https://orcid.org/0000-0003-0301-0805
Figures
Abstract
Introduction
Using body mass index (BMI) as a proxy, previous Mendelian randomization (MR) studies found total causal effects of general obesity on polycystic ovarian syndrome (PCOS). Hitherto, total and direct causal effects of general- and central obesity on PCOS have not been comprehensively analyzed.
Objectives
To investigate the causality of central- and general obesity on PCOS using surrogate anthropometric markers.
Methods
Summary GWAS data of female-only, large-sample cohorts of European ancestry were retrieved for anthropometric markers of central obesity (waist circumference (WC), hip circumference (HC), waist-to-hip ratio (WHR)) and general obesity (BMI and its constituent variables–weight and height), from the IEU Open GWAS Project. As the outcome, we acquired summary data from a large-sample GWAS (118870 samples; 642 cases and 118228 controls) within the FinnGen cohort. Total causal effects were assessed via univariable two-sample Mendelian randomization (2SMR). Genetic architectures underlying causal associations were explored. Direct causal effects were analyzed by multivariable MR modelling.
Results
Instrumental variables demonstrated no weak instrument bias (F > 10). Four anthropometric exposures, namely, weight (2.69–77.05), BMI (OR: 2.90–4.06), WC (OR: 6.22–20.27), and HC (OR: 6.22–20.27) demonstrated total causal effects as per univariable 2SMR models. We uncovered shared and non-shared genetic architectures underlying causal associations. Direct causal effects of WC and HC on PCOS were revealed by two multivariable MR models containing exclusively the anthropometric markers of central obesity. Other multivariable MR models containing anthropometric markers of both central- and general obesity showed no direct causal effects on PCOS.
Conclusions
Both and general- and central obesity yield total causal effects on PCOS. Findings also indicated potential direct causal effects of normal weight-central obesity and more complex causal mechanisms when both central- and general obesity are present. Results underscore the importance of addressing both central- and general obesity for optimizing PCOS care.
Citation: De Silva K, Demmer RT, Jönsson D, Mousa A, Teede H, Forbes A, et al. (2022) Causality of anthropometric markers associated with polycystic ovarian syndrome: Findings of a Mendelian randomization study. PLoS ONE 17(6): e0269191. https://doi.org/10.1371/journal.pone.0269191
Editor: Renato Polimanti, Yale University School of Medicine, UNITED STATES
Received: February 11, 2022; Accepted: May 16, 2022; Published: June 9, 2022
Copyright: © 2022 De Silva et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data used in this study are freely available in the IEU GWAS database at: https://gwas.mrcieu.ac.uk/.
Funding: KDS is supported by a PhD scholarship funded by the Australian Government under Research Training Program (RTP). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Polycystic ovarian syndrome (PCOS) is the most common endocrine disorder in women of reproductive age [1], with a prevalence ranging 8%-18% among this group [2]. It is also the major cause of anovulatory female infertility [3]. While a multitude of reproductive and metabolic abnormalities associate with PCOS, its convoluted etiology, presumably multifactorial and heterogeneous, is still not completely known. Classic clinico-pathologic features of PCOS comprise hyperandrogenism, oligo-anovulation, excessive weight or obesity and a range of metabolic manifestations such as glucose intolerance, insulin resistance, and dyslipidemia while a number of obstetric, cardiometabolic, oncological, and psychological complications may also ensue [4]. A recent review highlighted the correlation between obesity, hyperandrogenism, and insulin resistance, and suggested these might act in concert forming a vicious cycle to induce PCOS [5]. In addition, neuroendocrine changes such as gonadotropin secretory abnormalities [6], fetal programming alterations within the intrauterine microenvironment [7], genetic- and epigenetic factors [8], an array of environmental predictors [9] and certain inflammatory regulators [10] have also been implicated in the pathogenesis of PCOS. Constraints to the present PCOS care practices include a lack of uniform diagnostic criteria, limitations intrinsic to the evidence base emanating from observational epidemiological studies such as the concerns of confounding and reverse causality [11] as well as diagnostic delays and difficulties [12].
