterile 1.5 ml Eppendorf tubes and stored at − 80 °C until analysis. Stool samples were put on ice at collection and then aliquoted into sterile 2 ml tubes and stored at − 80 °C for gut microbial analysis. The remainder of the sample was weighed, diluted 1:1 with PBS (Sigma) and homogenised in a stomacher (Seward Stomacher 400 Circulator) at 200 bts/min for 2 min. The homogenates were transferred into polypropylene tubes and centrifuged at 65,000 ×g, for 2 h, at 4 °C using an ultra-speed centrifuge system (Thermo Scientific Sorvall LYNX 6000 Superspeed Centrifuge). The supernatants were then filtered through a 0.44 µm polyethersulfone (PES) syringe filter prior to a second filtration through a 0.22 µm PES syringe filter (Fisher Scientific). The supernatants, representing the faecal water fraction, were aliquoted into sterile 1.5 ml Eppendorf tubes and stored at − 80 °C for analysis of faecal water genotoxicity and metabolites.
## Faecal water genotoxicity
The human colon adenocarcinoma cell line Caco-2 was used to test faecal water genotoxicity using a CometChip assay (Trevigen), a high-throughput version of the standard alkaline comet assay [23].
Caco-2 cells were purchased from the ECACC and used between passage 10 and 22. Then they were routinely cultured in Dulbecco’s modified *Eagle medium* (DMEM) supplemented with 0.1 mM of non-essential amino acids, 100 U/mL of penicillin, 0.1 g/mL of streptomycin, and $20\%$ foetal bovine serum (FBS) (all Lonza UK). In the final passage prior to experimental procedures, the FBS was reduced to $10\%$ of culture medium and experiments were carried out with this as carrier control.
The comet assay was performed using the CometChip protocol (Trevigen, USA). To ensure adequate blinding, an independent analyst prepared $10\%$ (v/v) faecal extract in carrier media, prepared by passing the previously prepared faecal water through a 0.22 µm filter immediately prior to treating cells. $10\%$ v/v faecal extract preparations were shown to be non-cytotoxic in a 24 h MTT assay (viability was maintained at over $90\%$) and thus this concentration was deemed appropriate for genotoxicity.
Caco-2 cells were split at $80\%$ confluence, adjusted to 1 × 105 cells per ml and loaded on to the CometChip in 100 μl aliquots. After allowing cells to embed, excess media was removed and 100 μl of ($10\%$ v/v) treatment added. After 30 min, the treatment media was removed, and the chip was washed with PBS and sealed in low melting point agarose. The comet assay was completed under alkali conditions. Sybr Gold stained comets were visualised under fluorescence (LEICA DM5000 Fluouresence Microscope) at 5× magnification. Image acquisition was completed by a second investigator with no knowledge of treatments.
Comet tails were assessed using the automated CometChip software (Trevigen, USA) which counted a minimum of 150 cell nuclei per well. Data represents a mean of four wells per treatment. 50 μM H2O2 and a carrier control were included as inter-assay controls. T test revealed a significant difference in DNA damage between inter-assay controls ($$P \leq 0.02$$).
Faecal water genotoxicity significantly reduced from baseline after the Mycoprotein phase (− 8.28 ± $3.60\%$ DNA in tail, $$P \leq 0.05$$), while following the Meat phase, there was a non-significant increase (+ 4.91 ± $2.65\%$ DNA in tail, $$P \leq 0.09$$). The difference between these changes was also significant (13.19 ± $4.41\%$ DNA in tail, $$P \leq 0.01$$) (Fig. 2).Fig. 2Effects of Meat and Mycoprotein phases on faecal genotoxicity, assessed by percentage (%) DNA in tail following exposure to faecal water. Data represent a mean of four wells per treatment. 50 uM H2O2 and a carrier control were included as inter-assay controls, and t tests revealed a significant difference between 50 uM H2O2 and carrier control ($$P \leq 0.02$$). Mycoprotein phase: change from baseline, − 8.28 ± $3.60\%$, $$P \leq 0.05.$$ Meat phase: change from baseline, + 4.91 ± $2.65\%$, $$P \leq 0.09$$; Difference in study phase effects, 13.19 ± $4.41\%$, $$P \leq 0.01.$$ Error bars represent standard deviation. Changes within study phases and differences between study phases assessed using mixed-effects models (P ˂0.05 considered significant). * Indicates significant difference from baseline within the Mycoprotein study phase. # Indicates significant difference from baseline within the Meat study phase. †Indicates significant difference between the Mycoprotein and Meat study phase effects. § Indicates significant difference between 50 uM H2O2 and carrier control
## Faecal nitroso compounds
Faecal nitrates, nitrites and total NOC were determined using chemiluminescence as described previously [24]. To determine nitrite concentrations, 50 μl of faecal water was injected into a purge vessel containing 8 ml glacial acetic acid and 2 ml aqueous potassium iodide (50 mg/ml). Nitrogen was bubbled through a glass frit to mix the sample and transfer released nitric oxide to a Sievers NOA 280 analyser (Sievers, Boulder, CO, USA) via a condenser, an NaOH (1 mol/L) trap and a polypropylene filter (0.2 μm; Whatman, USA). The signal was processed using the instrument software. After every six injections, the purge vessel was emptied and refilled with fresh reagents. For quantification, known standards of sodium nitrite (1 to 10,000 nmol) were injected into the purge vessel filled with 8 ml glacial acetic acid and 2 ml aqueous potassium iodide (50 mg/ml).
For nitrate determination, the faecal water was incubated twofold with methanol for 30 min, followed by centrifugation at 14,000g at 4 °C for 5 min. 50 μl of the supernatant was injected into the purge vessel containing 8 ml vanadium (III) chloride solution (~ 0.4 g vanadium (III) chloride in 50 ml 1 M hydrochloric acid). The purge vessel was fitted with a water jacket to allow heating of the reagent to 96 °C and a cold water condenser (6 °C), using a circulating bath. Thereafter, the purge vessel was replenished with reagents as described above. The samples were quantified by comparing the area to the area of known standards of sodium nitrate (1 to 10,000 nmol). Results are expressed as nanomoles of total NOC.
## Microbial composition analysis by 16S ribosomal RNA amplicon sequencing
Stool aliquots were thawed prior to microbial DNA extraction using a Fast DNA Stool Mini Kit (Qiagen) according to the manufacturer’s instructions. 16S ribosomal RNA amplicon sequencing (rRNA) was performed at the NU-OMICS DNA sequencing research facility (Northumbria University). Sequencing libraries were prepared for targeted sequencing of the V4 region of the 16S rRNA gene from PCR amplicons as per the Schloss standard operating procedure (SOP) [25] using primers 515F and 806R [26]. Libraries were sequenced on the Illumina MiSeq platform (CA), using V2 (2 × 250), chemistry. Extraction kit and sequencing negative controls were prepared and sequenced simultaneously with all samples. Paired end reads were trimmed, merged, and processed by alignment to the SILVA database, followed by de novo clustering into operational taxonomic units (OTUs), and taxonomic assignment in Mothur, following the MiSeq SOP [25].
## Faecal volatile compounds
Automated headspace solid-phase microextraction (SPME) gas chromatography mass spectrometry (GC–MS) was conducted on an Agilent 7890A gas chromatograph, coupled to a CTC-PAL autosampler and a BencthTOF mass spectrometer (Markes Intl, Laitrisant, UK). Homogenised faecal samples (100–200 mg) were accurately weighed into 10 mL headspace vials, followed by the addition of 2 mL $26\%$ NaCl and $5\%$ metaphosphoric acid prior to sonication (5 min). The samples were then placed into the autosampler tray and incubated at 60 °C for 30 min, followed by a 30 s extraction onto a 75 μm Carboxen/ PDMS fibre (Supelco, Bellefonte, PA) and desorption in the GC injector set at 250 °C for 8 min. The injector was operated in the split mode at a split ratio of 20:1, using helium as a carrier gas at a constant flow of 1.15 mL/min. Chromatographic separation was achieved on VF-WaxMS ((L) 60 m × (D) 0.25 mm × (FT) 0.25 μm) capillary column (Agilent). The oven was held at 40 °C for 8 min followed by a temperature ramp at 4 °C/min to 100 °C, then 15 °C/min to 260 °C and held for 5 min. The mass spectrometer was operated in the total ion scan mode (m/z 30–450) with the ion source and transfer line temperatures held at 250 °C and 245 °C, respectively. The SCFA acetate, propionate, butyrate, valerate and hexanoate, branched chain fatty acids (BCFA) isobutyrate and isolvalerate, as well as cresol and phenol, were quantified using an external calibration curve.
Total SCFA excretion increased following the Mycoprotein phase, with an increase in all quantified SCFA, none of which reached statistical significance (Table 2). In contrast, there was a non-significant reduction in total SCFA after the Meat phase, with reductions across all SCFA quantified, none which were significant. Valerate was the only SCFA to be significantly different between study phases (144.31 ± 53.76 μg/g, $$P \leq 0.02$$). Total BCFA excretion was reduced following both diets, with a significant reduction after the Meat phase (− 190.15 ± 87.24 μg/g, $$P \leq 0.01$$). The difference between diets also reached statistical significance for BCFA quantified, as well as total BCFA (151.81 ± 86.67 μg/g, $$P \leq 0.04$$). Both diets caused reductions in faecal cresol, with increases in phenol. The effects within and between diets did not reach significance. Table 2Effects of 2-wk dietary interventions on faecal volatile compounds ($$n = 20$$)a,bVariableMean change from baseline, meatPcMean change from baseline, mycoproteinPcOverall difference, meat and mycoproteinPcShort chain fatty acids Acetate− 493.25 ± 233.320.76 + 874.20 ± 671.690.26 + 1367.45 ± 951.470.20 Propionate− 313.35 ± 304.540.40 + 637.10 ± 434.280.29 + 950.45 ± 442.450.09 Butyrate− 173.90 ± 170.300.66 + 147.65 ± 117.990.74 + 321.55 ± 316.920.54 Valerate− 63.24 ± 50.680.12 + 81.07 ± 58.720.39 + 144.31 ± 53.760.02† Hexanoate− 13.05 ± 10.910.51 + 20.16 ± 15.220.19 + 33.21 ± 22.340.17 Total SCFA− 1056.79 ± 806.370.58 + 1760.17 ± 1229.620.34 + 2816.96 ± 1749.880.17Branch chain fatty acids Isobutyrate− 98.10 ± 42.650.01*− 13.29 ± 9.220.63 + 84.81 ± 46.400.04† Isovalerate− 92.05 ± 45.060.02*− 25.05 ± 23.460.42 + 67.00 ± 41.250.05† Total BCFA− 190.15 ± 87.240.01*− 38.34 ± 32.040.52 + 151.81 ± 86.670.04†Proteolytic end products Cresol− 19.83 ± 17.250.18− 12.79 ± 9.710.31 + 7.04 ± 6.900.56 Phenol + 1.88 ± 1.160.15 + 3.52 ± 1.720.40− 1.64 ± 0.550.83SCFA short chain fatty acids, BCFA branch chain fatty acids*Mean change significantly different from baseline†Mean change significantly different between mycoprotein and meat dietary periodsaValues are presented as least square means ± SEs. Differences between variables at the beginning and end of each diet are shown. The overall difference column shows the differences between variables at the end of the Mycoprotein phase compared with the end of the Meat phase. P values were calculated for changes within and differences between study phases using mixed-effects modelsbResults are expressed in μg/gcA P ˂0.05 was considered significant
## Metabolomics
Frozen aliquots of urine were thawed prior to normalisation through dilution with sterile distilled water to the lowest specific gravity measured by refractometry as previously described [27]. Following this, 225 μL of the sample was transferred to 9 mm glass screw thread vials (Thermo Scientific), followed by the addition of 75 μL acetonitrile and stored at − 80 °C until the time of analysis.
For faecal water, frozen aliquots of sample were thawed before centrifugation at 18,000g at 4 °C for 30 min. 225 μL of the supernatant was transferred to 9 mm glass screw thread vials (Thermo Scientific), followed by the addition of 75 μL acetonitrile and stored at − 80 °C until the time of analysis.
Hydrophilic liquid interaction chromatography (HILIC) metabolite profiling of the urine and faecal water samples was performed on a Vanquish Liquid Chromatography chromatographic separation system connected to an IDX High Resolution Mass Spectrometer (Thermo Scientific). The HILIC positive and negative data sets were processed via Compound Discoverer 3.2 according to the following settings: untargeted metabolomic workflow with online database: mass tolerance 10 ppm, maximum shift 0.3 min, alignment model adaptive curve, minimum intensity 500 K, S/N threshold 3, compound consolidation, mass tolerance 10 ppm, RT tolerance 0.3 min. Database matching was performed at MS2 level using Thermo scientific m/z cloud with a similar index of $80\%$ or better.
A total of 1214 faecal metabolites were detected within the samples. PLS-DA between diets revealed a small distinction by the primary and secondary components (Supplementary Material: Fig. 2). Together, these two components accounted for ~ $26.6\%$ of the variability (9.6 and $17.8\%$ for the first and second components, respectively). Metabolites considered a VIP > 1.5 between diets are included in Supplementary Material: Table 4. Pathway analysis using the metabolites with VIP > 1.5 revealed thiamine metabolism was enriched, as well as pathways of fatty acid beta oxidation and carnitine synthesis following the Mycoprotein phase, whereas after the Meat phase, sphingolipid metabolism and purine metabolism were enriched (Supplementary Material: Fig. 2).
In urine, a total of 1,302 metabolites were detected within the samples. PLS-DA revealed a distinction between the diets by the primary and secondary components, however, with no defined clustering (Supplementary Material: Fig. 3). Together, these two components accounted for ~ $15.27\%$ of the variability (7.73 and $7.54\%$ for the first and second components, respectively). The metabolites considered a VIP > 1.5 are included in Supplementary Material: Table 5. Pathway analysis using the metabolites with VIP > 1.5 revealed thiamine metabolism was enriched after the Mycoprotein phase, followed by ammonia recycling, beta-alanine, -histidine, -glycine and -serine metabolism. However, taurine and hypotaurine metabolism was enriched the greatest after the Meat phase, followed by beta oxidation of very long chain fatty acids, phospholipid, steroid and bile acid biosynthesis. Both study phases led to an enrichment in the oxidation of BCFA (Supplementary Material: Fig. 3).
## Statistical analysis
All statistical analyses and visualisations were performed in RStudio [28]. Data were assessed for normality by visualising Q–Q plots and performing Shapiro–Wilk tests before statistical analysis. If data were assessed to be non-normally distributed, then non-parametric statistical tests were performed.
Excluding the metagenomic and metabolomic data, changes from baseline within study phases and differences between study phases were assessed using mixed-effects models. In all models, age, BMI, habitual alcohol intake and diet order were included as fixed effects, with the participant as the random effect.
For gut microbial analysis, Bray–Curtis dissimilarity was performed to assess beta diversity and ordination plots generated for visualisation. Differences within and between study phases in beta diversity were compared using a permutational multivariate analysis of variance. Alpha diversity was measured using Shannon index, inverse Simpson index, Fisher’s index and species richness (CHAO1 estimates). As microbiome sequencing data can contain a considerable number of zeros, we applied geometric mean of pairwise ratio (GMPR), a normalization method for zero-inflated data [29]. Using the normalised data, abundances of taxonomic ranks (i.e. phyla, genus) were determined. Changes in alpha diversity metrics and bacterial taxonomic abundance within and differences between study phases were compared using generalised mixed-effects models with participants as a random-effects factor. The models were adjusted for age, BMI and habitual alcohol intake; factors which can impact the microbiome, as well as the interaction of the randomisation order to determine any carry over effects. P values were adjusted using the Benjamini and Hochberg false discovery rate [30].
For metabolomics, multivariate statistical analysis was performed using MetaboAnalyst 5.0 [31]. Partial least squares discriminant analysis (PLS-DA) was performed to determine changes in metabolite profiles within and differences between study phases. The variable importance in projection (VIP) > 1.5 was taken to identify the features significantly differentiating within and between study phases, then the fold change ratio was obtained for each feature. Hierarchical cluster analysis heat maps were obtained using ward clustering algorithm and Euclidean distance calculation to further confirm the results of PLS-DA and to show the distribution of metabolites among all individuals. Taking the VIPs > 1.5 between diets, pathway analyses were performed to reveal which pathways were enriched in the diets. Within the metabolomic datasets, potential genotoxins of interest were identified, including the bile acids apocholic and 7-ketodeoxycholic in stool, and p-cresol sulphate in urine. To assess the differences within and between study phases in the excretion of these genotoxins, mixed-effects models were implemented as outlined above.
For all statistical analyses, a $P \leq 0.05$ was defined significant.
## Diet
This intervention involved changes in the composition of the volunteers’ diets across both study arms. During the Meat phase, volunteers’ average daily meat consumption increased from baseline by 18 g, and red and processed meat intake by 160 g, whilst during the Mycoprotein phase, individuals did not consume any meat and increased their consumption of mycoprotein by an average 240 g per day from baseline. Mean self-reported intake of total energy, protein, fat, saturated fat and sodium were not significantly changed from baseline by either diet. The Mycoprotein phase was associated with higher self-reported fibre intake, both from baseline (+ 16.99 ± 2.79 g/day, $P \leq 0.001$) and compared to the Meat phase (+ 16.74 ± 3.65 g/day, $P \leq 0.001$) (Supplementary Material: Table 2).
## Potential genotoxins
Faecal NOC increased after the Meat phase ($$P \leq 0.20$$), whereas there was a significant reduction following the Mycoprotein phase ($$P \leq 0.007$$). The difference between the diet effects was also statistically significant ($$P \leq 0.01$$) (Fig. 3a). As well as NOC, other potential genotoxins were identified in the metabolomics data: p-cresol sulphate in urine and the bile acids apocholic and 7-ketodeoxycholic in stool. Fig. 3Effects of Meat and Mycoprotein phases on potential genotoxins. a Faecal nitroso compounds. Presented in nmol. Mycoprotein phase: change from baseline, − 1044.00 ± 377.00 nmol, $$P \leq 0.02.$$ Meat phase: change from baseline, + 609.50 ± 541.00 nmol, $$P \leq 0.20$$; Difference in the study phase effects, 1653.50 ± 677.00 nmol, $$P \leq 0.01.$$ b Urinary P-cresol sulphate. Presented in intensity. Mycoprotein phase: change from baseline, -3.35 × 108 ± 9.8 × 108intensity, $$P \leq 0.002$$; Meat phase: change from baseline, + 1.19 × 108 ± 1.41 × 108intensity, $$P \leq 0.40.$$ Difference in study phase effects, 4.54 × 108 ± 1.89 × 108intensity, $$P \leq 0.02.$$ c Faecal apocholic acid. Presented in intensity. Mycoprotein phase: change from baseline, + 2.23 × 107 ± 2.01 × 107intensity, $$P \leq 0.27.$$ Meat phase: change from baseline, −2.70 × 107 ± 2.42 × 107intensity, $$P \leq 0.27.$$ Difference in study phase effects, 4.93 × 107 ± 3.37 × 107intensity, $$P \leq 0.15.$$ d Faecal 7-ketodeoxycholic acid. Presented in intensity. Mycoprotein phase: change from baseline, + 2.86 × 107 ± 1.89 × 107intensity, $$P \leq 0.82.$$ Meat phase: change from baseline, −8.03 × 106 ± 2.91 × 106intensity, $$P \leq 0.009.$$ Difference in study phase effects, 3.66 × 107 ± 2.00 × 107intensity, $$P \leq 0.17.$$ Error bars represent standard deviation. For all data, changes within study phases and differences between study phases assessed using mixed-effects models (P ˂0.05 considered significant). * Indicates significant difference from baseline within the Mycoprotein study phase. # Indicates significant difference from baseline within the Meat study phase. †Indicates significant difference between Mycoprotein and Meat study phase effects Following the Mycoprotein phase, there was a significant reduction in urinary p-cresol sulphate excretion ($$P \leq 0.002$$), and with an increase after the Meat phase ($$P \leq 0.40$$), the difference between diet effects was also significant ($$P \leq 0.02$$) (Fig. 3b). Apocholic acid increased in stool from baseline after the Mycoprotein phase ($$P \leq 0.27$$) and reduced following the Meat phase ($$P \leq 0.27$$), with the difference between diet effects not significant ($$P \leq 0.15$$) (Fig. 3c). 7-Ketodeoxycholic acid followed a similar pattern, increasing after the Mycoprotein phase ($$P \leq 0.82$$) and reducing after the Meat phase ($$P \leq 0.009$$), the difference between diet effects was also not significant ($$P \leq 0.17$$) (Fig. 3d).
## Microbial composition
Neither intervention affected alpha or beta diversity (data not shown). Bacterial phylotype analysis revealed significant changes in the relative abundances of bacterial communities. There were significant increases in the relative abundance of the bacterial phyla Proteobacteria ($$P \leq 0.004$$) and Verrucomicrobia ($$P \leq 0.02$$) following Mycoprotein, while Verrucomicrobia significantly reduced after Meat ($$P \leq 0.04$$), the difference between diet effects on Verrucomicrobia was also significant ($$P \leq 0.03$$) (Fig. 4).Fig. 4Effects of Meat and Mycoprotein phases on gut microbial phylum composition. Gut microbial phylum composition at baseline and completion for both Meat and Mycoprotein phases. There were significant increases in Proteobacteria ($$P \leq 0.004$$) and Verrucomicrobia ($$P \leq 0.02$$) following the Mycoprotein phase and a significant reduction in Verrucomicrobia ($$P \leq 0.04$$) after the Meat phase. The difference in diet effects on Verrucomicrobia ($$P \leq 0.03$$) was also significant. Changes within study phases and differences between study phases assessed using generalised mixed-effects models (P ˂0.05 considered significant) At the genera level, Akkermansia significantly increased following the Mycoprotein phase ($$P \leq 0.02$$), while there was a reduction in abundance after the Meat phase, the difference between diets being significant ($$P \leq 0.03$$). Mycoprotein and Meat phases had opposing effects on Roseburia, which increased after Mycoprotein ($P \leq 0.001$) and reduced following Meat ($P \leq 0.001$), with the difference between the diets being highly significant ($P \leq 0.001$). The diets also had contrasting influences on Faecalibacterium, which significantly increased after Meat ($$P \leq 0.006$$) and reduced following the Mycoprotein phase, with the difference between diets also being significant ($$P \leq 0.02$$). Oscillibacter was significantly increased following the Meat phase compared to the Mycoprotein phase ($$P \leq 0.004$$). Other notable effects were a significant reduction in Ruminococcus following the Meat phase ($$P \leq 0.03$$) and a significant increase in Lactobacillus after the Mycoprotein phase ($$P \leq 0.05$$). *The* genera significantly affected within the study phases are shown in Fig. 5 and those significantly different between study phases in Fig. 6.Fig. 5Effects of Meat and Mycoprotein phases on gut microbial genera. Gut microbial genera identified as significant for difference between study phase effects. Differences between study phases assessed using generalised mixed-effects models (P ˂0.05 considered significant)Fig. 6Effects of Meat and Mycoprotein phases on gut microbial genera. a Gut microbial genera identified as significant for change in abundance from baseline after the Mycoprotein phase. b Gut microbial genera identified as significant for change in abundance from baseline after the Meat phase. Changes within study phases assessed using generalised mixed-effects models (P ˂0.05 considered significant) In total, the analysis identified 392 OTUs in the samples. The OTUs identified as significantly different within and between study phases are included in the Supplementary Material: Table 3.
## Discussion
The present study investigated the impact of replacing a high red and processed meat diet with mycoprotein on faecal water genotoxicity and markers related to intestinal health. We recognise faecal water genotoxicity is poorly validated against tumour incidence in humans; however, it has been widely used as a non-invasive surrogate risk marker in short-term interventions exploring dietary components (notably meat) and CRC risk [5, 32, 33]. Further, it predicts subsequent tumour incidence in rodent models [34] and is used on the well-founded assumption that DNA damage is an initiating event in carcinogenesis [35], and therefore high genotoxic exposures may be viewed as a risk factor for neoplasia.
We observed that the Meat phase increased faecal water genotoxicity which corroborates previous findings [5, 21]. We also observed that the Mycoprotein phase significantly reduced genotoxicity, both relative to the Meat phase and to baseline (i.e., a standard western diet). This is a novel finding which suggests consuming mycoprotein may be protective against DNA damage, either via a displacement effect on harmful constituents of the diet, or independently via the introduction of antigenotoxic factors in the gut. Consistent with earlier studies, we also observed an increase in faecal NOC following the Meat phase, which, akin to faecal water genotoxicity, was significantly reduced by the Mycoprotein phase. The close association between genotoxicity and NOC suggests the latter as the predominant contributor to the former.
Both diets significantly influenced the composition of the gut microbiota. The most notable changes were increases in the relative abundance of Lactobacilli, Roseburia, and Akkermansia following the Mycoprotein phase. In models, Lactobacilli consistently exert significant protection against chemically induced tumours [36]. In vitro studies suggest one of the mechanisms for this anticancer effect may be via binding and chemical modification of NOC [37]. It is also evidenced that certain Lactobacilli augment intestinal barrier function by improving tight junction integrity [38] and enhancing colonic mucin production [39], conferring a further layer of defence against genotoxic insult. While this study, to our knowledge, is the first to report increases in Lactobacilli following mycoprotein consumption, it suggests that the fibre fraction of mycoprotein may have prebiotic potential [40]. The butyrate-producing Roseburia were also increased in relative abundance following the Mycoprotein phase, but decreased following the Meat phase. Roseburia have been associated with suppression of gut inflammatory processes [41] and are reduced in both inflammatory bowel disease [42] and CRC [43]. Butyrate is a potent anti-neoplastic agent in the gut [44]. Akkermansia is an abundant inhabitant of the human gut, degrading intestinal mucin which enhances cell turnover, in addition to priming trophic chains [45]. Further, propionic and acetic acids are metabolic by-products from the degradation of mucins, which can be absorbed and utilised by host tissue for energy or leveraged by other species to produce butyrate [45].
In contrast, the Meat phase caused the enrichment of the bile-resistant, putrefactive Alistipes and Oscillibacter. This might be an adaptive response to bile, or due to a greater proclivity for digestion-resistant myofibrillar protein arriving in the gut from meat versus mycoprotein. Acute feeding studies with mycoprotein lead to very high muscle protein synthesis rates, suggesting a high bioavailability of mycoprotein, which may culminate in low levels of dietary protein reaching the colon [46]. While observational data are conflicting, with studies showing Alistipes to be associated with both health and disease [47], Oscillobacter has been linked with weight gain, metabolic dysfunction and a leaky gut [48].
Prior to our study, gut fermentation of mycoprotein has only been reported in an in vitro model, which found an increase in total SCFA production [15]. We report increases in the faecal concentrations of all quantified SCFA following the Mycoprotein phase, which we attribute to the difference in fibre intake relative to participants’ habitual diet (+ 16.99 g day−1). The increased load of SCFA coincides with the enrichment of SCFA-producing genera following the Mycoprotein phase. Future work may consider a mycoprotein vs non-mycoprotein diet matched for fibre intake to elucidate if the 1,$\frac{3}{1}$,6 β-glucan/chitin present in mycoprotein is associated with a different SCFA response to other functional fibres. The reduction in SCFA excretion following the Meat phase was expected; diets rich in animal protein and lower in fibre are characterised by lower faecal SCFA compared to diets higher in fibre and lower in animal protein [49]. We also report reductions in BCFA following both diets. BCFA are considered markers of proteolytic fermentation [50] and concentrations tend to rise with increased protein intake [51]. Our observations could be due to the reduction in participants’ self-reported protein intake from baseline during intervention diets (− 11.41 g/day mycoprotein; − 8.61 g/day meat). There was also a significant difference in BCFA excretion between diets, with higher values following the Mycoprotein phase despite lower self-reported protein intake than the Meat phase. The difference in animal protein may not be a factor, as previous work found no differences in BCFA between vegans and omnivores [52]. The concentrations of excreted BCFA correlate weakly with the abundance of the largely saccharolytic genera Blautia (isobutyrate $r = 0.235$, $P \leq 0.05$, and isovalerate $r = 0.229$, $P \leq 0.05$), Coprococcus (isobutyrate $r = 0.243$, $P \leq 0.05$ and isovalerate $r = 0.252$, $P \leq 0.05$) and Odoribacter (isovalerate $r = 0.225$ $P \leq 0.05$). The higher concentrations of BCFA with the Mycoprotein relative to the Meat phase may therefore simply reflect higher saccharolytic microbial activity on that arm in response to the relative availability of fibre.
Applying untargeted metabolomics, we were able to decipher the predominant metabolic pathways in urine and stool. We found an enrichment in sphingolipid metabolism after the Meat phase. Sphingolipids are important mammalian signalling molecules with various biological activities, and importantly increased sphingolipid metabolism may be a response to DNA damage [53].
The rich metabolomic datasets also enabled the identification of other potential genotoxins as well as our targeted NOC analysis. We identified p-cresol sulphate in urine, a product of tyrosine fermentation in the gut [54], which is absorbed and sulphated for excretion in urine [55]. Studies have demonstrated p-cresol reaches genotoxic concentrations in in vitro gut fermentation models, and notably can predict the genotoxicity of the fermentation supernatant [56]. Like faecal genotoxicity and NOC, urinary p-cresol sulphate was increased by the Meat phase and significantly reduced by the Mycoprotein phase. These findings are further indicative of negative and positive effects of meat and mycoprotein, respectively. We also identified the bile acids apocholic and 7-ketodeoxycholic in stool. 7α-dehydroxylating bacteria convert and activate bile acids in the gut [57]. The transformed secondary bile acids have the potential to induce oxidative stress and DNA damage [58, 59] and are implicated in tumorigenesis [60]. We observed reductions in these bile acids after the Meat and increases following the Mycoprotein phase. Thus, we would argue they are not the main contributors to the genotoxic load in the faecal samples. Bile acid metabolism plays a predominant role in cholesterol homeostasis [61], and increased bile acids in stool may be indicative of mycoprotein’s ability to bind and sequester bile acids [62]. This may be a mechanism underpinning observations of cholesterol reduction by mycoprotein [63–65].
The present study has several strengths. We used a randomised crossover control investigator-blind study design, thus significantly reducing bias. The meat and mycoprotein products included are regularly consumed by the public, and the dose is achievable in a real-world setting. We also incorporated a variety of products in the dietary regimens to aid participant compliance, and to reflect the variety in real-world food exposures. It is noteworthy that once randomised to the study, we encountered no participant dropouts, which may be in part due to this approach. The Thermo IDX LCMS platform uses automated data acquisition and peak annotation resulting in improved metabolite profiling efficacy in comparison to alternative platforms. This allowed the ability to obtain a greater number of identifiable metabolites, which in turn enabled a rich dataset to determine metabolic pathways hitherto relatively unexplored in this type of study.
Our work also has limitations. The cohort included exclusively healthy, non-obese adult males. This cohort was selected to control external factors, but future studies might use larger, more diverse cohorts. We advised participants to avoid certain foods and supplements; however, we did not control dietary intake other than the supplied study products. Other dietary factors during the study phases may have influenced our findings. 16S amplicon sequencing was used for microbial analysis which limits the coverage to genus. We also collected single-day faecal samples, but the stool microbiome can show intraday variation.
In conclusion, this work demonstrates that substituting red and processed meat for the meat alternative mycoprotein reduces faecal genotoxins and the genotoxic load and increases SCFA production, as well as the abundance of beneficial genera in the human gut. Thus, mycoprotein may be considered a good alternative to meat when consumed as part of a balanced diet in the context of gut health and long-term CRC risk.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 716 KB)
## References
1. Bernstein AM, Song M, Zhang X, Pan A, Wang M, Fuchs CS. **Processed and unprocessed red meat and risk of colorectal cancer: analysis by tumor location and modification by time**. *PLoS ONE* (2015.0) **10** e0135959. DOI: 10.1371/journal.pone.0135959
2. Carr PR, Holleczek B, Stegmaier C, Brenner H, Hoffmeister M. **Meat intake and risk of colorectal polyps: results from a large population-based screening study in Germany**. *Am J Clin Nutr* (2017.0) **105** 1453-1461. DOI: 10.3945/ajcn.116.148304
3. 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. DOI: 10.1371/journal.pmed.0040325
4. Key TJ, Bradbury KE, Murphy N. **Diet and colorectal cancer in UK Biobank: a prospective study**. *Int J Epidemiol* (2019.0). DOI: 10.1093/ije/dyz064
5. Hughes R, Pollock JR, Bingham S. **Effect of vegetables, tea, and soy on endogenous N-nitrosation, fecal ammonia, and fecal water genotoxicity during a high red meat diet in humans**. *Nutr Cancer* (2002.0) **42** 70-77. DOI: 10.1207/s15327914nc421_10
6. Joosen AM, Lecommandeur E, Kuhnle GG, Aspinall SM, Kap L, Rodwell SA. **Effect of dietary meat and fish on endogenous nitrosation, inflammation and genotoxicity of faecal water**. *Mutagenesis* (2010.0) **25** 243-247. DOI: 10.1093/mutage/gep070
7. Gao J, Guo X, Wei W, Li R, Hu K, Liu X. **The association of fried meat consumption with the gut microbiota and fecal metabolites and its impact on glucose homoeostasis, intestinal endotoxin levels, and systemic inflammation: a randomized controlled-feeding trial**. *Diabetes Care* (2021.0) **44** 1970-1979. DOI: 10.2337/dc21-0099
8. Bouvard V, Loomis D, Guyton KZ, Grosse Y, Ghissassi FE, Benbrahim-Tallaa L. **Carcinogenicity of consumption of red and processed meat**. *Lancet Oncol* (2015.0) **16** 1599-1600. DOI: 10.1016/S1470-2045(15)00444-1
9. Willett W, Rockström J, Loken B, Springmann M, Lang T, Vermeulen S. **Food in the anthropocene: the EAT-lancet commission on healthy diets from sustainable food systems**. *Lancet* (2019.0) **393** 447-492. DOI: 10.1016/s0140-6736(18)31788-4
10. Sanchez-Sabate R, Sabaté J. **Consumer attitudes towards environmental concerns of meat consumption: a systematic review**. *Int J Environ Res Public Health* (2019.0) **16** 1220. DOI: 10.3390/ijerph16071220
11. Kumar P, Chatli MK, Mehta N, Singh P, Malav OP, Verma AK. **Meat analogues: Health promising sustainable meat substitutes**. *Crit Rev Food Sci Nutr* (2017.0) **57** 923-932. DOI: 10.1080/10408398.2014.939739
12. Hu FB, Otis BO, McCarthy G. **Can plant-based meat alternatives be part of a healthy and sustainable diet?**. *JAMA* (2019.0) **322** 1547-1548. DOI: 10.1001/jama.2019.13187
13. 13.Finnigan T, Needham L, Abbott C, editors (2017) Mycoprotein: a healthy new protein with a low environmental impact
14. Derbyshire E, Ayoob K-T. **Mycoprotein: nutritional and Health properties**. *Nutr Today* (2019.0) **54** 7-15. DOI: 10.1097/nt.0000000000000316
15. Harris HC, Edwards CA, Morrison DJ. **Short chain fatty acid production from mycoprotein and mycoprotein fibre in an in vitro fermentation model**. *Nutrients* (2019.0) **11** 800. DOI: 10.3390/nu11040800
16. Wang X, Gibson GR, Costabile A, Sailer M, Theis S, Rastall RA. **Prebiotic supplementation of in vitro fecal fermentations inhibits proteolysis by gut bacteria, and host diet shapes gut bacterial metabolism and response to intervention**. *Appl Environ Microbiol* (2019.0). DOI: 10.1128/aem.02749-18
17. Cherta-Murillo A, Lett AM, Frampton J, Chambers ES, Finnigan TJA, Frost GS. **Effects of mycoprotein on glycaemic control and energy intake in humans: a systematic review**. *Br J Nutr* (2020.0) **123** 1321-1332. DOI: 10.1017/s0007114520000756
18. Coelho MOC, Monteyne AJ, Dunlop MV, Harris HC, Morrison DJ, Stephens FB. **Mycoprotein as a possible alternative source of dietary protein to support muscle and metabolic health**. *Nutr Rev* (2020.0) **78** 486-497. DOI: 10.1093/nutrit/nuz077
19. Hughes R, Cross AJ, Pollock JRA, Bingham S. **Dose-dependent effect of dietary meat on endogenous colonic N-nitrosation**. *Carcinogenesis* (2001.0) **22** 199-202. DOI: 10.1093/carcin/22.1.199
20. Collins SM, Gibson GR, Kennedy OB, Walton G, Rowland I, Commane DM. **Development of a prebiotic blend to influence in vitro fermentation effects, with a focus on propionate, in the gut**. *FEMS Microbiol Ecol* (2021.0). DOI: 10.1093/femsec/fiab101
21. Hebels DG, Sveje KM, de Kok MC, van Herwijnen MH, Kuhnle GG, Engels LG. **Red meat intake-induced increases in fecal water genotoxicity correlate with pro-carcinogenic gene expression changes in the human colon**. *Food Chem Toxicol* (2012.0) **50** 95-103. DOI: 10.1016/j.fct.2011.10.038
22. 22.Nutritics. *Education Edition, v566, Dublin, Nutritics, 2021.
23. Ge J, Prasongtanakij S, Wood DK, Weingeist DM, Fessler J, Navasummrit P. **CometChip: a high-throughput 96-well platform for measuring DNA damage in microarrayed human cells**. *J Visualized Exp: JoVE* (2014.0). DOI: 10.3791/50607
24. Kuhnle GGC, Story GW, Reda T, Mani AR, Moore KP, Lunn JC. **Diet-induced endogenous formation of nitroso compounds in the GI tract**. *Free Radic Biol Med* (2007.0) **43** 1040-1047. DOI: 10.1016/j.freeradbiomed.2007.03.011
25. Kozich JJ, Westcott SL, Baxter NT, Highlander SK, Schloss PD. **Development of a dual-index sequencing strategy and curation pipeline for analyzing amplicon sequence data on the MiSeq Illumina sequencing platform**. *Appl Environ Microbiol* (2013.0) **79** 5112-5120. DOI: 10.1128/AEM.01043-13
26. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Lozupone CA, Turnbaugh PJ. **Global patterns of 16S rRNA diversity at a depth of millions of sequences per sample**. *Proceed National Acad Sci United States of America* (2011.0) **108** 4516-4522. DOI: 10.1073/pnas.1000080107
27. Edmands WMB, Ferrari P, Scalbert A. **Normalization to specific gravity prior to analysis improves information recovery from high resolution mass spectrometry metabolomic profiles of human urine**. *Anal Chem* (2014.0) **86** 10925-10931. DOI: 10.1021/ac503190m
28. 28.R Core Team (2020) R: A language and environment for statistical computing: R Foundation for Statistical Computing
29. Chen L, Reeve J, Zhang L, Huang S, Wang X, Chen J. **GMPR: a robust normalization method for zero-inflated count data with application to microbiome sequencing data**. *PeerJ* (2018.0). DOI: 10.7717/peerj.4600
30. Benjamini Y, Hochberg Y. **Controlling the false discovery rate: a practical and powerful approach to multiple testing**. *J Royal Statistical Soc: Series B (Methodological)* (1995.0) **57** 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x
31. Pang Z, Chong J, Zhou G, de Lima Morais DA, Chang L, Barrette M. **MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights**. *Nucleic Acids Res* (2021.0) **49** W388-W396. DOI: 10.1093/nar/gkab382
32. Costabile A, Fava F, Röytiö H, Forssten SD, Olli K, Klievink J. **Impact of polydextrose on the faecal microbiota: a double-blind, crossover, placebo-controlled feeding study in healthy human subjects**. *Br J Nutr* (2012.0) **108** 471-481. DOI: 10.1017/S0007114511005782
33. Eid N, Osmanova H, Natchez C, Walton G, Costabile A, Gibson G. **Impact of palm date consumption on microbiota growth and large intestinal health: a randomised, controlled, cross-over, human intervention study**. *Br J Nutr* (2015.0) **114** 1226-1236. DOI: 10.1017/S0007114515002780
34. Klinder A, Forster A, Caderni G, Femia AP, Pool-Zobel BL. **Fecal water genotoxicity is predictive of tumor-preventive activities by inulin-like oligofructoses, Probiotics (**. *Nutr Cancer* (2004.0) **49** 144-155. DOI: 10.1207/s15327914nc4902_5
35. Alhmoud JF, Woolley JF, Al Moustafa A-E, Malki MI. **DNA damage/repair management in cancers**. *Cancers* (2020.0) **12** 1050. DOI: 10.3390/cancers12041050
36. Brasiel PGA, Dutra Luquetti SCP, Peluzio M, Novaes RD, Gonçalves RV. **Preclinical evidence of probiotics in colorectal carcinogenesis: a systematic review**. *Dig Dis Sci* (2020.0) **65** 3197-3210. DOI: 10.1007/s10620-020-06062-3
37. Caldini G, Trotta F, Villarini M, Moretti M, Pasquini R, Scassellati-Sforzolini G. **Screening of potential lactobacilli antigenotoxicity by microbial and mammalian cell-based tests**. *Int J Food Microbiol* (2005.0) **102** 37-47. DOI: 10.1016/j.ijfoodmicro.2004.11.015
38. Karczewski J, Troost FJ, Konings I, Dekker J, Kleerebezem M, Brummer R-JM. **Regulation of human epithelial tight junction proteins by**. *A J Physiol- Gastrointestinal Liver Physiol* (2010.0) **298** G851-G859. DOI: 10.1152/ajpgi.00327.2009
39. Chen CM, Wu CC, Huang CL, Chang MY, Cheng SH, Lin CT. *Probiotics Antimicrob Proteins* (2021.0). DOI: 10.1007/s12602-021-09814-3
40. Costabile A, Kolida S, Klinder A, Gietl E, Bäuerlein M, Frohberg C. **A double-blind, placebo-controlled, cross-over study to establish the bifidogenic effect of a very-long-chain inulin extracted from globe artichoke (**. *Br J Nutr* (2010.0) **104** 1007-1017. DOI: 10.1017/S0007114510001571
41. Atarashi K, Tanoue T, Shima T, Imaoka A, Kuwahara T, Momose Y. **Induction of colonic regulatory T cells by indigenous clostridium species**. *Science* (2011.0) **331** 337-341. DOI: 10.1126/science.1198469
42. Takahashi K, Nishida A, Fujimoto T, Fujii M, Shioya M, Imaeda H. **Reduced abundance of butyrate-producing bacteria species in the fecal microbial community in Crohn’s disease**. *Digestion* (2016.0) **93** 59-65. DOI: 10.1159/000441768
43. Geng J, Fan H, Tang X, Zhai H, Zhang Z. **Diversified pattern of the human colorectal cancer microbiome**. *Gut Pathogens* (2013.0) **5** 2. DOI: 10.1186/1757-4749-5-2
44. Bernardazzi C, Xu H, Tong H, Laubitz D, Figliuolo da Paz V, Curiel L. **An indisputable role of NHE8 in mucosal protection**. *Am J Physiol Gastrointest Liver Physiol* (2020.0) **319** G421-G431. DOI: 10.1152/ajpgi.00246.2020
45. de Vos WM. **Microbe Profile: Akkermansia muciniphila: a conserved intestinal symbiont that acts as the gatekeeper of our mucosa**. *Microbiology* (2017.0) **163** 646-648. DOI: 10.1099/mic.0.000444
46. Monteyne AJ, Coelho MOC, Porter C, Abdelrahman DR, Jameson TSO, Jackman SR. **Mycoprotein ingestion stimulates protein synthesis rates to a greater extent than milk protein in rested and exercised skeletal muscle of healthy young men: a randomized controlled trial**. *Am J Clin Nutr* (2020.0) **112** 318-333. DOI: 10.1093/ajcn/nqaa092
47. Parker BJ, Wearsch PA, Veloo ACM, Rodriguez-Palacios A. **The genus alistipes: gut bacteria with emerging implications to inflammation, cancer, and mental health**. *Front Immunol* (2020.0). DOI: 10.3389/fimmu.2020.00906
48. Lam YY, Ha CWY, Campbell CR, Mitchell AJ, Dinudom A, Oscarsson J. **Increased gut permeability and microbiota change associate with mesenteric fat inflammation and metabolic dysfunction in diet-induced obese mice**. *PLoS ONE* (2012.0) **7** e34233. DOI: 10.1371/journal.pone.0034233
49. Ríos-Covián D, Ruas-Madiedo P, Margolles A, Gueimonde M, de Los Reyes-Gavilán CG, Salazar N. **Intestinal short chain fatty acids and their link with diet and human health**. *Front Microbiol* (2016.0). DOI: 10.3389/fmicb.2016.00185
50. Windey K, De Preter V, Verbeke K. **Relevance of protein fermentation to gut health**. *Mol Nutr Food Res* (2012.0) **56** 184-196. DOI: 10.1002/mnfr.201100542
51. Russell WR, Gratz SW, Duncan SH, Holtrop G, Ince J, Scobbie L. **High-protein, reduced-carbohydrate weight-loss diets promote metabolite profiles likely to be detrimental to colonic health**. *Am J Clin Nutr* (2011.0) **93** 1062-1072. DOI: 10.3945/ajcn.110.002188
52. Trefflich I, Dietrich S, Braune A, Abraham K, Weikert C. **Short- and branched-chain fatty acids as fecal markers for microbiota activity in vegans and omnivores**. *Nutrients* (2021.0) **13** 1808. DOI: 10.3390/nu13061808
53. Xu R, Garcia-Barros M, Wen S, Li F, Lin C-L, Hannun YA. **Tumor suppressor p53 links ceramide metabolism to DNA damage response through alkaline ceramidase 2**. *Cell Death Differ* (2017.0). DOI: 10.1038/s41418-017-0018-y
54. Saito Y, Sato T, Nomoto K, Tsuji H. **Identification of phenol- and p-cresol-producing intestinal bacteria by using media supplemented with tyrosine and its metabolites**. *FEMS Microbiol Ecol* (2018.0). DOI: 10.1093/femsec/fiy125
55. Wikoff WR, Anfora AT, Liu J, Schultz PG, Lesley SA, Peters EC. **Metabolomics analysis reveals large effects of gut microflora on mammalian blood metabolites**. *Proc Natl Acad Sci* (2009.0) **106** 3698-3703. DOI: 10.1073/pnas.0812874106
56. Al Hinai EA, Kullamethee P, Rowland IR, Swann J, Walton GE, Commane DM. **Modelling the role of microbial p-cresol in colorectal genotoxicity**. *Gut Microbes* (2018.0). DOI: 10.1080/19490976.2018.1534514
57. Ridlon JM, Kang DJ, Hylemon PB. **Bile salt biotransformations by human intestinal bacteria**. *J Lipid Res* (2006.0) **47** 241-259. DOI: 10.1194/jlr.R500013-JLR200
58. Dvorak K, Payne CM, Chavarria M, Ramsey L, Dvorakova B, Bernstein H. **Bile acids in combination with low pH induce oxidative stress and oxidative DNA damage: relevance to the pathogenesis of Barrett’s oesophagus**. *Gut* (2007.0) **56** 763-771. DOI: 10.1136/gut.2006.103697
59. Lechner S, Müller-Ladner U, Schlottmann K, Jung B, McClelland M, Rüschoff J. **Bile acids mimic oxidative stress induced upregulation of thioredoxin reductase in colon cancer cell lines**. *Carcinogenesis* (2002.0) **23** 1281-1288. DOI: 10.1093/carcin/23.8.1281
60. Ocvirk S, O'Keefe SJ. **Influence of bile acids on colorectal cancer risk: potential mechanisms mediated by Diet–Gut microbiota interactions**. *Curr Nutr Rep* (2017.0) **6** 315-322. DOI: 10.1007/s13668-017-0219-5
61. Chiang JYL. **Bile acid metabolism and signaling**. *Compr Physiol* (2013.0) **3** 1191-1212. DOI: 10.1002/cphy.c120023
62. Colosimo R, Mulet-Cabero A-I, Warren FJ, Edwards CH, Finnigan TJA, Wilde PJ. **Mycoprotein ingredient structure reduces lipolysis and binds bile salts during simulated gastrointestinal digestion**. *Food Funct* (2020.0) **11** 10896-10906. DOI: 10.1039/D0FO02002H
63. Coelho MOC, Monteyne AJ, Dirks ML, Finnigan TJA, Stephens FB, Wall BT. **Daily mycoprotein consumption for 1 week does not affect insulin sensitivity or glycaemic control but modulates the plasma lipidome in healthy adults: a randomised controlled trial**. *Br J Nutr* (2021.0) **125** 147-160. DOI: 10.1017/S0007114520002524
64. Turnbull WH, Leeds AR, Edwards DG. **Mycoprotein reduces blood lipids in free-living subjects**. *Am J Clin Nutr* (1992.0) **55** 415-419. DOI: 10.1093/ajcn/55.2.415
65. Turnbull WH, Leeds AR, Edwards GD. **Effect of mycoprotein on blood lipids**. *Am J Clin Nutr* (1990.0) **52** 646-650. DOI: 10.1093/ajcn/52.4.646
|
---
title: Pooled analysis of epigenome-wide association studies of food consumption in
KORA, TwinsUK and LLS
authors:
- Fabian Hellbach
- Lucy Sinke
- Ricardo Costeira
- Sebastian-Edgar Baumeister
- Marian Beekman
- Panayiotis Louca
- Emily R. Leeming
- Olatz Mompeo
- Sarah Berry
- Rory Wilson
- Nina Wawro
- Dennis Freuer
- Hans Hauner
- Annette Peters
- Juliane Winkelmann
- Wolfgang Koenig
- Christa Meisinger
- Melanie Waldenberger
- Bastiaan T. Heijmans
- P. Eline Slagboom
- Jordana T. Bell
- Jakob Linseisen
journal: European Journal of Nutrition
year: 2022
pmcid: PMC10030421
doi: 10.1007/s00394-022-03074-9
license: CC BY 4.0
---
# Pooled analysis of epigenome-wide association studies of food consumption in KORA, TwinsUK and LLS
## Abstract
### Purpose
Examining epigenetic patterns is a crucial step in identifying molecular changes of disease pathophysiology, with DNA methylation as the most accessible epigenetic measure. Diet is suggested to affect metabolism and health via epigenetic modifications. Thus, our aim was to explore the association between food consumption and DNA methylation.
### Methods
Epigenome-wide association studies were conducted in three cohorts: KORA FF4, TwinsUK, and Leiden Longevity Study, and 37 dietary exposures were evaluated. Food group definition was harmonized across the three cohorts. DNA methylation was measured using Infinium MethylationEPIC BeadChip in KORA and Infinium HumanMethylation450 BeadChip in the Leiden study and the TwinsUK study. Overall, data from 2293 middle-aged men and women were included. A fixed-effects meta-analysis pooled study-specific estimates. The significance threshold was set at 0.05 for false-discovery rate-adjusted p values per food group.
### Results
We identified significant associations between the methylation level of CpG sites and the consumption of onions and garlic [2], nuts and seeds [18], milk [1], cream [11], plant oils [4], butter [13], and alcoholic beverages [27]. The signals targeted genes of metabolic health relevance, for example, GLI1, RPTOR, and DIO1, among others.
### Conclusion
This EWAS is unique with its focus on food groups that are part of a Western diet. Significant findings were mostly related to food groups with a high-fat content.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s00394-022-03074-9.
## Introduction
Examining epigenetic modifications is a crucial step in exploring the effects of diet on human metabolism. Such modifications can occur at different biological levels, including DNA methylation, modification of histones and noncoding RNAs. The availability of precise measurement tools, the level of inter-individual variation and the expected effect sizes make DNA methylation the most appropriate research tool for diet and epigenetics studies [1].
DNA-methyl-transferase enzymes (DNMT) catalyze the generation of 5-methylcytosine, the main contributor of DNA methylation patterns, by utilizing methyl groups. Since 5-methylcytosine is degradable and insufficient activity of a maintenance DNMT enzyme can lead to loss of methylation with each cell division [2], there is a steady need for methyl group supply. Dietary intake represents the main source for methyl groups. Methionine, choline and its metabolite betaine [3], are all embedded in the C1 metabolism, contributing to the synthesis of the main methyl donor in human metabolism: s-adenosylmethionine. This makes the C1 metabolism the hypothesized primary link between diet and DNA methylation. However, research examining this link showed inconclusive results [4, 5], thus indicating that dietary methyl group donors and vitamins involved in the C1 metabolism are not major determinants for DNA methylation pattern changes. Analysis of food consumption data may better reflect synergistic effects of various food components as compared to single nutrients. Another link between diet and DNA methylation could be through modulation of inflammatory processes. Dietary compounds have been shown to be associated with systemic inflammation [6], which in turn can lead to disturbances in the balance of DNA methylation patterns [3].
So far, some analyses on the link between diet and global DNA methylation patterns [7], as well as diet and site-specific epigenetic changes [3], have been performed. In terms of site-specific analysis, the main focus of nutri-epigenomic research has been on epigenome-wide association studies (EWAS) of nutrients involved in human C1 metabolism [3, 4]. EWAS have also been carried out with dietary patterns and few single food groups [8–10]. However, a comprehensive EWAS at the food group level is lacking. Thus, our aim was to explore the association between food consumption and DNA methylation in population-based studies. We aimed to identify DNA methylation associations with food groups that (i) provide nutrients involved in the human C1 metabolism, (ii) are known in the literature for being associated with systemic inflammation (like red meat, cabbage or nuts), or (iii) were shown to be associated with cardio-metabolic disease risks (like sugar-sweetened beverages or vegetables) previously. The results of the EWAS conducted in three cohorts, KORA FF4 (KORA), TwinsUK (TUK) and Leiden Longevity Study (LLS), were included in this meta-analysis.
## Methods
The “Strengthening the Reporting of Observational Studies in Epidemiology—Nutritional Epidemiology (STROBE-nut)” checklist was used to report the findings of the present study [11]. For an overview of key points of methodology used in respective cohorts, see Table 1. Table 1Key points of methodology used in all three cohortsKORAFF4LLSTwinsUKDietary dataUsual dietary intake in g/day (methodology described as in [20]. Repeated 24 h food lists (246 items) and FFQ as adjusting variableFFQ (218 items), calculated in g/dayFFQ (131 items), calculated in g/dayMethylation dataInfinium MethylationEPIC BeadCHip (~ 850 k loci); preprocessing with the package minfiIllumina HumanMethylation450 array (~ 450 k loci); quality control using MethylAidIllumina HumanMethylation450 array (~ 450 k loci); preprocessing with ENmix and minfiStatistical modelLinear multivariable regression with technical covariates and food intake residual as exposure and methylation beta values as outcomeLinear multivariable regression with technical covariates and food intake residual as exposure and methylation beta values as outcomeMixed-model with cohort-specific random-effects and food intake residual as exposure and methylation beta values as outcome
## Populations
The KORA (Cooperative Health Research in the Region of Augsburg) FF4 study is the second follow-up of the population-based KORA S4 examination. It was conducted between 1999 and 2001 in the city of Augsburg and two surrounding counties in Germany. 4261 subjects aged 25–74 years were randomly drawn and agreed to participate in the S4 baseline study. 2279 of them also participated in the FF4 follow-up study ($\frac{2013}{2014}$). Details regarding the recruitment procedure have been published elsewhere [12]. Methylation data was available for 1928 subjects, and after exclusion of outliers (as described in the DNA methylation section), 1888 subjects remained. Further we excluded cases without available nutrition data ($$n = 541$$) or with blood cancer ($$n = 4$$). All participants met the criteria of acceptable caloric intake (500 kcal/d < x < 5000 kcal/d). Finally, 1322 subjects had full information on all covariates and were included in the EWAS.
The LLS consists of 1671 members of long-lived families (mean age 60 years) and their 744 partners (mean age: 60 years) as population controls. Dietary intake data in grams per day was collected from 1716 individuals. Members of long-lived families are very similar to the general population, although they have more favorable glucose tolerance [13], more favorable lipid parameters [14], and a lower prevalence of type-2 diabetes and myocardial infarction [15]. We analyzed them as one cohort of middle-aged people, and the current study was restricted to unrelated individuals. EWAS data and nutritional data was available on 507 individuals. All LLS participants met the criteria of acceptable caloric intake (500 kcal/d < x < 5000 kcal/d). Finally, 485 subjects had full information on all covariates and therefore were included in the EWAS.
The TwinsUK registry included over 14,000 research volunteer twin participants from the United Kingdom since 1992 [16]. Volunteers are monozygotic and dizygotic same-sex twins, predominately female ($82\%$), middle-aged (mean age 59) and over 18 years-old. Volunteers were recruited without selecting for any particular disease or trait and are mostly of European descent. Data on volunteers were collected through longitudinal questionnaires and clinical visits. The registry collected biological samples and further data through analysis of biological samples. Dietary data was collected for > 3000 female twins, and blood DNA methylation data obtained within two years of food frequency questionnaires was available for 493 of the female twins. The caloric intake of all twins included in this study was within the 500–5000 kcal/day range. A total of 487 female twins had information on all covariates and were included in the food group EWASs. A flowchart for the study samples and final analysis sample is given in Fig. 1.Fig. 1Flow chart of participant selection
## Dietary intake
In the KORA FF4 study, dietary data was collected via repeated 24 h food lists, comprising 246 items and a food frequency questionnaire (FFQ), including 148 items. The 24 h food list was derived from the NAKO Health study [17] and subjects were asked to report the type of food they consumed. The FFQ was adapted from the German version of the multilingual European Food Propensity Questionnaire [18]. Usual dietary intake was modeled with the amount consumed (if consumed at all) based on portion sizes from the Bavarian consumption study II [19], multiplied by the probability of consumption for an individual subject from at least two non-consecutive 24 h food lists. This was done to reduce measurement error, which is prominent in surveyed dietary data. Further information regarding assessment of dietary intake data and estimation of usual dietary intake is provided elsewhere [20]. The dietary data is classified in 17 main food groups and 71 food subgroups according to the EPIC SOFT classification [21]. Nutrient intake data was calculated based on the German food composition database, Bundeslebensmittelschlüssel, version 3.01 [22].
As part of the LLS study, participants were sent a 218-item FFQ constructed from the 104-item VetExpress FFQ, combined with the Dutch National Food Survey [23]. Food items were categorized into 17 main food groups and 67 subgroups, with combination formulae used to split intake where appropriate.
Dietary data in TwinsUK was collected through a 131-item FFQ comprising the food and drink items originally included in the EPIC Norfolk study [24]. The processing of this data was first described elsewhere [25]. Here, the daily intake of each item was calculated in g/day using the FETA software [26], and the default nutritional database used was McCance and Widdowson’s The Composition of Foods (5th edition) [27]. Food items were then allocated to food groups following the EPIC-Soft classification, matching items successfully to 32 of 33 food groups.
After regressing food group intake against energy intake, the predicted food group intake was added for the mean energy intake of the study population to the residuals in all three cohorts to improve interpretability. Furthermore, two dietary patterns were calculated in each study: the Alternate Healthy Eating Index 2010 (AHEI 2010) [28] and the Mediterranean Diet Score (MDS) [29]. The AHEI scoring system assesses foods and nutrients predictive of chronic disease risk (e.g. vegetables, nuts, alcohol). A lower score is associated with higher risk of chronic diseases of major importance for public health. Due to a lack of data, trans fats had to be excluded in the calculation of AHEI, resulting in a maximum of 100 points instead of 110. Usual dietary intake was transformed to servings per day with references reported in [28]. A high MDS reflects high adherence to a dietary pattern followed by people living in Mediterranean countries, relative to the sex-specific population median, except for alcohol, where a moderate amount of consumption is ranked highest. The MDS emphasizes the consumption of fish, legumes, fruits and nuts, cereals, and a high ratio of unsaturated to saturated lipids. The modification of the MDS is depicted in the fat ratio as a sum of monounsaturated and polyunsaturated fatty acids divided by saturated fatty acids. The MDS is a population-based dietary score. The definition of food groups was harmonized based on the EPIC-Soft classification that was used to classify each food in all three cohorts, ensuring that individual food items were attributed to the same food (sub-) group. Harmonization was not entirely possible for mushrooms, milk, yogurt, eggs and plant oils, because at least one study did not capture these items.
## DNA methylation data
KORA FF4: Using the EZ-96 DNA Methylation Kit (Zymo Research, Orange, CA, USA) in two separate batches ($$n = 488$$, $$n = 1440$$), genomic DNA from white blood cells (750 ng) from 1928 participants of the KORA FF4 study was bisulfite-converted. According to standard protocols provided by Illumina, subsequent methylation analysis was performed on an Illumina (San Diego, CA, USA) iScan platform using the Infinium MethylationEPIC BeadChip. For initial quality control and to generate methylation data export files, GenomeStudio software version 2011.1 with Methylation Module version 1.9.0 was used.
Further preprocessing and quality control of the data were performed in R v3.5.1 [30] with the package minfi v1.28.3 [31] and following primarily the CPACOR pipeline [32]. Raw intensities were read into R (command read.metharray) and background corrected (bgcorrect.illumina). Hereafter probes with detection p values > 0.01 were set to missing.
We removed problematic samples and probes before normalization. Forty samples were removed: 33 had median intensity < $50\%$ of the experiment-wide mean, or < 2000 arbitrary units, 9 (overlap of 4 with previous) had > $5\%$ missing values on the autosomes and 2 showed a mismatch between reported sex and that predicted by minfi. A total of 59,631 probes were removed (some overlapping multiple categories): 5786 with > $5\%$ missing values, cross-reactive probes as given in published lists ($$n = 44$$,493) [33, 34] and probes with SNPs with minor allele frequency < $5\%$ at the CG position ($$n = 11$$,370) or the single base extension ($$n = 5597$$) as given by minfi. Finally, probes from the Y chromosome ($$n = 379$$) and the X chromosome ($$n = 17$$,743, following quality control) were excluded from the analysis. A total of 788,106 probes remained.
Quantile normalization was then performed separately on the signal intensities divided into the 6 probe types: type I green unmethylated, type I green methylated, type I red unmethylated, type I red methylated, type II red, type II green [32]. For the X and Y chromosomes, men and women were processed separately; for the autosomes, Quantile normalization was performed for all samples together. Methylation beta values, a measure from 0 to 1 indicating the percentage of cells methylated at a given locus, were generated out of the transformed intensities. The threshold for exclusion of beta-value outliers was set at ± 3* interquartile range.
The Infinium MethylationEPIC Manifest file (available at www.illumina.com via product files) was used to map probes to genes and chromosomes using genome build 37. The Manifest file uses the gene database of the University of California Santa Cruz (UCSC). Informed consent for genetic studies was obtained from all subjects. The protocol for each study was approved by the institutional review board of each cohort.
LLS: Venous blood samples were taken from 732 unrelated individuals aged between 40 and 75 for whole blood DNA methylation profiling. The Zymo EZ DNA methylation kit (Zymo Research, Irvine, CA, USA) was used to bisulfite-convert 500 ng of genomic DNA, and 4 μl of bisulfite-converted DNA was measured on the Illumina HumanMethylation450 array using the manufacturer’s protocol (Illumina, San Diego, CA, USA). Preprocessing and normalization of the data were done as described in the DNAmArray workflow (https://molepi.github.io/DNAmArray_workflow/).
In brief, IDAT files were read using the minfi, while sample-level quality control (QC) was performed using MethylAid. Filtering of individual measurements was based on detection p value ($p \leq 0.01$), number of beads available (≤ 2), or zero values for signal intensity. Normalization was done using functional normalization as implemented in minfi, using five principal components extracted using the control probes for normalization. All samples or probes with more than $5\%$ of their values missing were removed.
TwinsUK: Whole-blood DNA methylation profiles in TwinsUK have previously been described [35]. Briefly, measurement of whole blood DNA methylation was performed using the Infinium HumanMethylation450 BeadChip (Illumina Inc, San Diego, CA) which profiles methylation levels at > 450,000 sites of the human genome. Processing of signals was performed using ENmix [36] for quality control, and minfi [31] to exclude samples with median methylated and unmethylated signals below 10.5. Both tools are available as Bioconductor software packages in R. During ENmix quality control checks, background and dye bias correction were performed as well as quantile normalization of signals. Bad probes and outlier samples were identified using standard parameter values, and signals with detP > 0.000001 and nbead < 3 were excluded. Beta-values were estimated after adjusting for differences in the distribution of type I and type II probe signals with the Regression on Correlated Probes (RCP) method. Beta-values out of the ± 3* interquartile distribution range were further excluded to match KORA FF4 exclusion criteria during association analyses. Maximum probe and sample missingness were set to $5\%$, and probes that mapped to multiple locations in the genome were removed. Overall, a total 430,768 autosomal probes and 487 individuals were included in our analysis.
Here we present the results of CpG sites that overlap between the Infinium MethylationEPIC and the Infinium HumanMethylation450 BeadChip, leaving a final number of at least 393,223 CpG sites per food group.
## Statistical analysis
The EWAS was carried out using linear regression analysis of the overlap of CpGs that were common in all three cohorts after quality control ($$n = 393$$,427). We performed a fixed-effect meta-analysis, because the estimated tau is considered imprecise with a small sample of studies [37]. In addition, we did a random-effects meta-analysis as a sensitivity analysis to follow-up on significant signals by evaluating the unadjusted p value. In context of the often high heterogeneity observed, we reported the I2 confidence interval, which is recommended in a small sample meta-analysis [38]. $$n = 1321$$ subjects from KORA FF4, $$n = 507$$ subjects from LLS and $$n = 487$$ subjects from TUK were included in the analysis, resulting in a sample size of $$n = 2315$.$ The primary outcome of this study was methylation beta values. We tested 37 food groups, nutrients and diet quality scores: potatoes, total vegetables, leafy vegetables, fruit vegetables, root vegetables, cabbage vegetables, onions and garlic, legumes, total fruits, nuts and seeds, milk, yogurt, cheese, cream, grain products, whole grain products, total meat, fresh red meat, processed meat, total fish, eggs, plant oils, butter, margarine, total sweets, cakes, sugar-sweetened beverages, coffee, tea, wine, beer, spirits, AHEI, MDS and folic acid. The residual method was used in each cohort to get intake estimates independent of total energy intake [39]. The p values were false-discovery rate (FDR) corrected ($p \leq 0.05$) using the Benjamini and Hochberg procedure. Methylation as beta values were regarded as the dependent variable. Exposures were food groups (g/day), dietary pattern scores (integer) and additionally folic acid in µg/day. Selected covariates for the model were sex, age (continuous), age squared, BMI (continuous), BMI squared, total caloric intake (continuous), alcohol in g/day (continuous—not applied in the analysis of wine, beer, spirits, AHEI and MDS), measured or estimated cell counts (using the Houseman-method [40]), smoking behavior (regular, former, never) and methylation plate and/or plate position as a technical variable. These were selected based on the literature and our own assessment of confounding with the disjunctive cause criterion [41]. Neutrophile granulocytes were excluded as a covariate due to multicollinearity. Only complete cases for every covariate were included in the analysis. To account for heterogeneity, we inspected and reported the p value of the Q-statistic and I2 for all CpGs that reached statistical significance. All statistical analyses were carried out with R statistical software version 4.0.4 [30]. Meta-analysis was performed with the metagen function of the meta package version 4.17.0 [42]. Figures were created using the ggplot2 package [43]. To evaluate whether CpGs were occuring in differentially methylated regions, DMRfinder [44] was used to test for the occurrence of significant CpGs < 1 kb apart as implemented in DNAmArray.
## Results
Overall, the results of 2316 participants were included in the meta-analysis. In KORA FF4, LLS and TUK, participants had a median age of 58, 59, and 60 years; a median BMI of 26.8, 25.1, and 25.6 kg/m2; and a median total energy intake of 1820, 1883, and 1808 kcal/day, respectively (Table 2). Intake of food groups for all cohorts can be found in Online Resource 1. Following a false-discovery rate adjustment with an alpha threshold at 0.05 (Table 3), we found 2 significant associations for onions and garlic consumption, 18 for nuts and seeds (Figs. 2a and 3), one for milk (Fig. 4), 11 for cream (Figs. 2b and 5), 13 for butter (Figs. 2c and 6), four for plant oils (Fig. 2d), five for wine, 16 for beer and six for spirits (for alcoholic beverages results, see Online Resource 2). We obtained no statistically significant signals for other food groups or dietary patterns. All significant CpGs were located in distinct regions (inter-CpG-distance > 1 kb). Some interesting annotated genes that are linked to metabolism include: GLI1 (Fig. 3), ATP5H, MYC, RPTOR, ASAM, FOXA2, and DIO1. Cg26633077 lies within the gene body of RPTOR, which could lead to suppressed gene expression with more cream consumption, as indicated by the negative effect size. *This* gene is involved in a signaling pathway that regulates cell growth in response to nutrient levels. Cg11798857 is positioned at the promoter of the FOXA2 gene. Combined with a positive effect size, this would indicate gene suppression as well. FOXA2 is a transcriptional activator for liver-specific genes. Figure 5 shows the forest plot of the CpG associated with MYC, which is a pro-fibrotic regulator. See Table 3 for information on all annotated genes and locations of the CpGs. Figure 7 displays examples of effect size estimates for the association of different food groups with DNA methylation. Two of the identified CpGs were detected in two distinct food groups, namely wine and beer. The first locus was annotated to the PHGDH gene, which is involved in the early steps of L-serine synthesis (cg14476101) and the second to TRA2B, which plays a role in mRNA processing (cg12825509).Table 2Population characteristics stratified by sex and cohortKORALLSTUKMaleFemaleOverallMaleFemaleOverallMaleFemalen6207021322240267507NA487Age in years (median [IQR])59.0 [49.0, 67.0]58.0 [48.0, 65.0]58.0 [49.0, 66.0]60.7 [55.9, 64.9]57.5 [53.0, 61.8]58.9 [54.5, 63.5]NA59.5 [52.2, 65.5]BMI in kg/m2 (median [IQR])27.4 [25.2, 30.5]26.1 [23.2, 29.8]26.8 [24.1, 30.2]25.3 [23.6, 27.2]24.6 [22.4, 26.9]25.1 [23.0, 27.1]NA25.6 [23.1, 29.3]Total energy intake (median [IQR])2093.8 [1885.9, 2332.7]1578.9 [1427.9, 1791.8]1819.8 [1550.8, 2114.8]2215.3 [1771.9, 2576.9]1730.5 [1465.0, 2008.3]1882.5 [1573.9, 2341.5]NA1808.1 [1473.1, 2199.3]Alcohol in g/day (median [IQR])13.2 [5.1, 24.6]2.7 [1.7, 5.3]5.0 [2.4, 13.9]16.1 [8.4, 28.0]9.0 [2.9, 19.1]12.5 [4.5, 89.6]NA5.4 [0.9, 12.6]Smoking behavior (%) Regular smoker96 (15.5)97 (13.8)193 (14.6)41 (11.5)44 (11.3)85 (11.6)NA60 (12.3) Former smoker283 (45.6)222 (31.6)505 (38.2)196 (54.9)162 (41.8)358 (48.9)NA162 (33.2) Never smoker241 (38.9)383 (54.6)624 (47.2)67 (18.8)132 (34.0)199 (27.2)NA265 (54.4) Physical activity: active (%)361 (58.2)452 (64.4)813 (61.5)240 [100]267 [100]507 [100]NANAValues are presented as median [Interquartilrange]Table 3Significant results of the meta-analyzed EWAS of KORA FF4, TwinsUK and Leiden Longevity StudyProbeIDStudies*Effect- size**p value FDRp value Q-statisticI2***FoodgroupChrRefGene nameRefGene groupRelation to CpG Islandcg06618277K–L–T− 1.39e-040.0500.2150.349 [0.000;0.789]Onions-garlic13N/AN/AN/Acg13970894K–L–T− 4.16e-040.0500.7060.000 [0.000;0.702]Onions-garlic10N/AN/AN_Shorecg03046445K–L–T9.94e-050.0190.0120.774 [0.266;0.930]Nuts-seeds12BHLHE41*1stExon*N_Shorecg05275153K–L–T− 4.39e-054.96e-053.14e-060.921 [0.801;0.969]Nuts-seeds4RGS12*Body*N/Acg08633290K–L–T− 6.72e-041.70e-052.79e-050.905 [0.748;0.964]Nuts-seeds19N/AN/AN_Shorecg09418283K–L–T7.31e-050.0050.0130.770 [0.252;0.929]Nuts-seeds12PAWR*1stExon*Islandcg10530560K–L–T− 1.40e-046.49e-070.0270.722 [0.060;0.918]Nuts-seeds12GLI1*5'UTR*S_Shelfcg11701148K–L–T2.64e-040.0050.1340.503 [0.000;0.856]Nuts-seeds8MYOM2TSS200N/Acg12430457K–L–T− 2.00e-040.0474.47e-040.870 [0.630;0.955]Nuts-seeds12SYT1*5'UTR*N/Acg12611195K–L–T− 1.22e-040.0389.64e-110.957 [0.905;0.980]Nuts-seeds6PPP1R14CBodyN/Acg13471114K–L–T2.21e-040.0360.1900.397 [0.000;0.814]Nuts-seeds2OTX1TSS1500Islandcg14436861K–L–T− 2.70e-044.96e-052.89e-070.934 [0.840;0.972]Nuts-seeds11WEE1*3'UTR*N/Acg14828673K–L–T− 1.17e-040.0270.0220.739 [0.126;0.922]Nuts-seeds8TOP1MTBodyN/Acg15864779K–L–T− 2.77e-050.0050.9950.000 [0.000;0.000]Nuts-seeds17ATP5H*TSS200*Islandcg16790682K–L–T8.13e-050.0380.0070.798 [0.360;0.936]Nuts-seeds12PIP4K2C*TSS200*Islandcg21251785K–L–T− 3.12e-040.0202.08e-100.955 [0.901;0.980]Nuts-seeds9TRPM3*TSS1500*N/Acg23415756K–L–T5.25e-050.0010.2730.229 [0.000;0.920]Nuts-seeds17NTN1BodyIslandcg25554998K–L–T− 2.43e-040.0270.0050.811 [0.411;0.940]Nuts-seeds11N/AN/AN_Shorecg27344289K–L–T− 2.49e-041.19e-040.0020.834 [0.497;0.945]Nuts-seeds5FLJ41603BodyN/Acg27496650K–L–T5.30e-050.0080.3630.012 [0.000;0.897]Nuts-seeds8TOXTSS1500Islandcg14732699K–T5.60e-060.0490.1400.540 [0.000;0.887]Milk8MYCBodyIslandcg03846926K–L–T2.26e-040.0480.0070.799 [0.363;0.936]Cream10C10orf1405'UTRS_Shorecg06947913K–L–T2.82e-040.0203.95e-140.968 [0.934;0.984]Cream12FAIM2TSS200Islandcg08846079K–L–T− 4.73e-040.0440.5180.000 [0.000;0.842]Cream1N/AN/AN/Acg09398214K–L–T1.21e-040.0385.70e-080.940 [0.859;0.975]Cream17MARCH10*Body*Islandcg10156125K–L–T3.86e-040.0205.53e-110.958 [0.908;0.981]Cream4UGT8*5'UTR*Islandcg13331940K–L–T1.82e-040.0367.79e-160.971 [0.943;0.986]Cream15MYO1EBodyIslandcg13923646K–L–T− 3.30e-040.0480.2340.311 [0.000;0.928]Cream1N/AN/AN/Acg17353893K–L–T− 2.15e-040.0202.07e-310.986 [0.975;0.992]Cream7CLIP2*Body*Islandcg22028181K–L–T1.31e-040.0267.67e-210.978 [0.959;0.989]Cream15CYFIP1TSS200Islandcg25734572K–L–T1.46e-040.0369.21e-050.892 [0.707;0.960]Cream19N/AN/AIslandcg26633077K–L–T− 6.99e-040.0485.44e-280.984 [0.972;0.991]Cream17RPTOR*Body*S_Shelfcg02488288K–L3.29e-040.0440.0060.868 [0.481;0.967]Plant-oils5SPRY4*TSS1500*N_Shorecg03995571K–L7.05e-050.0100.0150.833 [0.303;0.960]Plant-oils8FAM49BTSS1500Islandcg11189177K–L− 4.92e-040.0443.95e-040.920 [0.726;0.977]Plant-oils17ABR*Body*S_Shorecg18419070K–L− 4.78e-040.0280.0180.820 [0.239;0.957]Plant-oils3SEMA5BBodyN/Acg02924347K–L–T6.31e-050.0490.8750.000 [0.000;0.220]Butter16ABCA17P*TSS1500*Islandcg05781609K–L–T− 2.57e-040.0202.48e-140.968 [0.935;0.984]Butter1COL24A1BodyN/Acg07410571K–L–T5.11e-050.0490.7180.000 [0.000;0.685]Butter20DDRGK1TSS200Islandcg11798857K-L–T7.20e-040.0340.6930.000 [0.000;0.716]Butter20FOXA2*TSS1500*N_Shorecg11934386K–L–T4.75e-052.49e-045.55e-100.953 [0.896;0.979]Butter7C7orf41*1stExon*Islandcg13934553K–L–T− 3.64e-050.0205.21e-040.868 [0.621;0.954]Butter1PLEKHG5*Body*Islandcg14046757K–L–T5.29e-051.63e-041.65e-200.978 [0.959;0.988]Butter14ZC3H14*TSS200*Islandcg14981983K–L–T3.95e-050.0341.03e-120.964 [0.924;0.983]Butter19ZNF527TSS200Islandcg18247124K–L–T− 2.85e-045.45e-045.87e-120.961 [0.918;0.982]Butter11ASAM3'UTRN/Acg19200140K–L–T− 4.27e-040.0490.1880.402 [0.000;0.816]Butter20N/AN/AN_Shelfcg19526600K–L–T1.70e-040.0340.8250.000 [0.000;0.461]Butter1DIO1*1stExon*N/Acg26351764K–L–T6.36e-050.0241.03e-120.964 [0.924;0.983]Butter20FAM83DTSS200Islandcg26502414K–L–T4.70e-050.0250.7980.000 [0.000;0.538]Butter17METT10D5'UTRN_ShoreUCSC RefGene Name *Target* gene names from the UCSC database, UCSC RefGene Group Describing CpG position. TSS1500 = 200–1500 bases upstream of the Transcription start site (TSS), 5-UTR Within the 5' untranslated region, between the TSS and the ATG start site; Body = Between the ATG and stop codon, irrespective of the presence of introns, exons, TSS or promoters, 3'UTR Between the stop codon and the poly A signal, Relation to UCSC CpG Island The location of the CpG relative to the CpG Island. Shore = 0–2 kb from Island; Shelf = 2–4 kb from Island, N Upstream (5') of CpG Island, S Downstream (3') of CpG Island [66]*K is short for KORA FF4 inclusion; L is short for Leiden longevity study inclusion; T is short for TwinsUK inclusion**Effect sizes are %-methylation change per g/day residual intake***I2 is reported with $95\%$-confidence interval in brackets as calculated by the metagen function of the meta package**After a* gene name indicates available splice variantsFig. 2Volcano plots with the unadjusted p value on the y-axis. Every significant CpG after FDR adjustment is marked with its probeID. Effect size on the x-axis is %-methylation change per gram residual/day. a nuts and seeds, b cream, c butter, d plant oils in g/day residualsFig. 3Forest plot for the association between cg10530560 methylation level and nuts and seeds consumption. Effect size on the x-axis is %-methylation change per gram residual/day with a $95\%$ confidence intervalFig. 4Forest plot for the association between cg14732699 methylation level and milk consumption. Effect size on the x-axis is %-methylation change per gram residual/day with a $95\%$ confidence intervalFig. 5Forest plot for the association between cg26633077 methylation level and cream consumption. Effect size on the x-axis is %-methylation change per gram residual/day with a $95\%$ confidence intervalFig. 6Forest plot for the association between cg11798857 methylation level and butter consumption. Effect size on the x-axis is %-methylation change per gram residual/day with a $95\%$ confidence intervalFig. 7Combined forest plot of pooled estimators. One significant finding in different food groups is shown to get a perspective for the different effect sizes. Effect size on the x-axis is %-methylation change per gram residual/day with a $95\%$ confidence interval Many of the food groups for which we observed significant associations are high in fat content. However, in contrast to this statement, we found no significant signals in case of cheese, eggs or margarine consumption. We explored whether significant CpGs identified in one food group may also be associated with another (high-fat) food group. We chose the example of the findings for nuts and seeds, and Table 4 displays the results. In total for all explored food groups, 10 signals from the food group nuts and seeds showed an unadjusted p value < 0.05 in other high-fat food groups, and only two of them had the same direction of effect [cg09418283, cg10530560]. We did not observe any significant association for the consumption of food groups that are well known for their specific phytochemical content, such as leafy vegetables, cabbage vegetables and fruits, or coffee and tea. We also did not observe any DNA methylation association with AHEI or MDS.Table 4p values for high-fat food groups for loci with significant associations with the food group nuts and seedsCpGNuts-seedsButterCheeseCreamEggsMargarinePlant-oilsProcessed-meatcg030464455.27e–07*0.2170.4130.3280.5490.1020.9010.387cg052751534.10e–10*0.7250.7600.1890.7660.3770.1700.600cg086332908.66e–11*0.1180.1180.8530.4890.1700.7120.915cg094182831.04e–07*0.4270.9420.1850.021*0.7580.4520.082cg105305601.65e–12*0.8210.046*0.9290.043*0.7200.1710.088cg117011481.11e–07*0.2510.2300.5740.6720.7850.2990.764cg124304572.15e–06*0.6360.9860.3430.7900.3110.3040.423cg126111951.66e–06*0.2230.033*0.7480.4790.013*0.4430.051cg134711141.37e–06*0.8700.9790.6080.2400.3720.4520.085cg144368615.04e–10*0.8650.5980.2460.9680.013*0.8430.053cg148286738.97e–07*0.6100.7510.9420.6270.9930.9460.589cg158647798.95e–08*0.2320.7730.2100.8100.8190.8980.082cg167906821.57e–06*0.5260.4690.9290.7440.021*0.4720.169cg212517856.06e–07*0.5180.5470.3210.4270.6650.2540.890cg234157561.68e–08*0.045*0.9990.7450.8120.9560.6060.317cg255549989.77e–07*0.0880.8760.4090.5510.4010.6880.502cg273442891.51e–09*0.1550.4700.9210.3960.7300.9210.148cg274966502.11e–07*0.8120.3940.0800.002*0.4610.1730.284Underlining indicates same direction of effect*Indicates p value < 0.05 In many cases, heterogeneity between studies was high, with I2 > 0.8 (Table 3). Reasons could be differences in dietary assessment methods across studies or differences between populations. To explore this further, we also performed a random-effects meta-analysis, which reproduced 2 of 2 signals in onions and garlic [cg06618277; cg13970894], 7 out of 18 in nuts and seeds [cg03046445; cg11701148; cg13471114; cg15864779; cg23415756; cg27344289; cg27496650], 0 of 1 in milk, 3 of 11 in cream [cg03846926; cg08846079; cg13923646], 6 of 13 in butter [cg02924347; cg07410571; cg11798857; cg19200140; cg19526600; cg26502414], 2 of 4 in plant oils [cg02488288; cg18419070], 5 of 5 in wine [cg06690548; cg07856667; cg08033640; cg12825509; cg14476101], 10 of 16 in beer [cg01794805; cg03044533; cg03725309; cg06469895; cg07714319; cg08984272; cg10797552; cg11100157; cg11376147; cg15821562], and 1 of 6 in spirits [cg09307985]. Detailed results are listed in Online Resource 3. For further information regarding heterogeneity and effect size distribution, see Online Resource 4, where the p value distribution, I2 distribution and estimated tau distribution for every analyzed food group with significant signals are displayed. Online Resource 5 presents volcano plots for every analyzed food group.
## Discussion
This work explored many food groups that have not been studied in context of human DNA methylation, e.g., nuts and seeds, or added fats and oils. Our main finding is that the majority of analyzed food groups did not show significant associations with blood DNA methylation, and that significant associations with methylation levels were observed primarily for food groups high in fat content.
We evaluated whether the CpGs we found to be associated with food groups in this analysis had been previously identified in EWAS for other traits using the EWAS catalog [45]. Many significant associations (cg12825509, cg14476101, cg06690548, cg11376147, cg14476101, cg06469895, cg12825509, cg18120259, cg03725309, cg07714319, cg16246545, cg15821562, cg03044533, cg26282731, cg11100157, cg01794805) observed in our analysis on alcoholic beverages could be attributed to their ethanol content, and are already reported in the EWAS catalog for their association with alcohol consumption. Loci cg12430457 (nuts and seeds), cg06947913 (cream) and cg14046757, cg13934553, cg26502414, cg07410571 (butter) were all reported to be associated with rheumatoid arthritis [45]. One signal in nuts and seeds, cg14828673, was previously reported to be associated with waist-to-hip-ratio [45]. Surprisingly, cg13331940, which was significantly associated with cream, was previously reported to be associated with alcohol consumption per day [45]. None of our remaining significant signals were associated with metabolic traits, metabolic diseases or dietary exposures in past EWAS.
We found several interesting signals in the food group nuts and seeds for which there is a reported connection in the literature. Cg10530560 maps to the gene GLI1 and showed a significant association with the food group nuts and seeds. GLI1 is a transcription factor which gets activated by and is a marker of the sonic hedgehog pathway [46]. A negative effect size and the location in the gene body could be interpreted as a downregulation in gene expression, which would fit the downregulation of genes in the hedgehog pathway triggered by a diet high in either saturated or unsaturated fatty acids as reported by Mehmood et al. [ 46]. Deactivation of the hedgehog pathway is suggested to be associated with fat accumulation [47]. Another significant signal (cg15864779, located within the ATP5H gene) could possibly be explained by the high-methionine content in nuts. A high-methionine diet alters the ATP5H expression dependent on the paraoxonase genotype. Paraoxonase-positive mice have downregulated ATP5H, whereas paraoxonase-negative mice had upregulated ATP5H. This interaction is tightly linked to energy generation in the hyperhomocysteinemic liver [48].
The one CpG linked to milk consumption, cg14732699, is associated with MYC, a pro-fibrotic regulator. Butyric acid as a component in bovine milk triglycerides [49] could have affected the methylation of this MYC CpG site. One study identified butyrate as a protective agent for diet-induced non-alcoholic hepatic steatosis and liver fibrosis by downregulating, among other, MYC [50]. Another study observed an association between oleic acid, the main monounsaturated fatty acid in bovine milk, and the gene MYC. It showed that oleic acid promotes colorectal cancer development by upregulation of MYC, among others [51].
We also observed significant associations with cream consumption, another high-fat food group. CLIP2 associated with cg17353893 is reported to be downregulated under a high-fat diet regimen [52]. This downregulation also fits our results, where cg17353893 has a negative effect size and is located within the gene body [53]. The CYFIP1 (cg22028181) gene is a homolog of CYFIP2, which was described as a genetic factor underlying compulsive-like binge eating in mice [54]. CYFIP1 haploinsufficiency shows similar properties by increasing compulsive-like behavior and modulation of palatable food intake in mice [55]. Cream is a food with very high energy density; thus, dependent on the direction of the relationship, gene methylation could be either the cause or effect of cream consumption. Calorie intake impacts the gene associated with cg26633077, RPTOR, as shown in the stabilization of the MTOR-RPTOR association by nutrient deprivation, leading to inhibition of MTOR activity [56]. Despite the inhibition of the anabolic regulator MTOR, one study found that RPTOR null mice gained less weight, most likely due to reduced food intake in a high-fat diet, when compared to wild type mice [57]. It is worth noting that there was very high heterogeneity observed for cg26633077.
More insight into the association between CpG methylation and adiposity can be given by significant associations with butter intake. Cg18247124 is located in adipocyte adhesion molecule (ASAM), which was found to be correlated with BMI in human subcutaneous adipose tissue, and ASAM mRNA is increased during adipocyte differentiation in mice and humans [58]. Also, cg11798857 in the transcription start site of FOXA2 was a significant finding in our analysis. FOXA2 mRNA, related to fatty acid oxidation in the liver, was increased in mice fed with pre- and probiotics, along with improved insulin sensitivity and reduced adipocyte size [59]. DIO1 (cg19526600) encodes for type I iodothyronine deiodinase and can affect lipid metabolism through its effects on thyroid hormones. Xia et al. [ 60] reported that mice with an obese phenotype experienced ameliorated hepatic steatosis if the intervention was exercise, low-fat, quercetin or calorie restriction, possibly by affecting miRNAs, e.g. miR-383 and miR-146b to elevate DIO1 expression.
Comparing all of our results to previous EWAS is quite difficult because of the lack of EWAS analyzing food groups. Karabegovic et al. performed an EWAS in four European cohorts analyzing tea and coffee consumption. We tried to replicate the findings of Karabegovicet al. [ 61] for coffee with a Bonferroni adjusted alpha (0.05) solely in the KORA FF4 study, but failed, except for cg25648203, for which we could confirm the direction of effect. We did not observe significant signals in our meta-analysis of coffee and DNA methylation. There are obvious differences that could explain the failed replication. The study from Karabegovic et al. has ten times the sample size that our study has, which greatly increases the power to detect such signals. Also, while Karabegovic et al. used their coffee intake in cups per day, ours is measured as usual dietary intake in g/day and used as residuals in the linear regression.
Several pathways could assist in explaining the associations between food groups and methylation changes. One of our hypotheses was that the link between diet and inflammation could influence DNA methylation levels. Nuts are known for their high unsaturated and low saturated fatty acid content, which can affect homeostasis of inflammation and therefore impact DNA methylation patterns [3]. However, this argument has to be evaluated for every food group separately. Nuts, butter, plant oils and cream have a high-fat content in common, which could also either trigger or reduce inflammation in mice [62], but not in obese humans without metabolic disturbances [63]. Other food groups like red meat or cabbage that were associated with inflammatory processes in the past have not yielded any signals. Further studies are needed to confirm our results that the association of, for example, red meat and cabbage with inflammation are independent of DNA methylation.
Although our results hint at a pattern suggesting that the high-fat content of the food groups seems to be a major determinant in the modification of methylation patterns, the results as described in Table 4 do not confirm this regarding the significant signals found for the food group nuts and seeds. Additionally, we observed only a few or no significant signals in other high-fat content food groups like fish, processed meat and cheese.
Despite the focus on food groups, we also analyzed folic acid intake in this meta-analysis. We found no significant association here, which supports the theory that nutrients involved in the pathway that leads to the main methyl donor S-adenosylmethionine have at most a weak isolated impact on DNA methylation, as already demonstrated by Mandaviya et al. [ 4] and Dugué et al. [ 5].
Our study has several strengths. It is the first study which examined in three independent studies the intake of many food groups and subgroups for their association with DNA methylation. We harmonized the dietary intake data of KORA, LLS and TUK to ensure that same food classification scheme was applied. Residual confounding by energy intake was best considered by calculating food group residuals and using these in our models.
The analytic method to estimate the methylation level was similar across studies; the larger set of CpG sites measured in KORA was not considered here since the analyses were based on overlapping CpG sites across all studies. Our study also has limitations. We did not perform a food substitution model. Thus, we could not exclude the possibility that another food can act as a compensating mechanism. Also, since we have no gene expression data, conclusions about the effect of methylation change have to be confirmed in mechanistic studies. Additionally, we only had access to whole blood cells; therefore, we cannot draw any tissue-specific conclusions. Finally, there could be limited correlation of the same CpGs in the Illumina 450 k Chip used by TwinsUK and LLS and in the EPIC 850 k Chip used by KORA [64]. These results need replication to further clarify the association of food groups with white blood cell DNA methylation. As a fixed-effect model was chosen, extrapolating conclusions to different populations has to be done carefully. Although the random-effects meta-analysis more closely resembles the data reality than a fixed-effects analysis, because of the assumption of underlying distinct true means, the results should not be valued over the fixed-effects analysis, since an imprecise tau is included in our random-effects model [37]. We are aware of the debate around the focus on p values [65], but since we needed a threshold to decide if a CpG in this explorative study represents a meaningful finding, we deemed this the best fit. Due to the design of this study, we cannot draw conclusions regarding causality. Lastly, since dietary intake was assessed by FFQ’s (TUK, LLS) or a blended approach using repeated 24 h food list and an FFQ, exposure data may suffer from differential bias(including self-reporting bias).