It should be noted that a substantial proportion of the PCOS-related evidence base stems from observational studies which are precluded by various biases and unmeasured confounding. However, causal inference is fundamental to unravelling disease etiologies and mechanistic underpinnings. The gold standard of etiological inference remains randomized controlled trials which could overcome key limitations of observational studies. Given the practical drawbacks of randomized trials especially exorbitant financial costs and longer time spans entailed, Mendelian randomization (MR) methods offer a viable and robust alternative to inferring causality using observational data. Using genetic variants as instrumental variables (IVs) and leveraging on their strengths, especially stability and random assortment of alleles, the causality between risk factors and diseases can be plausibly deduced by MR. With the increasing availability of genome-wide association studies (GWAS) data, MR is making considerable contributions to uncovering complex, polygenic disease etiologies [13]. Well-conducted MR studies meeting the three key assumptions of relevance, independence, and exclusion restriction could thus provide high-level evidence on causality [14]. To this end, guidelines such as the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization (STROBE-MR) [15] have been developed to facilitate uniform and comprehensive presentation of MR studies.
Obesity is frequently observed in women with PCOS, predominantly in the form of abdominal obesity, aggravating metabolic and reproductive sequalae and PCOS-associated complications [16]. Available evidence suggests that the close link between obesity and PCOS is presumably mediated by multiple mechanisms such as insulin resistance-driven metabolic changes, steroidogenic and reproductive effects of hyperinsulinemia, and augmented adipokine synthesis by subcutaneous and visceral fat [17]. De Segher et al underscored the absence of a gonadotropic and/or ovarian disorder in a majority of girls with PCOS and theorized that central obesity and central adiposity are the main drivers of PCOS development [18]. Putative genetic links between obesity and adiposity with PCOS are reinforced by current evidence [19–21], for instance, the influence of FTO gene variants [22, 23]. A caveat in the relationship between obesity and PCOS is that obesity alone may be neither a necessary nor a sufficient condition for its development among all PCOS-susceptible females [19], given the presence of lean PCOS phenotypes [24], exemplifying the complex, multifactorial etiology of PCOS.
Notably, multiple MR studies revealed that obesity potentially causes PCOS, all of which exclusively used body mass index (BMI) as the surrogate anthropometric measure of obesity [25–28]. However, BMI is essentially a marker of general obesity whereas central obesity in particular seems to confer a formidable influence on PCOS pathogenesis [16–18]. Moreover, the use of multiple anthropometric traits could yield not only complementary but also exclusive information. For example, a recent consensus statement highlighted the unequivocal evidence that waist circumference (WC) provides additive and independent information to BMI and recommended that WC be used as a vital sign in clinical practice [29]. Also, concordant and consistent findings emanating from MR analyses using multiple anthropometric markers of central- and general- obesity could strengthen the evidence pertaining to their putative causal roles in ensuing PCOS. Anthropometric traits are customarily exploited in MR studies as valid and reliable surrogate markers for operationalizing central- and general obesity [30–32]. In these studies, multiple anthropometric markers were incorporated in univariable MR to gain evidence on causal roles of central- and general obesity [30–32].
In this study, we examined existing data from GWAS, and aimed to investigate the causality of multiple anthropometric markers of central- and general- obesity associated with PCOS via MR analyses. Specifically, we conducted univariable two-sample MR (2SMR) to determine total causal effects, multivariable MR to assess direct causal effects, and bidirectional MR- analyses to evaluate reverse causality, using anthropometric traits of central obesity (WC, hip circumference (HC), waist-to-hip ratio (WHR)) and general obesity (BMI and its constituent variables–weight and height).
Materials and methods
This study was conducted according to STROBE-MR guidelines [15], as described in S1 Table. Formal ethics approval was not required, being an analysis of publicly available, deidentified, summarized data.
Data sources for exposures
For the six anthropometric traits selected as exposures, we performed a comprehensive search on the IEU Open GWAS Project database (https://gwas.mrcieu.ac.uk/) to find the most suitable GWAS summary data for MR analyses. Given the female-exclusivity of PCOS and sex-based variations in genetically-proxied anthropometric traits [33], we retrieved female-only GWASs. In order to alleviate confounding by ancestry and population stratification effects, we resorted to cohorts with participants of European ancestry. We prioritized GWASs with large sample sizes and ultimately identified appropriate studies from the Genetic Investigation of ANthropometric Traits (GIANT) consortium (Table 1).
For weight, we selected the study with the GWAS-ID “ieu-a-107” reporting summary statistics from a meta-analysis of 46 cohorts over 73137 females of European ancestry and 2747007 SNPs. Data had been age-adjusted and details of the original study is available elsewhere [33].
The study selected for height with the GWAS-ID “ieu-a-97” report summary information from a meta-analysis of 46 studies spanning 73137 females of European ancestry and 2748546 SNPs. Data had been age-adjusted and further information of the study is published elsewhere [33].