## Conclusions
This study analyzed a broad range of different food groups and subgroups from three cohorts for their association with CpG methylation level. There were no significant associations for almost all vegetable or fruit food (sub-) groups. Rather, we observed interesting signals in food groups rich in fat, such as nuts and seeds, cream, butter, and plant oils. Some of the annotated genes seem to support the frequently observed effects of high-fat diets on DNA methylation in experimental studies. However, the results need replication in other cohorts with appropriate sample sizes to overcome some of the limitations present in this study.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (XLSX 16 KB)Supplementary file2 (XLSX 13 KB)Supplementary file3 (XLSX 19 KB)Supplementary file4 (PDF 1073 KB)Supplementary file5 (PDF 3114 KB)
## References
1. Sapienza C, Issa J-P. **Diet, nutrition, and cancer epigenetics**. *Annu Rev Nutr* (2016.0) **36** 665-681. DOI: 10.1146/annurev-nutr-121415-112634
2. Moore LD, Le T, Fan G. **DNA methylation and its basic function**. *Neuropsychopharmacology* (2013.0) **38** 23-38. DOI: 10.1038/npp.2012.112
3. Mahmoud AM, Ali MM. **Methyl donor micronutrients that modify DNA methylation and cancer outcome**. *Nutrients* (2019.0). DOI: 10.3390/nu11030608
4. Mandaviya PR, Joehanes R, Brody J. **Association of dietary folate and vitamin B-12 intake with genome-wide DNA methylation in blood: a large-scale epigenome-wide association analysis in 5841 individuals**. *Am J Clin Nutr* (2019.0) **110** 437-450. DOI: 10.1093/ajcn/nqz031
5. Dugué P-A, Chamberlain JA, Bassett JK. **Overall lack of replication of associations between dietary intake of folate and vitamin B-12 and DNA methylation in peripheral blood**. *Am J Clin Nutr* (2020.0) **111** 228-230. DOI: 10.1093/ajcn/nqz253
6. Shivappa N, Steck SE, Hurley TG, Hussey JR, Hébert JR. **Designing and developing a literature-derived, population-based dietary inflammatory index**. *Public Health Nutr* (2014.0) **17** 1689-1696. DOI: 10.1017/S1368980013002115
7. Noro F, Marotta A, Bonaccio M. **Fine-grained investigation of the relationship between human nutrition and global DNA methylation patterns**. *Eur J Nutr* (2022.0) **61** 1231-1243. DOI: 10.1007/s00394-021-02716-8
8. Ek WE, Tobi EW, Ahsan M. **Tea and coffee consumption in relation to DNA methylation in four European cohorts**. *Hum Mol Genet* (2017.0) **26** 3221-3231. DOI: 10.1093/hmg/ddx194
9. Do WL, Whitsel EA, Costeira R. **Epigenome-wide association study of diet quality in the Women's Health Initiative and TwinsUK cohort**. *Int J Epidemiol* (2021.0) **50** 675-684. DOI: 10.1093/ije/dyaa215
10. Ma J, Rebholz CM, Braun KVE. **Whole blood DNA methylation signatures of diet are associated with cardiovascular disease risk factors and all-cause mortality**. *Circ Genom Precis Med* (2020.0) **13** e002766. DOI: 10.1161/CIRCGEN.119.002766
11. Lachat C, Hawwash D, Ocké MC. **Strengthening the reporting of observational studies in epidemiology-nutritional epidemiology (STROBE-nut): an extension of the STROBE statement**. *PLoS Med* (2016.0) **13** e1002036. DOI: 10.1371/journal.pmed.1002036
12. Kowall B, Rathmann W, Stang A. **Perceived risk of diabetes seriously underestimates actual diabetes risk: the KORA FF4 study**. *PLoS ONE* (2017.0) **12** e0171152. DOI: 10.1371/journal.pone.0171152
13. Rozing MP, Westendorp RGJ, de Craen AJM. **Favorable glucose tolerance and lower prevalence of metabolic syndrome in offspring without diabetes mellitus of nonagenarian siblings: the Leiden longevity study**. *J Am Geriatr Soc* (2010.0) **58** 564-569. DOI: 10.1111/j.1532-5415.2010.02725.x
14. Vaarhorst AAM, Beekman M, Suchiman EHD. **Lipid metabolism in long-lived families: the Leiden Longevity Study**. *Age (Dordr)* (2011.0) **33** 219-227. DOI: 10.1007/s11357-010-9172-6
15. Westendorp RGJ, van Heemst D, Rozing MP. **Nonagenarian siblings and their offspring display lower risk of mortality and morbidity than sporadic nonagenarians: the Leiden Longevity Study**. *J Am Geriatr Soc* (2009.0) **57** 1634-1637. DOI: 10.1111/j.1532-5415.2009.02381.x
16. Spector TD, MacGregor AJ. **The St. Thomas' UK adult twin registry**. *Twin Res* (2002.0) **5** 440-443. DOI: 10.1375/twin.5.5.440
17. Freese J, Feller S, Harttig U. **Development and evaluation of a short 24-h food list as part of a blended dietary assessment strategy in large-scale cohort studies**. *Eur J Clin Nutr* (2014.0) **68** 324-329. DOI: 10.1038/ejcn.2013.274
18. Illner A-K, Harttig U, Tognon G. **Feasibility of innovative dietary assessment in epidemiological studies using the approach of combining different assessment instruments**. *Public Health Nutr* (2011.0) **14** 1055-1063. DOI: 10.1017/S1368980010003587
19. 19.Himmerich S, Gedrich K, Karg G, Wolfram G, Seiler H, Linseisen J (2022) Bayerische Verzehrsstudie (BVS) II: Abschlussbericht [Cited 2022 May 13] Available from: URL: http://ernaehrungsdenkwerkstatt.de/fileadmin/user_upload/EDWText/TextElemente/Ernaehrungserhebungen/Bayerische_Verzehrsstudie_zwei.pdf
20. Mitry P, Wawro N, Six-Merker J. **Usual Dietary intake estimation based on a combination of repeated 24-h food lists and a food frequency questionnaire in the KORA FF4 cross-sectional study**. *Front Nutr* (2019.0) **6** 145. DOI: 10.3389/fnut.2019.00145
21. Slimani N, Deharveng G, Charrondière RU. **Structure of the standardized computerized 24-h diet recall interview used as reference method in the 22 centers participating in the EPIC project**. *Comput Methods Programs Biomed* (1999.0) **58** 251-266. DOI: 10.1016/S0169-2607(98)00088-1
22. 22.Max-Rubner Institut (MRI)). Bundeslebensmittelschlüssel: BLS-Version 3.02 [Cited 2021 April 21] Available from: URL: https://www.blsdb.de/
23. Streppel MT, de Vries JHM, Meijboom S. **Relative validity of the food frequency questionnaire used to assess dietary intake in the Leiden Longevity Study**. *Nutr J* (2013.0) **12** 75. DOI: 10.1186/1475-2891-12-75
24. Day N, Oakes S, Luben R. **EPIC-Norfolk: study design and characteristics of the cohort. European Prospective Investigation of Cancer**. *Br J Cancer* (1999.0) **80** 95-103. PMID: 10466767
25. Teucher B, Skinner J, Skidmore PML. **Dietary patterns and heritability of food choice in a UK female twin cohort**. *Twin Res Hum Genet* (2007.0) **10** 734-748. DOI: 10.1375/twin.10.5.734
26. Mulligan AA, Luben RN, Bhaniani A. **A new tool for converting food frequency questionnaire data into nutrient and food group values: FETA research methods and availability**. *BMJ Open* (2014.0) **4** e004503. DOI: 10.1136/bmjopen-2013-004503
27. 27.McCance RA, Widdowson EM (2004) McCance and Widdowson's the composition of foods. 6th summary ed. Repr. Royal Society of Chemistry, Cambridge
28. Chiuve SE, Fung TT, Rimm EB. **Alternative dietary indices both strongly predict risk of chronic disease**. *J Nutr* (2012.0) **142** 1009-1018. DOI: 10.3945/jn.111.157222
29. Couto E, Boffetta P, Lagiou P. **Mediterranean dietary pattern and cancer risk in the EPIC cohort**. *Br J Cancer* (2011.0) **104** 1493-1499. DOI: 10.1038/bjc.2011.106
30. 30.R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2021. Available from: https://www.R-project.org. Accessed 1 June 2022
31. Aryee MJ, Jaffe AE, Corrada-Bravo H. **Minfi: a flexible and comprehensive Bioconductor package for the analysis of Infinium DNA methylation microarrays**. *Bioinformatics* (2014.0) **30** 1363-1369. DOI: 10.1093/bioinformatics/btu049
32. Lehne B, Drong AW, Loh M. **A coherent approach for analysis of the Illumina HumanMethylation450 BeadChip improves data quality and performance in epigenome-wide association studies**. *Genome Biol* (2015.0) **16** 37. DOI: 10.1186/s13059-015-0600-x
33. Pidsley R, Zotenko E, Peters TJ. **Critical evaluation of the Illumina MethylationEPIC BeadChip microarray for whole-genome DNA methylation profiling**. *Genome Biol* (2016.0) **17** 208. DOI: 10.1186/s13059-016-1066-1
34. McCartney DL, Walker RM, Morris SW, McIntosh AM, Porteous DJ, Evans KL. **Identification of polymorphic and off-target probe binding sites on the Illumina Infinium MethylationEPIC BeadChip**. *Genom Data* (2016.0) **9** 22-24. DOI: 10.1016/j.gdata.2016.05.012
35. Kurushima Y, Tsai P-C, Castillo-Fernandez J. **Epigenetic findings in periodontitis in UK twins: a cross-sectional study**. *Clin Epigenetics* (2019.0) **11** 27. DOI: 10.1186/s13148-019-0614-4
36. Xu Z, Niu L, Li L, Taylor JA. **ENmix: a novel background correction method for Illumina HumanMethylation450 BeadChip**. *Nucleic Acids Res* (2016.0) **44** e20. DOI: 10.1093/nar/gkv907
37. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. **A basic introduction to fixed-effect and random-effects models for meta-analysis**. *Res Synth Methods* (2010.0) **1** 97-111. DOI: 10.1002/jrsm.12
38. von Hippel PT. **The heterogeneity statistic I(2) can be biased in small meta-analyses**. *BMC Med Res Methodol* (2015.0) **15** 35. DOI: 10.1186/s12874-015-0024-z
39. Willett W. *Nutritional epidemiology* (2013.0)
40. Houseman EA, Accomando WP, Koestler DC. **DNA methylation arrays as surrogate measures of cell mixture distribution**. *BMC Bioinform* (2012.0) **13** 86. DOI: 10.1186/1471-2105-13-86
41. VanderWeele TJ. **Principles of confounder selection**. *Eur J Epidemiol* (2019.0) **34** 211-219. DOI: 10.1007/s10654-019-00494-6
42. Balduzzi S, Rücker G, Schwarzer G. **How to perform a meta-analysis with R: a practical tutorial**. *Evid Based Ment Health* (2019.0) **22** 153-160. DOI: 10.1136/ebmental-2019-300117
43. 43.ggplot2: elegant graphics for data analysis. Springer-Verlag New York; 2016. Available from: URL: https://ggplot2.tidyverse.org. Accessed 1 June 2022
44. Slieker RC, Bos SD, Goeman JJ. **Identification and systematic annotation of tissue-specific differentially methylated regions using the Illumina 450k array**. *Epigenetics Chromatin* (2013.0) **6** 26. DOI: 10.1186/1756-8935-6-26
45. Battram T, Yousefi P, Crawford G. **The EWAS Catalog: a database of epigenome-wide association studies**. *Wellcome Open Res* (2021.0). DOI: 10.12688/wellcomeopenres.17598.1
46. Mehmood R, Sheikh N, Khawar MB. **High-fat diet induced hedgehog signaling modifications during chronic kidney damage**. *Biomed Res Int* (2020.0) **2020** 8073926. DOI: 10.1155/2020/8073926
47. Qiu S, Cho JS, Kim JT. **Caudatin suppresses adipogenesis in 3T3-L1 adipocytes and reduces body weight gain in high-fat diet-fed mice through activation of hedgehog signaling**. *Phytomedicine* (2021.0) **92** 153715. DOI: 10.1016/j.phymed.2021.153715
48. Suszyńska-Zajczyk J, Jakubowski H. **Paraoxonase 1 and dietary hyperhomocysteinemia modulate the expression of mouse proteins involved in liver homeostasis**. *Acta Biochim Pol* (2014.0) **61** 815-823. DOI: 10.18388/abp.2014_1851
49. Haug A, Høstmark AT, Harstad OM. **Bovine milk in human nutrition—a review**. *Lipids Health Dis* (2007.0) **6** 25. DOI: 10.1186/1476-511X-6-25
50. Gart E, van Duyvenvoorde W, Toet K. **Butyrate protects against diet-induced NASH and liver fibrosis and suppresses specific non-canonical TGF-β signaling pathways in human hepatic stellate cells**. *Biomedicines* (2021.0). DOI: 10.3390/biomedicines9121954
51. Zhang Y, Wang Di, Lv B. **Oleic acid and insulin as key characteristics of T2D promote colorectal cancer deterioration in xenograft mice revealed by functional metabolomics**. *Front Oncol* (2021.0) **11** 685059. DOI: 10.3389/fonc.2021.685059
52. Dreja T, Jovanovic Z, Rasche A. **Diet-induced gene expression of isolated pancreatic islets from a polygenic mouse model of the metabolic syndrome**. *Diabetologia* (2010.0) **53** 309-320. DOI: 10.1007/s00125-009-1576-4
53. Tirado-Magallanes R, Rebbani K, Lim R, Pradhan S, Benoukraf T. **Whole genome DNA methylation: beyond genes silencing**. *Oncotarget* (2017.0) **8** 5629-5637. DOI: 10.18632/oncotarget.13562
54. Kirkpatrick SL, Goldberg LR, Yazdani N. **Cytoplasmic FMR1-interacting protein 2 is a major genetic factor underlying binge eating**. *Biol Psychiatry* (2017.0) **81** 757-769. DOI: 10.1016/j.biopsych.2016.10.021
55. Babbs RK, Beierle JA, Ruan QT. **Cyfip1 haploinsufficiency increases compulsive-like behavior and modulates palatable food intake in mice: dependence on Cyfip2 genetic background, parent-of origin, and sex**. *G3 (Bethesda)* (2019.0) **9** 3009-3022. DOI: 10.1534/g3.119.400470
56. Kim D-H, Sarbassov DD, Ali SM. **mTOR interacts with raptor to form a nutrient-sensitive complex that signals to the cell growth machinery**. *Cell* (2002.0) **110** 163-175. DOI: 10.1016/S0092-8674(02)00808-5
57. Onufer EJ, Tay S, Barron LK, Courtney CM, Warner BW, Guo J. **Intestinal epithelial cell-specific Raptor is essential for high fat diet-induced weight gain in mice**. *Biochem Biophys Res Commun* (2018.0) **505** 1174-1179. DOI: 10.1016/j.bbrc.2018.10.040
58. Eguchi J, Wada J, Hida K. **Identification of adipocyte adhesion molecule (ACAM), a novel CTX gene family, implicated in adipocyte maturation and development of obesity**. *Biochem J* (2005.0) **387** 343-353. DOI: 10.1042/BJ20041709
59. Jangra S, Pothuraju R, Sharma RK, Bhakri G. **Co-administration of soluble fibres and**. *Endocr Metab Immune Disord Drug Targets* (2020.0) **20** 1543-1551. DOI: 10.2174/1871530320666200526123621
60. Xia S-F, Jiang Y-Y, Qiu Y-Y, Huang W, Wang J. **Role of diets and exercise in ameliorating obesity-related hepatic steatosis: insights at the microRNA-dependent thyroid hormone synthesis and action**. *Life Sci* (2020.0) **242** 117182. DOI: 10.1016/j.lfs.2019.117182
61. Karabegović I, Portilla-Fernandez E, Li Y. **Epigenome-wide association meta-analysis of DNA methylation with coffee and tea consumption**. *Nat Commun* (2021.0) **12** 2830. DOI: 10.1038/s41467-021-22752-6
62. Ding Q, Guo R, Pei L. **N-acetylcysteine alleviates high fat diet-induced hepatic steatosis and liver injury via regulating the intestinal microecology in mice**. *Food Funct* (2022.0). DOI: 10.1039/d1fo03952k
63. Lu M, Wan Y, Yang B, Huggins CE, Li D. **Effects of low-fat compared with high-fat diet on cardiometabolic indicators in people with overweight and obesity without overt metabolic disturbance: a systematic review and meta-analysis of randomised controlled trials**. *Br J Nutr* (2018.0) **119** 96-108. DOI: 10.1017/S0007114517002902
64. Logue MW, Smith AK, Wolf EJ. **The correlation of methylation levels measured using Illumina 450K and EPIC BeadChips in blood samples**. *Epigenomics* (2017.0) **9** 1363-1371. DOI: 10.2217/epi-2017-0078
65. Rafi Z, Greenland S. **Semantic and cognitive tools to aid statistical science: replace confidence and significance by compatibility and surprise**. *BMC Med Res Methodol* (2020.0) **20** 244. DOI: 10.1186/s12874-020-01105-9
66. 66.Illumina Inc. Infinium MethylationEPIC Manifest Column Headings; 2020 [Cited 25 Nov 2021 ] Available from: URL: https://emea.support.illumina.com/bulletins/2016/08/infinium-methylationepic-manifest-column-headings.html?langsel=/fo/
|
---
title: 'The polyphenol epigallocatechin gallate lowers circulating catecholamine concentrations
and alters lipid metabolism during graded exercise in man: a randomized cross-over
study'
authors:
- Rachel Churm
- Liam M. Williams
- Gareth Dunseath
- Sarah L. Prior
- Richard M. Bracken
journal: European Journal of Nutrition
year: 2023
pmcid: PMC10030435
doi: 10.1007/s00394-023-03092-1
license: CC BY 4.0
---
# The polyphenol epigallocatechin gallate lowers circulating catecholamine concentrations and alters lipid metabolism during graded exercise in man: a randomized cross-over study
## Abstract
### Purpose
Physical exercise is shown to mitigate catecholamine metabolites; however, it is unknown if exercise-induced increases in sympatho-adrenal activity or catecholamine metabolites are influenced by ingestion of specific catechins found within green tea. This study explored the impact of epigallocatechin gallate (EGCG) ingestion on catecholamine metabolism during graded cycle exercise in humans.
### Methods
Eight males (22.4 ± 3.3 years, BMI:25.7 ± 2.4 kg.m2) performed a randomised, placebo-controlled, single-blind, cross-over trial after consumption (1450 mg) of either EGCG or placebo (PLAC) and performed graded cycling to volitional exhaustion. Venous bloods were taken at rest, 2 h post-ingestion and after every 3-min stage. Blood variables were analysed for catecholamines, catecholamine metanephrines and metabolic variables at rest, 2 h post-ingestion (POST-ING), peak rate of lipid oxidation (FATpeak), lactate threshold (LT) and peak rate of oxygen consumption (VO2peak). Data were analysed using SPSS (Version 26).
### Results
Resting catecholamine and metanephrines were similar between trials. Plasma adrenaline (AD) was lower in ECGC treatment group between trials at FATpeak ($P \leq 0.05$), LT ($P \leq 0.001$) and VO2peak ($P \leq 0.01$). Noradrenaline (NA) was lower under EGCG at POST ($P \leq 0.05$), FATpeak ($P \leq 0.05$), LT ($P \leq 0.01$) and VO2peak ($P \leq 0.05$) compared to PLAC. Metanephrines, glucose and lactate increased similarly with exercise intensity in both trials. Lipid oxidation rate was $32\%$ lower in EGCG at FATpeak (EGCG 0.33 ± 0.14 vs. PLAC 0.49 ± 0.11 g.min−1, $P \leq 0.05$). Cycle time to exhaustion was similar (NS).
### Conclusion
Acute EGCG supplementation reduced circulating catecholamines but not; metanephrine, glucose or lactates, response to graded exercise. Lower circulating catecholamines may explain a lower lipid oxidation rate.
## Introduction
Catechins are polyphenolic flavonoids and are bioactive ingredients in green tea. The four main types of catechins found in green tea are epigallocatechin-3-gallate (EGCG), epicatechin-3-gallate (ECG), epigallocatechin (EGC) and epicatechin (EC) with the most abundant and pharmacologically active being EGCG [1]. Combined these catechins account for ~ $30\%$ of dry weight of green tea leaves [2].
Green tea or its individual constituents have been shown to alter resting energy metabolism. Several short-term (< 3 days) supplementation studies have found an increase in resting energy expenditure [3–6]; however, this has not always been the case [7–9]. In studies that have found increased energy expenditure, this was related to an increase in lipid metabolism [4, 6]. For example, green tea ingestion (EGCG 245 mg.d−1, caffeine 270 mg.d−1) increased resting energy expenditure by $2.9\%$ and lipid oxidation rates by $12\%$ higher over the 24-h period compared with water [6]. Some studies have demonstrated no change in resting energy expenditure but found an increased lipid oxidation rate [7, 9]. However, altered lipid oxidation rates have not been replicated in all studies [3, 5]. Longer term interventional studies (6–12 weeks) have found greater body weight loss and/or prevention of weight gain after consuming green tea or green tea extract [10–12].
Whole body lipid oxidation rates may increase from resting values during submaximal exercise [13]. In this setting increased lipid combustion may allow for tissue glycogen sparing [14] and provide a greater calorific content per gram of fuel compared to carbohydrate. Acute green tea extract has been shown to alter lipid oxidation rates during exercise [15, 16]. The study by [16], examined how decaffeinated green tea extract (EGCG 366 mg) influenced metabolism during moderate intensity exercise. Over a 24-h period subjects consumed 3 capsules totalling 890 mg total polyphenols. Participants then performed a 30-min cycle at $60\%$ VO2 max. The results of this study noted no change in energy expenditure but a $17\%$ increase in lipid oxidation (0.35 ± 0.03 to 0.41 ± 0.03 g.min−1). Longer term supplementation studies have also noted increased lipid oxidation rates during exercise [17] however, most studies report no change in lipid utilisation rates following supplementation periods ranging from 1 to 28 days [18–23]. With increases in metabolic markers of fat oxidation under resting conditions following 7 days of ingestion of decaffeinated green tea extract (dGTE), is shown with an increase in 3-hydroxybutyrate at fasting and resting conditions following supplementation [21]. However, no increases in either glycerol and non-esterfied fatty acids were observed after 7 days of dGTE ingestion, indicating that dGTE did not enhance lipolysis. Furthermore, supplementation did not increase markers of fat oxidation during exercise. Interestingly, several studies have demonstrated efficacy of chronic green tea supplementation in losing weight [10–12]. However, [24] found no greater change in body mass, body composition, energy or substrate metabolism following supplementation when participants who were undergoing an energy restricted diet were also provided EGCG.
Few studies have explored the working mechanisms of green tea catechins in modulating resting or exercise energy metabolism that might translate to alterations in lipid metabolism or body mass loss. The sympatho-adrenal system is important in altering basal and exercise metabolism [25]. Circulating catecholamines can stimulate adipocyte and skeletal muscle hormone sensitive lipase to increase conversion of triglyceride to non-esterified fatty acids [25] and improve lipid combustion [26]. It has been hypothesized that the mechanism behind the short-term effects of green tea extract used in many studies may be directly related to the active green tea ingredient catechins. Catechins have been hypothesised to regulate the sympatho-adrenal system directly through inhibition of catechol-O-methyltransferase (COMT) activity [27] although this has been disputed [28]. COMT is an intracellular enzyme found in all tissues including skeletal muscle and adipose tissue, and degrades adrenaline and noradrenaline. Whilst in vitro results support flavonoid-mediated COMT inhibition [29] there is limited research evidence in vivo to support this potential mechanism. Although catecholamine metabolites like plasma metanephrine and normetanephrine have been shown to change in response to physical exercise [30] it is currently unknown if exercise-induced increases in sympatho-adrenal activity or catecholamine metabolites are influenced by ingestion of specific catechins found within green tea. Graded exercise is a useful modality to explore a well-characterised rise in energy metabolism and an increase in sympatho-adrenal activity [31].
Thus, the aim of this study was to explore the impact of acute ingestion of the polyphenol epigallocatechin gallate (EGCG) on catecholamine, catecholamine metabolite, systemic metabolic and cardio-vascular variables across a range of exercise intensities during graded cycle exercise in man.
## Participants
This study was conducted according to Declaration of Helsinki and all procedures involving human subjects were approved by the University Research Ethics Committee; PG/$\frac{2014}{28.}$ Eight apparently healthy male participants (age: 22.3 ± 3.3 years, estimated body fat 15 ± $5.2\%$, BMI 25.7 ± 2.4 kg m−2) were recruited to this study. Participants were included if they had a habitual caffeine intake ≤ 400 mg.d−1 (less than four cups of tea/coffee or caffeinated soda beverages per day) and performed habitual exercise three to five times per week for 30–90 min per exercise session. Written informed consent was obtained from all participants and all participants completed a health screening questionnaire to ascertain their health status and to determine their eligibility to partake in this study accordance with American Heart Association/American College of Sports Medicine (AHA/ACSM) (clinicaltrials.gov NCT03199430, http://clinicaltrials.gov/show/NCT03199430).
## Experimental design
In a randomised, placebo-controlled, single-blind, cross-over design study participants completed two trials after acute consumption of either EGCG or a placebo (PLAC) supplement after an overnight fast. Principal investigator was responsible for generating the random allocation sequence, enrolling participants, and assigning interventions. Following a two-hour monitoring period participants performed a continuous graded cycle exercise test to volitional exhaustion. There was at least a 7-day washout period between trials. No reported losses and exclusions after randomisation in any group. Due to the non-clinical exploratory nature of the randomised cross-over study it was registered following ethical approval.
## Supplementation
Participants were randomly assigned to either the intervention or placebo trial. After overnight fast participants arrived to the laboratory and were observed ingesting two capsules each of EGCG (minimum $94\%$ EGCG < $0.1\%$ caffeine) from a commercially available brand (TEAVIGO™; TAIYO GmbH, 1450 mg) or a placebo (1450 mg Corn Flour). Capsules were weighed and sorted to within ± $5\%$. The supplement was consumed in two size 00 vegetarian gelatin capsules alongside a standardized amount of distilled water (200 ml).
## Experimental protocol
Prior to participation in the experimental trials participants were familiarized with the laboratory equipment and the test procedures. On the morning of the test participants reported to the Exercise Physiology Laboratory following an 8–10 h fast where measures of body mass (weighting scales; Seca 770 Digital Scales, Seca Ltd, Birmingham, UK), height (stadiometer; Holtain Stadiometer, Holtain Ltd, Cymrych, Wales) and estimated fat percentage using bioelectrical impedance analysis (Bodystat Quadscan 4000, Bodystat Ltd, Isle of Man, UK) were made whilst wearing minimal clothing.
Participants were then seated for a 10-min period while a cannula was inserted into an antecubital vein. This was connected to a three-way stopcock for the repeated collection of venous blood at rest and during the exercise test. Saline (2–3 ml) was infused regularly keep the cannula patent. After a 10-min rest period, a venous blood sample (7 ml) was collected into a lithium-heparinised vacutainer. In addition, baseline measures of heart rate (Polar RS800CX), and a 10-min sample of expired air (Jaeger Vyuntus CPX, Erich Jaeger GmbH, CareFusion Hoechbegh, Germany) were also taken. Expired air was collected throughout the duration of the exercise test and measured for volume, the fractional concentration of oxygen (FEO$2\%$) and carbon dioxide (FECO$2\%$) (SentrySuite Software, Erich Jaeger GmbH, CareFusion Hoechberg, Germany). This allowed for the determination of volumes of O2 utilization and CO2 production. These data were then used to determine oxidative energy expenditure using principles of indirect calorimetry [32].
Following collection of resting parameters participants were remained in a semi-reclined position for two hours post ingestion of green tea extract. Then after a 5-min transition period, participants mounted a cycle ergometer (Lode Excalibur Sport Ergometer, Lode BV Groningen, The Netherlands) to perform a graded exercise test. Participants were instructed to cycle between 60 and 70 rpm at an initial power output of 60 Watts (W) with an increase in 30 W every 3 min. Verbal encouragement was provided to the participant throughout. Heart rate was measured constantly throughout the exercise test (Polar RS800CX) alongside respiratory gas measurements. Two and a half minutes into each 3-min stage, a rating of perceived exertion (RPE [33]) was taken from the participant and a venous blood sample obtained. The test continued until volitional exhaustion defined by the following criteria [1] cadence dropping below 50 rpm, [2] heart rate within 10 beats of age-predicted maximum, [3] levelling of VO2 though workload had increased. At this point cardio-respiratory variables were recorded and a final blood sample was taken at exhaustion. The participants dismounted the ergometer, reclined on a couch and were provided with water ad libitum. The indwelling cannula was removed and participants were monitored for 30 min before leaving the laboratory. No harms or unintended effects were reported in either group.
## Diet control
Participants were given food and physical activity diaries to complete in the 72 h prior to the first experimental trial and were strongly encouraged and reminded via SMS to replicate this diary prior to the second trial. Participants were also instructed to avoid alcohol, foods with high polyphenol content (i.e. fruits, dark chocolate and cereal bran), caffeinated beverages during this period. Participants were also instructed not to perform any physical activity in the 24 h period prior to each trial. The food diary was analysed using CompEat software (CompEat 5.7 Pro). In the 72 h preceding the first arm the macronutrient composition of the participants’ diets was carbohydrate $40.20\%$, fat $37.61\%$ and protein $22.20\%$, this was replicated for the alternate arm of the study and included a $5\%$ margin of error for diet replications across trial arms.
## Blood analyses
Venous blood samples were analysed immediately for lactate and glucose levels (Biosen C-Line, EKF Diagnostics). Thereafter, the remainder of the sample was centrifuged (Heraeus Megafuge 8, Thermo Scientific) for 10 min at 3,000 rpm with ~ 3 ml of plasma extracted into individual 1 ml microcentrifuge tubes and frozen immediately (− 80 °C) for later analysis of metanephrine, normetanephrine and catecholamine (adrenaline and noradrenaline) concentrations using commercially available enzyme linked absorbent assays (ELISA, Eagle Biosciences Inc, Nashua, New Hampshire, USA). The lower limit of detection for the adrenaline and noradrenaline assay was 5 pg ml−1 and 16 pg ml−1, respectively. Average intra-assay coefficients of variation were $8.35\%$ for adrenaline and $9.7\%$ for noradrenaline. Average recoveries of $97\%$ and $94\%$ were obtained for adrenaline and noradrenaline. The lower limit of detection for both the metanephrine and normetanephrine assays was 7 pg ml−1. Average intra-assay coefficients of variation were $7.95\%$ for metanephrine and $6.45\%$ for normetanephrine. Average recoveries of $94\%$ and $95\%$ were obtained for metanephrine and normetanephrine.
## Metabolic domains
On completion of the exercise tests metabolic and physiological data were grouped into domains to allow comparisons between participants relative to workload and physiological responses, namely (i) baseline (REST), (ii) two hours post ingestion at rest (POST-ING), (iii) highest lipid oxidation rate during exercise (FATpeak), (iv) lactate threshold, i.e. the value of LT estimated using simple linear regression by fitting to a model and identifying the workload LT, corresponding to the model with minimum Mean Squared Error (LT; calculated using Lactate-E software [34] and (v) peak rate of oxygen consumption (VO2peak).
## Calculations
From the recorded variables of VO2 and VCO2, fat and carbohydrate oxidation rates (g.min−1) were calculated using the stoichiometric equations at rest [32] and during exercise [35] under the assumption that protein utilisation was negligible.
## Data analysis
The data were analysed using the Statistical Package for the Social Sciences software (Version 26, SPSS, Inc). Data were reported as means ± SD with P ≤ 0.05 accepted. All data were assessed for normality (Shapiro–Wilk’s test) and subsequently analysed using two-way repeated measures ANOVAs (condition × time) with post hoc dependant t tests conducted with Bonferroni corrections where appropriate. Paired samples t test was used to compare performance parameters. This pilot study could not generate a priori power on primary endpoints due to lack of published information. However, in our pilot, post-priori data analysis on the primary catecholamine of interest (peak adrenaline concentration difference between EGCG and PLAC) revealed a statistical power of $99\%$ in eight participants at an alpha error level (two-sided significance) of $5\%$.
## Cardio-respiratory changes and ratings of perceived exertion
Cardio-respiratory changes and rating of perceived exertion under both PLAC and EGCG at rest and during exercise are reported in Table 1. Although there was a clear effect of exercise intensity on ventilation, oxygen consumed, carbon dioxide produced and respiratory exchange ratio, heart rate and RPE there was no effect of supplementation on any of these cardio-respiratory variables (NS). Fatpeak occurred at a similar percentage of VO2peak in both trials (PLAC 37.4 ± 4.8 vs. EGCG 37.4 ± $4.0\%$, NS). LT occurred at a similar percentage of VO2peak (PLAC 79.5 ± 8.3 vs. EGCG 79.0 ± $9.9\%$, NS).Table 1Cardiorespiratory markers and rating of perceived exertion scores at rest and during exercise during the placebo and EGCG trials ($$n = 8$$)VariableConditionRestFatpeakLTVO2peakVE (L.min−1)PLAC11.8 ± 428.4 ± 585.0 ± 24139.4 ± 22EGCG11.1 ± 432.1 ± 488.3 ± 22141.6 ± 28VO2 (L.min−1)PLAC0.3 ± 0.11.3 ± 0.22.9 ± 0.43.6 ± 0.6EGCG0.3 ± 0.11.3 ± 0.22.8 ± 0.33.4 ± 0.8VCO2 (L.min−1)PLAC0.3 ± 0.11.0 ± 0.22.7 ± 0.84.0 ± 0.6EGCG0.3 ± 0.11.2 ± 0.23.0 ± 0.44.0 ± 0.8RERPLAC0.86 ± 0.090.78 ± 0.11.03 ± 0.11.12 ± 0.1EGCG0.88 ± 0.070.85 ± 0.11.05 ± 0.1 1.13 ± 0.04HR (bpm)PLAC66 ± 8100 ± 9172 ± 9191 ± 9EGCG67 ± 9105 ± 6172 ± 9191 ± 8RPE (6–20)PLAC7 ± 216 ± 119 ± 0.7EGCG7 ± 115 ± 218 ± 0.83Only between group differences indicated. Statistical significance over time not reported for ease of interpretationData reported as mean ± SD
## Energy expenditure, carbohydrate and lipid oxidation
The resting and exercise-induced changes in energy expenditure, carbohydrate and lipid oxidation are located in Fig. 1. There was a 12–14-fold increase in oxidative energy expenditure during the graded cycle exercise test that was similar between conditions (NS). Interestingly, though an isoenergetic domain, rates of lipid oxidation at FATpeak were reduced by $32\%$ in the EGCG trial (EGCG 0.33 ± 0.14 vs. PLAC 0.49 ± 0.11 g.min−1, $P \leq 0.05$) with CHO utilisation correspondingly greater at this point (EGCG 0.87 ± 0.39 vs. PLAC 0.44 ± 0.34 g.min−1, $P \leq 0.05$).Fig. 1A Total energy expenditure and contribution from lipid (FAT) and carbohydrate (CHO) oxidation and B percentage contribution of lipid (FAT) and carbohydrate (CHO) oxidation to total energy expenditure; at rest, at highest lipid oxidation rate (FATpeak), point of lactate threshold (LT) and peak rate of oxygen consumption (VO2peak). * Indicates significant difference in energy derived from lipids and carbohydrate oxidation between conditions (P ≤ 0.05). †Indicates significant difference in energy derived from carbohydrate oxidation across time compared to rest (P ≤ 0.05), ‡indicates significant difference in energy derived from lipid oxidation across time compared to rest (P ≤ 0.05), unless annotated by non-significant time point (NS; B, Rest vs. FATpeak, EGCG treatment). Statistical analysis conducted using a 2-way repeated measures ANOVA and paired sample T Test
## Blood lactate and glucose
Blood lactate concentrations (Fig. 2) were influenced by time ($P \leq 0.05$) but not by trial (NS). At FATpeak blood lactate rose by ~ $50\%$ from rest within each condition and graded exercise produced a 12-fold increase in blood lactate under both PLAC and EGCG at VO2peak. Likewise, blood glucose concentrations were also influenced by time ($P \leq 0.05$), but did not differ between conditions (NS). Interestingly, under EGCG blood glucose post-ingestion was lower compared to resting values (EGCG Rest 4.39 ± 0.30 vs. Post-Ingestion 4.15 ± 0.30 mmol.l−1, P ≤ 0.05). There was a small rise in blood glucose concentration from rest to VO2peak values in the PLAC trial (REST 4.34 ± 0.4 vs. VO2peak 4.62 ± 0.3 mmol.l−1 $P \leq 0.05$) and also in the EGCG trial (REST 4.39 ± 0.3 vs. VO2peak 4.68 ± 0.38 mmol.l−1 $P \leq 0.05$).Fig. 2A Blood lactate & B blood glucose responses during the Placebo (diagonal line fill) and EGCG (grey) trials at rest, at post ingestion (POST) at highest lipid oxidation rate (FATpeak), point of lactate threshold (LT) and peak rate of oxygen consumption (VO2peak). †Indicates significant difference across time compared to rest in EGCG treatment only (P ≤ 0.05). ‡Indicates significant difference across time compared to rest in placebo treatment only (P ≤ 0.05). Data was analysed by repeated measures ANOVA with subsequent post hoc analysis, reported as mean ± SD
## Graded cycle test performance
There was no change in performance time noted between trials (EGCG 1370 ± 152 vs. PLAC 1377 ± 150 s, NS). All participants attained similar power outputs under both PLAC and EGCG conditions (EGCG 270 ± 32 vs. PLAC 266 ± 25 W, NS). Likewise, similar VO2peak values were noted (PLAC 44.8 ± 4.7 vs. EGCG 43.0 ± 5.0 ml.kg.min−1, NS).
## Plasma catecholamine and catecholamine metabolites
Plasma AD and NA concentrations are reported in Fig. 3. Resting plasma AD concentrations were similar between conditions and were not altered following ingestion of either supplement. Post-ingestion NA concentrations were lower compared to resting values under both PLAC and EGCG (EGCG; Rest.vs. Post; 0.95 ± 0.44 vs. 0.48 ± 0.55 nmol.l−1, $P \leq 0.05.$ PLAC; Rest.vs. Post; 1.15 ± 0.44 vs. 1.86 ± 0.94 nmol.l−1, $P \leq 0.05$). Compared to PLAC, EGCG plasma AD concentrations were statistically lower at FATpeak (EGCG 0.18 ± 0.11 vs. PLAC 0.37 ± 0.27 nmol.l−1, $P \leq 0.05$), LT (EGCG 0.35 ± 0.16 vs. PLAC 1.59 ± 0.49 nmol.l−1, $P \leq 0.001$) and VO2peak (EGCG 0.91 ± 0.58 vs. PLAC 4.39 ± 2.42 nmol.l−1, $P \leq 0.001$). NA concentrations under EGCG were significantly lower at FATpeak (EGCG 0.93 ± 0.72 vs. PLAC 2.43 ± 1.04 nmol.l−1, $P \leq 0.05$), LT (EGCG 3.41 ± 2.27 vs. PLAC 7.12 ± 7.00 nmol.l−1, $P \leq 0.01$), and VO2peak (EGCG 12.52 ± 6.53 vs. PLAC 21.90 ± 2.66 nmol.l−1, $P \leq 0.05$).Fig. 3A Plasma adrenaline & B plasma noradrenaline responses during the placebo (diagonal line fill) and EGCG (grey) trials at rest, at post ingestion (POST-ING) at highest lipid oxidation rate (FATpeak), point of lactate threshold (LT) and peak rate of oxygen consumption (VO2peak). * Indicates significant difference between conditions at time point (P ≤ 0.05). †Indicates significant difference across time compared to rest in EGCG treatment only (P ≤ 0.05), unless annotated by non-significant time point (NS; B, Rest vs. Post). ‡Indicates significant difference across time compared to rest in placebo treatment only (P ≤ 0.05). Data was analysed by repeated measures ANOVA with subsequent post hoc analysis and reported as mean ± SD Resting MET and NORMET concentrations were similar and did not change after 2 h of ingestion of either supplement (NS). Resting MET concentrations increased similarly to 321 ± 235 pmol.l−1 (NS) at VO2peak (Fig. 4) and in the EGCG trial to 358 ± 215 pmol.l−1 ($P \leq 0.05$). NORMET concentrations rose progressively with exercise in both trials reaching similar peak values (EGCG 1203 ± 500 vs. PLAC 1390 ± 598 pmol.l−1, NS).Fig. 4A Metanephrine and B normetanephrine responses during the Placebo (black dashed) and EGCG (grey solid) trials at rest, at post-ingestion (POST) at highest lipid oxidation rate (FATpeak), point of lactate threshold (LT) and peak rate of oxygen consumption (VO2peak). †Indicates significant difference across time compared to rest in EGCG treatment only (P ≤ 0.05). ‡Indicates significant difference across time compared to rest in placebo treatment only (P ≤ 0.05). Data were analysed by repeated measures ANOVA with subsequent post hoc analysis and reported as mean values
## Discussion
This study explored the impact of acute ingestion of the polyphenol epigallocatechin gallate (EGCG) on catecholamine, catecholamine metanephrine, systemic metabolic and cardio-respiratory variables during continuous incremental cycle exercise. To the authors’ knowledge, this study is the first to demonstrate that EGCG supplementation resulted in lowered circulating catecholamine concentrations. In addition, EGCG altered oxidative energy provision from lipid and carbohydrate at low exercise intensities.
Increased physical exercise intensity elevated heart rate, ventilation, O2 consumption, CO2 production and ratings of perceived exertion. Research studies that administered green tea extract to human participants found no change [18, 36] or lower heart rates [23] in response to exercise when compared to a placebo. Our results agree with the findings [18, 36], where exercise-induced increases in heart rate were similar between control and EGCG conditions. In contrast, the differing findings [23] might be explained by the polyphenol mixture administered by the researchers compared to only EGCG ($94\%$ pure) in our study. It might be suggested that the lower circulating catecholamine concentrations under EGCG would reduce HR, however, that other hormonal, humoral or electrical stimulants of sino-atrial node pacing influence heart rate suggest a greater contribution of these other factors to heart rate regulation during exercise.
Using principles of stoichiometry to determine the lipid and carbohydrate proportions of energy utilisation during exercise at relatively similar metabolic domains, this study revealed a $32\%$ decreased oxidation rate of lipids following EGCG supplementation at FATpeak compared to placebo. Increased VE and raised VCO2 rates under EGCG help explain this finding and led to a greater respiratory exchange ratio (RER) value (PLAC 0.78 ± 0.05 vs. EGCG 0.85 ± 0.07) that is indicative of a lower rate of lipid oxidation (Fig. 1) and a compensatory increase in carbohydrate oxidation rate under EGCG at FATpeak. This is a novel finding and is contrary to previous studies where EGCG-containing supplements increased lipid oxidation during exercise [16, 17] or did not influence substrate utilisation [18, 20, 22, 23]. It is difficult to fully reconcile our data with those of other researchers but suggest some of the differences may be due to the different exercise protocols, dosages, timings of supplement administration, percentage EGCG in GTE and/or caffeine-containing GTE.
The lipids (non-esterified fatty acids) oxidised during exercise are derived from intramuscular depots or adipose tissue. However, during exercise above the lactate threshold, circulating FFA amounts cannot meet the tissue oxidative needs and intramuscular triglyceride stores are used more [37]. In both adipose and skeletal muscle tissues, the enzyme hormone sensitive lipase (HSL) stimulates triglyceride breakdown and consequent liberation of free fatty acids for oxidation. During exercise, intramuscular HSL activity is stimulated by muscle contraction [38] and adrenaline [25, 39] with the effects of both factors being additive [40]. Given the exercise characteristics were well controlled in our study (cadence 60 rpm, 30 W progressive power increase every 3 min), the decreased rate of lipid oxidation at FATpeak which occurred at a similar relative exercise intensity (PLAC 37.4 ± 4.8 vs. EGCG 37.4 ± $4.0\%$VO2peak) might be due to a lesser stimulation of hormone sensitive lipase on intramuscular triglyceride utilisation via a decreased sympatho-adrenal response.
Many studies researching green tea extract or EGCG have explored oxidative energy provision with scant attention paid to non-oxidative pathways. We measured resting and exercising blood lactate concentration as an indicator of non-oxidative energy metabolism. As graded exercise was mostly performed in an oxidative isoenergetic domain, the noted difference in RER between PLAC and EGCG might indicate an increase in CHO utilisation. Increased CHO flux through non-oxidative glycolysis preceding oxidation might result in elevated pyruvate and lactate formation, with an increase in both these markers [28], following green tea extract supplementation. That there were no between-group differences in blood lactate is interesting especially despite the dramatically lowered circulating catecholamine values and suggests a lesser role of catecholamines in stimulating production of muscle glycogenolysis compared to insulin suppression and increased circulating glucagon [41]. However, given the absolute lactate concentrations in the bloodstream at the point of highest lipid oxidation (FATpeak) were low (~ 1 mM) our data must be interpreted with some caution. The lack of change in blood lactate contradicts previous work by Hodgson et al. [ 28], that in response to a 60-min cycle at $56\%$ VO2max there was an increase in blood lactate concentrations following GTE supplementation; however, they noted no change in catecholamine concentrations, contrary to this study. Blood glucose levels after ingestion of EGCG were lower compared to rest (~ 0.2 mmol.l−1). This decrease may be due to an insulin sensitising effect of EGCG, with literature demonstrating a $13\%$ increased insulin sensitivity in the fed state following acute EGCG supplementation [16]. Alternatively, plasma catecholamines stimulate hepatic β-adrenoreceptors and increase glycogenolysis with increased glucose release to the bloodstream. That resting plasma NA was lower under EGCG might help explain the lower blood glucose concentration at this time point.
There was no change in performance time or power output after EGCG supplementation. This is a similar finding to previous research [18] and provides little evidence of ergogenic potential of EGCG in exercise performance. Given the complex regulatory, often compensatory processes involved in the provision of oxygen and fuel during physical exercise a small alteration in lipid use alongside reductions in sympatho-adrenal activity did not materialise in any alteration in the perception of exertion or functional capacity.
Green tea leaves contain components like caffeine that can alter circulating catecholamine concentrations in some [4, 42] but not all [3, 5, 8, 28] studies so is a confounding variable. Thus, in our study we explored the influence of high purity ($94\%$), low caffeine (< $0.1\%$) EGCG extract (the highest concentration polyphenol in green tea) on resting and exercising circulating catecholamines and catecholamine metabolites (that directly involve the activity of catechol-O-methyl transferase). Consumption of EGCG via capsule form generates an average peak plasma concentration (Cmax) 2.5–2.8 h post-consumption [43], this corresponds to circulating catecholamine AD concentrations being greatly lowered following acute EGCG supplementation at optimum bioavailability when compared to PLAC, with plasma AD concentrations at (i) FATpeak reduced by $53\%$, (ii) LT being reduced by $78\%$, and (iii) VO2peak being reduced by $79\%$. Although it has long been recognised that flavanoids inhibit COMT activity [29], only more recently has EGCG been shown to inhibit human hepatic cytosolic COMT activity in vitro [27, 44]. Inhibition of COMT could allow for sustained (if not elevated) plasma catecholamine concentrations and potentially reduce production of downstream catecholamine metabolites (metanephrine and normetanephrine). The results of our study throw into question this mechanism. As stated above, plasma catecholamines were greatly reduced after EGCG supplementation and we found no differences in the resting or exercise-induced increases in plasma metanephrine or normetanephrine compared to the placebo trial suggestive of similar COMT activity in each trial. Thus, the mechanism(s) by which plasma circulating catecholamines are lower during exercise following EGCG supplementation are unclear. EGCG (as opposed to green tea extract) has been shown to have no influence α-adrenergic stimulation of rat vascular and aortic tissue in vitro [45]. Yet catecholamine secretion from perfused rat adrenal medulla is reduced due to the influence of polyphenols (that also contained epicatechin) via interruption of sodium and calcium interchange in chromaffin tissue and/or inducing nitric oxide release in vascular tissue [46]. Adding to the complexity of responses, administration of caffeine to rats with or without EGCG reduced the increase in catecholamines and vascular responses associated with caffeine administration alone [47]. As an alternative suggestion to explain our data, EGCG has been shown to inhibit DOPA decarboxylase, slowing conversion of l-DOPA into dopamine; the upstream product of NA and AD within adrenal medulla chromaffin cells and sympathetic neuron [48]. However, further investigation is warranted to detail the exact in vivo mechanistic pathways involved during exercise in man.
In conclusion, this study explored the impact of acute ingestion of the polyphenol epigallocatechin gallate (EGCG) on catecholamine, catecholamine metanephrines, systemic metabolic and cardio-vascular variables across a range of exercise intensities during graded cycle exercise in man. Circulating catecholamine concentrations and peak lipid oxidation were lower in response to a graded exercise test following acute EGCG supplementation.
Study limitations: Blood samples collected at the end of each stage were able to be analysed for catecholamines and catecholamine metabolites; however, we established five metabolic time points which were examined to remove potential bias and mitigated this limitation, all samples were however analysed for plasma blood lactate and plasma blood glucose. Second, the lack of a post ingestion respiration sample, in any future study it would be beneficial to add this sample point to allow a more complete picture of any subtle changes which may have occurred during the rest period.
## References
1. Kao Y, Hiipakka RA, Liao S. **Modulation of endocrine systems and food intake by green tea epigallocatechin gallate**. *Endocrinology* (2000) **141** 980-987. DOI: 10.1210/endo.141.3.7368
2. Graham HN. **Green tea composition, consumption, and polyphenol chemistry**. *Prev Med* (1992) **21** 334-350. DOI: 10.1016/0091-7435(92)90041-F
3. Bérubé-Parent S, Pelletier C, Doré J, Tremblay A. **Effects of encapsulated green tea and Guarana extracts containing a mixture of epigallocatechin-3-gallate and caffeine on 24 h energy expenditure and fat oxidation in men**. *Br J Nutr* (2005) **94** 432-436. DOI: 10.1079/BJN20051502
4. Dulloo AG, Duret C, Rohrer D, Girardier L, Mensi N, Fathi M, Chantre P, Vandermander J. **Efficacy of a green tea extract rich in catechin polyphenols and caffeine in increasing 24-h energy expenditure and fat oxidation in humans**. *Am J Clin Nutr* (1999) **70** 1040-1045. DOI: 10.1093/ajcn/70.6.1040
5. Rudelle S, Ferruzzi MG, Cristiani I, Moulin J, Macé K, Acheson KJ, Tappy L. **Effect of a thermogenic beverage on 24-hour energy metabolism in humans**. *Obesity (Silver Spring)* (2007) **15** 349-355. DOI: 10.1038/oby.2007.529
6. Rumpler W, Seale J, Clevidence B, Judd J, Wiley E, Yamamoto S, Komatsu T, Sawaki T, Ishikura Y, Hosoda K. **Oolong tea increases metabolic rate and fat oxidation in men**. *J Nutr* (2001) **131** 2848-2852. DOI: 10.1093/jn/131.11.2848
7. Boschmann M, Thielecke F. **The effects of epigallocatechin-3-gallate on thermogenesis and fat oxidation in obese men: a pilot study**. *J Am Coll Nutr* (2007) **26** 389S-395S. DOI: 10.1080/07315724.2007.10719627
8. Gregersen NT, Bitz C, Krog-Mikkelsen I, Hels O, Kovacs EM, Rycroft JA, Frandsen E, Mela DJ, Astrup A. **Effect of moderate intakes of different tea catechins and caffeine on acute measures of energy metabolism under sedentary conditions**. *Br J Nutr* (2009) **102** 1187-1194. DOI: 10.1017/S0007114509371779
9. Thielecke F, Rahn G, Böhnke J, Adams F, Birkenfeld AL, Jordan J, Boschmann M. **Epigallocatechin-3-gallate and postprandial fat oxidation in overweight/obese male volunteers: a pilot study**. *Eur J Clin Nutr* (2010) **64** 704-713. DOI: 10.1038/ejcn.2010.47
10. Basu A, Sanchez K, Leyva MJ, Wu M, Betts NM, Aston CE, Lyons TJ. **Green tea supplementation affects body weight, lipids, and lipid peroxidation in obese subjects with metabolic syndrome**. *J Am Coll Nutr* (2010) **29** 31-40. DOI: 10.1080/07315724.2010.10719814
11. Suliburska J, Bogdanski P, Szulinska M, Stepien M, Pupek-Musialik D, Jablecka A. **Effects of green tea supplementation on elements, total antioxidants, lipids, and glucose values in the serum of obese patients**. *Biol Trace Elem Res* (2012) **149** 315-322. DOI: 10.1007/s12011-012-9448-z
12. Brown A, Lane J, Holyoak C, Nicol B, Mayes AE, Dadd T. **Health effects of green tea catechins in overweight and obese men: a randomised controlled cross-over trial**. *Br J Nutr* (2011) **106** 1880-1889. DOI: 10.1017/S0007114511002376
13. Achten J, Jeukendrup AE. **Optimizing fat oxidation through exercise and diet**. *Nutrition* (2004) **20** 716-727. DOI: 10.1016/j.nut.2004.04.005
14. Vukovich MD, Costill DL, Hickey MS, Trappe SW, Cole KJ, Fink WJ. **Effect of fat emulsion infusion and fat feeding on muscle glycogen utilization during cycle exercise**. *J Appl Physiol* (1993) **75** 1513-1518. DOI: 10.1152/jappl.1993.75.4.1513
15. Sugita M, Kapoor MP, Nishimura A, Okubo T. **Influence of green tea catechins on oxidative stress metabolites at rest and during exercise in healthy humans**. *Nutrition* (2016) **32** 321-331. DOI: 10.1016/j.nut.2015.09.005
16. Venables MC, Hulston CJ, Cox HR, Jeukendrup AE. **Green tea extract ingestion, fat oxidation, and glucose tolerance in healthy humans**. *Am J Clin Nutr* (2008) **87** 778-784. DOI: 10.1093/ajcn/87.3.778
17. Roberts JD, Roberts MG, Tarpey MD, Weekes JC, Thomas CH. **The effect of a decaffeinated green tea extract formula on fat oxidation, body composition and exercise performance**. *J Int Soc Sports Nutr* (2015) **12** 1. DOI: 10.1186/s12970-014-0062-7
18. Dean S, Braakhuis A, Paton C. **The effects of EGCG on fat oxidation and endurance performance in male cyclists**. *Int J Sport Nutr Exerc Metab* (2009) **19** 624-644. DOI: 10.1123/ijsnem.19.6.624
19. Eichenberger P, Mettler S, Arnold M, Colombani PC. **No effects of three-week consumption of a green tea extract on time trial performance in endurance-trained men**. *Int J Vitam Nutr Res* (2010) **80** 54. DOI: 10.1024/0300-9831/a000006
20. Iscoe K, Riddell M. **Continuous moderate-intensity exercise with or without intermittent high-intensity work: effects on acute and late glycaemia in athletes with type 1 diabetes mellitus**. *Diabet Med* (2011) **28** 824-832. DOI: 10.1111/j.1464-5491.2011.03274.x
21. Jacobs DM, Hodgson AB, Randell RK, Mahabir-Jagessar-T K, Garczarek U, Jeukendrup AE, Mela DJ, Lotito S. **Metabolic response to decaffeinated green tea extract during rest and moderate-intensity exercise**. *J Agric Food Chem* (2014) **62** 9936-9943. DOI: 10.1021/jf502764r
22. Randell RK, Hodgson AB, Lotito SB, Jacobs DM, Rowson M, Mela DJ, Jeukendrup AE. **Variable duration of decaffeinated green tea extract ingestion on exercise metabolism**. *Med Sci Sports Exerc* (2014) **46** 1185-1193. DOI: 10.1249/MSS.0000000000000205
23. Martin BJ, Tan RB, Gillen JB, Percival ME, Gibala MJ. **No effect of short-term green tea extract supplementation on metabolism at rest or during exercise in the fed state**. *Int J Sport Nutr Exerc Metab* (2014) **24** 656-664. DOI: 10.1123/ijsnem.2013-0202
24. Mielgo-Ayuso J, Barrenechea L, Alcorta P, Larrarte E, Margareto J, Labayen I. **Effects of dietary supplementation with epigallocatechin-3-gallate on weight loss, energy homeostasis, cardiometabolic risk factors and liver function in obese women: randomised, double-blind, placebo-controlled clinical trial**. *Br J Nutr* (2014) **111** 1263-1271. DOI: 10.1017/S0007114513003784
25. Watt MJ, Stellingwerff T, Heigenhauser GJ, Spriet LL. **Effects of plasma adrenaline on hormone-sensitive lipase at rest and during moderate exercise in human skeletal muscle**. *J Physiol* (2003) **550** 325-332. DOI: 10.1113/jphysiol.2003.043133
26. Watt MJ, Heigenhauser GJ, Dyck DJ, Spriet LL. **Intramuscular triacylglycerol, glycogen and acetyl group metabolism during 4 h of moderate exercise in man**. *J Physiol* (2002) **541** 969-978. DOI: 10.1113/jphysiol.2002.018820
27. Chen D, Wang CY, Lambert JD, Ai N, Welsh WJ, Yang CS. **Inhibition of human liver catechol-**. *Biochem Pharmacol* (2005) **69** 1523-1531. DOI: 10.1016/j.bcp.2005.01.024
28. Hodgson AB, Randell RK, Boon N, Garczarek U, Mela DJ, Jeukendrup AE, Jacobs DM. **Metabolic response to green tea extract during rest and moderate-intensity exercise**. *J Nutr Biochem* (2013) **24** 325-334. DOI: 10.1016/j.jnutbio.2012.06.017
29. Borchardt RT, Huber JA. **Catechol**. *J Med Chem* (1975) **18** 120-122. DOI: 10.1021/jm00235a030
30. Bracken RM, Brooks S. **Plasma catecholamine and nephrine responses following 7 weeks of sprint cycle training**. *Amino Acids* (2010) **38** 1351-1359. DOI: 10.1007/s00726-009-0343-7
31. Wackerhage H, Gehlert S, Schulz H, Weber S, Ring-Dimitriou S, Heine O. **Lactate thresholds and the simulation of human energy metabolism: contributions by the cologne sports medicine group in the 1970s and 1980s**. *Front Physiol* (2022) **13** 899670. DOI: 10.3389/fphys.2022.899670
32. Frayn K. **Calculation of substrate oxidation rates in vivo from gaseous exchange**. *J Appl Physiol* (1983) **55** 628-634. DOI: 10.1152/jappl.1983.55.2.628
33. Borg GA. **Psychophysical bases of perceived exertion**. *Med Sci Sports Exerc* (1982) **14** 377-381. DOI: 10.1249/00005768-198205000-00012
34. Newell J, Higgins D, Madden N, Cruickshank J, Einbeck J, McMillan K, McDonald R. **Software for calculating blood lactate endurance markers**. *J Sports Sci* (2007) **25** 1403-1409. DOI: 10.1080/02640410601128922
35. Jeukendrup AE, Wallis GA. **Measurement of substrate oxidation during exercise by means of gas exchange measurements**. *Int J Sports Med* (2005) **26** S28-37. DOI: 10.1055/s-2004-830512
36. Richards JC, Lonac MC, Johnson TK, Schweder MM, Bell C. **Epigallocatechin-3-gallate increases maximal oxygen uptake in adult humans**. *Med Sci Sports Exerc* (2010) **42** 739-744. DOI: 10.1249/MSS.0b013e3181bcab6c
37. Emhoff C-AW, Messonnier LA, Horning MA, Fattor JA, Carlson TJ, Brooks GA. **Gluconeogenesis and hepatic glycogenolysis during exercise at the lactate threshold**. *J Appl Physiol* (2013) **114** 297-306. DOI: 10.1152/japplphysiol.01202.2012
38. Langfort J, Ploug T, Ihlemann J, Holm C, Galbo H. **Stimulation of hormone-sensitive lipase activity by contractions in rat skeletal muscle**. *Biochem J* (2000) **351** 207-214. DOI: 10.1042/bj3510207
39. Langfort J, Ploug T, Ihlemann J, Saldo M, Holm C, Galbo H. **Expression of hormone-sensitive lipase and its regulation by adrenaline in skeletal muscle**. *Biochem J* (1999) **340** 459-465. DOI: 10.1042/bj3400459
40. Kjaer M, Howlett K, Langfort J, Zimmerman-Belsing T, Lorentsen J, Bülow J, Ihlemann J, Feldt-Rasmussen U, Galbo H. **Adrenaline and glycogenolysis in skeletal muscle during exercise: a study in adrenalectomised humans**. *J Physiol* (2000) **528** 371-378. DOI: 10.1111/j.1469-7793.2000.00371.x
41. Coker RH, Kjaer M. **Glucoregulation during exercise**. *Sports Med* (2005) **35** 575-583. DOI: 10.2165/00007256-200535070-00003
42. Gahreman D, Wang R, Boutcher Y, Boutcher S. **Green tea, intermittent sprinting exercise, and fat oxidation**. *Nutrients* (2015) **7** 5646-5663. DOI: 10.3390/nu7075245
43. Henning SM, Niu Y, Lee NH, Thames GD, Minutti RR, Wang H, Go VLW, Heber D. **Bioavailability and antioxidant activity of tea flavanols after consumption of green tea, black tea, or a green tea extract supplement**. *Am J Clin Nutr* (2004) **80** 1558-1564. DOI: 10.1093/ajcn/80.6.1558
44. Lu H, Meng X, Yang CS. **Enzymology of methylation of tea catechins and inhibition of catechol-**. *Drug Metab Dispos* (2003) **31** 572-579. DOI: 10.1124/dmd.31.5.572
45. Lim D-Y, Lee E-S, Park H-G, Kim B-C, Hong S-P, Lee E-B. **Comparison of green tea extract and epigallocatechin gallate on blood pressure and contractile responses of vascular smooth muscle of rats**. *Arch Pharmacal Res* (2003) **26** 214-223. DOI: 10.1007/BF02976833
46. Yu BS, Na DM, Kang MY, Lim DY. **Polyphenols of**. *Korean J Physiol Pharmacol* (2009) **13** 517-526. DOI: 10.4196/kjpp.2009.13.6.517
47. Han J-Y, Kim C-S, Lim K-H, Kim J-H, Kim S, Yun Y-P, Hong JT, Oh K-W. **Increases in blood pressure and heart rate induced by caffeine are inhibited by (−)-epigallocatechin-3-**. *J Cardiovasc Pharmacol* (2011) **58** 446-449. PMID: 21975870
48. Bertoldi M, Gonsalvi M, Voltattorni CB. **Green tea polyphenols: novel irreversible inhibitors of dopa decarboxylase**. *Biochem Biophys Res Commun* (2001) **284** 90-93. DOI: 10.1006/bbrc.2001.4945
|
---
title: 'The impact of alternative delivery strategies for novel tuberculosis vaccines
in low-income and middle-income countries: a modelling study'
authors:
- Rebecca A Clark
- Christinah Mukandavire
- Allison Portnoy
- Chathika K Weerasuriya
- Arminder Deol
- Danny Scarponi
- Andrew Iskauskas
- Roel Bakker
- Matthew Quaife
- Shelly Malhotra
- Nebiat Gebreselassie
- Matteo Zignol
- Raymond C W Hutubessy
- Birgitte Giersing
- Mark Jit
- Rebecca C Harris
- Nicolas A Menzies
- Richard G White
journal: The Lancet. Global Health
year: 2023
pmcid: PMC10030455
doi: 10.1016/S2214-109X(23)00045-1
license: CC BY 3.0
---
# The impact of alternative delivery strategies for novel tuberculosis vaccines in low-income and middle-income countries: a modelling study
## Summary
### Background
Tuberculosis is a leading infectious cause of death worldwide. Novel vaccines will be required to reach global targets and reverse setbacks resulting from the COVID-19 pandemic. We estimated the impact of novel tuberculosis vaccines in low-income and middle-income countries (LMICs) in several delivery scenarios.
### Methods
We calibrated a tuberculosis model to 105 LMICs (accounting for $93\%$ of global incidence). Vaccine scenarios were implemented as the base-case (routine vaccination of those aged 9 years and one-off vaccination for those aged 10 years and older, with country-specific introduction between 2028 and 2047, and 5-year scale-up to target coverage); accelerated scale-up similar to the base-case, but with all countries introducing vaccines in 2025, with instant scale-up; and routine-only (similar to the base-case, but including routine vaccination only). Vaccines were assumed to protect against disease for 10 years, with $50\%$ efficacy.
### Findings
The base-case scenario would prevent 44·0 million ($95\%$ uncertainty range 37·2–51·6) tuberculosis cases and 5·0 million (4·6–5·4) tuberculosis deaths before 2050, compared with equivalent estimates of cases and deaths that would be predicted to occur before 2050 with no new vaccine introduction (the baseline scenario). The accelerated scale-up scenario would prevent 65·5 million (55·6–76·0) cases and 7·9 million (7·3–8·5) deaths before 2050, relative to baseline. The routine-only scenario would prevent 8·8 million ($95\%$ uncertainty range 7·6–10·1) cases and 1·1 million (0·9–1·2) deaths before 2050, relative to baseline.
### Interpretation
Our results suggest novel tuberculosis vaccines could have substantial impact, which will vary depending on delivery strategy. Including a one-off vaccination campaign will be crucial for rapid impact. Accelerated introduction—at a pace similar to that seen for COVID-19 vaccines—would increase the number of lives saved before 2050 by around $60\%$. Investment is required to support vaccine development, manufacturing, prompt introduction, and scale-up.
### Funding
WHO ($\frac{2020}{985800}$-0).
### Translations
For the French, Spanish, Italian and Dutch translations of the abstract see Supplementary Materials section.
## Introduction
Tuberculosis is one of the leading causes of infectious disease death worldwide, second only to COVID-19.1 The negative impact of COVID-19 on tuberculosis-related health services, such as delays in diagnosis, treatment, and neonatal vaccination has paused and reversed slowly declining trends in mortality.1, 2 WHO established the End TB Strategy in 2015, with the goal of reducing disease incidence, deaths, and costs worldwide from tuberculosis.3 Targets for 2025 include reductions in the absolute number of deaths from tuberculosis by $75\%$ and in incidence by $50\%$, and targets for 2035 include reductions in the absolute number of deaths by $95\%$ and in incidence by $90\%$, both compared with 2015 rates.3 However, most countries are not on track to achieve these targets.1, 4 The 2035 End TB targets explicitly assumed the introduction of new tools, including a novel tuberculosis vaccine, by 2025.3 WHO has proposed preferred product characteristics for new tuberculosis vaccines,5 which were developed through a highly consultative process, including regulators and policy makers from high-burden countries. Although progress has been made, the 2025 target for novel tuberculosis vaccine introduction is unlikely to be achieved.
A phase 2b trial of the M72/AS01E candidate vaccine showed an efficacy of 49·$7\%$ ($95\%$ CI 2·1–74·2) for preventing disease in adults positive by interferon-gamma release assay after 3 years of follow-up,6 and a trial of BCG-revaccination appeared efficacious at preventing sustained infection in a cohort of adolescents negative for interferon-gamma release assay, with an efficacy of 45·$4\%$ (6·4–68·1).7 Unfortunately, the phase 3 trial of M72/AS01E has not started, and therefore the realistic licensure date, should a positive result be found, might not be for many years. Policy changes on BCG-revaccination in adolescents could happen sooner in settings such as South Africa, but BCG-revaccination has not been tested in individuals positive for tuberculosis infection—a population shown previously to be epidemiologically important for rapid population-level impact.8 Research in context Evidence before this study Two systematic reviews in the previous 7 years have highlighted the benefits that novel tuberculosis vaccines could have on reducing the tuberculosis burden globally, and that vaccines are likely to be crucial to achieve elimination. These studies indicate that the impact of novel tuberculosis vaccines will depend on the characteristics of the setting, the vaccine, and the delivery strategy. We searched PubMed on Nov 2, 2022, with no date or language restrictions, to find all studies modelling the impact of vaccines aligned with the WHO preferred product characteristics for new tuberculosis vaccines, using the search terms ((tuberculosis) OR (Mtb)) AND ((vaccine) OR (immunisation)) AND ((WHO) OR (World Health Organization)) AND (preferred product characteristics). We found no studies estimating the potential health impacts of introducing a vaccine with characteristics aligned with the WHO preferred product characteristics in low-income and middle-income countries, and existing literature remains limited in terms of how realistic the modelled vaccine introduction and scale-up scenarios were. Added value of this study We estimated the potential impact on tuberculosis cases and deaths of vaccines for infants and for adolescents and adults meeting WHO preferred product characteristics in 105 low-income and middle-income countries that accounted for $93\%$ of the global tuberculosis incidence and mortality in 2019. We evaluated more complex and realistic base-case vaccine delivery scenarios than previously modelled by including country-specific introduction years between 2028 and 2047, and scaling up to target vaccine coverage across 5 years upon initial country introduction. The vaccine for infants was assumed to be delivered routinely to neonates, and the vaccine for adolescents and adults was assumed to be introduced routinely to those aged 9 years and as a one-off campaign for those aged 10 years and older. We compared the base-case scenarios to accelerated introduction and scale-up in all countries in 2025, at a speed similar to the pace of COVID-19 vaccine introduction, to estimate the implications of not meeting the End TB strategy target to develop and license a new tuberculosis vaccine by 2025, and scale up quickly. We also compared the base-case scenario for the adolescent and adult vaccine with a less ambitious routine-only introduction (no one-off vaccination for those aged 10 years and older). We grouped countries by WHO region, income group, and tuberculosis burden to identify where the largest impacts of a novel vaccine could be realised and identified the key implications of these findings. We found novel tuberculosis vaccines meeting the WHO preferred product characteristics could have a substantial impact, which would vary depending on delivery and vaccine characteristics. Inclusion of a vaccination campaign would be crucial for rapid impact. Most lives could be saved by novel vaccine introduction in the WHO South-East Asian region and African region, and higher rate reductions could be seen in low-income countries. Failing to meet the End TB target to develop and license a vaccine for adolescents and adults by 2025, and to quickly scale up roll-out in all countries, could lead to around 3 million more deaths in low-income and middle-income countries, whereas introduction at a pace similar to that achieved with COVID-19 vaccines could increase the number of lives saved before 2050 by around $60\%$. Implications of all the available evidence Our new evidence supports investment decisions in vaccine development, manufacturing, and delivery. Millions of additional deaths could be averted with rapid development and licensing of novel tuberculosis vaccines, and preparations should be made for their prompt introduction, including in campaigns, ideally at the pace that COVID-19 vaccines have been introduced.
This situation raises crucial questions for global and country decision makers, including the following: how many lives will be lost if we fail to roll out a novel tuberculosis vaccine by 2025? *What is* the potential impact if, instead, vaccines are introduced and rolled out following more traditional timelines? And how would these impacts vary by WHO region, income level, and tuberculosis burden?
We aimed to estimate the potential impact of vaccines meeting the WHO specifications5 in low-income and middle-income countries (LMICs) across a range of introduction and scale-up scenarios.
## Model development and calibration
To estimate the impact of novel tuberculosis vaccines, we developed a compartmental age-stratified dynamic *Mycobacterium tuberculosis* transmission model by adapting features of previous models.8, 9 We represented tuberculosis natural history with eight compartments, allowing for M tuberculosis infection along a spectrum from uninfected to active clinical disease.10, 11 A detailed description is provided in appendix 5 (pp 3–14).
We incorporated an access-to-care structure to represent systematic differences in tuberculosis burden, social protection, and health-care access by socioeconomic status.12 The access-to-care structure contains a high-access-to-care category, representing the top three income quintiles (ie, $60\%$ of the population per country), and a low-access-to-care category, representing the bottom two income quintiles (ie, $40\%$ of the population per country). We assumed no transition between strata, and random mixing (appendix 5 pp 9, 10).
To account for the influences of HIV and antiretroviral therapy (ART) on the risk of infection and progression to disease,13, 14 we classified countries as having a higher tuberculosis burden due to HIV if more than $15\%$ of tuberculosis cases were among people living with HIV and HIV prevalence was greater than $1\%$ (appendix 5 pp 21, 22). We modelled an HIV structure including categories in which people were classified as HIV-uninfected, HIV-infected and not on ART, and HIV-infected and on ART. The tuberculosis mortality rate and progression risk were increased in both HIV-infected compartments, with greater increases in those not on ART.
For each country, we calibrated a model to epidemiological data using history matching with emulation through the hmer R package,15 generating at least 1000 fitted parameter sets per country. Each country model was independently fitted to nine calibration targets in 2019: the country-specific tuberculosis incidence rate (for all ages, those aged 0–14 years, and those 15 years and older, separately), country-specific tuberculosis case notification rate (for all ages, those aged 0–14 years, and those 15 years and older, separately), country-specific tuberculosis mortality rate (for all ages), the global fraction of subclinical tuberculosis among active tuberculosis, and the global risk ratio of active tuberculosis for high-access-to-care relative to low-access-to-care. Models for countries classified as having a high tuberculosis burden due to HIV were fit to four additional country-specific all-age targets in 2019: HIV prevalence, ART coverage, tuberculosis incidence rate in people living with HIV, and tuberculosis mortality rate in people living with HIV. We used the distribution of results produced by these parameter sets to quantify estimation uncertainty.16
## Policy scenarios
For each country, a primary baseline scenario with no novel vaccine introduction was simulated, assuming non-vaccine tuberculosis interventions continue at current trends (ie, the status quo, no-new-vaccine baseline scenario). Because reported country-level data include the high coverage of neonatal BCG vaccination and we anticipate no discontinuation across the model time horizon,17 neonatal BCG vaccination was not explicitly modelled.
Aligning with the product characteristics described in the WHO preferred product characteristics, we evaluated a novel tuberculosis vaccine for adolescents and adults, and a novel vaccine for neonates and infants.5 Vaccines were assumed to prevent disease by reducing progression to subclinical disease and confer a mean protection of 10 years. We assumed the vaccine for adolescents and adults would be efficacious in individuals in any tuberculosis infection state at the time of vaccination (ie, pre-infection and post-infection), with $50\%$ vaccine efficacy. We assumed the vaccine for infants would be efficacious in individuals who were not infected with M tuberculosis at the time of vaccination (ie, pre-infection), with $80\%$ efficacy (appendix 5 p 26).
Roll-out of the vaccine for infants was simulated in two scenarios, and, separately, roll-out of the vaccine for adolescents and adults was simulated in three scenarios, with assumptions confirmed through consultation with a range of global tuberculosis vaccine experts involved in research, government, academia, and policy making. The base-case and accelerated scale-up scenarios for the infant vaccine involved routine neonatal vaccination with $85\%$ coverage. The base-case and accelerated scale-up scenarios for the adolescent and adult vaccine involved routine vaccination of those aged 9 years ($80\%$ coverage), with a one-time vaccination campaign for all individuals aged 10 years and older ($70\%$ coverage). The routine-only scenario (ie, the vaccine for adolescents and adults only) assumed routine vaccination of those aged 9 years ($80\%$ coverage). We assumed no differential vaccination by HIV infection or access-to-care status.
We evaluated vaccine delivery scenarios by varying the introduction year and scale-up trends between scenarios and countries (table 1; appendix 5 pp 26–30). In the base-case and routine-only scenarios, based on data from historical vaccine introduction, vaccines were assumed to be introduced in country-specific years and linearly scaled up to coverage targets across 5 years. To estimate introduction years, countries were divided into those that would be procuring with support from Gavi, the Vaccine Alliance and those that would be self-procuring. Factors influencing the timing of vaccine introduction were identified through expert consultation, and included disease burden, previous early adopter status, timelines for Gavi processes, capacity for immunisation, country-specific registration timelines, and commercial prioritisation. A scoring system was applied to each factor, and countries were assigned an aggregate score ranking their introduction position. The number of countries introducing the vaccine per year was informed by pneumococcal vaccine scale-up.18 In the accelerated scale-up scenarios, to more resemble the pace of COVID-19 vaccine introduction, all countries introduced vaccines in 2025 with coverage targets reached instantly. Table 1Characteristics of modelled vaccine delivery scenariosScenarios for the infant vaccineScenarios for the adolescent and adult vaccineBase-caseAccelerated scale-upBase-caseAccelerated scale-upRoutine-onlyAges targetedRoutine for infantsRoutine for infantsRoutine for those aged 9 years and a one-time vaccination campaign scaled up across 5 years for those aged 10 years or olderRoutine for those aged 9 years and a one-time vaccination campaign in 2025 for those aged 10 years or olderRoutine for those aged 9 yearsIntroduction yearCountry-specific2025Country-specific2025Country-specificVaccine roll-out trend5-year linear scale-up to coverageInstant scale-up to coverage5-year linear scale-up to coverageInstant scale-up to coverage5-year linear scale-up to coverageCoverage target (low, medium, and high)$75\%$, $85\%$, and $95\%$$75\%$, $85\%$, and $95\%$$70\%$, $80\%$, and $90\%$ for those aged 9 years; $50\%$, $70\%$, and $90\%$ for those aged 10 years and older$70\%$, $80\%$, and $90\%$ for those aged 9 years; $50\%$, $70\%$, and $90\%$ for those aged 10 years and older$70\%$, $80\%$, and $90\%$ for those aged 9 years; $50\%$, $70\%$, and $90\%$ for those aged 10 years and older
## Health impact indicators
We calculated the cumulative number of tuberculosis cases, treatments, and deaths averted between vaccine introduction and 2050, compared with the number estimated by the baseline scenario between the corresponding years, and we calculated tuberculosis incidence and mortality rate reductions in 2050 for each vaccine scenario compared with the rates estimated by the baseline in 2050. Incidence rates in 2035 for each vaccine scenario were estimated to investigate the feasibility of meeting the 2035 End TB target. Results are presented as the median and $95\%$ uncertainty range for all countries modelled, WHO region, World Bank income group,19 and WHO tuberculosis burden level.1
## Additional scenario analyses
We conducted scenario analyses to evaluate alternative assumptions regarding vaccine characteristics, delivery, and the baseline scenario. We simulated vaccine scenarios with lifelong protection for both vaccines, as well as scenarios with efficacy of the vaccine for adolescents and adults increased to $75\%$. For each scenario, low-coverage and high-coverage targets were compared with the medium-coverage targets used for the main analyses. We explored an alternative baseline: the 2025 End TB no-new-vaccine baseline, which assumed strengthening of non-vaccine tuberculosis interventions to meet the 2025 End TB incidence target,3 providing an alternative estimate of impact assuming more effective deployment of existing measures (appendix 5 p 25).
## Role of the funding source
The funder was involved in the development of the research question, study design, and provided comments on the manuscript draft, but had no role in the collection, analysis, and interpretation of the data, or writing of the report.
## Results
Epidemiological and demographic data were available to model 115 of 135 LMICs. We successfully calibrated 105 of 115 countries, accounting for $93\%$ of global tuberculosis cases and deaths in 2019. Calibrated model incidence and mortality rate trends for WHO regions, WHO tuberculosis burden levels, and World Bank income groups are given in appendix 5 (p 41). Country-specific vaccine introduction years (used in base-case and routine-only scenarios) ranged between 2028 and 2047 (appendix 5 pp 35–38). Figure 1 shows the cumulative number of countries introducing the vaccine per year, with $50\%$ of countries introducing the vaccine by 2034.Figure 1Assumed cumulative number of countries introducing the novel vaccine by year for the base-case and routine-only scenariosBase-case assumes introduction of routine vaccination for those aged 9 years and one-off vaccination for those aged 10 years and older. Routine-only assumes introduction of routine vaccination among those aged 9 years only. The earliest vaccine introduction occurs in 2028 and the latest in 2047. See appendix 5 (pp 26–33) for full details.
Our findings suggest that a vaccine for adolescents and adults with $50\%$ efficacy and 10-years of protection in the base-case scenario could avert approximately 44·0 million ($95\%$ uncertainty range 37·2–51·6) cases for all countries compared with the status quo no-new-vaccine baseline, including 1·4 million (1·2–1·6) cases of drug-resistant tuberculosis (table 2; appendix p 69). High numbers of cases overall could be averted in the WHO African region and South-East Asian region, which contribute the highest number to the global total, and 34·3 million (28·6–40·3) cases could be averted in lower-middle-income countries (table 2, figure 2). By 2050, 5·0 million ($95\%$ uncertainty range 4·6–5·4) deaths could be averted across all countries, including 2·2 million in the South-East Asian region, 2·1 million in the African region, and 4·1 million in lower-middle-income countries (table 2, figure 2). By 2050, 24·9 million ($95\%$ uncertainty range 21·9–27·3) treatments could be averted, with 11·7 million (10·1–13·4) averted treatments in the South-East Asian region alone. In the 27 countries categorised by WHO as having a high tuberculosis burden of the 105 countries modelled, 39·8 million ($95\%$ uncertainty range 33·7–46·7) cases, 22·6 million (19·9–24·8) treatments, and 4·5 million (4·2–4·9) deaths could be averted by 2050; around ten times higher than those averted in all other countries combined (table 2, figure 2).Table 2Cumulative cases, treatments, and deaths averted between vaccine introduction and 2050, and incidence and mortality rate reductions in 2050 by WHO region, WHO tuberculosis burden level, and World Bank income group for select vaccine scenarios (all 10-year duration of protection and medium coverage targets)All modelled countriesWHO regionWHO tuberculosis burden levelWorld Bank income groupAfrican regionRegion of the AmericasEastern Mediterranean regionEuropean regionSouth-East Asian regionWestern Pacific regionHigh-burden countriesAll other countriesLow-income countriesLower-middle-income countriesUpper-middle-income countriesAdolescent and adult vaccineBase-caseAverted cases before 2050, millions44·0 (37·2–51·6)13·9 (11·7–16·7)0·5 (0·5–0·6)3·9 (3·1–4·8)0·3 (0·3–0·4)19·5 (15·9–23·1)5·9 (5·0–6·9)39·8 (33·7–46·7)4·1 (3·4–4·9)5·0 (4·1–6·0)34·3 (28·6–40·3)4·7 (4·1–5·4)Averted deaths before 2050, millions5·0 (4·6–5·4)2·1 (1·9–2·3)0·04 (0·03–0·04)0·3 (0·2–0·4)0·03 (0·03–0·03)2·2 (2·0–2·6)0·3 (0·2–0·3)4·5 (4·2–4·9)0·5 (0·4–0·5)0·6 (0·5–0·6)4·1 (3·7–4·4)0·4 (0·3–0·5)Averted treatment before 2050, millions24·9 (21·9–27·3)6·3 (5·7–6·8)0·4 (0·3–0·4)2·4 (2·0–2·8)0·2 (0·2–0·3)11·7 (10·1–13·4)3·8 (3·3–4·2)22·6 (19·9–24·8)2·3 (2·0–2·6)2·9 (2·5–3·2)19·0 (16·6–21·2)2·9 (2·7–3·2)Incidence rate reduction in 2050, %25·$4\%$ (23·9–27·7)27·$0\%$ (25·7–31·3)15·$9\%$ (15·2–16·9)26·$7\%$ (23·7–31·6)20·$2\%$ (18·6–22·6)25·$4\%$ (23·3–28·2)19·$8\%$ (18·3–22·2)25·$4\%$ (23·8–27·9)25·$1\%$ (24·1–26·6)27·$3\%$ (26·0–29·1)26·$1\%$ (24·3–28·9)16·$7\%$ (15·8–18·0)Mortality rate reduction in 2050, %27·$1\%$ (25·6–30·1)27·$7\%$ (26·3–33·3)17·$7\%$ (16·8–18·7)28·$1\%$ (25·0–32·8)19·$9\%$ (18·6–21·6)26·$5\%$ (24·3–29·4)23·$1\%$ (21·2–25·8)27·$3\%$ (25·5–30·6)25·$9\%$ (25·0–27·1)27·$8\%$ (26·6–29·4)27·$6\%$ (25·8–31·3)19·$4\%$ (18·1–21·3)Accelerated scale-upAverted cases before 2050, millions65·5 (55·6–76·0)19·5 (16·7–23·1)0·8 (0·7–1·0)5·4 (4·3–6·7)0·6 (0·5–0·7)31·0 (25·8–36·4)8·1 (6·9–9·5)58·6 (49·9–67·9)7·0 (5·8–8·2)7·5 (6·2–9·0)51·7 (43·6–60·2)6·4 (5·6–7·2)Averted deaths before 2050, millions7·9 (7·3–8·5)3·1 (2·9–3·4)0·1 (0·1–0·1)0·5 (0·4–0·6)0·1 (0·1–0·1)3·8 (3·3–4·3)0·4 (0·4–0·4)7·0 (6·4–7·6)0·8 (0·8–0·9)0·9 (0·8–1·0)6·5 (5·9–7·0)0·5 (0·4–0·6)Averted treatment before 2050, millions38·6 (34·4–42·3)9·2 (8·5–9·9)0·6 (0·5–0·7)3·4 (2·9–4·0)0·4 (0·4–0·5)19·5 (16·8–22·2)5·3 (4·8–5·9)34·6 (30·7–37·9)4·0 (3·5–4·4)4·5 (4·0–5·0)30·0 (26·5–33·3)4·1 (3·7–4·4)Incidence rate reduction in 2050, %25·$2\%$ (23·9–27·5)27·$6\%$ (26·3–32·1)15·$2\%$ (14·4–16·2)27·$1\%$ (24·5–31·4)18·$4\%$ (16·4–21·6)24·$7\%$ (22·8–27·3)19·$4\%$ (18·1–21·3)25·$2\%$ (23·8–27·6)25·$3\%$ (24·5–26·8)27·$5\%$ (26·3–29·2)25·$9\%$ (24·3–28·6)16·$3\%$ (15·5–17·3)Mortality rate reduction in 2050, %26·$7\%$ (25·2–29·9)28·$2\%$ (26·8–34·6)16·$2\%$ (15·3–17·3)27·$9\%$ (25·2–32·3)18·$1\%$ (16·5–20·7)25·$3\%$ (23·2–28·2)21·$8\%$ (20·2–24·3)26·$8\%$ (25·1–30·4)26·$1\%$ (25·3–27·2)27·$7\%$ (26·6–29·2)27·$2\%$ (25·5–31·0)18·$4\%$ (17·3–20·0)Routine-onlyAverted cases before 2050, millions8·8 (7·6–10·1)3·5 (3·0–3·9)0·04 (0·03–0·05)0·9 (0·7–1·2)0·02 (0·02–0·03)3·4 (2·6–4·4)1·0 (0·8–1·2)8·1 (7·0–9·3)0·7 (0·6–0·8)1·1 (0·9–1·3)7·2 (6·2–8·3)0·5 (0·4–0·7)Averted deaths before 2050, millions1·1 (0·9–1·2)0·5 (0·4–0·6)0·003 (0·003–0·004)0·1 (0·1–0·1)0·002 (0·002–0·003)0·4 (0·3–0·5)0·1 (0·0–0·1)1·0 (0·8–1·1)0·1 (0·1–0·1)0·1 (0·1–0·1)0·9 (0·7–1·0)0·1 (0·0–0·1)Averted treatment before 2050, millions4·1 (3·7–4·6)1·2 (1·1–1·4)0·03 (0·02–0·03)0·5 (0·4–0·6)0·01 (0·01–0·02)1·8 (1·4–2·2)0·6 (0·5–0·7)3·8 (3·4–4·2)0·3 (0·3–0·4)0·6 (0·5–0·6)3·3 (2·9–3·8)0·3 (0·2–0·3)Incidence rate reduction in 2050, %9·$9\%$ (9·0–11·6)11·$2\%$ (10·3–14·7)3·$4\%$ (3·1–3·9)11·$9\%$ (9·9–15·3)4·$1\%$ (3·4–5·2)9·$1\%$ (7·8–11·1)7·$7\%$ (6·5–9·5)10·$2\%$ (9·1–12·0)8·$0\%$ (7·3–9·2)10·$5\%$ (9·6–11·9)10·$4\%$ (9·2–12·5)5·$2\%$ (4·4–6·3)Mortality rate reduction in 2050, %9·$9\%$ (8·9–12·3)10·$7\%$ (9·7–15·2)3·$7\%$ (3·3–4·2)11·$9\%$ (9·9–15·1)3·$8\%$ (3·3–4·5)8·$7\%$ (7·3–10·7)9·$2\%$ (7·5–11·7)10·$2\%$ (9·1–12·9)7·$2\%$ (6·5–8·1)9·$6\%$ (8·8–10·7)10·$2\%$ (9·0–13·1)6·$2\%$ (5·2–7·8)Infant vaccineBase-caseAverted cases before 2050, millions6·7 (5·8–7·7)2·9 (2·5–3·4)0·03 (0·02–0·03)0·8 (0·6–1·1)0·02 (0·01–0·02)2·2 (1·6–2·8)0·8 (0·6–1·0)6·2 (5·3–7·1)0·5 (0·4–0·6)0·9 (0·7–1·1)5·4 (4·7–6·2)0·4 (0·3–0·5)Averted deaths before 2050, millions0·9 (0·8–1·0)0·5 (0·4–0·6)0·003 (0·002–0·003)0·1 (0·1–0·1)0·002 (0·002–0·002)0·3 (0·2–0·4)0·1 (0·0–0·1)0·8 (0·7–1·0)0·1 (0·1–0·1)0·1 (0·1–0·1)0·7 (0·6–0·9)0·1 (0·0–0·1)Averted treatment before 2050, millions2·7 (2·4–2·9)0·9 (0·8–0·9)0·02 (0·01–0·02)0·4 (0·3–0·5)0·009 (0·008–0·01)1·0 (0·8–1·2)0·4 (0·3–0·5)2·4 (2·2–2·7)0·2 (0·2–0·3)0·4 (0·4–0·5)2·1 (1·9–2·3)0·2 (0·1–0·2)Incidence rate reduction in 2050, %8·$8\%$ (7·9–10·4)11·$0\%$ (10·0–14·5)2·$7\%$ (2·4–3·1)12·$0\%$ (9·7–15·6)2·$9\%$ (2·5–3·4)6·$9\%$ (5·8–8·6)7·$2\%$ (5·9–9·2)9·$0\%$ (8·1–10·7)7·$1\%$ (6·4–8·2)9·$8\%$ (8·9–11·1)9·$1\%$ (8·1–11·1)4·$7\%$ (3·9–5·9)Mortality rate reduction in 2050, %9·$8\%$ (8·7–12·0)11·$3\%$ (10·1–15·7)3·$7\%$ (3·2–4·3)13·$4\%$ (10·5–18·1)3·$3\%$ (2·9–3·9)7·$2\%$ (5·9–9·6)11·$2\%$ (8·5–15·4)10·$1\%$ (8·9–12·5)7·$1\%$ (6·4–8·0)9·$9\%$ (9·0–11·2)10·$0\%$ (8·7–12·5)6·$6\%$ (5·4–8·5)Accelerated scale-upAverted cases before 2050, millions16·3 (14·0–18·8)6·3 (5·4–7·2)0·1 (0·1–0·1)1·7 (1·3–2·2)0·1 (0·1–0·1)6·6 (5·1–8·6)1·5 (1·2–1·9)14·7 (12·6–17·1)1·6 (1·3–1·9)2·2 (1·8–2·8)13·3 (11·4–15·5)0·8 (0·6–0·9)Averted deaths before 2050, millions2·3 (2·0–2·6)1·1 (0·9–1·2)0·007 (0·006–0·008)0·2 (0·1–0·2)0·007 (0·006–0·009)0·9 (0·7–1·2)0·1 (0·1–0·2)2·0 (1·8–2·3)0·2 (0·2–0·3)0·3 (0·2–0·3)1·9 (1·6–2·2)0·1 (0·1–0·1)Averted treatment before 2050, millions7·7 (6·9–8·6)2·2 (2·0–2·4)0·04 (0·04–0·05)0·9 (0·8–1·2)0·04 (0·04–0·05)3·6 (2·9–4·3)0·9 (0·7–1·0)6·9 (6·2–7·8)0·7 (0·7–0·8)1·1 (1·0–1·3)6·2 (5·5–7·0)0·4 (0·3–0·4)Incidence rate reduction in 2050, %14·$3\%$ (13·0–16·7)16·$7\%$ (15·4–21·6)4·$6\%$ (4·2–5·2)17·$6\%$ (14·5–22·3)7·$5\%$ (6·2–9·8)12·$9\%$ (11·0–15·8)10·$3\%$ (8·7–12·6)14·$4\%$ (13·0–17·0)13·$4\%$ (12·4–14·9)16·$3\%$ (15·1–18·1)14·$9\%$ (13·3–17·9)6·$5\%$ (5·6–7·8)Mortality rate reduction in 2050, %15·$9\%$ (14·2–19·3)17·$5\%$ (15·9–24·1)5·$8\%$ (5·2–6·6)19·$2\%$ (15·5–24·7)7·$7\%$ (6·5–9·5)13·$4\%$ (11·2–17·0)15·$0\%$ (12·0–19·3)16·$1\%$ (14·3–19·9)14·$1\%$ (13·1–15·3)16·$8\%$ (15·6–18·5)16·$3\%$ (14·4–20·3)8·$9\%$ (7·5–11·1)Data are median estimates ($95\%$ uncertainty range). Cumulative cases, treatments, and deaths averted are calculated for each vaccine scenario compared with the estimated number predicted by 2050 with the status quo no-new-vaccine baseline. Incidence and mortality rate reductions are calculated relative to the incidence and mortality rate predicted in 2050 relative to the status quo no-new-vaccine baseline. See appendix 5 for all scenarios (pp 55–68).Figure 2Cumulative cases, treatments, and deaths averted between vaccine introduction and 2050, and incidence and mortality rate reductions in 2050 for the vaccine for adolescents and adults with varying delivery scenarios ($50\%$ efficacy vaccine, medium coverage, 10-year duration of protection), by WHO region, WHO tuberculosis burden level, and World Bank income group, expressed relative to a baseline scenario with no new vaccineCumulative cases, treatments, and deaths averted are calculated for each vaccine scenario compared with the estimated number predicted by 2050 with the status quo no-new-vaccine baseline. Incidence and mortality rate reductions are calculated relative to the incidence and mortality rate predicted in 2050 by the status quo no-new-vaccine baseline. Base-case scenario: routine vaccination of those aged 9 years and a one-off campaign for those aged 10 years and older, introduced in country-specific years between 2028 and 2047 and scaled up over 5 years. Accelerated scale-up scenario: routine vaccination of those aged 9 years and a one-off campaign for those aged 10 years and older, introduced in 2025 and scaled up instantly in all countries. Routine-only scenario: routine vaccination of those aged 9 years, introduced in country-specific years between 2028 and 2047 and scaled up over 5 years.
Introducing the vaccine for adolescents and adults in the base-case scenario was predicted to reduce tuberculosis incidence by 25·$4\%$ (23·9–27·7) and deaths by 27·$1\%$ (25·6–30·1) in 2050, compared with the status quo no-new-vaccine baseline scenario (table 2). The incidence reduction ranged from 15·$9\%$ in the WHO region of the Americas to 27·$0\%$ in the African region (table 2, figure 2). Deaths from tuberculosis were estimated to reduce by 17·$7\%$ in the region of the Americas to 28·$1\%$ in the Eastern Mediterranean region by introducing the adolescent and adult vaccine in the base-case scenario. By income group, the relative impact of the adolescent and adult vaccine was higher in low-income and lower-middle-income countries than in upper-middle-income countries (table 2, figure 2).
In both the base-case and accelerated scale-up scenarios, a lower impact of the infant vaccine (compared with the vaccine for adolescents and adults) was estimated before 2050, including 0·4–0·6 times incidence and mortality rate reductions by 2050 and 0·1–0·3 times the number of cases, treatments, and deaths averted (table 2).
Under the accelerated scale-up scenario, a $50\%$ efficacy of the vaccine for adolescents and adults could prevent 7·9 million (7·3–8·5) deaths—2·9 million more than the base-case—and avert 65·5 million (55·6–76·0) cases and 38·6 million (34·4–42·3) treatments (table 2, figure 2). By contrast, by only routinely vaccinating those aged 9 years (ie, the routine-only scenario), 8·8 million (7·6–10·1) cases, 4·1 million (3·7–4·6) treatments, and 1·1 million (0·9–1·2) deaths would be averted compared with the status quo no-new-vaccine baseline scenario (table 2, figure 2).
Assuming non-vaccine interventions do not improve in the future (ie, the status quo no-new-vaccine baseline), the outcomes of the base-case scenario of introducing the vaccine for adolescents and adults suggest we would reach $34\%$ of the 2035 global target to reduce tuberculosis cases by $90\%$ compared with 2015 levels, and under the accelerated scale-up scenario, we would reach $41\%$ of the target. Assuming the 2025 End TB target of reducing the incidence by $50\%$ compared with 2015 levels is met (ie, the 2025 End TB no-new-vaccine baseline), progress would be increased further, with the base-case and accelerated scale-up scenarios reaching $82\%$ of the target.
Impact results from scenarios with lifelong protection, $75\%$ efficacy, and low-coverage and high-coverage targets are provided in appendix 5 (pp 55–68). Assuming lower coverage targets or the 2025 End TB no-new-vaccine baseline led to reduced vaccine impact compared with vaccines with medium coverage or the status quo no-new-vaccine baseline, and vaccines with higher coverage, $75\%$ efficacy, or lifelong protection led to increased vaccine impact compared with vaccines with medium coverage, $50\%$ efficacy, or 10 years protection.
## Discussion
Our results suggest that novel tuberculosis vaccines could substantially reduce the tuberculosis burden in the coming decades. Relative to the status quo no-new-vaccine baseline, the base-case scenario—in which a tuberculosis vaccine for adolescents and adults with $50\%$ efficacy was introduced during 2028–47—could prevent 44·0 million cases and 5·0 million deaths before 2050, including 2·2 million deaths in the WHO South-East Asian region and 2·1 million deaths in the African region. The more ambitious accelerated scale-up scenario could prevent 65·5 million cases and 7·9 million deaths relative to baseline (which is around $60\%$ more deaths than the base-case scenario). The less ambitious routine-only scenario could prevent 8·8 million cases and 1·1 million deaths relative to baseline (which is around a fifth of the base-case scenario).
Impact estimates for vaccine introduction varied by region in our results. Although incidence and mortality rate reductions achievable by 2050 were similar between high-tuberculosis-burden countries and all other countries, the number of cases, treatments, and deaths averted were around ten times higher than those averted in all other countries, emphasising the need to focus on high-burden countries to maximise health impact. Large numbers of averted cases, treatments, and deaths were predicted in the African region and South-East Asian region, and in lower-middle-income countries, which are arguably populations in the greatest need.
Our modelling suggests that campaigns will be important to expedite health gains from vaccination. The base-case and routine-only scenarios offer a direct comparison of implementing vaccination with and without a campaign for those 10 years and older. The base-case scenario averted up to six times as many cases, deaths, and treatments as the routine-only scenario, supporting the need to include a campaign in any future delivery strategy to maximise health impact.
A new vaccine will be an important tool to accelerate progress towards the 2035 End TB targets. Conservatively assuming non-vaccine interventions do not improve in the future (status quo no-new-vaccine baseline) and roll out from 2028 in line with the pace of historical vaccine uptake, the base-case scenario suggests we could reach around a third of the 2035 global target. More optimistic assumptions, in which the 2025 End TB targets are met before vaccine roll-out (2025 End TB no-new-vaccine baseline), combined with the accelerated scale-up scenario, suggest more than $80\%$ of the global 2035 target could be met.
Two systematic reviews have highlighted potential health impacts of novel tuberculosis vaccines.20, 21 Our study expands on their findings, and it addresses some identified gaps. We showed that a vaccine for adolescents and adults would have greater and more rapid health impacts than a vaccine for infants before 2050. The largest burden of pulmonary tuberculosis disease is often found in adults;1 and in our modelling the vaccine for adolescents and adults was delivered to ages with a higher burden of tuberculosis compared with the vaccine for infants. Because health outcomes are estimated for 2050, the maximum follow-up time between vaccine delivery and impact calculation is 25 years. Therefore, even with the duration of protection increased, the infant vaccine is unlikely to protect those at highest risk of progressing to active disease in most countries during our simulation.
Meeting the End TB target to develop and license a vaccine for adolescents and adults by 2025, and introducing this vaccine at a pace similar to that of COVID-19 vaccines (accelerated scale-up) could avert around $60\%$ more deaths compared with introduction at a historical pace (base-case). The pace of COVID-19 vaccine introduction in LMICs, which was albeit slower than in high-income countries, was much faster than our base-case introduction assumption. As of February, 2023, more than $10\%$ of the population in almost $95\%$ of LMICs (ie, 122 of 129 countries reporting data) have been fully vaccinated since COVID-19 vaccines have been available, showing that faster vaccine introduction in LMICs is possible with high political will and financial resources.22 This situation is more similar to our accelerated scale-up scenario, which averted up to 2·9 million more deaths, than our base-case scenario. Although the benefits of rolling out a vaccine from 2028 at a pre-COVID-19 pace are predicted to be large, the increase in deaths shows the consequences of failing to rapidly introduce a vaccine. Unlike COVID-19, tuberculosis is a disease of those on low-incomes, which does not have the associated novelty, nor the same effect on high-income countries. Therefore, tuberculosis vaccines need concerted, sustained policy attention to overcome these barriers.
We successfully calibrated 105 of 135 LMICs, representing $93\%$ of global tuberculosis incidence. Excluding 30 countries will underestimate the number of cases, deaths, and treatments averted, and could bias the generalisability of the relative impact results if the epidemic in the excluded countries is substantially different than those included. Model misspecification and structural uncertainty is possible if country-specific epidemiology does not align with our structure. We used the best available estimates from literature, combined with previous knowledge and expert opinion, to substantiate the prior distributions. Therefore, our results reflect the inherent uncertainty in our knowledge of tuberculosis natural history. For newer discoveries in the field (eg, subclinical disease and self-clearance) data are sparse, and uncertainty is wide, which could bias our vaccine impact estimates. We made assumptions on parameters (eg, assuming the same amount of protection against reinfection in the infection and resolved compartments), which might slightly underestimate vaccine impact. We predicted tuberculosis declining across time, but the projected declines are unlikely to match actual declines, primarily affecting estimates of reaching End TB strategy goals, and numbers averted.
Because there are no new vaccines for tuberculosis, we assumed the characteristics of the modelled vaccines aligned with the recommendations in the WHO preferred product characteristics. Our impact results could be overestimated or underestimated if values for efficacy and duration of protection are lower or higher than the actual characteristics of a new vaccine. We assumed the vaccine for adolescents and adults would be efficacious in all individuals, because testing for tuberculosis infection before vaccination would be costly and logistically difficult. However, most trials have only enrolled individuals who are either positive for interferon-gamma release assay or negative for interferon-gamma release assay.6, 7 If the vaccine will only be efficacious in those who are positive for interferon-gamma release assay or those who are negative, our results will be overestimates, as shown previously.8 We assumed equivalent vaccine efficacy in people living with HIV and those who are HIV-naive, but vaccines are not always as efficacious in individuals who are immunocompromised,23, 24 which would reduce impact in countries classified as having a high tuberculosis burden associated with HIV.
For vaccine delivery, we attempted to represent a reasonable breadth of possibilities by speaking to experts and evaluating low and high coverage, efficacy, and introduction scenarios. Should there be rapid developments in tuberculosis diagnostics and treatments, or if funding were substantially increased, the impacts could be overestimates or underestimates. Our more ambitious scenario, accelerated scale-up, is less realistic than the base-case scenario, particularly in some LMICs. The scenario assumes a vaccine candidate would be ready for licensure, the supply exists, and that countries are positioned to make an introduction decision resulting in immediate uptake, all within the next 2–3 years, which is unlikely to be attainable by all countries. No specific risk groups were vaccinated in our model; however, initial delivery within countries could be through a targeted approach, which was previously shown to have a large population impact per vaccinated individual.25, 26, 27, 28, 29 Some countries could initially vaccinate groups at the highest risk of developing disease or who contribute the most to transmission, whereas others could focus on vulnerable ages or those who have had contact with an individual with confirmed tuberculosis disease. Understanding how a new tuberculosis vaccine could be introduced in different settings is an important area for future research.
There are remaining gaps that modelling can address to provide evidence for investing in tuberculosis vaccine development and delivery to inform the Full Value of Vaccine Assessment.30 Estimates of the cost-effectiveness, budget effect, and wider benefits of specific tuberculosis vaccine candidates would support research investment decision making. Future modelling research can help to better understand potential vaccine effectiveness considering a variety of factors, such as age, sex, and specific risk groups. We included an access-to-care structure to account for differences in tuberculosis burden and health-care access, which could be used to investigate differential vaccine targeting. To maximise the potential evidence available to countries, creating detailed individual country models to inform vaccine introduction decision making would be beneficial.
Our results suggest that novel tuberculosis vaccines could have a substantial impact on cases of and deaths from tuberculosis, which would vary depending on vaccine and delivery characteristics. Vaccination campaigns will be crucial for rapid impact, and an accelerated introduction that is done at a similar pace to that of COVID-19 vaccine introduction could save around $60\%$ more lives before 2050 than the same vaccine introduced and scaled up across 20 years. The COVID-19 pandemic has shown the advantages that billions of dollars of investment can have on vaccine research and development, and it provides an illustration of what is possible to achieve with novel tuberculosis vaccines. Continued investment in tuberculosis vaccine research is required to strengthen vaccine development, trials, and manufacturing, and to support prompt introduction and scale-up.
## Data sharing
No individual level participant data were used for this modelling study. Epidemiological data used are available from the WHO Global Tuberculosis Report and are summarised in appendix 5 (pp 41–49). Population estimates and projections are available from the UN Department of Economic and Social Affairs World Population Prospects 2019. The analytic code will be available immediately following publication, indefinitely, for anyone who wishes to access the data for any purpose.
## Declaration of interests
SM reports employment by the International AIDS Vaccine Initiative, a non-profit product development partnership supporting the access-oriented development of vaccines for several disease areas, including tuberculosis, and grant funding from WHO. MJ is funded by the Bill & Melinda Gates Foundation, Gavi the Vaccine Alliance, the UK Research Institute, the National Institute for Health Research, the European Commission, and the Wellcome Trust, and reports leadership or fiduciary roles in the board, society, committee, or advocacy groups for WHO and Gavi. RCH reports employment by Sanofi Pasteur, unrelated to tuberculosis and outside the submitted work. NAM received consulting fees from The Global Fund to Fight AIDS, Tuberculosis and Malaria and WHO, and reports funding to their institution from the US Centers for Disease Control and Prevention, the Gates Foundation, the National Institute of Health, and the US Council of State and Territorial Epidemiologists. RGW is funded for other work by the Wellcome Trust (218261/Z/19/Z), the National Institute of Health (1R01AI147321–01), EDCTP (RIA208D-2505B), the UK's Medical Research Council (CCF17–7779 via SET Bloomsbury), the Economic and Social Research Council (ES/P$\frac{008011}{1}$), the Gates Foundation (OPP1084276, OPP1135288, and INV-001754), and WHO. All other authors declare no competing interests.
## Supplementary Materials
French translation of the abstract Spanish translation of the abstract Italian translation of the abstract Dutch translation of the abstract
Supplementary appendix 5
## Acknowledgments
This *Article is* part of a WHO-commissioned work on the Full Value Assessment of Tuberculosis Vaccines ($\frac{2020}{985800}$-0). We thank all the attendees at WHO meetings on the Full Value Assessment of Tuberculosis Vaccines for insightful advice and direction. NG and MZ acknowledge funding from the Netherlands (4000002598 [2019–21]) to their institution. CKW, DS, MJ, MQ, and RAC acknowledge funding from the Gates Foundation (INV-001754) to their institution. We thank Philippe Glaziou (WHO) for reviewing and providing helpful suggestions on the paper.
## References
1. **Global tuberculosis report**. (2021)
2. McQuaid CF, Vassall A, Cohen T, Fiekert K, White RG. **The impact of COVID-19 on TB: a review of the data**. *Int J Tuberc Lung Dis* (2021) **25** 436-446. PMID: 34049605
3. **The End TB Strategy**. (Aug 16, 2015)
4. **New data shows COVID-19 combined with funding shortfalls are devastating efforts to end TB by 2030**. (2021)
5. **WHO Preferred product characteristics for new tuberculosis vaccines**. (2018)
6. Tait DR, Hatherill M, Van Der Meeren O. **Final analysis of a trial of M72/AS01**. *N Engl J Med* (2019) **381** 2429-2439. PMID: 31661198
7. Nemes E, Geldenhuys H, Rozot V. **Prevention of**. *N Engl J Med* (2018) **379** 138-149. PMID: 29996082
8. Harris RC, Sumner T, Knight GM, Zhang H, White RG. **Potential impact of tuberculosis vaccines in China, South Africa, and India**. *Sci Transl Med* (2020) **12**
9. Knight GM, Griffiths UK, Sumner T. **Impact and cost-effectiveness of new tuberculosis vaccines in low- and middle-income countries**. *Proc Natl Acad Sci USA* (2014) **111** 15520-15525. PMID: 25288770
10. 10Richards AS, Sossen B, Emery JC, et al. The natural history of TB disease—a synthesis of data to quantify progression and regression across the spectrum. Lancet Glob Health (in press).
11. Emery JC, Dodd PJ, Banu S. **Estimating the contribution of subclinical tuberculosis disease to transmission—an individual patient data analysis from prevalence surveys**. *medRxiv* (2022). DOI: 10.1101/2022.06.09.22276188
12. Siroka A, Law I, Macinko J. **The effect of household poverty on tuberculosis**. *Int J Tuberc Lung Dis* (2016) **20** 1603-1608. PMID: 27931334
13. Kwan CK, Ernst JD. **HIV and tuberculosis: a deadly human syndemic**. *Clin Microbiol Rev* (2011) **24** 351-376. PMID: 21482729
14. Suthar AB, Lawn SD, del Amo J. **Antiretroviral therapy for prevention of tuberculosis in adults with HIV: a systematic review and meta-analysis**. *PLoS Med* (2012) **9**
15. Iskauskas A. **hmer: history matching and emulation package**. (2022)
16. Briggs AH, Weinstein MC, Fenwick EAL, Karnon J, Sculpher MJ, Paltiel AD. **Model parameter estimation and uncertainty analysis: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force Working Group-6**. *Med Decis Making* (2012) **32** 722-732. PMID: 22990087
17. **BCG: immunization coverage estimates by WHO region**. (2022)
18. **Immunization**. (2022)
19. **World Bank country and lending groups: historical classification by income in XLSX format**. (2022)
20. Harris RC, Sumner T, Knight GM, White RG. **Systematic review of mathematical models exploring the epidemiological impact of future TB vaccines**. *Hum Vaccin Immunother* (2016) **12** 2813-2832. PMID: 27448625
21. Weerasuriya CK, Clark RA, White RG, Harris RC. **New tuberculosis vaccines: advances in clinical development and modelling**. *J Intern Med* (2020) **288** 661-681. PMID: 33128834
22. **Coronavirus (COVID-19) vaccinations**. (2022)
23. Nicolini LA, Giacobbe DR, Di Biagio A, Viscoli C. **Insights on common vaccinations in HIV-infection: efficacy and safety**. *J Prev Med Hyg* (2015) **56** E28-E32. PMID: 26789829
24. Kumarasamy N, Poongulali S, Beulah FE. **Long-term safety and immunogenicity of the M72/AS01E candidate tuberculosis vaccine in HIV-positive and -negative Indian adults: results from a phase II randomized controlled trial**. *Medicine (Baltimore)* (2018) **97**
25. Shrestha S, Chihota V, White RG, Grant AD, Churchyard GJ, Dowdy DW. **Impact of targeted tuberculosis vaccination among a mining population in South Africa: a model-based study**. *Am J Epidemiol* (2017) **186** 1362-1369. PMID: 29253139
26. Shrestha S, Chatterjee S, Rao KD, Dowdy DW. **Potential impact of spatially targeted adult tuberculosis vaccine in Gujarat, India**. *J R Soc Interface* (2016) **13**
27. Awad SF, Critchley JA, Abu-Raddad LJ. **Epidemiological impact of targeted interventions for people with diabetes mellitus on tuberculosis transmission in India: modelling based predictions**. *Epidemics* (2019) **30**
28. Harris RC, Sumner T, Knight GM. **Age-targeted tuberculosis vaccination in China and implications for vaccine development: a modelling study**. *Lancet Glob Health* (2019) **7** e209-e218. PMID: 30630775
29. Weerasuriya CK, Harris RC, McQuaid CF. **The epidemiologic impact and cost-effectiveness of new tuberculosis vaccines on multidrug-resistant tuberculosis in India and China**. *BMC Med* (2021) **19** 60. PMID: 33632218
30. Hutubessy RCW, Lauer JA, Giersing B. **The Full Value of Vaccine Assessments (FVVA): a framework to assess and communicate the value of vaccines for investment and introduction decision making**. *SSRN* (2021). DOI: 10.2139/ssrn.3841999
|
---
title: Comprehensive characterization of maternal, fetal, and neonatal microbiomes
supports prenatal colonization of the gastrointestinal tract
authors:
- Jee Yoon Park
- Huiyoung Yun
- Seung-been Lee
- Hyeon Ji Kim
- Young Hwa Jung
- Chang Won Choi
- Jong-Yeon Shin
- Joong Shin Park
- Jeong-Sun Seo
journal: Scientific Reports
year: 2023
pmcid: PMC10030461
doi: 10.1038/s41598-023-31049-1
license: CC BY 4.0
---
# Comprehensive characterization of maternal, fetal, and neonatal microbiomes supports prenatal colonization of the gastrointestinal tract
## Abstract
In this study, we aimed to comprehensively characterize the microbiomes of various samples from pregnant women and their neonates, and to explore the similarities and associations between mother-neonate pairs, sample collection sites, and obstetrical factors. We collected samples from vaginal discharge and amniotic fluid in pregnant women and umbilical cord blood, gastric liquid, and meconium from neonates. We identified 19,597,239 bacterial sequences from 641 samples of 141 pregnant women and 178 neonates. By applying rigorous filtering criteria to remove contaminants, we found evidence of microbial colonization in traditionally considered sterile intrauterine environments and the fetal gastrointestinal track. The microbiome distribution was strongly grouped by sample collection site, rather than the mother-neonate pairs. The distinct bacterial composition in meconium, the first stool passed by newborns, supports that microbial colonization occurs during normal pregnancy. The microbiome in neonatal gastric liquid was similar, but not identical, to that in maternal amnionic fluid, as expected since fetuses swallow amnionic fluid in utero and their urine returns to the fluid under normal physiological conditions. Establishing a microbiome library from various samples formed only during pregnancy is crucial for understanding human development and identifying microbiome modifications in obstetrical complications.
## Introduction
The human microbiome potentially carries the answer to many secrets of the human body. It has been linked to maintaining homeostasis in health and is associated with numerous diseases1,2. Recent research has shifted to explore the microbiome in less-studied populations, such as infants or pregnant women, to better understand its role in human development. Microbiome development is likely to start from the in-utero environment and changes in a lifetime, continuously affecting the immune system and metabolism. Pregnancy has been shown to alter microbial populations within the maternal body and may impact future maternal, fetal, and neonatal health3. Pregnancy allows temporary immunotolerance to a foreign body, facilitating microbiome remodelling and potential adaptations to the immune system and metabolism4. Some microbiome studies in pregnancy have proposed that fetal environments, including placenta and amniotic fluid, traditionally known as sterile, contain several characteristic microbiotas not identified in routinely performed culture techniques5,6. However, the biomass of these microbiotas is small and the reliability of the sequencing methods and potential for contamination have been criticized. The association between those microbiota and specific obstetric conditions has not yet been proven and warrants further investigation.
The vagina is the most commonly studied site of bacteria in the female reproductive organ, as it is connected to the uterus through the cervix and is exposed to the skin. Microbiome research in pregnancy, however, has advanced slowly due to ethical concerns and difficulties in accessing samples. Aagaard et al. found that the vaginal microbiome changes during pregnancy based on gestational age and that Lactobacillus species play a role in preventing pathogenic bacterial growth7. More specifically, pregnancy leads to decreased diversity, increased proportion of Lactobacillus, and higher stability in the vaginal microbiome8,9. Some vaginal bacteria have been linked to preterm birth via intrauterine inflammation or infection10–14, yet there are no clinical guidelines for testing or monitoring these microbiota. Other sites that had been evaluated for microbiome in pregnancy are maternal15, oral cavity16, placenta5, amniotic fluid17,18, and neonatal gut19; but previous studies were fragmentary and more systematic research is needed.
In this study, we have comprehensively characterized the microbiome in vaginal discharge (VD) and amniotic fluid (AF) from pregnant women and in umbilical cord blood (CB), gastric liquid (GL), and meconium (M) from their neonates. The goal was to determine the relationships between these samples and various obstetric conditions.
## Description of the study populations and clinical characteristics
A total of 141 low-risk pregnant women were enrolled sequentially and 178 neonates were born from the study population with 37 cases being twin pregnancies. All women were of Asian ethnicity (Korean), and the median age was 34 (interquartile range 31–37) years (Table 1). The proportion of nulliparity was slightly over half of the population ($67\%$), and the median values of height, weight, and body mass index (BMI) were 162 cm, 70 kg, and 27 kg/m2, respectively. About $30\%$ were conceived by assisted reproductive technology (ART), including intrauterine insemination (IUI) and in vitro fertilization with embryo transfer (IVF-ET). As mentioned above, twin pregnancy was approximately one-fourth of the total population, and among them, $19\%$ were monochorionic. The median gestational age at delivery was 37.7 weeks (interquartile range 36.9–38.6), and preterm birth before 37 weeks of gestation was $26.2\%$ ($\frac{37}{141}$). The rate of cesarean section was $55\%$ ($\frac{77}{141}$). Seven neonates had congenital structural anomalies (atrioventricular septal defect, absence of corpus callosum in the brain, achondroplasia, cleft lip, polydactyly, and syndactyly), which did not directly affect the neonatal survival. The frequencies of other obstetric complications or underlying maternal diseases are described in Table 1.Table 1Clinical characteristics of the study population. CharacteristicsValuesAge (years)34 (31–37)Nulliparity$67.4\%$ ($\frac{95}{141}$)Height (cm)162.4 (159.5–165.1)Weight (kg)69.9 (65.7–77.0)BMI (kg/m2)27.0 (25.1–29.8)Pregnancy from IVF-ET$24.1\%$ ($\frac{34}{141}$)Pregnancy from IUI$5.0\%$ ($\frac{7}{141}$)Twin pregnancy$26.2\%$ ($\frac{37}{141}$) Monochorionic twins$18.9\%$ ($\frac{7}{37}$)Gestational age at delivery (weeks)37.7 (36.9–38.6) Preterm birth before 37 weeks$26.2\%$ ($\frac{37}{141}$)Cesarean section$54.6\%$ ($\frac{77}{141}$)Epidural anesthesia$47.5\%$ ($\frac{67}{141}$)Birthweight (g)a2800 (2480–3124)Male neonatesa$50.0\%$ ($\frac{89}{178}$)Low Apgar score < 7 in 1 mina$3.9\%$ ($\frac{5}{178}$)Low Apgar score < 7 in 5 mina$0.6\%$ ($\frac{1}{178}$)Meconium staininga$2.2\%$ ($\frac{4}{178}$)Congenital structural anomalya$3.9\%$ ($\frac{7}{178}$)Obstetric complications and underlying diseases Use of tocolytics due to preterm labor$8.5\%$ ($\frac{12}{141}$) Preterm premature rupture of membranes$3.5\%$ ($\frac{5}{141}$) Cerclage operation$4.3\%$ ($\frac{6}{141}$) Preeclampsia$12.1\%$ ($\frac{17}{141}$) Chronic hypertension$2.1\%$ ($\frac{3}{141}$) Fetal growth restriction$2.8\%$ ($\frac{4}{141}$) Oligohydramnios in the 3rd trimester$5.0\%$ ($\frac{7}{141}$) Gestational thrombocytopenia$3.5\%$ ($\frac{5}{141}$) Gestational diabetes$13.5\%$ ($\frac{19}{141}$) Pregestational diabetes$0.7\%$ ($\frac{1}{141}$) Placenta previa$2.1\%$ ($\frac{3}{141}$) Placental abruption0 ($\frac{0}{141}$) Myoma uteri on ultrasound$12.1\%$ ($\frac{17}{141}$) Endometriosis confirmed before pregnancy$1.4\%$ ($\frac{2}{141}$) Hypothyroidism$9.9\%$ ($\frac{14}{141}$) Hyperthyroidism$0.7\%$ ($\frac{1}{141}$) Allergic diseasesb$4.3\%$ ($\frac{6}{141}$) Psychologic diseases on medicationc$2.8\%$ ($\frac{4}{141}$)BMI, body mass index; IVF-ET, in-vitro fertilization and embryo transfer; IUI, intrauterine insemination. Values are expressed as the median (interquartile range) for continuous variables and percentage for categorical variables.aThe denominator is the number of newborns.bAsthma for three cases, allergic rhinitis, angioedema, and cholinergic urticaria.cMajor depressive disorder for two cases, anxiety disorder, and panic disorder.
## Maternal and neonatal microbiome landscape during delivery
We identified 19,597,239 bacterial sequences and 22,412 unique amplicon sequence variants (ASVs) from 641 samples, including cervicovaginal discharge ($$n = 154$$), amniotic fluid ($$n = 40$$), gastric liquid ($$n = 100$$), umbilical cord blood ($$n = 125$$), meconium ($$n = 160$$), and negative controls ($$n = 62$$). The ASVs were taxonomically annotated, but we found evidence of batch effects in our sequencing data for all sample types except VD (Fig. 1 and Supplementary Fig. S1). The batch effects were likely introduced during library construction for next-generation sequencing (NGS) and not during sequencing itself (Supplementary Fig. S2). However, this was expected because our samples were collected from body sites with low-biomass specimens, making our samples prone to contamination20. Therefore, we expected to find many false positives and applied a series of filters, as outlined in Supplementary Fig. S3. Notably, we found and removed 203 ASVs that were statistically determined as contaminants because they were highly prevalent in negative controls (Supplementary Fig. S4) or they showed higher frequencies in low-concentration samples (Supplementary Figs. S5 and S6).Figure 1Batch effect detection in 16S rRNA amplicon sequencing data. Center log-ratio transformation was used to normalize the filtered ASV table before generating a hierarchically clustered heatmap based on correlation coefficients. AF, amniotic fluid; CB, umbilical cord blood; GL, gastric liquid; M, meconium; VD, cervicovaginal discharge; NC, negative control.
We measured the alpha diversity of the samples by calculating Shannon indices (Fig. 2A). The alpha diversity decreased in the following order: GL, AF, M, CB, and VD. The negative control group showed a slightly higher diversity compared to the VD sample, suggesting that negative controls for 16S amplicon sequencing can have microbiome diversity as rich as real biological specimens. Next, we estimated the beta diversity of our samples by computing the weighted UniFrac distances (Fig. 2B). The samples were moderately well separated by sample collection site when projected using principal coordinates analysis (PCoA).Figure 2Alpha and beta diversity of the Korean maternal and neonatal microbiome. ( A) Alpha diversity: The filtered ASV table was rarefied before Shannon index was computed for each sample. The VD group exhibited the least amount of alpha diversity. AF, amniotic fluid; CB, umbilical cord blood; GL, gastric liquid; M, meconium; VD, cervicovaginal discharge; NC, negative control; (B) Beta diversity: The filtered ASV table was rarefied before the samples were projected into 2D-space with principal coordinates analysis using the weighted UniFrac distance.
To determine if mothers share microbes with their newborns, we compared the average number of ASVs shared between mother-neonate pairs and 1000 pairs randomly selected from different families. We conducted a one-tailed t-test and found that mother-neonate pairs did not have a higher number of shared ASVs compared to randomly selected pairs ($$n = 0$.7$ vs. $$n = 1$.8$, respectively; $$p \leq 1.0$$).
## Clinical relevance of microbiome in pregnancy
To better understand the sources of variation seen in the beta diversity of our samples, we carried out the permutational multivariate analysis of variance (PERMANOVA) using different factors, including clinical information. As shown in Table 2, when all sample types were included in the analysis, the variable “Site” explained $17.2\%$ of the variation (p-value = 0.001), and the variable “LibraryMonth,” was $7.4\%$ (p-value = 0.002). This result indicates that the samples could still be separated well based on the microbiome pattern unique to their body site, despite the significant batch effects present within our dataset. When the analysis was restricted to each sample type, except for the sample VD group, the variable “LibraryMonth” was found to be significant for all sample types. The explanatory power increased to a range between 24.5 and $48.9\%$. These results align with the hypothesis that our samples are predominantly low-biomass specimens and prone to contamination. Table 2Summary of the results (R2 and p-values) from permutational multivariate analysis of variance (PERMANOVA).VariableAll sitesAFCBGLMVDSite0.172 (0.001)–––––LibraryMonth0.074 (0.002)0.489 (0.002)0.338 (0.008)0.245 (0.016)0.26 (0.001)0.051 (0.834)Age0 (0.947)0.021 (0.506)0.012 (0.2)0.019 (0.248)0.016 (0.151)0.016 (0.211)PretermBirth370.002 (0.445)0.029 (0.317)0.003 (0.826)0.012 (0.469)0.026 (0.017)0.004 (0.7)DeliveryMethod0.003 (0.206)0.01 (0.676)0.011 (0.225)0.038 (0.069)0.008 (0.473)0.06 (0.005)HasGDM0.002 (0.365)0.025 (0.372)0.015 (0.134)0.012 (0.446)0.018 (0.086)0.002 (0.891)IVFET0.001 (0.934)0.01 (0.845)0.002 (0.934)0.014 (0.364)0.004 (0.854)0 (0.99)Epidural0.003 (0.162)0.03 (0.328)0.018 (0.108)0.005 (0.76)0.006 (0.559)0.006 (0.47)InducedLabor0.002 (0.373)0.014 (0.676)0.003 (0.839)0.02 (0.226)0.01 (0.378)0.002 (0.828)Hypertension0.009 (0.135)0.038 (0.573)0.047 (0.119)0.011 (0.875)0.03 (0.363)0.019 (0.434)Weight0.002 (0.483)0.011 (0.776)0.022 (0.039)0.003 (0.938)0.006 (0.59)0.011 (0.321)HasTwins0.001 (0.544)0.009 (0.802)0.006 (0.451)0.016 (0.338)0.01 (0.369)0.007 (0.463)BabySex0.005 (0.263)0.024 (0.839)0.019 (0.261)0.031 (0.398)0.015 (0.63)0.022 (0.387)AntibioticsUse0.004 (0.086)–0.025 (0.06)0.007 (0.665)0.031 (0.035)0.014 (0.23)Residuals0.7190.290.4780.5680.560.786Total111111AF, amniotic fluid; CB, umbilical cord blood; GL, gastric liquid; M, meconium; VD, cervicovaginal discharge. Significant values are in bold.
Additionally, the variable “DeliveryMethod” was returned as significant for the VD group, the variables “PretermBirth37” and “AntibioticsUse” for the M group, and the variable “Weight” for the CB group (Fig. 3). We explored the significant variables in each group using PCoA with weighted UniFrac distance. Several ASVs of Lactobacillus and one ASV of Gardnerella were found in the VD group. In the M group *Staphylococcus showed* a strong association with preterm birth. Lastly, the lists of bacterial taxa were connected to the weights of neonates in the CB group. Table 3 shows the analysis of the composition of microbiomes (ANCOM) for various clinical data to study any statistically significant relevance with bacteria in multiple sample types. Figure 3Beta diversity results of the PERMANOVA analysis. Principal coordinates analysis using weighted UniFrac distance is shown for (A) the cervicovaginal discharge samples, (B) and (C) the meconium samples, and (D) the umbilical cord blood samples. Table 3Summary of the results from analysis of composition of microbiomes (ANCOM) at the genus level. VariableSiteTaxonW scoreResultsDeliveryMethodAFd__Bacteria;p__Firmicutes;c__Clostridia;o__Peptostreptococcales-Tissierellales; f__Peptostreptococcales-Tissierellales;g__Finegoldia88Higher in vaginal deliveryDeliveryMethodCB––DeliveryMethodGLd__Bacteria;p__Firmicutes;c__Bacilli;o__Mycoplasmatales; f__Mycoplasmataceae;g__Ureaplasma49Higher in vaginal deliveryDeliveryMethodM–––DeliveryMethodVD–––EpiduralaAFN/AN/AN/AEpiduralCB–––EpiduralGLd__Bacteria;p__Firmicutes;c__Bacilli;o__Mycoplasmatales; f__Mycoplasmataceae;g__Ureaplasma64Higher with epidural useEpiduralM–––EpiduralVD–––PretermBirth37AF–––PretermBirth37CBUnassigned;__;__;__;__;__b27Lower in preterm birthPretermBirth37CBd__Bacteria;__;__;__;__;__c25Lower in preterm birthPretermBirth37GL–––PretermBirth37M–––PretermBirth37VD–––HasGDMAF–––HasGDMCBd__Bacteria;p__Actinobacteriota; c__Actinobacteria;o__Frankiales; f__Nakamurellaceae; g__Nakamurella37Higher with GDMHasGDMGL–––HasGDMM–––HasGDMVD–––InducedLaborAF–––InducedLaborCB–––InducedLaborGL–––InducedLaborM–––InducedLaborVD–––IVFETAF–––IVFETCB–––IVFETGL–––IVFETM–––IVFETVD–––HypertensionAFd__Bacteria;p__Proteobacteria;c__Gammaproteobacteria;o__Enterobacterales; f__Enterobacteriaceae;g__Escherichia-Shigella27Higher with chronic hypertensionHypertensionCBd__Bacteria;p__Actinobacteriota;c__Actinobacteria;o__Actinomycetales; f__Actinomycetaceae;g__Actinomyces87Higher with preeclampsiaHypertensionGLd__Bacteria;p__Actinobacteriota;c__Actinobacteria;o__Bifidobacteriales; f__Bifidobacteriaceae;g__Bifidobacterium38Higher with preeclampsiaHypertensionGLd__Bacteria;p__Bacteroidota;c__Bacteroidia;o__Bacteroidales; f__Porphyromonadaceae;g__Porphyromonas29Higher with preeclampsiaHypertensionMd__Bacteria;p__Bacteroidota;c__Bacteroidia;o__Chitinophagales; f__Chitinophagaceae;g__Vibrionimonas35Higher with chronic hypertensionHypertensionVDd__Bacteria;p__Campilobacterota;c__Campylobacteria;o__Campylobacterales; f__Campylobacteraceae;g__Campylobacter45Higher with chronic hypertensionHypertensionVDd__Bacteria;p__Firmicutes;c__Clostridia;o__Lachnospirales; f__Lachnospiraceae;g__[Ruminococcus]_torques_group35Higher with chronic hypertensionHypertensionVDd__Bacteria;p__Actinobacteriota;c__Coriobacteriia;o__Coriobacteriales; f__Coriobacteriaceae;g__Collinsella34Higher with chronic hypertensionHypertensionVDd__Bacteria;p__Firmicutes;c__Clostridia;o__Peptostreptococcales-Tissierellales; f__Peptostreptococcales-Tissierellales;g__Fenollaria34Higher with chronic hypertensionAntibioticsUseAF–––AntibioticsUseCB–––AntibioticsUseGLd__Bacteria;p__Bacteroidota;c__Bacteroidia;o__Bacteroidales; f__Tannerellaceae;g__Parabacteroides40Higher with antibiotics useAntibioticsUseGLd__Bacteria;p__Proteobacteria;c__Gammaproteobacteria;o__Burkholderiales; f__Burkholderiaceae;g__Cupriavidus39Higher with antibiotics useAntibioticsUseM–––AntibioticsUseaVDN/AN/AN/ABabySexAF–––BabySexCB–––BabySexGL–––BabySexM–––BabySexVD–––AF, amniotic fluid; CB, umbilical cord blood; GL, gastric liquid; M, meconium; VD, cervicovaginal discharge.aSignificant hits were found by ANCOM, but these results were discarded as they have a very low W score (zero in many cases) and are likely artifacts; note that this is a known bug in ANCOM, typically caused by small sample size for a given test.bAmplicon sequence variants were labelled ‘Unassigned’ if it was not possible to classify them at the highest taxonomic level at the required confidence level.cThese amplicon sequence variants could not be classified beyond the domain level at the required confidence level.
## The resemblance of twin microbiome in delivery
To test the hypothesis that samples from twins, both monochorionic and dichorionic, have higher similarity in microbiome composition than randomly chosen samples, we compared the mean of weighted UniFrac distance between twin samples and randomly selected samples. More specifically, for each of the AF, CB, GL, and M groups, we performed bootstrapping hypothesis testing by randomly sampling pairwise distances with replacement from all samples 1000 times to build a $95\%$ confidence interval with the means of the sampled distances. We rejected the null hypothesis that there was no difference between the twin samples and randomly selected samples for all four sample types because the mean pairwise distance for twin samples was below the confidence interval (Fig. 4 and Supplementary Fig. S7). Next, we divided the twins into monochorionic and dichorionic twins and repeated hypothesis testing. We found that we could still reject the null hypothesis for all four sample types for dichorionic twins. For monochorionic twins, however, only the CB and M groups passed the test. Figure 4Higher similarity of microbiome composition in twin samples than in randomly chosen samples. For each sample type, the means of weighted UniFrac distances are shown for the twin samples. A $95\%$ confidence interval was constructed by randomly sampling pairwise distances with replacement from the samples for 1000 times.
## Characterization of the vaginal health-related microbiome
Several pathogenic and commensal vaginal microbiota have been shown to have important consequences for a woman’s reproductive and general health. To establish reference ranges of vaginal microbiota with known clinical associations in generally healthy pregnant women, we searched for bacterial targets commonly tested for assessing vaginal health within VD samples. More specifically, we focused on 31 bacterial targets (15 genera and 16 species) that are tested by the “SmartJane” assay from uBiome Inc., including Lactobacillus, Sneathia, and Gardnerella21. Of the 31 bacterial taxa of clinical importance, 12 were identified in our samples (Fig. 5).Figure 5Relative abundance of bacteria associated with vaginal health. Only bacterial targets in uBiome’s SmartJane assay that are also present in the vaginal discharge samples are shown.
We observed a higher relative abundance of Lactobacillus at the genus level but lower abundances of Aerococcus, Fusobacterium, Gardnerella, Peptoniphilus, Porphyromonas, and Prevotella. Most of our patients did not have any severe pregnancy-related complications. In addition, the majority of preterm birth ranged in the late preterm period from 34 + 0 weeks to 36 + 6 weeks. Therefore, the “SmartJane” assay did not capture almost any pathogenic microbiome. The specification level was examined and is listed in Fig. 5. We found Lactobacillus iners and *Lactobacillus jensenii* from the assay lists, but *Lactobacillus crispatus* was not commonly found in the vaginal microbiome. This could be simply because the SILVA reference database we used omitted Lactobacillus crispatus. We confirmed that some of the ASVs from the *Lactobacillus genus* were indeed *Lactobacillus crispatus* using the National Center for Biotechnology Information (NCBI) database (data not shown).
## Controversies surrounding in utero colonization
Since contamination is a critical issue in microbiome research, we used several up-to-date methods to confirm the presence of bacteria and found evidence of in utero colonization. The distinct bacterial composition in meconium, the first stool passed by newborns, supports that microbial colonization occurs in the intrauterine environment during normal pregnancy22,23. The microbiome in neonatal gastric liquid was similar to that in maternal amnionic fluid, as expected since fetuses swallow amnionic fluid in utero and their urine returns to the fluid under normal physiological conditions. However, the microbiome in gastric liquid was not exactly the same as in amnionic fluid, indicating the existence of unknown mechanisms for flora formation in the fetal oral cavity or proximal gastrointestinal tract, such as esophagus, from the intrauterine environment.
## Do different samples from mothers and newborns share the same microbiome?
The study aimed to determine whether samples from various body sites of pregnant women and their infants would have similar microbiomes, or if the maternal microbiome would be passed on to her fetus. Our results suggest that the microbiome primarily differed based on the body compartment where it was obtained, not the mother-fetus pair. That is, out of all factors, including various obstetric conditions, the sampling site was the most significant factor in determining microbiome similarity.
## Establishing a representative microbiome library of various samples to understand the microbiomes of typical pregnant women, fetuses, and neonates
This study’s key strength is its study population, which comprised of pure Asians and reflected general, low-risk pregnancies. The maternal age range was between 20 and 45 years, which is considered typical for reproductive age. There were roughly equal numbers of nulliparous women, caesarean sections, and male and female neonates. Other than a small number of instances of fetal distress, such as low Apgar scores and meconium staining, newborns with extremely pathological conditions that could alter the microbiome, such as severe preterm birth and treatment in a neonatal intensive care unit (NICU), were excluded. As a result, the microbiome analyzed in this study population is likely to represent typical pregnancy. It is crucial to establish a microbiome library for low-risk pregnant women and their normal neonates as a basis for comparison with pathological conditions, to better understand the microbiome composition during pregnancy.
## Association between microbiome and various pregnancy-related phenotypes
To identify the microbiomes associated with pregnancy-related conditions, such as delivery method, we conducted statistical analysis of differential abundance. Despite the challenges posed by the low microbial biomass and difficulties in controlling study subjects, which can result in false positive results, the bacteria listed in Table 3 seem to align with previous findings. For example, Finegoldia and Bifidobacterium have been previously linked to a healthier pregnancy, and our data confirms this association24,25. Other taxa listed in the table also have links to inflammation and pregnancy complications, such as gestational diabetes mellitus, preeclampsia, and preterm birth. The presence of *Campylobacter and* Lachnospiraceae in vaginal discharge, for example, is in line with previous research showing that these bacterial infections can lead to inflammation and preterm birth26,27.
By cross-referencing with clinical databases, our analysis revealed several significant associations. First, the abundant presence of Lactobacillus and Gardnerella in vaginal discharge is a well-known indicator of the pregnancy microbiome. Lactobacillus plays a protective role in the maternal microbiome during pregnancy, while *Gardnerella is* considered a pathogen and is strongly associated with preterm birth or pregnancy complications11,13,26. The presence of Faecalibacterium in cord blood is noteworthy, as it has been shown to be depleted in gestational diabetes mellitus28, even though the number of cases in our study population was relatively small. Additionally, *Staphylococcus was* found to be strongly associated with preterm birth in meconium. This result coincides with previous findings that suggest *Staphylococcus infections* can lead to preterm birth29,30.
Regarding the effect of antibiotics, we analyzed the relationship between antibiotic use and meconium samples, but the results showed limited association due to the small sample size. As Tormo-Badia et al. reported, antibiotics can alter the gut microbiome of offspring in pregnant mice31. Given that the existence of a “healthy microbiome” during pregnancy is considered crucial for maintaining a normal pregnancy, it is easy to imagine the potential negative consequences of antibiotics administration during pregnancy. Since antibiotics are only given to pregnant women who have signs of infection or inflammation, specific diseases, or preterm premature rupture of membranes with the risk of ascending infection to the fetus, it is practically challenging to determine the effect of antibiotics on the modification of the birth-related microbiome.
The meconium samples showed the presence of microbiome taxa such as Lactobacillus, Staphylococcus, and Ureaplasma, which are collectively known as the vaginal flora11,32. We attempted to evaluate the relationship between delivery mode and the microbiome in meconium, but we did not find any statistically significant differences in composition or diversity. According to a study by Dominguez-Bello et al., there are differences in the bacterial communities in the guts of infants depending on the mode of delivery33. Neonates born vaginally have a microbiome resembling their mother's vaginal microbiota, dominated by Lactobacillus. Conversely, infants born via cesarean section have a microbiome dominated by Staphylococcus, Corynebacterium, and Propionibacterium, which are commonly found on their mothers’ skin surfaces.
## Twin pregnancy and microbiome
Approximately a quarter of the pregnancies in our study were twin pregnancies ($\frac{37}{141}$). To the best of our knowledge, this is the first study to assess the microbiomes of twin newborns. Generally, our twin samples (AF, CB, GL, and M) showed a more similar composition compared to randomly selected samples, even for dichorionic twins who have separate intrauterine compartments. The only exception was CB and M samples from monochorionic twins, where randomly selected samples showed greater similarity, which is likely due to the small sample size of monochorionic twins.
## Conclusion
Exploring the microbiologic features related to pregnancy has been a challenging and controversial task for many years. Microbial invasion of the gestational cavity such as amniotic fluid or placenta can lead to serious obstetric complications such as preterm birth and severe neonatal morbidities that may persist throughout life. Despite the importance of research on the microbiome in pregnancy, progress has been limited due to ethical and accessibility issues. We have collected various samples from pregnant women and their neonates using a standardized protocol and established a microbiome database, which can serve as a reference library for studying samples with other pregnancy-related or pathologic conditions.
## Study design and sample collection
A prospective study was performed on live births delivered between March 2020 and January 2021. Samples were collected from women who had delivered at Seoul National University Bundang Hospital and their newborns. Women with unstable vital signs or those requiring urgent management such as transfusion and neonates admitted to the NICU or who had unstable vital signs after birth were excluded from the study. Samples for microbiome analysis included maternal VD, AF, CB, neonatal GL, and M. As a pregnant woman was hospitalized with expectancy of delivery, the VD sample was obtained using a polyester swab inserted into the posterior fornix of the vagina, assisted by sterile speculum examination. For those who had undergone cesarean section for delivery or amniocentesis for specific indications (i.e., for detection of intraamniotic inflammation/infection), approximately 10 cc of AF was obtained through a syringe for the study. During delivery, both cesarean section and vaginal delivery, approximately 20 cc of CB was taken through a syringe from the vein of the umbilical cord immediately after clamping. The syringe needle was directly inserted into the umbilical cord at the delivery site surrounded by sterile drapes to minimize surgical field contamination. Since removing amniotic fluid or other liquid from the newborn’s mouth and stomach after birth is a part of initial management to help the airway and to stimulate spontaneous breathing, most neonates received suctioning procedures, and the liquid collected in the suction bottle (approximately 15 ml) was carried into a conical tube for analysis of GL. The M sample, the newborn's very early stool, was carefully obtained within 24 h after birth using a polyester swab inserted into the anus as the neonate stabilized after initial management. We tried to collect all five different samples from each woman and neonate(s), nonetheless, a small part of samples from mother-neonate pairs were not obtained or missed for clinical circumstances. The primary outcome was the distribution and composition of the microbiome of the above samples from pregnant women and their neonates. To determine the association between the microbiome from different compartments and obstetric factors, medical records were collected and thoroughly reviewed. Data included maternal age, gestational age at delivery, delivery mode (vaginal delivery or cesarean section), the use of ART, other obstetric complications, and neonatal outcomes such as sex and birth weight.
## Ethics approval and consent to participate
This study was performed with the informed consent of appropriate participants in compliance with the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the Seoul National University Bundang Hospital (B-$\frac{1606}{350}$-003).
## Microbial DNA isolation
Microbial deoxyribonucleic acid (DNA) was extracted from the VD, GL, AF, and CB samples with the ZymoBIOMICS DNA Miniprep Kit (Zymo Research, Irvine, CA) and the sample M using the DNeasy PowerSoil Pro Kit (Qiagen, Germantown, MD) according to the manufacturer’s instructions. Briefly, samples were enzymatically and mechanically lysed by bead beating, followed by washing and filtering in the provided column. Extracted DNA concentrations were measured using a Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). The total amounts of extracted DNA were varied based on sample types, such as 1–10 ug for VD, 3 μg for CB, 30–200 ng for M, and 50 ng for GL and AF. For each box of the DNA extraction kit used, no material was used as a negative control. The blanks were processed in the entire protocol and analyzed.
## 16S rRNA gene amplification
The 16S ribosomal ribonucleic acid (rRNA) gene was amplified using the two-step polymerase chain reaction (PCR) protocol in the 16S Metagenomic Sequencing Library Preparation (Illumina, San Diego, CA). In the first PCR step, the V3–V4 hypervariable region of the 16S rRNA gene was amplified using 10 ng of each sample, 10 µM of 341F/785R primers, and Herculase II fusion DNA polymerase (Agilent, Santa Clara, CA). In the below primer sequence, ‘N’ base is selected from any random base, ‘W’ base is A or T, ‘H’ base is A, C or T, and ‘V’ base is A, C, or G.
341F: 5′- TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′ 785R: 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC-3′ PCR cycling was performed with an initial cycle at 95 °C for 3 min, followed by 25 cycles of 95 °C for 30 s, 55 °C for 30 s, 72 °C for 30 s, and a final extension cycle at 72 °C for 5 min. The amplicons were cleaned with AMPure XP beads (Beckman Coulter, Brea, CA, USA). In the second PCR, index primers from the Nextera DNA CD Index Kit (Illumina, San Diego, CA) were added to the ends of the amplicons generated in the first PCR. PCR cycling was performed with an initial cycle at 95 °C for 3 min, followed by ten cycles of 95 °C for 30 s, 55 °C for 30 s, 72 °C for 30 s, and a final extension cycle at 72 °C for 5 min. Each sample was cleaned with AMPure XP beads (Beckman Coulter, Brea, CA, USA) and eluted in UltraPure DNase/RNase-Free Water (Thermo Fisher Scientific, Waltham, MA). The amplified DNA was checked using a 2100 Bioanalyzer system using an Agilent DNA 1000 Kit (Agilent, Santa Clara, CA, USA). For each library production, no template was used as a negative control.
## 16S rRNA gene sequencing and analysis
Based on the DNA size and concentration, the amplicons were pooled in equimolar amounts and spiked with $30\%$ PhiX (Illumina, San Diego, CA). These were then sequenced on the Illumina MiSeq platform using paired-end 250 cycle MiSeq Reagent Kit V2 (Illumina, San Diego, CA) and a 300 cycle MiSeq Reagent Kit V3 (Illumina, San Diego, CA). Negative controls from the DNA extraction and library were sequenced.
## Sequencing data generation
We divided the samples into nine batches (Runs 1–9) and sequenced the V3-V4 region of the 16S rRNA gene using Illumina MiSeq machines with a target depth of 100,000 per sample (Supplementary Fig. S8). Sequencing was performed with 250 bp paired-end reads for all of the sequencing runs except for the last one (Run 9), where sequencing was performed with 300 bp paired-end reads for practical reasons. The read quality scores for each sequencing run are shown in Supplementary Fig. S9. The bcf2fastq program of Illumina was used to demultiplex raw sequencing data (BCL files) and output forward and reverse FASTQ files for each sample. Of note, some samples were sequenced more than once to assess the impact of batch effects. These included “sequencing duplicates” in which the identical NGS library of one sample was sequenced in separate runs and “library duplicates” in which multiple NGS libraries were prepared from the identical sample at different dates and then sequenced separately.
## Data analysis and visualization
Unless stated otherwise, all analyses were carried out using the QIIME 2 platform, a powerful community-developed platform for microbiome bioinformatics34. For each sequencing run, FASTQ files were imported to QIIME 2 and the DADA2 plugin35 to identify ASVs by trimming low-quality parts of sequence reads, denoising trimmed reads, and then merging the forward and reverse reads (Supplementary Fig. S8). The observed ASVs from individual sequencing runs were then merged into one ASV table. To detect and remove potential contaminants, we ran the decontam program on our samples, which looked for ASVs per sequencing batch that appeared at higher frequencies in low-concentration samples and were repeatedly found in the negative control36. Taxonomy classification was performed using a naive Bayes classifier using the SILVA database37. To visualize the outputs from QIIME 2, we developed the Dokdo program (https://github.com/sbslee/dokdo), an open-source and MIT-licensed Python package for microbiome sequencing analysis using QIIME 2. Dokdo internally uses the application programming interface of QIIME 2 and therefore does not require any other dependencies. Dokdo can be used to perform a variety of secondary analyses or create publication-quality figures from QIIME 2 files/objects (e.g. a taxonomic bar plot or an alpha rarefaction plot).
## Diversity analysis
We used the QIIME 2 command “qiime diversity core-metrics-phylogenetic” to compute the alpha and beta diversity metrics of our samples. When running the command, to normalize for the difference in read depth across the samples, we used the “-p-sampling-depth” option to rarefy our samples to 5,000 sequence reads and have an equal depth of coverage. We also ensured that all samples were sequenced to a sufficient depth of coverage for diversity analysis by creating rarefaction curves (Supplementary Fig. S10). Additionally, we used the “-i-phylogeny” option to provide a rooted phylogenetic tree of observed ASVs, which is required for performing PCoA based on the weighted UniFrac distance38.
## Statistical analysis
To assess the differential abundance of the microbiome in the context of clinical information such as preterm birth, we used the QIIME 2 command “qiime composition ancom” to perform ANCOM, which compares the centered log-ratio (CLR) of relative abundance between two or more groups of samples39. To determine whether groups of samples are significantly different from one another in beta diversity, we carried out PERMANOVA using the QIIME 2 command “qiime diversity adonis” which fits linear model assumptions to a distance matrix (e.g., weighted UniFrac) with the chosen variables. We performed bootstrapping hypothesis testing by building a $95\%$ confidence interval with the “scipy.stats.t.interval” method in the scipy package to compare similarities in microbiome composition between twins and randomly chosen samples40.
## Supplementary Information
Supplementary Figures. The online version contains supplementary material available at 10.1038/s41598-023-31049-1.
## References
1. Mesa MD. **The evolving microbiome from pregnancy to early infancy: A comprehensive review**. *Nutrients* (2020). DOI: 10.3390/nu12010133
2. Knight R. **Best practices for analysing microbiomes**. *Nat. Rev. Microbiol.* (2018) **16** 410-422. DOI: 10.1038/s41579-018-0029-9
3. Fox C, Eichelberger K. **Maternal microbiome and pregnancy outcomes**. *Fertil. Steril.* (2015) **104** 1358-1363. DOI: 10.1016/j.fertnstert.2015.09.037
4. Chen X, Liu S, Tan Q, Shoenfeld Y, Zeng Y. **Microbiome, autoimmunity, allergy, and helminth infection: The importance of the pregnancy period**. *Am. J. Reprod. Immunol.* (2017). DOI: 10.1111/aji.12654
5. Aagaard K. **The placenta harbors a unique microbiome**. *Sci. Transl. Med.* (2014) **6** 237ra265. DOI: 10.1126/scitranslmed.3008599
6. Perez-Munoz ME, Arrieta MC, Ramer-Tait AE, Walter J. **A critical assessment of the "sterile womb" and "in utero colonization" hypotheses: Implications for research on the pioneer infant microbiome**. *Microbiome* (2017) **5** 48. DOI: 10.1186/s40168-017-0268-4
7. Aagaard K. **A metagenomic approach to characterization of the vaginal microbiome signature in pregnancy**. *PLoS ONE* (2012) **7** e36466. DOI: 10.1371/journal.pone.0036466
8. Neuman H, Koren O. **The pregnancy microbiome**. *Nestle Nutr. Inst. Workshop Ser.* (2017) **88** 1-9. DOI: 10.1159/000455207
9. Nuriel-Ohayon M, Neuman H, Koren O. **Microbial changes during pregnancy, birth, and infancy**. *Front. Microbiol.* (2016) **7** 1031. DOI: 10.3389/fmicb.2016.01031
10. Staude B. **The microbiome and preterm birth: A change in paradigm with profound implications for pathophysiologic concepts and novel therapeutic strategies**. *Biomed. Res. Int.* (2018) **2018** 7218187. DOI: 10.1155/2018/7218187
11. DiGiulio DB. **Temporal and spatial variation of the human microbiota during pregnancy**. *Proc. Natl. Acad. Sci. USA* (2015) **112** 11060-11065. DOI: 10.1073/pnas.1502875112
12. Stout MJ. **Early pregnancy vaginal microbiome trends and preterm birth**. *Am. J. Obstet. Gynecol.* (2017) **217** 356.e351-356.e318. DOI: 10.1016/j.ajog.2017.05.030
13. Callahan BJ. **Replication and refinement of a vaginal microbial signature of preterm birth in two racially distinct cohorts of US women**. *Proc. Natl. Acad. Sci. USA* (2017) **114** 9966-9971. DOI: 10.1073/pnas.1705899114
14. Brown RG. **Vaginal dysbiosis increases risk of preterm fetal membrane rupture, neonatal sepsis and is exacerbated by erythromycin**. *BMC Med.* (2018) **16** 9. DOI: 10.1186/s12916-017-0999-x
15. Kuperman AA, Koren O. **Antibiotic use during pregnancy: How bad is it?**. *BMC Med.* (2016) **14** 91. DOI: 10.1186/s12916-016-0636-0
16. Fujiwara N. **Significant increase of oral bacteria in the early pregnancy period in Japanese women**. *J. Investig. Clin. Dent.* (2017). DOI: 10.1111/jicd.12189
17. Collado MC, Rautava S, Aakko J, Isolauri E, Salminen S. **Human gut colonisation may be initiated in utero by distinct microbial communities in the placenta and amniotic fluid**. *Sci. Rep.* (2016) **6** 23129. DOI: 10.1038/srep23129
18. DiGiulio DB. **Prevalence and diversity of microbes in the amniotic fluid, the fetal inflammatory response, and pregnancy outcome in women with preterm pre-labor rupture of membranes**. *Am. J. Reprod. Immunol.* (2010) **64** 38-57. DOI: 10.1111/j.1600-0897.2010.00830.x
19. Akagawa S. **Effect of delivery mode and nutrition on gut microbiota in neonates**. *Ann. Nutr. Metab.* (2019) **74** 132-139. DOI: 10.1159/000496427
20. Moreno I, Simon C. **Relevance of assessing the uterine microbiota in infertility**. *Fertil. Steril.* (2018) **110** 337-343. DOI: 10.1016/j.fertnstert.2018.04.041
21. Bik EM. **A novel sequencing-based vaginal health assay combining self-sampling, HPV detection and genotyping, STI detection, and vaginal microbiome analysis**. *PLoS ONE* (2019) **14** e0215945. DOI: 10.1371/journal.pone.0215945
22. Hansen R. **First-pass meconium samples from healthy term vaginally-delivered neonates: An analysis of the microbiota**. *PLoS ONE* (2015) **10** e0133320. DOI: 10.1371/journal.pone.0133320
23. Martin R. **Early-life events, including mode of delivery and type of feeding, siblings and gender, shape the developing gut microbiota**. *PLoS ONE* (2016) **11** e0158498. DOI: 10.1371/journal.pone.0158498
24. Kumar M. **Vaginal microbiota and cytokine levels predict preterm delivery in Asian women**. *Front. Cell Infect. Microbiol.* (2021) **11** 639665. DOI: 10.3389/fcimb.2021.639665
25. Nuriel-Ohayon M. **Progesterone increases bifidobacterium relative abundance during late pregnancy**. *Cell Rep.* (2019) **27** 730-736.e733. DOI: 10.1016/j.celrep.2019.03.075
26. Serrano MG. **Racioethnic diversity in the dynamics of the vaginal microbiome during pregnancy**. *Nat. Med.* (2019) **25** 1001-1011. DOI: 10.1038/s41591-019-0465-8
27. Smith JL. *J. Food Prot.* (2002) **65** 696-708. DOI: 10.4315/0362-028x-65.4.696
28. Ferrocino I. **Changes in the gut microbiota composition during pregnancy in patients with gestational diabetes mellitus (GDM)**. *Sci. Rep.* (2018) **8** 12216. DOI: 10.1038/s41598-018-30735-9
29. Kriebs JM. **Staphylococcus infections in pregnancy: Maternal and neonatal risks**. *J. Perinat. Neonatal. Nurs.* (2016) **30** 115-123. DOI: 10.1097/JPN.0000000000000165
30. Tang Y, Yu F, Hu Z, Peng L, Jiang Y. **Characterization of aerobic vaginitis in late pregnancy in a Chinese population: A STROBE-compliant study**. *Medicine* (2020) **99** e20732. DOI: 10.1097/MD.0000000000020732
31. Tormo-Badia N. **Antibiotic treatment of pregnant non-obese diabetic mice leads to altered gut microbiota and intestinal immunological changes in the offspring**. *Scand. J. Immunol.* (2014) **80** 250-260. DOI: 10.1111/sji.12205
32. Peretz A. **Mycoplasma and Ureaplasma carriage in pregnant women: The prevalence of transmission from mother to newborn**. *BMC Pregnancy Childbirth* (2020) **20** 456. DOI: 10.1186/s12884-020-03147-9
33. Dominguez-Bello MG. **Delivery mode shapes the acquisition and structure of the initial microbiota across multiple body habitats in newborns**. *Proc. Natl. Acad. Sci. USA* (2010) **107** 11971-11975. DOI: 10.1073/pnas.1002601107
34. Bolyen E. **Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2**. *Nat. Biotechnol.* (2019) **37** 852-857. DOI: 10.1038/s41587-019-0209-9
35. Callahan BJ. **DADA2: High-resolution sample inference from Illumina amplicon data**. *Nat. Methods* (2016) **13** 581-583. DOI: 10.1038/nmeth.3869
36. Davis NM, Proctor DM, Holmes SP, Relman DA, Callahan BJ. **Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data**. *Microbiome* (2018) **6** 226. DOI: 10.1186/s40168-018-0605-2
37. Quast C. **The SILVA ribosomal RNA gene database project: Improved data processing and web-based tools**. *Nucleic Acids Res.* (2013) **41** D590-596. DOI: 10.1093/nar/gks1219
38. Lozupone C, Knight R. **UniFrac: A new phylogenetic method for comparing microbial communities**. *Appl. Environ. Microbiol.* (2005) **71** 8228-8235. DOI: 10.1128/AEM.71.12.8228-8235.2005
39. Mandal S. **Analysis of composition of microbiomes: A novel method for studying microbial composition**. *Microb. Ecol. Health Dis.* (2015) **26** 27663. DOI: 10.3402/mehd.v26.27663
40. Virtanen P. **SciPy 1.0: Fundamental algorithms for scientific computing in Python**. *Nat. Methods* (2020) **17** 261-272. DOI: 10.1038/s41592-019-0686-2
|
---
title: SB2301-mediated perturbation of membrane composition in lipid droplets induces
lipophagy and lipid droplets ubiquitination
authors:
- Jinjoo Jung
- Jongbeom Park
- Mingi Kim
- Jaeyoung Ha
- Hana Cho
- Seung Bum Park
journal: Communications Biology
year: 2023
pmcid: PMC10030462
doi: 10.1038/s42003-023-04682-9
license: CC BY 4.0
---
# SB2301-mediated perturbation of membrane composition in lipid droplets induces lipophagy and lipid droplets ubiquitination
## Abstract
Lipid droplets (LDs) are involved in various biological events in cells along with their primary role as a storage center for neutral lipids. Excessive accumulation of LDs is highly correlated with various diseases, including metabolic diseases. Therefore, a basic understanding of the molecular mechanism of LD degradation would be beneficial in both academic and industrial research. Lipophagy, a selective autophagy mechanism/LD degradation process, has gained increased attention in the research community. Herein, we sought to elucidate a novel lipophagy mechanism by utilizing the LD-degrading small molecule, SB2301, which activates ubiquitin-mediated lipophagy. Using a label-free target identification method, we revealed that ethanolamine-phosphate cytidylyltransferase 2 (PCYT2) is a potential target protein of SB2301. We also demonstrated that although SB2301 does not modulate PCYT2 function, it induces the cellular translocation of PCYT2 to the LD surface and spatially increases the phosphatidylethanolamine (PE)/phosphatidylcholine (PC) ratio of the LD membrane, causing LD coalescence, leading to the activation of lipophagy process to maintain energy homeostasis.
The small molecule SB2301 activates lipophagy and alters lipid droplet membrane composition, with the ethanolamine-phosphate cytidylyltransferase 2 (PCYT2) being a potential target of the compound.
## Introduction
Lipid droplets (LDs) are specialized organelles that store cellular free fatty acids (FFAs) as neutral lipids, such as triacylglycerol (TG) or sterol esters (SE), to avoid FFA lipotoxicity1,2. In LDs, neutral lipids are surrounded by phospholipids such as phosphatidylcholine (PC) and phosphatidylethanolamine (PE)3, and surface proteins such as the perilipin family4 and lipases5. *Cells* generate or degrade LDs dynamically in response to environmental changes to maintain energy homeostasis and regulate lipid metabolism6.
Lipases on the LD surface mainly degrade neutral lipids when cells require LD degradation7,8. Singh et al. reported that LDs could be a substrate of selective autophagy, called lipophagy, that sequestrates LD within autophagosomes under starvation conditions9. The biomedical research community has been working on the lipophagy mechanism to reveal various roles of LDs beyond the storage of neutral lipids10, such as cellular stress modulation11,12, protein functional regulation13,14, and storage of other biomolecules15,16. Therefore, studying the LD regulatory mechanism can provide valuable information to define the underlying biological pathways owing to its relevance in chemical biology and drug discovery fields. Furthermore, the physiological relevance of LDs in metabolic diseases has recently been emphasized. The LD accumulation in the muscle and liver is a hallmark of metabolic diseases, including steatosis17,18, type 2 diabetes19,20, and atherosclerosis21,22. Thus, lipophagy has emerged as a new strategy for treating such diseases, especially steatosis and liver diseases23–25.
In addition, the phenotype-based approach has been a major strategy for discovering new molecular entities with novel modes of action. In this study, we performed image-based high-thoughput screening to monitor the number and size of cellular LDs as crucial phenotypes in live cells. We further assessed a new small molecule, SB2301, to explore ubiquitin-mediated lipophagy as a new regulatory mechanism of LDs. Using SB2301, we revealed a new lipophagy-activating mechanism that induces LD degradation by spatially altering the lipid composition of the LD membranes.
## Discovery of LD-reducing small molecules
First, we performed image-based phenotypic screening of cellular LDs in live cells to identify potential modulators of cellular LDs. Previously, we reported the fluorogenic probe, SF44, which has a hydrophobic LD-selective turn-on property in live cells26. We also demonstrated the application of SF44 to fluorescence imaging in a high-throughput manner with an excellent Z′ factor27. The SF44-based LD monitoring system utilizing SF44 allows real-time observation of the cellular LD dynamics without the need for washing steps. Thus, we applied this high-content LD-monitoring system in live cells to identify new chemical entities lacking cellular toxicity with minimal influence from other external factors.
We previously reported a series of compounds that reduced cellular LDs28. These compounds were derived from a privileged substructure-based diversity-oriented synthesis (pDOS) library designed with a central isoxazole moiety and its substitution at the C3 and C5 positions with privileged structures, such as indole, benzopyran, quinoline, and pyrimidine (Supplementary Fig. 1). Among them, only 3-(quinolin-6-yl)phenol-substituted isoxazole derivatives showed excellent efficacy for cellular LD reduction. However, these compounds showed some degree of cytotoxicity. To address this issue, we bioisosterically replaced isoxazole with 1,2,3-triazole as a structural modification to avoid cytotoxicity while improving the desired LD-reducing activities29. With 3-(quinolin-6-yl) phenol-substituted isoxazoles as a starting point, we constructed a 1,4-disubstituted 1,2,3-triazoles library containing quinoline-privileged structures for structure-activity relationship (SAR) studies (Supplementary Note 1). A series of 17 analogues were designed and synthesized based on 3-(quinolin-6-yl)phenol and 1,2,3-triazole scaffolds (Supplementary Note 2). As shown in Supplementary Scheme 1, alkynyl quinoline was subjected to a copper(I)-catalyzed click reaction with various aromatic azides in the presence of CuSO4·H2O and sodium ascorbate.
The LD reduction activity of these analogues (1–20) was initially evaluated in human cervical cancer HeLa cells at a fixed concentration (10 μM each) to determine the efficacy of LD reduction. In addition, potency was determined by measuring the half-maximal inhibitory concentration (IC50) (Supplementary Table 1). Compounds with high cytotoxicity and low water solubility were excluded during high-content phenotypic screening. Based on our SAR study, we confirmed that 1-(2-trifluoromethyl phenyl)-4-(3-hydroxyphenyl quinolone)triazole [4] displayed the most potent LD-reducing activity among our derivatives and was named SB2301 (Fig. 1A). We observed a dose-dependent LD reduction activity in HeLa cells with an IC50 of 4.4 μM at 24 h after treatment with SB2301 (Fig. 1B, C). Even though SB2301 showed some cytotoxicity at high concentrations (Fig. 1D), we used SB2301 at the concentrations where this is not an issue offering a window of opportunity for the subsequent biological studies. We also observed a similar pattern of LD reduction in human hepatocellular carcinoma HepG2 cells (Supplementary Fig. 2A) and mouse hepatocyte AML12 cells (Supplementary Fig. 2B). Consistent with the reduction in cellular LDs, the cellular triglyceride (TG) content was reduced in a dose-dependent manner (Fig. 1E). Based on these findings, we conducted a further study to determine the mechanism by which SB2301 reduced cellular LDs. Fig. 1SB2301 reduces cellular LDs without inducing cytotoxicity. A Chemical structure of SB2301 [4]. B Representative LD fluorescence images captured during phenotypic screening. C The quantification of cellular LD count in HeLa cells upon treatment with various SB2301 doses for 24 h. All data are shown as the mean ± standard deviation (SD). D Dose-response curves of cell viability upon SB2301 treatment for the indicated times in HeLa cells. All data are shown as the mean ± SD. E TG quantification results upon SB2301 treatment for 24 h in HepG2 cells. All data are represented as dot plots with the mean ± SD, ($$n = 3$$). Data were analyzed using an unpaired t-test. ** $$P \leq 0.0068$$ vs. dimethyl sulfoxide (DMSO), ##$$P \leq 0.0029$$ vs. DMSO.
## LD degradation by lipophagy activation
Cellular LD can be reduced either by preventing their formation (anabolic pathway) or promoting their degradation (catabolic pathway). We did not observe any drastic changes in the expression of lipid biosynthesis-related genes upon SB2301 treatment by qPCR analysis (Supplementary Fig. 3A) and proteins regulating TG/CE synthesis by western blot analysis (Supplementary Fig. 3B). Additionally, SB2301-mediated LD reduction was not inhibited by co-treatment with Orlistat, a common lipase inhibitor (Supplementary Fig. 3C). Instead, we confirmed that SB2301 treatment activated LD degradation via lipophagy, a selective autophagic process. During the progression of autophagy, the microtubule-associated protein 1 A/1B-light chain 3 (LC3) I is lipidated to become LC3 II on the surface of the phagophore. According to western blot analysis, LC3 I to II conversion occurred upon SB2301 treatment in both dose- and time-dependent manners (Fig. 2A, B). When we reduced autophagy related gene 5 (ATG5), one of the key proteins for the phagophore elongation of autophagy, SB2301-mediated conversion of LC3 I to II was completely abolished (Fig. 2C). We then examined SB2301-mediated autophagic flux upon treatment with bafilomycin A1 (Baf), a known modulator of vacuolar-type H+-ATPase, inhibiting the autolysosomal degradation of autophagic cargo by preventing its acidification, thereby blocking the late-stage autophagic flux. We observed increased LC3 II accumulation upon co-treatment with SB2301 and Baf in a dose-dependent manner (Fig. 2D). We also performed live-cell imaging with the mCherry-GFP-LC3 protein (Supplementary Fig. 4)30; Owing to the intrinsic quenching property of GFP in acidic conditions, we could determine whether mCherry-GFP-fused LC3 proteins are located in either autophagosomes (neutral) or autolysosomes (acidic) by simply monitoring changes in the GFP signal while tracking the LC3 protein with a pH-independent mCherry signal. Rapamycin (Rap) activates autophagy by inhibiting the mammalian target of rapamycin complex 1 (mTORC1) signaling pathway and inducing the formation of red puncta due to GFP signal quenching in acidic autolysosomes. In contrast, Baf treatment arrestes the late-stage autophagic flux by blocking lysosomal acidification, which was confirmed by the formation of yellow puncta via live-cell imaging with mCherry-GFP-LC3. Based on our control experiments with the autophagy inducer (Rap) and inhibitor (Baf), SB2301 phenocopied the autophagy inducer, confirming that SB2301 activates autophagy. Fig. 2SB2301 reduces cellular LDs via lipophagy. A Western blot analysis with various doses of SB2301, rapamycin (Rap, 2 μM), and bafilomycin A1 (Baf, 20 nM) treated in HepG2 cells for 12 h. Western blot analysis with 10 μM of SB2301 for indicated times (B), with SB2301 treatment under the autophagy related gene 5 (Atg5)-knockdown condition, followed by SB2301 treatment for 9 h (C), and with SB2301 treatment in the absence and presence of 5 nM of Baf for 6 h (D). E Representative live-cell fluorescence images of LC3 (red, mCherry) and LDs (green, SF44) on mCherry-LC3 transfected HeLa cells upon treatment with rapamycin (1 μM), chloroquine (10 μM), and SB2301 (5 μM) for 12 h. F Representative live-cell fluorescence images of lysosomes (red, lysotracker) and LDs (green, SF44) upon treatment with rapamycin (1 μM) and SB2301 (10 μM) on HeLa cells for 12 h. G Representative immunofluorescence images of ubiquitin (red) and LDs (green, BODIPY $\frac{493}{503}$) on HepG2 cells upon SB2301 treatment (10 μM) for 12 h. After cells were fixed, ubiquitin was labeled with anti-ubiquitin antibody and LD was stained with BODIPY $\frac{495}{503.}$ H Fluorescence intensity quantification of images captured in G. Images were selected randomly from biological triplicates ($$n = 21$$ from DMSO, $$n = 32$$ from SB2301). Cytoplasmic ubiquitin signals were subtracted as a background, and the fluorescent signal ratio (ubiquitin/LD) on each pixel was calculated. All data are represented as dot plots with the mean ± SD. Data were analyzed using an unpaired t-test. *** $$P \leq 0.0001.$$
We then monitored the cellular location of LD, LC3 (autophagosome), and lysosomes upon SB2301 treatment to determine whether LDs were used as autophagy substrates. As shown in Fig. 2E and Supplementary Fig. 5A, Raf (an autophagy activator, not a lipophagy activator) did not induce LDs to be surrounded by LC3 as LC3-mediated autophagosome substrate selection. Meanwhile, with chloroquine (a lysosomal degradation inhibitor) treatment, LDs were trapped in the autophagosome. SB2301 induced autophagosome formation via autophagy activation, which was consistent with the enhanced fluorescent signal of LC3. Furthermore, SB2301 treatment increased the co-localization of LD with LC3 proteins more than dimethyl sulfoxide (DMSO) control along with a significant reduction in LD fluorescence signals. Collectively, our results indicate that SB2301 induced the sequestration of whole or partial LDs in autophagosomes, which eventually fuse with lysosomes to hydrolyze LDs (Fig. 2F and Supplementary Fig. 5B). To ascertain whether SB2301 induces autophagy in organelles other than cellular LDs, we measured mitochondrial DNA content upon SB2301 treatment and confirmed that mitochondrial contents were unaffected by this treatment (Supplementary Fig. 6), indicating that SB2301 selectively reduced cellular LDs by inducing lipophagy.
Ubiquitination is actively involved in recognizing selective autophagy cargos31, such as the mitochondria32,33, protein aggregates34, peroxisomes35, and pathogens36. However, ubiquitin-mediated lipophagy has not yet been reported. To determine whether SB2301-mediated LD reduction via lipophagy is regulated by a ubiquitin-mediated mechanism, we performed immunofluorescence imaging against ubiquitin in HepG2 cells. As shown in Fig. 2G, Supplementary Figs. 7, 8, substantial ubiquitination occurred on the LD surface upon SB2301 treatment compared to that with DMSO treatment. In addition, when we quantified the ratio of ubiquitin to the LD signal after background subtraction with cytoplasmic ubiquitin signals, the ratio was significantly higher under SB2301-treated condition (Fig. 2H), confirming that LD surface proteins were more selectively ubiquitinated than cytoplasmic proteins upon SB2301 treatment.
## Target protein identification
To investigate the underlying mechanism of SB2301 on lipophagy, we conducted a target identification study using the thermal stability shift-based fluorescence difference in two-dimensional gel electrophoresis (TS-FITGE). This technique is based on characteristic changes in protein stability induced by thermal denaturation when the protein specifically interacts with its ligand37. Briefly, the DMSO- and SB2301-treated cells were subjected to heat shock at increasing temperatures, and the resulting lysates were chemically conjugated with Cy3- and Cy5-N-hydroxysuccinimide esters, respectively. Thereafter, two proteomes (fluorescently labeled with Cy3 and Cy5) from each temperature condition were combined and analyzed by 2D gel electrophoresis. Finally, fluorescence image-based analysis of 2D gel revealed color changes of individual protein spots, which were caused by changes in the thermal stability of a protein upon engagement with SB2301.
As shown in Fig. 3A and Supplementary Fig. 9, approximately nine red or green spots repeatedly appeared in the 2D-gel images of samples treated at higher temperatures, indicating enhanced (red) or decreased (green) protein thermal stability upon SB2301 engagement. Fourteen protein candidates were identified by mass spectrometry (Supplementary Table 2). A cellular thermal shift assay (CETSA) was performed on several candidate proteins to validate the target ID experiments (Fig. 3B and Supplementary Fig. 10). To prioritize the target protein candidates, we conducted a literature survey on their functions and selected three proteins, PCYT2 (ethanolamine-phosphate cytidylyltransferase 2), ACSL4 (long-chain-fatty-acid-CoA ligase 4), and IDH1 (isocitrate dehydrogenase [NADP] cytoplasmic), which are closely related to cellular lipid metabolism. We then verified whether each target protein could influence the quantity or size of cellular LDs with loss-of-function study. As shown in Fig. 3C and Supplementary Fig. 11, we observed that the cellular level of PCYT2 was inversely correlated with cellular LD counts, suggesting that PCYT2 might influence the lipophagy activation mechanism of SB2301.Fig. 3Identification of PCYT2 as a target protein of SB2301 by TS-FITGE.A Overlaid images of the Cy3 (green, DMSO-treated proteome) and Cy5 channel (red, SB2301-treated proteome). Indicated spots (white arrow) turned red at 59.1 °C then disappeared at 63.1 °C. B Immunoblot from CETSA representing the specific binding of SB2301 with PCYT2. 20 μM of SB2301 and negative compound [10] were treated on HepG2 cells for 1 h. C LD count changes upon depletion of Pcyt2, Acsl4, and Idh1 using the respective siRNAs. HepG2 cells were transfected with siRNAs for 48 h, and then the fluorescent LD images were obtained. All data are represented as dot plots with the mean ± SD, ($$n = 3$$). Data were analyzed using an unpaired t-test. * $$P \leq 0.0123$$ vs. si-scr. D Representative sensorgrams of surface plasmon resonance (SPR) analysis of SB2301 binding to human PCYT2. The equilibrium dissociation constant (KD) was calculated after 1:1 fitting with kinetic (upper) or affinity mode (lower).
We then conducted a biophysical experiment to examine the specific binding of SB2301 to PCYT2 using surface plasmon resonance (SPR) analysis and confirmed the direct engagement of SB2301 with PCYT2 in a 1:1 binding mode (Fig. 3D). Since the estimated equilibrium dissociation constant (Kd) value of SB2301 for PCYT2 was ~26 μM, PCYT2 may not be the sole protein target of SB2301. However, its synthetic analogues (3 and 10), which lack LD-reducing activity, showed no thermal stability shift with PCYT2 in CETSA (Supplementary Fig. 12) and no binding events in the SPR analysis (Supplementary Fig. 13A, B). In contrast, in the case of synthetic analogue 14 (moderate LD-reducing activity with cytotoxicity) or 13 (minimal LD-reducing activity), we observed direct binding with PCYT2 with a similar tendency to their LD-reducing activity (Supplementary Fig. 13C, D). Therefore, we concluded that PCYT2 could be a potential target protein correlated with the SB2301-mediated reduction of cellular LDs.
## Spatial regulation of PCYT2 followed by composition changes in LD membrane
PCTY2 is known to catalyze the synthesis of CDP-ethanolamine, a precursor of PE synthesis in the cytosol38. Choline/ethanol-amine phosphotransferase (CEPT) synthesizes PE using CDP-ethanolamine on the ER membrane, and the PCYT2-mediated step is the rate-determining step of cellular PE synthesis (Supplementary Scheme 2)39. Based on this information, we investigated the changes that might occur when SB2301 binds to PCYT2. Although the enzymatic function of PCYT2 was not affected by SB2301 treatment (Supplementary Fig. 14), we observed translocation of PCYT2 to the LD surface even after a short treatment time (within 1 h) (Fig. 4A, B for SB2301, Supplementary Fig. 15 for DMSO). Furthermore, SB2301 induced translocation of PCYT2 to the LD surface in a time- and dose-dependent manner (Supplementary Fig. 16 and Supplementary Video 1–4). Notably, large-sized LDs began to form within a similar time frame of PCYT2 translocation to the LD surface (Fig. 4A, C, Supplementary Figs. 7A, 17 for SB2301 and Supplementary Fig. 18 for DMSO). Thus, we inferred that LD enlargement might arise with SB2301-induced PCYT2 translocation, which allows for the direct and fast supply of CDP-ethanolamine to the LD surface and the subsequent increase in PE on the LD surface. The main phospholipid components of the LD membrane are PE and PC3. PC adopts a cylindrical shape, whereas PE adopts a conical shape owing to its smaller polar head40. Due to the different biophysical properties and overall shapes of the two phospholipids, the PE/PC ratio in the LD membrane influences the curvature and size of LD to remain in a state of minimum entropy40. Therefore, we hypothesized that SB2301 induces the translocation of PCYT2 at the LD surface, and this spatiotemporal relocation of PCYT2 preferentially increase the levels of PE at the LD surface, which consequently leads to the coalescence of unstable small LDs into large LDs to minimize the surface tension by reducing their curvature (Fig. 4D). We measured the PE/PC ratio in cellular LDs to test this hypothesis. Although we failed to observe any changes in the quantity of PE or PC in the SB2301-treated whole cells (Supplementary Fig. 19), substantial increases in the PE/PC ratio of the isolated LDs were observed (Fig. 4E). Thus, the amount of PE on the LD surface was increased by SB2301-induced PCYT2 translocation, and the resulting PE/PC ratio at the LD surface perturbed its biophysical stability, subsequently leading to LD coalescence. Fig. 4Spatial regulation of PCYT2 followed by dynamic changes in the LD size. A Representative immunofluorescence images of PCYT2 (green) and LD (blue, BODIPY $\frac{493}{503}$). HepG2 cells were treated with 10 μM SB2301 at the indicated times. The cells were fixed and immunostained for PCYT2 detection. B Quantification of PCYT2 on the LD. The superimposed area between LD and PCYT2 on each experimental condition was analyzed using the ImageJ software. Images were selected randomly from biological triplicates. All data are represented as dot plots with the mean ± SD. Data were analyzed using an unpaired t-test. * $$P \leq 0.0423$$, **$$P \leq 0.0038$$, ***$$P \leq 0.0002.$$ C The size distribution of purified LDs was measured by dynamic light scattering. HepG2 cells were treated with 10 μM SB2301 for 9 h prior to the purification of cellular LDs. D Schematic illustration of LD coalescence via modulating their PE/PC ratio. E The PE/PC ratio measurement in purified LDs shows that SB2301 treatment spatially increases the PE/PC ratio. All data are represented as dot plots with the mean ± SD, ($$n = 3$$). Data were analyzed using the paired t-test. ** $$P \leq 0.0077.$$ F The degree of contribution and remaining LDs were analyzed from LD fluorescence images at each time point. The degree of contribution was calculated according to the definition. The number of remaining LDs was obtained by dividing the total LD area by the cell counts. Upon SB2301 treatment, the mode of the blue graph (the SB2301-treated condition) shifted to the larger LD diameter, then returned to its original value. LD fluorescence images were captured from HepG2 cells treated with 20 μM of SB2301 for the indicated times. All data are represented as the mean ± SD. G, H Representative LD fluorescence images and quantification results on Atg5-knockdown HepG2 cells. ATG5-depleted cells were treated with 20 μM of SB2301 was treated for 48 h. The degree of contribution of LDs was analyzed as described in E. All data are represented as the mean ± SD. I, J Representative LD fluorescence images and quantification results from Ubb-knockdown HepG2 cells. UBB-depleted cells were treated with 20 μM of SB2301 for 48 h. The degree of contribution of LDs was then analyzed as described in E. All data are represented as the mean ± SD.
To elucidate the correlation between lipophagy and PE/PC ratio increase at the LD membrane, we systematically analyzed the dynamic changes in the LD size. The coalescence of small LDs during the earlier period of SB2301 treatment explained the decrease in cellular LD counts without fully activating the autophagy process (Supplementary Fig. 2A). The proportion of large LDs gradually increased around 6 h (Fig. 4A and Supplementary Fig. 16). Within the same period, the autophagy pathway was also activated (Fig. 2B). By analyzing the degree of contribution of cellular LDs over time, we observed a dynamic change in cellular LD volumes (Fig. 4F). Upon SB2301 treatment, LD coalescence occurred for up to 18 h; thereafter, the proportion of large LDs no longer increased and started to decrease owing to the continuous LD degradation. Eventually, the LD size distribution returned to its original value, but the total cellular LD volume was reduced to $64.2\%$ at 24 h (Fig. 4F). However, the LD volume distribution did not return to its original pattern even after SB2301 treatment under Atg5- (Fig. 4G, H, and Supplementary Fig. 21 for SB2301; Supplementary Figs. 20, 21 for DMSO) or Ubb-knockdown conditions (Fig. 4J, I, and Supplementary Fig. 21 for SB2301; Supplementary Figs. 20, 21 for DMSO). From these observations, we inferred that SB2301 treatment induced the translocation of PCYT2 at the LD surface and that the PE/PC ratio increased at the LD membrane, leading to LD coalescence. LD fusion via a rise in the PE/PC ratio occurred first regardless of autophagy. The resulting unfavorable large LDs may trigger ubiquitin-mediated lipophagy.
## Free fatty acid (FFA) consumption by the mitochondria
We performed a fluorescent-free fatty acid pulse-chase experiment41 to determine the fate of FFAs produced by lipophagy. The BODIPY-labeled FFA (BODIPY-C12) is one of the best reagents to track the intracellular movement of FFA from LD simply by monitoring the fluorescence signal of BODIPY-C12. For example, cells with high evergy demands can activate lipophagy and extend mitochondrial β-oxidation of FFAs to produce more ATPs for their survival42. When cells were incubated with BODIPY-C12, and then were depleted with serum, more FFAs were found in the mitochondria than in LDs (Fig. 5A, B). Similarly, we observed SB2301 treatment induced the degradation of cellular LDs via lipophagy and produced FFAs that accumulated in the mitochondria for β-oxidation (Fig. 5A, B). We measured the oxygen consumption rate (OCR) of cells in the absence and presence of SB2301 and confirmed that the cells increased their spare respiratory capacity (SRC) in response to SB2301 treatment (Fig. 5C, D, and Supplementary Fig. 22). Cells adjust their metabolic states with enhanced mitochondrial SRC under high energy demands or high-stress conditions to protect themselves from environmental stimuli43–45. Similar to the SRC increase, cells respond to environmental stimuli through AMP-activated protein kinase (AMPK), a representative energy sensor, and activate its downstream signaling to regulate β-oxidation46; AMPK activation allows the phosphorylation and inactivation of acetyl-CoA carboxylase (ACC), a potent inhibitor of mitochondrial fatty acid oxidation, leading to increased fatty acid oxidation47,48. By western blot analysis, we confirmed that AMPK and ACC were phosphorylated by SB2301 (Supplementary Fig. 23). The increase in SRC and AMPK-mediated ACC phosphorylation upon SB2301 treatment indicated that the cells turned on the buffering system to consume or remove the FFAs produced by SB2301-mediated lipophagy, thereby reducing cellular stress such as lipotoxicity. Fig. 5Cellular consumption of FFA is increased upon SB2301 treatment. A Representative live-cell fluorescence images of FFA pulse-chase assay with BODIPY-labeled FFA. 2 μM of BODIPY-C12 was treated on HeLa cells for 21 h. The cells were then treated with 5 μM of SB2301 or incubated in a serum-free medium for an additional 24 h. Mitochondria were stained with Mitotracker, and LD was stained with SF44. The images were merged by designating red for mitochondria, blue for BODIPY-C12, and green for LD. B Pearson correlation coefficient was calculated to demonstrate the colocalization of BODIPY-C12 and Mitotracker signals. Images were selected randomly from biological triplicates. All data are represented as dot plots with the mean ± SD. Data were analyzed using an unpaired t-test. ** $$P \leq 0.0052$$ vs. DMSO, ##$$P \leq 0.0070$$ vs. DMSO. C Oxygen consumption rate measurement. AML12 cells were treated with 20 μM of SB2301 for 24 h. All data are represented as the mean ± SD. Rot + Ant; rotenone + antimycin A. D Calculated spare respiratory capacity (maximum respiratory capacity—basal respiration) from C. All data are represented as dot plots with the mean ± SD, ($$n = 4$$). Data were analyzed using an unpaired t-test. *** $$P \leq 0.0004.$$
## Potential application of SB2301 to steatosis disease in a cell culture model
LD degradation can be a unique and beneficial phenotype for drug discovery as excessive LD accumulation is highly correlated with major causes of various metabolic diseases18. In addition, the amount of FFAs regulates non-alcoholic fatty liver diseases49. In this context, the mechanism for SB2301-activated lipophagy might be a potential therapeutic strategy as SB2301 effectively reduces LD and consumes FFAs. To explore this novel lipophagy activation mechanism as a therapeutic strategy, we applied SB2301 to in vitro disease models, particularly hepatic steatosis. To develop an in vitro steatosis model, we employed two model systems by treating HepG2 cells with oleic acid (OA)50 and tamoxifen (TM)51 to mimic a simple high-fat dietary system and severe steatohepatitis system, respectively. In both hepatic steatosis models, SB2301 reduced the cellular LD counts and area in a dose-dependent manner (Fig. 6). These findings confirm that the newly investigated lipophagy-mediated LD-reducing mechanism could be used as a therapeutic strategy for metabolic diseases, including non-alcoholic fatty liver disease and non-alcoholic steatohepatitis. Fig. 6LD-reducing effects on in vitro drug-induced steatosis models. HepG2 cells were treated with 75 μM of oleic acid (OA) or 10 μM of tamoxifen (TM) for 24 h to induce steatosis followed by SB2301 treatment for an additional 24 h.A Representative live-cell fluorescence LD images of SB2301 in OA- and TM-treated hepatic steatosis models. B Quantification results of cellular LD count and area from A. All data are represented as dot plots with the mean ± SD, ($$n = 3$$). Data were analyzed using an unpaired t-test. **** $P \leq 0.0001$, ***$$P \leq 0.001$$, $$$$$P \leq 0.0002$$ vs. DMSO, ###$$P \leq 0.0004$$, **$$P \leq 0.0093$$ vs. DMSO.
## Discussion
Cellular LD regulates energy homeostasis and lipid metabolism by changing its quantity and size in response to environmental changes, such as disease states and energy conditions52. LD research has gained considerable interest in academia and industry because of its relevance in pathophysiological conditions. However, only a few studies have explored the regulatory mechanisms of cellular LDs, especially by selective autophagy, namely lipophagy. Herein, we discovered a bioactive small molecule, SB2301, that reduces cellular LDs without severe cytotoxicity via a phenotype-based approach. We then demonstrated that SB2301 reduced cellular LDs by activating selective autophagy by monitoring LD sequestration into autophagosomes and autolysosomes, leading to lysosomal degradation. Interestingly, SB2301 treatment promoted ubiquitination on the LD surface, allowing specific substrate recognition for selective autophagy31,32,35, although ubiquitinated LDs and their association with lipophagy have not yet been reported. Therefore, we discovered a novel molecular mechanism of intracellular ubiquitin-mediated lipophagy. Several questions remain unanswered, including the identity of ubiquitinated proteins on the LD surface, and the mechanism by which phagophore recognize and sequester the ubiquitinated LDs53. Nevertheless, SB2301 combined with quantitative proteomics can serve as a unique research tool to investigate the novel ubiquitin-mediated lipophagy mechanisms.
PCYT2 was identified as a target protein of SB2301 using TS-FITGE. SB2301 treatment induced the translocation of PCYT2 to cellular LDs without affecting its enzymatic activity. We believe that SB2301-mediated translocation of PCYT2 to the LD surface might activate an increase in the PE/PC ratio specifically at LDs, thereby influencing their biophysical stability. As a result, the destabilized LD coalesced to minimize its curvature, and cells activate lipophagy to maintain homeostasis by eliminating unfavorable large LDs. Currently, it is unclear how SB2301 induces the translocation of PCYT2 to the LD surface and needs to conduct the PCYT2 knock-out study to demonstrate the requirement of PCYT2 for the activity of SB2301. However, the results obtained in this study serve as a pilot study to elucidate the detailed molecular mechanism of PCYT2 translocation and ubiquitin-mediated lipophagy processes.
Herein, we showed that SB2301 alters the membrane composition of cellular LDs through the spatial regulation of PCYT2, which directly supplies CDP-ethanolamine to the LD membrane. CEPT mediates the linking of CDP-ethanolamine with diacylglycerol, the final step for PE synthesis, and CEPT only exists on the ER surface54. However, it is known that CDP-choline does not necessarily require CEPT to synthesize PC on LD. Furthermore, the translocation of choline-phosphate cytidylyltransferase (PCYT1A) to LD itself can increase the amount of PC and prevent LD fusion55. The similarity between PE and PC in their biosynthetic mechanism and spatial synthesis56 might support the proposed mechanism in this study. Collectively, SB2301 induced translocation of PCYT2, leading to a spatial increase in the PE/PC ratio in the LD membrane. The resulting biophysical instability of the LD membrane causes LD coalescence and activates ubiquitin-mediated lipophagy to maintain cellular homeostasis. Further proteomics studies for the interactome of the target protein PCTY2 and the identification of ubiquitinated LD surface proteins would be essential to demonstrate how the coalescent LD surface induces protein ubiquitination, activates lipophagy, and identify the key players in this lipophagy-activating mechanism. Nevertheless, this ubiquitin-mediated lipophagy might provide a novel strategy for reducing the intracellular lipid content without inducing lipotoxicity because FFAs produced from degraded LDs can be consumed in mitochondria, as shown in this phenotype-based study. In addition, cellular LD reduction in in vitro hepatic steatosis models was demonstrated as a new therapeutic strategy to treat metabolic diseases caused by fat accumulation, including non-alcoholic steatohepatitis.
## Antibodies
Anti-LC3B (ab51520), anti-PCYT2 (ab126142), anti-IDH1 (ab81653), anti-ubiquitin (ab7780), anti-ATG5 (ab108327), TRITC-conjugated anti-rabbit IgG secondary antibodies (ab6718), and anti-DGAT1 (ab178711) were purchased from Abcam. Anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (CST 2118), HRP-labeled anti-mouse IgG (CST 7076), and HRP-labeled anti-rabbit IgG secondary antibodies (CST 7074) were purchased from Cell Signaling Technology. Anti-ACSL4 (sc-271800) and anti-WDR1 (sc-393159) were purchased from SantaCruz. Anti-SOAT1 (Novus biologicals, NB400-141) was purchased from Novus biologicals. Anti-SOAT2 (Cayman, 100027) was purchased from Cayman. Anti-DGAT2 (Thermo, PA5-21722) was purchased from Thermo Scientific.
## Plasmids
mCherry-GFP-LC3 plasmid (pBabe vector) was given from Dr. Heesun Cheong, Division of Chemical Biology, Research Institute, National Cancer Center, Korea. mCherry-LC3 plasmid was purchased from Addgene [40827].
## Imaging instrument
DeltaVision Elite imaging system from GE Healthcare was used for high resolution imaging experiment. Objective lenses were equipped with Olympus IX-71 inverted microscope. sCMOS camera and InSightSSI fluorescence illumination module were equipped with the system. For live cell imaging, a CO2 supporting chamber with an objective air heater were installed with the system. Images were analyzed with SoftWorks program supported by GE Healthcare.
## Cell culture
HepG2 cells were cultured in Dulbecco modified eagle medium (DMEM) with $10\%$ (v/v) fetal bovine serum (FBS), and $1\%$ (v/v) antibiotic-antimycotic (AA) solution. HeLa human cervical cancer cells were cultured in RPMI 1640 medium with $10\%$ (v/v) FBS, and $1\%$ AA solution. AML12 cells were cultured in DMEM/F12 (1:1) (Gibco, 11330-032), 1× insulin-transferrin-selenium-G Supplement (Gibco, 41400-045), dexamethasone (final 40 ng/ml), $10\%$ (v/v) FBS, and $1\%$ (v/v) AA solution. Cells were maintained in a 100-mm cell culture dish in an incubator at 37 °C, in a humidified atmosphere with $5\%$ CO2.
## LD imaging with SF44
Live cells were treated with SF44 (10 μM) and Hoechst 33342 (2 μg/ml). Serum-free condition (as a positive control) was changed to complete media before treatment with SF44 and Hoechst. After 30-min incubation, automatic fluorescence imaging was performed with InCell Analyzer 2000 [GE Healthcare] or DeltaVision Elite [GE Healthcare] without washing. Using InCell Analyzer 2000, images of four randomly selected spots per individual wells in a 96-well plate were automatically captured. Images were taken in auto-focusing mode at a 20× magnification. Fluorescence imaging was performed using the following filter settings: excitation_emission, $\frac{430}{24}$ nm_$\frac{605}{64}$ nm for LD; $\frac{350}{50}$ nm_$\frac{455}{50}$ nm for nucleus in InCell Analyzer 2000 or $\frac{438}{24}$ nm_$\frac{559}{38}$ nm for LD; $\frac{380}{18}$ nm_$\frac{435}{48}$ nm for nucleus in DeltaVision Elite. According to the manufacturer’s protocol, data were analyzed with InCell Developer program. The fluorescence intensity of LD was interpreted as a cellular organelle using a granularity module, and the area of individual cells was recognized by nuclei staining using collar segmentation.
## Cell viability assay
Cells were seeded in a transparent 96-well plate followed by compound treatment for the indicated times. Cell viability assay was done with WST assay following the manufacturer’s protocol.
## TG assay
HepG2 cells were seeded on a 12-well plate followed by compound treatment for the indicated time. The amount of triglyceride was measured by a triglyceride assay kit (Abcam, ab65336), following the manufacturer’s protocol.
## Mitochondrial DNA extraction and quantification with qPCR
HepG2 cells were treated with SB2301 for indicated times. The genomic DNA was extracted from HepG2 cells with DNeasy DNA extraction kit (Qiagen, 69581) according to the manufacturer’s instructions. 20 ng of genomic DNA was subjected to quantitative PCR (qPCR). KAPA SYBR FAST ABI Prime qPCR master mix (KK4605) and forward/reverse primers (200 nM) against human mitochondrial DNA or human GAPDH were mixed in a final volume of 20 μl with nuclease-free water. The PCR was conducted with StepOnePlus (Applied Biosystems), and the PCR cycling was used; initial denaturing at 95 °C for 3 min, followed by 40 cycles of denaturing at 95 °C for 3 s, and extension at 60 °C for 25 s. The data were analyzed by the comparative CT method, and the mitochondrial DNA levels were normalized to GAPDH levels.
## qPCR analysis of lipid biosynthesis related genes
Total mRNA was isolated from HepG2 cells treated with SB2301 using the RNeasy Plus mini kit (Qiagen 74134). Reverse transcription reaction (RT) was performed with AccuPower Cycle Script RT PreMix (Bioneer, K-2044) with 1 μg of total mRNA in a final volume 20 μl with nuclease-free water in the following conditions: 12 cycles with primer annealing at 30 °C for 1 min and DNA synthesis at 50 °C for 4 min. 1 μl of RT product was subjected to quantitative PCR (qPCR). KAPA SYBR FAST ABI Prime qPCR master mix (KK4605) and forward/reverse primers (200 nM) against each gene were mixed in a final volume 20 μl with nuclease-free water. The PCR was conducted with StepOnePlus (Applied Biosystems), and the PCR cycling was used; initial denaturing at 95 °C for 3 min, followed by 40 cycles of denaturing at 95 °C for 3 s, and extension at 60 °C for 30 s. The data were analyzed by the comparative CT method, and each expression levels were normalized to GAPDH levels.
## Western Blotting
Cells were lysed with radio-immunoprecipitation assay (RIPA) buffer (50 mM Tris, pH 7.8, 150 mM NaCl, $0.5\%$ deoxycholate, $1\%$ IGEPAL CA-630) and protease inhibitor cocktail (Roche). Lysates were obtained after centrifugation at 15,000 rpm for 20 min, by transferring the supernatant. Protein concentration was quantified with BCA protein assay kit. Overall protein sampling procedures were done at 4 °C. Prepared protein samples were analyzed with SDS-PAGE followed by a western blot procedure. Proteins were transferred into the PVDF membrane, and it was blocked with $2\%$ BSA in TBST over 1 h at room temperature (r.t.). Primary antibodies were treated overnight at 4 °C [Anti-LC3B; 1:2000, anti-ATG5; 1:1000, anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH); 1:2000, anti-DGAT1; 1:1000, anti-DGAT2; 1:1000, anti-SOAT1; 1:500, anti-SOAT2; 1:500] followed by washing with TBST. HRP-labeled anti-rabbit IgG or HRP-labeled anti-mouse IgG secondary antibody (1:5000) were treated at r.t. for 1 h. After washing with TBST, the membrane was developed by Amersham ECL prime solution. The chemiluminescent signal was measured by the ChemiDocMP imaging system.
## Plasmid transfection
HeLa cells were seeded on Lab-Tek II Chambered Coverglass w/Cover #1.5 Borosilicate Sterile/8 Well (Nunc 155409), 24 h before transfection. mCherry-GFP-LC3 plasmid or mCherry-hLC3 plasmid was transfected to HeLa cells using Lipofectamine 2000 reagent. Transfection was proceeded according to the manufacturer’s protocol.
## mCherry-GFP-LC3 puncta imaging
Fluorescence images from HeLa cells transfected mCherry-GFP-LC3 were obtained with 60× scale, using mCherry/mCherry, GFP/GFP (Excitation/Emission) filter sets. mCherry (excitation: $\frac{575}{25}$ nm, emission: $\frac{625}{45}$ nm); and GFP (excitation: $\frac{475}{28}$ nm, emission: $\frac{525}{48}$ nm). Images were analyzed with SoftWorks deconvolution software.
## mCherry-hLC3 and LD imaging
mCherry-hLC3-transfected HeLa cells were treated with SF44 (10 μM) for 30 min before imaging. Fluorescence images were obtained at a magnification of 100× using mCherry/mCherry (excitation_emission, $\frac{575}{25}$ nm_$\frac{625}{45}$ nm) and CFP/YFP (excitation_emission, $\frac{438}{24}$ nm_$\frac{559}{38}$ nm) filter sets. Images were analyzed using the SoftWorks deconvolution software.
## Lysosomes and LD imaging
Before fluorescence imaging, lysosomes were stained with Lysotracker DeepRed (50 nM, Thermo Scientific, L7528) for 1.5 h. LDs were stained with SF44 (10 μM), and nuclei were stained with Hoechst 33342 (2 μg/ml) for 30 min. Images were obtained at a magnification of 100× using Cy5/Cy5, (excitation_emission, $\frac{632}{22}$ nm_$\frac{676}{34}$ nm), FITC/TRITC (excitation_emission, $\frac{475}{28}$ nm_$\frac{594}{45}$ nm), and DAPI/DAPI (excitation_emission, $\frac{380}{18}$ nm_$\frac{435}{48}$ nm). Images were analyzed using the SoftWorks deconvolution software.
## Immunofluorescence
Cells were washed with cold phosphate-buffered saline (PBS) and fixed with $4\%$ paraformaldehyde in PBS. The cells were incubated in $0.1\%$ Triton X-100 in PBS for 15 min at r.t. for permeabilization. The samples were washed three times with ice-cold PBS, followed by incubation with $2\%$ BSA in PBS for 1 h at r.t. Fixed cells on dishes were exposed to the diluted primary antibody solution (ubiquitin; 1:300, PCYT2; 1:200) in PBS with $1\%$ BSA and 3 μM BODIPY $\frac{493}{503}$ (Thermo, D3922) at 4 °C overnight. The primary antibody was decanted and washed three times with PBS. Thereafter, a diluted anti-rabbit IgG-TRITC antibody (1:200) solution with 3 μM BODIPY $\frac{493}{503}$ and Hoechst 33342 (2 μg/ml) was applied to the samples and incubated at r.t. for 1 h. After washing three times with PBS, fluorescence images were captured using DeltaVision Elite fluorescence microscopy (100×) using TRITC/TRITC, (excitation_emission, $\frac{542}{27}$ nm_$\frac{594}{45}$ nm), FITC/FITC (excitation_emission, $\frac{475}{28}$ nm_$\frac{525}{48}$ nm), and DAPI/DAPI (excitation_emission, $\frac{380}{18}$ nm_$\frac{435}{48}$ nm). Images were analyzed using the SoftWorks deconvolution software.
## TS-FITGE
Human hepatocellular carcinoma HepG2 cells were incubated in serum-free DMEM with 20 μM SB2301 ($0.2\%$ (v/v) of DMSO in final concentration) for 1 h at 37 °C. Heat shock was applied to the cells for 3 min. The resulting cells were stabilized at 25 °C for 3 min, washed once with PBS, and resuspended in PBS containing $0.4\%$ NP-40 and protease inhibitor. Cells were subjected to three freeze(liquid nitrogen)/thaw cycles for cell lysis. The cell lysate was clarified by centrifugation at 20,000 g for 20 min at 4 °C. 50 mg of protein was precipitated, and the residual pellet was resuspended in 10 μl of labeling buffer (30 mM Tris-HCl at pH 8.6, 2 M thiourea, 7 M urea, and $4\%$ (w/v) CHAPS) with sonication. The soluble proteomes were mixed with 0.4 mM Cy3-NHS (for DMSO-treated group) or Cy5-NHS (for SB2301-treated group), and incubated at 4 °C for 45 min. The dye-conjugated proteomes were precipitated with cold acetone and resuspended in 75 μl of rehydration buffer (7 M urea, 2 M thiourea, $2\%$ (w/v) CHAPS, 40 mM DTT, and $1\%$ IPG buffer). DMSO-treated SB2301-treated samples were mixed, and 150 μl (75 μl for Cy3- and Cy5-labeled each) of proteomes was loaded on a 24-cm Immobiline Drystrip gel [GE Healthcare]. Isoelectric focusing was performed using Ettan IPGphor 3 [GE Healthcare] followed by 2-dimensional SDS-PAGE with an Ettan DALTsix system [GE Healthcare]. The gel was scanned using a Typhoon Trio [GE Healthcare]. For LCMS/MS, the gel was excised after silver staining.
## In-gel digestion and mass spectrometry
The protein spots from the silver-stained gel were excised, destained, and digested with trypsin. The mixture was evaporated in SpeedVac and then dissolved in $10\%$ acetonitrile with $0.1\%$ formic acid. The resulting peptides were desalted in a trap column (180 μm × 20 mm, Symmetry C18) and separated on a C18 reversed-phase analytical column (75 μm × 200 mm, 1.7 μm, BEH130 C18) [Waters] with an electrospray ionization Pico Tip (±10 μm i.d.) [ New objective]. The data were converted to.pkl files by Protein Lynx Global Server and searched by MASCOT engine with the SwissProt database.
## CETSA
Trypsinized HepG2 cells were incubated in serum-free DMEM media with the 20 μM compound (final $0.2\%$ (v/v) DMSO concentration) for 1 h at 37 °C. Heat shock applied to cell for 3 min and cells were stabilized at 25 °C for 3 min. Wash cell with PBS once and resuspend cells with PBS containing $0.4\%$ NP-40 and protease inhibitor. Lysis cells with freeze (liquid nitrogen)/thaw cycle, three times. Prepare soluble proteins and detect protein quantity with western blotting. ( Primary antibody dilution/anti-ACSL4; 1:1000, anti-PCYT2; 1:1000, anti-IDH1; 1:1000, anti-WDR1; 1:200, anti-GAPDH; 1:2000).