For BMI, the study with the GWAS-ID “ieu-a-974” was selected, which report summary data from a large-scale meta-analysis of 82 GWASs over 171977 females of European ancestry and 2494613 SNPs. Data had been age-adjusted and details are given elsewhere [34].
For WC and HC, we selected studies with the GWAS-IDs “ieu-a-63” and “ieu-a-51” respectively, which report pooled statistics from large-scale meta-analyses of 127997 females of European ancestry and 2444355 SNPs. Data had been adjusted for age and other study-specific covariates and details are available elsewhere [35].
The study with the GWAS-ID “ieu-a-75” was selected for WHR, which report pooled results from a meta-analysis encompassing 118003 females of European ancestry and 2466102 SNPs. Data had been adjusted for age and other study-specific covariates and details are available elsewhere [35].
Data source for PCOS
In order to avoid sample-overlapping with exposures, we selected PCOS GWAS summary statistics from a different source i.e. the FinnGen cohort on the IEU Open GWAS Project database. The FinnGen study entails a growing repository of genomic and clinical data emanating from a nationwide network of Finnish biobanks (https://finngen.gitbook.io/documentation/). We used the GWAS with the specific ID “finn-b-E4_POCS” (E4_POCS is the FinnGen phenocode for PCOS) which consisted of 118870 samples (642 cases; 118228 controls) and 16379676 genotyped SNPs in total (Table 1). All PCOS cases were clinically diagnosed from hospital discharge registries and cause of death registries using female-specific clinical endpoints (ICD-10: E282, ICD-8: 25690).
Selection of genetic variants as IVs
The SNPs selected as IVs need to fulfil three key assumptions for MR to yield valid results: (1) strongly associated with the exposure (relevance); (2) not associated with the outcome due to confounding (independence); (3) affect the outcome only through the exposure (exclusion restriction). In order to fulfil the first criterion of relevance, we selected biologically and statistically plausible SNPs at a genome-wide significance threshold of p < 5e-08. We assessed the statistical power of individual SNPs and the potential weak instrument bias via F-statistic, as defined below.
F = v2 × (n − 2)/(1 − v2), in which v2 indicates the variance of the exposure phenotype attributable to a given SNP and n stands for the sample size. Variance estimates were calculated using the following formula:
in which β denotes the per-allele effect size of the association between a given SNP and the exposure phenotype, eaf stands for the effect allele frequency, and se is the standard error of β [36]. Weak instrument bias may ensue when F < 10 [37].
We minimized confounding through the inclusion of female-only cohorts of European ancestry from a single consortium while the summary data had been adjusted for age and other study specific covariates. The assumption of exclusion restriction may be violated in the presence of horizontal pleiotropy, especially unbalanced (directional) horizontal pleiotropy–where one or more IVs exert a net effect on the outcome via a pathway not involving the exposure, biasing the MR estimate. In order to overcome horizontal pleiotropy, we conducted robust MR methods, post-hoc pleiotropy analyses, heterogeneity tests and outlier analyses [38], the details of which are provided later in the manuscript.
As the absence of linkage disequilibrium (LD) is a prerequisite for most MR methods, we performed clumping to prune statistically plausible IVs (p < 5e-08), selecting those SNPs with LD-R2 < 0.001 and clumping-distance > 10000 kb. When any exposure SNP was not present in the outcome data, we included proxy SNPs via LD tagging, with following specifications: minimum LD-R2 value = 0.8; minor allele frequency threshold for aligning palindromic SNPs = 0.3. We performed allele harmonization by aligning strands for all SNPs including palindromes, to ensure that the effects of the SNPs on the exposure correspond to the same allele as their effects on the outcome.
2SMR analyses
We applied four 2SMR methods i.e. inverse variance weighted (IVW) method using multiplicative random effects model as the primary analysis along with three robust methods: MR-Egger (MRE), weighted median (WME), and weighted mode (WMO) [38].
The IVW method is the recommended main MR method to be used with summarized data and multiple, uncorrelated genetic variants, since it is the most efficient approach in the presence of valid IVs, yielding causal estimates that are accounted for heterogeneity [38]. This method pools Wald ratio estimates for individual SNPs using inverse of the variance as weights.
The MRE method provides an asymptotically consistent causal effect measure adjusted for horizontal pleiotropy by pooling individual SNP-specific Wald ratios via an adapted Egger regression. In fact, its regression intercept is an estimate of the net pleiotropic effect which can therefore be used to assess horizontal pleiotropy. It allows for invalid IVs, provided the Instrument Strength Independent of Direct Effect (INSIDE) assumption (i.e. instrument strengths are independent of horizontal pleiotropic effects) holds true. However, this method is underpowered in the presence of relatively homogeneous SNP-exposure effect sizes, susceptible to regression dilution bias, and the causal estimation is heavily affected by outliers [38–40].