## si-RNA treatment
20 μM of si-RNA was applied to HepG2 cells for 48 h with Lipofectamine RNAiMAX (Invitrogen, 13778-100) following manufacture’s protocols.
## SPR analysis
The equilibrium dissociation constant (Kd) toward PCYT2 was determined by SPR using a Biacore T100 instrument [GE Healthcare]. The carboxyl group on the surface of the CM5 sensor chip was replaced with reactive succinimide ester using a combination of 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC) and N-hydroxysuccinimide (NHS) in flow cells 1 and 2. Human PCYT2 (Prospec, enz-221) was immobilized on flow cell 2 (10000 RU) through the formation of amide bonds with the resulting NHS ester. The remaining NHS ester on flow cells 1 and 2 was quenched via injection with 1 M ethanolamine-HCl (pH 8.0) solution. During the immobilization process, PBS was used as the running buffer. After immobilization, different concentrations of the compounds were injected for 60 s at a flow rate of 10 μL·min−1. At the same flow rate, the dissociation from the sensor surface was monitored for 300 s. For the running buffer, we used 1× PBS (pH 7.3) containing $3\%$ DMSO and $0.05\%$ P20. The binding events were measured at 25 °C. Data analysis was performed using the Biacore T100 Evaluation software [GE Healthcare]. Final sensorgrams were obtained after the elimination of responses from flow cell 1 and buffer-only control. The Kd was calculated by fitting the sensorgrams to a 1:1 binding model.
## PCYT2 enzymatic assay
PCYT2 activity was assayed as described previously with minor modification57. Briefly, prepare reaction mixture (50 μl) in 20 mM Tris-HCl buffer (pH 7.8), 5 mM DTT, 10 mM MgCl2, 650 μM phosphoethanolamine, and CTP with various concentration. Then the reaction mixture was incubated with 0.1 μg of purified PCYT2 at 37 °C for 3 min. Reactions were terminated by boiling for 2 min. The concentration of product (pyrophosphate) was measured by pyrophosphate assay kit (Lonza, LT07-610), and the reaction rate was calculated. DMSO and SB2301 were added at the indicated concentration in the reaction mixture.
## Quantification of the superimposed area between LD and PCYT2
Images were analyzed with ImageJ software [National Institutes of Health, Bethesda, MD, USA]. Background images of PCYT2 channel were obtained from Gaussian blur using Radius 8, and the original images of PCYT2 channel were subtracted from the background images to correct for inhomogeneous background. LD mask was prepared by adjusting the LD signal according to an intensity threshold (Image > adjust > threshold) using the default algorithm. For the LD edge mask, Find edges (Process > Find edges) was used to the thresholded LD images. The superimposed area between PCYT2 of which background was corrected and LD mask or LD edge mask was quantified by Image calculator (Process > Image calculator) using AND operation. The resulting images were thresholded, and the area was quantified by Analyze particles (Analyze > Analyze particles) using a size filter of 5-infinity (pixel).
## LD fractionation
LDs were isolated from HepG2 cell (~108 cells) using LD isolation kits (Cell Biolabs, MET-5011) according to the manufacturer’s instructions. The membrane, cytosol, and LD fraction were collected for purity check. 1 ml of chloroform/acetone (1:1, v/v) was added to separate proteins and lipids from the isolated LDs. Next, the organic phase was collected for further phospholipid composition analysis. Repeat this process twice with the pellet to dissolve the lipids. After removing the organic phase, the pellet was dried at r.t. and resuspended with SDS sampling buffer to prepare protein samples for purity check.
## LD size distribution by volume
Dynamic light scattering data was obtained with purified LDs in PMMA cuvette (Ratiolab, 2810100) by Malvern Zetasizer Nano-S. Dispersant and temperature were selected as water and 25 °C, respectively.
## PE/PC ratio analysis
To measure the amount of PE and PC in the lipid sample, PE assay kit (Biovision, K499) and PC assay kit (Biovision, K576) were used. Extracted lipids from whole cell lysates or purified cellular LDs were prepared via solvent evaporation using a rotary evaporator. According to the manufacturer’s instructions, dried lipids reconstituted with PE assay buffer and PC assay buffer were employed to measure each quantity.
## LD volume and size distribution analysis
Ten fluorescence images were captured at one point with 0.8 µm z-depth. Projected images were generated using SoftWorks deconvolution software. Fifteen randomly selected images from each condition were analyzed using InCell Developer [GE Healthcare], and the diameters of all LDs present in the given image were quantified. Assuming that the LD is a perfect sphere, the volume of each lipid droplet was calculated using the relationship between the radius and volume ($V = 4$/3 πr3). The LDs were divided into nine groups according to the length of the radius. The LD size distribution was calculated by dividing the sum of the volumes of LDs corresponding to each group by the sum of the volumes of total LDs.
## Fluorescent FFA pulse-chase experiments
HeLa cells were incubated with 2 μM BODIPY $\frac{558}{568}$ C12 (Thermo, D3835) in complete media for 21 h. Cells were then washed twice with complete media, incubated for an additional 1 h without BODIPY $\frac{558}{568}$ C12. SB2301 was treated for 24 h. For positive control, serum-starved (16 h) cells were prepared, separately. Mitochondria were labeled with 100 nM MitoTracker DeepRed (Thermo, M22426), and LDs were labeled with 10 μM SF44 for 30 min simultaneously prior to fluorescence imaging. Live cell imaging was conducted within 2 h.
## OCR measurement
AML12 cells were seeded on an Agilent Seahorse XFe24 plate and subsequently treated with SB2301. A sensor cartridge with Seahorse XF Calibrant (Agilent, 100840-000) was hydrated at 37 °C overnight in a non-CO2 incubator. OCR was measured in Seahorse XF DMEM (Agilent, 103757-100) containing 17 mM glucose (Agilent, 103577-100), 2.5 mM glutamine (Agilent, 103579-100), and 0.5 mM pyruvate (Agilent, 103578-100) in response to 1.5 μM oligomycin, 1 μM fluorocarbonylcyanide phenylhydrazone (FCCP), and 0.5 μM rotenone + antimycin A (Seahorse XF Cell Mito Stress Test Kit, 103015-100) with Seahorse XFe24 [Agilent].
## Statistics and reproducibility
The type of used statistical test and the number of biological replicates are specified in each figure’s legend. The figures of each experiment from each biological replicate can be found in the Supplementary Data 1.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Video 1 Supplementary Video 2 Supplementary Video 3 Supplementary Video 4 Reporting Summary The online version contains supplementary material available at 10.1038/s42003-023-04682-9.
## Peer review information
This manuscript has been previously reviewed at another Nature Portfolio journal. The manuscript was considered suitable for publication without further review at Communications Biology. Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editor: Manuel Breuer.
## References
1. Olzmann JA, Carvalho P. **Dynamics and functions of lipid droplets**. *Nat. Rev. Mol. Cell Biol.* (2019) **20** 137-155. DOI: 10.1038/s41580-018-0085-z
2. Bailey Andrew P. **Antioxidant role for lipid droplets in a stem cell niche of Drosophila**. *Cell* (2015) **163** 340-353. DOI: 10.1016/j.cell.2015.09.020
3. Penno A, Hackenbroich G, Thiele C. **Phospholipids and lipid droplets**. *Biochim. Biophys. Acta - Mol. Cell Biol. Lipids* (2013) **1831** 589-594. DOI: 10.1016/j.bbalip.2012.12.001
4. Itabe H, Yamaguchi T, Nimura S, Sasabe N. **Perilipins: a diversity of intracellular lipid droplet proteins**. *Lipids Health Dis* (2017) **16** 83. DOI: 10.1186/s12944-017-0473-y
5. Bersuker K. **A proximity labeling strategy provides insights into the composition and dynamics of lipid droplet proteomes**. *Dev. Cell* (2018) **44** 97-112 e117. DOI: 10.1016/j.devcel.2017.11.020
6. Welte MA. **Expanding roles for lipid droplets**. *Curr. Biol.* (2015) **25** R470-R481. DOI: 10.1016/j.cub.2015.04.004
7. Rizack MA. **Activation of an epinephrine-sensitive lipolytic activity from adipose tissue by adenosine 3′,5′-phosphate**. *J. Biol. Chem* (1964) **239** 392-395. DOI: 10.1016/S0021-9258(18)51690-2
8. Vaughan M, Steinberg D. **Effect of hormones on lipolysis and esterification of free fatty acids during incubation of adipose tissue in vitro**. *J. Lipid Res.* (1963) **4** 193-199. DOI: 10.1016/S0022-2275(20)40346-3
9. Singh R. **Autophagy regulates lipid metabolism**. *Nature* (2009) **458** 1131-1135. DOI: 10.1038/nature07976
10. Welte MA, Gould AP. **Lipid droplet functions beyond energy storage**. *Biochim. Biophys. Acta Mol. Cell Biol, Lipids* (2017) **1862** 1260-1272. DOI: 10.1016/j.bbalip.2017.07.006
11. Bosma M. **Sequestration of fatty acids in triglycerides prevents endoplasmic reticulum stress in an in vitro model of cardiomyocyte lipotoxicity**. *Biochim. Biophys. Acta. Mol. Cell Biol. Lipids* (2014) **1841** 1648-1655. DOI: 10.1016/j.bbalip.2014.09.012
12. Bensaad K. **Fatty acid uptake and lipid storage induced by HIF-1α contribute to cell growth and survival after hypoxia-reoxygenation**. *Cell Rep* (2014) **9** 349-365. DOI: 10.1016/j.celrep.2014.08.056
13. Moldavski O. **Lipid droplets are essential for efficient clearance of cytosolic inclusion bodies**. *Dev. Cell* (2015) **33** 603-610. DOI: 10.1016/j.devcel.2015.04.015
14. Camus G. **Diacylglycerol acyltransferase-1 localizes Hepatitis C virus NS5A protein to lipid droplets and enhances NS5A interaction with the viral capsid core**. *J. Biol. Chem.* (2013) **288** 9915-9923. DOI: 10.1074/jbc.M112.434910
15. Spicher L, Kessler F. **Unexpected roles of plastoglobules (plastid lipid droplets) in vitamin K1 and E metabolism**. *Curr. Opin. Plant Biol.* (2015) **25** 123-129. DOI: 10.1016/j.pbi.2015.05.005
16. Papackova Z, Cahova M. **Fatty acid signaling: the new function of intracellular lipases**. *Int. J. Mol. Sci.* (2015) **16** 3831-3855. DOI: 10.3390/ijms16023831
17. Fabbrini E, Sullivan S, Klein S. **Obesity and nonalcoholic fatty liver disease: Biochemical, metabolic, and clinical implications**. *Hepatology* (2010) **51** 679-689. DOI: 10.1002/hep.23280
18. Gluchowski NL, Becuwe M, Walther TC, Farese RV. **Lipid droplets and liver disease: from basic biology to clinical implications**. *Nat. Rev. Gastroenterol. Hepatol.* (2017) **14** 343. DOI: 10.1038/nrgastro.2017.32
19. Loon LJCV. **Intramyocellular lipid content in type 2 diabetes patients compared with overweight sedentary men and highly trained endurance athletes**. *Am. J. Physiol. Endocrinol. Metab.* (2004) **287** E558-E565. DOI: 10.1152/ajpendo.00464.2003
20. McGavock JM. **Cardiac steatosis in diabetes mellitus: a 1H-magnetic resonance spectroscopy study**. *Circulation* (2007) **116** 1170-1175. DOI: 10.1161/CIRCULATIONAHA.106.645614
21. Chiu H-C. **A novel mouse model of lipotoxic cardiomyopathy**. *J. Clin. Invest.* (2001) **107** 813-822. DOI: 10.1172/JCI10947
22. Larigauderie G. **Adipophilin enhances lipid accumulation and prevents lipid efflux from THP-1 macrophages: potential role in atherogenesis**. *Asterioscler, Thromb, Vasc. Biol* (2004) **24** 504-510. DOI: 10.1161/01.ATV.0000115638.27381.97
23. Zubiete-Franco I. **Methionine and S-adenosylmethionine levels are critical regulators of PP2A activity modulating lipophagy during steatosis**. *J. Hepatol.* (2016) **64** 409-418. DOI: 10.1016/j.jhep.2015.08.037
24. Kurahashi T. **An SOD1 deficiency enhances lipid droplet accumulation in the fasted mouse liver by aborting lipophagy**. *Biochem. Biophys. Res. Commun.* (2015) **467** 866-871. DOI: 10.1016/j.bbrc.2015.10.052
25. Zhang Z. **Lipophagy and liver disease: New perspectives to better understanding and therapy**. *Biomed. Pharmacother.* (2018) **97** 339-348. DOI: 10.1016/j.biopha.2017.07.168
26. Kim E, Lee S, Park SB. **A Seoul-Fluor-based bioprobe for lipid droplets and its application in image-based high throughput screening**. *Chem. Commun.* (2012) **48** 2331-2333. DOI: 10.1039/c2cc17496k
27. Lee S, Kim E, Park SB. **Discovery of autophagy modulators through the construction of a high-content screening platform via monitoring of lipid droplets**. *Chem. Sci.* (2013) **4** 3282-3287. DOI: 10.1039/c3sc51344k
28. Kim M, Hwang YS, Cho W, Park SB. **Synthesis of 3,5-disubstituted isoxazoles containing privileged substructures with a diverse display of polar surface area**. *ACS Comb. Sci.* (2017) **19** 407-413. DOI: 10.1021/acscombsci.7b00032
29. Patani GA, LaVoie EJ. **Bioisosterism: A rational approach in drug design**. *Chem. Rev.* (1996) **96** 3147-3176. DOI: 10.1021/cr950066q
30. Klionsky DJ. **Guidelines for the use and interpretation of assays for monitoring autophagy (4th edition)1**. *Autophagy* (2021) **17** 1-382. DOI: 10.1080/15548627.2020.1797280
31. Kirkin V, McEwan DG, Novak I, Dikic I. **A role for ubiquitin in selective autophagy**. *Mol. Cell* (2009) **34** 259-269. DOI: 10.1016/j.molcel.2009.04.026
32. Koyano F. **Ubiquitin is phosphorylated by PINK1 to activate parkin**. *Nature* (2014) **510** 162-166. DOI: 10.1038/nature13392
33. Kane LA. **PINK1 phosphorylates ubiquitin to activate Parkin E3 ubiquitin ligase activity**. *J. Cell Biol* (2014) **205** 143-153. DOI: 10.1083/jcb.201402104
34. Lu K, Psakhye I, Jentsch S. **Autophagic clearance of PolyQ proteins mediated by ubiquitin-Atg8 adaptors of the conserved CUET protein family**. *Cell* (2014) **158** 549-563. DOI: 10.1016/j.cell.2014.05.048
35. Kim PK, Hailey DW, Mullen RT, Lippincott-Schwartz J. **Ubiquitin signals autophagic degradation of cytosolic proteins and peroxisomes**. *Proc. Natl. Acad. Sci.* (2008) **105** 20567-20574. DOI: 10.1073/pnas.0810611105
36. Thurston TLM, Ryzhakov G, Bloor S, von Muhlinen N, Randow F. **The TBK1 adaptor and autophagy receptor NDP52 restricts the proliferation of ubiquitin-coated bacteria**. *Nat. Immunol.* (2009) **10** 1215-1221. DOI: 10.1038/ni.1800
37. Park H, Ha J, Koo JY, Park J, Park SB. **Label-free target identification using in-gel fluorescence difference via thermal stability shift**. *Chem. Sci.* (2017) **8** 1127-1133. DOI: 10.1039/C6SC03238A
38. Bakovic M, Fullerton MD, Michel V. **Metabolic and molecular aspects of ethanolamine phospholipid biosynthesis: the role of CTP:phosphoethanolamine cytidylyltransferase (Pcyt2)**. *Biochem. Cell Biol.* (2007) **85** 283-300. DOI: 10.1139/O07-006
39. Sundler R. **Ethanolaminephosphate cytidylyltransferase. Purification and characterization of the enzyme from rat liver**. *J. Biol. Chem* (1975) **250** 8585-8590. DOI: 10.1016/S0021-9258(19)40710-2
40. Thiam AR, Farese RV, Walther TC. **The biophysics and cell biology of lipid droplets**. *Nat. Rev. Mol. Cell Biol.* (2013) **14** 775-786. DOI: 10.1038/nrm3699
41. Rambold Angelika S, Cohen S, Lippincott-Schwartz J. **Fatty acid trafficking in starved cells: Regulation by lipid droplet lipolysis, autophagy, and mitochondrial fusion dynamics**. *Dev. Cell* (2015) **32** 678-692. DOI: 10.1016/j.devcel.2015.01.029
42. Liu K, Czaja MJ. **Regulation of lipid stores and metabolism by lipophagy**. *Cell Death Differ* (2013) **20** 3-11. DOI: 10.1038/cdd.2012.63
43. van der Windt Gerritje JW. **Mitochondrial respiratory capacity is a critical regulator of CD8+ T cell memory development**. *Immunity* (2012) **36** 68-78. DOI: 10.1016/j.immuni.2011.12.007
44. Yamamoto H. **Amla enhances mitochondrial spare respiratory capacity by increasing mitochondrial biogenesis and antioxidant systems in a murine skeletal muscle cell line**. *Oxid. Med. Cell. Longev.* (2016) **2016** 1735841. DOI: 10.1155/2016/1735841
45. Choi SW, Gerencser AA, Nicholls DG. **Bioenergetic analysis of isolated cerebrocortical nerve terminals on a microgram scale: spare respiratory capacity and stochastic mitochondrial failure**. *J. Neurochem* (2009) **109** 1179-1191. DOI: 10.1111/j.1471-4159.2009.06055.x
46. Hardie DG, Ross FA, Hawley SA. **AMPK: a nutrient and energy sensor that maintains energy homeostasis**. *Nat. Rev. Mol. Cell Biol.* (2012) **13** 251-262. DOI: 10.1038/nrm3311
47. Ruderman NB, Saha AK, Vavvas D, Witters LA. **Malonyl-CoA, fuel sensing, and insulin resistance**. *Am. J. Physiol. Endocrinol. Metab.* (1999) **276** E1-E18. DOI: 10.1152/ajpendo.1999.276.1.E1
48. Zang MW. **AMP-activated protein kinase is required for the lipid-lowering effect of metformin in insulin-resistant human HepG2 cells**. *J. Biol. Chem.* (2004) **279** 47898-47905. DOI: 10.1074/jbc.M408149200
49. Schweiger M. **Pharmacological inhibition of adipose triglyceride lipase corrects high-fat diet-induced insulin resistance and hepatosteatosis in mice**. *Nat. Commun.* (2017) **8** 14859. DOI: 10.1038/ncomms14859
50. Cui W, Chen SL, Hu KQ. **Quantification and mechanisms of oleic acid-induced steatosis in HepG2 cells**. *Am. J. Transl. Res.* (2010) **2** 95-104. PMID: 20182586
51. Zhao F. **The effect and mechanism of tamoxifen-induced hepatocyte steatosis in vitro**. *Int. J. Mol. Sci.* (2014) **15** 4019-4030. DOI: 10.3390/ijms15034019
52. Henne WM, Reese ML, Goodman JM. **The assembly of lipid droplets and their roles in challenged cells**. *EMBO J* (2018) **37** e98947. DOI: 10.15252/embj.201898947
53. Gatica D, Lahiri V, Klionsky DJ. **Cargo recognition and degradation by selective autophagy**. *Nat. Cell Biol.* (2018) **20** 233-242. DOI: 10.1038/s41556-018-0037-z
54. Henneberry AL, Wright MM, McMaster CR. **The major sites of cellular phospholipid synthesis and molecular determinants of fatty acid and lipid head group specificity**. *Mol. Biol. Cell* (2002) **13** 3148-3161. DOI: 10.1091/mbc.01-11-0540
55. Krahmer N. **Phosphatidylcholine synthesis for lipid droplet expansion is mediated by localized activation of CTP:Phosphocholine cytidylyltransferase**. *Cell Metab* (2011) **14** 504-515. DOI: 10.1016/j.cmet.2011.07.013
56. Henneberry AL, Wistow G, McMaster CR. **Cloning, genomic organization, and characterization of a human cholinephosphotransferase**. *J. Biol. Chem.* (2000) **275** 29808-29815. DOI: 10.1074/jbc.M005786200
57. Gohil VM. **Meclizine inhibits mitochondrial respiration through direct targeting of cytosolic phosphoethanolamine metabolism**. *J. Biol. Chem* (2013) **288** 35387-35395. DOI: 10.1074/jbc.M113.489237
|
---
title: 'Factors influencing influenza, pneumococcal and shingles vaccine uptake and
refusal in older adults: a population-based cross-sectional study in England'
authors:
- Pui San Tan
- Martina Patone
- Ashley Kieran Clift
- Hajira Dambha-Miller
- Defne Saatci
- Tom A Ranger
- Cesar Garriga
- Francesco Zaccardi
- Baiju R Shah
- Carol Coupland
- Simon J Griffin
- Kamlesh Khunti
- Julia Hippisley-Cox
journal: BMJ Open
year: 2023
pmcid: PMC10030484
doi: 10.1136/bmjopen-2021-058705
license: CC BY 4.0
---
# Factors influencing influenza, pneumococcal and shingles vaccine uptake and refusal in older adults: a population-based cross-sectional study in England
## Abstract
### Objectives
Uptake of influenza, pneumococcal and shingles vaccines in older adults vary across regions and socioeconomic backgrounds. In this study, we study the coverage and factors associated with vaccination uptake, as well as refusal in the unvaccinated population and their associations with ethnicity, deprivation, household size and health conditions.
### Design, setting and participants
This is a cross-sectional study of adults aged 65 years or older in England, using a large primary care database. Associations of vaccine uptake and refusal in the unvaccinated with ethnicity, deprivation, household size and health conditions were modelled using multivariable logistic regression.
### Outcome measure
Influenza, pneumococcal and shingles vaccine uptake and refusal (in the unvaccinated).
### Results
This study included 2 054 463 patients from 1318 general practices. 1 711 465 ($83.3\%$) received at least one influenza vaccine, 1 391 228 ($67.7\%$) pneumococcal vaccine and 690 783 ($53.4\%$) shingles vaccine. Compared with White ethnicity, influenza vaccine uptake was lower in Chinese (OR 0.49; $95\%$ CI 0.45 to 0.53), ‘Other ethnic’ groups (0.63; $95\%$ CI 0.60 to 0.65), black Caribbean (0.68; $95\%$ CI 0.64 to 0.71) and black African (0.72; $95\%$ CI 0.68 to 0.77). There was generally lower vaccination uptake among more deprived individuals, people living in larger household sizes (three or more persons) and those with fewer health conditions. Among those who were unvaccinated, higher odds of refusal were associated with the black Caribbean ethnic group and marginally with increased deprivation, but not associated with higher refusal in those living in large households or those with lesser health conditions.
### Conclusion
Certain ethnic minority groups, deprived populations, large households and 'healthier' individuals were less likely to receive a vaccine, although higher refusal was only associated with ethnicity and deprivation but not larger households nor healthier individuals. Understanding these may inform tailored public health messaging to different communities for equitable implementation of vaccination programmes.
## Background
Older adults are often more susceptible to infectious diseases circulating in the community, and may develop more severe health outcomes when infected due to lower immune responses associated with ageing1 and comorbidities. National influenza, pneumococcal and shingles vaccination programmes for older adults have been implemented in the UK in various phases.2–4 Through these national vaccination programmes, ‘seasonal’ influenza vaccines are offered annually, pneumococcal vaccines are offered as a single dose to adults aged 65 years and above, while the shingles vaccine is offered as a single dose to adults aged 70–79 years.2–4 The WHO recommends a target of $75\%$ population vaccination coverage.5 Recent reports from Public Health England have reported $81\%$ influenza vaccination coverage and $69\%$ pneumococcal vaccination coverage in adults aged 65 years and above, and $47\%$–$77\%$ for shingles vaccination coverage in adults aged 71 and 78, respectively.2–4 However, some evidence suggests that there could be differences in terms of vaccination coverage, potentially varying by geographical region, ethnicity, deprivation, household size and health conditions.2–4 6 7 For the purposes of equitable public health strategy, it is important to understand factors associated with uptake of vaccinations, and refusal of vaccinations in the unvaccinated population. Prior studies have demonstrated differential uptake of existing vaccinations across sociodemographic groups, however, many studies have either studied single vaccinations, not captured the appreciable casemix inherent to sociodemographic groups (such as by using broad ethnic categories), analysed a small set of relevant health conditions, and relied on potentially imprecise or biased self-report measures.7–9 In addition, although household size is known to increase the risk of transmission for infectious diseases, evidence on the association between household size and vaccination uptake remains limited.10 A few previous studies have suggested that individuals from larger households were less likely to be vaccinated, although these studies were small and mainly focused on childhood vaccinations.11 12 Further, it is of interest to understand the pathway events leading to the lack of vaccine uptake, and to what extent these are driven by patient refusal.
Here, we evaluated factors associated with uptake and refusal of existing national vaccination programmes (influenza, pneumococcal and shingles) in older adults (aged 65 years and above) in England and their associations with ethnic group, deprivation, household size and health conditions.
## Study population and data source
We performed a population-based cross-sectional study using QResearch (V.45). QResearch is a database with over 10 million current patients registered at more than 1800 practices in England. QResearch is an electronic healthcare primary care database in the UK with individual patient level records for general practices using the EMIS computer record system. The database captures information from general pratitioner (GP) consultations; including patient demographics, socioeconomic status, diagnoses, laboratory test results, treatments and vaccinations. The database has good representation of the general population of England, particularly in terms of different ethnic groups with proportions close to those reported by Office for National Statistics.13 *In this* study, we included adults aged 65–99 years currently registered with 1318 practices during the period 24 January 2020 to 31 October 2020, which comprised 2 054 463 of approximately 13.7 million patients aged 65 and over registered with a GP in England.14 We assessed the uptake and refusal of influenza, pneumococcal and shingles vaccines from 1 January 1989 to 31 October 2020 (last database update) as our main study outcome. As the shingles vaccination was rolled out nationally in England in 2013 for those aged 70 and up until 79,15 we included in our shingles vaccine analysis only those aged 70 and above, excluding those aged 80 and above in year 2013 as they were not eligible at the time. Uptake was defined as the last recorded instance of receiving the vaccines of interest within the study period. This was mostly in GP surgeries (~$99\%$), but also in-hospital or pharmacy administrations. Refusal was analysed in those with no record of vaccination, defined as last recorded instances of explicit refusal ($74\%$–$82\%$ of recorded code instances), consent not being given ($18\%$–$26\%$) or non-attendance to a scheduled vaccination appointment (0.03–$0.3\%$).16 Outcomes were defined using code dictionaries comprising relevant Read and SNOMED codes as inputted into the EMIS software by healthcare practitioners.
We extracted demographic data including age, sex, self-reported ethnic group, Townsend deprivation index quintile,17 18 geographical region within England ($$n = 10$$, see table 1), housing status and household size. Townsend deprivation score is commonly used in the UK to measure socioeconomic status. It uses the following characteristics to measure deprivation by postcode; proportion of [1] unemployment, [2] non-car ownership, [3] non-home ownership and [4] household crowding— a higher score suggesting greater deprivation. In this study, the scores were reported in quintiles, that is, the first quintile indicates the least deprived, while the fifth quintile indicates most deprived. Ethnicity was grouped into nine categories—white (white British, white Irish, other white), Indian, Pakistani, Bangladeshi, Other Asian, black Caribbean, black African, Chinese, Other ethnic group (white and black, white and Asian, other mixed, other black, other ethnic group). We also extracted data using GP Read and SNOMED codes from primary care records and International Classification of Diseases 10th Revision (ICD-10) codes from hospital records (where available) for diagnoses of asthma, chronic obstructive pulmonary disease (COPD), diabetes mellitus (types 1 and 2), hypertension, coronary heart disease, atrial fibrillation (AF), congenital heart disease, congestive cardiac failure (CCF), chronic neurological diseases (Parkinson’s disease, epilepsy, cerebral palsy), learning disability, dementia and severe mental illness (schizophrenia, severe depression, bipolar affective disorder and psychosis) and immune suppression (based on use of immunosuppressant medications). For each vaccination outcome (uptake and refusal), people with health conditions diagnosed prior to the vaccination outcome were defined as exposed, while those diagnosed with health conditions after the outcome were defined as unexposed. The most recently recorded body mass index (BMI) and smoking status were identified for each individual.
**Table 1**
| Characteristics | Characteristics.1 | Study population | Vaccine uptake | Vaccine uptake.1 | Vaccine uptake.2 |
| --- | --- | --- | --- | --- | --- |
| Characteristics | Characteristics | Overall | Influenza | Pneumococcal | Shingles* |
| Total | N (row %) | 2 054 463 | 1 711 465 (83.3) | 1 391 228 (67.7) | 690 783 (53.4) |
| Age | Mean (SD) | 75.5 (7.7) | 76.3 (7.7) | 77.1 (7.5) | 77.2 (4.4) |
| Age | 65–69 | 541 272 (26.3) | 373 566 (21.8) | 232 831 (16.7) | – |
| Age | 70–79 | 922 198 (44.9) | 793 150 (46.3) | 665 037 (47.8) | 469 684 (68.0) |
| Age | 80–89 | 471 167 (22.9) | 434 074 (25.4) | 395 456 (28.4) | 221 099 (32.0) |
| Age | 90–99 | 119 826 (5.8) | 110 675 (6.5) | 97 904 (7.0) | – |
| Sex | Female | 1 100 957 (53.6) | 926 592 (54.1) | 749 022 (53.8) | 365 203 (52.9) |
| Sex | Male | 953 506 (46.4) | 784 873 (45.9) | 642 206 (46.2) | 325 580 (47.1) |
| Ethnicity | White | 1 522 868 (74.1) | 1 293 856 (75.6) | 1 064 331 (76.5) | 539 237 (78.1) |
| Ethnicity | Indian | 35 618 (1.7) | 31 062 (1.8) | 25 454 (1.8) | 11 293 (1.6) |
| Ethnicity | Pakistani | 17 555 (0.9) | 15 588 (0.9) | 12 090 (0.9) | 4388 (0.6) |
| Ethnicity | Bangladeshi | 8138 (0.4) | 7635 (0.4) | 6264 (0.5) | 2076 (0.3) |
| Ethnicity | Other Asian | 17 848 (0.9) | 15 171 (0.9) | 11 890 (0.9) | 5135 (0.7) |
| Ethnicity | Black Caribbean | 22 859 (1.1) | 18 010 (1.1) | 14 102 (1.0) | 5791 (0.8) |
| Ethnicity | Black African | 16 880 (0.8) | 13 530 (0.8) | 9545 (0.7) | 3518 (0.5) |
| Ethnicity | Chinese | 6553 (0.3) | 4835 (0.3) | 3507 (0.3) | 1502 (0.2) |
| Ethnicity | Other ethnic groups | 25 410 (1.2) | 19 778 (1.2) | 14 569 (1.0) | 5832 (0.8) |
| Ethnicity | Ethnicity not recorded | 380 734 (18.5) | 292 000 (17.1) | 229 476 (16.5) | 112 011 (16.2) |
| Region | East Midlands | 46 002 (2.2) | 38 777 (2.3) | 30 526 (2.2) | 16 779 (2.4) |
| Region | East of England | 93 217 (4.5) | 77 645 (4.5) | 64 843 (4.7) | 34 167 (4.9) |
| Region | London | 322 941 (15.7) | 261 176 (15.3) | 204 112 (14.7) | 92 174 (13.3) |
| Region | North East | 47 496 (2.3) | 40 081 (2.3) | 33 271 (2.4) | 15 848 (2.3) |
| Region | North West | 417 970 (20.3) | 354 779 (20.7) | 292 600 (21.0) | 140 099 (20.3) |
| Region | South Central | 283 054 (13.8) | 239 109 (14.0) | 199 347 (14.3) | 102 632 (14.9) |
| Region | South East | 268 594 (13.1) | 220 952 (12.9) | 179 031 (12.9) | 91 516 (13.2) |
| Region | South West | 256 384 (12.5) | 213 037 (12.4) | 169 824 (12.2) | 87 179 (12.6) |
| Region | West Midlands | 237 881 (11.6) | 197 414 (11.5) | 161 606 (11.6) | 81 942 (11.9) |
| Region | Yorkshire & Humber | 80 924 (3.9) | 68 495 (4.0) | 56 068 (4.0) | 28 447 (4.1) |
| Deprivation quintile | 1 (most affluent) | 674 004 (32.8) | 569 701 (33.3) | 471 575 (33.9) | 251 660 (36.4) |
| Deprivation quintile | 2 | 547 862 (26.7) | 456 956 (26.7) | 373 336 (26.8) | 191 172 (27.7) |
| Deprivation quintile | 3 | 385 476 (18.8) | 318 962 (18.6) | 258 842 (18.6) | 123 090 (17.8) |
| Deprivation quintile | 4 | 267 458 (13.0) | 219 941 (12.9) | 175 665 (12.6) | 78 550 (11.4) |
| Deprivation quintile | 5 (most deprived) | 174 280 (8.5) | 141 551 (8.3) | 108 526 (7.8) | 44 651 (6.5) |
| Deprivation quintile | Not recorded | 5383 (0.3) | 4354 (0.3) | 3284 (0.2) | 1660 (0.2) |
| Home category | Neither in care home nor homeless | 2 005 725 (97.6) | 1 665 389 (97.3) | 1 356 313 (97.5) | 682 316 (98.8) |
| Home category | Care home | 47 655 (2.3) | 45 263 (2.6) | 34 352 (2.5) | 8301 (1.2) |
| Home category | Homeless | 1083 (0.1) | 813 (<0.01) | 563 (<0.01) | 166 (<0.01) |
| Household size | 1 person | 875 588 (42.6) | 726 447 (42.4) | 596 361 (42.9) | 285 715 (41.4) |
| Household size | 2 people | 849 357 (41.3) | 721 411 (42.2) | 594 481 (42.7) | 326 499 (47.3) |
| Household size | 3–5 people | 255 089 (12.4) | 199 611 (11.7) | 152 373 (11.0) | 65 031 (9.4) |
| Household size | 6–9 people | 30 961 (1.5) | 24 934 (1.5) | 18 767 (1.3) | 6678 (1.0) |
| Household size | 10 or more | 43 468 (2.1) | 39 062 (2.3) | 29 246 (2.1) | 6860 (1.0) |
| No of health conditions† | 0 | 667 163 (32.5) | 483 507 (28.3) | 566 398 (40.7) | 213 919 (31.0) |
| No of health conditions† | 1 | 786 798 (38.3) | 671 330 (39.2) | 559 648 (40.2) | 281 353 (40.7) |
| No of health conditions† | 2 | 428 751 (20.9) | 393 220 (23.0) | 215 126 (15.5) | 145 583 (21.1) |
| No of health conditions† | 3+ | 171 751 (8.4) | 163 408 (9.5) | 50 056 (3.6) | 49 928 (7.2) |
## Analyses
Descriptive analyses compared the uptake and refusal of the three vaccinations of interest by ethnic group, Townsend deprivation quintiles, household size and individual health conditions. Percentage uptake of each vaccination in individual GPs was plotted to display between-region variations.
Multivariable logistic regression models examined associations between ethnic group, deprivation, household size, health conditions and vaccination uptake and refusal by calculating adjusted OR and their $95\%$ CIs. Clustered robust SEs were used to account for clustering of individuals within GPs. Refusals were evaluated in never-receivers of each vaccine (no uptake). Individual models for each exposure (ethnic group, deprivation, household size, health conditions) and outcome (vaccination uptake and refusal for each vaccine) were fitted separately, allowing for adjustment of confounders: age, sex, geographical region, type of home, smoking status and/or BMI as relevant according to directed acyclic graphs—[1] ethnicity—no adjustments; [2] deprivation—adjusted for age, sex, region, ethnicity, household size; [3] household size—adjusted for age, sex, region, ethnicity, deprivation, [4] health conditions—age, sex, region, ethnicity, deprivation, household size, house type, smoking and BMI (online supplemental figure S1).
**Figure 1:** *Box and whiskers diagrams summarising influenza, pneumococcal and shingles vaccination uptake/refusal rates in practices across different regions in England. The midline of box represents median uptake/refusal rate, lower and upper boundaries of box represent first and third quartile, lower and upper whiskers represent minimum and maximum rates. Each individual dot represents individual practice uptake/refusal rates.*
Missing data for ethnic group ($18.5\%$), BMI ($5.6\%$), deprivation quintiles ($0.3\%$) and smoking status ($1.0\%$) were multiply imputed using chained equations under the missing at random assumption. Five imputations were generated using a single rich imputation model incorporating all outcomes, exposures and confounder covariates. Models were fitted in each of the five imputed datasets with model coefficients and their SEs pooled in accordance with Rubin’s rules.19 We also performed sensitivity analyses of results using complete-case analysis.
In addition, we performed post-hoc interaction analyses to explore potential interactive effects for vaccine uptake between ethnicity and deprivation, household size and number of health conditions.
The reporting of studies using observational rotinely-collected data (RECORD) guidelines were used for reporting.20 Statistical analyses were performed using STATA V.17.0.21
## Patient and public involvement reporting
Two public representatives advised on interest and appropriateness of the research questions, were involved in writing the protocol for the wider study and input on lay-summaries describing the planned study.
## Results
This study included 2 054 463 patients aged 65 years and older registered with 1318 GPs. Characteristics of the study population are shown in table 1 and S1. At least one influenza vaccine was received by 1 711 465 ($83.3\%$) patients, a pneumococcal vaccine by 1 391 228 ($67.7\%$) and a shingles vaccine by 690 783 ($53.4\%$ of over 70s). Figure 1 shows a descriptive overview of the rate of vaccination uptake and refusals by different regions in England at the practice level. For example, the median level of shingles vaccine uptake in London practices was ~$50\%$, compared with ~$60\%$ in East England. Overall, uptake of influenza vaccine (~$80\%$) was the highest among all three vaccine types, followed by pneumococcal vaccine (~$70\%$) and shingles vaccine (~$50\%$) (figure 1).
## Vaccination uptake
Vaccination uptake differed by ethnicity, deprivation, household size and health conditions (figure 1). In multivariable analysis compared with the white population, those from black Caribbean, black African, Chinese and Other ethnic groups showed lower uptake for all three vaccines (figure 2). Influenza vaccination uptake was significantly lower in black Caribbean (OR 0.68, $95\%$ CI 0.64 to 0.71), black African (OR 0.72; $95\%$ CI 0.68 to 0.77), Chinese (OR 0.49; $95\%$ CI 0.45 to 0.53) and ‘other ethnic group’ (OR 0.63; $95\%$ CI 0.60 to 0.65), but there was significantly higher uptake in Indian (OR 1.21; $95\%$ CI 1.14 to 1.28), Pakistani (OR 1.39; $95\%$ CI 1.28 to 1.52) and Bangladeshi (OR 2.68; $95\%$ CI 2.38 to 3.01) ethnic groups compared with the white group.
**Figure 2:** *Associations of ethnicity, deprivation, household size and number of health conditions on influenza, pneumococcal and shingles vaccine uptake. Logistic models for ethnicity, deprivation, household size and health conditions were run separately as each exposure factor required different sets of adjustment variables as informed by DAG evaluation. The following adjustment covariates were included in each of these models as the following: (1) Ethnicity—no adjustment; (2) Deprivation—adjusted for age, sex, region, ethnicity, household size; (3) Household size—adjusted for age, sex, region, ethnicity, deprivation, (4) Health conditions—adjusted for age, sex, region, ethnicity, deprivation, household size, house type, smoking and BMI. BMI, body mass index; DAG, directed acyclic graph,*
There was a similar pattern observed for pneumococcal vaccination uptake: black Caribbean (OR 0.70; $95\%$ CI 0.66 to 0.75), black African (OR 0.56; $95\%$ CI 0.51 to 0.62), Chinese (OR 0.49; $95\%$ CI 0.45 to 0.53), ‘other ethnic group’ (OR 0.58; $95\%$ CI 0.55 to 0.61) and also additionally for other Asian (OR 0.87; $95\%$ CI 0.80 to 0.93). Pneumococcal vaccine uptake was significantly higher only in the Bangladeshi ethnic group (OR 1.46; $95\%$ CI 1.29 to 1.65) compared with the white group. For shingles vaccine uptake, there was significantly lower uptake in all ethnic minority groups except in Indians (OR 0.98; $95\%$ CI 0.91 to 1.05).
For all three vaccines, vaccine uptake was generally lower among the more deprived, with the most deprived having $6\%$–$33\%$ lower odds of vaccine uptake (ORs 0.67–0.94) compared with the most affluent. People in households with two people had $22\%$–$32\%$ higher odds of having a vaccine compared with one-person households. However, the odds were lower in household sizes above three, with people in households of 10 or more people having $17\%$–$63\%$ lower odds of vaccine uptake compared with one-person households.
The uptake of each vaccination was also generally associated with increasing number of health conditions, with asthma being associated with higher uptake of all three vaccines, while AF, CCF, dementia, and severe mental illness were associated with lower uptake of all three vaccines. Individuals with COPD, diabetes and immunosuppression were also associated with higher uptake of both influenza and pneumococcal vaccines, but not shingles vaccine (online supplemental figure S2).
## Vaccination refusals in the unvaccinated
There were consistently significantly higher odds of vaccine refusal among the black Caribbean group compared with the white group for all three vaccines; influenza (OR 1.45; $95\%$ CI 1.34 to 1.56), pneumococcal (OR 1.29; $95\%$ CI 1.14 to 1.46) and shingles (OR 1.35; $95\%$ CI 1.23 to 1.49). Indian, Pakistani, Bangladeshi, other Asian, Black African, Chinese and other ethnic groups were significantly less likely to refuse all three vaccines compared with the white ethnic group, except for Pakistani and Bangladeshi, which showed no significant association with shingles vaccine refusal (figure 3).
**Figure 3:** *Associations of ethnicity, deprivation, household size and number of health conditions on influenza, pneumococcal and shingles vaccine refusal in the unvaccinated population. Logistic models for ethnicity, deprivation, household size and health conditions were run separately as each exposure factor required different sets of adjustment variables as informed by DAG evaluation. The following adjustment covariates were included in each of these models as the following: (1) Ethnicity—no adjustment; (2) Deprivation—adjusted for age, sex, region, ethnicity, household size; (3) Household size—adjusted for age, sex, region, ethnicity, deprivation, (4) Health conditions—adjusted for age, sex, region, ethnicity, deprivation, household size, house type, smoking and BMI. BMI, body mass index; DAG, directed acyclic graph.*
There was a general trend of refusal with increasing deprivation, particularly with shingles vaccine in the two most deprived quintiles (OR 1.21; $95\%$ CI 1.15 to 1.28 and OR 1.23; $95\%$ CI 1.14 to 1.33) (4th and 5th deprivation quintiles, respectively). Higher household size was associated with lower odds of refusal of all three vaccines in households of 3+ people and more (figure 3).
In unvaccinated individuals with three or more health conditions, the odds of refusal were: influenza vaccine (OR 10.29; $95\%$ CI 7.38 to 14.37), pneumococcal vaccine (OR 2.55; $95\%$ CI 2.24 to 2.90), shingles vaccine (1.60; $95\%$ CI 1.48 to 1.73). Individuals with type 2 diabetes consistently showed higher vaccine refusal for all three vaccines and individuals with COPD was also associated with higher refusal for influenza and pneumococcal vaccines (online supplemental figure S3).
## Additional analyses
Further, we explored interactions for vaccine uptake between ethnicity and deprivation, house size and number of health conditions. First, results suggested that individuals from certain ethnic minority groups who were more deprived could be more likely to receive a vaccine, particularly Bangladeshi and Black African (online supplemental figure S4). Second, across all three vaccines evaluated, Bangladeshi individuals living in larger households could be more likely to receive a vaccine (online supplemental figure S5). Third, vaccine uptake was generally more likely in individuals with higher number of health conditions, although the magnitude of effect varied slightly across different ethnic groups (online supplemental figure S6).
Finally, we performed sensitivity analyses to evaluate associations of vaccine uptake and refusal using complete-case analyses. In this analysis, we excluded individuals with missing information on covariates that is, ethnicity, deprivation, BMI and smoking. Results in the online supplemental figures S7 and 8 showed that estimates were comparable with the multiply imputed analysis presented as our main findings above.
## Summary
In this study, we observed generally lower uptake of influenza, pneumococcal and shingles vaccinations in particular ethnic minority groups and deprived populations. Black Caribbean, black African, Chinese and other ethnic groups consistently showed lower uptake of all three vaccines studied compared with the white ethnic group. In the unvaccinated population, the black Caribbean ethnic group consistently showed increased odds of refusal for all three vaccines. More deprived populations also showed lower vaccine uptake with higher recorded refusals in the unvaccinated. Household sizes above three persons were associated with lower vaccine uptake, but were not associated with higher refusal. Further, a lower number of pre-existing health conditions was generally associated with lower odds of vaccine uptake, although this was not reflected in terms of higher odds of refusal.
## Comparison with existing literature
Our observations that influenza vaccination uptake is inversely correlated with deprivation and varies across ethnic groups build on results from a recent study of adults between 2011 and 2016 using the CPRD database.7 This study analysed seasonal influenza vaccination uptake across five ‘seasons’ and similarly found that in the over 65s, black individuals were significantly less likely than white individuals to receive this vaccination. However, our study finds that South Asians may be more likely to have higher uptake of influenza vaccine, which may warrant further qualitative study to examine potential socioeconomic and behavioural factors driving this observation. Our examination of three vaccinations within a larger sample size (over 2 million vs 611 000), a more granular categorisation of ethnic groups (9 vs 4) and regions (10 vs 4), improved handling of missing data, and our analysis of vaccination refusals in the unvaccinated substantially improves our understanding of these complex public health behaviours. Our results showed that although four ethnic minority groups (black Caribbean, black African, Chinese and other ethnic group) had lower uptake of influenza vaccine, only the black Caribbean group showed increased odds of refusal among the unvaccinated.
We also found lower vaccine uptake in household sizes above three persons, although they also showed lower refusals in the unvaccinated population. This suggests that lower vaccine uptake in larger households could be driven by barriers to vaccine uptake other than due to refusal alone. A study in Hong Kong showed that vaccine uptake in the elderly living with younger family members was lower compared with elderly individuals living alone, or living with other elderly household members.6 This calls for further ethnographic research to explore social and household characteristics including age structure of household members and its potential association with vaccine uptake in the elderly in England.
Higher uptake of influenza and pneumococcal vaccinations in individuals with asthma, COPD, diabetes and immunosuppression could be related to clinical guidelines where individuals in these clinical risk groups would be more likely to be offered a vaccine by their healthcare providers.22 23 On the contrary, lower vaccine uptake in those with fewer health conditions could potentially be attributable to reduced contact with health services in the healthier population and hence, reduced likelihood to receive ‘opportunistic’ vaccination offers. Despite that, it is worth noting that our study also found that in the unvaccinated population there remains significant refusal in those with type 2 diabetes and COPD. Possibly relevant factors could be resistance to lifestyle and behaviour changes, in which individuals with diabetes and COPD who might be more likely to have unhealthy lifestyles, for example, smoking,24 25 might also be less receptive to health interventions, i.e. vaccines. However, this finding needs confirmation in other studies. In addition, interaction analyses in our study showed that certain ethnic minority groups such as Bangladeshis who were more deprived and living in larger households were more likely to receive a vaccine. This could potentially be due to availability of outreach programmes organised by local communities and GPs in these areas to create awareness and provide health education.26 27 Vaccine hesitancy findings from this study may also be relevant to ongoing COVID-19 vaccine hesitancy in the population. In a population study in older adults using National Immunisation Management System in the England, UK, it has been similarly shown that black African and black Caribbean and more deprived populations were less likely to receive COVID-19 vaccine.28 These similarities in findings across different vaccines suggest possible shared drivers of vaccine hesitancy, which might help inform future public health strategies for equitable implementation of vaccination programmes in general.
## Strengths and limitations
Use of the QResearch database offered a population-representative study sample with accurately coded data, enabling capture of vaccinations occurring outside GP (such as in pharmacies), as well as recorded invitations to vaccination sent by GPs and patient refusals. This permitted a robust evaluation of not only uptake, but also possible contributory mechanisms leading to uptake behaviours. Limitations include the lack of recording of variables such as religion, personal beliefs and reasons for refusal that predicate vaccine hesitancy in our sample. Further, our dataset also did not capture literacy levels, language barriers, access and education status, and hence were not able to evaluate the impact of these socioeconomic factors on vaccination uptake and refusal. These could be important factors influencing complex decision making and behavioural aspects, and hence would warrant further qualitative and ethnography studies. Classification of vaccination-related endpoints was reliant on individual practitioners using Read and SNOMED codes on the EMIS software system; however, as GP surgeries are financially incentivised through ‘Quality Outcome Framework’ payments to record vaccination services and we used an appropriately wide range of codes in our endpoint definitions, the risk of misclassification may be low.
## Implications for research and practice
Two key principles in health inequalities are Tudor-Hart’s inverse care law,29 where service provision is inversely proportional to the need for it, and the inverse equity hypothesis, which posits that new healthcare interventions are most likely to be taken up by those in less need and thus exacerbate pre-existing inequality in the short term. Our study may help inform policy makers regarding reducing inequity in the uptake of the studied vaccines, and tailor public health messaging to diverse communities. Elucidating the extent to which ethnic patterns in vaccine refusal are driven by cultural perceptions, institutional mistrust, variation in penetrance of misinformation and structural barriers for example, transport, language and occupational barriers in different ethnic groups requires further study in robust surveys and qualitative research. This may inform tailoring of information dissemination strategies and misinformation countermeasures to specific groups and geographical areas. Furthermore, judicious, longitudinal monitoring of the uptake and refusal rates of vaccines in different ethnic and social groups should enable real-time assessment of developing inequalities, which may inform adaptive public health strategies. Data from this may help develop strategies for increasing uptake in these groups including developing information about vaccines in different languages for use by community leaders, faith groups, local healthcare providers and community champions.30
## Conclusions
Certain ethnic minority, deprived populations, large households and healthier individuals were less likely to receive a vaccine, although in the unvaccinated population, higher odds of refusal were only associated with ethnicity and deprivation, but not larger households nor comorbidities. Understanding these associations may inform tailored public health messaging to different communities for equitable implementation of vaccination programmes.
## Data availability statement
Data may be obtained from a third party and are not publicly available. To guarantee the confidentiality of personal and health information, only the authors have had access to the data during the study in accordance with the relevant licence agreements. Access to QResearch data is according to the information on the QResearch website (www.qresearch.org).
## Patient consent for publication
Not applicable.
## References
1. Bartleson JM, Radenkovic D, Covarrubias AJ. **SARS-cov-2, COVID-19 and the ageing immune system**. *Nat Aging* (2021.0) **1** 769-82. DOI: 10.1038/s43587-021-00114-7
2. **Pneumococcal polysaccharide vaccine (PPV) coverage report, england, april 2019 to march 2020**
3. **Seasonal influenza vaccine uptake in GP patients: winter season 2020 to 2021**
4. England PH. **Shingles coverage report (adults eligible from april to september 2020 andvaccinated to end of december 2020): englandquarter 2 report of financial year 2020 to 2021**
5. Europe W. **Evaluation of seasonal influenza vaccination policies and coverage in the WHO european region**
6. Chan DPC, Wong NS, Wong ELY. **Household characteristics and influenza vaccination uptake in the community-dwelling elderly: a cross-sectional study**. *Prev Med Rep* (2015.0) **2** 803-8. DOI: 10.1016/j.pmedr.2015.09.002
7. Loiacono MM, Mahmud SM, Chit A. **Patient and practice level factors associated with seasonal influenza vaccine uptake among at-risk adults in england, 2011 to 2016: an age-stratified retrospective cohort study**. *Vaccine X* (2020.0) **4** 100054. DOI: 10.1016/j.jvacx.2020.100054
8. Coupland C, Harcourt S, Vinogradova Y. **Inequalities in uptake of influenza vaccine by deprivation and risk group: time trends analysis**. *Vaccine* (2007.0) **25** 7363-71. DOI: 10.1016/j.vaccine.2007.08.032
9. Fisher H, Trotter CL, Audrey S. **Inequalities in the uptake of human papillomavirus vaccination: a systematic review and meta-analysis**. *Int J Epidemiol* (2013.0) **42** 896-908. DOI: 10.1093/ije/dyt049
10. House T, Keeling MJ. **Household structure and infectious disease transmission**. *Epidemiol Infect* (2009.0) **137** 654-61. DOI: 10.1017/S0950268808001416
11. Rockliffe L, Waller J, Marlow LAV. **Role of ethnicity in human papillomavirus vaccination uptake: a cross-sectional study of girls from ethnic minority groups attending London schools**. *BMJ Open* (2017.0) **7**. DOI: 10.1136/bmjopen-2016-014527
12. Sarker AR, Akram R, Ali N. **Coverage and factors associated with full immunisation among children aged 12-59 months in Bangladesh: insights from the nationwide cross-sectional demographic and health survey**. *BMJ Open* (2019.0) **9**. DOI: 10.1136/bmjopen-2018-028020
13. **Population of england and wales**
14. **Patients registered at a GP practice**
15. **Vaccination against shingles guide**. (2022.0)
16. **QResearch web**
17. 17Townsend P and D. N. The black report. Penguin, 1982.. *The black report* (1982.0)
18. **2011 UK townsend deprivation scores - townsend deprivation scores report**. (2011.0)
19. Rubin DB. *Multiple Imputation for Non-response in Surveys* (1987.0). DOI: 10.1002/9780470316696
20. Benchimol EI, Smeeth L, Guttmann A. **The reporting of studies conducted using observational routinely-collected health data (record) statement**. *PLoS Med* (2015.0) **12**. DOI: 10.1371/journal.pmed.1001885
21. **Stata statistical software: release 17**
22. **Green book - pneumococcal**. (2018.0)
23. **Seasonal influenza vaccine uptake in GP patients: winter season 2020 to 2021**
24. Bhatt SP, Kim Y-I, Harrington KF. **Smoking duration alone provides stronger risk estimates of chronic obstructive pulmonary disease than pack-years**. *Thorax* (2018.0) **73** 414-21. DOI: 10.1136/thoraxjnl-2017-210722
25. Śliwińska-Mossoń M, Milnerowicz H. **The impact of smoking on the development of diabetes and its complications**. *Diab Vasc Dis Res* (2017.0) **14** 265-76. DOI: 10.1177/1479164117701876
26. Abbott S, Riga M. **Delivering services to the Bangladeshi community: the views of healthcare professionals in East London**. *Public Health* (2007.0) **121** 935-41. DOI: 10.1016/j.puhe.2007.04.014
27. Latif S, Ahmed I, Amin MS. **Exploring the potential impact of health promotion videos as a low cost intervention to reduce health inequalities: a pilot before and after study on bangladeshis in inner-city London**. *London Journal of Primary Care* (2016.0) **8** 66-71. DOI: 10.1080/17571472.2016.1208382
28. Nafilyan V, Dolby T, Razieh C. **Sociodemographic inequality in COVID-19 vaccination coverage among elderly adults in England: a national linked data study**. *BMJ Open* (2021.0) **11**. DOI: 10.1136/bmjopen-2021-053402
29. Victora CG, Joseph G, Silva ICM. **The inverse equity hypothesis: analyses of institutional deliveries in 286 national surveys**. *Am J Public Health* (2018.0) **108** 464-71. DOI: 10.2105/AJPH.2017.304277
30. Hanif W, Ali SN, Patel K. **Cultural competence in covid-19 vaccine rollout**. *BMJ* (2020.0). DOI: 10.1136/bmj.m4845
|
---
title: 'Effects of television viewing on brain structures and risk of dementia in
the elderly: Longitudinal analyses'
authors:
- Hikaru Takeuchi
- Ryuta Kawashima
journal: Frontiers in Neuroscience
year: 2023
pmcid: PMC10030518
doi: 10.3389/fnins.2023.984919
license: CC BY 4.0
---
# Effects of television viewing on brain structures and risk of dementia in the elderly: Longitudinal analyses
## Abstract
### Introduction
TV viewing in the elderly and in children is associated with subsequent greater decline of various cognitive functions including verbal working memory, but results of its association with subsequent risk of dementia were divided.
### Methods
In this longitudinal cohort study of UK Biobank, we investigated the associations of TV viewing length with subsequent risk of dementia and longitudinal changes of brain structural measures after corrections of a wide range of potential confounders.
### Results
Our results showed longer TV viewing was associated with increased risk of subsequent onset of dementia, as well as subsequent greater decline in intracellular volume fraction (ICVF) in the extensive areas of right lateral temporal cortex and the right medial temporal cortex, in the area around the left middle and inferior temporal cortex as well as the left fusiform gyrus, and the area adjacent to the left inferior frontal gyrus, and left insula.
### Discussion
These results may suggest prolonged TV viewing was associated with decline in density of neurites (axon, dendrites) in areas particularly implicated in language, communication, and memory, which are altered in dementia.
## Introduction
Television is an ubiquitous tool of modern life. However, its adverse effects on cognitive mechanisms have been well investigated. Longitudinal observational or intervention studies of children have shown that longer TV viewing is associated with subsequent declines in attention and cognitive ability, especially for verbal abilities including verbal working memory and verbal IQ (Gadberry, 1981; Sergeant et al., 2002; Zimmerman and Christakis, 2005; Landhuis et al., 2007; Takeuchi et al., 2015). In addition, recent large cohort studies of the elderly have linked longer TV viewing habits to subsequent declines in cognitive functions, including executive functions and visual and verbal short-term memory (Hoang et al., 2016; Bakrania et al., 2018; Fancourt and Steptoe, 2019). In the elderly, longer TV viewing has also been associated with an increased risk of cardiovascular disease (Grøntved and Hu, 2011). Moreover, our previous neuroimaging study involving children revealed that longer TV viewing is associated with greater increases in regional gray matter and lower developmental decreases in regional gray matter volume (rGMV) decreases in the fronto-polar areas (Takeuchi et al., 2015). In addition, ever since the association between TV viewing and dementia risk has been suggested (Aronson, 1993), a few studies investigated this association using small sample size and the results are divided and one reported the positive association and another failed to find the association (Lindstrom et al., 2005; Fajersztajn et al., 2021).
Despite these studies, the following issues have not been investigated; (a) associations between the length of TV viewing and subsequent changes in gray and white matter structural volume in the elderly, (b) the associations between the length of TV viewing and subsequent changes in brain microstructure properties of the brain, and (c) the associations between the length of TV and risk of dementia with the large sample size and corrections of a wide range of potential confounding variables. This study thus aims to investigate both of these issues. As longer TV viewing has been associated with greater declines in cognitive function of the elderly, we reasoned that this is likely to also increase the risk of dementia. Similarly, we predict that longer TV viewing is associated with lower brain volume, lower fractional anisotropy, and mean diffusivity, and other microstructural properties that are observed in aging and dementia in the fronto-polar areas, language and memory related areas. This is because TV viewing is consistently associated with declines of verbal functions and memory problems in the elderly (Takeuchi et al., 2015; Bakrania et al., 2018), and these neuroimaging characteristics characterize lower cognitive functions and dementia, and our previous study found effects of length of TV viewing in the fronto-polar areas in children (Nir et al., 2013; Takeuchi et al., 2015).
Identifying and describing the effects of TV viewing on cognitive decline, neural properties, and dementia development in the elderly is socially and scientifically important. For this study, we utilized data from the UK Biobank. Longitudinal design studies enable generation of associations between TV viewing and its subsequent impact on neurocognitive and neurological mechanisms in the aging brain.
To measure microstructural properties of the brain, we used metrics related to diffusion tensor imaging (DTI) and neurite orientation dispersion and density imaging (NODDI) in white matter (Edwards et al., 2017). Mean diffusivity (MD) describes the amount of water molecule diffusion regardless of direction, while axial diffusivity (AD) represents water molecule diffusion parallel to the tract within the voxel of interest and radial diffusivity (RD) measures the magnitude of water diffusion perpendicular to that tract. In addition, fractional anisotropy specifies the level of anisotropy of water diffusion. The intracellular volume fraction (ICVF) is used to represent neurite compartment density, as verified by histology in animal experiments (Sepehrband et al., 2015) while the isotropic volume fraction (ISOVF) reports the extracellular free water diffusion as well as the interstitial and cerebrospinal fluids (CSF). Finally, the orientation dispersion index quantifies the spread of fibers within an intracellular compartment. By combining these metrics with information on white matter volume, white matter changes can be comprehensively evaluated.
## Participants
For our study, we used data from the UK Biobank, which was obtained from a prospective cohort study of a middle-aged population in the United Kingdom and the procedures of which have been described elsewhere.1 Approval for these experiments was obtained from the North-West Multi-center Research Ethics Committee and written informed consent was obtained from each participant. Briefly, the participants went to one of 22 assessment centers throughout UK for data collection, with baseline data obtained from 502,505 participants. Our study included data for this cohort obtained at the first assessment visit (2006–2010), the first imaging data collection visit, which corresponded to the third assessment visit (2014–present), and the follow-up visit for imaging data collection, corresponding to the fourth assessment visit (2019–present). The schema of the study was presented in Figure 1. Sample of each analysis consists of those who have all the effective dependent and independent variables in the analyses.
**FIGURE 1:** *Study scheme.*
## Assessment of time spent watching TV
Time spent watching TV was assessed by the following question: “*In a* typical day, how many hours do you spend watching TV?”. Answers were: “less than an hour a day” or any integer value between 0 and 24, with an answer of “less than an hour a day” was regarded as 0 h. In addition, Responses of more than 6 h were treated as 6 h in the analysis in line with those of recent representative studies of the effects of length of TV viewing (Hamer et al., 2017; Celis-Morales et al., 2018).
## Sociodemographic and lifestyle measurements used as covariates
Self-reported gender data was used. From the database, the neighborhood-level socioeconomic status at recruitment (cov1), education level at recruitment (cov2), household income (cov3), current employment status (cov4), metabolic equivalent of task hours (MET) (cov5), number in household (cov6), body mass index (BMI) (cov7), self-reported health status (cov8), and sleep duration (cov9) were extracted or calculated and included as covariates. For additional details, refer to the Supplementary methods section.
## Structural MRI acquisition and pre-processing for volumetric analyses
For the UK Biobank study cohort, MRI imaging data was obtained during the third and fourth assessment visits. Images were obtained from 3 imaging centers equipped with identical scanners (Siemens Skyra 3T running VD13A SP4 with a Siemens 32-channel RF receive head coil, Munich, Germany).
T1-weighted structural images were used for voxel-based morphometry analyses. Images were segmented and segmented gray matter and white matter were normalized using the diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) procedure, modulated, smoothed (8 mm full width at half maximum (FWHM)], and the resultant maps representing rGMV and regional white matter volume (rWMV) analyzed. Finally, the signal change in rGMV and rWMV between pre-(3rd visit) and post-(4th visit) images was computed at each voxel for each participant. In this computation, we included only voxels that showed GMV or rWMV values >0.10 in both pre- and post-images to avoid possible partial volume effects around the borders between different tissues. The resulting maps representing the rGMV and rWMV change between the pre- and post-MRI experiments (rGMV post–rGMV pre, rWMV post–rWMV pre) were then forwarded to the second level analysis, described in the next section. For details, see Supplementary methods.
For microstructure analysis, we used DTI and NODDI measurements released by the UK Biobank Imaging Study. Normalization was performed based on a previously validated protocol (Takeuchi et al., 2013). Briefly, diffusion images using the information of MD and FA, and modified DARTEL procedure which took account the FA signal distribution within white matter areas (to align images and tracts within white matter areas), were used to normalize all DTI and NODDI images as well as regional gray matter density (rGMD), regional white matte density (rWMD), and regional cerebrospinal fluid maps (rCSFD).
From the pre- and post-normalized images of the normalized MD, AD, RD, ICVF, ISOVF, OD maps, areas not strongly likely to be gray or white matter in our custom template (defined by “gray matter tissue probability + white matter tissue probability < 0.99”) were removed. Then, from the pre- and post-intervention normalized images of normalized FA map, areas not strongly likely to be white matter in our custom template (defined by “white matter tissue probability < 0.99”) were removed. Subsequently, normalized MD, AD, RD, ICVF, ISOVF, OD as well as FA images were smoothed by convolution with an isotropic Gaussian kernel of 8- and 6-mm full-width at half maximum, respectively. The images representing the baseline to the follow-up change of normalized smoothed images [e.g., ICVF (4th visit) − ICVF (3rd visit)] were used for the second level analyses described below.
For more details on these procedures, refer to the Supplementary methods. We did not use Tract Based Spatial Statistics (Smith et al., 2006), for many reasons, but the one of the reasons is that as far as we checked the values of diffusion measures in each tract in UK *Biobank data* that was generated by TBSS were generating apparent too many outliers, and we need to take the procedures we are used to correct these problems. The pre-processing method we used is shown to have validity, effectively solved all problems raised by voxel-based DTI measures raised by Smith et al. [ 2006], and generate analytical results that are similar to those generated by TBSS (Takeuchi et al., 2013).
## Psychological and non-whole brain imaging data analyses
Psychological and non-whole brain imaging data were analyzed using Predictive Analysis Software, version 22.0.0 (SPSS Inc., Chicago, IL, USA; 2010).
Cox proportional hazards models were used to examine the relationships between TV viewing length and dementia of all causes, as previously described (Lourida et al., 2019). All-cause dementia was ascertained using hospital inpatient records and linkage to death register data. This is a widely taken method in UK Biobank studies involving dementia (Lourida et al., 2019). For more details, see Supplementary methods. Participants with (a) self-reported dementia or Alzheimer’s disease or cognitive impairment without a diagnosis of all-cause dementia in either hospital inpatient records or death register data, (b) subjects already diagnosed with dementia at baseline or within 5 years after baseline, (c) those who died within 5 years after baseline, and (d) those with visuospatial memory performance lower than 2SD were excluded from the analyses The time scale considered spanned from the time of the first assessment visit and until 30 September 2021. Covariates were sex, age at the first assessment visit, values of cov1–cov9 at the first assessment visit, length of TV viewing at the first assessment visit, and visuospatial memory performance at the first assessment visit (for details, see Supplementary methods, fluid intelligence data was not available for a majority of subjects). For these analyses, we conducted both of analyses that treated the length of TV viewing as a continuous variable as well as analyses that treated the length of TV viewing as a categorical variable and separated 0–1 h, 2–3 h, 4–5 h, and 6 h or more, consistent with previous work (Celis-Morales et al., 2018). Although, we included a wide range of variables in this study, this study has a rich sample size, which mitigates the problem of overfitting (Riley et al., 2020).
## Imaging data analysis
Statistical analyses of imaging data were performed with SPM12.
Longitudinal whole-brain multiple regression analyses were employed to look for associations between TV viewing length at the third assessment visit and pre-(3rd assessment visit) to post-(4th assessment) scan changes of brain images [rGMV, rWMV, and DTI (FA, MD, AD, and RD) and NODDI images (ICVF, ISOVF, and OD)], with imaging data available from the third and fourth assessment visits. For rGMV and rWMV analyses, only voxels with a signal intensity of >0.05 for maps of subjects whose images were used to create the template were included for whole brain analyses. The analyses of DTI and NODDI maps were limited areas of the masks that were created above (white matter mask in the FA analysis and gray + white matter mask in other analyses).
In all imaging analyses, the independent variables were sex, age at the third assessment visit, the number of interval days between the third and fourth assessment visits, values of cov1–cov9 at the third assessment visit (except for cov1 and cov2, which refer to values at recruitment), head size ratio at the third assessment visit (calculated using UK Biobank output, in UK Biobank, normalization for head size is done by using T1-based “headsize scaling factor,” data id = 25,000, which is scaling factor estimated when transforming from native to standard space), and the length of TV viewing at the third assessment visit. We did not control the scanner site as was the case with the representative MRI studies of UK Biobank (Miller et al., 2016). There are more than 2 sites and it is difficult to model the effects in whole brain analyses. The time-lapse changes and site-specific effect may exist even when the scanners and parameters are identical, but there are no reasons to assume, those will confound the associations between TV viewing and imaging outcomes.
A multiple comparison correction was performed using threshold-free cluster enhancement (TFCE) (Smith and Nichols, 2009) with randomized (5,000 permutations) non-parametric testing using the TFCE toolbox.2 We applied a threshold of family-wise error corrected at $P \leq 0.05.$
## Basic data
Baseline psychological data for all participants is provided in Supplementary Table 1. TV viewing length at baseline was not associated with the covariates used in this study at the level greater than | r| > 0.4 at baseline. Among other covariates, there were no correlations among covariates that are greater than | r| < 0.4, except in the case of the association of age and current employment (r = −0.54). These results erase the concern of multicollinearity among psychological covariates. Note the associations between longer TV viewing length and longitudinal decline of non-verbal fluid reasoning and short-term numerical memory in UK Biobank have been previously reported (Bakrania et al., 2018).
## Prospective analysis of dementia
Among the data of 502,505 participants in the present project, 121 participants had dementia record only through self-report and among remaining participants, 109 participants had dementia record that are diagnosed before the baseline date. Among the rest, 750 participants had dementia records diagnosed within 5 years after baseline, and 8,462 had died for other reasons during this period. Among the rest, after excluding who does not have data of any one of variables in the analysis, a total of 373,345 participants were included in our analysis. Among these patients, 4,086 cases of incident all-cause dementia were observed. Cox proportional hazard models in which TV viewing lengths were divided into four categories variables (0–1 h, 2–3 h, 4–5 h, 6 + h) revealed the overall group differences and compared with subjects with TV viewing lengths of 0–1 h at baseline, subjects with TV viewing length of 2–3 h, those of 4–5 h, those of 6 h or more. Further, compared with subjects with TV viewing lengths of 2–3 h at baseline, subjects with TV viewing length of 6 h or more showed higher risk. And compared with subjects with TV viewing length of 4–5 h at baseline, subjects with TV viewing length of 6 h or more showed higher risk. Statistical values and number of participants and cases in each group were presented in Figure 2.
**FIGURE 2:** *Statistical values and hazard ratios (HRs) with 95% confidence intervals (95% CIs) for the associations between length of TV viewing and all-cause dementia incidence from the main analyses and sub-analyses. Results of analyses that treated the TV viewing length as the continuous variable and analyses that treated the TV viewing length as the categorical variable were presented. P-values of each analysis, adjusted HR, cases and the participant of each category were presented.*
Even if we use continuous variable for TV viewing length as are the cases of multiple regression analyses, the results showed the robust associations of TV viewing length with the subsequent risk of incident dementia [$$p \leq 1.4$$×10–5, increasing length of TV viewing by 1 h is associated with HR of 1.049 (CI: 1.027–1.072)]. These are not effects of group of suprathreshold and although we used 6 h as the longest TV viewing length based on the previous study and the fact that there are few subjects with longer TV viewing, using 8 or 10 h as the longest TV viewing also showed the association with subsequent risk of dementia [8 h: $$p \leq 3.8$$ × 10–7, increasing length of TV viewing by 1 h is associated with HR of 1.053 [CI: 1.032–1.074; 10: $$p \leq 1.3$$ × 10–7, increasing length of TV viewing by 1 h is associated with HR of 1.053 (CI: 1.033–1.074)].
We then conducted analyses limited to males and analyses limited to females to determine whether there is a difference in the relationship between TV viewing length and risk of dementia in males and females. When TV viewing length was treated as a continuous variable, there was a significant relationship between TV viewing length and risk of all cause dementia in both the male-only and female-only analyses; when TV viewing time was treated as a categorical variable, there were significant overall group differences only in the male-only analysis. However, a similar risk trend was observed in the analysis of women only. The statistical values were presented in Figure 2.
To resolve additional concerns that this association only reflects that prolonged TV viewing is a symptom of undiagnosed dementia or a reflection of serious medical and cardiovascular conditions, we further added analyses that included systolic blood pressure, tobacco smoking level, loneliness, baseline heart attack, angina, stroke, diabetes, cancer, other serious medical conditions, as covariates. This additional analysis also revealed the significant associations of TV viewing and subsequent onset of dementia and the effect size was comparable. The details were provided in Supplementary methods and Figure 2.
To further investigate the impact of loneliness, we stratified subjects who said they often felt lonely at baseline and those who did not and conducted analyses with same conditions as in the main analysis. The results showed similar effect sizes for both analyses, with the presence of significant overall group differences and significant continuous variable effects. The details were provided in Figure 2.
## Longitudinal brain imaging analysis
For brain imaging data analysis, using the data from the third and fourth assessment visits was used. A total of 2,448 and 2,406 participants were included in our analyses for volumetric analyses and microstructural property analyses (a part of entire UK Biobank projects’ participants participated in the MRI experiments of third and fourth assessments). The sample characteristics of the 2,448 subjects in volumetric analyses were provided in Table 1.
**TABLE 1**
| Unnamed: 0 | No incident dementia (n = 369,259) | Incident dementia (n = 4086) |
| --- | --- | --- |
| | | Mean |
| Age | 55.78(8.06) | 63.97(4.83) |
| Townsend deprivation index | –1.44(2.99) | –1.14(3.19) |
| Education length | 14.43(5.05) | 12.91(5.23) |
| BMI | 27.35(4.73) | 27.81(4.86) |
| MET | 31.76(35.53) | 32.43(38.29) |
| Sleep length | 7.15(1.03) | 7.21(1.22) |
| Visuospatial memory (errors) | 3.69(2.4) | 4.39(2.51) |
| Systolic BP | 137.17(18.43) | 144.04(19.19) |
| | | Number |
| TV viewing length | TV viewing length | TV viewing length |
| (a) 0–1 h | 80,106(21.7%) | 505(12.4%) |
| (b) 2–3 h | 189,697(51.4%) | 1,838(45%) |
| (c) 4–5 h | 82,186(22.3%) | 1,309(32%) |
| (d) 6h- | 17,270(4.7%) | 434(10.6%) |
| Male number | 173,437(47%) | 2,331(57%) |
| Household income | Household income | Household income |
| (a) Less than £18,000 | 75,786(20.5%) | 1,748(42.8%) |
| (b) £18,000 to £30,999 | 92,650(25.1%) | 1,280(31.3%) |
| (c) £31,000 to £5,1999 | 99,601(27%) | 670(16.4%) |
| (d) £52,000 to £100,000 | 79,807(21.6%) | 316(7.7%) |
| (e) Greater than £100,000 | 21,415(5.8%) | 72(1.8%) |
| Currently employed | 231,480(62.7%) | 1,010(24.7%) |
| Household number | Household number | Household number |
| (a) 1 | 68,506(18.6%) | 1,068(26.1%) |
| (b) 2 | 166,959(45.2%) | 2,432(59.5%) |
| (c) 3 | 59,026(16%) | 375(9.2%) |
| (d) 4≤ | 74,768(20.2%) | 211(5.2%) |
| Overall health (4 levels) | Overall health (4 levels) | Overall health (4 levels) |
| (a) Poor | 13,425(3.6%) | 365(8.9%) |
| (b) Fair | 72,264(19.6%) | 1,158(28.3%) |
| (c) Good | 217,800(59%) | 2,127(52.1%) |
| (d) Excellent | 65,770(17.8%) | 436(10.7%) |
| Current smoking level (3 levels) | Current smoking level (3 levels) | Current smoking level (3 levels) |
| (a) No | 331,522(89.8%) | 3,646(89.2%) |
| (b) Only occasionally | 10,310(2.8%) | 88(2.2%) |
| (c) On most or all days | 27,321(7.4%) | 350(8.6%) |
| Often feel lonely | Often feel lonely | Often feel lonely |
| (a) No | 299,972(81.2%) | 3,192(78.1%) |
| (b) Yes | 64,631(17.5%) | 829(20.3%) |
| Diabetes* | Diabetes* | Diabetes* |
| (X) | 351,899(95.3%) | 3,543(86.7%) |
| (O) | 16,718(4.5%) | 536(13.1%) |
| Heart attack* | Heart attack* | Heart attack* |
| (X) | 361,482(97.9%) | 3,827(93.7%) |
| (O) | 7,399(2%) | 256(6.3%) |
| Angina* | Angina* | Angina* |
| (X) | 358,913(97.2%) | 3,688(90.3%) |
| (O) | 9,968(2.7%) | 395(9.7%) |
| Stroke* | Stroke* | Stroke* |
| (X) | 364,330(98.7%) | 3,881(95%) |
| (O) | 4,551(1.2%) | 202(4.9%) |
| Cancer* | Cancer* | Cancer* |
| (X) | 340,670(92.3%) | 3,679(90%) |
| (O) | 28,583(7.7%) | 405(9.9%) |
| Other serious medical conditions* | Other serious medical conditions* | Other serious medical conditions* |
| (X) | 293,304(79.4%) | 2,696(66%) |
| (O) | 70,525(19.1%) | 1,312(32.1%) |
Whole brain multiple regression analyses using VBM analysis revealed there were no significant associations between TV viewing length at the third assessment visit and longitudinal changes in rGMV as well as rWMV from the third to the fourth assessment visit. There was a trend of positive association between TV viewing and longitudinal change in rGMV in the area of the ventromedial prefrontal cortex (x = −18, $y = 36$, z = −18, $t = 4.24$, cluster size = 1,269 mm3), which is close to the area where significant positive associations between TV viewing length at baseline and longitudinal change in rGMV was observed in our previous study of children (Takeuchi et al., 2015).
Whole brain multiple regression analyses involving DTI and NODDI analyses revealed significant results only in the analysis of ICVF. There were significant positive associations between TV viewing length at the third assessment visit and longitudinal changes in ICVF between from third to the fourth assessment visit in (a) the anatomical extensive cluster that spread in the areas involving the right amygdala, right hippocampus, right parahippocampal gyrus, right fusiform gyrus, right occipital lobe, right lateral temporal gyrus, right temporal pole, and right sagittal stratum and right stria terminalis, (b) an anatomical cluster that spread in the left inferior and middle temporal gyrus, and left fusiform gyrus, (c) an anatomical cluster that spread in the right Heschl gyrus, right insula, right putamen, right *Rolandic operculum* and right internal capsules, and (d) anatomical clusters that spread in the areas of left inferior frontal gyrus, left insula, and left anterior corona radiata (Figure 3).
**FIGURE 3:** *Associations between TV viewing length and longitudinal changes in intracellular volume fraction (ICVF). (A,B) Results are overlaid on a “single subject” T1 image from SPM. Results were obtained using a threshold of threshold-free cluster enhancement (TFCE) of p < 0.05 based on 5,000 permutations. The color represents the strength of the t-value. The color bar represents the TFCE score. It reflects both voxel’s height and the sum of the spatially contiguous voxels supporting it; therefore, it reflects both the strength and extent of effects. Profiles of imaging values at pre and post scans (which correspond to the third and the fourth assessment visits) in the significant cluster of the extensive cluster around the right temporal gyrus (C), and the cluster in the white matter area of the left inferior frontal gyrus (D). *p < 0.05, **p < 0.01, ***p < 0.001, in the t-tests comparing the raw pre to post differences of ICVF in the significant clusters between groups.*
Length of TV viewing was weakly but robustly negatively correlated with total brain volume that is normalized for head size at the third visit in this cohort (UK *Biobank data* field 25,009) after correction of age, sex and covariates 1–9 which were described in section “Materials and methods” ($$p \leq 1.8$$ × 10–5, $t = 4.292$, standardized beta = 0.020, $$n = 35949$$). Aside from the direction of causality of this cross-sectional association, we then investigate the impact of this effect (i.e., the impact of brain atrophy at the baseline (third visit) of brain imaging acquisition) on the above significant results of ICVF. We examined how inclusion of total brain volume that is normalized for head size at the third visit, changes the standardized beta of associations between TV viewing length and mean ICF values of significant clusters in the multiple regression analysis in addition to corrections of covariates of the main analysis. The results are presented in Table 2. In all significant clusters, the standardized beta were affected little by the inclusion of this covariate.
**TABLE 2**
| Included gray matter areas* (number of significant voxels in left and right side of each anatomical area) | Included large bundles** (number of significant voxels in left and right side of each anatomical area) | x | y | z | TFCE value | Corrected p-value (FWE) | Cluster size (voxel) | Standardized beta (wiyhout TBV correction, with TBV correction)*** |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Amygdala (R:52)/Angular gyrus (R:4)/Cuneus (R:1)/Fusiform gyrus (R:331)/Heschl gyrus (R:1)/Hippocampus (R:128)/Inferior occipital lobe (R:74)/Middle occipital lobe (R:430)/Superior occipital lobe (R:32)/Parahippocampal gyrus (R:229)/Inferior temporal gyrus (R:774)/Middle temporal gyrus (R:642)/Temporal pole (R:89)/Superior temporal gyrus (R:223)/ | Sagittal stratum (R:41)/Cingulum (R:19)/Heschl gyrus (R:19)/Stria terminalis (R:19)/Uncinate fasciculus (R:7)/ | 56 | −14 | −20 | 961.28 | 0.002 | 3490 | −0.078, −0.077 |
| Fusiform gyrus (L:12)/Inferior occipital lobe (L:2)/Inferior temporal gyrus (L:128)/Middle temporal gyrus (L:33)/ | | −50 | −58 | −8 | 532.14 | 0.032 | 177 | −0.072, −0.073 |
| Heschl gyrus (R:4)/Insula (R:34)/Putamen (R:6)/Rolandic operculum (R:4)/ | Posterior limb of internal capsule (R:19)/Retrolenticular part of internal capsule (R:9)/External capsule (R:15)/ | 34 | −20 | 14 | 492.6 | 0.041 | 104 | −0.073, −0.073 |
| | Anterior corona radiata (L:16)/ | −22 | 38 | 12 | 480.77 | 0.044 | 56 | −0.077, −0.077 |
| Inferior frontal triangular (L:4)/Insula (L:10)/ | Anterior corona radiata (L:1)/ | −30 | 28 | 8 | 468.25 | 0.047 | 15 | −0.071, −0.071 |
| | Anterior corona radiata (L:10)/ | −26 | 30 | 22 | 465.86 | 0.048 | 16 | −0.071, −0.072 |
## Discussion
Our study revealed new associations between the length of TV viewing and subsequent changes in ICVF of the brain, and dementia onset in the elderly. Consistent with our hypothesis, the longer TV viewing was associated with a slight but increased risk of dementia development. Brain imaging analyses revealed that longer TV viewing is associated with subsequent greater decline in ICVF in the extensive areas of right lateral temporal cortex and the right medial temporal cortex, in the area around the left middle and inferior temporal cortex, and the area adjacent to the left inferior frontal gyrus, and left insula. The significant decline of ICVF in these areas in ones with prior longer TV viewing were partly consistent with our hypothesis that longer TV viewing would be associated with alterations in the fronto-polar areas and areas relevant to the language and memory as described below and the direction of change is consistent with the aging-related changes in this measure (Cox et al., 2016). These changes were not attributed to physical activity levels or education levels, as this was controlled for in our regression analyses. These are not attributed to other diseases or symptoms of undiagnosed as dementia as well, as controlling for having cancer, diabetes, heart attack, angina, stroke, other serious medical conditions, or blood pressure and removed subjects who were diagnosed as dementia or died within 5 years after baseline does not impact the significance and effect size of findings of dementia. As subjects with particularly low memory functioning at baseline is removed in analyses of dementia these are also not attributed to effects of inaccurate reporting of subjects with high risk of subsequent incident dementia.
In the present study, among neuroimaging measures, the significant findings were only observed in analyses of ICVF. Decline in ICVF may reflect decline in density of neurites (axon, dendrites) (Deligianni et al., 2016). In addition, although, Direction of ICVF’s change may be not uniform, but ICVF generally declines in aging, and it is also reduced in patients with dementia, consistent with neural atrophy in aging and dementia (Nir et al., 2013). Decline in ICVF perhaps might be induced by the principle of use it or lose it in neural systems (Frick and Benoit, 2010; Shors et al., 2012), or the occlusion of capillaries due to lifestyle with prolonged TV watching. The reason why the significant reduction of diffusivity measures is not observed is clear, However, we speculate this could be due to a decrease of blood flow arising from a decline in brain activity or the subtle occlusion of capillaries, and reduction may be density of neurites may be cancelled. The latter possibility is supported by finding showing that even among the physically active elderly, reducing the length of TV viewing results in a reduced risk of cardiovascular disease (Patterson et al., 2020).
The associations between longer TV viewing and changes in ICVF were found in the areas relevant to language, communication, memory and so on. The largest cluster of the associations between longer TV viewing and decline in ICVF were found in the cluster that primarily spread in the right lateral temporal cortex and medial temporal cortex. Both of the right temporal and left temporal gyrus are suggested to play key roles in a number of verbal processes (Binder et al., 1994; Herholz et al., 1994; McGuire et al., 1996). However, the right one may more focus on non-verbal processes such as the processing of non-verbal sound discrimination, recognition and comprehension (McGlone and Young, 1992), processing of linguistic context (Kircher et al., 2001), irony and metaphor comprehension (Eviatar and Just, 2006), and gaze recognition (Akiyama et al., 2006). On the other hand, hippocampus and the parahippocampal gyrus are critically involved in memory processes (Squire and Zola-Morgan, 1991), and atrophy of these area is related to progression of Alzheimer’s dementia (Köhler et al., 1998). In addition, ICVF of the hippocampus is shown to be related to the accumulation of tau protein which is involved in the disease progression of Alzheimer’s dementia and memory function (Nir et al., 2020). The left fusiform has multiple functions, but the function relevant to here may be this region’s role in reading and letter recognition (Thesen et al., 2012). Finally, the significant association in the area in and adjacent to the left insula and left inferior frontal gyrus, may be related to the left insula’s function of speech processing, left inferior frontal gyrus’ function of process of word generation (Indefrey and Levelt, 2000; Oh et al., 2014), and important roles in the phonological loop of the working-memory system (Schulze et al., 2011). Through these neural changes of areas related to abovementioned functions, longer TV viewing may be related to the reduction of verbal abilities, memory functions and increased risk of dementia.
This study has a few limitations. First, this study is an observational longitudinal study and not an intervention study. Although we corrected for a wide range of potential confounding factors, including activity level (MET), socioeconomic status, baseline dependent variable values, sleep duration, BMI, household number, and health status in our analyses, still uncorrected pre-dispositions toward longer TV viewing may affect our outcomes. Alternatively, TV viewing length may be itself a sign of later neurocognitive changes. These issues could ultimately be solved through well-designed intervention studies. In addition, in this project, the contents of viewed TV programs, or how TV was watched was not assessed and effects of these may be a matter of future investigations.
Previously, longer TV viewing was associated with subsequent cognitive decline in the children and elderly, regional gray matter volume changes during child development, an increased risk of cardiovascular disease and divided results of altered risk of dementia. In this current study, we advanced understanding of the effects of TV viewing by showing that longer TV viewing in the middle to old age was associated with a decline in ICVF which is supposed to reflect density of neurites (axon, dendrites) in areas particularly implicated in language, memory and increased risk of subsequent risk of dementia after corrections of a wide range of confounding variables. Attention should thus be paid to elderly who have daily habits of watching TV for prolonged periods of time.
## Data availability statement
The datasets presented in this article are not readily available because the data is accessible upon the request to UK Biobank. Requests to access the datasets should be directed to https://www.ukbiobank.ac.uk/.
## Ethics statement
The studies involving human participants were reviewed and approved by North-West Multi-center Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
HT conceptualized the study, pre-processed, analyzed the data, and wrote the manuscript. RK played a key role in obtaining the relevant funding and supervised the study. Both 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.2023.984919/full#supplementary-material
## References
1. Akiyama T., Kato M., Muramatsu T., Saito F., Nakachi R., Kashima H.. **A deficit in discriminating gaze direction in a case with right superior temporal gyrus lesion.**. (2006) **44** 161-170. DOI: 10.1016/j.neuropsychologia.2005.05.018
2. Aronson M.. **Does excessive television viewing contribute to the development of dementia?**. (1993) **41** 465-466. DOI: 10.1016/0306-9877(93)90128-D
3. Bakrania K., Edwardson C., Khunti K., Bandelow S., Davies M., Yates T.. **Associations between sedentary behaviors and cognitive function: Cross-sectional and prospective findings from the UK biobank.**. (2018) **187** 441-454. DOI: 10.1093/aje/kwx273
4. Binder J., Rao S., Hammeke T., Yetkin F., Jesmanowicz A., Bandettini P.. **Functional magnetic resonance imaging of human auditory cortex.**. (1994) **35** 662-672. DOI: 10.1002/ana.410350606
5. Celis-Morales C., Lyall D., Steell L., Gray S., Iliodromiti S., Anderson J.. **Associations of discretionary screen time with mortality, cardiovascular disease and cancer are attenuated by strength, fitness and physical activity: Findings from the UK Biobank study.**. (2018) **16**. DOI: 10.1186/s12916-018-1063-1
6. Cox S., Ritchie S., Tucker-Drob E., Liewald D., Hagenaars S., Davies G.. **Ageing and brain white matter structure in 3,513 UK Biobank participants.**. (2016) **7**. DOI: 10.1038/ncomms13629
7. Deligianni F., Carmichael D., Zhang G., Clark C., Clayden J.. **NODDI and tensor-based microstructural indices as predictors of functional connectivity.**. (2016) **11**. DOI: 10.1371/journal.pone.0153404
8. Edwards L., Pine K., Ellerbrock I., Weiskopf N., Mohammadi S.. **NODDI-DTI: Estimating neurite orientation and dispersion parameters from a diffusion tensor in healthy white matter.**. (2017) **11**. DOI: 10.3389/fnins.2017.00720
9. Eviatar Z., Just M.. **Brain correlates of discourse processing: An fMRI investigation of irony and conventional metaphor comprehension.**. (2006) **44** 2348-2359. DOI: 10.1016/j.neuropsychologia.2006.05.007
10. Fajersztajn L., Di Rienzo V., Nakamura C., Scazufca M.. **Watching TV and cognition: The SPAH 2-year cohort study of older adults living in low-income communities.**. (2021) **12**. DOI: 10.3389/fneur.2021.628489
11. Fancourt D., Steptoe A.. **Television viewing and cognitive decline in older age: Findings from the English Longitudinal Study of Ageing.**. (2019) **9**. DOI: 10.1038/s41598-019-39354-4
12. Frick K., Benoit J.. **Use it or lose it: Environmental enrichment as a means to promote successful cognitive aging.**. (2010) **10** 1129-1141. DOI: 10.1100/tsw.2010.111
13. Gadberry S.. **Effects of restricting first graders’ TV-viewing on leisure time use, IQ change, and cognitive style.**. (1981) **1** 45-57. DOI: 10.1016/0193-3973(80)90061-1
14. Grøntved A., Hu F.. **Television viewing and risk of type 2 diabetes, cardiovascular disease, and all-cause mortality: A meta-analysis.**. (2011) **305** 2448-2455. DOI: 10.1001/jama.2011.812
15. Hamer M., Yates T., Demakakos P.. **Television viewing and risk of mortality: Exploring the biological plausibility.**. (2017) **263** 151-155. DOI: 10.1016/j.atherosclerosis.2017.06.024
16. Herholz K., Pietrzyk U., Karbe H., Würker M., Wienhard K., Heiss W.. **Individual metabolic anatomy of repeating words demonstrated by MRI-guided positron emission tomography.**. (1994) **182** 47-50. DOI: 10.1016/0304-3940(94)90202-X
17. Hoang T., Reis J., Zhu N., Jacobs D., Launer L., Whitmer R.. **Effect of early adult patterns of physical activity and television viewing on midlife cognitive function.**. (2016) **73** 73-79. DOI: 10.1001/jamapsychiatry.2015.2468
18. Indefrey P., Levelt W., Gazzaniga M. S.. **The neural correlates of language production**. (2000)
19. Kircher T., Brammer M., Tous Andreu N., Williams S., McGuire P.. **Engagement of right temporal cortex during processing of linguistic context.**. (2001) **39** 798-809. DOI: 10.1016/S0028-3932(01)00014-8
20. Köhler S., Black S., Sinden M., Szekely C., Kidron D., Parker J.. **Memory impairments associated with hippocampal versus parahippocampal-gyrus atrophy: An MR volumetry study in Alzheimer’s disease.**. (1998) **36** 901-914. DOI: 10.1016/S0028-3932(98)00017-7
21. Landhuis C., Poulton R., Welch D., Hancox R.. **Does childhood television viewing lead to attention problems in adolescence? Results from a prospective longitudinal study.**. (2007) **120** 532-537. DOI: 10.1542/peds.2007-0978
22. Lindstrom H., Fritsch T., Petot G., Smyth K., Chen C., Debanne S.. **The relationships between television viewing in midlife and the development of Alzheimer’s disease in a case-control study.**. (2005) **58** 157-165. DOI: 10.1016/j.bandc.2004.09.020
23. Lourida I., Hannon E., Littlejohns T., Langa K., Hyppönen E., Kuźma E.. **Association of lifestyle and genetic risk with incidence of dementia.**. (2019) **322** 430-437. DOI: 10.1001/jama.2019.9879
24. Maldjian J., Laurienti P., Burdette J.. **Precentral gyrus discrepancy in electronic versions of the Talairach atlas.**. (2004) **21** 450-455. DOI: 10.1016/j.neuroimage.2003.09.032
25. Maldjian J., Laurienti P., Kraft R., Burdette J.. **An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets.**. (2003) **19** 1233-1239. DOI: 10.1016/S1053-8119(03)00169-1
26. McGlone J., Young G., Baker A.. **Cerebral localization**. (1992)
27. McGuire P., Silbersweig D., Frith C.. **Functional neuroanatomy of verbal self-monitoring.**. (1996) **119** 907-917. DOI: 10.1093/brain/119.3.907
28. Miller K., Alfaro-Almagro F., Bangerter N., Thomas D., Yacoub E., Xu J.. **Multimodal population brain imaging in the UK Biobank prospective epidemiological study.**. (2016) **19** 1523-1536. DOI: 10.1038/nn.4393
29. Nir T., Jahanshad N., Villalon-Reina J., Toga A., Jack C., Weiner M.. **Effectiveness of regional DTI measures in distinguishing Alzheimer’s disease, MCI, and normal aging.**. (2013) **3** 180-195. DOI: 10.1016/j.nicl.2013.07.006
30. Nir T., Salminen L., Reina J., Tubi M., Thomopoulos S., Maiti P.. **Hippocampal subfield microstructure abnormalities mediate associations between tau burden and memory performance: Neuroimaging/multi-modal comparisons.**. (2020) **16**. DOI: 10.1002/alz.039622
31. Oh A., Duerden E., Pang E.. **The role of the insula in speech and language processing.**. (2014) **135** 96-103. DOI: 10.1016/j.bandl.2014.06.003
32. Patterson F., Mitchell J., Dominick G., Lozano A., Huang L., Hanlon A.. **Does meeting physical activity recommendations ameliorate association between television viewing with cardiovascular disease risk? A cross-sectional, population-based analysis.**. (2020) **10**. DOI: 10.1136/bmjopen-2019-036507
33. Riley R., Ensor J., Snell K., Harrell F., Martin G., Reitsma J.. **Calculating the sample size required for developing a clinical prediction model.**. (2020) **368**. DOI: 10.1136/bmj.m441
34. Schulze K., Zysset S., Mueller K., Friederici A., Koelsch S.. **Neuroarchitecture of verbal and tonal working memory in nonmusicians and musicians.**. (2011) **32** 771-783. DOI: 10.1002/hbm.21060
35. Sepehrband F., Clark K., Ullmann J., Kurniawan N., Leanage G., Reutens D.. **Brain tissue compartment density estimated using diffusion-weighted MRI yields tissue parameters consistent with histology.**. (2015) **36** 3687-3702. DOI: 10.1002/hbm.22872
36. Sergeant J., Geurts H., Oosterlaan J.. **How specific is a deficit of executive functioning for attention-deficit/hyperactivity disorder?**. (2002) **130** 3-28. DOI: 10.1016/S0166-4328(01)00430-2
37. Shors T., Anderson M., Curlik I. D., Nokia M.. **Use it or lose it: How neurogenesis keeps the brain fit for learning.**. (2012) **227** 450-458. DOI: 10.1016/j.bbr.2011.04.023
38. Smith S., Jenkinson M., Johansen-Berg H., Rueckert D., Nichols T., Mackay C.. **Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data.**. (2006) **31** 1487-1505. DOI: 10.1016/j.neuroimage.2006.02.024
39. Smith S., Nichols T.. **Threshold-free cluster enhancement: Addressing problems of smoothing, threshold dependence and localisation in cluster inference.**. (2009) **44** 83-98. DOI: 10.1016/j.neuroimage.2008.03.061
40. Squire L., Zola-Morgan S.. **The medial temporal lobe memory system.**. (1991) **253** 1380-1386. DOI: 10.1126/science.1896849
41. Takeuchi H., Taki Y., Hashizume H., Asano K., Asano M., Sassa Y.. **The impact of television viewing on brain structures: Cross-Sectional and longitudinal analyses.**. (2015) **25** 1188-1197. DOI: 10.1093/cercor/bht315
42. Takeuchi H., Taki Y., Thyreau B., Sassa Y., Hashizume H., Sekiguchi A.. **White matter structures associated with empathizing and systemizing in young adults.**. (2013) **77** 222-236. DOI: 10.1016/j.neuroimage.2013.04.004
43. Thesen T., McDonald C., Carlson C., Doyle W., Cash S., Sherfey J.. **Sequential then interactive processing of letters and words in the left fusiform gyrus.**. (2012) **3**. DOI: 10.1038/ncomms2220
44. Tzourio-Mazoyer N., Landeau B., Papathanassiou D., Crivello F., Etard O., Delcroix N.. **Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain.**. (2002) **15** 273-289. DOI: 10.1006/nimg.2001.0978
45. Zimmerman F., Christakis D.. **Children’s television viewing and cognitive outcomes: A longitudinal analysis of national data.**. (2005) **159** 619-625. DOI: 10.1001/archpedi.159.7.619
|
---
title: 'Alcohol consumption: context and association with mortality in Switzerland'
authors:
- Flurina Suter
- Giulia Pestoni
- Janice Sych
- Sabine Rohrmann
- Julia Braun
journal: European Journal of Nutrition
year: 2022
pmcid: PMC10030531
doi: 10.1007/s00394-022-03073-w
license: CC BY 4.0
---
# Alcohol consumption: context and association with mortality in Switzerland
## Abstract
### Purpose
Non-communicable diseases generate the largest number of avoidable deaths often caused by risk factors such as alcohol, smoking, and unhealthy diets. Our study investigates the association between amount and context of alcohol consumption and mortality from major non-communicable diseases in Switzerland.
### Methods
Generalized linear regression models were fitted on data of the cross-sectional population-based National Nutrition Survey menuCH (2014–2015, $$n = 2057$$). Mortality rates based on the Swiss mortality data (2015–2018) were modeled by the alcohol consumption group considering the amount and context (i.e., during or outside mealtime) of alcohol consumption and potential confounders. The models were checked for spatial autocorrelation using Moran’s I statistic. Integrated nested Laplace approximation (INLA) models were fitted when evidence for missing spatial information was found.
### Results
Higher mortality rates were detected among drinkers compared to non-drinkers for all-cancer (rate ratio (RR) ranging from 1.01 to 1.07) and upper aero-digestive tract cancer (RR ranging from 1.15 to 1.20) mortality. Global Moran’s I statistic revealed spatial autocorrelation at the Swiss district level for all-cancer mortality. An INLA model led to the identification of three districts with a significant decrease and four districts with a significant increase in all-cancer mortality.
### Conclusion
Significant associations of alcohol consumption with all-cancer and upper aero-digestive tract cancer mortality were detected. Our study results indicate the need for further studies to improve the next alcohol-prevention scheme and to lower the number of avoidable deaths in Switzerland.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s00394-022-03073-w.
## Introduction
Non-communicable diseases (NCDs), such as cardiovascular diseases (CVD), diabetes, and cancers are the leading causes of death [1]. Moreover, NCDs cause most avoidable deaths often related to well-known risk factors, such as alcohol consumption, tobacco, and unhealthy diets [2, 3].
Previous studies have investigated the association of alcohol consumption with non-communicable diseases. However, the relationship between amount of alcohol consumption and health risks are not fully resolved yet [4–11]. A Swiss study found a J-shaped curve association between coronary heart disease risk and alcohol consumption, indicating a protective effect of moderate alcohol intake [12]. Nevertheless, other studies have shown the opposite effect, reporting an association of moderate alcohol consumption with liver disease and specific cancer site risk [4, 13, 14]. Alcohol is known to be carcinogenic for humans and has been classified by the International Agency for Research on Cancer (IARC) as a group 1 carcinogen [15]. It is a drug with a toxic effect on the human’s organs and tissue and its consumption can lead to psychoactive effects, which can in turn lead to injuries and accidents [16]. The IARC classifies alcoholic beverages as a carcinogen with sufficient evidence for the following cancer sites: oral cavity, pharynx, larynx, upper digestive tract, esophagus, colorectal, liver, bile duct and breast (in women) and with suggestive evidence for stomach, lung, and pancreatic cancer [17, 18].
The safe-drinking guideline by the Swiss Federal Commission for Alcohol Issues (EKAL) recommends to not exceed a maximum daily intake of 12 g for a healthy woman and 24 g for a healthy man [19]. The recommendation for women corresponds to a maximum of one standard glass which is about 3 dl beer, 1 dl wine, or 0.25 dl liquor and twice this amount for men [19]. In 2017, $8.4\%$ of all deaths in Switzerland between 15 and 74 years of age were caused by alcohol consumption, indicating an urgent need for a targeted alcohol-prevention scheme [20].
Previous studies suppose that not only the amount but also the context of alcohol consumption influences the risk of non-communicable disease [21, 22]. Evidence suggests that the consumption of alcoholic beverages without a meal might be more detrimental to health than consumption of alcohol with a meal [21, 22].
The aim of our study was to investigate amount and context of alcohol consumption, using dietary, sociodemographic, anthropometric, and lifestyle data from the first National Nutrition Survey, the menuCH study. Since the relationship between alcohol consumption and health risks are not fully resolved yet, we investigated the association between amount and context of alcohol consumption and mortality from major non-communicable diseases in Switzerland.
## Methods
The structure of this report was based on the STROBE-nut guidelines [23]. The data used for this study were combined from three different sources: the menuCH study (2014–2015), the Swiss population census data (2015–2018), and the Swiss mortality data (2015–2018). The three data sources were combined at the district level.
## Study design and participants of menuCH
The menuCH study is a cross-sectional population-based study conducted between January 2014 and February 2015 in ten centers across Switzerland [24, 25]. It included two 24-h dietary recalls (24HDR) and one questionnaire about sociodemographic, dietary, and lifestyle factors [25, 26]. The first 24HDR was conducted on-site in one of the centers, and the second one took place two to six weeks later by telephone [25].
In collaboration with the Federal Statistical Office (FSO), a target sample of 4,627,878 Swiss residents was drawn. The stratified, random sample included adults of 18–75 years of age, which represented both sexes, five age categories (18–29, 30–39, 40–49, 50–64, and ≥ 65 years old), the three main language regions in Switzerland (CH-German, CH-French, and CH-Italian), and the twelve most populated Swiss cantons of the seven major regions [25, 27].
From the source population consisting of 13,606 participants, 5496 participants were successfully contacted and eligible for the study [28]. Excluded from the study were 3410 non-responders and 29 participants who did not complete the dietary assessment [28]. The final study sample consisted of 2057 participants [28]. Detailed information on the study recruitment are presented in Online Resource Fig. S1.
## Anthropometric, lifestyle and demographic factors
The participants’ lifestyle and demographic factors were derived from the self-administered questionnaire [26]. The following variables were used in our study: sex (male, female), age (divided into eleven categories: 18–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–75 years old), Swiss language region (CH-German, CH-French, CH-Italian), education level (primary, secondary, tertiary), physical activity (low, moderate, high; based on the short-form International Physical Activity Questionnaire (IPAQ) definitions [29]), and smoking habits (never, former, current). The anthropometric factors of height, weight, and hip circumference were measured during the first 24HDR by trained interviewers following the WHO-MONICA protocol [25, 30]. The participants’ body weight and height were used to calculate their body mass index (BMI). For pregnant or lactating women, their self-reported weight before pregnancy was used to calculate the BMI. According to the World Health Organization (WHO) definitions, the participants were classified as ‘underweight’ (BMI < 18.5 kg/m2), ‘normal weight’ (18.5 kg/m2 ≤ BMI < 25.0 kg/m2), ‘overweight’ (25.0 kg/m2 ≤ BMI < 30.0 kg/m2) or ‘obese’ (BMI ≥ 30 kg/m2) [31].
## Dietary assessment
The dietary data were assessed by two 24HDR, which were distributed over all weekdays and seasons [32]. The data were collected by trained dietitians who used the trilingual Swiss version (0.2014.02.27) of the automated software GloboDiet® (GD, formerly EPIC-Soft®, IARC, Lyon, France [33, 34]), which was adapted by the Federal Food Safety and Veterinary Office, Bern, Switzerland. To facilitate the quantification of consumed amounts, a book with 119 series of six graduated portion-size pictures [35] and about 60 actual household measures were presented to the participants [28]. To ensure the quality of the collected data, the data were screened and cleaned according to the IARC’s guidelines using an updated version of GD® (0.2015.09.28) [36]. Afterwards, the foods, recipes, and ingredients obtained by the GD® software were linked using the matching tool FoodCASE (Premotec GmbH, Winterthur, Switzerland) to the most suitable item found in the Swiss Food Composition Database [37].
Each menuCH participant was categorized into one of six alcohol consumption groups considering the participant’s pure alcohol intake from alcoholic beverages in the 24HDR as well as information on general alcohol avoidance from the self-administered questionnaire. Participants, who did not consume alcoholic beverages in the 24HDR and reported alcohol avoidance were categorized as ‘abstainer’, whereas participants who did not report alcohol consumption in the 24 HDR but did not report alcohol avoidance were categorized as ‘safe_no’. Participants, who did consume alcoholic beverages in the 24HDR were categorized into four groups. On the one hand, the categorization was based on whether the participants consumed more pure alcohol during or outside mealtime (‘during’ and ‘outside’ drinkers, respectively). Since the 24HDR did not directly assess the context of drinking, the definition by Sieri et al. [ 38] was used: an alcoholic beverage was consumed outside mealtime if during that specific time of the day less than $10\%$ of the total daily energy intake (excluding energy intake from alcoholic beverages) was consumed and the consumed amount of pure alcohol was at least 5 g. On the other hand, the categorization was based on whether the participant drank on average more than the maximum daily recommendation (women: 12 g pure alcohol; men: 24 g pure alcohol) proposed by the EKAL [19]. If the participants drank on average more than recommended, they were categorized as heavy drinker (‘heavy’) and otherwise categorized as safe drinker (‘safe’). Therefore, the resulting six alcohol consumption groups were: ‘abstainer’, ‘safe_no’, ‘safe_during', ‘safe_outside’, ‘heavy_during’, or ‘heavy_outside’. An overview of the six alcohol consumption groups is given in the Online Resource Table S1.
To investigate diet quality, the alternate healthy eating index (AHEI) was calculated for each participant for each of the two 24HDR interviews [32]. Then, the average AHEI was determined and used in the analyses of alcohol consumption data. The AHEI is based on eleven components, each having a score between 0 and 10 points [32]; these components are: vegetables, fruits, whole grains, sugar sweetened beverages, nuts, meat, trans fat, long chain omega-3 fatty acids, polyunsaturated fatty acids (PUFA), sodium, and alcohol. We excluded the component ‘alcohol’ since the effect of alcohol consumption should only be contained in the alcohol group variable. Therefore, in the present study, AHEI scores between 0 and 100 were possible, with higher scores indicating a healthier diet.
## Swiss population and mortality data
The Swiss population census data and the mortality data were provided by the FSO [39]. For both databases, only residents between 18 and 75 years old were included to ensure the same age range as in the menuCH study. All-cause mortality and cause-specific mortality determined by the final cause of death (encoded using the 10th revision of the international classification of diseases (ICD-10) [40]) were investigated. The following cause-specific mortalities were investigated: CVD (ICD-10: I00-I99), all-cancer (ICD-10: C00-C97, D32-D33, and D37-D48), colorectal cancer (ICD-10: C18-C21), liver cancer (ICD-10: C22), upper aero-digestive tract (UADT) cancer (all organs and tissues of the respiratory tract, upper part of the digestive tract, and the upper esophagus (ICD-10: C00-C15 and C32), but excluding the stomach), breast cancer (only in women; ICD-10: C50), diabetes (ICD-10: E10-E14.9).
In addition, based on evidence for alcohol-related carcinogenic effects on human organs [17, 18], the following eight cancer sites were combined into one group: colorectal, liver, UADT, breast, prostate (ICD-10: C61), pancreatic (ICD-10: C25), urinary tract (ICD-10: C67-C68), stomach (ICD-10: C16) [41, 42].
Mortality ratios standardized for sex, age, and year of death (SMR) were calculated at the district level by dividing the number of observed deaths by the number of expected deaths in the overall Swiss population, based on the Swiss population census data. The latter was determined using an indirect method based on the standardized Swiss population mortality rates. The district of each menuCH participant was determined using their postal code and data provided by the FSO dated on the 1st of January 2019 [43, 44].
## Statistical analysis
Since not all Swiss inhabitants had the same probability to be included in the menuCH study sample, the analyses of the participants’ data of the final sample were weighted based on sex, age, marital status, major living region, nationality, household size, weekday, and season of the 24HDR day [27].
Descriptive statistics (absolute numbers, median, interquartile range, and percentages) were used to characterize the study population. Additionally, descriptive maps were used to show the geographic distribution of chronic diseases district-level SMR. The association between alcohol consumption and mortality was investigated by modeling mortality rates, fitting generalized linear regression models. Negative binomial regression models were used to handle overdispersion (for all-cause, CVD, all-cancer, and breast cancer mortality) and Quasipoisson models to handle underdispersion (for colorectal cancer, liver cancer, UADT cancer, and diabetes mortality). For each of the menuCH participants, the total number of observed deaths in the participant’s sex, age, and district category was used as outcome variable. The explanatory variable was the alcohol consumption group and the participant’s sex, age (continuous variable), smoking status, physical activity level, BMI group, education level, and average AHEI were included as further covariates. The log of the total number of residents stratified by the participant’s sex, age, and district category was included as offset term.
Missing values for physical activity level ($$n = 524$$), education level ($$n = 3$$), and smoking category ($$n = 4$$) (Table 1) were imputed using multivariate imputation by chained equations (MICE) [45]. The results of the 30 imputed data sets were pooled using Rubin’s rule [46]. *In* general, the results obtained with the imputed data sets were similar to the results of the complete case analyses. Therefore, only the results based on the imputed data sets are presented and used for further analyses. Table 1Characteristics of menuCH participants ($$n = 2057$$) stratified by alcohol consumption groupa,bVariableOverallAbstainerSafe_noSafe_duringSafe_outsideHeavy_duringHeavy_outsiden2057192678503115452117Women, n (%. %*) 1124 ($54.6\%$, $50.2\%$)135 ($70.3\%$, $65.6\%$)415 ($61.2\%$, $56\%$)252 ($50.1\%$, $47\%$)37 ($32.2\%$, $27\%$)232 ($51.3\%$, $46.5\%$)53 ($45.3\%$, $42.7\%$)Age (IQR)45 [33, 58]45 [35, 56]41 [29, 54]48 [35, 61]40 [30, 53]50 [40, 62]40 [27, 54]Age group, n (%, %*) 18-29 years old400 ($19.4\%$, $18.8\%$)40 ($20.8\%$, $17.9\%$)183 ($27\%$, $25.9\%$)63 ($12.5\%$, $13.4\%$)26 ($22.6\%$, $20\%$)48 ($10.6\%$, $12.3\%$)40 ($34.2\%$, $30.7\%$) 30-44 years old533 ($25.9\%$, $29.9\%$)50 ($26\%$, $30.8\%$)206 ($30.4\%$, $34.4\%$)127 ($25.2\%$, $28.9\%$)36 ($31.3\%$, $38.1\%$)83 ($18.4\%$, $22.8\%$)31 ($26.5\%$, $30.5\%$) 45-59 years old625 ($30.4\%$, $29.8\%$)62 ($32.3\%$, $32.6\%$)168 ($24.8\%$, $23.1\%$)169 ($33.6\%$, $31.5\%$)36 ($31.3\%$, $28.2\%$)162 ($35.8\%$, $36.8\%$)28 ($23.9\%$, $23.4\%$) 60-75 years old499 ($24.3\%$, $21.6\%$)40 ($20.8\%$, $18.6\%$)121 ($17.8\%$, $16.6\%$)144 ($28.6\%$, $26.3\%$)17 ($14.8\%$, $13.7\%$)159 ($35.2\%$, $28.1\%$)18 ($15.4\%$, $15.3\%$)Language region, n (%, %*)c German1341 ($65.2\%$, $68.8\%$)112 ($58.3\%$, $66.4\%$)466 ($68.7\%$, $70.6\%$)326 ($64.8\%$, $69.7\%$)95 ($82.6\%$, $81\%$)256 ($56.6\%$, $62.1\%$)86 ($73.5\%$, $75.6\%$) French502 ($24.4\%$, $25.7\%$)53 ($27.6\%$, $26.5\%$)151 ($22.3\%$, $25\%$)115 ($22.9\%$, $22.8\%$)15 ($13\%$, $16.3\%$)144 ($31.9\%$, $32.5\%$)24 ($20.5\%$, $20.2\%$) Italian214 ($10.4\%$, $5.6\%$)27 ($14.1\%$, $7.1\%$)61 ($9\%$, $4.4\%$)62 ($12.3\%$, $7.5\%$)5 ($4.3\%$, $2.7\%$)52 ($11.5\%$, $5.4\%$)7 ($6\%$, $4.2\%$)Education level, n (%, %*) Primary89 ($4.3\%$, $4.7\%$)15 ($7.8\%$, $7.6\%$)32 ($4.7\%$, $5.8\%$)21 ($4.2\%$, $3.9\%$)3 ($2.6\%$, $1.7\%$)13 ($2.9\%$, $3.5\%$)5 ($4.3\%$, $3.8\%$) Secondary968 ($47.1\%$, $43.2\%$)93 ($48.4\%$, $46.1\%$)332 ($49.2\%$, $44.9\%$)229 ($45.5\%$, $41.5\%$)49 ($42.6\%$, $41.1\%$)205 ($45.4\%$, $41.2\%$)60 ($51.3\%$, $47.1\%$) Tertiary997 ($48.5\%$, $52.1\%$)84 ($43.8\%$, $46.2\%$)311 ($46.1\%$, $49.2\%$)253 ($50.3\%$, $54.6\%$)63 ($54.8\%$, $57.2\%$)234 ($51.8\%$, $55.3\%$)52 ($44.4\%$, $49.1\%$) NA3 ($0.1\%$, $0.3\%$)0 ($0\%$, $0\%$)3 ($0.4\%$, $0.9\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)BMI (IQR)24.3 (21.8, 27.1)24.4 (21.5, 27.8)24.0 (21.6, 26.8)24.4 (21.9, 26.6)25.7 (22.8, 28.1)24.7 (22.0, 27.9)24.4 (22.2, 26.8)BMI group, n (%, %*) Underweight51 ($2.5\%$, $2.3\%$)5 ($2.6\%$, $2.6\%$)18 ($2.7\%$, $2.6\%$)10 ($2\%$, $1.9\%$)4 ($3.5\%$, $2.4\%$)12 ($2.7\%$, $2.4\%$)2 ($1.7\%$, $1.6\%$) Normal1115 ($54.2\%$, $54.4\%$)108 ($56.2\%$, $57.2\%$)395 ($58.3\%$, $58.3\%$)267 ($53.1\%$, $55\%$)53 ($46.1\%$, $40.1\%$)227 ($50.2\%$, $50.7\%$)65 ($55.6\%$, $55.6\%$) Overweight629 ($30.6\%$, $30.7\%$)42 ($21.9\%$, $19.3\%$)191 ($28.2\%$, $27.6\%$)176 ($35\%$, $34\%$)42 ($36.5\%$, $45.3\%$)145 ($32.1\%$, $33.5\%$)33 ($28.2\%$, $28.4\%$) Obese262 ($12.7\%$, $12.5\%$)37 ($19.3\%$, $20.8\%$)74 ($10.9\%$, $11.4\%$)50 ($9.9\%$, $9.2\%$)16 ($13.9\%$, $12.2\%$)68 ($15\%$, $13.5\%$)17 ($14.5\%$, $14.4\%$)Physical activity, n (%, %*) Low219 ($14.3\%$, $14.9\%$)21 ($15.1\%$, $24.1\%$)78 ($15.5\%$, $15.5\%$)54 ($14.1\%$, $13.7\%$)8 ($8.8\%$, $14.4\%$)46 ($13.6\%$, $12.6\%$)12 ($15\%$, $12.1\%$) Moderate487 ($31.8\%$, $31.7\%$)40 ($28.8\%$, $23.7\%$)148 ($29.4\%$, $29.3\%$)127 ($33.2\%$, $34.2\%$)28 ($30.8\%$, $30.8\%$)120 ($35.6\%$, $36.6\%$)24 ($30\%$, $27.2\%$) High827 ($53.9\%$, $53.4\%$)78 ($56.1\%$, $52.2\%$)278 ($55.2\%$, $55.3\%$201 ($52.6\%$, $52.1\%$)55 ($60.4\%$, $54.8\%$)171 ($50.7\%$, $50.8\%$)44 ($55\%$, $60.7\%$) NA524 ($25.5\%$, $24.8\%$)53 ($27.6\%$, $29.1\%$)174 ($25.7\%$, $25.4\%$)121 ($24.1\%$, $22.3\%$)24 ($20.9\%$, $21.3\%$)115 ($25.4\%$, $25.2\%$)37 ($31.6\%$, $26.1\%$)Smoking, n (%, %*) Never914 ($44.5\%$, $42.1\%$)121 ($63\%$, $64.1\%$)332 ($49.3\%$, $46.9\%$)226 ($44.9\%$, $42.9\%$)55 ($47.8\%$, $48.2\%$)153 ($33.8\%$, $30.1\%$)27 ($23.1\%$, $18.9\%$) Former688 ($33.5\%$, $35.5\%$)53 ($27.6\%$, $29.2\%$)202 ($30\%$, $31.9\%$)190 ($37.8\%$, $40\%$)35 ($30.4\%$, $27.8\%$)172 ($38.1\%$, $40.4\%$)36 ($30.8\%$, $33.4\%$) Current451 ($22\%$, $22.4\%$)18 ($9.4\%$, $6.8\%$)140 ($20.8\%$, $21.1\%$)87 ($17.3\%$, $17.1\%$)25 ($21.7\%$, $24\%$)127 ($28.1\%$, $29.6\%$)54 ($46.2\%$, $47.7\%$) NA4 ($0.2\%$, $0.3\%$)0 ($0\%$, $0\%$)4 ($0.6\%$, $1.0\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)0 ($0\%$, $0\%$)Pure alcohol intake [g/day] (IQR)5.7 (0.0, 20.6)0 [0, 0]0 [0, 0]8.0 (5.0, 11.8)9.8 (6.4, 15.0)30.7 (21.4, 43.8)33.0 (25.1, 49.6)AHEI (IQR)d39.9 (31.7, 48.7)42.4 (33.1, 52.3)40.4 (32.1, 48.8)40.7 (32.4, 49.5)36.4 (29.4, 43.3)39.0 (31.2, 47.1)38.6 (30.2, 43.9)BMI body mass index, AHEI alternate healthy eating index, NA missing valuesaParticipants not consuming alcoholic beverages in the 24HDR were categorized as ‘abstainer’ if reporting alcohol avoidance, and as ‘safe-no’ if not. Participants consuming alcoholic beverages in the 24HDR were categorized based on whether the participants consumed more alcohol during or outside mealtime (‘during’ and ‘outside’, respectively) and on whether their consumption was below or above the maximum daily recommendations (‘safe’ and ‘heavy’, respectively) [19, 38]bCategorical variables are expressed as absolute number (n), unweighted percentage (%), and weighted percentage (%*). The weighted percentages (%*) are weighted according to the menuCH weighting strategy [27] for sex, age, marital status, major living region in Switzerland, nationality, household size, weekday, and season of the recall day. Continuous variables are expressed as weighted median and weighted interquartile range. The weighted median and IQR are weighted according to the menuCH weighting strategy [27] for sex, age, marital status, major living region in Switzerland, nationality, household size, weekday, and season of the recall daycGerman language region: canton Aargau, Basel City, Basel Country, Berne, Lucerne, Zurich, and St. Gallen. French language region: canton Jura, Neuchâtel, Vaud, and Geneva. Italian language region: canton TicinodThe AHEI is calculated as average AHEI of the two 24HDR. The AHEI is based on ten components each contributing between 0 and 10 points: vegetables, fruits, whole grains, sugar sweetened beverages, nuts, meat, trans fat, long chain omega-3 fatty acids, polyunsaturated fatty acids (PUFA), and sodium. The healthier the participant’s diet, the higher the AHEI score The districts were defined as neighboring districts based on a first order neighborhood structure with rook contiguity. Additionally, the neighbors’ data were weighted by taking the inverse of the total number of neighbors of the corresponding district. The residuals of the regression models were investigated for spatial autocorrelation at the district level using global and local Moran’s I. The global Moran’s I statistic is an indicator for the existence and degree of spatial autocorrelation [47]. The statistic can range from -1 indicating spatial dispersion up to + 1 indicating spatial cluster-building [47]. A one-sided P value based on the Z-score was calculated [48]. To check the robustness, a one-sided P value based on 1000 Monte Carlo (MC) simulations was calculated additionally. Local Moran’s I were checked for significance based on a permutation test ($$n = 1000$$). No correction for multiple testing was included, since the number of Monte Carlo simulations determined the lower limit of the P value [49]. Local Moran’s I values were visualized using local indicators of spatial autocorrelation (LISA) cluster maps.
If evidence for spatial autocorrelation was detected, an integrated nested Laplace approximation (INLA) model was fitted. The structured spatial component was a Besag model [50] and the unstructured spatial component was an iid model (random noise). The default LogGamma prior distribution (shape = 1; rate = 0.00005) was used for both components. The results for each imputed data set were pooled by calculating the average of the estimates.
The analyses were performed in GeoDa (version 1.14.0) and in the R programming language (version 4.1.0; [51]). In R, the packages popEpi (version 0.4.8) and Epi (version 2.44) were used to calculate the SMR, mice (version 3.13.0) to impute missing values, survey (version 4.1.1) and DescTools (version 0.99.42) to conduct weighted analyses, MASS (version 7.3.54) to fit generalized linear regression models, spdep (version 1.1.8) and rgeos (version 0.5.5) to conduct spatial analyses, ggplot2 (version 3.3.5), ggsn (version 0.5.0), and sf (version 1.0.2) for creating figures, and INLA (version 21.11.22) to set up INLA models. For all analyses, the statistical significance level was set to 0.05.
## Descriptive results
The menuCH participants’ baseline characteristics stratified by alcohol consumption group are shown in Table 1. The largest group were the occasional drinkers (safe_no) with 678 participants and the two smallest were the two outside mealtime groups with 117 and 115 participants, respectively. All six alcohol consumption groups were characterized by differences across the variables investigated in comparison to the overall study population, e.g., abstainers and occasional drinkers were more likely to be never smokers, whereas during mealtime drinkers were more likely to be current or former smokers and outside mealtime drinking was more common among younger and during mealtime drinking among older participants. Median pure alcohol consumption per day stratified by sex and context of drinking can be seen in Fig. S2 and Fig. S3 (Online Resource).
Figure 1 shows the average pure alcohol intake per person stratified by sex, weekday, and context of drinking. Regardless of weekday and context, men drank more than women. Both sexes consumed more pure alcohol during mealtime than outside. The amount of pure alcohol intake was lower at the beginning of the week, increased from Thursday onwards and reached its peak on Saturday. On Saturday, men consumed on average 23.7 g pure alcohol during and 8.5 g outside mealtime and women on average 11.1 g pure alcohol during and 3.4 g outside mealtime. On Saturday only, both sexes consumed on average more than the maximum daily recommendation given by the EKAL [19].Fig. 1Average per person pure alcohol intake stratified by sex (M = male; F = female), recall day, and context of drinking (unweighted data, $$n = 2057$$). The threshold for pure alcohol intake per day proposed by the EKAL [19] is shown by a dashed line for women (12 g) and as a dotted line for men (24 g) Between the years 2015 and 2018, the following number of deaths were documented in Switzerland: 84,959 all-cause deaths, 16,082 CVD deaths, 37,202 all-cancer deaths, 3327 colorectal cancer deaths, 1821 liver cancer deaths, 2360 UADT cancer deaths, 3064 breast cancer deaths among women, 1256 diabetes deaths, and 17,362 deaths attributable to the eight specific cancer sites group.
The SMR at the district level are shown in Fig. 2. All-cause and all-cancer maps revealed a similar pattern with high SMR mainly in the western region and low SMR mainly in the central region. For CVD mortality high SMR were detected in the northwestern and eastern region. In contrast, many districts with low SMR were observed in the southwestern region. No clear pattern was detected for any of the specific cancer type mortalities, except for liver cancer: CH-French and CH-Italian districts tended to have higher SMR than CH-German districts. High diabetes SMR were more prevalent in central and northwestern region, whereas lower diabetes SMR were observed in the northeastern and southwestern region. Fig. 2Standardized mortality ratios (SMR) at the district level (unweighted data, number of districts = 143). The SMR are standardized for sex, age, and year of death. The SMR were calculated using an indirect method with the whole Swiss population as reference population. Breast cancer SMR (F) were calculated only for women. For all other causes of death (A, B, C, D, E, G, H), the data of both sexes were included to calculate the SMR
## Main results
Table 2 shows the results of the generalized linear regression models, with occasional drinkers used as reference group. *In* general, the consumption of alcoholic beverages tended to increase mortality rates, especially for all-cancer and UADT cancer mortality. Heavy, during mealtime drinkers had a higher risk of all-cancer (RR = 1.05, $95\%$ CI 1.00, 1.10), breast cancer (RR = 1.10, $95\%$ CI 1.00, 1.21), and UADT cancer (RR = 1.19, $95\%$ CI 1.09, 1.31) mortality. Furthermore, heavy, outside mealtime drinkers had an increased UADT cancer (RR = 1.20, $95\%$ CI 1.00, 1.43) and diabetes mortality (RR = 1.35, $95\%$ CI 1.07, 1.71). Even for safe drinkers there was evidence for an increased risk of all-cancer (safe_during: RR = 1.07, $95\%$ CI 1.02, 1.13), liver cancer (safe_outside: RR = 1.27, $95\%$ CI 1.07, 1.51), and UADT cancer (safe_during: RR = 1.15, $95\%$ CI 1.04, 1.26) mortality. Interestingly, the abstainers showed evidence for an increased all-cancer (RR = 1.07, $95\%$ CI 1.00, 1.14) and UADT cancer (RR = 1.21, $95\%$ CI 1.07, 1.37) mortality. Table 2Association of alcohol consumption with sex-, age-, and district-specific mortality rate ($$n = 2057$$) (rate ratios and $95\%$ confidence intervals)aAll-causeb,d,gCVDb,d,gAll-cancerb,d,gColorectal cancerc,d,gRR$95\%$ CIRR$95\%$ CIRR$95\%$ CIRR$95\%$ CIAlcohol consumption groupf Abstainer1.010.96, 1.070.970.88, 1.071.07*1.00, 1.140.970.88, 1.09 Safe_no (ref.)1.001.001.001.00 Safe_during0.990.96, 1.030.990.93, 1.061.07*1.02, 1.131.070.99, 1.16 Safe_outside1.010.95, 1.081.010.90, 1.141.010.93, 1.111.150.99, 1.32 Heavy_during1.000.96, 1.030.990.93, 1.061.05*1.00, 1.101.050.97, 1.13 Heavy_outside1.060.99, 1.131.070.95, 1.211.040.95, 1.131.080.93, 1.25Liver cancerc,d,gBreast cancerb,e,gUADT cancerc,d,gDiabetesc,d,gRR$95\%$ CIRR$95\%$ CIRR$95\%$ CIRR$95\%$ CIAlcohol consumption groupf Abstainer1.080.94, 1.241.100.97, 1.241.21*1.07, 1.371.000.83, 1.20 Safe_no (ref.)1.001.001.001.00 Safe_during0.960.87, 1.071.070.97, 1.181.15*1.04, 1.260.930.81, 1.07 Safe_outside1.27*1.07, 1.511.200.95, 1.521.070.90, 1.281.040.81, 1.33 Heavy_during1.090.99, 1.201.10*1.00, 1.211.19*1.09, 1.310.910.79, 1.04 Heavy_outside0.960.79, 1.171.090.90, 1.321.20*1.00, 1.431.35*1.07, 1.71CVD cardiovascular diseases, UADT upper aero-digestive tract, RR rate ratio, CI confidence intervalaThe menuCH participants’ data were weighted according to the weighting strategy [27] for sex, age, marital status, major living region in Switzerland, nationality, household size, weekday, and season of the recall daybA negative binomial regression model was fittedcA Quasipoisson regression model was fitteddThe analysis included data of both sexes and were further adjusted for sex, age, smoking category, physical activity, BMI group, education level, and alternate health eating indexeThe analysis included data of only women and was further adjusted for age, smoking category, physical activity, BMI group, education level, and alternate health eating indexfParticipants not consuming alcoholic beverages in the 24HDR were categorized as ‘abstainer’ if reporting alcohol avoidance, and as ‘safe-no’ if not. Participants consuming alcoholic beverages in the 24HDR were categorized based on whether the participants consumed more alcohol during or outside mealtime (‘during’ and ‘outside’, respectively) and on whether their consumption was below or above the maximum daily recommendations (‘safe’ and ‘heavy’, respectively)[19, 38]gBetween the years 2015 and 2018, the following number of deaths were documented in Switzerland: 84,959 all-cause deaths, 16,082 CVD deaths, 37,202 all-cancer deaths, 3327 colorectal cancer deaths, 1821 liver cancer deaths, 3064 breast cancer deaths among women, 2360 UADT cancer deaths, and 1256 diabetes deaths*Statistical significance (significance level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α= 0.05)
## Spatial analyses
The residuals of the generalized regression models were investigated for spatial autocorrelation at the district level using global Moran’s I statistic. The results of the spatial autocorrelation analysis are shown in Table 3. Only for all-cancer mortality, there was evidence for spatial autocorrelation (observed global Moran’s I: 0.144; expected global Moran’s I: − 0.014). The significant local Moran’s I values are visualized in a LISA cluster map (Online Resource Fig. S4). In total, 5 districts showed evidence for spatial clusters or spatial outliers (Fig. 3).Table 3Global Moran’s I statistic based on generalized linear regression model residuals at the district level ($$n = 75$$)Cause of deathObserved Moran’s IExpected Moran’s IVariance Moran’s IPZ-scoreaPMCaAll-cause– 0.013– 0.0140.0070.50.44CVD0.101– 0.0140.0070.0860.11All-cancer0.144– 0.0140.0080.035*0.039*Colorectal cancer0.091– 0.0140.0080.120.13Liver cancer– 0.056– 0.0140.0070.30.28Breast cancer– 0.097– 0.0140.0080.170.18UADT cancer0.016– 0.0140.0010.220.18Diabetes– 0.032– 0.0140.0050.40.44CVD cardiovascular diseases, UADT upper aero-digestive tract, MC Monte CarloaOne-sided P value with significance level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α= 0.05*Statistical significance (significance level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α= 0.05)Fig. 3Geographic visualization at district level for the structured spatial component of the integrated nested Laplace approximation (INLA) model. Districts with a statistically significant structured spatial component are colored either red or blue, indicating significantly increased or significantly decreased mortality, respectively Based on results of the INLA model (Table 4), there was evidence for an increased all-cancer mortality among abstainers (RR = 1.07, $95\%$ CI 1.00, 1.14); safe, during mealtime drinkers (RR = 1.08, $95\%$ CI 1.03, 1.13); and heavy, during mealtime drinkers (RR = 1.04, $95\%$ CI 1.00, 1.09) compared to occasional drinkers. Table 4INLA model for association of alcohol consumption with sex-, age-, and district-specific all-cancer mortality: fixed effects ($$n = 2057$$) (rate ratios, standard deviations, $95\%$ credible intervals)All-canceraRRSD$95\%$ CIAlcohol consumption groupb Abstainer1.07*0.031.00, 1.14 Safe_no (ref.)1.00 Safe_during1.08*0.021.03, 1.13 Safe_outside1.020.040.94, 1.11 Heavy_during1.04*0.021.00, 1.09 Heavy_outside1.050.040.97, 1.14RR rate ratio, SD standard deviation, CI credible intervalaThe negative binomial model part of the INLA model included data of both sexes and was further adjusted for the following fixed effects: sex, age, smoking category, physical activity, BMI group, education level, and alternate healthy eating index. The menuCH participants’ data were weighted according to the weighting strategy [27] for sex, age, marital status, major living region in Switzerland, nationality, household size, weekday, and season of the recall daybParticipants not consuming alcoholic beverages in the 24HDR were categorized as ‘abstainer’ if reporting alcohol avoidance, and as ‘safe-no’ if not. Participants consuming alcoholic beverages in the 24HDR were categorized based on whether the participants consumed more alcohol during or outside mealtime (‘during’ and ‘outside’, respectively) and on whether their consumption was below or above the maximum daily recommendations (‘safe’ and ‘heavy’, respectively) [19, 38]*Statistical significance (significance level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α= 0.05) Districts with a statistically significant structured spatial component are shown in Fig. 3. The districts Uster (RR = 0.91, $95\%$ CI 0.82, 0.99), Hinwil (RR = 0.91, $95\%$ CI 0.81, 1.00), and Meilen (RR = 0.89, $95\%$ CI 0.78, 0.99) revealed evidence for a decreased all-cancer mortality rate, whereas the districts canton Neuchâtel (RR = 1.12, $95\%$ CI 1.01, 1.27), Jura Bernois (RR = 1.13, $95\%$ CI 1.01, 1.31), Seeland (RR = 1.14, $95\%$ CI 1.01, 1.33), and Biel (RR = 1.18, $95\%$ CI 1.01, 1.40) revealed evidence for an increased all-cancer mortality rate.
## Sensitivity analyses
The adjustments for diet quality via the AHEI did not meaningfully change the results of our study. The SMR for the cancer sites with evidence for a carcinogenic effect of alcoholic beverages [17, 18], which are not presented in Fig. 2, are shown in Online Resource Fig. S5. The results of the negative binomial regression model for mortality from the eight specific cancer sites were similar to those for all-cancer mortality (Online Resource Table S2). The global Moran’s I statistic revealed no evidence for spatial autocorrelation (observed Moran’s I: − 0.060, expected Moran’s I: − 0.014) and therefore, no INLA model was fitted.
## Discussion
In our study, descriptive differences in alcohol consumption were observed for anthropometric and lifestyle factors, revealing risk groups that should be targeted by alcohol-prevention strategies. Significantly higher mortality rates with increasing alcohol consumption were detected especially for all-cancer and UADT cancer, consistent with evidence-based carcinogenic effects of alcohol reported in previous studies [17, 18]. For the other investigated causes of death, the results pointed in the same direction, i.e., increase in mortality with increasing alcohol consumption, but were mostly not statistically significant. The INLA model for all-cancer mortality revealed Swiss districts with a significantly lower or higher all-cancer baseline mortality rate, indicating the existence of additional factors influencing all-cancer mortality.
Similarly as in a European study [38], alcohol consumption was higher for men than women, and higher during mealtime than outside. The amount of pure alcohol intake was lower at the beginning of the week, increased from Thursday onwards and reached its peak on Saturday. On Saturdays, both sexes consumed on average more than the maximum daily recommendation given by the EKAL [19]. This weekday pattern is in line with the current literature, which reports an increase in alcohol consumption toward the weekend and reaching a peak on Friday and Saturday [52, 53].
Observed differences in amounts and context of alcohol intake indicate that Switzerland has a similar general drinking pattern as other European countries. For example, a high amount of alcohol consumption was more prominent in men [19], in the CH-German and CH-French regions [19, 54], and in individuals with a high education level [19, 55], an increased, unhealthy BMI [56, 57], high physical activity level [58, 59], and in current smokers [60], compared with corresponding references. In our study, age differences were observed with respect to the context of drinking: younger participants were more prominent in the outside mealtime alcohol consumption groups, whereas older participants were more prominent in the during mealtime groups. Therefore, Switzerland could adopt an already existing and successfully implemented alcohol-prevention scheme of another country with a similar general drinking behavior.
The investigation of Swiss mortality data revealed a cause-of-death specific pattern at the district level. Regional variations such as the (diet) culture, socioeconomic factors, or urbanization of the districts might have influenced the SMR. Overall, the generalized linear regression models revealed a general trend of increased mortality rates across alcohol drinkers and abstainers compared to occasional drinkers. The resulting relationship was often a J-shaped curve that is commonly reported in the current literature for several causes of death [61]. Nevertheless, some researchers are questioning the J-shaped curve since the composition of the abstainer group is often heterogeneous, including former heavy drinkers or participants who avoid alcohol for reasons of poor health, and this leads to distorted results [62–64]. The increased rate among abstainers in our study might be due to the heterogeneous group composition, because the reason for alcohol avoidance was not assessed in the menuCH survey [64].
In our study, weak evidence was detected for an association of alcohol consumption with all-cause, CVD, and colorectal cancer mortality. In contrast, for all-cancer and UADT cancer mortality, strong evidence was observed for a higher mortality risk among abstainers and during mealtime drinkers in comparison to occasional drinkers. Overall, the highest rate ratios were observed in outside mealtime drinking groups for liver cancer, breast cancer, and diabetes mortality. Generally, the mortality rate ratios were all pointing in the same direction, indicating an increased mortality rate with increasing alcohol consumption, which is in line with the current literature on all-cause [63, 64] and noncommunicable disease mortality [17, 62, 65–71]. Therefore, our study suggests that there exists no safe alcohol drinking level. However, most rate ratios in our study were statistically not significant, presumably due to small sample sizes and small numbers of observed deaths.
Sieri et al. [ 38] postulated that alcoholic beverages might be more harmful when consumed outside mealtime. In our study, we found increased diabetes, UADT cancer, and liver cancer mortality rates for outside mealtime drinkers compared to occasional drinkers. In contrast, during mealtime drinkers had increased all-cancer, UADT cancer, and breast cancer (the latter only among heavy drinkers) mortality rates compared to occasional drinkers. Nevertheless, the outside and during mealtime alcohol consumption groups, when compared to occasional drinkers, revealed similar estimates overall. However, the latter does not imply that the context of alcohol consumption is not associated with mortality. Possible reasons for the lack of significance for alcohol consumption outside mealtime could be the low numbers of observed deaths, small sample sizes, and age differences among the alcohol consumption groups, leading to less observed deaths in alcohol consumption groups with mainly younger participants.
The Moran’s I statistic and the LISA map of all-cancer mortality indicated evidence for five districts to be spatial clusters and spatial outliers, respectively. The fixed effects estimates of the INLA model were similar to the estimates of the generalized linear regression model. The structured spatial component revealed evidence for an increased all-cancer mortality rate for four districts in cantons Neuchâtel and Berne and a decreased rate for three districts in canton Zurich. The detected geographic variation at the district level could have been caused by differences in (diet) cultures, urbanization of the districts, or socioeconomic factors. Further studies are needed to investigate the latter associations.
Our study had limitations that could have impacted the results. First, the drinking behavior of only two 24HDR was assumed to represent the general drinking behavior of each menuCH participant. In addition, since data sets were provided at different levels, our study assumed that the menuCH participants were correctly assigned to their district and were representative for their district’s alcohol consumption. Recall bias in the menuCH study could have led to over- or underestimation of alcohol consumption. Lastly, the menuCH study is not a longitudinal study but a cross-sectional study with possible reverse causation.
An important strength of our study was the survey weighting strategy, which enabled the 2057 menuCH participants to be representative for the target population. The postal code information enabled us to link the alcohol consumption data with mortality data. Moreover, the menuCH study provided information on all participants’ baseline characteristics, enabling us to adjust for known confounders. Lastly, the questionnaire enabled us to distinguish non-drinkers, who answered alcohol avoidance (abstainer), from those, who did not (safe_no).
In conclusion, significant associations of alcohol consumption with all-cancer and UADT cancer mortality were detected, indicating an increased mortality rate with increasing alcohol intake. For the other investigated causes of death, the results pointed in the same direction, but were statistically not significant. Significant spatial dependencies were observed for all-cancer mortality, revealing Swiss districts with evidence for a lower or higher all-cancer baseline mortality rate. Lastly, the present study highlighted important descriptive differences in alcohol consumption among sexes, age groups, education and physical activity levels, BMI, and smoking categories, revealing risk groups that should be the focus of future Swiss alcohol-prevention schemes. Therefore, the results of our study indicate the need for further studies to improve the next alcohol-prevention scheme and to lower the number of avoidable deaths in Switzerland.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (PDF 809 KB)
## References
1. 1.World Health Organization (2022) The Global Health Observatory—Noncommunicable diseases. Available at https://www.who.int/data/gho/data/themes/noncommunicable-diseases. Accessed 5 Apr 2022
2. 2.Lankester T, Grills N (2019) Setting up community health and development programmes in low and middle income settings. Available at https://books.google.ch/books?hl=de&lr=&id=IMKKDwAAQBAJ&oi=fnd&pg=PP1&ots=q64w91q6i3&sig=Oi8JDe_LG051Rc0C8DLnv4FHVqE&redir_esc=y#v=onepage&q&f=false. Accessed 5 Apr 2022
3. 3.World Health Organization (2021) Noncommunicable diseases. Available at https://www.who.int/news-room/fact-sheets/detail/noncommunicable-diseases. Accessed 5 Apr 2022
4. O’Keefe JH, Bhatti SK, Bajwa A. **Alcohol and cardiovascular health: the dose makes the poison or the remedy**. *Mayo Clin Proc* (2014.0) **89** 382-393. DOI: 10.1016/j.mayocp.2013.11.005
5. Larsson SC, Wallin A, Wolk A, Markus HS. **Differing association of alcohol consumption with different stroke types: a systematic review and meta-analysis**. *BMC Med* (2016.0) **14** 1-11. DOI: 10.1186/s12916-016-0721-4
6. Reynolds K, Lewis B, David J. **Alcohol consumption and risk of stroke—a meta-analysis**. *J Am Med Assoc* (2003.0) **290** 579-588. DOI: 10.1001/jama.289.5.579
7. Ronksley PE, Brien SE, Turner BJ. **Association of alcohol consumption with selected cardiovascular disease outcomes: a systematic review and meta-analysis**. *BMJ* (2011.0) **342** d671. DOI: 10.1136/bmj.d671
8. Chiva-Blanch G, Badimon L. **Benefits and risks of moderate alcohol consumption on cardiovascular disease: current findings and controversies**. *Nutrients* (2020.0) **12** 1-19. DOI: 10.3390/nu12010108
9. Li XH, Yu FF, Zhou YH, He J. **Association between alcohol consumption and the risk of incident type 2 diabetes: a systematic review and dose-response meta-analysis**. *Am J Clin Nutr* (2016.0) **103** 818-829. DOI: 10.3945/ajcn.115.114389
10. Joosten MM, Chiuve SE, Mukama KJ. **Changes in alcohol consumption and subsequent risk of type 2 diabetes in men**. *Diabetes* (2011.0) **60** 74-79. DOI: 10.2337/db10-1052
11. Griswold MG, Fullman N, Hawley C. **Alcohol use and burden for 195 countries and territories, 1990–2016: a systematic analysis for the global burden of disease study 2016**. *The Lancet* (2018.0) **392** 1015-1035. DOI: 10.1016/S0140-6736(18)31310-2
12. Foerster M, Marques-Vidal P, Gmel G. **Alcohol drinking and cardiovascular risk in a population with high mean alcohol consumption**. *Am J Cardiol* (2009.0) **103** 361-368. DOI: 10.1016/j.amjcard.2008.09.089
13. Faeh D, Minder C, Gutzwiller F, Bopp M. **Culture, risk factors and mortality: can Switzerland add missing pieces to the European puzzle?**. *J Epidemiol Commun Health* (2009.0) **63** 639-645. DOI: 10.1136/jech.2008.081042
14. Stockwell TR, Chikritzhs TN. **Late-life increases in alcohol consumption among postmenopausal women appear associated with greater breast cancer risk and less coronary heart disease risk**. *Evid Based Med* (2016.0) **21** 195. DOI: 10.1136/ebmed-2016-110503
15. 15.International Agency for Research on Cancer (2012) Personal Habits and Indoor Combustions—IARC Monographs on the Evaluation of Carcinogenic Risks to Humans. Available at https://www.publications.iarc.fr/Book-And-Report-Series/Iarc-Monographs-On-The-Identification-Of-Carcinogenic-Hazards-To-Humans/Personal-Habits-And-Indoor-Combustions-2012. Accessed 5 Apr 2022
16. 16.World Health Organization (2018) Global status report on alcohol and health. Available at https://www.who.int/publications/i/item/9789241565639. Accessed 5 Apr 2022
17. 17.World Cancer Research Fund, American Institute for Cancer Research (2018) Alcoholic drinks and the risk of cancer. Continuous Update Project. Available at http://www.dietandcancerreport.org. Accessed 5 Apr 2022
18. Baan R, Straif K, Grosse Y. **Policy watch—carcinogenicity of alcoholic beverages**. *Lancet Oncol* (2007.0) **8** 292-293. DOI: 10.1016/s1470-2045(07)70099-2
19. 19.Federal Statistical Office (2019) BFS Aktuell—Alkoholkonsum 2017. Available at https://www.bfs.admin.ch/asset/de/10887924. Accessed 5 Apr 2022
20. 20.Gmel G (2020) Alkoholbedingte Sterblichkeit in der Schweiz im Jahr 2017. In: Sucht Schweiz. Available at https://www.bag.admin.ch/dam/bag/de/dokumente/npp/forschungsberichte/forschungsberichte-alkohol/alkoholbedingte-todesfaelle-ch-2017.pdf. Accessed 5 Apr 2022
21. Vieira BA, Luft VC, Schmidt MI. **Timing and type of alcohol consumption and the metabolic syndrome—ELSA-Brasil**. *PLoS ONE* (2016.0) **11** 1-17. DOI: 10.1371/journal.pone.0163044
22. Trevisan M, Schisterman E, Mennotti A. **Drinking pattern and mortality: the Italian risk factor and life expectancy pooling project**. *Ann Epidemiol* (2001.0) **11** 312-319. DOI: 10.1016/S1047-2797(00)00183-6
23. Lachat C, Hawwash D, Ocké MC. **Strengthening the reporting of observational studies in epidemiology—nutritional epidemiology (STROBE-nut): an extension of the STROBE statement**. *Nutr Bull* (2016.0) **41** 240-251. DOI: 10.1111/nbu.12217
24. Krieger JP, Pestoni G, Cabaset S. **Dietary patterns and their sociodemographic and lifestyle determinants in Switzerland: results from the national nutrition survey menuCH**. *Nutrients* (2019.0) **11** 1-16. DOI: 10.3390/nu11010062
25. 25.Bochud M, Chatelan A, Blanco J-M, Beer-Borst S (2017) Anthropometric characteristics and indicators of eating and physical activity behaviors in the Swiss adult population—results from menuCH 2014–2015. Available at https://www.boris.unibe.ch/101641/. Accessed 5 Apr 2022
26. 26.Federal Food Safety and Veterinary Office (2017) menuCH—Fragebogen zum Ernährungs- und Bewegungsverhalten. Available at https://www.boris.unibe.ch/101641/. Accessed 5 Apr 2022
27. 27.Pasquier J, Chatelan A, Bochud M (2017) Weighting strategy. Available at https://www.menuch.unisante.ch/index.php/catalog/4/download/17. Accessed 5 Apr 2022
28. Chatelan A, Beer-Borst S, Randriamiharisoa A. **Major differences in diet across three linguistic regions of Switzerland: results from the first national nutrition survey menuCH**. *Nutrients* (2017.0) **9** 1-17. DOI: 10.3390/nu9111163
29. 29.International Physical Activity Questionnaire Group (2004) Guidelines for data processing and analysis of the international physical activity questionnaire (IPAQ)—short form. Available at https://www.ipaq.ki.se. Accessed 5 Apr 2022
30. 30.World Health Organization (1998) MONICA_Manual_Part III: population survey, section 1: population survey data component. Available at https://www.menuch.unisante.ch/index.php/catalog/4/download/59. Accessed 7 Apr 2022
31. 31.World Health Organization (2000) WHO Technical report series—obesity: preventing and managing the global epidemic. Available at https://www.apps.who.int/iris/handle/10665/42330. Accessed 5 Apr 2022
32. Pestoni G, Krieger JP, Sych JM. **Cultural differences in diet and determinants of diet quality in Switzerland: results from the national nutrition survey menuCH**. *Nutrients* (2019.0) **11** 1-15. DOI: 10.3390/nu11010126
33. Crispim SP, de Vries JHM, Geelen A. **Two non-consecutive 24h recalls using EPIC-Soft software are sufficiently valid for comparing protein and potassium intake between five European centres—results from the European food consumption validation (EFCOVAL) study**. *Br J Nutr* (2011.0) **105** 447-458. DOI: 10.1017/S0007114510003648
34. Slimani N, Casagrande C, Nicolas G. **The standardized computerized 24-h dietary recall method EPIC-Soft adapted for pan-European dietary monitoring**. *Eur J Clin Nutr* (2011.0) **65** S5-S15. DOI: 10.1038/ejcn.2011.83
35. 35.Camenzind-Frey E, Zuberbühler C (2014) Schweizerisches Fotobuch/Livre Photo Suisse/Manuale Fotografico Svizzero. Available at https://www.iumsp.ch. Accessed 5 Apr 2022
36. Crispim SP, Nicolas G, Casagrande C. **Quality assurance of the international computerised 24 h dietary recall method (EPIC-Soft)**. *Br J Nutr* (2014.0) **111** 506-515. DOI: 10.1017/S0007114513002766
37. 37.Federal Food Safety and Veterinary Office (2021) Home—the Swiss food composition database. Available at https://www.valeursnutritives.ch/en/. Accessed 5 Apr 2022
38. Sieri S, Agudo A, Kesse E. **Patterns of alcohol consumption in 10 European countries participating in the European prospective investigation into cancer and nutrition (EPIC) project**. *Public Health Nutr* (2002.0) **5** 1287-1296. DOI: 10.1079/phn2002405
39. 39.Federal Statistical Office (2021) STAT-TAB—interaktive Tabellen (BFS)—Ständige und nichtständige Wohnbevölkerung nach institutionellen Gliederungen, Staatsangehörigkeit (Kategorie), Geschlecht und Alter. Available at https://www.pxweb.bfs.admin.ch/pxweb/de/px-x-0102010000_101/px-x-0102010000_101/px-x-0102010000_101.px. Accessed 5 Apr 2022
40. 40.World Health Organization (2019) ICD-10 version: 2019—international statistical classification of diseases and related health problems 10th revision. Available at https://www.icd.who.int/browse10/2019/en. Accessed 5 Apr 2022
41. Pestoni G, Karavasiloglou N, Braun J. **Does diet map with mortality? Ecological association of dietary patterns with chronic disease mortality and its spatial dependence in Switzerland**. *Br J Nutr* (2022.0) **127** 1037-1049. DOI: 10.1017/S0007114521001525
42. Lohse T, Faeh D, Bopp M, Rohrmann S. **Adherence to the cancer prevention recommendations of the world cancer research fund/American institute for cancer research and mortality: a census-linked cohort**. *Am J Clin Nutr* (2016.0) **104** 678-685. DOI: 10.3945/ajcn.116.135020
43. 43.Federal Statistical Office (2019) Eidgenössisches Gebäude- und Wohnungsregister—Umwandlung von PLZ in BFS-Gemeindenummer. Available at https://www.bfs.admin.ch/bfs/en/home/statistics/catalogues-databases/publications.assetdetail.7226419.html. Accessed 5 Apr 2022
44. 44.Federal Statistical Office (2019) Raumgliederungen—Resultat. Available at https://www.agvchapp.bfs.admin.ch/de/typologies/results?SnapshotDate=01.01.2019&SelectedTypologies%5B0%5D=HR_REGCH. Accessed 5 Apr 2022
45. van Buuren S, Groothuis-Oudshoorn K. **Mice: multivariate imputation by chained equations in R**. *J Stat Softw* (2011.0) **45** 1-67. DOI: 10.18637/jss.v045.i03
46. Royston P. **Multiple imputation of missing values**. *Stand Genomic Sci* (2004.0) **4** 227-241
47. Overmars KP, de Koning GHJ, Veldkamp A. **Spatial autocorrelation in multi-scale land use models**. *Ecol Model* (2003.0) **164** 257-270. DOI: 10.1016/S0304-3800(03)00070-X
48. Al-Kindi KM, Alkharusi A, Alshukaili D. **Spatiotemporal assessment of COVID-19 spread over Oman using GIS techniques**. *Earth Syst Environ* (2020.0) **4** 797-811. DOI: 10.1007/s41748-020-00194-2
49. Abrams AM, Kleinman K, Kulldorff M. **Gumbel based p-value approximations for spatial scan statistics**. *Int J Health Geogr* (2010.0) **9** 1-12. DOI: 10.1186/1476-072X-9-61
50. Besag J, York J, Mollié A. **Bayesian image restoration, with two applications in spatial statistics**. *Ann Inst Stat Math* (1991.0) **43** 1-59. DOI: 10.1007/BF00116466
51. 51.R Core Team (2021) The R project for statistical computing. Available at https://www.r-project.org/. Accessed 5 Apr 2022
52. Kushnir V, Cunningham JA. **Event-specific drinking in the general population**. *J Stud Alcohol Drugs* (2014.0) **75** 968-972. DOI: 10.15288/jsad.2014.75.968
53. Lau-Barraco C, Braitman AL, Linden-Carmichael AN, Stamates AL. **Differences in weekday versus weekend drinking among nonstudent emerging adults**. *Exp Clin Psychopharmacol* (2016.0) **24** 100-109. DOI: 10.1037/PHA0000068
54. 54.Federal Statistical Office (2017) Statistik Schweiz, Fachbereich 14—Gesundheit, Fachbereich 14–Gesundheit—Alkohol—Gewohnheitsmässiger Alkoholkonsum, 1992, 1997, 2002, 2007, 2012, 2017. Available at http://www.portal-stat.admin.ch/sgb2017/files/de/02a.xml. Accessed 5 Apr 2022
55. Reavley NJ, Jorm AF, McCann Tv, Lubman DI. **Alcohol consumption in tertiary education students**. *BioMed Central Public Health* (2011.0) **11** 1-9. DOI: 10.1186/1471-2458-11-545
56. Lourenço S, Oliveira A, Lopes C. **The effect of current and lifetime alcohol consumption on overall and central obesity**. *Eur J Clin Nutr* (2012.0) **66** 813-818. DOI: 10.1038/ejcn.2012.20
57. Sayon-Orea C, Martinez-Gonzalez MA, Bes-Rastrollo M. **Alcohol consumption and body weight: a systematic review**. *Nutr Rev* (2011.0) **69** 419-431. DOI: 10.1111/j.1753-4887.2011.00403.x
58. Dodge T, Clarke P, Dwan R. **The relationship between physical activity and alcohol use among adults in the united states: a systematic review of the literature**. *Am J Health Promot* (2017.0) **31** 97-108. DOI: 10.1177/0890117116664710
59. Musselman JRB, Rutledge PC. **The incongruous alcohol-activity association: physical activity and alcohol consumption in college students**. *Psychol Sport Exerc* (2010.0) **11** 609-618. DOI: 10.1016/j.psychsport.2010.07.005
60. Istvan J, Matarazzo JD. **Tobacco, alcohol, and caffeine use: a review of their interrelationships**. *Psychol Bull* (1984.0) **95** 301-326. DOI: 10.1037/0033-2909.95.2.301
61. Fekjær HO. **Alcohol—a universal preventive agent? A critical analysis**. *Addiction* (2013.0) **108** 2051-2057. DOI: 10.1111/add.12104
62. Goel S, Sharma A, Garg A. **Effect of alcohol consumption on cardiovascular health**. *Curr Cardiol Rep* (2018.0) **20** 1-10. DOI: 10.1007/s11886-018-0962-2
63. Knott CS, Coombs N, Stamatakis E, Biddulph JP. **All cause mortality and the case for age specific alcohol consumption guidelines: pooled analyses of up to 10 population based cohorts**. *BMJ* (2015.0) **350** 1-13. DOI: 10.1136/bmj.h384
64. Stockwell T, Zhao J, Panwar S. **Do “moderate” drinkers have reduced mortality risk? A systematic review and meta-analysis of alcohol consumption and all-cause mortality**. *J Stud Alcohol Drugs* (2016.0) **77** 185-198. DOI: 10.15288/jsad.2016.77.185
65. Fillmore KM, Kerr WC, Stockwell T. **Moderate alcohol use and reduced mortality risk: systematic error in prospective studies**. *Addict Res Theory* (2006.0) **14** 101-132. DOI: 10.1080/16066350500497983
66. 66.Norat T, Vieira AR, Abar L, et al (2017) World cancer research fund international systematic literature review—the associations between food, nutrition and physical activity and the risk of colorectal cancer. Available at https://www.wcrf.org/wp-content/uploads/2021/02/colorectal-cancer-slr.pdf
67. 67.Norat T, Chan D, Vingeliene S, et al (2017) World cancer research fund international systematic literature review—the associations between food, nutrition and physical activity and the risk of breast cancer. Available at https://www.wcrf.org/wp-content/uploads/2021/02/breast-cancer-slr.pdf. Accessed 5 Apr 2022
68. Schlesinger S, Neuenschwander M, Ballon A. **Adherence to healthy lifestyles and incidence of diabetes and mortality among individuals with diabetes: a systematic review and meta-analysis of prospective studies**. *J Epidemiol Commun Health* (2020.0) **74** 481-487. DOI: 10.1136/jech-2019-213415
69. Li Y, Mao Y, Zhang Y. **Alcohol drinking and upper aerodigestive tract cancer mortality: a systematic review and meta-analysis**. *Oral Oncol* (2014.0) **50** 269-275. DOI: 10.1016/j.oraloncology.2013.12.015
70. Naimi TS, Stockwell T, Zhao J. **Selection biases in observational studies affect associations between ‘moderate’ alcohol consumption and mortality**. *Addiction* (2017.0) **112** 207-214. DOI: 10.1111/add.13451
71. Ko H, Chang Y, Kim HN. **Low-level alcohol consumption and cancer mortality**. *Sci Rep* (2021.0) **11** 1-9. DOI: 10.1038/s41598-021-84181-1
|
---
title: 'Repeat two-stage exchange arthroplasty for recurrent periprosthetic hip or
knee infection: what are the chances for success?'
authors:
- A. C. Steinicke
- J. Schwarze
- G. Gosheger
- B. Moellenbeck
- T. Ackmann
- C. Theil
journal: Archives of Orthopaedic and Trauma Surgery
year: 2022
pmcid: PMC10030533
doi: 10.1007/s00402-021-04330-z
license: CC BY 4.0
---
# Repeat two-stage exchange arthroplasty for recurrent periprosthetic hip or knee infection: what are the chances for success?
## Abstract
### Introduction
Two-stage revision is a frequently chosen approach to treat chronic periprosthetic joint infection (PJI). However, management of recurrent infection after a two-stage exchange remains debated and the outcome of a repeat two-stage procedure is unclear. This study investigates the success rates of repeat two-stage exchange arthroplasty and analyzes possible risk factors for failure.
### Materials and methods
We retrospectively identified 55 patients (23 hips, 32 knees) who were treated with repeat resection arthroplasty and planned delayed reimplantation for recurrent periprosthetic joint infection between 2010 and 2019 after a prior two-stage revision at the same institution. The minimum follow-up was 12 months with a median follow-up time of 34 months (IQR 22–51). The infection-free survival, associated revision surgeries, and potential risk factors for further revision were analyzed using Kaplan–Meier survival curves and comparative non-parametric testing.
### Results
$78\%$ ($\frac{43}{55}$) underwent reimplantation after a repeat implant removal. Of those who completed the second-stage surgery, $37\%$ ($\frac{16}{43}$) underwent additional revision for infection and $14\%$ ($\frac{6}{55}$) underwent amputation. The reinfection-free implant survivorship amounted to $77\%$ ($95\%$ CI 64–$89\%$) after 1 year and $38\%$ ($95\%$ CI 18–$57\%$) after 5 years. Patients with a higher comorbidity score were less likely to undergo second-stage reimplantation (median 5 vs. 3, $$p \leq 0.034$$). Furthermore, obese patients ($$p \leq 0.026$$, Fisher’s exact test) and diabetics ($p \leq 0.001$, log-rank test) had a higher risk for further infection. Most commonly cultures yielded polymicrobial growth at the repeat two-stage exchange ($27\%$, $\frac{15}{55}$) and at re-reinfection ($32\%$, $\frac{9}{28}$). Pathogen persistence was observed in $21\%$ ($\frac{6}{28}$) of re-reinfected patients.
### Conclusion
The success rates after repeat two-stage exchange arthroplasty are low. Patients must be counseled accordingly and different modes of treatment should be considered.
## Introduction
Periprosthetic joint infection (PJI) is a severe complication of total joint arthroplasty and occurs in around 1–$2\%$ of primary knee or hip arthroplasties [1]. As the demand for total joint arthroplasty is on the rise due to an aging population, the revision burden due to PJI is expected to increase as well [1–3].
Two-stage exchange usually using an antibiotic-loaded spacer is considered the gold standard in treating chronic PJI [1]. A two-stage approach with removal of the implant and all foreign material, debridement, and irrigation during first-stage surgery allows for a thorough debridement of all infected tissue and can be considered appropriate for all chronic infection regardless of culture results prior to surgery, soft tissue conditions, and timing of the infection. However, despite these general advantages, two-stage revision surgery is associated with a great deal of morbidity due to the two surgeries required and the period in between stages and the rate of reinfection can be as high as $30\%$ in some cases [4–10]. Recurrence of infection can be associated with further morbidity as well as a high mortality, particularly if further surgeries are needed [11–13]. One option in these cases is to perform a repeat two-stage procedure, i.e., another sequence of explantation, spacer insertion, and delayed reimplantation of the infected prosthesis [11–13]. Nonetheless, it is debated whether a repeat two-stage exchange is adequate, as poor infection-free survival has been reported with failure rates ranging from 22 to $49\%$ [11–15]. When taken into consideration that many patients do not undergo second-stage reimplantation, failure rates might be even higher [16].
However, despite the high expected rate of reinfection, there is a scarcity of studies on the outcome of a repeat two-stage exchange and potential risk factors for failure remain unknown.
This study investigates the success rates of repeat two-stage procedures for hip and knee PJI at a single institution, analyzes microbiological findings, and describes possible risk factors for failure.
## Material and methods
The approval of the local ethics committee (2019-042-F-s Ethikkommission der Aerztekammer Westfalen-Lippe und der Westfaelischen-Wilhelms Universitaet Muenster) was obtained before initiation of this retrospective cohort study. Patients were included if they met the following criteria: history of a completed two-stage exchange arthroplasty for chronic hip or knee PJI at our institution, diagnosis of further periprosthetic joint infection of the same joint analog to the criteria published by the Musculoskeletal Infection Society (MSIS) from 2011 [17], treatment with repeat resection arthroplasty and planned delayed reimplantation at our institution between 2010 and 2019, and a minimum follow-up period of 1 year. However, patients who did undergo revision surgery or died prior to that were included [18]. Patients with prior resection of a bone tumor and subsequent infection were excluded from this study. Using our prospectively maintained institutional joint registry, we identified 305 patients who had undergone two-stage exchange arthroplasty of a hip or knee prosthesis due to chronic PJI at our institution between 2010 and 2017. Of these, 55 were treated with repeat resection arthroplasty and planned delayed reimplantation between 2010 and 2019 due to reinfection. The median follow-up period was 34 months (interquartile range (IQR) 22–51).
Success of the repeated two-stage exchange arthroplasty was defined following the Delphi-based consensus definition that includes healed wounds, no further surgical procedure for infection, and no PJI-related mortality [19].
Data regarding the patients’ surgical history, clinical course, medication, and preexisting comorbidities were collected from electronic files. An age-adjusted Charlson Comorbidity Index (CCI) was calculated for each patient [20]. Microbiological findings are presented for the initial two-stage infection, second two-stage, and potential further revisions. Infections were classified as persistent rather than new infections, if at least one pathogen that was cultured at the explantation stage of the preceding two-stage exchange arthroplasty was cultured again at the subsequent two-stage exchange or any further revision for PJI failure [21].
Patient demographics of the study cohort at the time of the repeat two-stage exchange are showcased in Tables 1 and 2. The median age at the time of the repeat two-stage exchange was 73 years (IQR 64–78). An overview of the patients’ course and outcomes is provided in Fig. 1.Table 1Patient demographics for patients with or without further revision for reinfection after a repeat two-stage exchangeVariableEntire study cohortReinfectionNo reinfectionp (Fisher’s exact test)Female$49\%$ ($\frac{27}{55}$)$48\%$ ($\frac{13}{27}$)$52\%$ ($\frac{14}{27}$)0.790TKA$58\%$ ($\frac{32}{55}$)$53\%$ ($\frac{17}{32}$)$47\%$ ($\frac{15}{32}$)0.787Diabetes mellitus$40\%$ ($\frac{22}{55}$)$68\%$ ($\frac{15}{22}$)$32\%$ ($\frac{7}{22}$)0.054Obesity$64\%$ ($\frac{35}{55}$)$63\%$ ($\frac{22}{35}$)$37\%$ ($\frac{13}{35}$)0.026Chronic kidney disease$27\%$ ($\frac{15}{55}$)$53\%$ ($\frac{8}{15}$)$47\%$ ($\frac{7}{15}$)1.000Hypertension$85\%$ ($\frac{47}{55}$)$51\%$ ($\frac{24}{47}$)$49\%$ ($\frac{23}{47}$)1.000Heart disease$51\%$ ($\frac{28}{55}$)$57\%$ ($\frac{16}{28}$)$43\%$ ($\frac{12}{28}$)0.423Depression$25\%$ ($\frac{14}{55}$)$50\%$ ($\frac{7}{14}$)$50\%$ ($\frac{7}{14}$)1.000COPD$18\%$ ($\frac{10}{55}$)$50\%$ ($\frac{5}{10}$)$50\%$ ($\frac{5}{10}$)1.000Anticoagulation$29\%$ ($\frac{16}{55}$)$38\%$ ($\frac{6}{16}$)$63\%$ ($\frac{10}{16}$)0.245Tobacco use$20\%$ ($\frac{11}{55}$)$27\%$ ($\frac{3}{11}$)$73\%$ ($\frac{8}{11}$)0.101Additional DAIR before repeat two-stage revision$25\%$ ($\frac{14}{55}$)$43\%$ ($\frac{6}{14}$)$57\%$ ($\frac{8}{14}$)0.547p values for Fisher’s exact testSignificant differences are marked in boldTable 2Patient demographics for patients with or without further revision for reinfection after a repeat two-stage exchangeVariableEntire study cohortPatients reinfectedPatients not reinfectedp (Mann–Whitney U test)Age at second PJI73 (IQR 64–78)72 (IQR 64–78)73 (IQR 64–79)0.655Body mass index31.4 (IQR 27.7–35.4)32 (IQR 30–38)29 (IQR 26–33)0.040Age-adjusted CCI4 (IQR 3–5)4 (IQR 3–5)4 (IQR 3–5)0.314Months between first and second PJI13 (IQR 4–32)16 (IQR 5–35)10 (IQR 2–26)0.162Number of previous surgeries before second PJI5 (IQR 4–7)5 (IQR 4–7)5 (IQR 4–7)0.939p values for Mann–Whitney U testSignificant differences are marked in boldFig. 1Flowchart of the treatment course and final outcome
## Surgical approach
The first stage of a two-stage exchange arthroplasty included the thorough removal of all foreign material and bradytrophic tissue, irrigation, and if needed resection of osteomyelitic bone. Three to seven tissue samples were obtained and sent to our hospital’s institute for microbiology, where they were cultured for 7–14 days on Columbia blood agar, Schaedler agar, and Chocolate agar. Concluding the procedure, handmade interim spacers were implanted. Articulating spacers were used for hips, whereas knees underwent implantation of static spacers, both of which were made from polymethylmethacrylate (PMMA) bone cement (Copal G + C or Palacos G + C, Heraeus medical, Wehrheim, Germany) and had 5–$10\%$ of its weight added in antibiotics based on resistance testing (usually 2 g Vancomycin per 40 g cement, 4 g of Meropenem for gram negatives, 600 mg of Voriconazole, or 200 mg of Amphotericin for fungal organisms). In some individual cases, definitive resection arthroplasty of the hip (Girdlestone) was performed, usually if it was likely prior to first-stage surgery that the patient did not wish for any further surgeries due to limited life expectancy, cognitive disorders, or severe comorbidities.
Joint aspiration was not routinely performed before reimplantation given the expected poor sensitivity and specificity [22]. Systemic antibiotic therapy based on the infecting organism was administered for at least two weeks intravenously, followed by four weeks of oral treatment. If inflammatory serum markers (C-reactive protein and interleukin 6) had declined and soft tissues were healed, reimplantation was planned [23]. Eradication of infection was defined based on the criteria defined by Diaz-Ledezma et al. [ 19] citing healed wounds, absence of PJI-related mortality, and no further revision surgery for infection. For early (within 4 weeks after the last surgery) or acute infections, a DAIR procedure was performed, which includes irrigation, debridement, antibiotic therapy, and the retention of the implant with component exchange, while for later, chronic infections or after a failed debridement procedure, removal of the prosthesis and a repeat staged revision was recommended.
The implants for reimplantation were chosen in consideration of individual factors, such as age, defect size, and bone quality. In three cases, an arthrodesis implant was selected for the second-stage reimplantation. During the reimplantation surgery, deep tissue samples were obtained and sent for microbiological analysis. After reimplantation was completed, antibiotics were administered for 2 weeks in case of negative intraoperative culture results, and for six weeks if positive cultures at the time of reimplantation were obtained.
## Statistical analysis
We pseudonymized all patient data before conducting any statistical analysis.
Descriptive statistics were investigated for data distribution and categorical variables. Means and ranges are used to report parametric data, whereas non-parametric data are displayed using medians and interquartile ranges (IQR). Differences in groups of binary variables were compared using Fisher’s exact test, whereas metric variables were compared using the Mann–Whitney U test for non-parametric distributions or student’s t test for parametric distributions of data. Re-reinfection, i.e., failure of the repeat two-stage exchange arthroplasty defined by the Delphi-based criteria [19], was set as the primary outcome measure. Secondary outcome measures were as follows: completion of the repeat two-stage exchange arthroplasty with reimplantation, and amputation.
Implant survival was assessed using the Kaplan–Meier survival analysis with $95\%$ confidence intervals (CI) presented [24]. Differences in survival were compared using the log-rank test [25]. Again, primary endpoint was the diagnosis of re-reinfection in line with the *Delphi consensus* criteria [19].
All tests were two tailed with an alpha level of $5\%$ considered significant.
## Reimplantation rates
Among the 55 patients who underwent the first stage of a planned repeat two-stage exchange, five knee patients ($16\%$, $\frac{5}{32}$) and seven hip patients ($30\%$, $\frac{7}{23}$) did not complete the two-stage procedure with reimplantation of the prosthesis, making a total of twelve patients ($22\%$, $\frac{12}{55}$) who did not undergo reimplantation. Among these, six are considered reinfected or persistently infected, while in the other six patients the infection is considered eradicated.
Of the six patients who remained infection-free after the explantation stage, two knee patients have died of unrelated cause before the planned reimplantation surgery and four hip patients were successfully treated with a Girdlestone resection arthroplasty due to low functional demand and high surgical risk and remained infection-free until the last follow-up.
Among the patients with persisting infection or reinfection, three patients (two knees, one hip) underwent amputation, one hip patient died from sepsis that was most likely PJI related, one knee patient who was treated with Girdlestone resection arthroplasty underwent a subsequent debridement procedure due to deep tissue reinfection, and one patient with a retained hip spacer had recurrent wound infection.
Patients who did not undergo second-stage reimplantation had a higher median age-adjusted CCI (5 vs. 3 ($$p \leq 0.034$$)). Furthermore, patients with chronic kidney disease (CKD) ($$p \leq 0.068$$) and patients with a shorter time between the first two-stage and reinfection ($$p \leq 0.061$$) appeared to be at increased risk to not undergo reimplantation, although no statistical significance could be ascertained with the numbers available.
## Additional surgical performances
$25\%$ ($\frac{14}{55}$) of patients (8 knees, 6 hips) underwent debridement, irrigation and exchange of the mobile and modular implant components prior to an eventual repeat implant removal. These patients, however, did not have a different outcome than the patients who had not undergone an additional DAIR procedure regarding the rate of re-reinfection after repeat two-stage exchange ($$p \leq 0.547$$) or the length of infection-free implant survival ($$p \leq 0.750$$).
During the prosthesis-free interval of the repeat two-stage procedure, ten patients (4 knees, 6 hips) underwent a singular spacer exchange procedure and one hip patient underwent three consecutive spacer exchange procedures, resulting in a total of $20\%$ of patients ($\frac{11}{55}$) having undergone at least one spacer exchange. The reason for a spacer exchange was spacer dislocation in four cases and clinical signs of persisting infection in nine cases. In our hospital, a spacer exchange due to mechanical reasons, like dislocation, was indicated if the spacer caused pain or disability, threatened neurovascular structures, or could potentially cause skin necrosis, analog to the criteria postulated at the International Consensus Meeting in 2018 [26, 27]. Spacer exchange due to persisting infection was indicated following the Delphi-based consensus criteria [19]. While the cultures taken during spacer exchange were culture negative in $45\%$ ($\frac{5}{11}$) of spacer exchanges, a different, new organism was cultured during $55\%$ ($\frac{6}{11}$) procedures compared to the first-stage surgery.
## Re-reinfection rates and infection-free implant survival after reimplantation
Ultimately, $78\%$ ($\frac{43}{55}$) of patients underwent reimplantation after a median time of 89 days (IQR 71–143). However, further revision for re-reinfection was performed in $51\%$ ($\frac{22}{43}$) of cases (14 knees, 8 hips) after reimplantation. To treat re-reinfection, a total of 15 DAIR procedures, twelve repeat two-stage exchange procedures, and nine amputation surgeries were performed in this subgroup. One patient refused further surgical treatment of his reinfected prosthesis and thus remained with a fistula until last follow-up.
For the 43 cases in which the repeat two-stage procedure was completed, a Kaplan–*Meier analysis* of the infection-free implant survival was conducted (Fig. 2). After 1 year, the cumulative infection-free survival probability was $77\%$ ($95\%$ CI 64–$89\%$), $65\%$ after 2 years ($95\%$ CI 50–$79\%$), and $38\%$ after 5 years ($95\%$ CI 18–$57\%$).Fig. 2Kaplan–Meier survival curve for reinfection-free survivorship after completed repeat two-stage revision Patients suffering from diabetes mellitus had a significantly lower infection-free survival rate than patients without diabetes mellitus ($33\%$ ($95\%$ CI 9–$57\%$) vs $82\%$ ($95\%$ CI 67–$96\%$) after 2 years, $p \leq 0.001$).
The infection-free survival was not different between TKA and THA ($70\%$ ($95\%$ CI 53–$88\%$) vs $56\%$ ($95\%$ CI 31–$80\%$) after 2 years, $$p \leq 0.483$$).
## Overall re-reinfection rates, amputation rates, and mortality
Among all 55 patients who underwent first-stage resection arthroplasty, $51\%$ ($\frac{28}{55}$) ultimately suffered from re-reinfection. This includes seven amputation surgeries in the subgroup of knee PJI ($22\%$, $\frac{7}{32}$) and two amputations that were performed on hip patients ($9\%$, $\frac{2}{23}$), resulting in a total amputation rate of $16\%$ ($\frac{9}{55}$) in our cohort. The overall rate of re-reinfection was not different between hip ($48\%$, $\frac{11}{23}$) and knee ($53\%$, $\frac{17}{32}$) arthroplasties ($$p \leq 0.787$$). However, the rate of re-reinfection was significantly higher in obese patients (body mass index ≥ 30 kg/m2) ($$p \leq 0.026$$).
Overall, $29\%$ ($\frac{16}{55}$) of patients have died during the follow-up period after a mean time of 32 months (range 1–92). One-year mortality was $11\%$ ($\frac{6}{55}$) and 30-day mortality was $4\%$ ($\frac{2}{55}$).
## Aseptic complications
Aseptic complications after the repeat two-stage exchange arthroplasty occurred in seven cases: two patients underwent revision for aseptic TKA loosening, one femoral and tibial each (8 and 10 months postoperatively), and five patients suffered hip dislocation and underwent closed reduction in all cases at a median follow-up of 3 months.
## Microbiological findings
The microbiological findings at the explantation stage of the first and the repeat two-stage exchange surgeries and at re-reinfection are presented in Table 3.Table 3Causative organisms at each infectionOrganismFirst two-stage exchangeSecond two-stage exchangeRe-reinfectionCoagulase-negative staphylococci17 ($31\%$)14 ($25\%$)5 ($18\%$)Staphylococcus aureus10 ($18\%$)4 ($7\%$)4 ($14\%$)Streptococci4 ($7\%$)4 ($7\%$)1 ($4\%$)Gram-negative bacteria3 ($5\%$)10 ($18\%$)5 ($18\%$)Candida spp.0 ($0\%$)1 ($2\%$)0 ($0\%$)Enterococci0 ($0\%$)1 ($2\%$)1 ($4\%$)Anaerobic3 ($5\%$)0 ($0\%$)1 ($4\%$)Culture-negative6 ($11\%$)6 ($11\%$)1 ($4\%$)Polymicrobial10 ($18\%$)15 ($27\%$)9 ($32\%$)Missing2 ($4\%$)0 ($0\%$)1 ($4\%$)Total55 ($100\%$)55 ($100\%$)28 ($100\%$) While coagulase-negative staphylococci, mostly Staphylococcus epidermidis, were the most common finding during surgery for the first episode of PJI ($31\%$, $\frac{17}{55}$), polymicrobial findings increased with the second and third PJI, being the most reported microbiological finding at reinfection ($27\%$, $\frac{15}{55}$) and re-reinfection ($32\%$, $\frac{9}{28}$). However, with the numbers available, there was no difference in polymicrobial infections between the first, second, and third PJI ($$p \leq 0.320$$).
In $9\%$ ($\frac{5}{55}$) of all cases (two knees, three hips), the microorganism isolated at the explantation stage of the first two-stage exchange was persistent at the explantation stage of the second two-stage exchange. $67\%$ ($\frac{37}{55}$) of reinfections were caused by a new pathogen that had not been isolated at the explantation stage of the first two-stage exchange. $24\%$ ($\frac{13}{55}$) of the patients in this study cohort were culture negative at either one or both two-stage exchanges or, in two cases, had missing information about the microbiological findings at the first two-stage exchange. The persistence rates among knee patients ($6\%$, $\frac{2}{32}$) and hip patients ($13\%$, $\frac{3}{23}$) do not statistically differ ($$p \leq 0.639$$).
Of all 28 patients who are considered re-reinfected after the repeat two-stage exchange arthroplasty, $68\%$ ($\frac{19}{28}$) of re-reinfections are considered new infections. A persistent pathogen was observed in $21\%$ ($\frac{6}{28}$) of re-reinfected patients. Two patients ($7\%$, $\frac{2}{28}$) were culture negative at the second two-stage exchange and one patient’s microbiological findings at re-reinfection are missing, as the diagnosis has been made at an outside hospital. Knee patients had a higher persistence rate of $24\%$ ($\frac{4}{17}$) than hip patients with a rate of $18\%$ ($\frac{2}{11}$), however, without a significant statistical difference between the two groups ($$p \leq 1.000$$).
$28\%$ ($\frac{13}{55}$) of patients had positive cultures during reimplantation surgery at the first episode of PJI. During the second two-stage procedure, $37\%$ ($\frac{16}{43}$) of patients had positive cultures during reimplantation. However, this difference was not significant between the two episodes ($$p \leq 0.182$$).
With the numbers available, no microbiological factors (culture-negative infection, positive culture at reimplantation, or polymicrobial infection) were associated with the risk of re-reinfection or reduced implant survivorship.
## Discussion
Two-stage exchange arthroplasty remains the gold standard for the treatment of chronic periprosthetic joint infection. Still, infection recurrence is observed in $6\%$ to $33\%$ of the cases [4–10] and patients suffering from reinfection often face a long course of multiple additional surgeries with a potentially poor outcome [28]. However, given the relatively low number of repeat two-stage exchange arthroplasty procedures reported, data regarding the outcome of patients undergoing this procedure are scarce. This study provides additional data about the success rate of repeat two-stage exchange procedures in a single tertiary revision arthroplasty center and identifies possible risk factors associated with further revision surgery as well as microbiological insights with a high percentage of polymicrobial infections in repeat two-stage procedures and frequent positive cultures during second-stage reimplantation. In our cohort, the probability to remain infection-free amounted to only around $65\%$ after 2 years and more than half the patients in this cohort ultimately suffered from re-reinfection following this procedure. We found that obese patients and diabetics were at increased risk for further infection and reduced infection-free implant survival, respectively, and patients with a higher comorbidity score were more likely not to undergo second-stage reimplantation.
Prior studies have reported re-reinfection rates after repeat two-stage exchange arthroplasty ranging from 22.2 to $49\%$ [11–15]. While the comparability to previous studies is somewhat limited by different demographic and surgical details of the patients included, the failure rate presented in our study is relatively high. It should be noted that surgeons must consider various factors in the success of two-stage procedures that are difficult to account for in the retrospective study designs of our and related studies.
Kheir et al. investigated 60 patients who underwent further surgical intervention after a failed two-stage exchange of infected THA or TKA, including 26 patients, with a repeat two-stage procedure that was successful in $62\%$ of patients [11]. The authors note a higher risk for reinfection after a debridement procedure with retention of the prosthesis compared to a repeat two-stage procedure and propose a more aggressive approach with repeat staged revision. While we also noted that $25\%$ of patients who ultimately underwent a repeat two-stage procedure had a failed attempt to retain the implant, they did not have a worse outcome after the two-stage procedure. Nonetheless, considering that a DAIR procedure is far less invasive than a complete implant removal, we suggest that if early (within 4 weeks) or acute reinfection occurs, patients should be counseled regarding the high risk of failure after a DAIR procedure, but if soft tissue conditions and the patients general state allows for it, it can be an option to avoid further staged revision.
Khan et al. investigated repeat two-stage procedures of infected THA and reported a success rate of $57\%$ at two-year follow-up in 42 patients, which is comparable with our study results [12]. With $33\%$ of patients deceased before the two-year follow-up, the mortality presented in their cohort also corresponds with our results. Khan et al. concluded that an underlying host problem might be the cause for poor outcome regardless of treatment modality and recommended further research on host optimization and other treatment methods for repeat PJI. Considering the equally poor results of repeat two-stage exchanges in our cohort, we agree that patients should be informed regarding this dismal perspective and a mutual decision should be made if further staged revision is planned.
Considering the poor outcome and high risk for re-revision in repeat two-stage procedures in our and the aforementioned studies, there is an urgent need for further investigation on alternative treatment options for repeat PJI.
The retention of a spacer can be a good option in selected patients [29, 30]. Nevertheless, a risk for recurrent infection should be taken into account, particularly if one considers that in the present study, even if no reimplantation is performed, there is a high risk of further revision surgeries for infection.
Amputation is considered a last resort treatment that is usually chosen if the infection cannot be controlled with any other surgical or medical treatment option. The possibility of curing the infection comes at the cost of low functional status and high mortality [31–33]. Nevertheless, amputation can still yield good patient satisfaction [33].
With Girdlestone resection arthroplasty, partial limb function can be preserved, while the infected prosthesis or spacer is removed. However, the rates of complication, mortality, and reoperation are reported to be high [34]. On the other hand, a conversion back to a THA can achieve satisfactory results, if the patient is deemed suitable [35]. This also applies to arthrodesis of the knee [35]. Wu et al. recommended performing arthrodesis as the treatment of choice after a failed two-stage exchange of TKA [36]. In our cohort, only three patients were treated with an arthrodesis instead of a second stage reimplantation and all three subsequently suffered from re-reinfection. Because of the low occurrence in our study, however, we cannot make any general statement concerning the outcome after knee arthrodesis.
Chronic antibiotic suppression can be an option for high surgical risk patients or in cases with large bone defects in which reconstruction would be likely to fail [35]. However, surgeons must consider long-term side effects of suppression therapy and the risk of uncontrolled infection. Furthermore, Leitner et al. investigated the mortality rate of patients with these salvage procedures and found a long-term mortality of $44\%$ at a median follow-up of 8 years. They recommended a chronic fistula instead of further resection arthroplasty [37]. Regarding repeated revisions in TKA, they recommended considering amputation early after multiple failed revision arthroplasties [37]. Surgeons must always consider that these patients can have a very limited life expectancy and high mortality and further surgical intervention must be weighed regarding the high risk of failure. Nevertheless, if a repeat two-stage exchange for PJI is planned, these patients should also be counseled regarding alternative approaches, but further research is much needed to determine the value of different treatment modalities with regard to reinfection rates and quality of life.
In order to potentially optimize host factors and assess individual risk, patient- and procedure-related predisposing factors are relevant. However, to our knowledge there are no studies that specifically investigate risk factors in the setting of a repeat two-stage exchange. In the present study, diabetics had a lower implant survival probability. Although this factor has not been described for repeat two-stage revisions, diabetes mellitus is a known risk factor for infection of total hip or knee arthroplasty [38]. Additionally, it has been shown that poorly controlled diabetes appears to be linked to even worse outcomes [39] and in many cases diabetes and obesity are connected [40]. Considering that Fehring et al. reported an increase of obese patients from $30.4\%$ in 1990 to $52.1\%$ in 2005 [41], not only the risk for infection after primary arthroplasty but also the risk for further infections must be expected to increase if re-revision is performed. It appears warranted to emphasize diabetes control and to address related complications, such as foot ulcers early on as they might pose as a source of PJI [42]. Future studies should investigate the effect of potential host optimization on the outcome of two-stage procedures and repeat revision.
Polymicrobial infections were most common in patients with repeat two-stage revision but were not associated with a worse outcome. However, the high prevalence of polymicrobial infections must be taken into account to plan the systemic treatment. Contrary to this, Tan et al. found that polymicrobial PJI has a worse outcome and higher PJI-related mortality than PJI caused by one organism or culture-negative PJI, although this was done in patients with first PJI [43]. In our study cohort, most of the patients suffering re-reinfection after the repeat two-stage exchange arthroplasty presented with a new pathogen. Zmistowski et al. suggested that especially patients with poor health status and a high number of comorbidities are at risk for reinfection with new organisms because of their higher disposition to infections rather than due to a failed treatment [21]. However, we found that in this study the percentage of persisting infections increased from the second PJI to re-reinfection. Zmistowski et al. found that Staphylococcal organisms, especially methicillin-resistant Staphylococcus aureus, have a high risk to be persistent in recurrent PJI. Further research on pathogen persistence rates, microbiological resistance patterns, and antimicrobial treatment could provide interesting additional insight.
This study’s findings must be interpreted considering several limitations: Our study data were collected at a single institution and therefore might include bias that comes from local preferences or routines. There may have also been a selection bias since our study cohort only consists of patients who were deemed fit enough for a repeat two-stage exchange arthroplasty; however, in our practice generally further surgery was recommended if reinfection was present. Nonetheless, it is possible that patients had undergone surgery elsewhere or that suppression treatment has been initiated. Furthermore, for some statistical analysis there might be sparse data bias as the number of risk factors and number of events are limited and a multivariate analysis is not helpful in this situation. Nevertheless, we think that our findings are a valuable addition to the existing research on this rare issue.
## Conclusion
Repeat two-stage revision for recalcitrant infection after a two-stage procedure for PJI leads to further infection in around $50\%$ of patients at mid-term follow-up. Particularly, obese patients and diabetics are at high risk of reinfection. Surgeons should be aware of the high percentage of polymicrobial infections encountered at a repeat two-stage procedure and plan antibiotic treatment accordingly. Furthermore, particularly in comorbid patients an additional share of patients did not reach second-stage reimplantation; therefore, alternative strategies should be discussed in order to avoid re-revisions in this challenging group of patients.
## References
1. Kapadia BH, Berg RA, Daley JA. **Periprosthetic joint infection**. *Lancet* (2016) **387** 386-394. DOI: 10.1016/S0140-6736(14)61798-0
2. Murphy BPd, Dowsey MM, Spelman T. **The impact of older age on patient outcomes following primary total knee arthroplasty**. *Bone Joint J* (2018) **100-B** 1463-1470. DOI: 10.1302/0301-620X.100B11.BJJ-2017-0753.R6
3. Helwig P, Morlock J, Oberst M. **Periprosthetic joint infection–effect on quality of life**. *Int Orthop* (2014) **38** 1077-1081. DOI: 10.1007/s00264-013-2265-y
4. Berend KR, Lombardi AV, Morris MJ. **Two-stage treatment of hip periprosthetic joint infection is associated with a high rate of infection control but high mortality**. *Clin Orthop Relat Res* (2013) **471** 510-518. DOI: 10.1007/s11999-012-2595-x
5. Wolf M, Clar H, Friesenbichler J. **Prosthetic joint infection following total hip replacement: results of one-stage versus two-stage exchange**. *Int Orthop* (2014) **38** 1363-1368. DOI: 10.1007/s00264-014-2309-y
6. Pangaud C, Ollivier M, Argenson J-N. **Outcome of single-stage versus two-stage exchange for revision knee arthroplasty for chronic periprosthetic infection**. *EFORT Open Rev* (2019) **4** 495-502. DOI: 10.1302/2058-5241.4.190003
7. Kunutsor SK, Whitehouse MR, Lenguerrand E. **Re-Infection outcomes following one- and two-stage surgical revision of infected knee prosthesis: a systematic review and meta-analysis**. *PLoS ONE* (2016) **11** e0151537. DOI: 10.1371/journal.pone.0151537
8. Kunutsor SK, Whitehouse MR, Blom AW. **Re-infection outcomes following one- and two-stage surgical revision of infected hip prosthesis: a systematic review and meta-analysis**. *PLoS ONE* (2015) **10** e0139166. DOI: 10.1371/journal.pone.0139166
9. Sabry FY, Buller L, Ahmed S. **Preoperative prediction of failure following two-stage revision for knee prosthetic joint infections**. *J Arthroplasty* (2014) **29** 115-121. DOI: 10.1016/j.arth.2013.04.016
10. Barton CB, Wang DL, An Q. **Two-stage exchange arthroplasty for periprosthetic joint infection following total hip or knee arthroplasty is associated with high attrition rate and mortality**. *J Arthroplasty* (2020) **35** 1384-1389. DOI: 10.1016/j.arth.2019.12.005
11. Kheir MM, Tan TL, Gomez MM. **Patients with failed prior two-stage exchange have poor outcomes after further surgical intervention**. *J Arthroplasty* (2017) **32** 1262-1265. DOI: 10.1016/j.arth.2016.10.008
12. Khan N, Parmar D, Ibrahim MS. **Outcomes of repeat two-stage exchange hip arthroplasty for prosthetic joint infection**. *Bone Joint J* (2019) **101-B** 110-115. DOI: 10.1302/0301-620X.101B6.BJJ-2018-1556.R1
13. Azzam K, McHale K, Austin M. **Outcome of a second two-stage reimplantation for periprosthetic knee infection**. *Clin Orthop Relat Res* (2009) **467** 1706-1714. DOI: 10.1007/s11999-009-0739-4
14. Brown TS, Fehring KA, Ollivier M. **Repeat two-stage exchange arthroplasty for prosthetic hip re-infection**. *Bone Joint J* (2018) **100-B** 1157-1161. DOI: 10.1302/0301-620X.100B9.BJJ-2018-0470.R1
15. Fehring KA, Abdel MP, Ollivier M. **Repeat two-stage exchange arthroplasty for periprosthetic knee infection is dependent on host grade**. *J Bone Joint Surg Am* (2017) **99** 19-24. DOI: 10.2106/JBJS.16.00075
16. Gomez MM, Tan TL, Manrique J. **The fate of spacers in the treatment of periprosthetic joint infection**. *J Bone Joint Surg Am* (2015) **97** 1495-1502. DOI: 10.2106/JBJS.N.00958
17. Parvizi J, Zmistowski B, Berbari EF. **New definition for periprosthetic joint infection: from the Workgroup of the Musculoskeletal Infection Society**. *Clin Orthop Relat Res* (2011) **469** 2992-2994. DOI: 10.1007/s11999-011-2102-9
18. Xu C, Tan TL, Li WT. **Reporting outcomes of treatment for periprosthetic joint infection of the knee and hip together with a minimum 1-year follow-up is reliable**. *J Arthroplasty* (2020) **35** 1906-1911.e5. DOI: 10.1016/j.arth.2020.02.017
19. Diaz-Ledezma C, Higuera CA, Parvizi J. **Success after treatment of periprosthetic joint infection: a Delphi-based international multidisciplinary consensus**. *Clin Orthop Relat Res* (2013) **471** 2374-2382. DOI: 10.1007/s11999-013-2866-1
20. Yang C-C, Fong Y, Lin L-C. **The age-adjusted Charlson comorbidity index is a better predictor of survival in operated lung cancer patients than the Charlson and Elixhauser comorbidity indices**. *Eur J Cardiothorac Surg* (2018) **53** 235-240. DOI: 10.1093/ejcts/ezx215
21. Zmistowski B, Tetreault MW, Alijanipour P. **Recurrent periprosthetic joint infection: persistent or new infection?**. *J Arthroplasty* (2013) **28** 1486-1489. DOI: 10.1016/j.arth.2013.02.021
22. Hoell S, Moeller A, Gosheger G. **Two-stage revision arthroplasty for periprosthetic joint infections: what is the value of cultures and white cell count in synovial fluid and CRP in serum before second stage reimplantation?**. *Arch Orthop Trauma Surg* (2016) **136** 447-452. DOI: 10.1007/s00402-015-2404-6
23. Hoell S, Borgers L, Gosheger G. **Interleukin-6 in two-stage revision arthroplasty: what is the threshold value to exclude persistent infection before re-implanatation?**. *Bone Joint J* (2015) **97-B** 71-75. DOI: 10.1302/0301-620X.97B1.33802
24. Kaplan EL, Meier P. **Nonparametric estimation from incomplete observations**. *J Am Stat Assoc* (1958) **53** 457-481. DOI: 10.1080/01621459.1958.10501452
25. Mantel N. **Evaluation of survival data and two new rank order statistics arising in its consideration**. *Cancer Chemother Rep* (1966) **50** 163-170. PMID: 5910392
26. Kozaily E, Chisari E, Parvizi J. **Is there a role for spacer exchange in two-stage exchange arthroplasty for periprosthetic joint infection?**. *J Clin Med* (2020). DOI: 10.3390/jcm9092901
27. Abdel MP, Barreira P, Battenberg A. **Hip and knee section, treatment, two-stage exchange spacer-related: proceedings of international consensus on orthopedic infections**. *J Arthroplasty* (2019) **34** 427-438. DOI: 10.1016/j.arth.2018.09.027
28. Shahi A, Tan TL, Chen AF. **In-hospital mortality in patients with periprosthetic joint infection**. *J Arthroplasty* (2017) **32** 948-952.e1. DOI: 10.1016/j.arth.2016.09.027
29. Choi H-R, Freiberg AA, Malchau H. **The fate of unplanned retention of prosthetic articulating spacers for infected total hip and total knee arthroplasty**. *J Arthroplasty* (2014) **29** 690-693. DOI: 10.1016/j.arth.2013.07.013
30. Lee K-J, Min B-W, Bae K-C. **Unintended retention of temporary articulating spacers in the treatment of periprosthetic hip joint infection**. *Orthopedics* (2020) **43** e251-e257. DOI: 10.3928/01477447-20200404-08
31. Böhle S, Rohe S, Kirschberg J. **Poor outcome of above-knee amputation after septic failure of revision total knee arthroplasty**. *J Knee Surg* (2020). DOI: 10.1055/s-0040-1716416
32. Fedorka CJ, Chen AF, McGarry WM. **Functional ability after above-the-knee amputation for infected total knee arthroplasty**. *Clin Orthop Relat Res* (2011) **469** 1024-1032. DOI: 10.1007/s11999-010-1577-0
33. Orfanos AV, Michael RJ, Keeney BJ. **Patient-reported outcomes after above-knee amputation for prosthetic joint infection**. *Knee* (2020) **27** 1101-1105. DOI: 10.1016/j.knee.2019.10.007
34. Malcolm TL, Gad BV, Elsharkawy KA. **Complication, survival, and reoperation rates following girdlestone resection arthroplasty**. *J Arthroplasty* (2015) **30** 1183-1186. DOI: 10.1016/j.arth.2015.02.011
35. Fagotti L, Tatka J, Salles MJC. **Risk factors and treatment options for failure of a two-stage exchange**. *Curr Rev Musculoskelet Med* (2018) **11** 420-427. DOI: 10.1007/s12178-018-9504-1
36. Wu CH, Gray CF, Lee G-C. **Arthrodesis should be strongly considered after failed two-stage reimplantation TKA**. *Clin Orthop Relat Res* (2014) **472** 3295-3304. DOI: 10.1007/s11999-014-3482-4
37. Leitner L, Posch F, Amerstorfer F. **The dark side of arthroplasty: competing risk analysis of failed hip and knee arthroplasty with periprosthetic joint infection**. *J Arthroplasty* (2020) **35** 2601-2606.e1. DOI: 10.1016/j.arth.2020.04.078
38. Maradit Kremers H, Lewallen LW, Mabry TM. **Diabetes mellitus, hyperglycemia, hemoglobin A1C and the risk of prosthetic joint infections in total hip and knee arthroplasty**. *J Arthroplasty* (2015) **30** 439-443. DOI: 10.1016/j.arth.2014.10.009
39. Wang S-H, Xu C, Tan TL. **Increased postoperative glucose variability is associated with adverse outcome following two-stage exchange arthroplasty for periprosthetic joint infection**. *J Arthroplasty* (2020) **35** 1368-1373. DOI: 10.1016/j.arth.2019.11.046
40. Dowsey MM, Choong PFM. **Obese diabetic patients are at substantial risk for deep infection after primary TKA**. *Clin Orthop Relat Res* (2009) **467** 1577-1581. DOI: 10.1007/s11999-008-0551-6
41. Fehring TK, Odum SM, Griffin WL. **The obesity epidemic: its effect on total joint arthroplasty**. *J Arthroplasty* (2007) **22** 71-76. DOI: 10.1016/j.arth.2007.04.014
42. Rakow A, Perka C, Trampuz A. **Origin and characteristics of haematogenous periprosthetic joint infection**. *Clin Microbiol Infect* (2019) **25** 845-850. DOI: 10.1016/j.cmi.2018.10.010
43. Tan TL, Kheir MM, Tan DD. **Polymicrobial periprosthetic joint infections: outcome of treatment and identification of risk factors**. *J Bone Joint Surg Am* (2016) **98** 2082-2088. DOI: 10.2106/JBJS.15.01450
|
---
title: 'The quality of antiretroviral medicines: an uncertain problem'
authors:
- Ngan Thi Do
- Phonepasith Boupha
- Paul N Newton
- Céline Caillet
journal: BMJ Global Health
year: 2023
pmcid: PMC10030546
doi: 10.1136/bmjgh-2022-011423
license: CC BY 4.0
---
# The quality of antiretroviral medicines: an uncertain problem
## Abstract
### Objectives
Substandard and falsified (SF) antiretrovirals (ARVs) risk poor outcomes and drug resistance, potentially affecting millions of people in need of treatment and prevention. We assessed the available evidence on SF ARV and related medical devices to discuss their potential public health impact.