Using the weighted median of Wald ratios, the WME method produces an asymptotically consistent estimate of the causal effect, provided 50% or more of the variants are valid IVs that do not violate the exclusion restriction criterion [38]. The WMO method clusters SNPs into groups based on the similarity of their individual ratio estimates, calculates the inverse variance weighted number of SNPs in each cluster, and produces a causal estimate based on the cluster having the largest weighted number of SNPs [41]. Both WME and WMO methods require some genetic variants to be valid instruments and are robust to outliers [38].
Results from 2SMR analyses were summarized in tabular format, along with odds ratios (OR) and 95% confidence intervals (CI) for all β estimates.
Method comparison plots
Scatter plots and trend lines pertaining to different 2SMR methods were generated for each anthropometric exposure-outcome analysis. Slopes and directions of trend lines represent the magnitudes and directions of causal estimates, respectively.
Single SNP analyses
Causal effects of each SNP were determined individually and were visualized in forest plots along with pooled estimates using all SNPs, under IVW and MRE methods. With respect to each exposure, we investigated for the presence of any significantly causally associated individual SNPs at Bonferroni multiple testing corrected p-value thresholds.
Leave-one-out sensitivity analyses
We conducted leave-one-out sensitivity analyses to assess whether causal estimates were significantly influenced by a single SNP. Wald ratios from IVW-MR analyses conducted excluding each SNP as well as pooled IVW-MR estimates encompassing all SNPs were visualized in forest plots.
Heterogeneity analyses
Heterogeneity is a measure of the consistency of the causal estimate across all SNPs whereby lower heterogeneity is indicative of a reliable MR estimate. We assessed heterogeneity via Cochran’s Q statistic and associated p-values. Funnel plots were also generated to visually assess heterogeneity. Lower values on the y-axis denote less precise estimates while those with increasing precision tend to ‘funnel’ in. Larger distributions suggest heterogeneity which may have resulted from horizontal pleiotropy whereas asymmetric distributions are indicative of unbalanced horizontal pleiotropy.
Analysis of horizontal pleiotropy and outliers
The MRE regression intercept is an estimator of the magnitude of horizontal pleiotropy [39]. For each anthropometric exposure, we calculated the MRE intercept, its standard error, and directionality p-value. The Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MRPRESSO) method presents a unified framework to assess pleiotropy and outliers via a three-step process: identification of pleiotropy and outliers (MRPRESSO global test); rectification of pleiotropy by removing outliers (MRPRESSO outlier test); analysis of the distortion in the causal estimate before and after removing outliers (MR-PRESSO distortion test) [42]. We applied the MRPRESSO method to evaluate pleiotropy and outliers with respect to all exposure-outcome associations. Radial plots have been proposed as an improved strategy for visualizing outliers in 2SMR analyses [43]. We generated radial plots for all exposure-outcome associations that were investigated with 2SMR in this study.
Analyses were conducted in R (version 4.1.2) [44], using “TwoSampleMR” [45], “MRPRESSO” [42], and “RadialMR” [43] packages.
Exploring genetic architectures underlying causality
Subsequent to causality testing, MR studies increasingly conduct downstream analyses to explore genetic underpinnings [46, 47]. For instance, previous MR studies showed that functional annotations of highly significant SNPs that act as major drivers of causality in MR models could assist in gaining etiologic insights [46] and identifying putative drug targets [47]. For significant causal associations, we predicted nearest gene(s)/transcriptional start site(s) ascribed to corresponding SNPs in harmonized datasets, using the Open Targets Genetics Portal [48]. In order to identify shared and non-shared genetic architectures underlying these significant causal associations, we explored overlapping and non-overlapping SNPs and genes across exposures.
Multivariable MR analyses
Standard, univariable MR estimates the total causal effect of an exposure on the outcome whereas multivariable MR measures the direct causal effect of an exposure on an outcome, conditioned on other exposure(s) [49]. Multivariable MR is an advancement of MR to account for IVs that are associated with multiple exposures. Its application is envisaged in situations such as when the exposures are biologically related and when exposures potentially formulate a network of causal effects. Specifically, its benefits in the presence of complexities such as confounders, colliders, vertical or functional pleiotropy, and mediation have been demonstrated [49, 50]. The three assumptions of IVs in multivariable MR are that the SNPs: associate with one or more exposures (relevance); do not associate with the outcome via confounding (independence); do not to affect the outcome except via its association(s) with single or multiple exposure(s) being assessed (exclusion restriction) [38].