### Methods
Searches were conducted in Embase, PubMed, Google, Google Scholar, Web of Science and websites with interest in ARV quality in English and French up to 30 November 2021. Publications reporting on the prevalence of SF ARV were assessed in a quantitative analysis using the Medicine Quality Assessment Reporting Guidelines (MEDQUARG).
### Results
We included 205 publications on SF ARV and 11 on SF medical devices. Nineteen prevalence surveys of SF ARV, published between 2003 and 2021, were included, with no surveys relevant to SF medical devices. The prevalence survey sample size ranged from 3 to 2630 samples (median (Q1–Q3): 16.0 (10.5–44.5); 3 ($15.8\%$) used random outlet sampling methods. Of the 3713 samples included in the prevalence surveys, $1.4\%$ ($$n = 51$$) failed at least one test. Efavirenz, nevirapine and lamivudine-nevirapine-stavudine combination were the most surveyed ARV with failure frequencies of $3.6\%$ ($\frac{7}{193}$), $2.6\%$ ($\frac{5}{192}$) and $2.8\%$ ($\frac{5}{177}$), respectively. The median (Q$1\%$–Q$3\%$) concordance with the MEDQUARG criteria was $42.3\%$ ($34.6\%$–$55.8\%$).
### Conclusion
These results suggest that there are few data in the public domain of the quality of ARV in supply chains; the proportion of SF ARV is relatively low in comparison to other classes of essential medicines. Even a low proportion of the ARV supply chain being poor quality could make a large difference in the HIV/AIDS international landscape. The 95-95-95 target for 2026 and other international targets could be greatly hampered if even $1\%$ of the millions of people taking ARV (for both prevention and prophylaxis) receive medicines that do not meet quality standards. More surveillance of SF ARV is needed to ensure issues are detected.
## Introduction
Antiretrovirals (ARVs) are primarily used for the treatment and prevention of infection by the human immunodeficiency virus (HIV).1 According to the WHO, approximately 38.4 million people were living with HIV at the end of 20212 and by July 2022, the HIV/AIDS had caused 40.1 million deaths globally.2 Approximately 850 children became infected with HIV and approximately 310 children died each day in 2021 from AIDS-related causes.3 Globally, $75\%$ of HIV-infected people were receiving antiretroviral therapy (ART) at the end of 2021.2 4 With no cure or vaccine currently available, access to quality ART is crucial to control the infection and help prevent transmission. The WHO estimated that between 2000 and 2019 ARV saved 15.3 million lives and reduced the percentage of new HIV infections by $39\%$ and HIV-related deaths by $51\%$.5 HIV drug resistance (HIVDR) affects the efficacy of ART, resulting in increased HIV-associated morbidity and mortality and transmission. According to surveys conducted in 10 countries in sub-Saharan Africa (2012–2020), nearly one-half of infants born to mothers infected with HIV presented with HIVDR to one or more non-nucleoside reverse transcriptase inhibitors (NNRTIs), one of the key classes of medicines for treatment and prevention of HIV transmission.6 7 Minimising the spread of HIVDR is critical to ensure long-term efficacy and durability of ARV.
The global ARV drugs market value exceeded US$ 24.7 billion in 2018.8 Projections suggest that it will be US$ 22.5 billion by 2024.
Substandard (due to within factory or supply chain errors) and falsified (due to fraud) (substandard and falsified, SF) medical products of all therapeutic classes have been found in many countries.9 10 The WHO estimated that around $10.5\%$ of medical products are SF in L/MIC, with an estimated US%30.5 billion financial loss.11 A variety of defects have been found in SF medicines. They may contain one or several unexpected toxic active ingredients, too low or too high amounts of the expected active ingredients, they may contain none of the expected active ingredient(s) and they may also fail to dissolve properly, hence preventing the active ingredient(s) from reaching the blood stream, thus losing their efficacy. Hence, SF represent a serious public health problem. They also have a significant impact on clinical practice and the economy, and they generate loss of confidence in healthcare professionals and healthcare systems.11 Antibiotics and antimalarials are the most studied classes of medicines.12–15 A recent systematic review of the scientific literature showed that $17.4\%$ of the 13 555 antibiotics tested for quality failed at least one quality test.13 In another systematic review, $15.4\%$ of the 3414 medicines used for cardiovascular diseases failed at least one quality test.15 In both reviews, samples were mainly collected from low-income and middle-income countries and the number of samples tested per country was relatively small compared with the amount of medicines used globally. There is little scientific evidence publicly available on the quality of medicines available in high-income countries but the number and types of recalls by regulatory authorities show that these countries are not immune.16–20 Good quality ARVs are vital in the management of HIV infection and AIDS. The high number of people affected, the cost, the length of treatment and impaired access raise the risk of ARV falsification. Cases of SF ARV have been identified over the past decades and ARV are often quoted as medicines with common/recurring quality issues.21–24 However, as far as, we are aware there is no clear understanding on the epidemiology of SF ARV globally. This systematic review was conducted with the key objective to summarise the available evidence on ARV medicines quality globally, to discuss their potential impact for patients and society.
## Search strategy
Search terms relevant to pharmaceutical quality (eg, ‘falsified’, ‘substandard’) were combined with search terms relevant to ARV and HIV/AIDS (online supplemental file 1). Systematic searches were conducted in Embase, PubMed, Google, Google Scholar and Web of Science in English and French up to 30 November 2021. The search terms were adapted for searches in MRA websites, and other websites with interest in medicines quality in English and French (online supplemental file 2). The articles from the first 20 pages of Google search results were screened for eligibility. Titles and abstracts were first screened and full texts of the identified articles were then assessed for eligibility. A manual search of the reference lists of the included articles was performed. Articles identified in previous systematic reviews by our group that included ARV medicines but not captured in our searches were also included.
## Eligibility criteria
Scientific articles and grey literature in English or French assessing or discussing the quality of ARV medicines were included. Articles containing scientific data on the prevalence of ARV medicines quality were the most relevant publications for this review. Other scientific articles included studies describing new tests or validation of innovative techniques to determine the quality of medicines in which ARV medicine samples were used to validate the technique, equivalence studies and quality control analyses. We also included reports of seizures, recalls, alerts by the MRAs or pharmaceutical companies and patients describing adverse reactions where the quality of the medicine was suspected to be the cause. The different types of study included in this review are described in online supplemental file 3.
We excluded data from publications describing the development/validation of analysis technique(s) for quality assessment of ARV medicines without sufficient information on the samples used and publications on the quality of herbal/mineral/animal part remedies claimed to treat HIV/AIDS.
We included medical devices for the diagnosis of HIV.
## Key definitions
Following the 2017 WHO definitions, falsified medicines are those that ‘deliberately/fraudulently misrepresent their identity, composition or source’.25 Substandard medicines are ‘authorised medical products that fail to meet either their quality standards or their specifications, or both’.25 This may result from negligence/errors during the manufacturing process or degradation through deterioration because of inappropriate storage/transport in the supply chain. There is inadequate evidence to distinguish poor quality medicines resulting from errors during the manufacturing process from subsequent degradation in the supply chain due to heat and humidity.
Pharmaceutical analysis relies on compendial tests described in pharmacopoeial monographs. For finished medicines, monographs commonly include the identification and quantification of Active Pharmaceutical Ingredient (API) content (using sophisticated standardised techniques such as liquid chromatography coupled with various detectors), dissolution testing, detection of specific levels of predetermined impurities/related substances, uniformity of dosage units and additional attributes depending on the formulation of the product (eg, tablet friability). In many studies included in this review, not all pharmacopoeial analyses were conducted and also a variety of non-pharmacopoeial assays were used, for example, for investigating specific contaminants or unstated APIs. Assay details were not always provided making it difficult to standardise the definition of a ‘failed sample’. Consequently, we define a failed sample as one for which at least one quality analysis test performed by the investigators gave a fail result, irrespective of the number and type of assays used.
As it is not possible to reliably classify a medicine as substandard or falsified without packaging analysis, products without packaging authentication that failed at least one quality test (ie, the results are outside the acceptable limits of the chosen specifications reference, either pharmacopoeia monograph or in-house specifications) are defined as ‘substandard or falsified’ (SorF).14 However, all samples that contained incorrect or no API were assumed to be falsified, although there is a (limited) risk of misclassification of such samples as falsified when they are substandard, due to gross manufacturing errors.
As in previous systematic reviews by our group,13 15 26 we define ‘failure frequency’ (FF) as the proportion of samples included in a prevalence survey that failed at least one quality test described in the report. We define a ‘data point’ as a specific location where medicines were collected for quality analysis, at a given time and for a given study. For medicines purchased online the location where the samples were received was extracted.
## Data collection
Data were manually extracted into the ‘Online Medicine Quality Data Manager’, an online data entry tool developed by the Infectious Diseases Data Observatory (IDDO) Informatics and the Lao-Oxford-Mahosot-Wellcome Trust Research Unit Medicine Quality team. Publication type (eg, report, original research article), year of publication, sampling type, location (country and city, where available) and type of outlet where samples were collected, the total number of samples collected, API/API combination name, number of samples failing medicine quality test(s), quality defect and the techniques that were used to analyse samples were entered in the online tool.
## Data analysis
FlySpeed SQL Query (V.3.5.4.2) was used to extract data from the online database and Microsoft Excel 2013 was used for data analysis. Qualitative variables were expressed as numbers and percentages (n (%)). Quantitative variables were expressed as median with first and third quartiles (Q1 and Q3, respectively).
## Quality of studies assessment: Medicine Quality Assessment Reporting Guidelines
The methodology and reporting of prevalence surveys were evaluated using the Medicine Quality Assessment Reporting Guidelines (MEDQUARG). MEDQUARG is a checklist of 26 items that should be included in reports of medicine quality surveys.27 *All criteria* had to be fulfilled for each item to be awarded one point. Prevalence surveys were assessed independently by two reviewers with a third person resolving any disagreement. Only the prevalence surveys published as original articles in scientific journals, following the Introduction/Methods/Results/Discussion section or similar style and published as reports or PhD thesis, were assessed.
This review was registered in the International Prospective Register for Systematic Review (PROSPERO, Registration No: CRD42016039531) and is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (online supplemental file 4).
## Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
## Overall literature on ARV medicines quality
After duplicates removal, 21 462 out of 25 880 publications gathered through electronic searches were screened by title and abstract (figure 1).
**Figure 1:** *PRISMA flow chart of the selection process of the publications on antiretroviral medicines quality. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.*
In total, 216 publications were included in this review, of which more than half were original research articles ($57.9\%$ ($$n = 125$$)) and $13.9\%$ ($$n = 30$$) were lay press (figure 2). Most original research articles ($89.6\%$, $\frac{112}{125}$) were published in peer-reviewed journals. The number of publications related to ARV medicines quality per year was low between 1990 and 2003, reached a peak in 2016 ($$n = 28$$ publications) and then decreased (figure 2).
**Figure 2:** *Number of publications per type and year of publication. (Note: publications published up to the 30 November 2021 only were included).*
Of the 216 publications, 205 were on ARV medicines quality and eleven on the quality of medical devices used in HIV. Of the 205 publications on ARV quality, 76 ($37.1\%$) described the quality of ARV medicines in a specific location at a specific time with a total of 455 data points, and 129 ($62.9\%$) did not contain data point information. No publication on medical devices for HIV diagnosis contained data on their quality in a specific location at a specific time. Out of 76 publications with data points, 19 ($25.0\%$) were prevalence studies, 15 ($19.7\%$) analytical technique development/validation, 8 ($10.5\%$) routine quality control analysis, 4 ($5.3\%$) equivalence studies, 1 ($1.3\%$) bioavailability study and the data from the United States Pharmacopoeia's (USP) Medicines Quality Database were also included as one publication ($1.3\%$) (online supplemental file 5). Others were recall/warning/alerts ($$n = 16$$), seizures ($$n = 7$$) and case reports ($$n = 5$$) published in newspapers or medicines regulatory authorities websites.
A total of 4898 samples were collected and tested for quality, mainly in prevalence surveys ($$n = 3713$$, $75.8\%$) and routine MRA quality control analysis ($$n = 766$$, $15.6\%$). Of all samples, 59 ($1.2\%$) failed at least one quality test. Of the failing samples, 54 ($91.5\%$) were classified as SorF because no packaging analysis to assess the authenticity of the samples had been performed, 5 ($8.5\%$) were substandard and no samples were classified as falsified.
All data are mapped and can be downloaded on the IDDO Medicine Quality Surveyor system (https://www.iddo.org/mqsurveyor/%23antiretrovirals).
## Prevalence surveys
Nineteen prevalence surveys published between 2003 and 2021 were included. Overall 3713 samples of 22 different APIs or combinations of APIs were collected in 21 countries (168 data points) on 4 continents. The sample size per study ranged from 3 to 2630 samples with a median (Q1–Q3) of 16.0 (10.5–44.5) samples per prevalence survey. The overall FF in prevalence surveys was $1.4\%$ ($\frac{51}{3}$,713). Of the failing samples, 47 ($92.2\%$) were classified as SorF, 4 ($7.8\%$) were substandard and no samples were classified as falsified.
Three prevalence surveys used random sampling to select the outlets to be included (FF $2.1\%$, $\frac{9}{419}$), 14 used convenience sampling only (FF $1.2\%$, $\frac{38}{3}$,247), 1 used mixed random and convenience sampling designs (FF $0.0\%$, $\frac{0}{42}$), and the sampling strategy was not described in one survey (FF $80.0\%$, $\frac{4}{5}$) (online supplemental file 6).
We found no publicly available evidence for $\frac{174}{195}$ ($89.2\%$) of national states. About three-fourths ($75.8\%$, $$n = 2813$$/3713) of samples in prevalence surveys were collected from low-income countries, $18.7\%$ ($$n = 695$$/3,713) and $0.1\%$ ($$n = 37$$/3,713) were collected in middle-income and high-income countries, respectively (table 1). One hundred and sixty-eight samples ($4.5\%$) were part of a large multicountry study but the FF were not broken down by country. Over $90\%$ ($\frac{3675}{3713}$) of samples included in prevalence surveys were procured in Africa and Asia, representing $97.0\%$ ($\frac{3603}{3713}$) and $1.9\%$ ($\frac{72}{3713}$) of all the samples, respectively.
**Table 1**
| Continent | Income | Country | No of publications | No of data points | Failure frequency % (n/N) |
| --- | --- | --- | --- | --- | --- |
| Americas | | | | | 11.8 (2/17) |
| | HIC | USA | 1.0 | 6.0 | 12.5 (2/16) |
| | UMIC | Jamaica | 1.0 | 1.0 | 0.0 (0/1) |
| Europe | | | | | 9.5 (2/21) |
| | HIC | Lithuania | 2.0 | 5.0 | 40.0 (2/5) |
| | HIC | UK | 1.0 | 3.0 | 0.0 (0/16) |
| Asia | | | | | 2.8 (2/72) |
| | UMIC | China | 1.0 | 3.0 | 33.3 (1/3) |
| | LMIC | Cambodia | 1.0 | 1.0 | 14.3 (1/7) |
| | LMIC | India | 1.0 | 7.0 | 0.0 (0/17) |
| | UMIC | Thailand | 1.0 | 3.0 | 0.0 (0/3) |
| | Unknown | Unknown* | 1.0 | 8.0 | 0.0 (0/42) |
| Africa | | | | | 1.2 (45/3603) |
| | LIC | Ethiopia | 1.0 | 4.0 | 25.0 (1/4) |
| | LMIC | Senegal | 2.0 | 9.0 | 14.5 (8/55) |
| | UMIC | South Africa | 3.0 | 9.0 | 9.1 (1/11) |
| | LMIC | Nigeria | 2.0 | 11.0 | 5.7 (4/70) |
| | LIC | DR Congo | 2.0 | 11.0 | 3.9 (2/51) |
| | LMIC | Zambia | 5.0 | 17.0 | 3.1 (2/65) |
| | LMIC | Cameroon | 2.0 | 11.0 | 1.4 (1/69) |
| | LIC | Tanzania | 3.0 | 23.0 | 0.9 (24/2707) |
| | Unknown | Unknown† | 1.0 | 1.0 | 0.8 (1/126) |
| | LMIC | Kenya | 3.0 | 26.0 | 0.3 (1/394) |
| | LIC | Uganda | 2.0 | 9.0 | 0.0 (0/51) |
| Total | | | 19.0 | 168.0 | 1.4 (51/3713) |
The FF was the highest in the Americas ($11.8\%$, $\frac{2}{17}$), followed by Europe ($9.5\%$, $\frac{2}{21}$), but the total number of samples tested was low. The FF was $1.2\%$ ($\frac{45}{3603}$) in Africa and $2.8\%$ ($\frac{2}{72}$) in Asia. The highest number of samples was collected in Tanzania ($$n = 2707$$), with an FF of $0.9\%$ ($\frac{24}{2707}$).
The proportion of samples of Efavirenz collected in prevalence surveys was the highest ($5.2\%$, $\frac{193}{3713}$) with FF=$3.6\%$ ($\frac{7}{193}$), followed by nevirapine ($5.2\%$, $\frac{192}{3713}$) with FF=$2.6\%$ ($\frac{5}{192}$) and lamivudine-nevirapine-stavudine combination ($3.8\%$, $\frac{177}{3713}$) with FF=$2.8\%$ ($\frac{5}{177}$), respectively (table 2).
**Table 2**
| API/API combination | No of publications | No of data points | Failure frequency % (n/N) |
| --- | --- | --- | --- |
| Ritonavir | 1 | 2 | 100.0 (2/2) |
| Indinavir | 4 | 6 | 42.9 (6/14) |
| Lopinavir-ritonavir | 4 | 5 | 18.2 (8/44) |
| Lamivudine-zidovudine-nevirapine | 3 | 3 | 8.2 (7/85) |
| Stavudine | 6 | 13 | 4.2 (4/96) |
| Efavirenz | 10 | 23 | 3.6 (7/193) |
| Lamivudine-nevirapine-stavudine | 7 | 14 | 2.8 (5/177) |
| Nevirapine | 12 | 24 | 2.6 (5/192) |
| Zidovudine | 7 | 18 | 1.9 (2/103) |
| Lamivudine | 6 | 18 | 1.5 (2/132) |
| Lamivudine-zidovudine | 5 | 11 | 1.5 (2/134) |
| Antiretroviral-unspecified | 2 | 2 | 0.0 (1/2,325) |
| Abacavir | 3 | 3 | 0.0 (0/33) |
| Abacavir-lamivudine | 1 | 1 | 0.0 (0/1) |
| Amprenavir | 1 | 1 | 0.0 (0/1) |
| Didanosin | 4 | 4 | 0.0 (0/20) |
| Efavirenz-lamivudine-tenofovir disiproxil | 1 | 1 | 0.0 (0/29) |
| Emtricitabine-efavirenz-tenofovir disiproxil | 2 | 2 | 0.0 (0/28) |
| Emtricitabine-tenofovir disoproxil | 2 | 4 | 0.0 (0/30) |
| Lamivudine-stavudine | 4 | 5 | 0.0 (0/43) |
| Saquinavir | 1 | 2 | 0.0 (0/2) |
| Tenofovir disoproxil | 3 | 5 | 0.0 (0/25) |
| Tenofovir disoproxil-lamivudine | 1 | 1 | 0.0 (0/3) |
| Total | 19 | 168 | 1.4 (51/3713) |
The FF of samples of ritonavir was the highest ($100.0\%$, $\frac{2}{2}$), followed by that of indinavir ($42.9\%$, $\frac{6}{14}$) but only few samples were tested.
Most of samples collected in prevalence surveys were tested for more than one quality attributes ($93.8\%$, $\frac{3483}{3713}$). Fourteen samples ($1.4\%$, $\frac{14}{1034}$) failed the API content test and 8 samples ($1.3\%$, $\frac{8}{616}$) failed the dissolution test. No sample ($0.0\%$, $\frac{0}{495}$) failed impurity/contaminant/related substances tests (online supplemental file 7).
Six samples out of 3256 (0.2 %) failed visual inspection of sample units (shape/colour uniformity, presence of contamination etc) and/or non-comparative packaging analysis (check of the availability of specific information and in some cases the conformity to packaging and labelling requirements with reference to MRA guidelines) in prevalence surveys. Of 14 samples that failed API content tests, $50.0\%$ ($\frac{7}{14}$) contained lower API amount than stated, $42.9\%$ ($\frac{6}{14}$) higher API amount and for 1 sample (7.1 %, $\frac{1}{14}$) there was not enough information in the publication to determine whether it contained higher or lower amounts of API. Twelve out of 19 studies used High-Performance Liquid Chromatography (HPLC) methods (coupled with various detectors) for analysing API content ($79.6\%$, $\frac{823}{1034}$ samples).
The USP was the most commonly used (in $\frac{13}{19}$ studies), followed by the British Pharmacopoeia and the International Pharmacopoeia (in 5 and 4 studies, respectively) (online supplemental file 6).
The highest FF was observed in samples collected from private pharmacies ($28.0\%$, $\frac{7}{25}$), followed by hospital/health centres ($19.0\%$, $\frac{8}{98}$), websites ($7.7\%$, $\frac{2}{26}$) and other government outlets ($6.3\%$, $\frac{1}{16}$) (online supplemental file 8). In total, 1302 samples were collected in multiple types of facilities with an FF of $2.2\%$ ($\frac{29}{1302}$) but results of the quality tests were not given by outlet type. In additional, 2200 samples included in one study were collected in Tanzanian ports of entry with FF $0.0\%$ ($\frac{0}{2200}$). For 21 samples, there was no information on the health facility where the samples were collected.
For the majority of the samples ($93.3\%$ ($\frac{3464}{3713}$)) included in prevalence surveys, there were no details on the stated manufacturer, or no breakdown of the samples by country of origin of the manufacturer (online supplemental file 9). The FF of the samples stated as made by Asian manufacturers ($6.4\%$, $\frac{238}{3713}$), was of $3.8\%$ ($\frac{9}{238}$). The FF of samples stated as made by American manufacturers was the highest ($14.3\%$, $\frac{1}{7}$).
The median (Q$1\%$–Q$3\%$) concordance with MEDQUARG items of 15 prevalence surveys that met the inclusion criteria for appraisal using MEDQUARG was $42.3\%$ ($34.6\%$–$55.8\%$) (figure 3, online supplemental file 10).
**Figure 3:** *Percentage of concordance of the 15 prevalence surveys with the 26 items included in MEDQUARG checklist. MEDQUARG, Medicine Quality Assessment Reporting Guidelines.*
## Quality of studies assessment
Although 10 surveys were reported after the publication of the MEDQUARG in 2009, none stated that the MEDQUARG guidelines were followed to report the findings. Three ($20.0\%$) studies reported how the sample collectors presented to the seller (whether covert shopper, and what the sampler said/asked the seller) and 4 ($26.7\%$) outlined the sampling design with sufficient details (online supplemental file 10). Only $40.0\%$ ($\frac{6}{15}$) of the studies provided definitions on the quality of medicines or recognised the WHO definition. In $33.3\%$ ($\frac{5}{15}$) of the surveys, the samples were clearly categorised as genuine, falsified or substandard or another equivalent terminology (or an explanation of the reason why this was not done); $33.3\%$ ($\frac{5}{15}$) stated whether medicines were registered with the government in the location(s) sampled. Sixty per cent ($\frac{10}{15}$) of the studies reported with sufficient details the relationship between packaging and chemistry results. The MRA of the sampled country(ies) was either involved in the study (a representative of the MRA being an author in the paper) or was stated to be informed of its findings in four studies ($26.7\%$).
## Seizures, recalls, case reports
Twenty-eight publications describing recalls/warning/alerts ($$n = 16$$), seizures ($$n = 7$$) and case reports ($$n = 5$$) of SF ARV medicines were found during our searches (online supplemental file 11). Recalls of products of 14 APIs/combinations due to dissolution failure, API content or impurity/contaminant were found. In addition, 10 recalls/warning/alerts and seizures of HIV diagnostic test kit and HIV viral load for diagnostic test were identified (online supplemental file 12). Those include the substitution of 140 000 HIV rapid diagnostic test (RDT) kits by urinary pregnancy tests or resale of just-past expiry kits in India,28 and recall of one million of HIV testing kits in Kenya out of concern that they give false negative results.29 Other publications included in our review are listed in online supplemental file 13.
## Discussion
We synthesised the publicly available evidence on the quality of ARV medicines from different publicy accessible sources. Overall, $1.4\%$ of 3713 ARV samples collected in 21 countries failed at least 1 quality test in the 19 prevalence studies. The limited sample sizes of the studies impede interpretation of the results. Drawing conclusions on the impact of SF ARV for patients and the community is also rendered difficult by the limited reporting of the findings in the various prevalence surveys, and often by the bias generated by their limited methodology, as described by others.30 31 The observed FF in this review is lower than the $4.2\%$ ($\frac{43}{1}$,018) failure rate described in a recent review of the literature of studies conducted between 2007 and 2016 by the WHO.11 One recent study may result in underestimating the FF.32 *In this* study, from which more than half of the samples (2630 samples) described in the current review originated, 2200 samples collected at ports of entry in Tanzania over 4 years passed the Global Pharma Health Fund(GPHF)-Minilab initial screening tests, which included simple visual inspection of dosage units, API identification by thin-layer chromatography and disintegration tests. These 2200 samples were not further tested by reference testing in the laboratory. However, the same report describes that $10\%$ samples of samples collected in other health structures that passed GPHF-Minilab screening were further tested using laboratory reference testing, resulting in an FF of $3\%$. Though the GPHF-Minilab has shown good performances to identify falsified samples containing none of the stated API, its sensitivity to identify substandard medicines containing lower or higher amounts of API is much lower.33 If the same $3\%$ FF was applied to the 2200 samples collected in ports, the FF in this review would have been more than double ($3.1\%$ ($\frac{117}{3713}$)).
SF ‘HIV/hepatitis medicines’ represented $\frac{43}{1500}$ ($2.9\%$) of rapid alerts of reports to the Global Surveillance and Monitoring System between 2013 and 2017.9 Although ARVs are often quoted as one of the most affected products, together with other anti-infectives, the FF for ARV estimated here falls below that of other classes of medicines described in previous systematic reviews using the same methodology, such as for antibiotics (FF of $17.4\%$ ($\frac{2357}{13}$ 555)) and cardiovascular medicines (FF of $15.4\%$ ($\frac{525}{3414}$)).13 15 In those reviews, samples were frequently procured in private sector’s facilities such as retail pharmacies, unlike in the current review in which an FF of $28.0\%$ was observed in samples collected from private retail pharmacies, but only 25 samples were collected. ARV are often procured in LMIC within public or NGO vertical programmes which often follow stringent quality assurance systems and procure only WHO-prequalified medicines. However, in 2011 in Kenya nurses identified a falsified version of the ARV Zidolam-N, a WHO prequalified product, in Médecins Sans Frontières supplies relabelled fraudulently to extend its expiry date.34 The most common quality defects observed in prevalence surveys were lower or higher API content than stated on the label, failed dissolution tests (either too rapid or too slow), and impurity/contaminant/related substances tests. API in higher concentrations than expected risks not only poor outcomes to patients, but also lack of adherence through more frequent side effects. Using ARV medicines with too low API content and/or poor dissolution may lead to treatment failure, prolonged illness or death, and risks engendering the spread of drug resistant pathogens, although, as far as we are aware, the link between SF ARV and the emergence and spread of resistance has not been demonstrated.35 We found no publicly available evidence for almost $90\%$ of national states, and for 17 of the 30 countries that bear $89\%$ of the new HIV infections,36 which indicates an important lack of oversight of the risks. We found no study on the quality of dolutegravir, though this might be due to its only recent recommendation for use by the WHO (in combination with two NNRTIs) for newly diagnosed HIV patients.37 We also found limited information on tenofovir-based oral combinations recommended in 2015 by the WHO for pre-exposure prophylaxis (PrEP).38 An increasing number of countries are including self-testing of HIV in their national policies. Cases of SF RDTs show the importance of postmarket surveillance of diagnostic kits. However, no studies trying to better understand the extent of quality issues of RDTs were identified.
Due to convenience, increasing accessibility to, perceived economical and confidential advantages of the internet, especially in the context of HIV/AIDS associated stigma and discriminations, online purchase of ARV is likely to increase. This may be particularly relevant to people searching for oral PrEP when at high risk of infection. In 2020, 130 countries had adopted the WHO recommendations on oral PrEP in national guidelines.39 Only two prevalence studies described the quality of ARV purchased on the internet, with too few samples collected to comment on the results.40 41
## Limitations
Searches were conducted only in English and French, risking the exclusion of articles, for example in Latin America, and we identified recalls/seizures/case reports mainly from searches in a limited number of MRA’s websites and other websites interested in medicine quality. Unpublished postmarketing surveillance results from other MRAs and the pharmaceutical industry were not captured. Most studies were of small sample size and used convenient sampling which risk bias. The quality of reporting of prevalence surveys was poor as reflected by the low MEDQUARG scores. The quality of samples was assessed by different pharmacopoeia references. In most prevalence surveys, we found limited information on stated country of manufacture and more than one-third of the samples were collected in one study in different outlets but no details on the quality of the samples by type of outlet were given. We, thus, did not perform further analysis that could lead to misleading interpretation.
The diversity of and the often poor methodology and reporting of the studies renders the findings of systematic reviews of medicine quality difficult to interpret and extrapolate,30 31 though we believe it is the best method to summarise the current evidence on the quality of different classes of medicines.
## Recommendations
There are clear gaps in the understanding of the epidemiology of SF ARV and related diagnostic tests. Initiatives such as the Distributed Pharmaceutical Analysis Laboratory (DPAL), a collaboration established between 30 academic institutions around the world to determine the quality of medicines collected from partner organisations in L/MICs, may facilitate better understanding of the epidemiology of SF medicines and other medical products.42 Although packaging analysis is difficult, especially in obtaining voucher samples, it is vital to allow the objective distinction between substandard and falsified products. That $92.2\%$ of failing samples were classified as SorF is a major impediment for deciding on policy as interventions to counter substandard and falsified differ.
Key current global public health aims are the 95-95-95 target of the Sustainable Development Goals by 2026 and to end AIDS by 2030.36 43 Diagnosing $95\%$ and achieving viral suppression in $95\%$ of all HIV-positive individuals risks failure even if only $1\%$ of the ARV/RDT available on the market do not fulfil their roles because they are poor quality. With millions of people being treated or using ARV for the prevention of HIV, even a small proportion of poor quality ARV with impaired efficacy or increased toxicity will greatly endanger the lives of millions, not only those treated, but also those who may be infected as a result of transmission from people using SF ARV. A related issue is concern about the quality of condoms, with many incidents and seizures of tons of falsified condoms with holes,44–47 but the extent of the problem is also unknown. Gaps in the scientific evidence impede development of objective action plans on how best to secure the supply chains for ARV, RDT and other medical devices such as condoms. With the current goals set by international actors to scale up community based approaches for both treatment and prevention, such as community drug distribution, safeguards to ensure quality ARV and RDT will be crucial. More efforts also need to be put into controlling the quality of medicines available on the internet.
Shortages of good quality ARV create opportunities for substandard and falsified ARV medicines to reach supply chains. Shortages are exacerbated during the COVID-19 pandemic, as land, sea and air transport services shut down. People had difficulties to access ARV because of travel restrictions, disruptions in health services within countries and worsening of the economic situation because of the pandemic.48 Better preparedness is needed for the next pandemic, for medical products to treat the pandemic’s causing agent and for other medical products vital to millions such as ARV.
In view of the limitations described above, prevalence surveys with robust survey methodology adequate sample sizes, and better reporting of findings, in wider geographical regions including HIC and online sales are needed for a more comprehensive epidemiological information on the quality of ARV medicines. This would allow examination of trends over time and the impact of SF ARV on humans and their economy.
## Conclusion
Even a small proportion of SF ARV is unacceptable, as it may result in a myriad of HIV positive people not receiving the correct treatment, risking poor outcomes and resistance, and those using ARV as prophylaxis unknowingly being unprotected against infection. These results cannot represent an exact prevalence of poor quality ARV drugs globally but are a warning sign. The methodological limitations do not allow exptrapolation that $1.4\%$ of ARV globally are SF. There is clearly a risk and more data on the epidemiology of SF ARV, facilitation of packaging analysis and optimisation of devices for their screening of SF products in supply chains are needed.
## Data availability statement
All data relevant to the study are included in the article or uploaded as online supplemental information. All data are mapped and can be downloaded on the Infectious Diseases Data Observatory Medicine Quality Surveyor system(https://www.iddo.org/mqsurveyor/%23antiretrovirals).
## Patient consent for publication
Not applicable.
## References
1. Saag MS, Benson CA, Gandhi RT. **Antiretroviral drugs for treatment and prevention of HIV infection in adults: 2018 recommendations of the International antiviral society-USA panel**. *JAMA* (2018.0) **320** 379-96. DOI: 10.1001/jama.2018.8431
2. **HIV**
3. **HIV statistics - global and regional trends - UNICEF DATA 2022**
4. **Global HIV & AIDS statistics — 2020 fact sheet**
5. **HIV/AIDS**. (2021.0)
6. **Fact sheet: HIV drug resistance**
7. **HIV drug resistance report 2021**
8. **Global industry trends, share, size, growth, opportunity and forecast 2019-2024**
9. 9
World Health Organization
. WHO global surveillance and monitoring system for substandard and falsified medical products. 2017.. *WHO global surveillance and monitoring system for substandard and falsified medical products* (2017.0)
10. Newton PN, Bond KC. **Global access to quality-assured medical products: the oxford statement and call to action**. *Lancet Glob Health* (2019.0) **7** e1609-11. DOI: 10.1016/S2214-109X(19)30426-7
11. **A study on the public health and socioeconomic impact of substandard and falsified medical products**. *WHO* (2018.0) 4-8
12. Tabernero P, Fernández FM, Green M. **Mind the gaps -- the epidemiology of poor-quality anti-malarials in the malarious world -- analysis of the worldwide antimalarial resistance network database**. *Malar J* (2014.0) **13**. DOI: 10.1186/1475-2875-13-139
13. Zabala GA, Bellingham K, Vidhamaly V. **Substandard and falsified antibiotics: neglected drivers of antimicrobial resistance?**. *BMJ Glob Health* (2022.0) **7**. DOI: 10.1136/bmjgh-2022-008587
14. Saraswati K, Sichanh C, Newton PN. **Quality of medical products for diabetes management: a systematic review**. *BMJ Glob Health* (2019.0) **4**. DOI: 10.1136/bmjgh-2019-001636
15. Do NT, Bellingham K, Newton PN. **The quality of medical products for cardiovascular diseases: a gap in global cardiac care**. *BMJ Glob Health* (2021.0) **6**. DOI: 10.1136/bmjgh-2021-006523
16. Machado Reis AT, Berardo BFR, Loureiro R. **Quality of medicines in Portugal: a retrospective review of medicine recalls (2000-2015)**. *PDA J Pharm Sci Technol* (2018.0) **72** 44-9. DOI: 10.5731/pdajpst.2017.007567
17. Naughton BD, Akgul E. **Medicine quality in high-income countries: the obstacles to comparative prevalence studies**. *Med Access Point Care* (2021.0) **5** 23992026211052270. DOI: 10.1177/23992026211052272
18. Almuzaini T, Sammons H, Choonara I. **Quality of medicines in Canada: a retrospective review of risk communication documents (2005-2013)**. *BMJ Open* (2014.0) **4**. DOI: 10.1136/bmjopen-2014-006088
19. Almuzaini T, Sammons H, Choonara I. **Substandard and falsified medicines in the UK: a retrospective review of drug alerts (2001-2011)**. *BMJ Open* (2013.0) **3**. DOI: 10.1136/bmjopen-2013-002924
20. AlQuadeib BT, Alfagih IM, Alnahdi AH. **Medicine recalls in Saudi Arabia: a retrospective review of drug alerts (January 2010–january 2019)**. *Futur J Pharm Sci* (2020.0) **6** 91. DOI: 10.1186/s43094-020-00112-3
21. Penzak SR, Acosta EP, Turner M. **Antiretroviral drug content in products from developing countries**. *Clin Infect Dis* (2004.0) **38** 1317-9. DOI: 10.1086/383575
22. 22
World Health Organization
. Medicines regulatory authorities: current status and the way forward. 2006.. *Medicines regulatory authorities: current status and the way forward* (2006.0)
23. **Check combivir, serostim**. *Epogen*
24. **[E-drug] counterfeit arvs in DRC**. (2004.0)
25. 25
World Health Organization
. Member state mechanism on substandard/spurious/falsely-labelled/falsified/ counterfeit medical products. 2017.. *Member state mechanism on substandard/spurious/falsely-labelled/falsified/ counterfeit medical products* (2017.0)
26. Vidhamaly V, Bellingham K, Newton PN. **The quality of veterinary medicines and their implications for one health**. *BMJ Glob Health* (2022.0) **7**. DOI: 10.1136/bmjgh-2022-008564
27. Newton PN, Lee SJ, Goodman C. **Guidelines for field surveys of the quality of medicines: a proposal**. *PLoS Med* (2009.0) **6**. DOI: 10.1371/journal.pmed.1000052
28. **Calcutta high court (appellete side) - unknown vs the state of west bengal on 2014**. (2014.0)
29. **Kenya recalls ‘faulty’ south korean HIV kits. 2011**
30. McManus D, Naughton BD. **A systematic review of substandard, falsified, unlicensed and unregistered medicine sampling studies: a focus on context, prevalence, and quality**. *BMJ Glob Health* (2020.0) **5**. DOI: 10.1136/bmjgh-2020-002393
31. Mackey TK. **Prevalence of substandard and falsified essential medicines: still an incomplete picture**. *JAMA Netw Open* (2018.0) **1**. DOI: 10.1001/jamanetworkopen.2018.1685
32. Mziray S, Maganda BA, Mwamwitwa K. *Quality of selected anti-retroviral medicines: tanzania mainland market as a case study*
33. Vickers S, Bernier M, Zambrzycki S. **Field detection devices for screening the quality of medicines: a systematic review**. *BMJ Glob Health* (2018.0) **3**. DOI: 10.1136/bmjgh-2018-000725
34. Cohn J, von Schoen-Angerer T, Jambert E. **When falsified medicines enter the supply chain: description of an incident in Kenya and lessons learned for rapid response**. *J Public Health Policy* (2013.0) **34** 22-30. DOI: 10.1057/jphp.2012.53
35. Amon JJ. **Dangerous medicines: unproven AIDS cures and counterfeit antiretroviral drugs**. *Global Health* (2008.0) **4**. DOI: 10.1186/1744-8603-4-5
36. **Fast-track ending the AIDS epidemic by 2030**
37. **Global HIV programme - treatment & care**
38. **Global HIV programme - pre-exposure prophylaxis**
39. **Global state of prep**
40. Wang T, Hoag SW, Eng ML. **Quality of antiretroviral and opportunistic infection medications dispensed from developing countries and Internet pharmacies**. *J Clin Pharm Ther* (2015.0) **40** 68-75. DOI: 10.1111/jcpt.12226
41. Wang X, Nutland W, Brady M. **Quantification of tenofovir disoproxil fumarate and emtricitabine in generic pre-exposure prophylaxis tablets obtained from the Internet**. *Int J STD AIDS* (2019.0) **30** 765-8. DOI: 10.1177/0956462419841144
42. Bliese SL, Berta M, Lieberman M. **Involving students in the distributed pharmaceutical analysis laboratory: a citizen-science project to evaluate global medicine quality**. *J Chem Educ* (2020.0) **97** 3976-83. DOI: 10.1021/acs.jchemed.0c00904
43. **Global AIDS strategy 2021-2026 — end inequalities. end AIDS**
44. **Vietnam’s counterfeit condom crisis**
45. **Fake condoms warning**
46. **Too short’ condoms lead to counterfeit bust in spain**
47. **Counterfeit condom gang in china broken up**
48. **WHO: access to HIV medicines severely impacted by COVID-19 as AIDS response stalls 2020**
|
---
title: Clinical and genetic associations of deep learning-derived cardiac magnetic
resonance-based left ventricular mass
authors:
- Shaan Khurshid
- Julieta Lazarte
- James P. Pirruccello
- Lu-Chen Weng
- Seung Hoan Choi
- Amelia W. Hall
- Xin Wang
- Samuel F. Friedman
- Victor Nauffal
- Kiran J. Biddinger
- Krishna G. Aragam
- Puneet Batra
- Jennifer E. Ho
- Anthony A. Philippakis
- Patrick T. Ellinor
- Steven A. Lubitz
journal: Nature Communications
year: 2023
pmcid: PMC10030590
doi: 10.1038/s41467-023-37173-w
license: CC BY 4.0
---
# Clinical and genetic associations of deep learning-derived cardiac magnetic resonance-based left ventricular mass
## Abstract
Left ventricular mass is a risk marker for cardiovascular events, and may indicate an underlying cardiomyopathy. Cardiac magnetic resonance is the gold-standard for left ventricular mass estimation, but is challenging to obtain at scale. Here, we use deep learning to enable genome-wide association study of cardiac magnetic resonance-derived left ventricular mass indexed to body surface area within 43,230 UK Biobank participants. We identify 12 genome-wide associations (1 known at TTN and 11 novel for left ventricular mass), implicating genes previously associated with cardiac contractility and cardiomyopathy. Cardiac magnetic resonance-derived indexed left ventricular mass is associated with incident dilated and hypertrophic cardiomyopathies, and implantable cardioverter-defibrillator implant. An indexed left ventricular mass polygenic risk score ≥90th percentile is also associated with incident implantable cardioverter-defibrillator implant in separate UK Biobank (hazard ratio 1.22, $95\%$ CI 1.05-1.44) and Mass General Brigham (hazard ratio 1.75, $95\%$ CI 1.12-2.74) samples. Here, we perform a genome-wide association study of cardiac magnetic resonance-derived indexed left ventricular mass to identify 11 novel variants and demonstrate that cardiac magnetic resonance-derived and genetically predicted indexed left ventricular mass are associated with incident cardiomyopathy.
A genome-wide association study of cardiac magnetic resonance-derived left ventricular mass index including 43,000 UK Biobank participants reveals 12 associations (11 novel), implicating genes involved in cardiac contractility and cardiomyopathy.
## Introduction
Left ventricular hypertrophy (LVH) is defined as pathologically increased left ventricular mass (LVM)1 and is associated with increased risk of cardiovascular events including heart failure (HF)1–3, stroke1, atrial fibrillation (AF)4, and sudden cardiac death5. Increased LVM is also a hallmark of certain primary cardiomyopathies such as hypertrophic cardiomyopathy (HCM) and some dilated cardiomyopathies (DCM). Although LVM can be estimated using 12 lead electrocardiograms or echocardiography, cardiac magnetic resonance (CMR) offers more accurate and reproducible quantification, and has therefore emerged as the gold standard for diagnosing LVH6.
Imaging-based estimation of LVM typically requires LV segmentation, which is usually performed manually and requires substantial time and expertise. As a result, genetic analyses of imaging-based LVM have been limited by modest sample sizes. Genome-wide association studies (GWAS) of echocardiography-based LVM identified a single susceptibility locus downstream of SPCS37–9. More recently, a genome-wide association study within 19,000 individuals10 identified significant variants in the gene TTN associated with CMR-based LVM.
Here, we apply a validated deep learning approach to automate estimation of LVM using CMR images (Machine Learning for Health – Segmentation [ML4Hseg]), to maximize power to detect genetic associations underlying CMR-derived LVM11. Specifically, we implement ML4Hseg to estimate LVM using CMRs from nearly 50,000 participants in the UK Biobank. Given body size is a major determinant of LV size and mass12, we analyze LVMI (i.e., LVM indexed by body surface area) in our primary analyses, and assess unindexed LVM in secondary analyses. Our GWAS of LVMI identifies 12 independent variants meeting genome-wide significance, including 11 novel associations. Using expression quantitative trait loci (eQTLs), transcriptome-wide association testing (TWAS), and tissue-specific expression levels, we propose several candidate genes, many of which have been previously associated with cardiac contractility and cardiomyopathy. We additionally develop a polygenic risk score (PRS) for LVMI, and demonstrate that both phenotypic and genetic LVMI are associated with incident cardiovascular diseases including cardiomyopathy.
## Genome-wide association study of CMR-derived LVM
We conducted a multi-ancestry GWAS including 43,230 individuals ($91\%$ European ancestry) (Fig. 1, Supplementary Table 1). The analysis included 9.9 million common variants imputed at an INFO score ≥0.30 and having minor allele frequency (MAF) ≥$1\%$. The genomic control factor was 1.15 with a linkage disequilibrium score regression intercept of 1.00, consistent with polygenicity of the LVMI trait as opposed to inflation (Supplementary Fig. 1). Observed scale h2 for LVMI was 0.26 (standard error [SE] 0.02).Fig. 1Overview of study design and flow. We obtained CMR-derived LVM index in 44,375 individuals undergoing CMR imaging. We performed a genome-wide association study of CMR-derived LVMI and assessed for associations between CMR-derived LVMI and cardiovascular outcomes. Using GWAS results, we developed a polygenic risk score for LVMI, and applied it to 443,326 separate UK Biobank participants with genetic data (left), and 29,354 individuals from the independent Mass General Brigham Biobank (right), to assess for associations between genetically determined LVMI and cardiovascular outcomes.
The GWAS initially revealed 12 candidate SNPs associated with CMR-derived LVMI at genome-wide significance (Table 1 and Fig. 2). Conditional analyses identified an additional variant on chromosome 2, and that the two variants on chromosome 17 located 914 kb apart (r2 = 0.37) were not independent, ultimately resulting in 12 lead SNPs for LVMI. The SNP most strongly associated with LVMI (rs2255167, $$p \leq 1.4$$ × 10−26) was located at the TTN locus on chromosome 2 and has been previously associated with LVM. TTN is highly expressed in LV tissue (Supplementary Table 2)10. The remaining loci ($$n = 11$$) were novel, with many located at or proximate to genes implicated in arrhythmias, cardiomyopathy and cardiomyocyte function, including FLNC, MYOZ1, MAPT, WNT, CLCN6, MYBPC3 and SYNPO2L. Regional association plots for each genome-wide significant SNP are shown in Supplementary Fig. 2. Results for 18 additional variants having suggestive but not genome-wide significant associations are shown in Supplementary Table 3. A secondary GWAS of unindexed LVM revealed 12 genome-wide significant SNPs, of which 6 overlapped with the primary LVMI GWAS, and a 7th was a strong proxy (r2 = 0.87). Loci unique to analyses of unindexed LVM appeared primarily enriched for genes associated with body size (e.g., FTO, HMGA2, GDF5), although FTO has also been implicated in HF13 and CDKN1A has been associated with DCM in a recent multi-trait analysis14 (Supplementary Table 4 and Supplementary Fig. 3).Table 1Variants associated with CMR-derived left ventricular mass index in the mixed-ancestry GWASrsIDChrPosition (hg38)*Closest* gene(s)FunctionRisk/alt alleleRAFBetaSEP value*rs143800963111835418CLCN6IntronicC/A0.950.950.164.2 × 10−9rs2255167†2178693555TTNIntronicT/A0.810.970.093.2 × 10−26rs10497529‡2178975161CCDC141MissenseG/A0.961.280.202.2 × 10−9-5133066736HSPA4IndelCTT/C0.720.500.081.6 × 10−9rs93884986126552277CENPW-G/T0.81−0.550.104.1 × 10−9rs341632291073647154SYNPO2LMissenseG/T0.86−0.600.101.0 × 10−8rs37299891147348490MYBPC3MissenseT/C0.87−0.610.111.8 × 10−8rs2855251612121592356KDM2BIntronicC/T0.85−0.580.101.5 × 10−8rs65985411598727906IGF1RIntronicA/G0.36−0.420.084.6 × 10−8rs562527251614995819PDXDC1IntronicG/A0.750.540.093.7 × 10−9rs65034511745870981MAPTIntronicT/C0.67−0.520.081.1 × 10−10rs199501§1746785247WNT3IntronicA/G0.240.550.091.1 × 109rs62621197198605262ADAMTS10MissenseC/T0.961.110.202.9 × 10−8Chr chromosome, RAF risk allele frequency, OR odds ratio.*Denotes two-sided p value corresponding to BOLT-LMM χ2 statistic.†Locus previously reported for LVM10.‡Variant identified in conditional analysis conditioned on lead SNPs (beta, standard error, and p value are adjusted).§Association no longer observed in analysis conditioned on rs6503451.Fig. 2Manhattan plot of mixed-ancestry GWAS for CMR-derived LVM index. Depicted across increasing chromosome (x-axis) are the association results of the primary mixed-ancestry GWAS of left ventricular mass index. The y-axis plots the negative log10 of the two-sided p value corresponding to BOLT-LMM χ2 statistic. Variants meeting the standard multiplicity correction for genome-wide significance ($p \leq 5$ × 10−8, depicted by hashed horizontal line), are labeled by the closest gene to the lead variant.
In GWAS restricted to individuals of European ancestry, 14 loci met genome-wide significance, of which 12 were either a lead variant or a strong proxy (r2 > 0.8) for a lead variant in the primary GWAS (Supplementary Table 5 and Supplementary Figs. 4 and 5). The two loci unique to the European ancestry analysis were rs143973349, an insertion-deletion variant located near FLNC, a gene highly expressed in LV tissue and previously associated with familial hypertrophic, restrictive, and arrhythmogenic cardiomyopathies, and rs142032045, located in a gene-rich region closest to DOC2A and near several variants previously associated with body size15–18. The variant near FLNC had a suggestive association with LVMI in the primary multi-ancestry GWAS, while the variant near DOC2A did not ($$p \leq 3.2$$ × 10−7 and $$p \leq 1.1$$ × 10−5, respectively). The only variant meeting genome-wide significance in the primary mixed-ancestry GWAS that was not a lead variant in the European-only GWAS did have a suggestive association (rs6598541 near IGF1R $$p \leq 7.7$$ × 10−8).
Results of secondary GWAS analyses, including rank-based inverse normal transformed LVMI, LVMI indexed using the 2.7th power of height, LVMI indexed using lean body mass, LVMI with exclusions for prevalent myocardial infarction and heart failure, and unindexed LVM adjusted for height and weight, are shown in Supplementary Tables 6-10. Results obtained using alternative indexing methods were broadly consistent with the primary analysis in terms of variants identified and effect directions. A summary of association results for the lead variants identified in the primary GWAS tested across varying indexing methods is shown in Supplementary Table 11.
## Bioinformatics and in silico functional analyses to determine candidate genes
In total, of the 12 independent lead SNPs, eight (or their proxies at r2 ≥ 0.8) were significant eQTLs in LV and/or AA tissue samples (Fig. 3). The locus including variant rs143973349 unique to the European ancestry analysis also included eQTLs for LV and AA tissue. For a significant proportion of candidate genes, expression was identified in both LV and AA tissue samples. We then performed TWAS and identified 6 genes across 5 loci where predicted expression was associated with LVMI. Each of the genes implicated by TWAS was also an eQTL for either LV or AA (Fig. 3). Using Hi-C analysis, we observed several potentially relevant chromatin interactions, including between lead variant rs56252725 on chromosome 16 and gene MYH11, which encodes an isoform of the myosin heavy chain which is highly expressed in LV tissue and has been associated with electrocardiogram amplitude, and between lead variant rs143973349 (European-only analysis) and gene CCDC136, which encodes a membrane protein and in which variants have been previously associated with dilated and hypertrophic cardiomyopathies. Detailed results of eQTL, TWAS, and Hi-C analyses are shown in Supplementary Table 2.Fig. 3Candidate gene summary. Depicted is a summary of study results. We used a deep learning algorithm to perform a GWAS of CMR-derived LVMI in 43,230 individuals, finding 12 independent loci associated with LVMI. Using proximity to lead variants, expression quantitative trait locus (eQTL) analysis, transcriptome-wide association studies (TWAS), Hi-C analysis, LV tissue-specific expression levels, and prior evidence, we identified candidate genes across the 12 loci. *Candidate* genes were enriched for genes involved in stress response and neurohormonal regulation, cardiac structure and cardiomyopathy, and cell signaling/function (gray box).
Probable candidate genes at each locus of interest are summarized in Fig. 3. In several cases, the closest gene was additionally supported by either eQTL or TWAS prioritization, including SYNPO2L near rs56252725, IGF1R near rs6598541, PDXDC1 near rs56252725, MAPT near rs6598541, and WNT3 near rs199501. In selected instances, downstream analyses prioritized alternative genes, including NPPA near rs143800963 and ORAI1 near rs28552516, with both genes having substantial expression in LV tissue. *Selected* genes prioritized based on strong biologic plausibility or previous associations with LVM included TTN near rs255167, MYBPC3 near rs3729989, and FLNC near rs143973349 (EUR only subset). TTN, MYBPC3, and FLNC are also substantially expressed in LV tissue (Supplementary Table 2).
## Comparison to prior associations with LV measurements and cardiovascular traits
We assessed whether the significant loci we identified have been previously associated with LV measurements10, 19 and cardiovascular traits. Including the European-only analysis, a total of 4 loci have been previously associated with LV measurements. Variant rs2255167 is located on a region of TTN previously associated with LV mass, LV end diastolic volume, LV end-systolic volume, and LV ejection fraction. Variants rs6503451 near MAPT and rs199501 near WNT3 are located at regions previously associated with LV end-systolic volume. In the European-only analysis, variant rs143973349 near FLNC is at a locus previously associated with LV end-systolic volume and LV ejection fraction. Several additional loci have been implicated in other cardiovascular diseases such as heart failure (e.g., rs34163229 near SYNPO2L), cardiomyopathy (e.g., rs2255167 near TTN, rs3729989 near MYBPC3, rs143973349 near FLNC), and atrial fibrillation (e.g., rs6598541 near IGF1R), while others have been associated with cardiovascular risk factors such as blood pressure or diabetes. Several variants are located at regions previously associated with electrocardiographic traits such as PR interval (e.g., rs56252725 near PDXDC1), QRS duration (rs6598541 near IGF1R), and QRS amplitude (rs6503451 near MAPT). Variants rs28552516 near KDM2B, rs62621197 near ADAMTS10, and rs142032045 near DOC2A in the European-only analysis have not been previously associated with either LV or other cardiovascular traits. A summary of lead variants and their prior associations is shown in Supplementary Table 12.
To assess whether the variants we identified in association with LVMI have been previously associated with other LV measurements, we compared our loci to those reported to have genome-wide associations with other LV measurements in prior analyses by Pirruccello et al.19 and Aung et al.10. We performed an analogous search for associations with any cardiovascular disease or risk factor using the National Human Genome Research Institute GWAS Catalog61. For these analyses, we tabulated all associations including the same variant, a variant serving as a strong proxy (r2 ≥ 0.80), or a variant mapping to the same candidate gene.
## Associations between LVMI and cardiomyopathy
We assessed for associations between CMR-derived LVMI and incident cardiovascular disease. At a median follow-up of 2.7 years (Q1:1.9, Q3:4.1), greater LVMI was consistently associated with greater risk of multiple conditions, including AF, MI, HF, DCM, HCM, and ICD implant (Supplementary Table 13). CMR-derived LVH was strongly associated with incident DCM (HR 10.9, $95\%$ CI 4.67–20.2), HCM (HR 9.26, $95\%$ CI 3.20–26.8), and ICD implant (HR 8.42, $95\%$ CI 3.82–18.6). Cumulative risk of events stratified by presence versus absence of CMR-derived LVH is depicted in Fig. 4.Fig. 4Kaplan–Meier plots of the association between CMR-derived LVH and incident cardiovascular disease. Plots depicting the cumulative risk of atrial fibrillation (top left), heart failure (top right), myocardial infarction (bottom left), and ventricular arrhythmias (bottom right), stratified by the presence (orange) versus absence (teal) of CMR-derived LVH. LVH was defined as LVM index (LVMI) > 72 g/m2 in men and >55 g/m2 in women44. The number at risk within each stratum over time is depicted below each plot.
We next evaluated associations between LVMI genetic risk and incident outcomes. In a set of UK Biobank participants separate from the GWAS sample ($$n = 443$$,326), a greater LVMI PRS was associated with higher risk of multiple incident conditions including AF, HF, ventricular arrhythmias, DCM, and ICD implant (Table 2). In the independent MGB sample ($$n = 29$$,354), the LVMI PRS was again associated with incident ICD implant, along with suggestive associations with HCM and DCM (Table 2). In models of incident ICD risk, the relative hazard of ICD was consistently greatest at the highest levels of CMR-derived LVMI as well as LVMI PRS, with similar effect sizes in both the UK Biobank and MGB (Fig. 5). Disease association results were generally similar in analyses restricted to individuals of European ancestry (Supplementary Table 14), and when utilizing a PRS derived from GWAS performed after exclusion of individuals with prevalent myocardial infarction and heart failure (Supplementary Table 15).Table 2Associations between LVMI PRS and incident diseaseHazard ratio for covariate ($95\%$ CI)*N events/N total†Follow-up, yrs (Q1,Q3)PRS (per 1 SD)PRS (90th percentile)PRS (95th percentile)UK BiobankAtrial fibrillation$\frac{25050}{43591711.8}$ (11.0,12.6)1.01 (1.00–1.03)1.03 (0.98–1.07)1.04 (0.98–1.10)Myocardial infarction$\frac{13405}{43204411.8}$ (11.0,12.6)1.03 (1.01–1.05)1.05 (0.99–1.11)1.10 (1.02–1.18)Heart failure$\frac{13540}{44059011.9}$ (11.0,12.6)1.04 (1.02–1.05)1.06 (1.00–1.12)1.08 (1.00–1.16)Ventricular arrhythmias$\frac{4882}{44229511.9}$ (11.1,12.6)1.06 (1.03–1.09)1.13 (1.04–1.24)1.17 (1.04–1.32)Dilated cardiomyopathy‡$\frac{1023}{44301311.9}$ (11.1,12.6)1.10 (1.04–1.17)1.15 (0.95–1.40)1.29 (1.00–1.66)Hypertrophic cardiomyopathy‡$\frac{420}{44315011.9}$ (11.1,12.6)1.08 (0.98–1.09)0.95 (0.68–1.33)1.23 (0.82–1.86)Implantable defibrillator$\frac{1444}{44321611.9}$ (11.1,12.6)1.07 (1.02–1.13)1.22 (1.05–1.44)1.22 (0.98–1.51)Mass General BrighamAtrial fibrillation$\frac{1332}{253162.9}$ (2.0,4.1)1.01 (0.95–1.06)1.02 (0.85–1.22)1.03 (0.80–1.31)Myocardial infarction$\frac{695}{255922.9}$ (2.0,4.1)0.99 (0.92–1.06)0.97 (0.74–1.25)0.71 (0.47–1.07)Heart failure$\frac{1074}{250632.9}$ (2.0,4.1)0.97 (0.91–1.03)1.18 (0.97–1.42)1.00 (0.76–1.33)Ventricular arrhythmias$\frac{944}{269903.0}$ (2.0,4.2)0.99 (0.93–1.05)1.00 (0.81–1.24)1.03 (0.76–1.38)Dilated cardiomyopathy$\frac{492}{288213.0}$ (2.1,4.2)1.06 (0.97–1.16)1.27 (0.97–1.67)1.06 (0.70–1.59)Hypertrophic cardiomyopathy$\frac{183}{287313.0}$ (2.1,4.2)1.14 (0.98–1.32)1.04 (0.64–1.69)0.82 (0.38–1.75)Implantable defibrillator$\frac{152}{284543.0}$ (2.1,4.2)1.05 (0.89–1.24)1.75 (1.12–2.74)1.69 (0.91–3.12)CI confidence interval, PRS polygenic risk score, Q1 quartile 1, Q3 quartile 3, SD standard deviation.*Hazard ratios obtained using Cox proportional hazards models adjusted for age, sex, and principal components 1–5.†N includes all individuals without the prevalent condition at baseline.‡Includes $$n = 20$$ events with high confidence loss-of-function, deleterious missense, known pathogenic or likely pathogenic variant for HCM, and $$n = 50$$ events with high confidence loss-of-function, deleterious missense, known pathogenic or likely pathogenic rare variant for DCM (see text and Supplementary Table 18).Fig. 5Association between CMR-derived and genetically predicted LVMI and incident ICD implant. Depicted are plots showing the relative hazard of incident implantable cardioverter-defibrillator (ICD) implant as a function of increasing standardized CMR-derived LVM index (left), increasing standardized LVMI PRS in UK Biobank (middle) and increasing standardized LVMI PRS in Mass General Brigham (MGB, right). In each plot, the y-axis depicts the relative hazard of incident ICD compared to the hazard observed for individuals with an average LVMI (left) or average PRS value (middle and right), derived from Cox proportional hazards models adjusted for age and sex (left), and adjusted for age, sex, and the first five principal components of genetic ancestry (middle and right). The relative hazard is plotted on the logarithmic scale. The functional form of the association was selected empirically using a penalized spline approach, in which the degrees of freedom for the penalized spline fit were chosen based on minimization of the corrected Akaike Information Criterion75. The number of events and individuals included in each analysis are listed above each plot.
## Mendelian-randomization analyses of blood pressure and diabetes
To assess for potential causal associations between blood pressure and CMR-derived LVMI, we performed MR analyses using genetic instruments for SBP and DBP among individuals of European ancestry. We performed analogous analyses for diabetes. In an inverse-variance weighted two-sample MR, a 1-SD increase in genetically mediated SBP was associated with a 0.27 g/m2 increase in CMR-derived LVMI ($95\%$ CI 0.23–0.31, $$p \leq 1.75$$ × 10−41), and a 1-SD increase in genetically mediated DBP was associated with a 0.32 g/m2 increase in CMR-derived LVMI ($95\%$ CI, 0.25–0.39, $$p \leq 1.64$$ × 10−20). A 1-SD increase in genetically mediated risk of diabetes was associated with a 0.31 g/m2 increase in CMR-derived LVMI ($95\%$ CI, 0.05–0.56, $$p \leq 0.018$$). Weighted median and MR-Egger analyses demonstrated similar results for SBP and DBP, but associations with diabetes were no longer significant (weighted median: 0.19 g/m2, $95\%$ CI −0.15 to 0.53, $$p \leq 0.26$$; MR-Egger: 0.15 g/m2, $95\%$ CI −0.36 to 0.66, $$p \leq 0.56$$). MR-Egger analyses suggested no substantive directional pleiotropy in the SBP, DBP, and diabetes instruments (intercept 0.01, p-0.38 for SBP; intercept −0.02, $$p \leq 0.04$$ for DBP; intercept=0.01, $$p \leq 0.50$$ for diabetes). MR results were similar using unindexed LVM (Supplementary Table 16). MR plots are shown in Supplementary Fig. 6.
As a form of validation of our LVM estimation, we sought to identify evidence of known causal associations between elevated blood pressure and increased LVM40. We therefore conducted two-sample Mendelian-randomization (MR) within individuals of genetic European ancestry in the UK Biobank sample. Given strong epidemiologic associations between diabetes and LVM62, we performed analogous MR analyses for diabetes. Genetic instruments for systolic blood pressure (SBP) and diastolic blood pressure (DBP) were derived from a recent GWAS63. The same set of SNPs was used for both systolic and diastolic blood pressure, but weights specific to systolic versus diastolic blood pressure were used for the systolic and diastolic Mendelian-randomization analysis, respectively63. Utilizing an 865 SNP instrument for SBP and DBP, we prioritized inverse-variance weighted (IVW) meta-analyses of the effect of each SNP on CMR-derived LVMI (and LVM) divided by the effect of the same SNP on SBP and DBP, respectively. We performed an analogous procedure using a 337 SNP instrument for diabetes64. Linear regression models were adjusted for age, sex, genotyping array, and the first ten principal components of genetic ancestry, to determine the beta coefficients and standard errors for the association of each SNP with the outcome (CMR-derived LVMI). These SNP-specific estimates were combined to conduct two-sample Mendelian randomization using the ‘MendelianRandomization’ package in R. Weighted median and MR-Egger analyses were performed secondarily to address potential invalid instruments and directional pleiotropy.
## Discussion
In the current study, we utilized a deep learning segmentation algorithm to perform GWAS of CMR-derived LVMI in nearly 50,000 individuals. Leveraging favorable statistical power and a rich imaging-based phenotype, we identified 12 independent loci associated with LVMI at genome-wide significance. Of the loci identified, 11 are novel for LV mass, 9 have not been previously associated with any LV measurement, and 2 have not been associated with any cardiovascular trait or risk factor. A European-only analysis revealed 2 additional loci which are novel for LV mass. Downstream analyses prioritize several candidate genes, including multiple genes previously associated with cardiac structure and function, as well as cardiomyopathy. Importantly, CMR-derived and genetically determined LVMI were each associated with greater risk of incident cardiovascular events, including incident, DCM, and ICD implant.
Our analyses suggest that common variants in cardiac structural and functional genes appear to be important determinants of LVM. CMR-derived LVMI was strongly associated with variation at rs2255167, located within the gene encoding the large sarcomeric protein titin and previously associated with LV mass10, as well as LV volumes and ejection fraction19. MYOZ1, which encodes a sarcomeric protein involved in calcineurin signaling and was prioritized by both eQTL and TWAS analysis, has been previously associated with HF13 and AF20. A mouse knockout of MYOZ1 resulted in increased exercise capacity through activation of the nuclear factor of activated T-cells21. *Another* gene prioritized by both eQTL and TWAS, TNNT3, encodes a troponin T isoform which is highly expressed in LV tissue. The TNNT3 R63H variant has been shown to result in increased contractility in mouse skeletal muscle and is a cause of the human disease Arthrogryposis (Type 2B2)22, characterized by limb contractures (i.e., excessive muscular contraction). SYNPO2L, an actin-related protein expressed in LV myocardium, has been previously associated with AF23, HF24, HCM14, and voltage-duration product (a clinical indicator of LVH)25.
Several of the candidate genes we identified prioritize neurohormonal regulation and response to physiologic stress as potential genetic determinants of LVMI. Specifically, lead variant rs143800963 is located on chromosome 1 within 20 kb of NPPA and NPPB, genes that encode the natriuretic peptides Nppa and Nppb, respectively, with both proteins playing important roles in blood pressure regulation and salt homeostasis26. Both Nppa and Nppb are constitutively expressed in ventricular myocardium and upregulated in response to stress27. NPPB knockout in mice results in augmentation of the cardiac fibrosis response to pressure overload28. Conversely, cardiomyocyte-specific deletion of ORAI1, which encodes a regulator of calcium-induced calcium release, results in improved response to pressure overload and protection against angiotensin II-induced cardiac remodeling in adult myocardium29. IGFR1, an eQTL for LV tissue in which predicted expression in LV was associated with LVMI, encodes the insulin-like growth factor receptor 1, which has been implicated in organ growth and insulin resistance30.
Several LVMI candidate genes have previous links to cardiomyopathy and HF. The strongest association we observed was at rs2255167, a variant located in TTN, in which mutations have been previously associated with familial cardiomyopathy31 and early-onset AF32. One of the loci detected in the European ancestry analysis (and suggestive in the primary analysis), FLNC, encodes filamin C, an actin-related protein associated with familial HCM16, restrictive cardiomyopathy17, arrhythmogenic cardiomyopathy15, and LV contractile function19. A mouse knock-in of filamin C results in myofibrillar degeneration33. PPP3CB, which encodes the signaling protein calcineurin, has been implicated in pathologic cardiac hypertrophy34. Lead variant rs3729989 is located near MYBPC3, a gene encoding the cardiac myosin-binding protein. Mutations in MYBPC3 are a known cause of DCM and HCM35, 36. FTO, an obesity gene previously associated with HF13, was associated with unindexed LV mass, but not LVMI. Interestingly, we identified several loci which are novel for LVM but have prior associations with electrocardiographic traits37, 38. Future work is warranted to assess whether such associations may reflect electrical manifestations of LV mass or the presence of a cardiomyopathy.
Importantly, we observed that both phenotypic and genetically predicted LVMI were associated with increased risks of incident cardiovascular events. Increased LVMI and LVH are consistently associated with HF2. Here, we observed associations not only with HF, but also incident DCM, HCM, and insertion of an ICD (a surrogate for cardiomyopathy or ventricular arrhythmias). Consistent with the notion that LVMI may be an endophenotype for certain cardiomyopathies, we observed that genetically predicted LVMI (using a 465-variant PRS) was associated with greater risk of incident ICD implant in a separate set of UK Biobank participants as well as an external sample from the MGB healthcare system. Of note, we did not exclude individuals with DCM or HCM from our incident disease analyses since we hypothesized that polygenic risk may nevertheless contribute to the development of clinical outcomes39. In the context of low event rates, however, the LVMI PRS was associated with incident DCM only in the UK Biobank, and associations with incident HCM were not significant in either sample. Consistent with expectations40, 41, using Mendelian-randomization analyses, we observed associations between genetically predicted blood pressure and diabetes risk with greater LVM. Overall, our findings provide evidence that the genetic variation underlying increased LVM may be clinically relevant, and highlight the need for future research to evaluate the potential utility of a polygenic predictor of LVM to improve identification of individuals at risk of incident cardiomyopathy.
Our study has limitations. First, our analysis was a mixed-ancestry GWAS, but the sample is predominantly of European descent. Therefore, our results may not generalize to individuals of other ancestries. Second, we used a previously published deep learning model (ML4Hseg) to facilitate well-powered GWAS of CMR-derived LVM. ML4Hseg was trained using an imperfect segmentation method as ground truth11, 42, which may have led to lower agreement with true LVM as compared to some alternative approaches (e.g., $95\%$ limits of agreement −27g to 27 g with ML4Hseg versus −18 to 18 g by Bai et al. using a proprietary deep learning model43 and −5 to 8 g by Peterson et al. in a small set of hand-labeled measurements44). Nevertheless, estimates from ML4Hseg correlate strongly ($r = 0.86$) with hand-labeled CMR-derived LVM in the UK Biobank11, and MR analyses recapitulated a known causal relationship between elevated blood pressure and increased indexed LVM40. Third, our ability to assess for associations between CMR-derived LVMI and incident outcomes was limited by event rates and follow-up currently available after imaging. Fourth, generalizability may be affected by bias introduced by methods of enrollment, as UK Biobank participants are enriched for health and socioeconomic status compared to the general population45. Fifth, we analyzed LVM indexed to body surface area since this measure is in common clinical use, even though alternative methods of body mass correction exist. We therefore performed multiple analyses using alternative indexing methods (e.g., 2.7th power of height).
In summary, we performed GWAS of deep-learned CMR-derived LVM including nearly 50,000 individuals. We discovered 12 independent loci meeting genome-wide significance, including 11 that are novel. Using complementary downstream analyses, we identified multiple candidate genes, many of which are involved in cardiac structure and function, and several that have been previously implicated in cardiomyopathy. Both CMR-derived and genetically determined LVM were associated with incident ICD implant in independent datasets. Our findings add to our understanding of common genetic variation underlying LVM and demonstrate the potential to use deep learning to define rich phenotypes at scale to empower clinically relevant biological discovery.
## Study populations
The discovery sample comprised the UK Biobank, a population-based prospective cohort of 502,629 participants recruited between 2006–2010 in the United Kingdom to investigate the genetic and lifestyle determinants of disease. The design of the cohort has been described previously46, 47. Briefly, approximately 9.2 million individuals aged 40-69 years living within 25 miles of the 22 assessment centers in England, Wales, and Scotland were invited, and $5.4\%$ participated in the baseline assessment. Extensive questionnaire data, physical measures, and biological data were collected at recruitment, with ongoing data collection in large subsets of the cohort, including repeated assessments and multimodal imaging. At the time of the current analysis, over 450,000 individuals have genome-wide genotyping data available. All participants are followed up for health outcomes through linkage to national health-related datasets.
We utilized the MGB Biobank to replicate a LVMI PRS that we derived in the UK Biobank. The MGB *Biobank is* a biorepository comprising patients from a multi-institutional healthcare network spanning seven hospitals in the New England region of the United States. MGB Biobank participants are followed for health outcomes through linkage to electronic health record (EHR) data.
UK Biobank and MGB Biobank participants provided written informed consent. The UK Biobank was approved by the UK Biobank Research Ethics Committee (reference number 11/NW/0382) and the MGB Biobank by the MGB Institutional Review Board. Use of UK Biobank (application #17488) and MGB *Biobank data* were approved by the local MGB Institutional Review Board.
## Cardiac magnetic resonance acquisition
For all analyses, we included individuals who underwent CMR during a UK Biobank imaging assessment and whose bulk CMR data were available for download as of 04-01-2020 (Fig. 1). The full CMR protocol of the UK Biobank has been described in detail previously48. Briefly, all CMR examinations were performed in the United Kingdom on a clinical wide-bore 1.5 Tesla scanner (MAGNETOM Aera, Syngo Platform VD13A, Siemens Healthineers, Erlangen, Germany). All acquisitions used balanced steady-state free precession with typical parameters.
## Left ventricular mass estimation
We obtained CMR-derived LVM from all individuals with available CMR imaging using ML4Hseg11. ML4Hseg is a convolutional neural network which identifies pixels corresponding to LV myocardium, which are then summed to estimate LV area and multiplied by slice thickness to estimate LV myocardial volume. LV myocardial volume is then multiplied by myocardial density (1.05 g/cm3) to yield LVM. LVM estimates were calibrated to the sex-specific sample means using manually labeled LVM measurements which were available within a subset of the UK Biobank sample ($$n = 4910$$), where sex was classified using self-reported data. LVM estimates obtained using the described method have been shown to have very good correlation (Pearson r 0.86) and agreement (mean absolute error 10 g) against manually labeled LVM in the UK Biobank11. LVM estimates were indexed for body surface area using the DuBois formula to yield LVMI49. A total of 59 ($0.1\%$) individuals with outlying estimated LVM values (defined as falling outside 5 interquartile ranges from the median, or any value ≤0 g/m2 following calibration) were removed prior to analyses (Fig. 1). The distribution of CMR-derived LVM is shown in Supplementary Fig. 7.
## Genome-wide association study
To identify common genetic variation associated with CMR-derived LVM, we performed a GWAS of indexed LVM using BOLT-LMM v2.3.450, which accounts for ancestral heterogeneity, cryptic population structure, and sample relatedness by fitting a linear mixed model with a Bayesian mixture prior as a random effect19, 51, 52. Previous evidence supports the use of LMM approaches to perform GWAS of admixed populations, which may provide favorable statistical power51, 53, 54, and similar approaches have been taken previously19, 51, 52. The GWAS was performed among 43,230 individuals having undergone CMR imaging, after exclusion of individuals without genetic data meeting standard quality control metrics (e.g., no evidence of sex chromosome aneuploidy, outliers in heterozygosity and missing rates). Imputed variants were retained if the imputation information metric was ≥0.3. All variants with minor allele frequency <$1\%$ were excluded from the final analyses. Our model was adjusted for age at CMR acquisition, sex, array platform, and first five principal components of genetic ancestry, where sex was classified on the basis of genetic sex. Associations were considered statistically significant at the standard genome-wide significance level ($$p \leq 5$$ × 10−8). Lead single nucleotide polymorphisms (SNPs) were grouped into independent loci based on distance (±500 kb), with conditional analyses performed to assess for independent signals within windows. Variants having suggestive (i.e., $p \leq 1$ × 10−6) but not genome-wide significant associations were similarly tabulated. Genetic inflation was assessed by calculating the genomic control factor λ, inspecting quantile-quantile plots, and calculating the linkage disequilibrium score (LDSC) regression intercept using LDSC v1.0.155. Observed scale heritability (h2) was estimated using the slope of LDSC regression. We assessed for independent signals within genome-wide significant loci by a) performing GWAS while conditioning on the imputed allele dosage of each lead SNP found in the primary GWAS (excluding insertion-deletion variants), and b) performing GWAS while conditioning on the top variant on chromosome 17 alone (rs6503451), to assess whether the additional variant located 914 kb apart on chromosome 17 (rs199502, r2 = 0.37), was independent. The primary GWAS was performed among individuals of all genetic ancestries.
We performed several secondary GWAS analyses. First, we performed analogous GWAS restricted to individuals of *European* genetic ancestry ($$n = 39$$,187). Second, we performed GWAS of unindexed LV mass (with and without adjustment for height and weight), as well as LV mass alternatively indexed using the 2.7th power of height56. Third, we performed a GWAS of LVMI after rank-based inverse normal transformation. Fourth, we performed GWAS of LVMI excluding individuals with prevalent myocardial infarction and heart failure.
## Bioinformatics and in silico functional analyses
We assessed whether genes within 500 kb of lead SNPs were related to cardiac gene expression using GTEx57 version 8 cis-eQTL tissue data (dbGaP Study Accession phs000424.v8.p2). To maximize power to detect potential candidate genes, we considered eQTLs for both atrial appendage (AA) and LV tissue data19, 58. We included lead variants as well as strong proxy variants (r2 ≥ 0.8). We also quantified tissue-specific expression levels from bulk RNA sequencing data from GTEx57 version 8 (dbGaP Study Accession phs000424.v8.p2). We evaluated the effects of predicted gene expression levels on LVMI by performing a transcriptome-wide association study (TWAS) using S-PrediXcan59. GTEx genotypes and normalized expression data in AA and LV tissues provided in the software were used as training sets to develop the prediction models. Prediction models between each gene-tissue pair were developed using elastic net regression. In total, we tested 6636 and 6008 associations in AA and LV, respectively. The significance threshold for S-PrediXcan was therefore set at $$p \leq 0.05$$/(6636 + 6008), or 3.95 × 10−6. We assessed for potential long-range chromatin interactions using Hi-C analysis in adult heart tissues obtained from the Myocardial Applied Genomics Network (MAGNet, www.med.upenn.edu/magnet) at the University of Pennsylvania60.
We prioritized candidate genes on the basis of closest proximity to the lead variant, eQTLs, TWAS, tissue-specific expression levels, Hi-C analysis, and biologic plausibility based on previously reported data. All prioritized genes were supported by at least two lines of evidence.
## Polygenic risk score development
To develop a PRS as a genetic instrument for CMR-derived LVMI, we applied a pruning and thresholding approach to our LVMI GWAS results. After removing insertion-deletion variants and strand ambiguous (i.e., A/T and C/G) variants to facilitate replication, we developed and tested four separate candidate PRS utilizing each combination of two thresholds used to define index SNPs ($$p \leq 1$$ × 10−6 and $$p \leq 1$$ × 10−4) and two thresholds used to prune proxy SNPs (r2 = 0.3 and r2 = 0.5). We then selected the PRS explaining the greatest variance in LVMI within the derivation set, which ultimately comprised a set of 465 variants (r2 = 0.3, $$p \leq 1$$ × 10−4, variance of LVMI explained = 0.084; +3.56 g/m2 increase in LVMI per 1-standard deviation [1-SD] increase in PRS, $p \leq 0.01$).
## Outcomes association testing
We assessed for associations between CMR-derived LVMI and incident AF, myocardial infarction, HF, ventricular arrhythmias, DCM, HCM, and implantable cardioverter-defibrillator (ICD) within participants with follow-up clinical data available after the imaging visit. We assessed for analogous associations using LVH, which was defined as LVMI > 72 g/m2 in men and >55 g/m2 in women44, and alternatively as the sex-specific 90th percentile of LVM1. Diseases were defined using combinations of self-report and inpatient International Classification of Diseases, 9th and 10th revision codes (Supplementary Data 1). Start of follow-up was defined at the time of CMR acquisition and spanned until the earliest of an incident event, death, or last follow-up. The date of last follow-up was dependent upon the availability of linked hospital data, and was therefore defined as March 31, 2021 for participants enrolled in England ($93.6\%$) and Scotland ($6.1\%$), and February 28, 2018 for participants enrolled in Wales ($0.3\%$).
We performed analogous association testing between the LVMI PRS and the same set of incident cardiovascular events among individuals in the UK Biobank that did not undergo CMR ($$n = 443$$,326). Outcome and person-time definitions were similar, although start of follow-up was defined as the date of UK Biobank enrollment and blood sample collection. We also repeated association testing between the LVMI PRS and incident events in the independent MGB Biobank sample, using analogous models with person-time beginning at the date of blood sample collection and ending at an event, death, or last encounter in the electronic health record.
## Statistical analysis
We tested associations between CMR-derived LVM and incident AF, myocardial infarction, HF, ventricular arrhythmias, DCM, HCM, and ICD using Cox proportional hazards regression with adjustment for sex and age at CMR acquisition. We fit analogous models using LVH (defined using the thresholds described above) and the LVMI PRS as the primary exposures. Models including the PRS were additionally adjusted for the first five principal components of genetic ancestry. For the PRS outcomes analyses, we did not exclude individuals with pathogenic or likely pathogenic variants for HCM or DCM for the following reasons: (a) a substantial proportion of individuals with clinically confirmed HCM and DCM have no causal variant identified14, 65, (b) recent evidence suggests that polygenic background may play an important role in disease development even among individuals carrying mutations39, and (c) rare variant information is not available in all individuals in our UKBB or MGB replication samples. To assess the frequency of pathologic rare variants among individuals with incident HCM and DCM events, we did tabulate carrier status of high confidence loss of function, deleterious missense, and known pathogenic or likely pathogenic variants in HCM and DCM genes as cataloged in ClinVar as of $\frac{2}{9}$/2021. We also included high confidence loss-of-function variants using LOFTEE66, a plug-in of VEP67, and deleterious missense variants68 using 30 in silico prediction tools presented in v4.1a of the dbnsfp database69. A full list of variants is shown in Supplementary Table 17.
Validity of the proportionality assumption was assessed using the Grambsch-Therneau test of correlation70 as well as visual inspection of smoothed fits to Schoenfeld residuals versus time. Where present, substantial deviations from proportional hazards (observed only for age, sex, and certain principal components of ancestry), were modeled by including interaction terms with strata of person-time.
Statistical analyses were performed using R v4.0 (packages ‘data.table’ v1.13.6, ‘ggplot2’ v3.3.3,’survival’ v3.2-7,’prodlim’ v2019.11.13, ‘MendelianRandomization’ v0.5.0)71, 72. Except where otherwise noted, all two-tailed p-values <0.05 were considered statistically significant.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-37173-w.
## Peer review information
Nature Communications thanks Alistair Young and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
## References
1. Bluemke DA. **The relationship of left ventricular mass and geometry to incident cardiovascular events: the MESA (Multi-Ethnic Study of Atherosclerosis) study**. *J. Am. Coll. Cardiol.* (2008.0) **52** 2148-2155. DOI: 10.1016/j.jacc.2008.09.014
2. Kawel-Boehm N. **Left Ventricular Mass at MRI and long-term risk of cardiovascular events: the multi-ethnic study of atherosclerosis (MESA)**. *Radiology* (2019.0) **293** 107-114. DOI: 10.1148/radiol.2019182871
3. Lazzeroni D, Rimoldi O, Camici PG. **From left ventricular hypertrophy to dysfunction and failure**. *Circ. J.* (2016.0) **80** 555-564. DOI: 10.1253/circj.CJ-16-0062
4. Chrispin J. **Association of electrocardiographic and imaging surrogates of left ventricular hypertrophy with incident atrial fibrillation: MESA (Multi-Ethnic Study of Atherosclerosis)**. *J. Am. Coll. Cardiol.* (2014.0) **63** 2007-2013. DOI: 10.1016/j.jacc.2014.01.066
5. Haider AW, Larson MG, Benjamin EJ, Levy D. **Increased left ventricular mass and hypertrophy are associated with increased risk for sudden death**. *J. Am. Coll. Cardiol.* (1998.0) **32** 1454-1459. DOI: 10.1016/S0735-1097(98)00407-0
6. Lenstrup M, Kjaergaard J, Petersen CL, Kjaer A, Hassager C. **Evaluation of left ventricular mass measured by 3D echocardiography using magnetic resonance imaging as gold standard**. *Scand. J. Clin. Lab. Investig.* (2006.0) **66** 647-657. DOI: 10.1080/00365510600892233
7. Wild PS. **Large-scale genome-wide analysis identifies genetic variants associated with cardiac structure and function**. *J. Clin. Investig.* (2017.0) **127** 1798-1812. DOI: 10.1172/JCI84840
8. Kanai M. **Genetic analysis of quantitative traits in the Japanese population links cell types to complex human diseases**. *Nat. Genet.* (2018.0) **50** 390-400. DOI: 10.1038/s41588-018-0047-6
9. Mosley JD. **The polygenic architecture of left ventricular mass mirrors the clinical epidemiology**. *Sci. Rep.* (2020.0) **10** 7561. DOI: 10.1038/s41598-020-64525-z
10. Aung N. **Genome-wide analysis of left ventricular image-derived phenotypes identifies fourteen loci associated with cardiac morphogenesis and heart failure development**. *Circulation* (2019.0) **140** 1318-1330. DOI: 10.1161/CIRCULATIONAHA.119.041161
11. 11.Khurshid, S. et al. Deep learning to estimate cardiac magnetic resonance–derived left ventricular mass. Cardiovasc. Digit. Health J. S2666693621000232. 10.1016/j.cvdhj.2021.03.001 (2021).
12. Engel DJ, Schwartz A, Homma S. **Athletic cardiac remodeling in US professional basketball players**. *JAMA Cardiol.* (2016.0) **1** 80. DOI: 10.1001/jamacardio.2015.0252
13. Shah S. **Genome-wide association and Mendelian randomisation analysis provide insights into the pathogenesis of heart failure**. *Nat. Commun.* (2020.0) **11** 163. DOI: 10.1038/s41467-019-13690-5
14. Tadros R. **Shared genetic pathways contribute to risk of hypertrophic and dilated cardiomyopathies with opposite directions of effect**. *Nat. Genet.* (2021.0) **53** 128-134. DOI: 10.1038/s41588-020-00762-2
15. Begay RL. **Filamin C truncation mutations are associated with arrhythmogenic dilated cardiomyopathy and changes in the cell-cell adhesion structures**. *JACC Clin. Electrophysiol.* (2018.0) **4** 504-514. DOI: 10.1016/j.jacep.2017.12.003
16. Valdés-Mas R. **Mutations in filamin C cause a new form of familial hypertrophic cardiomyopathy**. *Nat. Commun.* (2014.0) **5** 5326. DOI: 10.1038/ncomms6326
17. Brodehl A. **Mutations in FLNC are associated with familial restrictive cardiomyopathy**. *Hum. Mutat.* (2016.0) **37** 269-279. DOI: 10.1002/humu.22942
18. Locke AE. **Genetic studies of body mass index yield new insights for obesity biology**. *Nature* (2015.0) **518** 197-206. DOI: 10.1038/nature14177
19. Pirruccello JP. **Analysis of cardiac magnetic resonance imaging in 36,000 individuals yields genetic insights into dilated cardiomyopathy**. *Nat. Commun.* (2020.0) **11** 2254. DOI: 10.1038/s41467-020-15823-7
20. Roselli C. **Multi-ethnic genome-wide association study for atrial fibrillation**. *Nat. Genet* (2018.0) **50** 1225-1233. DOI: 10.1038/s41588-018-0133-9
21. Frey N. **Calsarcin-2 deficiency increases exercise capacity in mice through calcineurin/NFAT activation**. *J. Clin. Investig.* (2008.0) **118** 3598-3608. DOI: 10.1172/JCI36277
22. Daly SB. **Exome sequencing identifies a dominant TNNT3 mutation in a large family with distal arthrogryposis**. *Mol. Syndromol.* (2014.0) **5** 218-228. DOI: 10.1159/000365057
23. Weng L-C. **Heritability of atrial fibrillation**. *Circ. Cardiovasc. Genet.* (2017.0) **10** e001838. DOI: 10.1161/CIRCGENETICS.117.001838
24. Schneider BP. **Genome-wide association study for anthracycline-induced congestive heart failure**. *Clin. Cancer Res* (2017.0) **23** 43-51. DOI: 10.1158/1078-0432.CCR-16-0908
25. van der Harst P. **52 genetic loci influencing myocardial mass**. *J. Am. Coll. Cardiol.* (2016.0) **68** 1435-1448. DOI: 10.1016/j.jacc.2016.07.729
26. Goetze JP. **Cardiac natriuretic peptides**. *Nat. Rev. Cardiol.* (2020.0) **17** 698-717. DOI: 10.1038/s41569-020-0381-0
27. Man J, Barnett P, Christoffels VM. **Structure and function of the Nppa-Nppb cluster locus during heart development and disease**. *Cell Mol. Life Sci.* (2018.0) **75** 1435-1444. DOI: 10.1007/s00018-017-2737-0
28. Tamura N. **Cardiac fibrosis in mice lacking brain natriuretic peptide**. *Proc. Natl Acad. Sci. USA* (2000.0) **97** 4239-4244. DOI: 10.1073/pnas.070371497
29. Segin S. **Cardiomyocyte-specific deletion of Orai1 reveals its protective role in angiotensin-II-induced pathological cardiac remodeling**. *Cells* (2020.0) **9** 1092. DOI: 10.3390/cells9051092
30. Cubbon RM, Kearney MT, Wheatcroft SB. **Endothelial IGF-1 receptor signalling in diabetes and insulin resistance**. *Trends Endocrinol. Metab.* (2016.0) **27** 96-104. DOI: 10.1016/j.tem.2015.11.009
31. Herman DS. **Truncations of titin causing dilated cardiomyopathy**. *N. Engl. J. Med.* (2012.0) **366** 619-628. DOI: 10.1056/NEJMoa1110186
32. Choi SH. **Association between titin loss-of-function variants and early-onset atrial fibrillation**. *JAMA* (2018.0) **320** 2354-2364. DOI: 10.1001/jama.2018.18179
33. Chevessier F. **Myofibrillar instability exacerbated by acute exercise in filaminopathy**. *Hum. Mol. Genet* (2015.0) **24** 7207-7220. DOI: 10.1093/hmg/ddv421
34. Wilkins BJ. **Calcineurin/NFAT coupling participates in pathological, but not physiological, cardiac hypertrophy**. *Circ. Res.* (2004.0) **94** 110-118. DOI: 10.1161/01.RES.0000109415.17511.18
35. Watkins H. **Mutations in the cardiac myosin binding protein-C gene on chromosome 11 cause familial hypertrophic cardiomyopathy**. *Nat. Genet.* (1995.0) **11** 434-437. DOI: 10.1038/ng1295-434
36. Daehmlow S. **Novel mutations in sarcomeric protein genes in dilated cardiomyopathy**. *Biochem Biophys. Res Commun.* (2002.0) **298** 116-120. DOI: 10.1016/S0006-291X(02)02374-4
37. Verweij N. **The genetic makeup of the electrocardiogram**. *Cell Syst.* (2020.0) **11** 229-238.e5. DOI: 10.1016/j.cels.2020.08.005
38. Ntalla I. **Multi-ancestry GWAS of the electrocardiographic PR interval identifies 202 loci underlying cardiac conduction**. *Nat. Commun.* (2020.0) **11** 2542. DOI: 10.1038/s41467-020-15706-x
39. Fahed AC. **Polygenic background modifies penetrance of monogenic variants for tier 1 genomic conditions**. *Nat. Commun.* (2020.0) **11** 3635. DOI: 10.1038/s41467-020-17374-3
40. Hendriks T. **Effect of systolic blood pressure on left ventricular structure and function: a Mendelian randomization study**. *Hypertension* (2019.0) **74** 826-832. DOI: 10.1161/HYPERTENSIONAHA.119.12679
41. Ai S. **Effects of glycemic traits on left ventricular structure and function: a Mendelian randomization study**. *Cardiovasc. Diabetol.* (2022.0) **21** 109. DOI: 10.1186/s12933-022-01540-6
42. Suinesiaputra A. **Fully-automated left ventricular mass and volume MRI analysis in the UK Biobank population cohort: evaluation of initial results**. *Int. J. Cardiovasc. Imaging* (2018.0) **34** 281-291. DOI: 10.1007/s10554-017-1225-9
43. Bai W. **Automated cardiovascular magnetic resonance image analysis with fully convolutional networks**. *J. Cardiovasc. Magn. Reson* (2018.0) **20** 65. DOI: 10.1186/s12968-018-0471-x
44. Petersen SE. **Reference ranges for cardiac structure and function using cardiovascular magnetic resonance (CMR) in Caucasians from the UK Biobank population cohort**. *J. Cardiovasc. Magn. Reson* (2017.0) **19** 18. DOI: 10.1186/s12968-017-0327-9
45. Fry A. **Comparison of sociodemographic and health-related characteristics of UK Biobank participants with those of the general population**. *Am. J. Epidemiol.* (2017.0) **186** 1026-1034. DOI: 10.1093/aje/kwx246
46. Sudlow C. **UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age**. *PLoS Med.* (2015.0) **12** e1001779. DOI: 10.1371/journal.pmed.1001779
47. Littlejohns TJ, Sudlow C, Allen NE, Collins R. **UK Biobank: opportunities for cardiovascular research**. *Eur. Heart J.* (2019.0) **40** 1158-1166. DOI: 10.1093/eurheartj/ehx254
48. Petersen SE. **UK Biobank’s cardiovascular magnetic resonance protocol**. *J. Cardiovasc. Magn. Reson* (2016.0) **18** 8. DOI: 10.1186/s12968-016-0227-4
49. Du Bois D, Du Bois EF. **A formula to estimate the approximate surface area if height and weight be known. 1916**. *Nutrition* (1989.0) **5** 303-311. PMID: 2520314
50. Loh P-R, Kichaev G, Gazal S, Schoech AP, Price AL. **Mixed-model association for biobank-scale datasets**. *Nat. Genet.* (2018.0) **50** 906-908. DOI: 10.1038/s41588-018-0144-6
51. Wojcik GL. **Genetic analyses of diverse populations improves discovery for complex traits**. *Nature* (2019.0) **570** 514-518. DOI: 10.1038/s41586-019-1310-4
52. Page GP. **Multiple-ancestry genome-wide association study identifies 27 loci associated with measures of hemolysis following blood storage**. *J. Clin. Investig.* (2021.0) **131** e146077. DOI: 10.1172/JCI146077
53. Lloyd-Jones LR. **Inference on the genetic basis of eye and skin color in an admixed population via Bayesian linear mixed models**. *Genetics* (2017.0) **206** 1113-1126. DOI: 10.1534/genetics.116.193383
54. Caliebe A. **Including diverse and admixed populations in genetic epidemiology research**. *Genet. Epidemiol.* (2022.0) **46** 347-371. DOI: 10.1002/gepi.22492
55. Bulik-Sullivan BK. **LD Score regression distinguishes confounding from polygenicity in genome-wide association studies**. *Nat. Genet* (2015.0) **47** 291-295. DOI: 10.1038/ng.3211
56. Cuspidi C. **Improving cardiovascular risk stratification in essential hypertensive patients by indexing left ventricular mass to height(2.7)**. *J. Hypertens.* (2009.0) **27** 2465-2471. DOI: 10.1097/HJH.0b013e32833105a6
57. **Genetic effects on gene expression across human tissues**. *Nature* (2017.0) **550** 204-213. DOI: 10.1038/nature24277
58. Ahlberg G. **Genome-wide association study identifies 18 novel loci associated with left atrial volume and function**. *Eur. Heart J.* (2021.0) **42** 4523-4534. DOI: 10.1093/eurheartj/ehab466
59. **Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics**. *Nat. Commun.* (2018.0) **9** 1825. DOI: 10.1038/s41467-018-03621-1
60. 60.Bianchi, V. et al. Detailed regulatory interaction map of the human heart facilitates gene discovery for cardiovascular disease. Preprint at bioRxiv10.1101/705715 (2019).
61. Buniello A. **The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019**. *Nucleic Acids Res* (2019.0) **47** D1005-D1012. DOI: 10.1093/nar/gky1120
62. Palmieri V. **Effect of Type 2 Diabetes Mellitus on Left Ventricular Geometry and Systolic Function in Hypertensive Subjects: Hypertension Genetic Epidemiology Network (HyperGEN) Study**. *Circulation* (2001.0) **103** 102-107. DOI: 10.1161/01.CIR.103.1.102
63. **Genetic analysis of over 1 million people identifies 535 new loci associated with blood pressure traits**. *Nat. Genet* (2018.0) **50** 1412-1425. DOI: 10.1038/s41588-018-0205-x
64. Mahajan A. **Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation**. *Nat. Genet.* (2022.0) **54** 560-572. DOI: 10.1038/s41588-022-01058-3
65. Walsh R. **Quantitative approaches to variant classification increase the yield and precision of genetic testing in Mendelian diseases: the case of hypertrophic cardiomyopathy**. *Genome Med.* (2019.0) **11** 5. DOI: 10.1186/s13073-019-0616-z
66. Karczewski KJ. **The mutational constraint spectrum quantified from variation in 141,456 humans**. *Nature* (2020.0) **581** 434-443. DOI: 10.1038/s41586-020-2308-7
67. McLaren W. **The ensembl variant effect predictor**. *Genome Biol.* (2016.0) **17** 122. DOI: 10.1186/s13059-016-0974-4
68. Jurgens SJ. **Analysis of rare genetic variation underlying cardiometabolic diseases and traits among 200,000 individuals in the UK Biobank**. *Nat. Genet* (2022.0) **54** 240-250. DOI: 10.1038/s41588-021-01011-w
69. Liu X, Wu C, Li C, Boerwinkle E. **dbNSFP v3.0: A One-Stop Database of Functional Predictions and Annotations for Human Nonsynonymous and Splice-Site SNVs**. *Hum. Mutat.* (2016.0) **37** 235-241. DOI: 10.1002/humu.22932
70. Grambsch PM, Thern’eau TM. **Proportional hazards tests and diagnostics based on weighted residuals**. *Biometrika* (1994.0) **81** 515-526. DOI: 10.1093/biomet/81.3.515
71. 71.R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing Vienna, Austria, 2015).
72. 72.Dowle, M. et al. data.table: extension of ‘data.frame’. Version 1.12.6. https://CRAN.R-project.org/package=data.table.
73. Lambert SA. **The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation**. *Nat. Genet.* (2021.0) **53** 420-425. DOI: 10.1038/s41588-021-00783-5
74. 74.Shaankhurshid. shaankhurshid/lvmass_gwas: v1.0.10.5281/ZENODO.7548696 (2023).
75. Hurvich CM, Simonoff JS, Tsai C-L. **Smoothing parameter selection in nonparametric regression using an improved Akaike information criterion**. *J. R. Stat. Soc.: Ser. B (Stat. Methodol.)* (1998.0) **60** 271-293. DOI: 10.1111/1467-9868.00125
|
---
title: Coronary artery restenosis and target lesion revascularisation in women by
pregnancy history
authors:
- Moa Pehrson
- Andreas Edsfeldt
- Giovanna Sarno
- Abigail Fraser
- Janet W. Rich-Edwards
- Isabel Goncalves
- Mats Pihlsgård
- Simon Timpka
journal: Open Heart
year: 2023
pmcid: PMC10030621
doi: 10.1136/openhrt-2022-002130
license: CC BY 4.0
---
# Coronary artery restenosis and target lesion revascularisation in women by pregnancy history
## Abstract
### Background
Women’s pregnancy history is associated with incident risk of coronary artery disease with some evidence also suggesting a relevance for prognosis following treatment.
### Objectives
To study the associations between maternal history of preterm delivery, a history of small for gestational age infant, parity and age at first delivery with clinical restenosis after percutaneous coronary intervention (PCI).
### Methods
In this prospective cohort study, we included 6027 women <65 years undergoing their first PCI 2006–2017, merging clinical register data on PCI procedures in Sweden with comprehensive registry data on deliveries since 1973. We used proportional hazards regression to study the association between aspects of pregnancy history and clinical restenosis in per-segment analyses, and with target lesion revascularisation (TLR) in per-patient analyses. We adjusted models for procedural-related and patient-related predictors of restenosis.
### Results
During 15 981 segment-years of follow-up, 343 ($3.7\%$) events of clinical restenosis occurred. We found no strong evidence of associations between the studied aspects of pregnancy history and clinical restenosis following PCI. For example, the restenosis HR for a history of preterm delivery in the fully adjusted model was 1.09 ($95\%$ CI 0.77 to 1.55) and the TLR HR was 1.18 ($95\%$ CI 0.91 to 1.52).
### Conclusion
Risk of restenosis following treatment with PCI did not differ by the studied aspects of pregnancy history, including preterm delivery, in young and middle-aged women. Larger studies are needed to obtain more precise estimates.
## Introduction
Female sex is associated with worse outcome following percutaneous coronary intervention (PCI).1 A woman’s pregnancy history, including pregnancy complications such as preterm delivery (PTD), small for gestational age infant (SGA), age at first delivery and total parity, is associated with her future risk of coronary artery disease (CAD).2–4 However, there are indications that the pregnancy history also is associated with women’s prognosis following treatment of CAD with PCI. We have previously shown that women with a history of PTD have an increased risk of major adverse cardiovascular events following coronary artery stenting compared with women who delivered at term.5 *Studying restenosis* risk among women with a history of hypertensive disorders of pregnancy, we found that late-onset pre-eclampsia is associated with a reduced risk following PCI.6 *While restenosis* remains a common complication of PCI,7 it is not known the extent to which other aspects of pregnancy history are associated with restenosis risk.
Here, we aimed to study association between aspects of pregnancy history and risk of symptomatic (ie, clinical) restenosis following PCI. In order to do so, we merged clinical register data on women undergoing PCI procedures in Sweden with comprehensive data on PTD history, SGA, age at first delivery and total parity. As each woman can have had more than one coronary artery segment targeted with PCI during the same procedure, we complementary studied the risk of clinical restenosis per segment and the risk of target lesion revascularisation (TLR) per patient.
## Methods
We conducted a prospective cohort study on a national sample originating from two comprehensive Swedish registers: Swedish Coronary Angiography and Angioplasty Registry (SCAAR) and the Medical Birth Register from the National Board of Health and Welfare. The cohort and statistical methodology have previously been described.6 In short, women were included based on the following criteria: PCI procedure recorded in SCAAR in 2006–2017 after first pregnancy and first pregnancy recorded in the Medical Birth Register (figure 1). We excluded procedures before 2006 as not all covariables were routinely registered before that time point. As age is an important factor to consider for the prognosis following PCI and the inclusion of older women was limited by the lack of delivery data prior to the start of registry registration in 1973, we excluded women >65 years at time of index PCI. We also excluded individuals with missing data on any segment variable ($$n = 3$$), individuals with a history of coronary artery bypass (CABG) or a planned CABG procedure at time of index PCI ($$n = 24$$), individuals with missing data on any exposure variables ($$n = 51$$) and/or individuals with coronary artery stenosis classified as ‘other’ ($$n = 21$$).
**Figure 1:** *Flow chart of study sample. Figure shows the inclusion and exclusion criteria for the study sample. CABG, coronary artery bypass; PCI, MBR, Medical Birth Register; PCI, percutaneous coronary intervention.*
Data on migration and death during follow-up originated from Statistics Sweden and the Swedish National Death Register, respectively. The information was linked using the Swedish identification number.8
## Pregnancy history
Data on pregnancy history originated from the Swedish Medical Birth Register. The Medical Birth Register has collected data on most pregnancies in Sweden since 1973.9 PTD was defined as delivery <36+6 weeks of gestation and further subcategorised into late PTD (34+0–36+6 weeks of gestation) and early PTD (22+0–33+6 weeks of gestation). PTD history was further defined according to the woman’s most PTD prior to her first PCI procedure. Parity was categorised as 1, 2–3 and >4, and based on all pregnancies before index PCI procedure.3 Any SGA was defined as >2 SDs below the normal weight by infant sex and length of pregnancy.10 Age at first delivery was categorised as <20 years, 20–34 years and >35 years.2
## Index PCI procedure
Data on the index PCI procedure originated from SCAAR, a national quality of care register aiming to record information on all coronary angiographies and PCI procedures in Sweden.11 12 Each procedure in SCAAR is described with angiographic, demographic and procedure-related variables. We included several variables as procedural predictors of restenosis. Indication for PCI was categorised as ST-elevation myocardial infarction, non-ST-elevation myocardial infarction, unstable CAD, stable CAD and other. We grouped coronary segments identified in SCAAR into specific vessels treated: left main stem, left anterior descending artery, right coronary artery, left circumflex coronary artery and other. Type of device(s) used were grouped into the following categories: bare metal stent, bare metal stent with balloon predilation, drug-eluting stent, drug-eluting stent with balloon predilation, drug coated balloon and not drug coated balloon. We also included length of stent and stent diameter >3 mm following procedure, using interaction terms with stent as the variables are only applicable to these procedures. In addition to these procedure-related variables, we included patient-related predictors for restenosis. Hypertension and/or dyslipidaemia was defined as a woman receiving antihypertensive treatment or lipid-lowering agents at time of PCI, respectively. Diabetes was defined as a patient having a known diabetes diagnosis at time of PCI, regardless of treatment. Smoking status was divided into the categories never smoker, ex-smoker or current smoker at time of PCI. Prior myocardial infarction was defined by SCAAR as myocardial infarction prior to the current hospitalisation, including silent myocardial infarction based on electrocardiography or echocardiography findings. To account for general improvement of care during the study period, we also included year of procedure and categorised it into three categories: 2006–2009, 2010–2013 and 2014–2017.
## Clinical endpoints
Restenosis was defined as in the SCAAR registry: a stenosis assessed by angiographic visual estimation (>$50\%$) or by fractional flow reserve <0.80 in a previously stented segment identified by coronary angiography for any clinical indication performed anywhere in Sweden.13 14 TLR was defined as repeated PCI targeting any segment included in the index procedure or CABG after the index procedure, whichever came first. Information on repeat PCI was obtained from SCAAR and CABG procedures from the Swedish in-patient care registry. To minimise the risk of including planned CABG procedures as events individuals with a planned CABG procedure at time of index PCI were excluded as described above.
## Statistical analysis
We summarised characteristics of the study sample as means or percentages and calculated the percentages of missing data for each variable. Event rates were estimated using the Kaplan-Meier method and comparisons made using the log-rank test.
To study pregnancy history as a prognostic maker of clinical restenosis we used proportional hazards regression in a per-segment analysis. To account for dependence between coronary artery segments in the same patient, we used a Jackknife (on patient level) estimator of variance. We adjusted for risk factors of restenosis in three steps. In model I, we included age at index PCI, in model II, we additionally accounted for procedure related variables, and in model III, we additionally adjusted for patient-related variables. Right-censoring during follow-up occurred at 2 years of follow-up, end of follow-up in 2017, migration out of Sweden or death, whichever came first.
To study pregnancy history as a prognostic marker of TLR, we used proportional hazards regression in a per-patient analysis. We adjusted for prognostic factors as described above for clinical restenosis with some exceptions. We did not include segment specific variables such as type of device used and class of stenosis. Furthermore, each treated vessel category was included as a not mutually exclusive binary variable, and we additionally adjusted for number of treated vessels. We repeated the analysis for each pregnancy history exposure variable. Right-censoring during follow-up occurred at 2 years of follow-up, end of follow-up in 2017, migration out of Sweden or death, whichever came first.
Multiple imputation was used to impute missing data on a patient level. Individuals with any missing data on a segment level were excluded as described above and 20 imputed datasets were created using multiple imputation by chained equations. Results were combined using PROC MIANALYZE. Statistical analyses were conducted using SAS V.9.4. A significance level of $p \leq 0.05$ was used for hypothesis testing.
## Results
Our study sample consisted of 6027 women with 9397 segments (figure 1). Median time from first delivery to index procedure was 31.0 years (IQR 25.0–35.4 years). Table 1 shows patient characteristics of the study sample by pregnancy history, and online supplemental table 1 shows segment characteristics of the study sample by pregnancy history. In short, women with a history of PTD more often presented with diabetes, hypertension and hyperlipidaemia at the time of PCI compared with women without a history of PTD. Women with a history of SGA were more often active smokers and presented with ST-elevation myocardial infarction (STEMI) at the time of PCI compared with women with no history of SGA. Women unipara at the time of PCI more often presented with diabetes compared to multiparous women. Women who experienced their first delivery at a younger age more often presented with STEMI at the time of PCI compared with women who experienced their first delivery at age >35 years. Segment characteristics were similar for all exposures.
**Table 1**
| Unnamed: 0 | Preterm delivery | Preterm delivery.1 | Small for gestational age | Small for gestational age.1 | Parity at time of PCI | Parity at time of PCI.1 | Parity at time of PCI.2 | Age at first delivery (years) | Age at first delivery (years).1 | Age at first delivery (years).2 | Unnamed: 11 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variables n, (%) unless stated | No PTD (n=5022) | Ever PTD (n=1005) | No SGA(n=5324) | Ever SGA(n=703) | Parity 1(n=1258) | Parity 2–3(n=4165) | Parity >4(n=604) | Age <20(n=788) | Age 20–34(n=4954) | Age >35(n=285) | |
| | | | | | | | | | | | Missing (%) |
| Age (SD) | 55.3 (6.4) | 54.0 (6.8) | 55.1 (6.5) | 54.9 (6.5) | 55.8 (6.7) | 55.0 (6.5) | 54.3 (5.9) | 53.4 (5.7) | 55.3 (6.5) | 55.4 (6.7) | – |
| Previous MI | 177 (3.5) | 55 (5.5) | 191 (3.6) | 41 (5.8) | 63 (5.0) | 154 (3.7) | 15 (2.5) | 37 (4.7) | 182 (3.7) | 13 (4.6) | 1.3 |
| Diabetes | 707 (14.1) | 237 (23.6) | 852 (16.0) | 92 (13.1) | 269 (21.4) | 577 (13.9) | 98 (16.2) | 138 (17.5) | 761 (15.4) | 45 (15.8) | 0.5 |
| Hypertension | 2179 (43.4) | 479 (47.7) | 2350 (44.1) | 308 (43.8) | 584 (46.4) | 1823 (43.8) | 251 (41.6) | 356 (45.2) | 2180 (44.0) | 122 (42.8) | 1.5 |
| Dyslipidaemia | 1457 (29.0) | 356 (35.4) | 1616 (30.4) | 197 (28.0) | 425 (33.8) | 1202 (28.9) | 186 (30.8) | 250 (31.7) | 1474 (29.8) | 89 (31.2) | 1.8 |
| Smoking | | | | | | | | | | | 3.5 |
| Never | 1422 (28.3) | 276 (27.5) | 1544 (29.0) | 154 (21.9) | 337 (26.8) | 1228 (29.5) | 133 (22.0) | 98 (12.4) | 1489 (30.1) | 111 (39.0) | |
| Ex-smoker | 1206 (24.0) | 223 (22.2) | 1283 (24.1) | 146 (20.8) | 309 (24.6) | 999 (24.0) | 121 (20.0) | 159 (20.2) | 1205 (24.3) | 65 (22.8) | |
| Smoker | 2218 (44.2) | 469 (46.7) | 2308 (43.4) | 379 (53.9) | 554 (44.0) | 1801 (43.2) | 332 (55.0) | 507 (64.3) | 2083 (42.1) | 97 (34.0) | |
| Year of index PCI | | | | | | | | | | | – |
| 2006–2009 | 1315 (26.2) | 239 (23.8) | 1355 (25.5) | 199 (28.3) | 356 (28.3) | 1066 (25.6) | 132 (21.9) | 184 (23.4) | 1278 (25.8) | 92 (32.3) | |
| 2010–2013 | 1685 (33.6) | 351 (34.9) | 1792 (33.7) | 244 (34.7) | 411 (32.7) | 1413 (33.9) | 212 (35.1) | 257 (32.6) | 1700 (34.3) | 79 (27.7) | |
| 2014–2017 | 2022 (40.3) | 415 (41.3) | 2177 (40.9) | 260 (37.0) | 491 (39.0) | 1686 (40.5) | 260 (43.1) | 347 (44.0) | 1976 (39.9) | 114 (40.0) | |
| Indication for PCI | | | | | | | | | | | – |
| STEMI | 1843 (36.7) | 359 (35.7) | 1902 (35.7) | 300 (42.7) | 446 (35.5) | 1540 (37.0) | 216 (35.8) | 297 (37.7) | 1810 (36.5) | 95 (33.3) | |
| NSTEMI | 696 (13.9) | 131 (13.0) | 732 (13.8) | 95 (13.5) | 161 (12.8) | 565 (13.6) | 101 (16.7) | 136 (17.3) | 653 (13.2) | 38 (13.3) | |
| Unstable CAD | 1614 (32.1) | 326 (32.4) | 1735 (32.6) | 205 (29.2) | 421 (33.5) | 1316 (31.6) | 203 (33.6) | 228 (28.9) | 1613 (32.6) | 99 (34.7) | |
| Stable CAD | 715 (14.2) | 150 (14.9) | 782 (14.7) | 83 (11.8) | 189 (15.0) | 606 (14.6) | 70 (11.6) | 98 (12.4) | 722 (14.6) | 45 (15.8) | |
| Other | 154 (3.1) | 39 (3.9) | 173 (3.3) | 20 (2.8) | 41 (3.3) | 138 (3.3) | 14 (2.3) | 29 (3.7) | 156 (3.2) | 8 (2.8) | |
## Clinical restenosis by pregnancy history
Figure 2 shows the cumulative incidence of clinical restenosis by pregnancy history. A history of PTD was not associated with an increased unadjusted rate of clinical restenosis, nor were any of the other studied pregnancy history exposures.
**Figure 2:** *Cumulative incidence of restenosis by aspects of pregnancy history. Figure shows the unadjusted rate of restenosis by aspects of pregnancy history. Event rates were estimated using the Kaplan-Meier method and comparisons made using the log-rank test. SGA, small for gestational age.*
Table 2 presents the results from the per-segment analysis on clinical restenosis following first PCI by pregnancy history. In total, 343 (3.7 %) events occurred following first PCI during a follow-up time of 15 981 segment-years. There was no strong evidence of an association between a history of PTD, a history of ever SGA, parity at time of PCI or age at first birth and clinical restenosis following first PCI, though point estimates for late PTD, SGA and greater parity were all greater than one.
**Table 2**
| Unnamed: 0 | Model I | Model I.1 | Model II | Model II.1 | Model III | Model III.1 |
| --- | --- | --- | --- | --- | --- | --- |
| | HR (95% CI) | P value | HR (95% CI) | P value | HR (95% CI) | P value |
| Preterm delivery (PTD) (events/segment years) | Preterm delivery (PTD) (events/segment years) | Preterm delivery (PTD) (events/segment years) | Preterm delivery (PTD) (events/segment years) | | | |
| No PTD (278/13 204) | 1 (reference) | | 1 (reference) | | 1 (reference) | |
| Ever PTD (65/2.776) | 1.08 (0.77 to 1.53) | 0.65 | 1.17 (0.83 to 1.65) | 0.37 | 1.09 (0.77 to 1.55) | 0.62 |
| Late PTD (50/1933) | 1.21 (0.82 to 1.78) | 0.35 | 1.28 (0.87 to 1.88) | 0.21 | 1.19 (0.81 to 1.76) | 0.37 |
| Very PTD (15/843) | 0.81 (0.44 to 1.50) | 0.50 | 0.91 (0.48 to 1.70) | 0.76 | 0.85 (0.45 to 1.59) | 0.60 |
| Small for gestational age (SGA) (events / segment years) | Small for gestational age (SGA) (events / segment years) | Small for gestational age (SGA) (events / segment years) | Small for gestational age (SGA) (events / segment years) | Small for gestational age (SGA) (events / segment years) | | |
| No SGA (299/14 073) | 1 (reference) | | 1 (reference) | | 1 (reference) | |
| Ever SGA (44/1907) | 1.09 (0.72 to 1.65) | 0.69 | 1.10 (0.72 to 1.66) | 0.69 | 1.13 (0.75 to 1.72) | 0.56 |
| Parity at time of PCI (events / segment years) | Parity at time of PCI (events / segment years) | Parity at time of PCI (events / segment years) | Parity at time of PCI (events / segment years) | | | |
| Parity 1 (70/3332) | 1 (reference) | | 1 (reference) | | 1 (reference) | |
| Parity 2–3 (235/11 029) | 1.00 (0.71 to 1.42) | 0.99 | 1.08 (0.76 to 1.53) | 0.69 | 1.13 (0.79 to 1.61) | 0.51 |
| Parity >4 (38/1618) | 1.09 (0.66 to 1.80) | 0.73 | 1.22 (0.73 to 2.02) | 0.45 | 1.27 (0.77 to 2.11) | 0.35 |
| Age at first delivery (years) (events/segment years) | Age at first delivery (years) (events/segment years) | Age at first delivery (years) (events/segment years) | Age at first delivery (years) (events/segment years) | | | |
| Age <20 (46/2041) | 1 (reference) | | 1 (reference) | | 1 (reference) | |
| Age 20–34 (283/13 187) | 1.00 (0.67 to 1.49) | 0.99 | 0.90 (0.60 to 1.35) | 0.63 | 0.90 (0.60 to 1.37) | 0.63 |
| Age >35 (14/751) | 0.87 (0.43 to 1.75) | 0.69 | 0.74 (0.36 to 1.53) | 0.42 | 0.73 (0.35 to 1.52) | 0.40 |
## TLR by pregnancy history
Figure 3 shows the cumulative incidence of TLR by pregnancy history. None of the pregnancy history variables studied were associated with an increased unadjusted rate of TLR.
**Figure 3:** *Cumulative incidence of target lesion revascularisation by aspects of pregnancy history. Figure shows the unadjusted rate of target lesion revascularisation by aspects of pregnancy history. Event rates were estimated using the Kaplan-Meier method and comparisons made using the log-rank test. SGA, small for gestational age.*
Online supplemental table 2 shows the results from the per-patient analysis on TLR by pregnancy history. In total, 383 (6.4 %) events occurred following first PCI during a follow-up time of 10 103 person-years. There was no strong evidence of associations between the studied pregnancy history variables and TLR following first PCI. Here too, point estimates for late PTD, SGA and greater parity were above one but CIs were wide.
## Discussion
In this prospective cohort study, we show that neither a history of PTD, nor several other aspects of pregnancy history, are strongly associated with clinical restenosis following PCI in parous women <65 years.
Restenosis is an adverse outcome of PCI characterised by an inflammatory response to vessel wall damage.15 Increased inflammatory markers, such as C reactive protein, are seen in patients who develop restenosis after PCI.16 The pregnancy history aspects examined in this study are associated with the development of future CAD, possibly due to inflammation and/or endothelial dysfunction.
PTD is a risk factor for future maternal CAD.4 As stated above, the association has been partly attributed to that PTD and the development of future CAD share common pathways.17–19 Up to a quarter of the association between PTD and future maternal CAD has been shown to be explained by placental disorders such as pre-eclampsia, another female-specific risk factor for future maternal CAD.20 Women with a history of SGA are also at an increased risk for future cardiovascular disease (CVD).21 Even though the pathophysiological background to this association is overall inadequately understood, it has been suggested that a delivery complicated by SGA is associated with endothelial dysfunction and thus future development of CVD. Endothelial dysfunction, and in turn placental dysfunction, is recognised as an underlying pathway for development of future maternal CVD in other pregnancy complications such as pre-eclampsia.22–24 *Parity is* associated with a non-linear increased risk of maternal CVD,3 25 and the increased risk in women with parity >4 could be explained by an association with subclinical atherosclerosis in these women.26 Pregnancy in itself can be seen as an atherogenic state with dyslipidaemia, insulin resistance and weight gain, and it has been hypothesised that a prolonged exposure time to this state is what leads to atherosclerosis and increased risk of CVD in these women.26 Alternatively, raising a larger family may increase CVD risk via adverse effects on diet, sleep, physical activity and other behaviours, as well as weight gain.
Studies have shown a possible inverse association between age at first delivery and future CVD.2 Although socioeconomic factors are highlighted as the most likely explanation, it has also been suggested that the physiological changes a pregnancy entails could affect an adolescent body differently compared with an adult body. It could also be explained by the fact that some studies show an association between low maternal age and adverse pregnancy outcomes that are known predictors of future CVD (eg, PTD and delivering an SGA infant).27 28 We have previously reported that PTD warrants consideration as a risk factor in the secondary prevention setting post-coronary artery stenting.5 Though we report no association between the exposures in this study and clinical restenosis, studies like this could still be clinically relevant. As restenosis is often associated with angina or acute coronary syndrome and patients with restenosis often undergo TLR,15 studies that contribute to a better understanding of groups with a higher risk of adverse outcomes post-PCI can in turn contribute to a better patient outcome. As we have previously mentioned, further studies are needed to investigate how existing strategies for secondary prevention in the post-PCI stetting can reduce the risk of adverse outcomes postcoronary artery stenting in women with a history of PTD.
## Strengths and limitations
The main strength of this study is the comprehensive and national study sample based on data from richly-completed well-known registers.11 13 The extensive nature of both the Medical Birth Register and SCAAR allowed us to include pregnancy data collected over decades and to adjust for several known predictors of restenosis, both procedure related and patient related. Another strength is our use of multiple imputation to account for missing data. However, this study also had some limitations. While our sample covers over a decade of national data, incident restenosis events are relatively rare, and our results do not exclude small to moderate associations for all exposures. Furthermore, we only included women age <65 years. As previously described, age is a strong predictor of worse outcome after PCI29 and older women have much less complete delivery history available in the Medical Birth Registry, which started in 1973. Additionally, we excluded women with PCI before 2006 as not all procedure-related variables were routinely collected before 2006. Our results are dependent on a consistency in PCI-related treatment of the women studied (eg, drug treatment before, during and after PCI). We believe an inconsistency to be unlikely given that pregnancy history is not considered in any relevant guidelines of acute cardiac care and was therefore not likely considered in the care of the women included in this study. Furthermore, we observed no major difference in the restenosis estimates by pregnancy history after adjusting for several patient-related and procedure-related predictors. *The* generalisation of the results could possibly be affected by the ethnic homogeneity of the study sample. Lastly, it should be mentioned that pregnancy dating using ultrasound was not widely used in Sweden until the 1970s, and at the beginning of the Medical Birth Register not all pregnancies were dated using ultrasound.
## Conclusion
In conclusion, neither a history of PTD nor the other aspects of pregnancy history studied were associated with clinical restenosis following PCI. However, larger studies are needed to obtain more precise estimates.
## Data availability statement
No data are available.
## Patient consent for publication
Not applicable.
## Ethics approval
This study involves human participants and was approved by Ethical Review Board in Lund, Sweden. Participants gave informed consent to be included in the SCAAR registry.
## References
1. Jackson EA, Moscucci M, Smith DE. **The association of sex with outcomes among patients undergoing primary percutaneous coronary intervention for ST elevation myocardial infarction in the contemporary era: insights from the blue cross blue shield of Michigan cardiovascular Consortium (BMC2)**. *Am Heart J* (2011) **161** 106-112. DOI: 10.1016/j.ahj.2010.09.030
2. Rosendaal NTA, Pirkle CM. **Age at first birth and risk of later-life cardiovascular disease: a systematic review of the literature, its limitation, and recommendations for future research**. *BMC Public Health* (2017) **17**. DOI: 10.1186/s12889-017-4519-x
3. Lv H, Wu H, Yin J. **Parity and cardiovascular disease mortality: a dose-response meta-analysis of cohort studies**. *Sci Rep* (2015) **5**. DOI: 10.1038/srep13411
4. Wu P, Gulati M, Kwok CS. **Preterm delivery and future risk of maternal cardiovascular disease: a systematic review and meta-analysis**. *J Am Heart Assoc* (2018) **7**. DOI: 10.1161/JAHA.117.007809
5. Pehrson M, Edsfeldt A, Sarno G. **Long-term outcome following coronary artery stenting by history of preterm delivery**. *JACC: Advances* (2022) **1** 100142. DOI: 10.1016/j.jacadv.2022.100142
6. Lin A, Pehrson M, Sarno G. **Coronary artery restenosis in women by history of preeclampsia**. *J Am Heart Assoc* (2022) **11**. DOI: 10.1161/JAHA.122.026287
7. Moussa ID, Mohananey D, Saucedo J. **Trends and outcomes of restenosis after coronary stent implantation in the United States**. *J Am Coll Cardiol* (2020) **76** 1521-31. DOI: 10.1016/j.jacc.2020.08.002
8. Ludvigsson JF, Otterblad-Olausson P, Pettersson BU. **The Swedish personal identity number: possibilities and pitfalls in healthcare and medical research**. *Eur J Epidemiol* (2009) **24** 659-67. DOI: 10.1007/s10654-009-9350-y
9. Cnattingius S, Villamor E, Johansson S. **Maternal obesity and risk of preterm delivery**. *JAMA* (2013) **309** 2362-70. DOI: 10.1001/jama.2013.6295
10. Marsál K, Persson PH, Larsen T. **Intrauterine growth curves based on ultrasonically estimated foetal weights**. *Acta Paediatr* (1996) **85** 843-8. DOI: 10.1111/j.1651-2227.1996.tb14164.x
11. Fokkema ML, James SK, Albertsson P. **Population trends in percutaneous coronary intervention: 20-year results from the SCAAR (Swedish coronary angiography and angioplasty registry)**. *J Am Coll Cardiol* (2013) **61** 1222-30. DOI: 10.1016/j.jacc.2013.01.007
12. Råmunddal T, Hoebers LP, Henriques JPS. **Prognostic impact of chronic total occlusions: a report from SCAAR (swedish coronary angiography and angioplasty registry)**. *JACC Cardiovasc Interv* (2016) **9** 1535-44. DOI: 10.1016/j.jcin.2016.04.031
13. Fröbert O, Lagerqvist B, Carlsson J. **Differences in restenosis rate with different drug-eluting stents in patients with and without diabetes mellitus: a report from the SCAAR (Swedish angiography and angioplasty registry)**. *J Am Coll Cardiol* (2009) **53** 1660-7. DOI: 10.1016/j.jacc.2009.01.054
14. Sarno G, Lagerqvist B, Fröbert O. **Lower risk of stent thrombosis and restenosis with unrestricted use of “ new-generation ” drug-eluting stents: a report from the nationwide Swedish coronary angiography and angioplasty registry (SCAAR)**. *Eur Heart J* (2012) **33** 606-13. DOI: 10.1093/eurheartj/ehr479
15. Buccheri D, Piraino D, Andolina G. **Understanding and managing in-stent restenosis: a review of clinical data, from pathogenesis to treatment**. *J Thorac Dis* (2016) **8** E1150-62. DOI: 10.21037/jtd.2016.10.93
16. Angioi M, Abdelmouttaleb I, Rodriguez RM. **Increased C-reactive protein levels in patients with in-stent restenosis and its implications**. *Am J Cardiol* (2001) **87** 1189-93. DOI: 10.1016/s0002-9149(01)01492-8
17. Ridker PM, Hennekens CH, Buring JE. **C-reactive protein and other markers of inflammation in the prediction of cardiovascular disease in women**. *N Engl J Med* (2000) **342** 836-43. DOI: 10.1056/NEJM200003233421202
18. Rich-Edwards JW, Fraser A, Lawlor DA. **Pregnancy characteristics and women’s future cardiovascular health: an underused opportunity to improve women’s health?**. *Epidemiol Rev* (2014) **36** 57-70. DOI: 10.1093/epirev/mxt006
19. Minissian MB, Kilpatrick S, Eastwood J-A. **Association of spontaneous preterm delivery and future maternal cardiovascular disease**. *Circulation* (2018) **137** 865-71. DOI: 10.1161/CIRCULATIONAHA.117.031403
20. Auger N, Potter BJ, He S. **Maternal cardiovascular disease 3 decades after preterm birth: longitudinal cohort study of pregnancy vascular disorders**. *Hypertension* (2020) **75** 788-95. DOI: 10.1161/HYPERTENSIONAHA.119.14221
21. Haug EB, Markovitz AR, Fraser A. **The role of cardiovascular risk factors in maternal cardiovascular disease according to offspring birth characteristics in the HUNT study**. *Sci Rep* (2021) **11**. DOI: 10.1038/s41598-021-99478-4
22. Chambers JC, Fusi L, Malik IS. **Association of maternal endothelial dysfunction with preeclampsia**. *JAMA* (2001) **285** 1607-12. DOI: 10.1001/jama.285.12.1607
23. Wu P, Haththotuwa R, Kwok CS. **Preeclampsia and future cardiovascular health: a systematic review and meta-analysis**. *Circ Cardiovasc Qual Outcomes* (2017) **10**. DOI: 10.1161/CIRCOUTCOMES.116.003497
24. Sankaran S, Kyle PM. **Aetiology and pathogenesis of IUGR**. *Best Pract Res Clin Obstet Gynaecol* (2009) **23** 765-77. DOI: 10.1016/j.bpobgyn.2009.05.003
25. Li W, Ruan W, Lu Z. **Parity and risk of maternal cardiovascular disease: a dose-response meta-analysis of cohort studies**. *Eur J Prev Cardiol* (2019) **26** 592-602. DOI: 10.1177/2047487318818265
26. Sanghavi M, Kulinski J, Ayers CR. **Association between number of live births and markers of subclinical atherosclerosis: the Dallas heart study**. *Eur J Prev Cardiol* (2016) **23** 391-9. DOI: 10.1177/2047487315571891
27. Chen X-K, Wen SW, Fleming N. **Teenage pregnancy and adverse birth outcomes: a large population based retrospective cohort study**. *Int J Epidemiol* (2007) **36** 368-73. DOI: 10.1093/ije/dyl284
28. Schummers L, Hacker MR, Williams PL. **Variation in relationships between maternal age at first birth and pregnancy outcomes by maternal race: a population-based cohort study in the United States**. *BMJ Open* (2019) **9**. DOI: 10.1136/bmjopen-2019-033697
29. Kunadian V, Qiu W, Lagerqvist B. **Gender differences in outcomes and predictors of all-cause mortality after percutaneous coronary intervention (data from United Kingdom and Sweden)**. *Am J Cardiol* (2017) **119** 210-6. DOI: 10.1016/j.amjcard.2016.09.052
|
---
title: 'Growth patterns in childhood and adolescence and adult body composition: a
pooled analysis of birth cohort studies from five low and middle-income countries
(COHORTS collaboration)'
authors:
- Natalia E Poveda
- Linda S Adair
- Reynaldo Martorell
- Shivani A Patel
- Manuel Ramirez-Zea
- Santosh K Bhargava
- Sonny A Bechayda
- Delia B Carba
- Maria F Kroker-Lobos
- Bernardo Lessa Horta
- Natália Peixoto Lima
- Mónica Mazariegos
- Ana Maria Baptista Menezes
- Shane A Norris
- Lukhanyo H Nyati
- Linda M Richter
- Harshpal Sachdev
- Fernando C Wehrmeister
- Aryeh D Stein
journal: BMJ Open
year: 2023
pmcid: PMC10030655
doi: 10.1136/bmjopen-2022-068427
license: CC BY 4.0
---
# Growth patterns in childhood and adolescence and adult body composition: a pooled analysis of birth cohort studies from five low and middle-income countries (COHORTS collaboration)
## Abstract
### Objective
We examined associations among serial measures of linear growth and relative weight with adult body composition.
### Design
Secondary data analysis of prospective birth cohort studies.
### Settings
Six birth cohorts from Brazil, Guatemala, India, the Philippines and South Africa.
### Participants
4173 individuals followed from birth to ages 22–46 years with complete and valid weight and height at birth, infancy, childhood and adolescence, and body composition in adult life.
### Exposures
Birth weight and conditional size (standardised residuals of height representing linear growth and of relative weight representing weight increments independent of linear size) in infancy, childhood and adolescence.
### Primary outcome measures
Body mass index, fat mass index (FMI), fat-free mass index (FFMI), fat mass/fat-free mass ratio (FM/FFM), and waist circumference in young and mid-adulthood.
### Results
In pooled analyses, a higher birth weight and relative weight gains in infancy, childhood and adolescence were positively associated with all adult outcomes. Relative weight gains in childhood and adolescence were the strongest predictors of adult body composition (β ($95\%$ CI) among men: FMI (childhood: 0.41 (0.26 to 0.55); adolescence: 0.39 (0.27 to 0.50)), FFMI (childhood: 0.50 (0.34 to 0.66); adolescence: 0.43 (0.32 to 0.55)), FM/FFM (childhood: 0.31 (0.16 to 0.47); adolescence: 0.31 (0.19 to 0.43))). Among women, similar patterns were observed, but, effect sizes in adolescence were slightly stronger than in childhood. Conditional height in infancy was positively associated with FMI (men: 0.08 (0.03 to 0.14); women: 0.11 (0.07 to 0.16)). Conditional height in childhood was positively but weakly associated with women’s adiposity. Site-specific and sex-stratified analyses showed consistency in the direction of estimates, although there were differences in their magnitude.
### Conclusions
Prenatal and postnatal relative weight gains were positive predictors of larger body size and increased adiposity in adulthood. A faster linear growth in infancy was a significant but weak predictor of higher adult adiposity.
## Introduction
Low and middle-income countries (LMICs) have witnessed an increase in the prevalence of cardiometabolic risk factors such as diabetes, hypertension and obesity in the adult population.1–3 Overweight and obesity are not only metabolic outcomes but also serve as risk factors for other cardiometabolic diseases.4 The Developmental Origins of Health and *Disease hypothesis* proposes that anatomical, physiological and metabolic adaptations in early life have long-lasting effects on the onset of adult cardiometabolic disease.5 Adaptations in body composition might have an important role in the programming of cardiometabolic disease ‘through its own programming by early growth, and/or through being a mediator of the programming process’.6 A recent review concluded that associations between infant growth and long-term cardiometabolic outcomes might be mediated by body composition after estimates were attenuated when controlling for current body mass index (BMI).7 Similarly, other systematic reviews and one meta-analysis showed that childhood obesity is a risk factor of cardiometabolic outcomes in adulthood. However, results were inconclusive in demonstrating whether or not this excessive weight gain in childhood was an independent factor or whether most of this effect was mediated through contemporary adult BMI.8–10 BMI in infancy, childhood and adolescence has a moderate to strong tracking into adult adiposity.8 11 However, the discrimination between body compartments such as fat and lean mass, which have different metabolic functions, is more relevant to understand cardiometabolic risk than BMI alone.12 The study of the role of growth across the life span on body composition has demonstrated similarities and differences when comparing body compartments. Overall studies have shown that birth weight is positively associated with lean mass later in life,6 and that it is a stronger predictor of fat-free mass (FFM) than fat mass (FM).13 14 Less consistency (positive, inverse and null findings) has been observed in studies that have examined the association between birth weight (low and high) with adiposity in childhood, adolescence and adulthood.6 15 16 Additionally, studies in high-income countries (HICs) have shown that a rapid weight gain (≥ 0.67 SD) in infancy (first 24 months after birth) is associated with higher obesity and FM in childhood and adulthood,7 17 whereas findings in LMICs have shown that weight gain in infancy is associated with higher FM and FFM in adult life, being a stronger predictor of FFM.6 7 17 *There is* a paucity of studies about the long-term effects of growth across the life span on adult body composition. Available evidence has primarily examined associations between prenatal and postnatal growth (mainly in infancy) with body composition in childhood and adolescence. In a recent literature review of publications between 2003 and 2018, only 6 out of 39 studies examined associations between weight gain during infancy and body composition in adult participants from LMICs.7 Less evidence is available about the role of linear growth (height) on adult body composition, and findings are mixed.7 *There is* a need for investigating the independent long-term role of linear growth and relative weight gain across the life course on body composition in young and mid-adulthood; stages in which the accumulation of cardiometabolic risk is higher.
Pooled analyses of the Consortium of Health-Oriented Research in Transitional Societies (COHORTS), a consortium of birth cohorts from five LMICs,18 showed that higher weight at birth, in infancy and in mid-childhood were associated with a higher risk of adult overweight (ages 18–31 years).13 These pooled analyses also showed that weight at birth and in infancy were stronger predictors of FFM than of FM in adult life, while weight gain in mid-childhood was a stronger predictor of FM than FFM.13 14 We expand on these previous studies by [1] Analysing for the first time the period between childhood and adolescence, [2] Examining the long-term role of child growth on body composition in young and mid-adulthood (ages 22–46 years) given that in previous analyses some study participants were adolescents, and [3] Using body composition indices (height squared adjustment) to assess body compartments (weight, FM, and FFM) independent of linear size. Specifically, we now examine the associations of linear growth and relative weight at birth, infancy, childhood and adolescence with measures of body size and composition (BMI, fat mass index (FMI), fat-free mass index (FFMI), FM/FFM ratio and waist circumference) in young and mid-adulthood.
## Study design and participants
We conducted a secondary data analysis of COHORTS.18 *This consortium* comprises six prospective birth cohorts in five study sites: [1] The 198219 20 and 1993 Pelotas Birth Cohorts in Brazil;21 [2] The Institute of Nutrition of Central America and Panama Nutrition Trial and Longitudinal Study in Guatemala;22 [3]The New Delhi Birth Cohort in India;23 [4] The Cebu Longitudinal Health and Nutrition Survey in the Philippines;24 and [5] The Birth to Twenty Plus Cohort in South Africa.25 Table 1 presents a general description of each birth cohort.
**Table 1**
| Country (cohort) | Study design | Cohort enrolment | Age at enrolment | Initial sample | Analytical sample | Age and year at most recent follow-up | Body composition methods |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Brazil (1982 Pelotas Birth Cohort) | Prospective cohort | 1982 | Birth | 5914 | 674 | 30(2012) | BodPod and anthropometry |
| Brazil (1993 Pelotas Birth Cohort) | Prospective cohort | 1993 | Birth | 5249 | 827 | 22.5(2015) | BodPod and anthropometry |
| Guatemala (INCAP Nutrition Trial and Longitudinal Study) | Community trial | 1969–1977 | Pregnancy to 7 years | 2392 | 163 | 45.5(2015–2017) | Deuterium oxide dilution technique and anthropometry |
| India (NDBC) | Prospective cohort | 1969–1972 | Before pregnancy | 8181 | 681 | 46(2016–2019) | Bioimpedance and anthropometry |
| The Philippines (CLHNS) | Prospective cohort | 1983–1984 | Birth | 3080 | 1197 | 34.5(2018) | Bioimpedance and anthropometry |
| South Africa (Birth to Twenty Plus Cohort) | Prospective cohort | 1990 | Pregnancy | 3273 | 595 | 22(2012) | DEXA (Dual-energy X-ray absorptiometry) and anthropometry |
We selected one analytical sample for each study site (figure 1). Of a total of 27 437 participants, we excluded participants with missing body composition in adulthood ($$n = 16$$ 029), implausible values of BMI ≤12 kg/m2 or ≥ 60 kg/m2 and waist circumference ≤51 cm or ≥ 190 cm ($$n = 1$$), and pregnancy at the moment of the anthropometric assessment ($$n = 135$$). We also excluded those individuals with missing anthropometric data at birth, childhood or adolescence ($$n = 6967$$), implausible values of weight and height considering the WHO Z-score thresholds ($$n = 168$$),26 27 and pregnancy at the adolescence measurement ($$n = 47$$).
**Figure 1:** *Flow chart of analytical samples selection, Consortium of Health-Oriented Research in Transitional Societies (COHORTS).*
## Outcomes: adult body composition
Study sites applied different standard methods to measure body composition in adult participants such as dual-energy X-ray absorptiometry, air-displacement plethysmography (BodPod), deuterium oxide dilution technique, bioimpedance19 28–30 and anthropometry (table 1). Our outcomes measures are: [1] BMI (weight in kg/height in metres2) as a proxy of body size; [2] FMI to represent the fat compartment adjusted for size (FM in kg/height in metres2); [3] FFMI to represent the lean compartment adjusted for size (FFM in kg/height in metres2); [4] FM/FMM ratio (a higher ratio indicates a greater increment in the FM component); and [5] The waist circumference (centimetres) as a proxy of abdominal adiposity. We transformed these outcomes measures into Z-scores (SD units) to compare effect sizes within strata of site and sex, and among the measures.
## Anthropometry at birth, infancy, childhood and adolescence
Birth weight was measured in all study sites, while birth length was available only in four of the six study sites (except Brazil 1982 and South Africa). In Brazil, anthropometric data at birth was measured at the hospitals by the research team (Brazil 1993) or by the hospital personnel (Brazil 1982) using paediatric scales.31 In Guatemala, the study team measured weight within the first 48 hours after birth and length within the first 15 days after the delivery, using a beam balance and a measuring board.32 In India, trained project staff measured birth and length within the first 72 hours of life using standard instruments.33 In the Philippines, the $60\%$ of the participants who were born at home were weighted by local birth attendants using hanging scales and the remaining participants were measured in hospitals, and birth length was measured within the first 6 days of life.34 In South Africa, birth weight measured by hospital personnel was obtained from hospital records.
Trained research staff at each study site followed standard procedures to take anthropometric measures of length and/or height and weight in subsequent follow-ups (childhood and adolescence). Growth measures that represent linear growth (length/height) and mass accretion independent of linear size (relative weight) were measured in different periods at each study site. We identified four common or near ages where anthropometric data were available across all cohorts: [1] Birth, [2] Infancy (all study sites registered weight and length data at 24 months of age except Brazil 1993 that collected these data at 12 months), [3] Childhood (five study sites collected growth data at 48 months of age, but the Philippines collected data at 102 months), and [4] Adolescence (all study sites collected anthropometric data at 180 months of age). To be able to compare growth measures between study sites, we estimated sex-specific length/height-for-age and weight-for-age Z-scores (HAZ and WAZ, respectively) using the WHO growth standards.35
## Exposures: conditional growth (height and relative weight)
Repeated measures of linear growth and relative weight are strongly correlated. To eliminate this correlation, we estimated conditional size measures (standardised residuals) by regressing current HAZ on previous HAZ and WAZ measures, and regressing current WAZ on current HAZ and all prior HAZ and WAZ measures, within strata of site and sex. This approach allowed us to identify if children’s growth deviates from their expected length/height or weight as estimated based on their own previous growth measurements and the growth of other children. We used birth weight as the anchor to estimate conditional growth measures (height and relative weight) in the main analysis of all cohorts. In separate models, we derived conditional growth measures using birth length as an anchor in the four study sites in which this variable was available. We estimated conditional height and conditional relative weight in four age intervals across all study sites: [1] Prenatal (from conception to birth captured though birth weight and birth length); [2] Infancy (birth to 12 months in Brazil 1993, and birth to 24 months in the other study sites); [3] Childhood (12–48 months in Brazil 1993, 24–48 months in Brazil 1982, Guatemala, India, and South Africa, and 24–102 months in the Philippines); and [4] Adolescence (48–180 months in five study sites and 102–180 months in the Philippines).
Covariates: We identified potential confounders available at each study site such as birth characteristics (gestational age, birth order), income and/or wealth index of child’s household at birth, maternal characteristics (height, age at first childbirth, schooling attainment, marital status), paternal characteristics (age, schooling attainment), reproductive factors (age at menarche) and age in adulthood (online supplemental table 1).
## Statistical analyses
We used R V.4.0 for all statistical analyses. Statistical significance was established at a value of $p \leq 0.05.$
## Site-specific regression models
We used multiple linear regressions with standardised coefficients to estimate the associations between growth patterns at the four age intervals and adult body composition since all outcomes were continuous and without major violations of the normal distribution (graphical examination). We built site-specific models stratified by sex given the biological differences in body composition.36 We ran two sets of models: [1] Unadjusted models for all study sites except Guatemala, in which we minimally adjust for birth year and intervention group considering the initial study design of this cohort;22 [2] Fully adjusted models controlling for potential confounders available at each study site (online supplemental table 1). Additionally, we adjusted for adult height in waist circumference models, for age at menarche and teenage childbearing in models of women, and for maternal skin colour in both Brazil cohorts as one proxy of socioeconomic status. In Brazil 1993 models, we used survey procedures given that this cohort collected data in subsamples (all children with low weight at birth and a random sample of other children) at 12 months and 48 months of age.31 *After a* visual examination of effect sizes from both sets of models at each study site, we observed slight changes in magnitude but not in direction (results not presented). We present the fully adjusted models to account for the confounding effect of this group of covariates.
## Pooling effect sizes
The cohorts come from different study sites and data were collected in different periods, resulting in heterogeneity among site-specific estimates. We used a random-effects meta-analysis approach to pool effect sizes to account for this underlying heterogeneity. Given that our outcomes were continuous, we used the restricted maximum likelihood estimator to calculate the between-study heterogeneity (variance of the distribution of true effect sizes).37 We also applied Knapp-Hartung adjustments in the estimation of CIs of pooled effects. We estimated Cochran’s Q with its corresponding p value to test whether there was heterogeneity between study sites, and calculated the I2 statistic to quantify the magnitude of between-study heterogeneity. As a rule of thumb an I2 equal to $25\%$ is interpreted as low heterogeneity, $50\%$ as moderate heterogeneity and $75\%$ as substantial heterogeneity.37 We used the Metafor package in R to run these analyses.
## Multiple imputation
To deal with missing data, we built sex-specific multiple imputation models at each study site. Missing data patterns are presented in online supplemental table 1. We assumed that the underlying missing pattern was missing at random (MAR). In brief, we used the R package Multivariate Imputation by Chained Equations to generate different multiple imputed data sets with a maximum of 50 iterations. We estimated the number of imputations based on the highest percentage of missingness within strata of site and sex. In our imputation models, we included all covariates used in the main analyses and the outcome variables. We used the ‘with and pool’ functions to run the main analyses in each imputed data set and to combine the estimates into a single result with its corresponding SEs and CIs.
## Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
## Results
We analysed data from a combined analytical sample of 4137 participants with complete and valid information of body composition in adulthood and anthropometry across the life course (figure 1). The comparison of general characteristics between participants included in the final analytical samples versus those excluded only showed significant differences in a couple of variables in selected cohorts (online supplemental table 2). General characteristics of the six cohorts, and anthropometric characteristics at birth and adulthood of men and women at each study site are presented in tables 1 and 2, respectively. Participants’ average adult age ranged from 22 years in Brazil 1993 and South Africa to 45–46 years in Guatemala and India. India had the lowest mean birth weight compared with other study sites. The proportion of excess weight and adiposity were correlated with the age of the cohorts, with the oldest cohorts having the greatest proportion of BMI ≥25 kg/m2 and the highest FMI. Among men, the proportion of excess weight ranged from $16.0\%$ in South Africa to $73.2\%$ in India, and among women it ranged from $44.5\%$ in Brazil 1993 to $85.4\%$ in Guatemala. Participants from Guatemala, India and Brazil 1982 had the highest FMI (women above 10 kg/m2 and men above 7 kg/m2). FFMI ranged from 15.5 kg/m2 to 20.2 kg/m2 in men, and from 13.7 kg/m2 to 17.7 kg/m2 in women. Other characteristics (at birth, childhood and parental) of study participants at each study site are presented in online supplemental table 3.
**Table 2**
| Unnamed: 0 | Brazil 1982 | Brazil 1982.1 | Brazil 1993 | Brazil 1993.1 | Guatemala | Guatemala.1 | India | India.1 | The Philippines | The Philippines.1 | South Africa | South Africa.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | Men(n=344) | Women(n=330) | Men(n=375) | Women(n=452) | Men(n=67) | Women(n=96) | Men(n=426) | Women(n=255) | Men(n=656) | Women(n=541) | Men(n=287) | Women(n=308) |
| Birth length (cm) | | | 49.4 (2.3) | 48.4 (2.1) | 49.6 (2.6) | 48.5 (2.2) | 48.7 (2.0) | 47.9 (2.1) | 49.3 (2.0) | 48.7 (2.0) | | |
| Birth weight (kg) | 3.3 (0.6) | 3.2 (0.5) | 3.3 (0.5) | 3.1 (0.5) | 3.1 (0.5) | 3.0 (0.5) | 2.9 (0.4) | 2.7 (0.4) | 3.0 (0.4) | 3.0 (0.4) | 3.1 (0.5) | 3.0 (0.5) |
| Adult height (cm) | 174.9 (7.1) | 161.9 (6.2) | 174.8 (7.6) | 160.8 (6.6) | 164.7 (5.7) | 152.0 (4.8) | 169.9 (6.8) | 154.5 (5.5) | 162.8 (5.6) | 150.9 (5.5) | 171.5 (6.2) | 159.7 (6.4) |
| Adult weight (kg) | 82.8 (16.3) | 70.1 (16.6) | 75.8 (16.1) | 66.4 (15.6) | 71.4 (14.3) | 69.4 (11.8) | 80.6 (15.0) | 70.0 (12.5) | 66.1 (12.8) | 57.3 (11.9) | 62.9 (11.3) | 64.6 (14.9) |
| Adult body mass index (BMI) (kg/m2†) | 27.0 (4.8) | 26.7 (6.1) | 24.6 (4.7) | 25.5 (5.9) | 26.3 (4.8) | 30.0 (4.7) | 27.9 (4.7) | 29.3 (4.8) | 24.9 (4.4) | 25.2 (5.0) | 21.4 (3.5) | 25.3 (5.6) |
| Adult excess weight† | 215 (62.5%) | 171 (51.8%) | 160 (40.4%) | 195 (44.5%) | 42 (62.7%) | 82 (85.4%) | 312 (73.2%) | 208 (81.6%) | 297 (45.3%) | 256 (47.3%) | 46 (16.0%) | 142 (46.1%) |
| Adult waist circumference (cm) | 89.9 (11.5) | 80.8 (11.7) | 82.2 (10.8) | 77.3 (11.6) | 93.7 (10.8) | 102.8 (10.7) | 100.6 (12.0) | 91.8 (10.5) | 82.1 (10.8) | 80.8 (11.5) | 75.8 (8.6) | 82.4 (13.5) |
| Adult fat mass index (FMI, kg/m2†) | 7.1 (3.6) | 10.5 (4.7) | 5.3 (3.7) | 9.6 (4.5) | 7.7 (2.8) | 12.7 (3.1) | 7.7 (2.8) | 11.6 (3.4) | 5.5 (2.5) | 9.0 (3.5) | 4.0 (2.1) | 9.6 (3.9) |
| Adult fat-free mass index (FFMI, kg/m2†) | 19.9 (2.1) | 16.2 (1.9) | 19.2 (1.9) | 15.8 (1.9) | 18.5 (2.6) | 17.3 (2.2) | 20.2 (2.2) | 17.7 (1.7) | 19.3 (2.0) | 16.1 (1.4) | 15.5 (1.8) | 13.7 (1.9) |
| Adult fat mass/fat-free mass ratio | 0.4 (0.2) | 0.6 (0.2) | 0.3 (0.2) | 0.6 (0.2) | 0.4 (0.1) | 0.7 (0.1) | 0.4 (0.1) | 0.6 (0.2) | 0.3 (0.1) | 0.5 (0.2) | 0.3 (0.1) | 0.7 (0.2) |
## Conditional relative weight and adult body composition
Table 3 shows the pooled adjusted associations of conditional growth with adult FMI and FFMI. Birth weight and conditional relative weight in infancy, childhood and adolescence were positively associated with FMI and FFMI in adulthood. A higher birth weight was a predictor of higher FMI (men: 0.07 ($95\%$ CI 0.01 to 0.14); women: 0.09 ($95\%$ CI 0.01 to 0.17)) and FFMI (men: 0.11 ($95\%$ CI 0.05 to 0.18); women: 0.12 ($95\%$ CI 0.05 to 0.19)). An increment of 1 SD in conditional relative weight in infancy was associated with an increment of 0.2–0.3 SD in adult FMI, while for every SD increment in conditional weight in childhood or adolescence there was a 0.4–0.5 SD increase in FMI in adulthood. Similarly, the associations of relative weight gain with adult FFMI were positive and stronger as boys and girls grew older. Increments in conditional relative weight by 1 SD in infancy, childhood and adolescence were associated with higher FFMI by 0.3 SD, 0.4–0.5 SD and 0.4 SD units in adult life, respectively. Table 4 shows the pooled adjusted associations of conditional growth with adult FM/FFM ratio. Birth weight was not significantly associated with FM/FFM ratio in adult life. Conditional relative weight in infancy, childhood and adolescence were associated with a higher FM/FMM ratio with increments ranging from 0.17 to 0.43 for every SD in conditional weight.
The between-study heterogeneity assessment showed low or moderate heterogeneity across almost all associations of conditional growth with FMI and FM/FFM, except conditional relative weight in childhood among women (I2=$83.9\%$) and conditional relative weight in adolescence among men (I2=$78.0\%$) (tables 3 and 4). The between-study heterogeneity assessment of FFMI models showed substantial variability in estimates of conditional relative weight in childhood and adolescence among women and men, and conditional height in childhood and adolescence among women (I2≥$75\%$) (table 3).
Site-specific estimates were consistent in magnitude and direction. Birth weight was positively associated with adult FMI in men and women from India and the Philippines, and in women from Brazil 1993. Conditional relative weights in infancy, childhood and adolescence were positively associated with adult FMI among men and women across all cohorts (online supplemental table 4). Site-specific analyses showed that birth weight was a predictor of a higher adult FFMI among men and women from four out of the six cohorts (Brazil 1982, India, the Philippines and South Africa), and among women from Brazil 1993 and Guatemala. Across all cohorts but Guatemala, a higher relative weight gain in childhood was significantly associated with a higher FFMI. Conditional relative weight in adolescence was a strong and significant predictor of a higher FFMI in adult life among all study sites and both sexes (online supplemental table 5). Birth weight was positively and significantly associated with FM/FFM among men and women from India and the Philippines, and women from Brazil 1993. Conditional relative weights in infancy, childhood and adolescence were positively associated with FM/FFM across all study sites with a few exceptions (online supplemental table 6).
Table 5 shows the pooled adjusted associations of conditional growth with adult BMI and waist circumference. Birth weight and all conditional relative weight measures were positively associated with BMI and waist circumference among adult men and women. Effect sizes at birth were small but significant predictors of BMI and waist circumference. Associations strengthened with age at measurement; for example, men’s BMI increased by 0.29 ($95\%$ CI 0.24 to 0.35) SD units, 0.50 ($95\%$ CI 0.37 to 0.64) SD units and 0.47 ($95\%$ CI 0.35 to 0.60) SD units per each SD in conditional relative weight in infancy, childhood and adolescence, respectively. Similar patterns were observed in women. Relative weight gains in infancy and childhood were associated with increments in waist circumference ranging from 0.25 SD to 0.26 SD and 0.40–0.43 SD among adult men and women, respectively. Increments in relative weight during adolescence predicted increments in waist circumference by 0.39 ($95\%$ CI 0.27 to 0.52) SD units in men and 0.49 ($95\%$ CI 0.42 to 0.56) SD units in women. The between-study heterogeneity was low or moderate for most of the associations, except some associations of conditional relative weight in childhood and adolescence (heterogeneities above $75\%$) (table 5).
**Table 5**
| Unnamed: 0 | BMI (SD units) | BMI (SD units).1 | BMI (SD units).2 | BMI (SD units).3 | BMI (SD units).4 | BMI (SD units).5 | Waist circumference (SD units) | Waist circumference (SD units).1 | Waist circumference (SD units).2 | Waist circumference (SD units).3 | Waist circumference (SD units).4 | Waist circumference (SD units).5 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | Men(n=2155) | Men(n=2155) | Men(n=2155) | Women(n=1982) | Women(n=1982) | Women(n=1982) | Men(n=2155) | Men(n=2155) | Men(n=2155) | Women(n=1982) | Women(n=1982) | Women(n=1982) |
| | β (95% CI) | I2† Statistic (%)‡ | Cochran’s Q test(P value) | β (95% CI) | I2† Statistic (%) | Cochran’s Q test (P value) | β (95% CI) | I2† Statistic (%) | Cochran’s Q test(P value) | β (95% CI) | I2† Statistic (%) | Cochran’s Q test (P value) |
| Conditional growth, Z-scores† | | | | | | | | | | | | |
| Birth weight | 0.10*(0.05 to 0.14) | 13.7 | 0.454 | 0.12*(0.03 to 0.20) | 47.0 | 0.056 | 0.08*(0.03 to 0.12) | 0.0 | 0.573 | 0.09*(0.01 to 0.16) | 0.1 | 0.260 |
| Conditional relative weight in infancy | 0.29*(0.24 to 0.35) | 0.0 | 0.602 | 0.30*(0.24 to 0.36) | 0.0 | 0.634 | 0.26*(0.18 to 0.33) | 17.9 | 0.345 | 0.25*(0.19 to 0.32) | 0.0 | 0.638 |
| Conditional relative weight in early/mid-childhood | 0.50*(0.37 to 0.64) | 58.6 | 0.044 | 0.53*(0.31 to 0.75) | 78.4 | 0.001 | 0.43*(0.35 to 0.52) | 0.0 | 0.551 | 0.40*(0.16 to 0.63) | 79.3 | 0.002 |
| Conditional relative weight in adolescence | 0.47*(0.35 to 0.60) | 85.5 | <0.001 | 0.58*(0.47 to 0.69) | 69.4 | 0.006 | 0.39*(0.27 to 0.52) | 79.9 | <0.001 | 0.49*(0.42 to 0.56) | 30.0 | 0.285 |
| Conditional height in infancy | 0.10*(0.03 to 0.16) | 36.5 | 0.097 | 0.12*(0.08 to 0.16) | 0.0 | 0.736 | 0.12*(0.03 to 0.21) | 48.2 | 0.089 | 0.08(−0.05 to 0.20) | 19.8 | 0.216 |
| Conditional height in early/mid-childhood | 0.07(−0.04 to 0.19) | 58.4 | 0.043 | 0.12(−0.01 to 0.25) | 53.2 | 0.066 | 0.11(−0.04 to 0.26) | 57.3 | 0.04 | 0.14*(0.06 to 0.23) | 0.0 | 0.866 |
| Conditional height in adolescence | 0.01(−0.07 to 0.09) | 0.0 | 0.307 | 0.01(−0.1 to 0.13) | 37.3 | 0.225 | 0.08(0.00 to 0.16) | 0.0 | 0.596 | −0.04(−0.26 to 0.18) | 41.7 | 0.161 |
Site-specific analyses of BMI and waist circumference showed similar effect sizes in direction and magnitude as pooled estimates. Relative weight gains in infancy, childhood and adolescence were predictors of higher BMI and waist circumference in adulthood across study sites, with a few exceptions (online supplemental tables 7 and 8).
## Conditional height and adult body composition
Conditional height in infancy was positively associated with FMI and FM/FFM ratio in adult life, but effect sizes were small (around 0.1) (tables 3 and 4). None of the conditional height variables were significantly associated with adult FFMI (table 3). A high conditional height in infancy was a significant but weak predictor of higher adult BMI among men and women, and waist circumference among women (increments of 0.10–0.12 SD units) (table 5). Conditional height in childhood was positively but weakly associated with women’s FMI, FM/FFM ratio and waist circumference (tables 3–5).
We used conditionals derived from birth length as the anchor in models in four of the study sites. In these pooled analyses, we found that birth length was a weak but statistically significant predictor of higher FFMI and BMI in adult women (online supplemental tables 9 and 10). Other effect sizes from these models were consistent in direction and magnitude with effect sizes observed in the main models (birth weight as anchor). Site-specific analyses using birth length as anchor are presented in online supplemental tables 11–14.
Body size and composition indices should control for allometric scaling (differences in body compartments given differences in linear size) and should be independent of their denominator (height).38 39 Thus, to assess the statistical validity of body size and composition indices, we examined the correlation between these indices with adult height. We observed that the three indices (BMI, FMI and FFMI) were not significantly correlated with height across all study sites and sexes (correlation coefficients ranged from −0.15 to 0.10). The correlation coefficients of waist circumference with adult height were significant but weak (0.15–0.26) (online supplemental table 15).
## Discussion
In a pooled analysis of six prospective birth cohorts from five LMICs, we found that a higher birth weight predicted higher adult FFMI, FMI, BMI and waist circumference, but not adiposity (FM/FFM). Birth length was a positive predictor of FFMI and BMI, only among adult women. Higher relative weight gains in infancy, childhood and adolescence were associated with higher FFMI, FMI, FM/FFM ratio, BMI and waist circumference in adult life. Associations strengthened with age, thus, effect sizes were the strongest in childhood and adolescence, almost twice the magnitude of effect sizes in infancy, and four to five times the magnitude of birth estimates. Our findings also showed that a higher conditional height in infancy was a significant but weak predictor of higher FMI and BMI among adult men and women, and higher waist circumference among women. A higher conditional height in childhood was positively but weakly associated with women’s FMI, FM/FFM ratio and waist circumference.
This updated analysis confirms previous results from COHORTS13 14 and similar site-specific studies28 33 40 showing that birth weight is a predictor of higher adult FFMI and FMI. Thus, our findings showed the independent and long-lasting role of prenatal growth (weight gain mainly) on body composition and size in young and middle-age adulthood. There are differences in effect sizes when comparing our estimates of FMI and FFMI with previous COHORTS' analysis because we adjusted by adult height, used a different approach to combine site-specific estimates, participants were older, and possibly due to the decreased association of birth weight with adult measures across time. The positive and significant associations between prenatal and postnatal relative weight gains with all outcomes in young and mid-adulthood suggest that relative weight is a predictor of overall body size than body composition. A recent review of literature highlighted a similar conclusion in relation to rapid weight gain in infancy (0–2 years of age) with body size in childhood, adolescence and adulthood.7 We expand on that conclusion by showing that excessive relative weight gain beyond 2 years of age is also a predictor of overall size in young and middle-age adulthood. Additionally, we infer that a higher FFMI reflects the accretion in FFM as a body response (expansion of bone and muscle mass) to support higher weight,41 42 since on average COHORTS participants had overweight or obesity in adulthood. This inference is also supported by the finding that conditional relative weight in infancy, childhood and adolescence were associated with a greater FM/FFM ratio, indicating that the increments in FM were faster than the increments in FMM. Thus, higher weight gains during these age intervals might be associated with a greater risk of adiposity in adult life.
Compared with previous combined analyses of COHORTS, we examined for the first time the interval from childhood to adolescence. Our findings showed that relative weight gain in adolescence was associated with higher adult adiposity (FMI, FM/FFM ratio and waist circumference) in consistency with earlier studies in LMICs.13 14 28 Studies in HICs, like an analysis of the Fels Longitudinal Study showed similar trends. This study found that timing and velocity of BMI in childhood, adolescence and postadolescence were predictors of higher BMI and adiposity in adulthood (ages 35–45 years), adolescence and postadolescence being the strongest predictors of adult adiposity.43 Whereas we observed similar effect sizes in childhood and adolescence suggesting that both periods are equally important for predicting the risk of higher adiposity in adult life. In a deeper understanding of the predicting role of growth on adult adiposity, recent analyses of Brazil 198244 and South Africa28 cohorts found that conditional relative weight gains across different stages of the life course were significant predictors of both visceral and subcutaneous fat at ages 30 years and 22 years, respectively.44 A combination of biological, environmental, behavioural and sociodemographic factors is associated with weight gain and adiposity in childhood and adolescence that may not only persist later in life but might undermine the positive role of a healthy birth weight and linear growth early in life. Some of these factors are the medium to strong tracking of child and adolescent BMI into adult adiposity,8 11 prenatal and postnatal factors,45 reproductive characteristics of women (earlier age at menarche, short periods between age at menarche and age at first pregnancy),46 and changes in lifestyle patterns (diets of obesogenic environments, a decline in physical activity, and an increment in sedentary activities) as consequence of the nutrition transition and transformations of the food systems, home and work environments in LMICs during the last decades.47 Fewer studies have examined the role of linear growth on later body composition and results are mixed.7 Our results showed that linear growth in infancy was a significant but weak predictor of adiposity in adult life (FMI and BMI in men and women, and waist circumference in women). These findings are also consistent with previous reports of COHORTS13 and South Africa birth cohort,28 but our estimates were smaller given the adjustment by adult height. Our findings showed that linear growth was not a predictor of adult FFMI, which suggests that linear growth is a predictor of adult height rather than FFM. In contrast, an earlier study in South Africa showed that linear growth across five age intervals (from 0 to 22 years of age) predicted a higher FFM in young adulthood,28 and an analysis conducted in Guatemala before the start of the obesity epidemic found that linear growth was associated with higher FFM in adolescence.48 Linear growth is determined by the hormonal regulation of the growth hormone (GH) secreted by the pituitary and the insulin-like growth factor 1 (IGF-1), an intermediary hormone between the GH and target tissues.49 Based on human models, growth during the first year after birth is described as an extension of fetal growth,50 and linear size in infancy is influenced prominently by IGF-1 that is regulated mainly by nutritional factors than by the GH.51 52 Thus, IGF-1 might be the hormonal link underlying the association between faster linear growth in infancy and adult adiposity given its metabolic functions (anabolism) and role in promoting linear growth.49 Studies of British cohorts have shown that infants (0–2 years) with a faster postnatal growth (catch-up growth) in weight or height, had higher levels of IGF-1 at age 5 years, independent of birth levels.53 IGF-1 levels in childhood were positively correlated with weight, height and body composition in childhood.53 Besides, children who are taller tend to be heavier, and rapid weight gains and catch-up growth in postnatal years have been associated with higher adiposity and risk of obesity in childhood and adult life.7 In turn, animal models and clinical trials have demonstrated that excess weight causes a downregulation of the GH/IGF-1 axis leading to higher adiposity in adulthood.54
Thus, increased growth in infancy, including a faster linear growth, might induce the long-lasting metabolic programming of the GH/IGF-1 axis and a subsequent increased risk of adiposity. However, the evidence on the role of linear growth on obesity risk and higher adiposity later in life is limited and inconclusive.7 For example, a recent study of a British cohort showed that infant linear growth (ages 0–3 months) modestly improved prediction models of childhood adiposity, but the estimates were small and with a negative trend.55 While, our findings showed that faster linear growth in infancy (0-24 months) has a weak but significant association with higher adult adiposity, a positive trend that contrasts with findings in children.
Overall, larger weight gains across the life course were associated with larger body size and adiposity in adult life among men and women. However, among women, estimates for size in adolescence tended to be stronger than those for childhood, in contrast to the men, in whom childhood estimated tended to be stronger. Regarding conditional height, a faster liner growth in infancy was positively but weakly associated with higher adult adiposity among both men and women, and faster linear growth in childhood was positively associated with higher adiposity among adult women but not in men.
Sex-based dimorphism in human growth patterns, body size and composition since the prenatal period, with pronounced differences in certain age intervals is well recognised.56 In infancy, growth patterns are similar among boys and girls, a period characterised by a deceleration of linear growth and fat accretion after the first year of birth.57 However, in childhood, after the adiposity rebound (around age 6 years), there is acceleration in linear growth and weight gain, accretion that is mainly due to gains in lean mass among boys and gains in FM among girls.57 In adolescence, women reach puberty earlier and growth stops at a younger age compared with men. Thus, men have a longer prepubertal period for growth and in adolescence they experience a greater accretion in lean mass than FM, whereas weight gain in women is characterised by a greater accretion in FM and their growth velocity is slower than men.50 57 These sex differences remain until adult life, where men and women on average have different body stature, composition, shape and fat distribution. These changes and sex differences in body size and composition are the result of hormonal regulations, mainly by the action of the GH/IGF-1 axis in infancy and childhood, plus the effect of steroid hormones in puberty.50 57 Thus, sex dimorphisms in growth patterns and body composition might explain the differences in the effect sizes and trends among men and women that we observed. Public health recommendations and interventions should pay special attention to these differences in growth and body composition among boys and girls mainly in childhood and adolescence, to prevent the excess of adiposity and high prevalence of obesity in adult life (higher among women), since early stages of life.
Some limitations of our analysis include: [1] Attrition as one source of selection bias given the long-lasting follow-up of these birth cohorts; [2] Residual confounding since we could not control for covariates such as diet, reproductive and life style factors; [3] *Missing data* of covariates (ranges between $0.1\%$–$41\%$), for which we assumed MAR and applied multiple imputation techniques; and [4] As expected there was substantial between-study heterogeneity, mostly in associations of conditional relative weight in childhood and adolescence. Some reasons that might explain the observed heterogeneity, beyond the baseline differences in places and contexts, include: (a) The age intervals definition was guided by data availability, as not all children were measured at the same intervals; (b) Differences in sample sizes across study sites given that we had to restrict the analysis to participants with complete anthropometric data in the four age intervals to be able to estimate conditional growth measures; (c) Differences in the techniques and instruments to measure body composition which might be a source of information bias (non-differential misclassification). In this regard, our research question intended to study the absolute increase in body composition not the relative estimate, and we assumed that all techniques, independent of the method and units used, ranked people in the same way. Additionally, we transformed all body composition measurements into SD units in each cohort and estimated standardised βs to be able to compare estimates between sites. To minimise the previous limitations, we used the random meta-analysis approach that not only considers the underlying heterogeneity when pooling estimates but also allows us to control for differences in sample sizes across study sites.
Among the strengths of our analyses, we highlight the prospective and long follow-up of these six birth cohorts in LMICs. We used large sample sizes (except Guatemala), and analysed two compartments of body composition (beyond BMI) that were measured using gold standards and/or validated techniques. We studied the independent association of linear growth and weight gain with body composition through the use of conditional growth measures, which allowed us to break the high correlation between weight and height and to study the independent role of linear growth and relative weight gain in specific age intervals. We selected the best model at each study site, fully adjusted by multiple confounders, and applied a random-effects meta-analysis that by default considers the heterogeneity between study sites.
We are cautious with the generalizability of our results since we used selected birth cohorts which design was not intended to be representative of countries were information was collected, and given the social, demographic, economic, political and cultural differences in contexts and periods in which each cohort started. However, the consistency in magnitude and direction of effect sizes across study sites (six birth cohorts from five different cities around the world), indicates that our findings are robust and suggests that the long-term role of growth on body composition might apply to individuals from other LMICs with similar characteristics.
In summary, our findings suggest that: [1] Growth in the first 1000 days of life has a small and long-lasting association with adult body size and composition; [2] Childhood and adolescence exhibit the strongest associations between relative weight gain with body size and composition in adulthood; and [3] A faster linear growth in infancy (girls and boys) and childhood (girls) predicts increased adiposity in adult life. Historical efforts to reduce growth faltering in LMICs have been successful, with little evidence to date of increases in BMI for age.30 58 59 However, once nutritional and other environmental constraints on growth are removed, further nutritional inputs beyond individual requirements are likely to increase adiposity. Thus, public health interventions might focus on promoting adequate linear growth in early life and a healthy weight gain throughout the life course.
## Data availability statement
COHORTS data are not freely available due to privacy considerations. Data are available upon reasonable request to the principal investigators at each study site.
## Patient consent for publication
Not applicable.
## Ethics approval
This study involves human participants and was approved by Brazil, Federal University of Pelotas, reference number 1.250.366; Guatemala, Institute of Nutrition for Central America and Panama, reference number CIE-REV-072-2017; India, Sitaram Bhartia Institute of Science and Research, reference numbers SBISR/$\frac{2012}{002}$, SBISR/RES$\frac{1}{3}$/2012, SBISR/IEC/$\frac{2014}{001}$, IEC/SBISR/$\frac{2015}{1}$ and FL/SBISR/IEC/2019–01; the Philippines, Research Ethics Committee at University of San Carlos, reference number $\frac{006}{2018}$-01-borja; South Africa, Human Research Ethics Committee at University of Witswatersrand, reference number M180225). Parents or guardians of participants granted their written or verbal consent to participate at baseline and in childhood/adolescence follow-ups, while in adulthood participants gave their own consent. The present analysis was approved by the Institutional Review Board of Emory University (IRB number 95960). Participants gave informed consent to participate in the study before taking part.
## References
1. NCD-RisC) NRFC. **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. DOI: 10.1016/S0140-6736(16)00618-8
2. NCD-RisC) NRFC. **Worldwide trends in blood pressure from 1975 to 2015: a pooled analysis of 1479 population-based measurement studies with 19·1 million participants**. *Lancet* (2017.0) **389** 37-55. DOI: 10.1016/S0140-6736(16)31919-5
3. NCD-RisC) NRFC. **Trends in adult body-mass index in 200 countries from 1975 to 2014: a pooled analysis of 1698 population-based measurement studies with 19·2 million participants**. *Lancet* (2016.0) **387** 1377-96. DOI: 10.1016/S0140-6736(16)30054-X
4. Ezzati M, Pearson-Stuttard J, Bennett JE. **Acting on non-communicable diseases in low- and middle-income tropical countries**. *Nature* (2018.0) **559** 507-16. DOI: 10.1038/s41586-018-0306-9
5. Barker DJ. **The developmental origins of well-being**. *Philos Trans R Soc Lond B Biol Sci* (2004.0) **359** 1359-66. DOI: 10.1098/rstb.2004.1518
6. Wells JCK, Chomtho S, Fewtrell MS. **Programming of body composition by early growth and nutrition**. *Proc Nutr Soc* (2007.0) **66** 423-34. DOI: 10.1017/S0029665107005691
7. Woo JG. **Infant growth and long-term cardiometabolic health: a review of recent findings**. *Curr Nutr Rep* (2019.0) **8** 29-41. DOI: 10.1007/s13668-019-0259-0
8. Umer A, Kelley GA, Cottrell LE. **Childhood obesity and adult cardiovascular disease risk factors: a systematic review with meta-analysis**. *BMC Public Health* (2017.0) **17**. DOI: 10.1186/s12889-017-4691-z
9. Lloyd LJ, Langley-Evans SC, McMullen S. **Childhood obesity and adult cardiovascular disease risk: a systematic review**. *Int J Obes (Lond)* (2010.0) **34** 18-28. DOI: 10.1038/ijo.2009.61
10. Lloyd LJ, Langley-Evans SC, McMullen S. **Childhood obesity and risk of the adult metabolic syndrome: a systematic review**. *Int J Obes (Lond)* (2012.0) **36** 1-11. DOI: 10.1038/ijo.2011.186
11. Singh AS, Mulder C, Twisk JWR. **Tracking of childhood overweight into adulthood: a systematic review of the literature**. *Obes Rev* (2008.0) **9** 474-88. DOI: 10.1111/j.1467-789X.2008.00475.x
12. Ward LC. **Human body composition: yesterday, today, and tomorrow**. *Eur J Clin Nutr* (2018.0) **72** 1201-7. DOI: 10.1038/s41430-018-0210-2
13. Adair LS, Fall CHD, Osmond C. **Associations of linear growth and relative weight gain during early life with adult health and human capital in countries of low and middle income: findings from five birth cohort studies**. *Lancet* (2013.0) **382** 525-34. DOI: 10.1016/S0140-6736(13)60103-8
14. Kuzawa CW, Hallal PC, Adair L. **Birth weight, postnatal weight gain, and adult body composition in five low and middle income countries**. *Am J Hum Biol* (2012.0) **24** 5-13. DOI: 10.1002/ajhb.21227
15. Weng SF, Redsell SA, Swift JA. **Systematic review and meta-analyses of risk factors for childhood overweight identifiable during infancy**. *Arch Dis Child* (2012.0) **97** 1019-26. DOI: 10.1136/archdischild-2012-302263
16. Rallis D, Balomenou F, Tzoufi M. **A systematic review indicates an association between birth weight and body fat in childhood**. *Acta Paediatr* (2021.0) **110** 2023-39. DOI: 10.1111/apa.15834
17. Yang Z, Huffman SL. **Nutrition in pregnancy and early childhood and associations with obesity in developing countries**. *Matern Child Nutr* (2013.0) **9 Suppl 1** 105-19. DOI: 10.1111/mcn.12010
18. Richter LM, Victora CG, Hallal PC. **Cohort profile: the Consortium of health-orientated research in transitioning societies**. *Int J Epidemiol* (2012.0) **41** 621-6. DOI: 10.1093/ije/dyq251
19. Horta BL, Gigante DP, Gonçalves H. **Cohort profile update: the 1982 pelotas (Brazil) birth cohort study**. *Int J Epidemiol* (2015.0) **44** 441. DOI: 10.1093/ije/dyv017
20. Victora CG, Barros FC. **Cohort profile: the 1982 pelotas (Brazil) birth cohort study**. *Int J Epidemiol* (2006.0) **35** 237-42. DOI: 10.1093/ije/dyi290
21. Victora CG, Hallal PC, Araújo CL. **Cohort profile: the 1993 pelotas (Brazil) birth cohort study**. *Int J Epidemiol* (2008.0) **37** 704-9. DOI: 10.1093/ije/dym177
22. Stein AD, Melgar P, Hoddinott J. **Cohort profile: the Institute of nutrition of central America and Panama (INCAP) nutrition trial cohort study**. *Int J Epidemiol* (2008.0) **37** 716-20. DOI: 10.1093/ije/dyn028
23. Bhargava SK, Sachdev HS, Fall CHD. **Relation of serial changes in childhood body-mass index to impaired glucose tolerance in young adulthood**. *N Engl J Med* (2004.0) **350** 865-75. DOI: 10.1056/NEJMoa035698
24. Adair LS, Popkin BM, Akin JS. **Cohort profile: the Cebu longitudinal health and nutrition survey**. *Int J Epidemiol* (2011.0) **40** 619-25. DOI: 10.1093/ije/dyq085
25. Richter L, Norris S, Pettifor J. **Cohort profile: Mandela’s children: the 1990 birth to twenty study in South Africa**. *Int J Epidemiol* (2007.0) **36** 504-11. DOI: 10.1093/ije/dym016
26. Mei Z, Grummer-Strawn LM. **Standard deviation of anthropometric Z-scores as a data quality assessment tool using the 2006 who growth standards: a cross country analysis**. *Bull World Health Organ* (2007.0) **85** 441-8. DOI: 10.2471/blt.06.034421
27. Jesson J, Schomaker M, Malasteste K. **Stunting and growth velocity of adolescents with perinatally acquired HIV: differential evolution for males and females. A multiregional analysis from the iedea global paediatric collaboration**. *J Int AIDS Soc* (2019.0) **22**. DOI: 10.1002/jia2.25412
28. Prioreschi A, Munthali RJ, Kagura J. **The associations between adult body composition and abdominal adiposity outcomes, and relative weight gain and linear growth from birth to age 22 in the birth to twenty plus cohort, South Africa**. *PLoS One* (2018.0) **13**. DOI: 10.1371/journal.pone.0190483
29. Gonçalves H, Wehrmeister FC, Assunção MCF. **Cohort profile update: the 1993 pelotas (Brazil) birth cohort follow-up at 22 years**. *Int J Epidemiol* (2018.0) **47** 1389-1390e. DOI: 10.1093/ije/dyx249
30. Kroker-Lobos MF, Ramirez-Zea M, Stein AD. **Overweight and obesity, cardiometabolic health, and body composition: findings from the follow-up studies of the INCAP longitudinal study**. *Food Nutr Bull* (2020.0) **41** S59-68. DOI: 10.1177/0379572120903222
31. Baptista Menezes AM, Oliveira PD, Wehrmeister FC. **Associations between growth from birth to 18 years, intelligence, and schooling in a Brazilian cohort**. *Am J Clin Nutr* (2020.0) **112** 187-94. DOI: 10.1093/ajcn/nqaa047
32. Li H, Stein AD, Barnhart HX. **Associations between prenatal and postnatal growth and adult body size and composition**. *Am J Clin Nutr* (2003.0) **77** 1498-505. DOI: 10.1093/ajcn/77.6.1498
33. Sachdev HS, Fall CHD, Osmond C. **Anthropometric indicators of body composition in young adults: relation to size at birth and serial measurements of body mass index in childhood in the new Delhi birth cohort**. *Am J Clin Nutr* (2005.0) **82** 456-66. DOI: 10.1093/ajcn.82.2.456
34. Adair LS. **Size at birth and growth trajectories to young adulthood**. *Am J Hum Biol* (2007.0) **19** 327-37. DOI: 10.1002/ajhb.20587
35. **The WHO child growth standards: world health organization**. (2016.0)
36. Li C, Ford ES, Zhao G. **Estimates of body composition with dual-energy X-ray absorptiometry in adults**. *Am J Clin Nutr* (2009.0) **90** 1457-65. DOI: 10.3945/ajcn.2009.28141
37. Harrer M, Cuijpers P, Furukawa TA. **Doing meta-analysis with R: A hands-on guide**
38. Judd SE, Ramirez-Zea M, Stein AD. **Relation of ratio indices of anthropometric measures to obesity in a stunted population**. *Am J Hum Biol* (2008.0) **20** 446-50. DOI: 10.1002/ajhb.20762
39. Wells JCK, Victora CG. **Indices of whole-body and central adiposity for evaluating the metabolic load of obesity**. *Int J Obes (Lond)* (2005.0) **29** 483-9. DOI: 10.1038/sj.ijo.0802899
40. Corvalán C, Gregory CO, Ramirez-Zea M. **Size at birth, infant, early and later childhood growth and adult body composition: a prospective study in a stunted population**. *Int J Epidemiol* (2007.0) **36** 550-7. DOI: 10.1093/ije/dym010
41. Savvidis C, Tournis S, Dede AD. **Obesity and bone metabolism**. *Hormones (Athens)* (2018.0) **17** 205-17. DOI: 10.1007/s42000-018-0018-4
42. Tomlinson DJ, Erskine RM, Morse CI. **The impact of obesity on skeletal muscle strength and structure through adolescence to old age**. *Biogerontology* (2016.0) **17** 467-83. DOI: 10.1007/s10522-015-9626-4
43. Guo SS, Huang C, Maynard LM. **Body mass index during childhood, adolescence and young adulthood in relation to adult overweight and adiposity: the fels longitudinal study**. *Int J Obes Relat Metab Disord* (2000.0) **24** 1628-35. DOI: 10.1038/sj.ijo.0801461
44. Araújo de França GV, De Lucia Rolfe E, Horta BL. **Associations of birth weight, linear growth and relative weight gain throughout life with abdominal fat depots in adulthood: the 1982 pelotas (Brazil) birth cohort study**. *Int J Obes (Lond)* (2016.0) **40** 14-21. DOI: 10.1038/ijo.2015.192
45. Larqué E, Labayen I, Flodmark C-E. **From conception to infancy-early risk factors for childhood obesity**. *Nat Rev Endocrinol* (2019.0) **15** 456-78. DOI: 10.1038/s41574-019-0219-1
46. Mazariegos M, Varghese JS, Kroker-Lobos MF. **Age at childbirth and change in BMI across the life-course: evidence from the INCAP longitudinal study**. *BMC Pregnancy Childbirth* (2022.0) **22**. DOI: 10.1186/s12884-022-04485-6
47. Popkin BM, Corvalan C, Grummer-Strawn LM. **Dynamics of the double burden of malnutrition and the changing nutrition reality**. *Lancet* (2020.0) **395** 65-74. DOI: 10.1016/S0140-6736(19)32497-3
48. Rivera JA, Martorell R, Ruel MT. **Nutritional supplementation during the preschool years influences body size and composition of Guatemalan adolescents**. *J Nutr* (1995.0) **125** 1068S-1077S. DOI: 10.1093/jn/125.suppl_4.1068S
49. Laron Z. **Insulin-Like growth factor 1 (IGF-1): a growth hormone**. *Mol Pathol* (2001.0) **54** 311-6. DOI: 10.1136/mp.54.5.311
50. Karlberg J. **A biologically-oriented mathematical model (ICP) for human growth**. *Acta Paediatr Scand Suppl* (1989.0) **350** 70-94. DOI: 10.1111/j.1651-2227.1989.tb11199.x
51. Fliesen T, Maiter D, Gerard G. **Reduction of serum insulin-like growth factor-I by dietary protein restriction is age dependent**. *Pediatr Res* (1989.0) **26** 415-9. DOI: 10.1203/00006450-198911000-00010
52. Ong KK, Elmlinger M, Jones R. **Growth hormone binding protein levels in children are associated with birth weight, postnatal weight gain, and insulin secretion**. *Metabolism* (2007.0) **56** 1412-7. DOI: 10.1016/j.metabol.2007.06.004
53. Ong K, Kratzsch J, Kiess W. **Circulating IGF-I levels in childhood are related to both current body composition and early postnatal growth rate**. *J Clin Endocrinol Metab* (2002.0) **87** 1041-4. DOI: 10.1210/jcem.87.3.8342
54. Berryman DE, Glad CAM, List EO. **The GH/IGF-1 axis in obesity: pathophysiology and therapeutic considerations**. *Nat Rev Endocrinol* (2013.0) **9** 346-56. DOI: 10.1038/nrendo.2013.64
55. Ong KK, Cheng TS, Olga L. **Which infancy growth parameters are associated with later adiposity? the Cambridge baby growth study**. *Ann Hum Biol* (2020.0) **47** 142-9. DOI: 10.1080/03014460.2020.1745887
56. **Who child growth standards based on length/height, weight and age**. *Acta Paediatr Suppl* (2006.0) **450** 76-85. DOI: 10.1111/j.1651-2227.2006.tb02378.x
57. Wells JCK. **Sexual dimorphism of body composition**. *Best Pract Res Clin Endocrinol Metab* (2007.0) **21** 415-30. DOI: 10.1016/j.beem.2007.04.007
58. Martorell R. **History and design of the INCAP longitudinal study (1969-1977) and its impact in early childhood**. *Food Nutr Bull* (2020.0) **41** S8-22. DOI: 10.1177/0379572120906062
59. Stein AD, Wang M, Ramirez-Zea M. **Exposure to a nutrition supplementation intervention in early childhood and risk factors for cardiovascular disease in adulthood: evidence from Guatemala**. *Am J Epidemiol* (2006.0) **164** 1160-70. DOI: 10.1093/aje/kwj328
|
---
title: Identification of mitochondria-related key gene and association with immune
cells infiltration in intervertebral disc degeneration
authors:
- Wei Guo
- Kun Mu
- Wen-Shuai Li
- Shun-Xing Gao
- Lin-Feng Wang
- Xiao-Ming Li
- Jian-Yong Zhao
journal: Frontiers in Genetics
year: 2023
pmcid: PMC10030706
doi: 10.3389/fgene.2023.1135767
license: CC BY 4.0
---
# Identification of mitochondria-related key gene and association with immune cells infiltration in intervertebral disc degeneration
## Abstract
Intervertebral disc (IVD) degeneration and its inflammatory microenvironment can result in discogenic pain, which has been shown to stem from the nucleus pulposus (NP). Increasing evidence suggests that mitochondrial related genes are strictly connected to cell functionality and, importantly, it can regulate cell immune activity in response to damaged associated signals. Therefore, identification of mitochondria related genes might offer new diagnostic markers and therapeutic targets for IVD degeneration. In this study, we identified key genes involved in NP tissue immune cell infiltration during IVD degeneration by bioinformatic analysis. The key modules were screened by weighted gene co-expression network analysis (WCGNA). *Characteristic* genes were identified by random forest analysis. *Then* gene set enrichment analysis (GSEA) was used to explore the signaling pathways associated with the signature genes. Subsequently, CIBERSORT was used to classify the infiltration of immune cells. Function of the hub gene was confirmed by PCR, Western blotting and ELISA. Finally, we identified MFN2 as a crucial molecule in the process of NP cell pyroptosis and NLRP3 inflammasome activation. We speculate that the increased MFN2 expression in NP tissue along with the infiltration of CD8+ T cells, NK cell and neutrophils play important roles in the pathogenesis of IVD degeneration.
## Introduction
Low back pain (LBP) is a major musculoskeletal disease with adversely affects in people of all ages and socioeconomic groups (Knezevic et al., 2021). It is the most common reason for medical consultation and the leading reason of disability worldwide (Knezevic et al., 2021). LBP can be caused by a variety of reasons, but intervertebral disc (IVD) degeneration has been indicated as the most important reason (Martin et al., 2008; Livshits et al., 2011). IVD is composed of nucleus pulposus (NP) located centrally, fibrous annulus in the periphery, and cartilaginous end plates that connect cranially and caudally (Colombier et al., 2014). The healthy NP tissue serves as the hydrogel-like core of the IVD, consisting mainly of NP cells and extracellular matrix (ECM) (Lyu et al., 2019). It is generally believed that NP cells are of great important for IVD function and ECM metabolism, which first exhibits degenerative changes during IVD degeneration (Lan et al., 2021). Therefore, a further study of the molecular mechanism of NP cell loss may provide a novel therapeutic target for improving the IVD degeneration.
Mitochondria is a closed, double-membrane organelle which is found in nearly all eukaryotes. *Mitochondria* generate adenosine triphosphate mainly via oxidative phosphorylation of substances. Mitochondrial membrane potential and intimal permeability are the main characteristics of healthy mitochondria. In order to maintain oxidative phosphorylation, various carbon substrates need to be metabolized through certain pathways, and finally recycle and polymerize with tricarboxylic acid (TCA) to produce reductive equivalent. Mitochondria contain an active calcium ion transport system that activates many enzymes associated with oxidative metabolic pathways. Previous study has linked mitochondria to cell damage and a wide range of age-related diseases (Giorgi et al., 2018). Cell death induced by mitochondria is an important mechanism leading to IVD degeneration (Sun et al., 2021). It has been confirmed that disruption of mitochondrial dynamics is also closely related to mitochondrial dysfunction and oxidative stress during IVD degeneration. Xu et al. [ 2019] found that the accumulation of progerin in human NP tissues was correlated with IVD degeneration progression, and further studies confirmed that progerin stimulation could shift mitochondrial dynamics to fission events by reducing the levels of mitochondrial fusion factors Opa1 and increasing the levels of mitochondrial fission factor Drp1 (Song et al., 2021). NP cell function loss occurs in the IVD degeneration progression stage due to cell senescence, death, inflammatory responses, and imbalances in anabolic and catabolic metabolism, which are also closely associated with mitochondrial damage (Wang et al., 2020a; Huang et al., 2020). Most studies on mitochondria-related genes focus on their regulation of mitochondrial function (Prasun, 2020; Sun et al., 2021). However, increasing evidence suggests that mitochondrial related genes are strictly connected to cell functionality and, importantly, it can regulate cell immune activity in response to damaged associated signals (Zhang et al., 2021; Deng et al., 2022). IVD was considered as an immunologically privileged organ that isolated NP tissue from the host immune system due to their unique structure. When the blood-NP barrier is breached, NP tissue triggers an immune response. This process plays an important role in IVD degeneration and result in a variety of subsequent pathological processes (Bridgen et al., 2017). Therefore, the relationship of mitochondrial related genes with NP cell inflammatory response in the process of IVD degeneration has become a research hotspot.
In this study, bioinformatics analysis based on transcriptome microarrays was used to identify mitochondria-associated genes in patients with IVD degeneration, providing new insights into the diagnosis and treatment of the disease. Moreover, we investigated the effects of Mitofusin-2 (MFN2) overexpression on human NP cells pyroptosis and NLRP3 inflammasome activation.
## Ethics statement
This study was approved by the ethics committees of Hebei Province Cangzhou Hospital of Integrated Traditional Chinese Medicine-Western Medicine [2020022]. Human NP tissue samples were obtained from patients undergoing surgery at Hebei Province Cangzhou Hospital of Integrated Traditional Chinese Medicine-Western Medicine. Written informed consent was obtained from all patients for the use of their tissue specimens for research purpose.
The studies involving human participants were reviewed and approved by Ethics committees of Hebei Province Cangzhou Hospital of Integrated Traditional Chinese Medicine-Western Medicine. The patients/participants provided their written informed consent to participate in this study.
## Clinical specimens
Human lumbar degenerative NP specimens were obtained from 10 patients with IVD degeneration undergoing discectomy. The control samples were taken from 10 age- and sex-matched patients with fresh traumatic vertebral fracture undergoing decompressive surgery because of neurological deficits.
## Data sources
In this study, two data sets were downloaded from the Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/), namely, GSE56081 [platform: GPL15314 Arraystar Human LncRNA microarray V2.0 (Agilent-033010 probe name version) and GSE165722 (GPL24676 Illumina NovaSeq) 6,000 (Homo sapiens)]. GSE56081 contained two groups of mRNA expression profiles, containing 5 NP tissue from IVD degeneration patients and five from controls. GSE165722 contained eight samples of Single-cell transcriptome profiling from NP tissue. A total of 1,136 mitochondria-related genes (Mito-Genes) were downloaded from the Mitocarta3.0 database (Rath et al., 2021).
## Weighted gene coexpression network analysis (WGCNA) was used to identify IVD degeneration genes
For this study, a R package named “WGCNA” was used to construct the co-expressed clusters by analysis the gene expression level of GSE56081 (Langfelder and Horvath, 2008). Prior to the network construction, the raw data was converted into a recognizable format using R package “affy,” then the missing values were estimated using k nearest neighbor based approach, and then the median method was used for normalization (Troyanskaya et al., 2001). The pickSoftThreshold function of WGCNA package is used to calculate the soft threshold power and adjacency relationship. Then, the adjacency matrix is converted into topological overlap matrix, and the corresponding dissimilarity is calculated to carry out hierarchical clustering analysis. The dynamic tree cutting method with a minimum module size of 20 were used to identify the co-expressed gene modules. The module associated with IVD degeneration was identified by correlation analysis. IVD degeneration genes were acquired by module membership within the modules.
## Identification of IVD degeneration-mitochondria related genes
The IVD degeneration-mitochondria related genes (IVDD-Mito_Genes) were obtained by intersecting the IVD degeneration genes and Mito-Genes. We then used R package clusterProfiler to analyze the GO and KEGG signaling pathways of IVDD-Mito_Genes and output the top 10 GO and 15 KEGG signaling pathways.
## Signature genes identification
Subsequently, we exerted machine learning algorithms called random forest to screen IVDD-Mito_Genes. We used R packet “randomforest” to classify IVDD-Mito_Genes. The random forest model calculated the average error rate of IVDD-Mito_Genes to determine the optimal number of variables (Xie et al., 2022; Yu et al., 2022). We then calculate the error rate for each tree and determine the optimal number of trees based on the lowest error rate. After the above parameters are determined, the random forest tree model is established. Finally, the characteristic importance score of each IVDD-Mito_Genes was performed, and the top10 genes were selected. We used the area under the curve (AUC) of the receiver operating characteristic curve (ROC) to evaluate the diagnostic efficiency of these signature genes. An AUC greater than 0.8 indicates a good diagnostic effect.
## Gene set enrichment analysis
We grouped the IVD degeneration cohort according to the median expression of signature genes, and performed gene set enrichment analysis (GSEA) according to different subgroups to determine the relationship between signature genes and signaling pathways (Subramanian et al., 2005).