In order to appraise direct causal effects, we performed multivariable MR analyses with various combinations of anthropometric markers that demonstrated significant total causal effects in preceding, univariable 2SMR analyses. This was performed using the multivariable inverse variance weighted method deployed through the ‘mr_mvivw’ function in the ‘MendelianRandomization’ R package [51, 52]. In brief, this method entails the following sequence of steps: acquisition of IVs for each exposure; collation of these into a set of all IVs; clumping them to alleviate LD between variants across exposures; re-extraction of all the final, clumped IVs from all of the exposures; harmonization of variants to be on the same reference allele; application of multivariable MR upon the harmonized data. Specifically, multivariable weighted linear regression was conducted and the direct causal effect was estimated by regression of the associations with PCOS on the associations with the anthropometric exposures, with the intercept set to zero and weights being the inverse variances of the associations with PCOS. A multiplicative random effects model was assumed in all multivariable analyses, accounting for heterogeneity between causal estimates and allowing for over-dispersion in regression models. Heterogeneity statistic (Cochran’s Q) and associated p-value tested the null hypothesis that all genetic variants estimated the same causal estimate and its rejection indicates the presence of single or multiple pleiotropic variants.
Bidirectional MR analyses
In order to assess the reverse causality hypothesis (i.e. whether PCOS in turn causes central or general obesity), we conducted a set of univariable 2SMR analyses, taking PCOS as the exposure and each anthropometric marker as the outcome. For this purpose, we used the same data sources described previously. We applied the same criteria and specifications as stated previously for the selection of genetic variants as IVs, while 2SMR analyses followed the same methodological approach detailed earlier.
Results
Genetic variants selected as IVs
Harmonized datasets containing details of the SNPs selected as IVs for each exposure are presented in S2 Table. Eleven SNPs underlying the weight-PCOS association were identified, which comprised no proxies and no palindromes. We included 51 SNPs with respect to the height-PCOS association, which consisted of no proxies and 5 palindromes. Thirty-seven SNPs were identified with respect to the BMI-PCOS association, which comprised 2 proxies and 2 palindromes. With respect to WC-PCOS and HC-PCOS associations, we identified 17 identical SNPs, which comprised no proxies and 2 palindromes. We identified 23 SNPs underlying the WHR-PCOS association, which consisted of no proxies and a single palindrome. The F-statistic of all IVs across the exposures was > 10, indicating no substantial weak instrument bias.
2SMR results
Results from 2SMR analyses are presented in Table 2. Four exposures, namely, weight, BMI, WC, and HC, were significantly causally associated with PCOS.
According to the IVW method, a unit (1 kg) increase in weight was associated with a significant increase in PCOS (OR = 2.69; 95% CI = 1.04, 6.95; p = 0.0413). As per the MRE model, a unit increase in weight yielded an OR of 77.05 (95% CI = 3.14, 1890.64; p = 0.0260).
According to the IVW method, a unit (1 kg/m2) increase in BMI was associated with a significant increase in PCOS (OR = 4.06; 95% CI = 2.29, 7.22; p = 1.68E-06). As per the WME model, a unit increase in BMI yielded an OR of 2.90 (95% CI = 1.22, 6.90; p = 0.0162).
The IVW method revealed that WC was significantly associated with a higher risk of PCOS (OR = 6.22; 95% CI = 2.25, 17.16; p = 0.0004). As per the WME method, a unit (1cm) increase in WC significantly increased PCOS risk (OR = 11.98; 95% CI = 3.30, 43.56; p = 0.0002). According to the WMO method, a unit increase in WC conveyed an OR of 20.27 (95% CI = 2.32, 177.49; p = 0.0152).
As revealed by the IVW method, a unit (1cm) increase in HC was significantly associated with a higher risk of PCOS (OR = 6.22; 95% CI = 2.25, 17.16; p = 0.0004). According to the WME method, a unit increase in HC was associated with a significant increase in PCOS (OR = 11.98; 95% CI = 3.35, 42.81; p = 0.0001). As per the WMO method, a unit increase in HC conveyed an OR of 20.27 (95% CI = 2.17, 189.65; p = 0.0179).
However, MRE regression estimates did not achieve statistical significance with BMI, WC, and HC as exposures. Also, height and WHR were not significantly causally associated with PCOS in any 2SMR analyses.
Method comparison plots
As shown in Figs 1–4, scatter plots and trend lines pertaining to the four significant exposures (weight, BMI, WC, HC) demonstrate positive causal associations with PCOS, according to different MR methods. Scatter plots and regression lines for the two non-significant exposures (height, WHR) are presented in S1 and S2 Figs.
Trend lines from the four different two-sample Mendelian randomization methods employed indicating the positive causal associations, are also included in each scatter plot.
Trend lines from the four different two-sample Mendelian randomization methods employed indicating the positive causal associations, are also included in each scatter plot.
Trend lines from the four different two-sample Mendelian randomization methods employed indicating the positive causal associations, are also included in each scatter plot.
Trend lines from the four different two-sample Mendelian randomization methods employed indicating the positive causal associations, are also included in each scatter plot.
Single SNP analyses
Results from single SNP analyses are provided in S3 Table. Forest plots depicting single SNP analyses and pooled causal estimates as per IVW and MRE methods for significant exposures are presented in Figs 5–8 while corresponding forest plots for non-significant exposures are given in S3 and S4 Figs. According to multiple testing corrected p-value thresholds, a single SNP (rs2867131; p = 0.00394; p < 4.54 × 10–3, 0.05/11) was individually associated with PCOS in the weight data. None of the SNPs were individually associated with PCOS in all other exposure data: height (p > 9.80 × 10−4, 0.05/51); BMI (p > 1.35 × 10−3; 0.05/37); WC and HC (p > 2.94 × 10−3; 0.05/17); WHR (p > 2.17 × 10−3; 0.05/23).
Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
Forest plot of hip circumference, against polycystic ovarian syndrome as the outcome. Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
Leave-one-out analyses
Results from leave-one-out analyses are presented in S4 Table. Leave-one-out analysis plots and pooled causal estimates as per IVW-MR method (for comparison) for significant exposures are presented in Figs 9–12 while corresponding leave-one-out analysis plots for non-significant exposures are provided in S5 and S6 Figs. According to Bonferroni multiple testing corrected p-value thresholds, on exclusion of alternating SNPs, statistical significance was still retained in three exposure datasets, indicating the robustness of their causal associations, despite the exclusion of any single SNP: BMI (p < 1.35 × 10–3; 0.05/37); WC and HC (p < 2.94 × 10–3; 0.05/17). Leave-one-out analyses of other anthropometric traits did not achieve statistical significance: weight (p > 4.54 × 10−3, 0.05/11); height (p > 9.80 × 10−4, 0.05/51); WHR (p > 2.17 × 10−3; 0.05/23).
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs via IVW-MR method (drawn for comparison).
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs via IVW-MR method (drawn for comparison).
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs via IVW-MR method (drawn for comparison).
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs via IVW-MR method (drawn for comparison).
Heterogeneity analyses
Results from heterogeneity analyses (Cochran’s Q- and p-values) are summarized in Table 3. We did not observe statistically significant heterogeneity in any 2SMR analyses (p > 0.05). Funnel plots for the significant exposures are illustrated in Figs 13–16 while those for non-significant exposures are provided in S7 and S8 Figs. In conformity with non-significant findings from heterogeneity analyses, asymmetric distributions indicating directional horizontal pleiotropy or large spreads suggesting considerable heterogeneity could not be discerned in funnel plots.
Horizontal pleiotropy and outliers
We present the results of horizontal pleiotropy analyses via MRE regression intercepts and directionality p-values in Table 4. These analyses revealed that there was no significant horizontal pleiotropy (p > 0.05). As shown in S5 Table, congruent findings were obtained from MRPRESSO analyses which confirmed the absence of outliers and no significant horizontal pleiotropy (global test p-values > 0.05). Radial plots for significant exposures are provided in Figs 17–20 while those for non-significant exposures are given in S9 and S10 Figs. All radial plots illustrate the absence of outliers.
No significant outliers were detected.
No significant outliers were detected.
No significant outliers were detected.
No significant outliers were detected.
Genetic architectures underlying causality
Table 5 presents SNPs and nearest genes/TSSs underlying the four significant causal associations: weight, BMI, WC and HC versus PCOS. Notably, both WC and HC had the same 17 SNPs underlying their significant causal associations with PCOS. There were 11 SNPs underlying the significant causal association between weight and PCOS and 37 SNPs underlying the association between BMI and PCOS. Two common SNPs (rs10938397, rs543874) were observed across these four associations. The number of genes underlying the associations of weight, BMI, and WC/HC versus PCOS were 11, 40, and 21, respectively. Of these, 5 genes (GNPDA2, MC4R, SEC16B, TMEM18, ZFP64) were shared across the four significant associations. Of the 11 neighboring genes underlying the weight-PCOS association, only 4 (H2BC7, HMGA2, MTIF3, ZBTB38) were exclusive/non-shared. Of the 40 neighboring genes underlying the BMI-PCOS association, only 19 (ADCY3, AMPD2, BEND5, ETV5, FOXG1, FTO, GPRC5B, HNF4G, LMX1B, MTCH2, NPC1, PDK4, PTBP2, RASA2, SKOR1, STK33, TLR4, UHRF1BP1, ZZZ3) were exclusive/non-shared. Of the 21 neighboring genes underlying the associations between WC/HC and PCOS, only 2 (GPR61, VRK2) were exclusive.