We analysis signaling pathways and GO associated with signature genes which had diagnostic efficiency by GSEA analysis. Figures 5A–C demonstrated the top six signal pathways. The results showed that MRPS17 was markedly associated with adherens junction, ether lipid metabolism, biosynthesis keratan sulfate, hematopoietic cell lineage, protein export, RNA degradation (Figure 5A). The expression of MFN2 significantly correlated with ether lipid metabolism, biosynthesis keratan sulfate, hematopoietic cell lineage, olfactory transduction, protein export and RNA degradation (Figure 5B). The expression of PDF significantly correlated with allograft rejection, autoimmune thyroid disease, graft versus host disease, propanoate metabolism and protein export (Figure 5C). The top six GO were demonstrated in Figures 5D–F. The results showed that MRPS17 was significantly correlated with myoblast fusion, positive regulation of monooxygenase activity, regulation of B cell mediated immunity, regulation of immunoglobulin production, regulation of muscle organ development and cytokine activity (Figure 5D). The expression of MFN2 significantly correlated with myoblast fusion, positive regulation of monooxygenase activity, regulation of B cell mediated immunity, regulation of immunoglobulin production, regulation of muscle organ development and cytokine activity (Figure 5E). The expression of PDF significantly correlated with kinetochore organization, neutral amino acid transport, pigment biosynthetic process, positive regulation of endothelial cell proliferation, regulation of extrinsic apoptotic signaling pathway via death domain receptors (Figure 5F).
**FIGURE 5:** *The GSEA of the signature genes in IVD degeneration. (A–C) KEGG enrichment results of signature genes (MRPS17, MFN2, PDF). (D–F) GO enrichment results of signature genes (MRPS17, MFN2, PDF).*
## Immune cell infiltration analysis
We used the principle of linear support vector regression to deconvolution the expression matrix of 22 human immune cell subtypes by CIBERSORT method to investigate the differences of immune cell between IVD degeneration patients and normal subjects. Subsequently, immune cells with significant differences in infiltration between patients with IVD degeneration and normal subjects were screened, and spearman correlation analysis was used to identify their correlation with signature genes.
## Signal cell RNA-seq data analysis
Raw RNA-seq data (GSE165722) were processed according to bioinformatics analysis principles. Cells with a unique feature count of <200 and a mitochondrial count of >$20\%$ in the sample were filtered out. t-distributed stochastic neighbor embedding (t-SNE) analysis, K-mean clustering and hierarchical clustering methods were used to analyze the data. The following R packages were used: limma, Seurat, dplyr, magrittr, celldex, SingleR, monocle.
## Cell communication analysis
Cell communication is determined by assessing the expression of ligand and receptor pairs within a cell population (CellChat R package). Interactions between different cell types were examined, and gene expression 0.2 was set as an effective cut-off point.
## Human NP cell culture
The NP tissue samples were separated and cut into 1 m pieces, and the nucleus pulposum tissue was digested with $0.25\%$ pronase and $0.2\%$ collagenase type II at 37°C. The digested suspension was filtered through a 70 μm pore mesh and cultured in DMEM medium containing $10\%$ fetal bovine serum and $1\%$ penicillin-streptomycin at a CO2 concentration of $5\%$ and a temperature of 37°C.
## Western blotting
Cells were lysed with a buffer containing a protease inhibitor, 0.25 M Tris-HCl, $20\%$ glycerol, $4\%$ sodium dodecyl sulfate (SDS), and $10\%$ mercaptoethanol (pH 6.8). The same amount of total protein (10 μg) was separated by $10\%$–$12\%$ SDS-polyacrylamide gel and electrotransferred to polyvinylidene fluoride membrane. Then, $5\%$ non-fat milk in Tris-buffered saline containing $0.1\%$ Tween-20 (TBST) was employed for blocking at ambient (1 h), followed by incubation with primary antibodies in TBST containing $5\%$ non-fat milk overnight at 4°C. Secondary antibody was added at room temperature for 1 h and Western blotting was performed using an enhanced chemiluminescence system.
## ELISA
The contents of human IL-1β in the cell culture supernatant were detected according to the manufacturer’s instructions.
## Quantitative reverse-transcription PCR (RT-qPCR)
After chloroform extraction and precipitation, the DNA in the sample was removed by DNase I treatment, followed by reverse transcription of purified RNA using RevertAid reverse transcriptase. The reverse transcription primers containing CMV promoter sequences were designed to target specific genes. RT-qPCR was performed with AriaMx Real-time PCR system and QuantStudio Real-time PCR system. Supplementary Table S3 showed the primers used in this study.
## Cell transfection
Lipofectamine 3,000 (Invitrogen) was used to transfect the plasmids or siRNA respectively with third-generation NP cells as recommended by the manufacturer and described previously (Guo et al., 2020). Cells were collected 48 h after transfection.
## Statistical analysis
All experiments were repeated three times or more. Continuous data are mean ± standard deviation (SD). Bioinformatic analyses in this study was performed by R software (version 4.1.2). Prism 7.0 (GraphPad Software, United States) and SPSS 22.0 (SPSS Inc., United States) were used for statistical analyses. * $p \leq 0.05$, **$p \leq 0.01$, and ***$p \leq 0.001$ were significance levels.
## Identification of IVD degeneration genes by WGCNA analysis
We use dynamic shear tree algorithm to segment the modules, and set the minimum size of the modules as 20 to build a scale-free co-representation network. Figure 1A shows the results of the cluster tree. Finally, the data were clustered into 19 modules (Figure 1B). Then the correlation between each module and IVD degradation is calculated. Supplementary Table S1 shows the genes in each module. Correlation analysis showed that blue module was significantly associated with IVD degradation. Therefore, genes in blue module was considered as IVD degradation genes. The crossover between the IVD degeneration genes and mitochondrial related genes was shown in Figure 1C (Supplementary Table S2).
**FIGURE 1:** *The gene modules significantly associated with IVD degeneration were identified by WGCNA. (A) WGCNA cluster dendrogram. (B) WGCNA cluster modules. (C) Venn diagram shows the interaction between mitochondria-associated genes and genes associated with IVD degradation.*
## Function enrichment analysis
By GO and KEGG enrichment analysis of 58 IVD degeneration-Mito_genes, 81 biological processes (BP), 50 cellular components (CCs), 32 molecular functions (mf) and 20 KEGG signaling pathways were identified. The enrichment analysis results of the top 10 GO and 10 KEGG signaling pathways are shown in Figures 2A, B, respectively. The results showed that these genes are enriched in biological processes such as protein targeting to mitochondrial, mitochondrial transport, mitochondrial protein localization, protein transmembrane transport, and KEGG signaling pathways such as those in Parkinson’s disease, and oxidative phosphorylation.
**FIGURE 2:** *Functional enrichment analysis of IVD degeneration-Mito_Genes. (A) Top 10 BP, MF, CC of GO functional enrichment analysis. (B) The KEGG analysis of degeneration-Mito_Genes.*
## Identification of signature genes by random forest algorithms
Eight signature genes with relative importance greater than 0.3 were identified by random forest analysis (Figures 3A, B), containing MRPS17, ECHDC2, MFN2, EFHD1, PDF, PXMP2, TOMM7, PANK2.
**FIGURE 3:** *The machine algorithms for signature genes. (A) Confidence intervals for error rates of random forest models. (B) The relative importance of genes in random forest models.*
## Diagnostic efficacy of signature genes in predicting IVD degeneration
Figures 4A–H showed that four of the signature genes screened through random forest were more highly expressed in patients with IVD degeneration than normal people, including MRPS17, MFN2, PDF and PANK2, suggesting that these genes may play a key role in IVD degeneration. Figures 4I–P shows the area under curve (AUC) of these signature genes. The results showed that three of the signature genes (MRPS17, MFN2, PDF) had significant diagnostic effect in predicting the degeneration of IVD. These results were further confirmed in 10 patients and 10 controls human NP samples using RT-qPCR analysis and Western blotting, MRPS17, MFN2, PDF were significantly increased in IVD degeneration NP tissues, MFN2 showed significant diagnostic effect in predicting IVD degeneration (Supplementary Figure S1).
**FIGURE 4:** *The performance of the signature genes. (A–H) Signature gene expression levels in the nucleus pulposus tissue of IVD degeneration patient and healthy subjects. (I–P) The diagnostic performance of signature genes was evaluated by ROC.*
## Immune cell infiltration
Compared with normal subjects, IVD degeneration patients have higher CD8+ T cells, NK cell activated, neutrophils infiltration and lower dendritic cell resting (Figure 6A). MFN2 was associated with the infiltration of CD8+ T cells, NK cell activated and neutrophils positively. PDF was associated with the infiltration of CD8+ T cells positively. MRPS17 was positively associated with the infiltration of CD8+ T cells and neutrophils (Figure 6B). Notably, MFN2 was positively correlated with all the IVD degeneration related immune cell infiltration. These results demonstrated that MFN2 might be a potential immune-related target, and further study of MFN2 might provide a better understanding of immune cell infiltration in IVD degeneration.
**FIGURE 6:** *Signature genes related to immune cell infiltration. (A) The infiltration of immune cells in the NP tissues of the IVD degeneration and healthy subjects. (B) The relationship between signature genes and immune cell infiltration.*
## Single-cell transcriptome profiling of the IVD tissue
We used scRNA-seq data (GSE165722) to analyze and identify different cell types in NP tissues during IVD degeneration, and visualized the results with dimensionality reduction algorithm (t-SNE). We identified 21 cluster and 11 major distinct cell types which were determined using singleR and cell markers (Figures 7A, B). Consistent with above results, we found a lot of immune cells in the tSNE map which may suggest the immune cell infiltration during the progress of IVD degeneration. Subsequently, we examined MFN2 expression in these cell types. As expected, MFN2 was expressed in NP cells and most immune cells (Figures 7C, D).
**FIGURE 7:** *Single cell sequencing data analysis and identification of different cell types. (A, B) t-SNE projections and cell type annotation of IVD tissue. (C, D) t-SNE and violin diagrams demonstrated MFN2 expression pattern at the single-cell level.*
We used CellChat analysis to identified potential interactions between different types of immune cells and NP cells. The results showed that NP cells interact with neutrophils, myelocyte, monocyte, macrophage, endothelial cells closely (Figures 8A–K).
**FIGURE 8:** *CellChat analysis. (A–K) Circle plots demonstrated the interactions in the cell-cell communication network between IVD degeneration and healthy subjects, respectively.*
## Mfn2 overexpression activated NLRP3 inflammasome
IL-1β secretion following exposure to MSU (a prototypical danger signals) was significantly decreased in MFN2 knockdown NP cells as compared to wild-type NP cells (Figure 9A). As shown in Figures 9B, C, MSU increased MFN2 mRNA and protein expression level in comparison with control group. Pyrosis is a programmed inflammatory cell death that occurs when inflammasome is activated. In this study, we found that the absence of MFN2 markedly attenuated the release of LDH from NP cells, suggesting that MFN2 had a regulatory effect on pyroptosis (Figure 9D). To elucidate the effect of MFN2 in the activation of NLRP3 inflammasome, we examined IL-1β cleavage and caspase-1 activation in MFN2 overexpressed NP cells. The results showed that the activation of Caspase-1 and the cleavage of IL-1β were significantly increased in MFN2 overexpressed NP cells (Figure 9E). Notably, MFN2 overexpression significantly increased NLRP3 protein and mRNA expression level, but had no effect on pro-IL-1β and pro-caspase-1 expression (Figures 9E, F). And NLRP3 knockdown markedly inhibited IL-1β secretion in MFN2 overexpression NP cells (Figure 9G). These results indicated that MFN2 can promote the expression of NLPR3 and activate the NLRP3 inflammasome, resulting in the pyroptosis of NP cells and the release of inflammatory cytokines.
**FIGURE 9:** *Overexpression of MFN2 activated NLRP3 inflammasome in NP cells. (A) NP cells were transfected as indicated. IL-1β released was assessed by ELISA. (B, C) NP cells were stimulated with MSU. MFN2 mRNA expression was assessed by RT-qPCR and Western blot. (D) Pyroptosis was determined by LDH assay. (E) The Western blot assay for NLRP3, IL-1β cleavage and caspase-1 activation. (F) NP cells were transfected as indicated. NLRP3, pro- IL-1β and pro-caspase-1 mRNA were assessed by RT-qPCR. (G) NP cells were transfected as indicated. IL-1β released was assessed by ELISA.*
## Discussion
MFN2 is a multifunctional protein involved in a variety of physiological and pathological processes, including mitochondrial fusion transport, cell metabolism, pyroptssis, and autophagy (Yu et al., 2018). Dysregulation of MFN2 can lead to neurodegeneration, sarcopenia, metabolic disease, heart disease and many other diseases (Sebastian et al., 2016; Kim et al., 2017; Chandhok et al., 2018; Rocha et al., 2018). Increasing evidence have demonstrated that immune cell infiltration acts as an important role in IVD degeneration (Wang et al., 2021). This study is the first time to reveal the relationship between MFN2 and immune cell infiltration during intervertebral disc degeneration.
Previous studies have shown that MFN2 can be phosphorylated by Jun N-terminal kinase under cellular stress, leading to MFN2 degradation through the ubiquitin-proteasome system (Leboucher et al., 2012). The ubiquitination of MFN2 promotes mitochondrial degradation and prevents fusion of damaged mitochondria (Tanaka et al., 2010). In this study, our results indicated that MFN2 has the ability to regulate immune response in addition to the function of regulating mitochondrial fusion. Previously research demonstrated that IVD degeneration is characterized by infiltration of CD68+ macrophages, T cells (CD4+, CD8+), and neutrophils (Wang et al., 2021). In combination with scRNA-seq, cell heterogeneity can be identified for in-depth study of biological structure and function. The scRNA-seq technique can generate the expression profile of individual cells for the analysis of heterogeneous cell populations and the identification of cell types. Understanding the phenotype of immune cells in the IVD microenvironment is critical to understanding the mechanisms of IVD degeneration progression. Immune infiltrating cells in the IVD degeneration have been shown to be very important in the IVD degeneration (Risbud and Shapiro, 2014; Wang et al., 2021). In this research, we found significantly increased infiltration of CD8+ T cells and neutrophils in degenerative NP tissues compared to controls. We also found a strong positive correlation between the MFN2 and the level of immune infiltration of three types of invasive immune cells (including CD8+ T cells, NK cell and neutrophils). This implied a strong association of MFN2 with immune cells in the IVD degeneration, which was subsequently validated using scRNA-seq datasets. CellChat can quantitatively infer and analyze intercellular communication networks from scRNA-seq data. Our results showed the number of interactions between NP cells and immune cells in the IVD degeneration including neutrophils, myelocyte, monocyte, macrophage. IVD is considered as an immunologically privileged organ because their unique structure insulates the NP tissue from the host immune system (Sun et al., 2020). When NP cells pyroptosis, NP releases inflammatory factors such as IL-1β, which induces immune cell infiltration and triggers an immune response. This process act as an important role in IVD degeneration and result in multiple pathological processes that eventually lead to fibrotic of NP tissue (Wang et al., 2020b). At the same time, immune cells recruited into the IVD microenvironment will release more inflammatory mediators, further damaging the IVD microenvironment and causing nucleus pulposus cell death (Silva et al., 2019).
NLRP3 is a member of the pattern recognition receptors that mediate the activation of the innate immune system. The NLRP3 inflammasome is important for the immune defense system. NLRP3 initiates inflammasome assembly, promotes recruitment of inflammasome complex by caspase-1 and activates caspase-1. Activated caspase-1 cleave the IL-1β precursor protein and converts it into a biologically active mature form, which promotes an immune response. Previous studies have shown that MFN2 acts as a major regulator of the immune response by binding to NLRP3 and promoting IL-1β secretion after infection with the virus (Ichinohe et al., 2013; Tur et al., 2020). In our research, we observed that MFN2 was highly expressed during IVD degeneration. MFN2 overexpression associated with the infiltration of CD8+ T cells, NK cell activated and neutrophils. MFN2 overexpression also activate NLRP3 inflammasome which results in IL-1β release and NP cell pyroptosis. This may be due to the overexpression of MFN2 in NP cells inducing the expression of NLRP3 and the activation of inflammasome. Combining the results of this study with the conclusions of previous studies, we hypothesized that a special type of immune microenvironment is generated during the process of IVD degeneration by release of IL-1β from NP cells that recruit CD8+ T cells, NK cell activated and neutrophils. This immune microenvironment, in turn, further promotes NP cell death and IVD pathological changes.
This study also has some limitations. First of all, the data in this study are mainly from public databases and involve a small number of samples, which may lead to certain bias. However, the reliability of our analysis was confirmed by the results of in vitro experiments. Second, this study still needs a larger clinical samples to further validate the results. Third, the specific mechanism of MFN2 cause NLRP3 induction and increase inflammasome activation remains to be explored.
Our research shows that immune cell infiltration, including T cells, NK cell activated and neutrophils, is participated in the pathological process of IVD degeneration. Via microarray data analysis, single-cell sequencing data analysis and in vitro experiments, we identified MFN2 as a signature gene, which present good diagnostic value and may serve as a molecular target for treatment of IVD degeneration.
## 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
WG designed the study and analysis. Experiments were performed by WG and W-SL. WG and KM analyzed data and typeset figures. The study was supervised by L-FW, X-ML, and J-YZ. WG wrote the manuscript with contributions from KM, W-SL, L-FW, X-ML, and J-YZ. All of the authors subsequently reviewed and edited 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/fgene.2023.1135767/full#supplementary-material
## References
1. Bridgen D. T., Fearing B. V., Jing L., Sanchez-Adams J., Cohan M. C., Guilak F.. **Regulation of human nucleus pulposus cells by peptide-coupled substrates**. *Acta Biomater.* (2017) **55** 100-108. DOI: 10.1016/j.actbio.2017.04.019
2. Chandhok G., Lazarou M., Neumann B.. **Structure, function, and regulation of mitofusin-2 in health and disease**. *Biol. Rev. Camb Philos. Soc.* (2018) **93** 933-949. DOI: 10.1111/brv.12378
3. Colombier P., Clouet J., Hamel O., Lescaudron L., Guicheux J.. **The lumbar intervertebral disc: From embryonic development to degeneration**. *Jt. Bone Spine* (2014) **81** 125-129. DOI: 10.1016/j.jbspin.2013.07.012
4. Deng H., Zhu S., Zhu L., Sun J., Ding Y., Li F.. **Mfn2 is responsible for inhibition of the RIG-I/IRF7 pathway and activation of NLRP3 inflammasome in Seneca Valley virus-infected PK-15 cells to promote viral replication**. *Front. Immunol.* (2022) **13** 955671. DOI: 10.3389/fimmu.2022.955671
5. Giorgi C., Marchi S., Simoes I. C. M., Ren Z., Morciano G., Perrone M.. **Mitochondria and reactive oxygen species in aging and age-related diseases**. *Int. Rev. Cell Mol. Biol.* (2018) **340** 209-344. DOI: 10.1016/bs.ircmb.2018.05.006
6. Guo W., Mu K., Zhang B., Sun C., Zhao L., Li H. R.. **The circular RNA circ-GRB10 participates in the molecular circuitry inhibiting human intervertebral disc degeneration**. *Cell Death Dis.* (2020) **11** 612. DOI: 10.1038/s41419-020-02882-3
7. Huang D., Peng Y., Li Z., Chen S., Deng X., Shao Z.. **Compression-induced senescence of nucleus pulposus cells by promoting mitophagy activation via the PINK1/PARKIN pathway**. *J. Cell Mol. Med.* (2020) **24** 5850-5864. DOI: 10.1111/jcmm.15256
8. Ichinohe T., Yamazaki T., Koshiba T., Yanagi Y.. **Mitochondrial protein mitofusin 2 is required for NLRP3 inflammasome activation after RNA virus infection**. *Proc. Natl. Acad. Sci. U. S. A.* (2013) **110** 17963-17968. DOI: 10.1073/pnas.1312571110
9. Kim Y. J., Park J. K., Kang W. S., Kim S. K., Han C., Na H. R.. **Association between mitofusin 2 gene polymorphisms and late-onset alzheimer's disease in the Korean population**. *Psychiatry Investig.* (2017) **14** 81-85. DOI: 10.4306/pi.2017.14.1.81
10. Knezevic N. N., Candido K. D., Vlaeyen J. W. S., Van Zundert J., Cohen S. P.. **Low back pain**. *Lancet* (2021) **398** 78-92. DOI: 10.1016/S0140-6736(21)00733-9
11. Lan T., Shiyu H., Shen Z., Yan B., Chen J.. **New insights into the interplay between miRNAs and autophagy in the aging of intervertebral discs**. *Ageing Res. Rev.* (2021) **65** 101227. DOI: 10.1016/j.arr.2020.101227
12. Langfelder P., Horvath S.. **WGCNA: an R package for weighted correlation network analysis**. *BMC Bioinforma.* (2008) **9** 559. DOI: 10.1186/1471-2105-9-559
13. Leboucher G. P., Tsai Y. C., Yang M., Shaw K. C., Zhou M., Veenstra T. D.. **Stress-induced phosphorylation and proteasomal degradation of mitofusin 2 facilitates mitochondrial fragmentation and apoptosis**. *Mol. Cell* (2012) **47** 547-557. DOI: 10.1016/j.molcel.2012.05.041
14. Livshits G., Popham M., Malkin I., Sambrook P. N., Macgregor A. J., Spector T.. **Lumbar disc degeneration and genetic factors are the main risk factors for low back pain in women: The UK twin spine study**. *Ann. rheumatic Dis.* (2011) **70** 1740-1745. DOI: 10.1136/ard.2010.137836
15. Lyu F. J., Cheung K. M., Zheng Z., Wang H., Sakai D., Leung V. Y.. **IVD progenitor cells: A new horizon for understanding disc homeostasis and repair**. *Nat. Rev. Rheumatol.* (2019) **15** 102-112. DOI: 10.1038/s41584-018-0154-x
16. Martin B. I., Deyo R. A., Mirza S. K., Turner J. A., Comstock B. A., Hollingworth W.. **Expenditures and health status among adults with back and neck problems**. *JAMA* (2008) **299** 656-664. DOI: 10.1001/jama.299.6.656
17. Prasun P.. **Mitochondrial dysfunction in metabolic syndrome**. *Biochim. Biophys. Acta Mol. Basis Dis.* (2020) **1866** 165838. DOI: 10.1016/j.bbadis.2020.165838
18. Rath S., Sharma R., Gupta R., Ast T., Chan C., Durham T. J.. **MitoCarta3.0: An updated mitochondrial proteome now with sub-organelle localization and pathway annotations**. *Nucleic Acids Res.* (2021) **49** D1541-D1547. DOI: 10.1093/nar/gkaa1011
19. Risbud M. V., Shapiro I. M.. **Role of cytokines in intervertebral disc degeneration: Pain and disc content**. *Nat. Rev. Rheumatol.* (2014) **10** 44-56. DOI: 10.1038/nrrheum.2013.160
20. Rocha A. G., Franco A., Krezel A. M., Rumsey J. M., Alberti J. M., Knight W. C.. **MFN2 agonists reverse mitochondrial defects in preclinical models of Charcot-Marie-Tooth disease type 2A**. *Science* (2018) **360** 336-341. DOI: 10.1126/science.aao1785
21. Sebastian D., Sorianello E., Segales J., Irazoki A., Ruiz-Bonilla V., Sala D.. **Mfn2 deficiency links age-related sarcopenia and impaired autophagy to activation of an adaptive mitophagy pathway**. *EMBO J.* (2016) **35** 1677-1693. DOI: 10.15252/embj.201593084
22. Silva A. J., Ferreira J. R., Cunha C., Corte-Real J. V., Bessa-Goncalves M., Barbosa M. A.. **Macrophages down-regulate gene expression of intervertebral disc degenerative markers under a pro-inflammatory microenvironment**. *Front. Immunol.* (2019) **10** 1508. DOI: 10.3389/fimmu.2019.01508
23. Song Y., Lu S., Geng W., Feng X., Luo R., Li G.. **Mitochondrial quality control in intervertebral disc degeneration**. *Exp. Mol. Med.* (2021) **53** 1124-1133. DOI: 10.1038/s12276-021-00650-7
24. Subramanian A., Tamayo P., Mootha V. K., Mukherjee S., Ebert B. L., Gillette M. A.. **Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles**. *Proc. Natl. Acad. Sci. U. S. A.* (2005) **102** 15545-15550. DOI: 10.1073/pnas.0506580102
25. Sun K., Jing X., Guo J., Yao X., Guo F.. **Mitophagy in degenerative joint diseases**. *Autophagy* (2021) **17** 2082-2092. DOI: 10.1080/15548627.2020.1822097
26. Sun Z., Liu B., Luo Z. J.. **The immune privilege of the intervertebral disc: Implications for intervertebral disc degeneration treatment**. *Int. J. Med. Sci.* (2020) **17** 685-692. DOI: 10.7150/ijms.42238
27. Tanaka A., Cleland M. M., Xu S., Narendra D. P., Suen D. F., Karbowski M.. **Proteasome and p97 mediate mitophagy and degradation of mitofusins induced by Parkin**. *J. Cell Biol.* (2010) **191** 1367-1380. DOI: 10.1083/jcb.201007013
28. Troyanskaya O., Cantor M., Sherlock G., Brown P., Hastie T., Tibshirani R.. **Missing value estimation methods for DNA microarrays**. *Bioinforma. Oxf. Engl.* (2001) **17** 520-525. DOI: 10.1093/bioinformatics/17.6.520
29. Tur J., Pereira-Lopes S., Vico T., Marin E. A., Munoz J. P., Hernandez-Alvarez M.. **Mitofusin 2 in macrophages links mitochondrial ROS production, cytokine release, phagocytosis, autophagy, and bactericidal activity**. *Cell Rep.* (2020) **32** 108079. DOI: 10.1016/j.celrep.2020.108079
30. Wang L., He T., Liu J., Tai J., Wang B., Zhang L.. **Revealing the immune infiltration landscape and identifying diagnostic biomarkers for lumbar disc herniation**. *Front. Immunol.* (2021) **12** 666355. DOI: 10.3389/fimmu.2021.666355
31. Wang Y., Che M., Xin J., Zheng Z., Li J., Zhang S.. **The role of IL-1β and TNF-α in intervertebral disc degeneration**. *Biomed. Pharmacother.* (2020) **131** 110660. DOI: 10.1016/j.biopha.2020.110660
32. Wang Y., Wang H., Zhuo Y., Hu Y., Zhang Z., Ye J.. **SIRT1 alleviates high-magnitude compression-induced senescence in nucleus pulposus cells via PINK1-dependent mitophagy**. *Aging (Albany NY)* (2020) **12** 16126-16141. DOI: 10.18632/aging.103587
33. Xie X., Zhang Y., Yu J., Jiang F., Wu C.. **Significance of m(6)A regulatory factor in gene expression and immune function of osteoarthritis**. *Front. Physiol.* (2022) **13** 918270. DOI: 10.3389/fphys.2022.918270
34. Xu X., Wang D., Zheng C., Gao B., Fan J., Cheng P.. **Progerin accumulation in nucleus pulposus cells impairs mitochondrial function and induces intervertebral disc degeneration and therapeutic effects of sulforaphane**. *Theranostics* (2019) **9** 2252-2267. DOI: 10.7150/thno.30658
35. Yu F., Xu T., Wang M., Chang W., Li P., Wang J.. **Function and regulation of mitofusin 2 in cardiovascular physiology and pathology**. *Eur. J. Cell Biol.* (2018) **97** 474-482. DOI: 10.1016/j.ejcb.2018.07.003
36. Yu J., Xie X., Zhang Y., Jiang F., Wu C.. **Construction and analysis of a joint diagnosis model of random forest and artificial neural network for obesity**. *Front. Med. (Lausanne).* (2022) **9** 906001. DOI: 10.3389/fmed.2022.906001
37. Zhang S., Feng F., Dai J., Li J., Bu X., Xie X.. **Recombinant high-mobility group box 1 (rHMGB1) promotes NRF2-independent mitochondrial fusion through CXCR4/PSMB5-mediated Drp1 degradation in endothelial cells**. *Oxidative Med. Cell. Longev.* (2021) **2021** 9993240. DOI: 10.1155/2021/9993240
|
---
title: Will they or won't they? Understanding New Zealand adults' attitudes towards
using digital interventions
authors:
- Holly Wilson
- Penelope Hayward
- Liesje Donkin
journal: Frontiers in Digital Health
year: 2023
pmcid: PMC10030707
doi: 10.3389/fdgth.2023.1008564
license: CC BY 4.0
---
# Will they or won't they? Understanding New Zealand adults' attitudes towards using digital interventions
## Abstract
### Background
Digital interventions deliver healthcare via the internet or smartphone application to support people's well-being and health. Yet uptake is relatively poor. Furthermore, several studies exploring attitudes towards digital interventions have found inconsistent attitudes. In addition to this, regional and cultural nuances may further influence attitudes to digital interventions.
### Objective
This study aimed to understand New Zealand adults' attitudes towards digital interventions and their influences.
### Results
A mixed-method design consisting of a cross-sectional survey and semi-structured interviews found that New Zealand adults hold varied and complex attitudes towards digital interventions. Attitudes were found to be influenced by group membership and the scenarios in which digital interventions are made available. In addition, beliefs about the benefits and concerns surrounding digital interventions, knowledge, perceived views of others, and previous experience and confidence influenced these attitudes.
### Conclusions
Findings indicated that digital interventions would be acceptable if offered as part of the healthcare service rather than a standalone intervention. Key modifiable factors that could positively influence attitudes were identified and could be leveraged to increase the perceived acceptability of digital interventions.
## Introduction
E-health, using information and web-based technology to deliver healthcare [1, 2], has numerous benefits for individuals and healthcare systems. E-health has given people a greater choice, access to more information and increased ability to manage healthcare [1, 3] through the development of patient portals [4, 5] and virtual delivery methods [2, 6]. Likewise, it has increased safety through more rapid and accurate communication between healthcare professionals [7, 8] and facilitating best care practices through electronic decisional aides and prompts [4, 9]. Healthcare systems have also become more cost-effective through E-health [2, 10] by maximising access to resources [11, 12] and lower-cost treatment options [9].
One subtype of E-health is the delivery of structured health programmes via the internet or smartphone applications, termed digital interventions (DI). DI supports mental and physical health by helping people to engage in behaviours that may prevent the development of illness (13–16), facilitating early detection [15, 17, 18], and improving the management of chronic conditions (19–21). For physical health, DI can support people to engage in exercise and healthy eating [13, 14], to better self-manage their health or chronic illnesses (19–21), and improve medication adherence [21]. For mental health, DI can support people to engage in well-being behaviours as prevention [15, 16] and help people who would not meet the threshold for secondary or tertiary services [15, 17, 18]. Likewise, DI can deliver evidence-based psychological techniques that improve mental health (15, 22–24).
Despite the benefits of DI, real-world engagement could be better [25, 26]. For instance, real-world uptake of DI that focuses on anxiety and depression ranges between $1\%$–$28\%$ [26]. Low uptake means that potential beneficiaries are unlikely to seek out and use DI in their everyday life. Likewise, DI may not be considered a viable treatment option if there is low uptake and high attrition.
Attitudes are people's overall evaluation of a stimulus, which guide behaviour [27, 28]. Poor attitudes to DI have been linked to less likelihood of engagement (19, 27–32), whilst favourable attitudes increase the possibility of use (19, 29–31). Positive attitudes have also been linked to greater benefits following DI use [33] than those with unfavourable attitudes. Poor or negative attitudes towards DI are thought to be one of the leading causes of poor uptake and attrition in usage [34, 35].
How people come to hold attitudes about DI is complex. Two key models related to attitude formation for technology are the Technology Acceptance Model [TAM; [36, 37]] and the Unified Theory of Technology Acceptance [UTAUT; [38, 39]]. According to these models, such attitudes are influenced by people's beliefs about technology, particularly regarding benefits and concerns, perceived ease of use, perceived effort to use the technology, the opinion of important others (social norms), and whether people have access to the required technology (36–39). Similarly, numerous studies have shown that people's experiences [40, 41], the influence of key social relationships [29, 41, 42], and specific beliefs held about DI [27, 28], such as the expected benefits (29, 42–46), perceived ease of use [41, 46], perceptions of data security [19, 47], and internet confidence [29, 48] influence attitudes to technology.
Despite attitudes influencing the likelihood of using DI, surprisingly, little is known about the attitudes people hold. Attitudes about DI for physical health have rarely been studied, with few existing studies suggesting somewhat positive attitudes [49, 50]. Attitudes about DI for mental health have been more widely studied, perhaps as DI are more commonly used to support mental health. Evidence suggests that attitudes towards DI to support mental health vary (40, 51–54) and can differ by population or by the purpose of the DI. For instance, previous use leads to more favourable attitudes [52, 53], and people prefer using DI to support mild conditions rather than severe [40, 55]. Health professionals have shown less favourable attitudes than the public [51, 56]. Taken together, attitudes towards DI are complex and influenced by a range of factors that can influence uptake and use.
In New Zealand (NZ), little is known about people's attitudes towards DI. Previous research has shown that people in NZ have positive attitudes towards DI for weight loss [57] and that positive attitudes are linked to prior use [58, 59]. However, despite not knowing if people want to use DI, the NZ government increasingly plans to integrate technology and DI into healthcare [60]. To date, no studies have looked at NZ adults' attitudes to or intended use of DI. Given that international literature has demonstrated wide variation in attitudes and uptake, it is essential to explore attitudes to DI if there will be significant resource investment in this space. Therefore, this study sought to understand NZ adults’ attitudes towards DI and what shapes these attitudes.
## Method
This study utilised a mixed-methods approach consisting of a cross-sectional survey and semi-structured interviews to explore NZ adults' attitudes towards DI and the factors influencing these attitudes.
## Participants
Participants were NZ residents or citizens over 18 years who could speak and understand English at a level to consent to participate in the study. For the interview, participants also needed access to technology that would allow them to complete the interview by video-calling platform or telephone.
## Recruitment
Participants were recruited online between October 2020 and March 2021. Recruitment methods consisted of unpaid advertising on social media through the researchers' networks and contacting organisations with groups of interest, such as caregivers and health professionals. To limit bias from recruiting people online who may be more comfortable using the internet, printed flyers and posters were distributed in areas with a high density of population of interest. Purposeful recruitment of specific populations aimed to obtain a representative sample of the NZ population with an emphasis on recruiting Māori, the indigenous people of NZ.
A power analysis was conducted using an online sample size calculator (surveymonkey.com) based on a population of 5,000,000 with a confidence level of $95\%$ and a margin of error of $5\%$. This calculation gave a minimum sample size of a minimum of 385 participants. We aimed for a slightly higher sample size to ensure that a diverse range of participants was represented in our sample. A target sample size of 10–15 interviews was estimated to reach theoretical saturation, typically between 6 to 12 interviews. Recruitment ended when 400 participants completed the survey, and theoretical saturation was met at 14 interviews.
## Study procedure
Participants could take part in the survey either online or in pen-and-paper format. The survey gathered information about demographics, health status, internet access, usage and confidence. Participants were then provided with the following definition: “digital interventions are programmes or tools delivered via the internet or smartphone that have a clear structure or program and target health or mental health outcomes. They could be things like apps, chatbots, or online treatment programmes”. They then rated their attitudes towards DI (37 items) and factors that may influence attitudes based on the literature (43 items). Health professionals were also asked to rate their attitudes towards using DI in their workplace and for personal use (13 items). Following the survey, participants could enter a draw to win a gift voucher as an acknowledgement for their time.
Participants who expressed interest in an interview were contacted to confirm the interview's purpose and process and their willingness to participate. Interviews took place in person or via Zoom or telephone, depending on participant preference and COVID-19 restrictions. Interviews lasted between 20 and 60 min. The interviews centred around one key question, “What are your views on digital interventions?” with further probing questions exploring participant attitudes about DI and the factors that contributed to these views – for example, “What experiences have you had with digital interventions?”. Each interviewee received a NZ$30.00 voucher.
## Cross-sectional survey design
As there was no single scale that covered all items that may influence attitudes toward DI, items were developed from existing scales and key models including the TAM [36], the UTAUT [39], the extension of the UTAUT [41], the attitudes towards psychological online interventions questionnaire [APOI; [61]], and the e-therapy attitudes and process questionnaire [e-TAP; [62]]. Items were also drawn from previous studies that measured people's attitudes towards e-health [40, 44]. Where relevant, items were kept in their original format (see Table 1 for item source).
**Table 1**
| Concept | Description of concept | Scales items drawn from/adapted from | Items in this study (see Multimedia Appendix 1 for the survey) |
| --- | --- | --- | --- |
| Behavioural intention | Participants’ intention to use DI in the future | A web-based acceptance facilitating intervention for identifying patient's acceptance, uptake and adherence to internet and mobile-based pain interventions: a randomised controlled trial. (44) | 43–45 |
| Knowledge | Participants’ knowledge of DI | Extension of the UTAUT (41) | 41, 42 |
| Ease of use | Participants’ perceptions of how easy a DI is to use | A web-based acceptance facilitating intervention for identifying patient's acceptance, uptake and adherence to internet and mobile-based pain interventions: a randomised controlled trial (44) | 90, 92–94 |
| Ease of use | Participants’ perceptions of how easy a DI is to use | Items adapted from UTAUT (45) | |
| Ease of use | Participants’ perceptions of how easy a DI is to use | - Item B21- Item B22- Item B24 | 868791 |
| Ease of use | Participants’ perceptions of how easy a DI is to use | Generated by researchers | 88, 89 |
| Performance expectancy | Participants’ perceptions of the benefits of using a DI | Items adapted from the APOI scale (61) | |
| Performance expectancy | Participants’ perceptions of the benefits of using a DI | - ABE1- ABE2- ABE4- SEC1- SEC2- SEC3 | 108109110111112113114 |
| Social influence | The opinion of important others about DI | Acceptability of internet treatment of anxiety and depression (40)UTAUT (38): item SI1Generated by researchers | 96–9910095 |
| Effort expectancy | Participants’ perception of the effort required to engage with a DI, such as time and energy demands | (40)effort expectancy itemAPOI scale (61) item SEE4 | 103104102 |
| Access to appropriate technologies | If participants had access to the technology required to use a DI | UTAUT (38) item FC1 | 101 |
| Data security | Participants’ perception of how secure information would be | (40): concerns regarding data security | 105, 106 |
| Accessibility | Participants’ perceptions of privacy, flexibility, cost and convenience | Generated by researchers | 107, 115–117 |
| Internet accessibility | Participants’ accessibility to the internet | Generated by researchers | 4 |
| Internet anxiety and confidence | If people experience anxiety while using the internet and people's confidence using technology | (44): Internet anxiety items(40): Facilitating conditions items 1&2Generated by researchers | 29, 3029, 3028 |
| Culturally acceptability (results presented elsewhere) | If DI fitted within participants’ culture | Generated by researchers | 118–123 |
| COVID-19 | If participants’ experience with COVID-19 shaped attitudes | Generated by researchers | 140–142 |
## Ethical approval
Ethical approval was obtained from The University of Auckland Human Participants Ethics Committee, reference number UAPHEC3037.
## Data analysis
The cross-sectional survey was analysed using SPSS version 27. Scores were calculated by averaging all items related to each relevant factor. This occurred for overall attitudes towards DI, attitudes to DI that support physical health (PH), and attitudes to DI that supported mental health (MH) and each factor that was hypothesised to influence attitudes. Each average score was measured on a 5-point Likert scale.
Data were examined for normality, and if the assumptions of normality were not met, nonparametric tests were used. Independent samples t-test, one-way ANOVA and chi-square tests were used to examine group differences in attitudes. Pearson's correlations and independent samples t-test were used to examine the effect of each variable on attitudes. Stepwise regressions were calculated with each item with a significant Pearson's correlation with that attitude (Overall, PH, MH).
Interviews were audio-recorded and transcribed verbatim by HW. Each interview transcript was then analysed using inductive thematic analysis to draw the themes from the interviews themselves [63]. HW coded each interview to reflect key themes, while LD coded a selection of interviews. Coding occurred with hard copies or in Microsoft Word. The authors discussed codes themes to ensure agreement and consistent interpretation of themes. Where there was disagreement about themes, these were discussed and resolved.
## Results
Four hundred and eight people participated in the survey; 17 ($4.2\%$) participants were excluded as they did not meet the minimum completion rate of reporting at least their exposure to DI, leaving 391 participants for analysis (see Table 2 for demographics). Most participants were female ($80.1\%$, $$n = 313$$) with an average age of 44.44 years (range: 18–86 years, Mdn = 43.00, SD = 16.99). Most participants ($68.5\%$) identified as NZ European/Pākehā, with $13.6\%$ identifying as Māori, $1.0\%$ as Samoan, $3.3\%$ as Chinese, $2.3\%$ as Indian and $11.3\%$ as another ethnicity. Most participants ($60\%$) held at least a university qualification, and $99.2\%$ reported having access to the internet at home ($$n = 388$$). There were no significant differences between the participants who completed the survey on paper ($$n = 35$$, $8.9\%$) or online ($$n = 356$$, $91.9\%$) in terms of demographics, self-reported health, internet access or confidence using technology.
**Table 2**
| Unnamed: 0 | Unnamed: 1 | Final Sample | Final Sample.1 | Final Sample.2 | Final Sample.3 |
| --- | --- | --- | --- | --- | --- |
| | Excluded from the final analysis (n = 17) | Included in the final analysis (n = 391) | Completed the survey (n = 351) | Did not complete the survey (n = 40 | Differences between those who completed and those who did not complete |
| Demographic | n (%) | n (%) | n (%) | n (%) | |
| Residency Status | | | | | x2(1) = 1.12, p = .291 |
| Resident | 13 (76.5) | 341 (87.2) | 47 (13.4) | 37 (92.5) | |
| Citizen | 3 (5.9) | 50 (12.8) | 304 (86.6) | 3 (7.5) | |
| Not Reported | 1 (5.9) | 0 (0) | 0 (0) | 0 (0) | |
| Education | | | | | x2(11) = 8.77, p = .643 |
| No formal education | 2 (11.8) | 19 (4.9) | 17 (4.8) | 2 (5.0) | |
| NCEA Level 1/ School Certificate | 0 (0) | 23 (5.9) | 20 (5.7) | 3 (7.5) | |
| NCEA Level 2/ Six Form/University Entrance | 0 (0) | 19 (4.9) | 18 (5.1) | 1 (2.5) | |
| NCEA Level 3/ Bursary | 3 (17.6) | 23 (5.9) | 21 (6.0) | 2 (5.0) | |
| Level 4 Certificate | 0 (0) | 19 (4.9) | 19 (5.4) | 0 (0) | |
| Level 5 Diploma | 0 (0) | 12 (3.1) | 11 (3.1) | 1 (2.5) | |
| Level 6 Diploma | 0 (0) | 11 (2.8) | 11 (3.1) | 0 (0) | |
| Bachelors | 2 (11.8) | 128 (32.7) | 117 (33.3) | 11 (27.5) | |
| Masters | 2 (11.8) | 66 (16.9) | 56 (16.0) | 10 (25.0) | |
| Doctorate | 1 (5.9) | 40 (10.4) | 33 (33.3) | 7 (17.5) | |
| Other | 0 (0) | 30 (7.7) | 27 (7.7) | 3 (7.5) | |
| Not reported | 7 (41.2) | 0 (0) | 1 (0.3) | 0 (0) | |
| Employment | | | | | x2(6) = 5.00, p = .543 |
| Employed full-time | 6 (35.3) | 184 (47.1) | 169 (48.1) | 15 (37.5) | |
| Employed part-time | 1 (5.9) | 80 (20.5) | 70 (19.9) | 10 (25.0) | |
| Student | 1 (5.9) | 36 (9.2) | 20 (5.7) | 2 (5.0) | |
| Unemployed | 0 (0) | 22 (5.6) | 13 (3.7) | 7 (17.5) | |
| Sickness or disability benefit | 0 (0) | 13 (3.3) | 29 (8.3) | 0 (0) | |
| Retired | 1 (5.9) | 47 (12.0) | 42 (12.0) | 5 (12.5) | |
| Prefer not to say | 1 (5.9) | 8 (2.0) | 7 (2.0) | 1 (2.5) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Gender | | | | | x2(6) = 2.066, p = .914 |
| Male | 2 (11.8) | 68 (17.4) | 63 (17.9) | 5 (12.5) | |
| Female | 8 (47.1) | 313 (80.1) | 278 (79.2) | 27 (87.5) | |
| Gender Neutral | 0 (0) | 2(0.5) | 2 (0.6) | 0 (0) | |
| Non-Binary | 0 (0) | 3 (0.8) | 3 (0.9) | 0 (0) | |
| Other | 0 (0) | 3 (0.8) | 3 (0.9) | 0 (0) | |
| Prefer not to say | 0 (0) | 1 (0.3) | 1 (0.3) | 0 (0) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Ethnicity | | | | | x2(1) = .4.22, p = .518 |
| Māori | 0 (0) | 53 (13.6) | 50 (14.2) | 3 (7.5) | |
| NZ European/Pākehā | 7 (41.2) | 268 (68.5) | 241 (68.7) | 27 (67.5) | |
| Samoan | 0 (0) | 4 (1.0) | 3 (0.9) | 1 (2.5) | |
| Chinese | 0 (0) | 13 (3.3) | 12 (3.4) | 1 (2.5) | |
| Indian | 1 (5.9) | 9 (2.3) | 7 (2.0) | 2 (5.0) | |
| Other | 2 (11.8) | 44 (11.3) | 38 (10.8) | 6 (15.0) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Hold a community services card | | | | | x2(1) = 3.69, p = .06) |
| Yes | 0 (0) | 60 (15.4) | 58 (16.6) | 2 (5.0) | |
| No | 10 (58.8) | 330 (84.6) | 292 (83.4) | 38 (95.0) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Caregiver | | | | | x2(1) = 2.11, p = .146 |
| Yes | 0 (0) | 60 (15.3) | 57 (16.2) | 3 (7.5) | |
| No | 10 (58.8) | 331 (84.7) | 294 (83.8) | 37 (92.5) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Currently have a mental illness | | | | | x2(1) = 3.73, p = .053 |
| Yes | 0 (0) | 131 (33.7) | 123 (35.0) | 32 (80.0) | |
| No | 10 (58.8) | 258 (66.3) | 226 (64.4) | 8 (20.0) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Currently have a physical illness | | | | | x2(1) = 2.93, p = .087 |
| Yes | 3 (17.6) | 114 (29.2) | 107 (30.5) | 33 (82.5) | |
| No | 7 (41.2) | 277 (70.8) | 244 (69.5) | 7 (17.5) | |
| Not reported | 7 (41.2) | 0 (0) | 0 (0) | 0 (0) | |
| Health Professional | | | | | x2(1) = 1.68, p = .146 |
| Yes | 0 (0) | 104 (29.3) | 104 (29.9) | 0 (0) | |
| No | 0 (0) | 251 (70.7) | 247 (70.4) | 4 (100) | |
| Internet at home | | | | | x2(1) = 1.76, p = .185 |
| Yes | 9 (52.94) | 388 (99.2) | 349 (99.4) | 39 (97.5) | |
| No | 0 (0) | 3 (0.8) | 2 (0.6) | 1 (2.5) | |
| Not reported | 8 (47.1) | 0 (0) | 0 (0) | 0 (0) | |
| Amount of internet at home | | | | | x2(4) = 1.33, p = .857 |
| Unlimited | 8 (47.0) | 354 (90.8) | 316 (90.0) | 39 (97.5) | |
| Capped 50–70 | 1 (5.9) | 22 (5.6) | 21 (6.0) | 1 (2.5) | |
| Capped 25–50 | 0 (0) | 10 (2.6) | 9 (2.6) | 1 (2.5) | |
| Capped <25 | 0 (0) | 2 (0.5) | 2 (0.6) | 0 (0) | |
| | 0 (0) | 2 (0.5) | 2 (0.6) | 0 (0) | |
| Not reported | 8 (47.1) | 0 (0) | 0 (0) | 0 (0) | |
| Internet access on a phone | | | | | x2(2) = 5.96, p = .06 |
| Yes | 9 (52.94) | 375 (96.2) | 337 (96.3) | 38 (95.0) | |
| No | 0 (0) | 10 (2.6) | 10 (2.0) | 2 (5.0) | |
| Unsure | 0 (0) | 5 (1.3) | 3 (0.9) | 0 (0) | |
| Not reported | 8 (47.1) | 0 (0) | 0 (0) | 0 (0) | |
## Attitudes about DI
Participants had an overall neutral attitude towards DI ($M = 3.13$, Mdn = 3.18, SD = 0.77, $$n = 368$$, see Figure 1), attitudes to DI for physical health ($M = 3.13$, Mdn = 3.14, SD = 0.82, $$n = 369$$; see Figure 2) and attitudes to DI for mental health ($M = 3.09$, Mdn = 3.10, SD = 0.88, $$n = 369$$; see Figure 3). There was no significant difference between attitudes to DI for physical and mental health [t[368] = 1.28, $$p \leq .203$$, $d = 0.05$]. Participants' overall attitude to DI ($r = .65$, $p \leq .001$), attitude to DI for physical health ($r = .59$, $p \leq .001$) and for mental health ($r = .61$, $p \leq .001$) had strong positive correlations with behavioural intention to use DI in the future ($M = 3.96$, Mdn = 4.00, SD = 0.95, $$n = 369$$).
**Figure 1:** *Frequency distribution with a normal curve of the average overall attitude towards DI.* **Figure 2:** *Frequency distribution with a normal curve of average attitude towards DI for physical health (PH).* **Figure 3:** *Frequency distribution with a normal curve of average attitude towards DI for mental health (MH).*
Participant attitudes varied by group membership (see Table 3). Health professionals had favourable attitudes toward the use of DI in their professional practice ($M = 3.46$, Mdn = 3.66, SD = 0.90, $$n = 104$$) and for their personal use [$M = 3.24$, SD = 0.82, $$n = 103$$; t[102] = −5.55, $p \leq .001$]. Health professionals also had more favourable attitudes to DI for mental health for their personal use (MMH = 3.24, SD = 0.93; $$n = 104$$) than the attitudes held by the general population [MMH = 3.02, SD = 0.85, $$n = 251$$; tMH[353] = 2.13, $p \leq .05$, $d = 0.25$]. No differences were observed in overall attitude [t[352] = 1.92, $$p \leq .056$$, $d = 0.22$] and attitudes to DI for physical health conditions [t[353] = 1.62, $$p \leq 1.07$$, $d = 0.18$].
**Table 3**
| Variable | Overall attitude to DI | Overall attitude to DI.1 | Overall attitude to DI.2 | Overall attitude to DI.3 | Significant difference |
| --- | --- | --- | --- | --- | --- |
| Variable | M | SD | M | SD | Significant difference |
| Gender a | Male (n = 67) | Female (n = 292) | t(356) = −1.49, p = .14, d = 1.09 | | |
| Gender a | 2.99 | 0.85 | t(356) = −1.49, p = .14, d = 1.09 | 3.15 | 0.75 |
| Community services card holder | Yes (n = 59) | No (n = 309) | t(72.26) = −2.83, p < .05, d = 0.45 | | |
| Community services card holder | 2.84 | 0.86 | t(72.26) = −2.83, p < .05, d = 0.45 | 3.18 | 0.74 |
| Distance to nearest medical centre | 30 minutes or above (n = 11) | 30 minutes or less (n = 358) | t(366) = −1.42, p = .160, d = 0.44 | | |
| Distance to nearest medical centre | 3.45 | 0.79 | t(366) = −1.42, p = .160, d = 0.44 | 3.11 | 0.77 |
| Previously used a DI | Yes (n = 184) | No (n = 168) | t(366) = −3.62, p = <.001, d = 0.38 | | |
| Previously used a DI | 3.28 | 0.73 | t(366) = −3.62, p = <.001, d = 0.38 | 2.99 | 0.79 |
| Completion method | Paper (n = 34) | Online (n = 335) | t(35.52) = 0.86, p = .395, d = 0.27 | | |
| Completion method | 2.98 | 1.00 | t(35.52) = 0.86, p = .395, d = 0.27 | 3.14 | 0.74 |
| Informal caregivers | Yes (n = 58) | No (n = 311) | t(366) = 0.49, p = .622 d = 0.08 | | |
| Informal caregivers | 3.17 | 0.89 | t(366) = 0.49, p = .622 d = 0.08 | 3.11 | 0.74 |
| Have a physical health condition | Yes (n = 111) | No (n = 257) | t(366) = 1.10, p = .273, d = 0.13 | | |
| Have a physical health condition | 3.06 | 0.79 | t(366) = 1.10, p = .273, d = 0.13 | 3.16 | 0.76 |
| Have a mental health condition | Yes (n = 127) | No (n = 240) | t(364) = −0.80, p = .423, d = 0.09 | | |
| Have a mental health condition | 3.17 | 0.81 | t(364) = −0.80, p = .423, d = 0.09 | 3.10 | 0.75 |
Participants who lived rurally had slightly more favourable attitudes towards DI than those living in urban centres, but this was insignificant ($$p \leq .106$$). Participants who held a community services card (those with low income or are high users of health services) had a significantly less favourable overall attitude to DI and for DI to support mental health (Mattitude = 2.84, SD = 0.86, MPH = 2.91, SD = 0.97, MMH = 2.73, SD = 1.02, $$n = 59$$) than participants without a community services card (Mattitude = 3.18, SD = 0.74, MPH = 3.17, SD = 0.79, MMH = 3.16, SD = 0.84, $$n = 309$$; tattitude(72.26) = −2.83, $p \leq .05$, $d = 0.45$; tMH(73.40) = −3.04, $p \leq .05$, $d = 0.49$). There were no differences in attitudes about DI for physical health [t(73.43) = −1.95, $$p \leq .06$$, $d = 0.31$] between these groups. Previous use led to a more favourable overall attitude ($p \leq .001$).
No differences in attitudes were observed between people who completed the survey in paper format and those that completed it online ($$p \leq .395$$), those with mental ($$p \leq .423$$) or physical health conditions ($$p \leq .273$$) and those who did not, caregivers and non-caregivers ($$p \leq .622$$), between the genders ($$p \leq .14$$) or education levels ($F = 0.66$, $$p \leq .619$$, est w2 = 0.00).
As indicated by the average rating of items, participants preferred DI to support conditions of “mild” severity for both physical and mental health than conditions that were “moderate” or “severe” (see Table 4). DI were not perceived as acceptable to support people who were suicidal ($M = 2.33$, SD = 1.28, $$n = 370$$) and were a less acceptable option than medication ($M = 2.91$, SD = 1.18, $$n = 369$$) or psychological therapy ($M = 2.52$, SD = 1.09, $$n = 370$$).
**Table 4**
| Unnamed: 0 | Mild | Mild.1 | Moderate | Moderate.1 | Severe | Severe.1 | Unnamed: 7 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | M | SD | M | SD | M | SD | F test |
| Physical Health (n = 370) | 2.98 | 1.16 | 2.30 | 1.01 | 1.57 | 1.03 | F(1.58,58.64) = 270.66, p < .001 |
| Mental Health (n = 369) | 2.90 | 1.2 | 2.22 | 1.06 | 1.52 | 1.04 | F(1.47,539.55) = 293.00, p < .001 |
## Factors that were associated with people's attitudes
Several factors were associated with attitudes (see Table 5). Age had weak to moderate negative correlations with overall attitude to DI (<.001) and attitude to DI for physical health (<.001), but not the attitude to DI for mental health ($$p \leq 0.69$$). Knowledge about DI ($M = 3.66$, Mdn = 4.00 SD = 1.24, $$n = 370$$) had significant weak to moderate correlations with overall attitude ($p \leq .001$), attitude to DI for physical health (<.001) and DI for mental health ($p \leq .001$).
**Table 5**
| Variable | Overall attitude | Overall attitude.1 | Attitude to DI for physical health | Attitude to DI for physical health.1 | Attitude to DI for mental health | Attitude to DI for mental health.1 |
| --- | --- | --- | --- | --- | --- | --- |
| Variable | r | p | r | p | r | p |
| Average age | −.19 | <.001 | −.10 | .069 | −.25 | <.001 |
| Average knowledge of DI | .33 | <.001 | .25 | <.001 | .29 | <.001 |
| Average accessibility of DI | .67 | <.001 | .61 | <.001 | .65 | <.001 |
| Average social influence | .55 | <.001 | .53 | <.001 | .52 | <.001 |
| Average how easy a DI was to use | .51 | <.001 | .45 | <.001 | .50 | <.001 |
| Average perceptions of effort to use a DI | −.39 | <.001 | −.37 | <.001 | −.35 | <.001 |
| Average performance expectancy | .53 | <.001 | .51 | <.001 | .50 | <.001 |
| Average data security | .34 | <.001 | .35 | <.001 | .34 | <.001 |
| Confidence in using the internet | .21 | <.001 | .19 | <.001 | .20 | <.001 |
| Anxiety about using the internet | −.07 | .185 | −.06 | .242 | −.09 | .098 |
| Facilitating conditions | .23 | <.001 | .21 | <.001 | .22 | <.001 |
## Social influence
Social influence ($M = 3.99$, Mdn = 4.00, SD = 0.67, $$n = 355$$) had the strongest correlation with overall attitude to DI ($p \leq .001$), attitude to DI physical health ($p \leq .001$) and DI for mental health ($p \leq .001$). Participants had significantly greater intention of using a DI when recommended by a doctor [$M = 4.03$, SD = 0.99, tdoctor[369] = 4.81, $p \leq .001$, $d = 0.07$] or therapist [$M = 4.01$, SD = 1.03, ttherapist[368] = 4.41, $p \leq .001$, $d = 0.05$] rather than self-seeking for something they are struggling with ($M = 3.96$, SD = 0.95). Those who had been recommended a DI by someone close had significantly more favourable attitudes towards DI ($M = 3.37$, SD = 0.72, $$n = 65$$) than those who had not [$M = 3.01$, SD = 0.77, $$n = 305$$; tattitude[366] = 2.86, $p \leq .05$, $d = 0.47$].
## Perceptions about qualities of DI
Perceptions of ease of use ($M = 3.86$, Mdn = 3.89, SD = 0.88, $$n = 356$$) had strong positive associations with overall attitude ($p \leq .001$), attitude to DI for physical health ($p \leq .001$) and DI for mental health ($p \leq .001$). Perception of how much effort it would take to use a DI ($M = 2.77$, Mdn = 2.67, SD = 0.85, $$n = 356$$) had a significant negative association with overall attitude ($p \leq .001$), with DI to support physical health ($p \leq .001$) and DI to support mental health ($p \leq .001$). Performance expectancy ($M = 2.71$, Mdn = 2.71, SD = 0.59, $$n = 355$$) had significant positive associations with overall attitude ($p \leq .001$), with DI to support physical health ($p \leq .001$) and DI to support mental health ($p \leq .001$). Perceptions of ease of accessibility of DI ($M = 3.55$, Mdn = 3.75, SD = 0.79, $$n = 355$$) had strong positive correlations with overall DI attitude ($p \leq .001$), attitude to DI for physical health ($p \leq .001$) and DI for mental health ($p \leq .001$).
## Technology beliefs and confidence
Low concern about data security ($M = 3.18$, Mdn = 3.00, SD = 1.01, $$n = 362$$) had a positive relationship with overall attitude ($p \leq .001$), with DI to support physical health ($p \leq .001$) and DI to support mental health ($p \leq .001$). Confidence using the internet ($M = 4.57$, Mdn = 5.00, SD = 0.67, $$n = 397$$) had significant positive associations with overall attitude ($p \leq .001$), attitude to DI for physical health ($p \leq .001$) and DI for mental health ($p \leq .001$). Anxiety about using the internet ($M = 2.15$, Mdn = 2.00, SD = 1.14, $$n = 390$$) had non-significant weak correlations with overall attitude ($$p \leq .185$$), attitude to DI for physical health ($$p \leq .242$$) and DI for mental health ($$p \leq 0.89$$). Access to necessary technology ($M = 4.67$, Mdn = 3.75, SD = 0.69, $$n = 355$$) was associated with a more favourable attitude overall ($p \leq .001$), with DI to support physical health ($p \leq .001$) and DI to support mental health ($p \leq .001$).
## Impact of COVID-19
Most participants reported that the COVID-19 pandemic did not change their intention to use DI ($53.1\%$, $$n = 187$$); however, $42.3\%$ ($$n = 1490$$) of participants reported being more likely to use DI as a result of the pandemic.
## Factors influence beliefs about DI
Participants with a community services card had a significantly lower perception of performance expectancy ($M = 2.54$, SD = 0.68, $$n = 58$$) than those without [$M = 2.75$, SD = 0.56, $$n = 296$$; t[352] = −2.52, $p \leq .05$, $d = 0.57$] and less knowledge about DI ($M = 3.15$, SD = 1.37, $$n = 59$$) than those without a community services card [$M = 3.75$, SD = 1.20, $$n = 310$$; t[367] = 3.42, $p \leq .001$, $d = 0.35$]. Health professionals had significantly greater self-reported knowledge of DI [Mhealth professionals = 4.11, SD = 0.95, $$n = 104$$; Mpublic = 3.48, SD = 1.29, $$n = 251$$; t(257.39) = 5.06, $p \leq .001$, $d = 0.52$]. Participants under the age of 40 had greater performance expectancy ($M = 4.16$, SD = 0.68, $$n = 157$$) than those aged between 41 to 64 years ($M = 3.76$, SD = 0.84, $$n = 150$$), while those aged over 65 had poorest performance expectancy [$M = 3.86$, SD = 0.85, $$n = 48$$; F[2,352] = 26.60, $p \leq .001$, est w2 = 0.13]. People who had previously used a DI placed less importance on social influence ($M = 4.10$, SD = 0.62, $$n = 168$$) and had greater knowledge ($M = 4.18$, SD = 0.94, $$n = 176$$) than participants who had not used a DI (Msocial influence = 3.89, SD = 0.73, $$n = 187$$; t[353] = −2.87, $p \leq .05$, $d = 0.31$; Mknowledge = 3.19, SD = 1.03, $$n = 194$$; t[368] = −8.33, $p \leq .001$, $d = 1.00$).
## Predicting attitudes to DI
Stepwise regressions were used to predict attitudes toward DI. For each regression, all items that had significant correlations with the attitudes type were entered into the model. The stepwise regression (Table 6) showed that the recommendation of someone close, the approval of those close, the flexibility, and the perceived effectiveness of DI were the most important in shaping people's overall attitude [F[15,334] = 57.67, $p \leq .001$, est w2 = 0.71]. For attitudes to DI for PH, the approval of those close, the perceived effectiveness, the privacy and the cost were the best predictors of participants' attitudes [F[10,339] = 65.90, $p \leq .001$, est w2 = 0.65]. Finally, for attitudes to DI for MH, the recommendation of someone close, the approval of those close and the perceived effectiveness of the DI were key predictors of participants' attitudes [F[11,338] = 62.77, $p \leq .001$, est w2 = 0.66].
**Table 6**
| Variable | B | 95%CI for B | 95%CI for B.1 | SE B | R | R2 |
| --- | --- | --- | --- | --- | --- | --- |
| Variable | B | LL | UL | SE B | R | R2 |
| Overall attitude to DI | | | | | | |
| | | | | | .85 | .71 |
| Item 95: I would use a digital intervention if recommended by someone close to me | 0.18*** | 0.12 | 0.23 | 0.30 | | |
| Item 99: Those people who are important to me would approve of me using digital interventions for physical health | 0.12*** | 0.05 | 0.19 | 0.17 | | |
| Item 115: A digital intervention is appealing because of their flexibility | 0.10*** | 0.04 | 0.15 | 0.13 | | |
| Have you previously used a digital intervention | 0.10* | 0.01 | 0.19 | 0.06 | | |
| Item 112: I do not expect digital interventions to for physical health to be effective in the long term | −0.09** | −0.15 | −0.04 | −0.13 | | |
| Item 116: A digital intervention is appealing because of their cost | 0.08** | 0.03 | 0.13 | 0.10 | | |
| Item 107: A digital intervention is appealing because of privacy | 0.08** | 0.01 | 0.12 | 0.10 | | |
| Item 113: By using a digital intervention, I would not need professional support | 0.08** | 0.02 | 0.11 | 0.09 | | |
| Item 98: Those people who are important to me would approve of me using digital interventions for mental health | 0.08* | 0.01 | 0.14 | 0.10 | | |
| Item 96: Other people would think badly about me if I would use a digital intervention or mental health problems | 0.07** | 0.02 | 0.12 | 0.10 | | |
| Item 122: People from my culture use digital interventions | 0.07** | 0.02 | 0.11 | 0.10 | | |
| Item 111: I do not expect digital interventions for mental health to be effective in the long term | −0.07* | −0.13 | −0.01 | −.09 | | |
| Item 90: I would know where to get help if using a digital intervention | 0.06** | 0.21 | 0.10 | 0.10 | | |
| Item 102: Digital interventions could increase isolation and loneliness | −0.05* | −0.09 | −0.001 | −0.07 | | |
| Item 108: By using a digital intervention, I can reveal my feelings more easily than with a therapist | 0.04* | 0.01 | 0.09 | 0.07 | | |
| Attitudes towards DI for physical health | | | | | | |
| | | | | | .81 | .66 |
| Item 99: Those who are important to me would approve of me using digital interventions for my physical health | 0.23*** | 0.17 | 0.28 | 0.29 | | |
| Item 112: I do not expect digital interventions for physical health to be effective in the long term | −0.21*** | −0.26 | −0.15 | −0.27 | | |
| Item 95: I would use a digital intervention if recommended by someone close to me | 0.15*** | 0.09 | 0.21 | 0.18 | | |
| Item 107: A digital intervention is appealing because of their privacy | 0.11*** | 0.04 | 0.15 | 0.12 | | |
| Item 116: A digital intervention is appealing because of their cost | 0.10*** | 0.04 | 0.15 | 0.12 | | |
| Item 110: By using a digital intervention, I would not have to fear that someone will find out that I have a psychological or mental health problem | 0.06** | 0.02 | 0.11 | 0.09 | | |
| Item 113: By using a digital intervention, I would not need professional support | 0.06* | 0.01 | 0.12 | 0.77 | | |
| Item 90: I would know where to get help if using a digital intervention | 0.06* | 0.01 | 0.10 | 0.09 | | |
| Item 122: People from my culture use digital interventions | 0.06* | 0.01 | 0.11 | 0.08 | | |
| Item 102: Digital interventions could increase isolation and loneliness | −0.05 | −0.09 | −0.01 | −0.07 | | |
| Average attitude towards DI to support mental health | | | | | | |
| | | | | | .82 | .67 |
| Item 95: I would use a digital intervention if recommended by someone close to me | 0.19*** | 0.12 | 0.26 | 0.22 | | |
| Item 111: I do not expect digital interventions for mental health to be effective in the long term | −0.18*** | −0.23 | −0.12 | −0.22 | | |
| Item 98: Those people who are important to me would approve of me using digital interventions for mental health | 0.15*** | 0.09 | 0.22 | 0.18 | | |
| Item 115: A digital intervention is appealing because of their flexibility | 0.11*** | 0.04 | 0.18 | 0.13 | | |
| Item 122: People from my culture use digital interventions | 0.10*** | 0.04 | 0.15 | 0.12 | | |
| Item 116: A digital intervention is appealing because of their cost | 0.10** | 0.04 | 0.17 | 0.12 | | |
| Item 90: I would know where to get help if using a digital intervention | 0.09*** | 0.04 | 0.14 | 0.13 | | |
| Item 113: By using a digital intervention, I would not need professional support | 0.08** | 0.02 | 0.14 | 0.92 | | |
| Item 109: I would be more likely to tell my friends that I use a digital intervention than that I visit a therapist | 0.07* | 0.10 | 0.12 | 0.09 | | |
| Item 108: By using a digital intervention, I can reveal my feelings more easily than with a therapist | 0.05 | −0.00 | −0.11 | −0.08 | | |
| Age | −0.01* | −0.01 | −0.01 | −0.09 | | |
## Qualitative findings
Twenty-one people l were interested in participating in interviews and were contacted to participate. Fourteen people took part in the interview (see Table 7 for demographics), with $71.4\%$ ($$n = 10$$) being female and $71.4\%$ ($$n = 10$$) identifying as NZ European/Pākehā. The average age was 43.60 years (range 27–64 years).
**Table 7**
| Interview Number | Age | Gender | Ethnicity |
| --- | --- | --- | --- |
| 1 | 55 | Female | NZ European/Pākehā |
| 2 | 27 | Female | NZ European/Pākehā |
| 3 | 64 | Female | NZ European/Pākehā |
| 4 | – | Female | NZ European/Pākehā |
| 5 | – | Male | Māori |
| 6 | 37 | Female | Pasifika/Māori |
| 7 | – | Female | NZ European/Pākehā |
| 8 | 37 | Female | Māori |
| 9 | – | Female | NZ European/Pākehā |
| 10 | 36 | Male | Māori |
| 11 | 60 | Male | NZ European/Pākehā |
| 12 | 57 | Female | NZ European/Pākehā |
| 13 | 28 | Male | NZ European/Pākehā |
| 14 | 35 | Female | NZ European/Pākehā |
## People’s attitudes towards DI are varied
Eleven participants had positive attitudes towards DI. However, for two participants, having a positive attitude did not mean that they believed they would use a DI in the future, but rather that they could see the benefits and appeal of DI for others. Several participants also identified concerns about DI despite being positive about them –leaving them unsure if they would use them. Only one participant had negative views about DI and no interest in using it. Participants frequently explained that the key influence on their positive attitude towards DI was its perceived benefit.
All participants had clear views about the role of DI and where DI should sit in healthcare delivery. For example, DI were perceived to need to supplement existing healthcare and not replace existing services, with many participants identifying that face-to-face healthcare was gradually moving to a DI format. Similarly, participants believed DI would be the most beneficial in supporting mild illnesses or conditions. At the same time, there were perceptions that DI would be inadequate or risky in complex or high-risk situations. Therefore, participants may be reluctant to use DI when considering their condition as “severe” or “high-risk”, and being offered a DI in this situation may invalidate their concerns.
## Factors that shaped people's beliefs about DI
Attitudes about DI were shaped by participant beliefs, knowledge, and experience with DI. Specifically, people's beliefs about the benefits and risks of DI, experience and confidence with technology, and the opinions of others influenced their attitudes. Where participants had a favourable experience with DI in the past or knew of others that had good experiences, they were more likely to have favourable attitudes towards DI and to reduce anxiety about its use.
## Perceived benefits of DI above traditional healthcare shaped positive attitudes
All participants reported several benefits of DI when compared to traditional services. These benefits included increased control and choice over traditional services and the convenience and flexibility of DI rather than fixed treatment appointments and requirements. Likewise, DI provided immediate support and could be completed without disclosing their health status to their workplace or educational institution. DI were therefore perceived as offering improved anonymity, reduced travel costs, and healthcare services without waiting times. Benefits were particularly pronounced for people who lived rurally, as DI provided access to healthcare that may not be available locally. For the six participants who had used DI, the benefits they experienced while using it were crucial in shaping their attitudes.
## Concerns about problem avoidance and exacerbation of health concerns influence negative attitudes
However, despite most participants having favourable attitudes towards DI and perceiving many benefits of DI, twelve participants still had concerns. Concerns commonly expressed were that DI could enable people to avoid health concerns or problems more easily than they could in traditional healthcare. Specifically, participants believed that it was easier to be avoidant with an object or a tool than with a health professional. Likewise, participants expressed concerns that a lack of engagement with DI could be detrimental and that there was no safety net around this – such as no accountability or individual follow-up. Additionally, DI were perceived to have the potential to enable self-diagnosis of symptoms or reinforce harmful beliefs that could fuel anxiety and distress.
## Security and privacy concerns shaped negative attitudes about DI
Another common concern of DI was the security and management of information within the DI itself. Eight, generally older participants, expressed concern that their data would not stay private and could be shared with organisations or people with malicious intentions. Those who indicated this often had concerns about wider internet use and acknowledged that these anxieties were not limited to DI. Despite participants having some concerns about DI, they were still open to using them in the future and were hopeful that DI would benefit users.
## More knowledge about DI would shape more favourable attitudes
Eight participants reported that they had low awareness of DI, and this made it difficult to discuss their attitudes. Low awareness was also identified as a barrier to engagement. However, despite a lack of knowledge, participants reported that more knowledge about DI, particularly the endorsement of experts in the field or health professionals, would shape more favourable attitudes. Knowledge about the effectiveness of DI was also identified as important as it could help people navigate to good quality DI, where many currently available apps are of poor quality and lack an evidence base.
## Positive beliefs about technology mean positive attitudes towards digital intervention
Participants' beliefs about using technology shaped their attitudes towards DI. Participants who were confident using technology (seven participants), usually with substantial technological experience, had favourable attitudes towards DI. However, people who were anxious or had poor confidence using technology (three participants), usually with less technological experience, were likelier to have unfavourable attitudes. Although this was not age-specific, older participants often held these concerns. However, it was also noted that assumptions should not be made based on age, as some older participants felt comfortable with the technology.
Eight participants cited that DI had to be easy to use in terms of set-up, interaction with the app, accessible language and effective use of people's time. People who believed that DI were complex or complicated to use, or had used one with poor functionality or high levels of jargon in the past, were less likely to have favourable attitudes towards DI.
One critical factor in shaping seven participants' technology beliefs, seemingly due to increased exposure to technology, was the COVID-19 lockdowns. The restrictions on movement and access to healthcare meant greater exposure to and engagement with e-health, such as virtual consultations. This increased experience with e-health resulted in participants having greater confidence in using technology. Many participants indicated that this had influenced their attitudes and that, accordingly, they could see the benefits and appeal of DI when face-to-face healthcare was not possible.
## The opinion of important other people influences attitudes about DI
Recommendations from close friends, health professionals or experts established the belief that DI were trustworthy and potentially beneficial for seven participants. Likewise, for participants who had used a DI, deciding to use one was often motivated by the recommendation of a trusted friend or health professional. Recommendations (or criticism) by healthcare professionals were compelling. Thus, social influence is a critical motivating factor for positive attitudes and the possible use of DI.
## Discussion
This study was the first to explore NZ adults' attitudes about DI and what shapes these attitudes. In line with international findings, we found that attitudes towards DI are complex and ambivalent. Attitudes varied by group membership, with health professionals and people in rural areas reporting more favourable attitudes towards DI. Low-income or high-health need populations had less favourable attitudes. People want DI to be used alongside traditional healthcare, not as a replacement, and to support people with mild health conditions rather than complex or high-risk situations, such as when someone is suicidal. Some people, despite favourable attitudes, would not personally use a DI, often due to other concerns or a lack of motivation to engage with technology. People's beliefs about perceived benefits, concerns, technology, previous experience with DI, knowledge and social norms were the most important influences on their attitudes. Specifically, beliefs about the benefits of accessibility, flexibility and convenience of DI over traditional healthcare were associated with more favourable attitudes. Despite this, people had concerns about the lack of accountability around engagement leading to low motivation and potentially detrimental effects. People also held concerns about the security and privacy of the information provided. Social norms were associated with favourable attitudes. Specifically, the endorsement of an expert in the field or a health professional encouraged people to use a quality DI. Likewise, people's beliefs about strong data security, confidence and low anxiety about the internet, and access to the necessary technology, were associated with more favourable attitudes.
These findings are consistent with previous research that demonstrates variation in attitudes [40, 54] by group membership (51–53, 56) and the comfort with DI within healthcare systems [40, 55, 56, 64]. Conflicting with previous findings [51, 56], health professionals in NZ had more favourable attitudes towards DI for their personal use to support mental health compared to attitudes held by the general population. This may be due to greater professional exposure to DI or technology in their workplace; alternatively, DI may offer more confidentiality for healthcare professionals. People who held a community services card in NZ had slightly less favourable attitudes, which may be due to difficulties accessing DI and technology due to financial hardship. Alternately, it could be that low-income earners are also high healthcare users, so they are familiar with and prefer a traditional healthcare model. Equally, high service use may be associated with frustration with healthcare systems, such that DI might be perceived as “fobbing them off”.
The range of attitudes, and the belief that DI should supplement traditional healthcare, are consistent with previous literature [40, 49, 50, 54]. Similarly, participants believed that DI is likely to benefit people with mild conditions [40, 55, 56], further suggesting that healthcare consumers want DI to complement the existing healthcare system rather than replace it. In offering DI, patients have more choices in their healthcare delivery, which could increase access to services. Perhaps the belief that DI are most suitable for mild conditions is driven by insufficient knowledge of the effectiveness of DI for severe or complex conditions (22–24) or a general lack of knowledge about what DI can offer to whom. Moreover, these findings demonstrate that attitudes do not always reflect behaviour [27, 28], as some participants reported favourable attitudes but had no intention of using DI. This could be due to participants perceiving that the interviewer may hold a positive attitude about DI, thus activating a social desirability bias. Conversely, the lack of intention could also indicate that although participants were interested in DI and thought DI could be helpful, there were too many barriers, such as difficulties engaging with technology. This suggests that positive attitudes alone are not enough, and barriers to use also need to be considered in evaluating people's likelihood of engaging with DI.
Consistent with previous literature [36, 37, 39], people's beliefs about the advantages of DI, typically regarding accessibility, anonymity and convenience, were influential in shaping people's attitudes. Interestingly, beliefs in perceived effectiveness or health improvements from using the DI were not commonly mentioned as a benefit, except for those who had previously used a DI. This suggests that patients' beliefs about effectiveness are shaped by previous experience, while those without previous experience base their attitudes on the perceived practical advantages of DI, such as accessibility. Thus, the factors that shape attitudes may shift over time and with experience. Social norms were also important in influencing people's attitudes. A health professional's recommendation helped people trust in the effectiveness of the DI and influenced more favourable attitudes.
Beliefs, previous experience and social norms were associated with attitudes, which is mostly consistent with previous literature [19, 22, 47, 52, 55] and supports the validity of the UTAUT and TAM model. However, the influences on attitudes regarding DI for physical and mental health seem different. For physical health DI, the approval of others, the effectiveness and the accessibility of DI contributed the most to positive attitudes. Perhaps the improved availability of support and ease of using DI can potentially overcome physical barriers to treatment seeking, such as physical discomfort related to travel for people with physical health issues, thus increasing the appeal. Whereas, for mental health DI, a recommendation from someone close to the individual, perceived effectiveness, flexibility, and the approval of others were the most important influences on people's attitudes. Given the stigma surrounding MH, it is unsurprising that social approval and recommendations are key in attitude formation, as is flexibility to minimise the impact on one's life. For overall attitudes, recommendation and approval from someone close, flexibility, previous experience, and effectiveness were most influential in shaping attitudes. Given the commonalities here, perceived effectiveness and social norms are important in shaping people's attitudes and engagement with DI and should be highlighted in the promotion of DI.
This study adds to this literature by demonstrating that experiences, including the recent restrictions in movement due to the COVID-19 pandemic, led to increased exposure to healthcare technology, shaping more favourable attitudes. Inconsistent with previous research [29, 48, 65], experiencing anxiety while using the internet was not associated with less favourable attitudes towards DI. Again, this may reflect increased online experience during COVID-19 and participants now recognising that digital tools may be a part of mainstream life.
This study provides insight into the complexity and interaction of factors that shape people's attitudes toward DI. This study suggests that DI are not likely to be appealing or used by everyone. Instead, health promotion should target factors influencing attitudes to increase engagement. Specifically, it is essential to provide accurate information about the benefits and effectiveness of DI and concerns around maximising data security. Additionally, the recommendation of a trusted individual, health professional, or expert may also be beneficial. These improved attitudes could influence people's uptake of DI and ensure that people are accessing good quality DI. Such practical considerations are essential considerations in the context of the NZ government's plan to utilise DI [60] and could help to improve the uptake of DI. Therefore, before implementing DI, there needs to be an effort to improve people's attitudes, including focusing on the influences established here.
This study is limited as the sample does not represent all NZ adults. Specifically, the views of a well-educated, internet-connected and confident female were over-represented in this survey. Similarly, the interviews are not representative of the views of all NZ adults. These findings represent the views of a sample of NZ European/Pākehā and a small group of Māori who are diverse, while the views of other ethnic groups were not well-represented. Likewise, all participants in this study had access to the technology required to use DI, while people who do not have as easy access may hold different views. The cross-sectional nature of this study does not allow the exploration of causal relationships, and it could be that there is a reciprocal relationship between attitudes and the factors that influence them. Future research could also investigate the causal nature of these relationships in an NZ population. It is also important to note that NZ is a country that provides universal healthcare, and although a small country, it does have isolated pockets where it is difficult to access care. Due to this, the findings may not be generalisable to all other countries. As with any qualitative research, the researcher's own bias and lens may shape the interpretation of the findings; however, to mitigate/ restrict this risk, discussion was held amongst all researchers in the interpretation of findings. Despite these limitations, the large sample size and diversity of people included in the study still provide useful information on attitudes about DI.
## Conclusion
In conclusion, NZ adults hold varied attitudes about DI that are shaped by a complex interaction of beliefs, previous experience and social norms. Overall positive attitudes toward DI depend upon the context and healthcare circumstances in which DI is used. Unexpectedly, while some participants had direct concerns that would prevent them from using DI, others held positive attitudes but still no intention of using DI themselves. People's beliefs about the beneficial accessibility of DI relative to traditional healthcare and security concerns were vital in shaping people's attitudes. The influence of other people helped some people to decide that DI would be beneficial for their health needs – this social influence could be a method of targeting interventions to improve attitudes towards DI. Practically, DI are potentially a viable way to support the NZ healthcare system, but that would not be accessible to all members of the population if negative attitudes were not addressed. Therefore, future research must consider how to improve NZ adults’ attitudes towards DI to optimise their benefits for people's health, including professional recommendations, modifying people's underlying beliefs about DI or increasing people's exposures to DI or health-related technology. The complex interaction of beliefs, experiences and social norms shapes people's attitudes about DI.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Requests should be sent to the corresponding author.
## Ethics statement
This study was reviewed and approved by The University of Auckland Human Participants Ethics Committee, reference number UAPHEC3037. The study participants provided their written or oral consent to participate as deemed appropriate and in accordance with the guidance from the ethics committee.
## Author contributions
HW designed the research, conducted recruitment and the interviews, completed the analysis of the interviews and survey and wrote the manuscript. PH provided cultural supervision on the project, was involved in the interpretation of the findings and contributed to the manuscript. LD designed the research, conducted recruitment, completed one interview, coded a sample of manuscripts, supervised HW in the research project, analysis and writing, and 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.
## References
1. Eysenbach G. **What is eHealth?**. *J Med Internet Res* (2001) **3** e20. DOI: 10.2196/jmir.3.2.e20
2. Luxton DD, Magnavita JJ. **Behavioural and mental health apps**. *Using technology in mental health practice* (2018) 43-61. DOI: 10.1037/0000085-004
3. Hollis C, Morriss R, Martin J, Amani S, Cotton R, Denis M. **Technological innovations in mental healthcare: harnessing the digital revolution**. *Br J Psychiatry* (2015) **206** 263-5. DOI: 10.1192/bjp.bp.113.142612
4. Kruse CS, Bolton K, Freriks G. **The effect of patient portals on quality outcomes and its implication to meaningful use: a systematic review**. *J Med Internet Res* (2015) **17** e44. DOI: 10.2196/jmir.3171
5. Carini E, Villani L, Pezzullo AM, Gentili A, Barbara A, Ricciardi W. **The impact of digital patient portals on health outcomes, system efficiency, and patient attitudes: updated systematic review**. *J Med Internet Res* (2021) **23** e26189. DOI: 10.2196/26189
6. Coulon SM, Monroe CM, West DS. **A systematic, multi-domain review of mobile smartphone apps for evidence-based stress management**. *Am J Prev Med* (2016) **51** 95-105. DOI: 10.1016/j.amepre.2016.01.026
7. Eden KB, Messina R, Li H, Osterweil P, Henderson CR, Gusie J. **Examining the value of electronic health records on labor and delivery**. *Am J Obstet* (2008) **199** e1-e9. DOI: 10.1016/j.ajog.2008.07.004
8. Keasberry J, Scott LA, Sullican C, Staib A, Ashby R. **Going digital: a narrative overview of the clinical and organisational impacts of eHealth technologies in hospital practice**. *Aust Health Rev* (2017) **41** 646-64. DOI: 10.1071/AHI6233
9. Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. **An overview of clinical decision support systems: benefits, risks, and strategies for success**. *npj Digit Med* (2020) **3** 17. DOI: 10.1038/s41746-020-0221-y
10. Kawamoto K, Houlihan CA, Balas EA, Lobach DF. **Improving clinical practice using clinical decision support systems: a systematic review of trials to identify features critical to success**. *BMJ* (2005) **330** 765-8. DOI: 10.1136/bmj.38398.500764.8F
11. Hailey D, Roine R, Ohinmaa A. **Systematic review of evidence for the benefits of telemedicine**. *J Telemed Telecare* (2002) **8** 1-7. DOI: 10.1177/1357633X020080S201
12. Smit F, Lokkerbol J, Riper H, Boon B, Blankers M. **Modeling the cost-effectiveness of health care systems for alcohol use disorders: how implementation of eHealth interventions improves cost-effectiveness**. *J Med Internet Res* (2011) **13** e56. DOI: 10.2196/jmir.1694
13. Spring B, Pellegrini C, McFadden HG, Pfammatter AF, Stump TK, Siddique J. **Multicomponent mHealth intervention for large, sustained change in multiple diet and activity risk behaviors: the make better choices 2 randomised controlled trial**. *J Med Internet Res* (2018) **20** e10528. DOI: 10.2196/10528
14. Oosterveen E, Tzelepis F, Ashton L, Hutchesson MJ. **A systematic review of eHealth behavioral interventions targeting smoking, nutrition, alcohol, physical activity and/or obesity for young adults**. *Prev Med* (2017) **99** 197-206. DOI: 10.1016/j.ypmed.2017.01.009
15. Linardon J, Cuijpers P, Carlbring P, Messer M, Fuller-Tyskiwwica M. **The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomised controlled trials**. *World Psychiatry* (2019) **18** 325-36. DOI: 10.1002/wps.20673
16. Harrer M, Adam SH, Fleischmann RJ, Baumeister H, Auerbach R, Bruffaerts R. **Effectiveness of an internet- and app-based intervention for college students with elevated stress: randomised controlled trials**. *J Med Internet Res* (2018) **20** e136. DOI: 10.2196/jmir.9293
17. Kawadler JM, Hemmings NR, Ponzo S, Morelli D, Bird G, Plans D. **Effectiveness of a smartphone app (BioBase) for reducing anxiety and increasing mental well-being: pilot feasibility and acceptability study**. *JMIR Form Res* (2020) **4** e18067. DOI: 10.2196/18067
18. Deady M, Choi I, Calvo RA, Glozier N, Christensen H, Harvey SB. **eHealth interventions for the prevention of depression and anxiety in the general population: a systematic review and meta-analysis**. *BMC Psychiatry* (2017) **17** 310. DOI: 10.1186/s12888-017-1473-1
19. Patel S, Akhtar A, Malins S, Wright N, Rowley E, Young E. **The acceptability and usability of digital health interventions for adults with depression, anxiety, and somatoform disorders: qualitative systematic review and meta-synthesis**. *J Med Internet Res* (2020) **22** e16228. DOI: 10.2196/16228
20. Merchant R, Inamdar R, Henderson K, Barrett N, Su JG, Riley J. **Digital health intervention for asthma: patient-reported value and usability**. *JMIR Mhealth Uhealth* (2018) **6** e133. DOI: 10.2196/mhealth.7362
21. Aardoom JJ, Loheide-Niesmann L, Ossebaard HC, Riper H. **Effectiveness of ehealth interventions in improving treatment adherence for adults with obstructive sleep apnea: meta-analytic review**. *J Med Internet Res* (2020) **22** e16972. DOI: 10.2196/16972
22. Harrison V, Proudfoot J, Wee PP, Parker G, Pavlovic DH, Manicavasagar V. **Mobile mental health: review of the emerging field and proof of concept study**. *J Ment Health* (2011) **20** 506-24. DOI: 10.3109/09638237.2011.608746
23. March S, Spence SH, DOnovan CL, Kenardy JA. **Large-scale dissemination of internet-based cognitive behavioral therapy for youth anxiety: feasibility and acceptability study**. *J Internet Med Res* (2018) **20** e234. DOI: 10.2196/jmir.9211
24. McCall HC, Helgadottir FD, Menzies RG, Hadjistavropoulos HD, Chen FS. **Evaluating a web-based social anxiety intervention among community users: analysis of real-world data**. *J Med Internet Res* (2019) **21** e11566. DOI: 10.2169/11566
25. Reilly R, Stephens J, Micklem J, Tufanaru C, Harfield S, Fisher I. **Use and uptake of web-based therapeutic interventions amongst Indigenous populations in Australia, New Zealand, the United States of America and Canada: a scoping review**. *Syst Rev* (2020) **9**. DOI: 10.1186/s13643-020-01374-x
26. Fleming T, Bavin L, Lucassen M, Stasiak K, Hopkins S, Merry S. **Beyond the trial: systematic review of real-world uptake and engagement with digital self-help interventions for depression, low mood, or anxiety**. *J Med Internet Res* (2018) **20** e199. DOI: 10.2196/jmir.9275
27. Ajzen I, Fishbein M. *Understanding attitudes and predicting social behavior* (1980)
28. Fishbein M, Ajzen I. *Belief, attitude, intention and behavior: An introduction to theory and research* (1975)
29. Askari M, Klaver NS, van Gestel TJ, Klundert JVD. **Intention to use medical apps among older adults in the Netherlands: cross-sectional study**. *J Med Internet Res* (2020) **22** e18080. DOI: 10.2196/18080
30. Wangberg SC, Gammon D, Spitznogle K. **In the eyes of the beholder: exploring psychologists’ attitudes towards and use of e-therapy in Norway**. *Cyberpsychol Behav* (2007) **10** 418-23. DOI: 10.1089/cpb.2006.9937
31. Zhang X, Guo X, Lai K, Guo F, Li C. **Understanding gender differences in m-Health adoption: a modified theory of reasoned action model**. *Telemed e-Health* (2014) **20** 39-46. DOI: 10.1089/tmj.2013.0092
32. Leventhal H, Phillips LA, Burns E. **The Common-Sense model of self-regulation (CSM): a dynamic framework for understanding illness self-management**. *J Behav Med* (2016) **39** 935-46. DOI: 10.1007/s10865-016-9782-2
33. Aerts JEM, van Dam A. **Blended e-Health in cognitive behavioural therapy: usage intensity, attitudes and therapeutic alliance in clinical practice**. *Psychology* (2018) **9** 2422-35. DOI: 10.4236/psych.2018.910139
34. Badawy SM, Cronin RM, Hankins J, Crosby L, DeBaun M, Thompson AA. **Patient-centered eHealth interventions for children, adolescents, and adults with sickle cell disease: systematic review**. *J Med Internet Res* (2018) **20** e10940. DOI: 10.2196/10940
35. Parks AC, Williams AL, Kackloudis GM, Stafford JL, Boucher EM, Honomichl RD. **The effects of a digital well-being intervention on patients with chronic conditions: observational study**. *J Med Internet Res* (2020) **22** e16211. DOI: 10.2196/16211
36. Davis FD. **Perceived usefulness, perceived ease of use and user acceptance of information technology**. *MIS Q* (1989) **13** 319-39. DOI: 10.2307/249008
37. Davis FD. **User acceptance of information technology: system characteristics, user perceptions and behavioural impacts**. *Int J Man-Mach Stud* (1993) **38** 465-87. DOI: 10.1006/imms.1993.1022
38. Venkatesh V, Thong YLJ, Xu X. **Consumer acceptance and use of information technology: extending the unified theory of acceptance and use of technology**. *MIS Q* (2012) **36** 157-78. DOI: 10.2307/41410412
39. Venkatesh V, Morris GM, Davis GB, Davis FD. **User acceptance of information technology: towards a unified view**. *MIS Q* (2003) **27** 425-78. DOI: 10.2307/30036540
40. Ebert DD, Berking M, Cuijpers P, Lehr D, Pörtner M, Baumeister H. **Increasing the acceptance of internet-based mental health interventions in primary care patients with depressive symptoms. A randomised controlled trial**. *J Affect Disord* (2015) **176** 9-17. DOI: 10.1016/j.jad.2015.01.056
41. Hennemann S, Beutel ME, Zwerenz R. **Drivers and barriers to acceptance of web-based aftercare of patients in inpatient routine care: a cross-sectional survey**. *J Med Internet Res* (2016) **18** e337. DOI: 10.2196/jmir.6003
42. Hennemann S, Beutel ME, Zwerenz R. **Ready for eHealth? Health Professionals’ Acceptance and Adoption of eHealth Interventions in Inpatient Routine Care**. *J Health Commun* (2017) **22** 274-84. DOI: 10.1080/10810730.2017.1284286
43. Holtz BR, McCarroll AM, Mitchell KM. **Perceptions and attitude towards a mobile phone app for mental health for college students: qualitative focus group study**. *JMIR Form Res* (2020) **4** e18347. DOI: 10.2196/18347
44. Lin J, Faust B, Ebert DD, Krämer L, Baumeister H. **A web-based acceptance-facilitating intervention for identifying patients’ acceptance, uptake and adherence of internet-and mobile-based pain interventions: randomised controlled trial**. *J Med Internet Res* (2018) **20** e244. DOI: 10.2196/jmir.9925
45. Liu L, Cruz AM, Rincon AR, Buttar V, Ranson Q, Goertzen D. **What factors determine therapists’ acceptance of new technologies for rehabilitation—a study using the Unified Theory of Acceptance and Use of Technology (UTAUT)**. *Disabil Rehabil* (2015) **37** 447-55. DOI: 10.3109/09638288.2014.0923529
46. Bruno R, Abbott JM. **Australian health professionals’ attitudes towards and frequency of use of internet supported psychological interventions**. *Int J Ment Health* (2015) **44** 107-23. DOI: 10.1080/00207411.2015.1009784
47. Proudfoot J, Parker G, Pavlovic DH, Manicavasagar V, Adler E, Whitton A. **Community attitudes to the appropriation of mobile phones for monitoring and managing depression, anxiety and stress**. *J Med Internet Res* (2010) **12** e64. DOI: 10.2196/jmir.1475
48. Beatty L, Kemp E, Binnion C, Turner J, Milne D, Butow P. **Uptake and adherence to an online intervention for cancer-related distress: older age is not a barrier to adherence but may be a barrier to uptake**. *Support Care Cancer* (2017) **25** 1905-14. DOI: 10.1007/s00520-017-3591-1
49. Baumeister H, Seifferth H, Lin J. **Impact of acceptance facilitating intervention on patients’ acceptance of internet-based pain interventions: a randomised controlled trial**. *Clin J Pain* (2015) **31** 528-35. DOI: 10.1097/AJP.0000000000000118
50. Sanders AB, Conroy DE, Schmitz KH, Gusani N. **Physical activity and sedentary behavior in older gastrointestinal cancer survivors: need and acceptability of digital health interventions**. *Int J Gastrointest Cancer* (2019) **50** 703-8. DOI: 10.1007/s12029-018-0128-x
51. Schröder J, Berger T, Meyer B, Lutz W, Hautzinger M, Späth C. **Attitudes towards internet interventions among psychotherapists and individuals with mild to moderate depression symptoms**. *Cogn Ther Res* (2017) **41** 745-56. DOI: 10.1007/s10608-017-9850-0
52. Wootton BM, Titov N, Dear BF, Spence J, Kemp A. **The acceptability of Internet-based treatment and characteristics of an adult sample with obsessive compulsive disorder: an Internet survey**. *PLoS One* (2011) **6** e20548. DOI: 10.1371/journal.pone.0020548
53. van der Krieke L, Wunderink L, Emerenia AC, Jonge PD, Sytema S. **E-mental health self-management for psychotic disorders: state of the art and future perspectives**. *Psychiatr Serv* (2014) **65** 33-49. DOI: 10.1176/appi.ps.201300050
54. Handley T, Perkins D, Kay-Lambkin F, Lewin T, Kelly B. **Familiarity with and intentions to use internet-delivered mental health treatments among older rural adults**. *Aging Ment Health* (2015) **19** 989-96. DOI: 10.1080/13607863.2014.981744
55. Clarke J, Proudfoot J, Vatiliotis V, Verge C, Holmes-Walker DJ, Campbell L. **Attitudes towards mental health, mental health research and digital interventions by young adults with type 1 diabetes: a qualitative analysis**. *Health Expect* (2018) **21** 668-77. DOI: 10.1111/hex.12662
56. Gun SY, Titov N, Andrews G. **Acceptability of Internet treatment of anxiety and depression**. *Australas Psychiatry* (2011) **19** 259-65. DOI: 10.3109/10398562.2011.562295
57. Gorton D, Dixon R, Maddison R, Mhurchu CN, Jull A. **Consumer views on the potential use of mobile phones for the delivery of weight-loss interventions**. *J Hum Nutr Diet* (2011) **24** 616-9. DOI: 10.1111/j.365-277X.2011.01163.x
58. Mhurchu CN, Whittaker R, McRobbie H, Ball K, Crawford D, Michie J. **Feasibility, acceptability and potential effectiveness of a mobile health (mHealth) weight management programme for New Zealand adults**. *BMC Obesity* (2014) **1** 1-10. DOI: 10.1186/2052-9538-1-10
59. Dobson R, Whittaker R, Jiang Y, Maddison R, Shepherd M, McNamara C. **Effectiveness of text message based, diabetes self management support programme (SMS4BG): two arm, parallel randomised controlled trial**. *Br Med J* (2018) **36** 1-10. DOI: 10.1136/bmj.k1959
60. 60.Ministry of Health. New Zealand Health Strategy: Future direction. Wellington, New Zealand: Ministry of Health (2016).. *New Zealand Health Strategy: Future direction* (2016)
61. Schröder J, Sautier L, Kriston L, Berger T, Meyer B, Späth C. **Development of a questionnaire measuring attitudes towards psychological online interventions-the APOI**. *J Affect Disord* (2015) **187** 136-41. DOI: 10.1016/j.jad.2015.08.044
62. Clough BA, Eigeland JA, Madden IR, Rowland D, Casey LM. **Development of the eTAP: a brief measure of attitudes and process in e-interventions for mental health**. *Internet Interv* (2019) **18** 100256. DOI: 10.1016/j.invent.2019.100256
63. Braun V, Clarke V. **Using thematic analysis in psychology**. *Qual Res Psychol* (2006) **3** 77-101. DOI: 10.1191/1478088706qp063oa
64. Meisel SF, Drury H, Perera-Delcourt R. **Therapists’ attitude to offering eCBT in an inner-city IAPT service: a survey study**. *Cogn Behav Therapist* (2018) **11** 1-16. DOI: 10.1017/S17544X18000107
65. Philippi P, Baumeister H, Apolinário-Hagen J, Ebert DD, Hennemann S, Kott L. **Acceptance towards digital health interventions - Model validation and further development of the Unified Theory of Acceptance and Use of Technology**. *Internet Interv* (2021) **26** 100456. DOI: 10.1016/j.invent.2021.100459
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.