Multivariable MR findings
Results of multivariable MR analyses are summarized in Table 6. No significant pleiotropic effects were observed in all multivariable MR models, as indicated by non-significant heterogeneity statistic (p > 0.05). Of the seven models assessed, only two models containing exclusively the anthropometric markers of central obesity produced significant direct causal effects: multivariable MR model 2 containing HC and WC as exposures and multivariable MR model 7 containing HC, WC, and WHR as exposures. In both cases, HC and WC emerged significant: in multivariable MR model 2, β = 1.827 and p = 0.00042 for both WC and HC; in multivariable MR model 7, β = 2.001 and p = 0.001, for both WC and HC. Other five multivariable models which contained surrogate anthropometric markers of both central- and general- obesity as exposures demonstrated no statistical significance, indicating lack of direct causal effects on PCOS in these models. Clinical implications of these findings alluding to the complexity of causal links between central- and general- obesity versus PCOS have been elaborated in Discussion.
Bidirectional MR findings
Bidirectional MR analyses taking PCOS as the exposure and each anthropometric trait as the outcome retrieved only a single functional variant/SNP (rs1531788) against the specified criteria: genome-wide significance threshold p < 5e – 08; clumping LD-R2 < 0.001; clumping distance > 10000 kb; proxy SNPs via LD-tagging with minimum LD-R2 > 0.8; MAF threshold for aligning palindromic SNPs = 0.3. Therefore, we could not reliably assess causal associations in the opposite direction using our data sources. In Discussion section, we contextualized findings from previous bidirectional MR studies which had evaluated the reverse causality between obesity and PCOS.
Discussion
To our knowledge, this is the first MR study to comprehensively and systematically analyze total and direct causal associations between surrogate anthropometric markers of both general- and central- obesity and PCOS, using large-sample, female-only cohorts of European ancestry. Since none of the previous studies assessing causality between obesity and PCOS conducted multivariable MR analyses to shed light on direct causal effects [25–28], our study offers novel and more profound insights, compared to prior research. We further report shared and non-shared genetic architectures with respect to potentially causal anthropometric traits in ensuing PCOS.
Findings from univariable 2SMR models: Total causal effects
We found that both central- and general- obesity potentially yield overall, total causal effects on PCOS, as signified by the four significant anthropometric markers–weight, BMI, WC, and HC, as per univariable 2SMR models. With respect to general obesity, our findings are congruent with previous 2SMR analyses which reported positive total causal effects of BMI on PCOS [25–28]. With regard to central obesity also, we observed consistently positive total causal effects of WC and HC on PCOS, as indicated by significant causal estimates across a number of univariable 2SMR models (OR range: 6.22–20.27). With regard to all four markers–weight, BMI, WC, and HC, we observed significance from more than a single 2SMR model, which was reassuring. Specifically, the standard IVW method as well as one or more robust methods (MRE, WME, WMO) demonstrated statistical significance, providing consistent evidence on significant total causal effects of both central- and general obesity in ensuing PCOS (Table 2).
We ensured minimal overlap between exposure-outcome samples–a key requirement for reducing bias in 2SMR estimates [14]. In fact, non-substantial sample overlaps, especially involving controls, are even permissible, and would not evoke bias in 2SMR studies [53]. No confounding by ancestry entailed as all participants were of European ancestry. However, this may have led to a lack of generalizability of our findings and transethnic validations via MR in independent cohorts would be required. Further, there was no weak instrument bias as all IVs had F > 10. Robustness of our findings was underscored by a range of sensitivity analyses, which uncovered no substantial heterogeneity, horizontal pleiotropy, and outliers. While the use of summarized public data has advantages such as transparency, reproducibility, and the greater ability to detect causal associations due to large samples, it has a few shortcomings as well. For example, we could not conduct subgroup analyses, rule out collider bias due to over-adjusted summarized association estimates, or assess potentially non-linear associations of obesity with PCOS, for which individual-level data are required [38]. The relatively small number of PCOS cases in the outcome GWAS (N = 642) might be a limiting factor, given the community prevalence of PCOS > 8% [2], as opposed to the substantially lower prevalence of clinically-diagnosed cases (~0.54%) in the Finnish cohort used in the present study. However, these are women with overt PCOS presenting to the clinicians with common, cardinal clinical features including obesity and predominant PCOS phenotypes, as confirmed by ICD codes. As we uncovered potential causalities despite the small number of cases, stronger findings with larger causal estimates might possibly be obtained from independent cohorts containing a larger number of women with PCOS, which need to be investigated in future studies. For future studies, we suggest meta-analyzing several cohorts from sources such as the UK Biobank and FinnGen as a judicious avenue for formidably increasing the sample size. Although we minimized confounding with female-only cohorts of European ancestry and summary estimates adjusted for multiple variables, there may still have been residual confounding influencing our causal estimates toward or away from the null.
Genetics underlying causality
We observed a considerable degree of overlap across significant causal findings, perhaps indicating similar genetically-driven causal mechanisms. However, there were SNPs and genes that were exclusive to each causal association as well, suggesting unique, exposure-specific pathways may also participate in PCOS pathogenesis. This suggests that differences may exist in the genetic basis of PCOS among women with central obesity alone (normal weight-central obesity), as compared to women with both general- and central obesity. Of the identified genes in our study, FTO, GNPDA2, MC4R, MTCH2, NEGR1, SH2B1, TMEM18 are unequivocally regarded as obesity genes associated with PCOS, as supported by contemporary evidence [54]. A gene set rather similar to those found in the present study (ADCY3, ATP2A1, BCDIN3D, BDNF, CADM2, ETV5, FTO, GNPDA2, GPRC5B, HNF4G, LINGO2, LMX1B, MC4R, MTCH2, NEGR1, POC5, PTBP2, RASA2, SEC16B, TFAP2B, TLR4, TMEM18, ZC3H4, ZFP64) was underlying the causal association between BMI and PCOS in a previous MR study as well [25].
Mechanisms underlying the causal effects of central-and general obesity on PCOS are not fully understood and need to be investigated. Towards this end, the SNPs and neighboring genes underlying causal associations uncovered by significant 2SMR models in the present study may be insightful. In S6 Table, we have presented congruent evidence from other studies supporting for putative mechanistic roles of these shared and non-shared genes underlying the causal associations between anthropometric markers of central- and/or general obesity and PCOS. However, it was beyond the scope of the current study to explore pathophysiological functions of these genes in the development of PCOS, and we recommend their putative pathobiological relevance for PCOS is explored in future experiments.
There is a formidable body of evidence supporting FTO gene’s association with both obesity and PCOS. While FTO is, in fact, the first locus identified as unequivocally associated with adiposity [55], a meta-analysis revealed a direct association between FTO variant and PCOS risk, independent of BMI [56]. Both FTO and MC4R gene variants are associated with obesity in PCOS [57] while observational evidence indicates a direct role of the interaction between FTO and MC4R polymorphisms in the development of PCOS [58]. TLR4 and toll-like receptor genes in general contribute to the development of chronic low-grade inflammation as well as insulin resistance and hyperandrogenism observed in PCOS [59] whereas a MR study revealed essential causal roles of systemic inflammatory regulators, especially cytokines, in the pathogenesis of PCOS [60]. However, mechanisms by which certain other genes found to be underlying the causal associations in the present study contribute to the development of PCOS, are not completely known.
Current evidence indicates that obesity contributes to the onset and progression of PCOS through multiple mechanisms including the escalation of insulin resistance and compensatory hyperinsulinemia which in turn leads to augmented adipogenesis and reduced lipolysis, sensitization of thecal cells to luteinizing hormone and intensification of functional ovarian hyperandrogenism by increasing ovarian androgen synthesis, and elevation of inflammatory adipokines which in turn leads to upregulation of insulin resistance and adipogenesis [19]. Indeed, obesity contributes to the pathogenesis of PCOS via multiple mechanisms that encompass the three cardinal clinical manifestations of PCOS, namely, hyperandrogenism, reproductive-, and metabolic dysfunction [61]. Moreover, our findings allude to the concept of secondary PCOS, which has only emerged in recent years. It has been proposed that adiposity and hyperinsulinemia (exogenous in type 1- and endogenous in type 2- diabetes) cause PCOS [62]. Recent research on polygenic risks scores has also demonstrated that both men and women manifest similar metabolic features of PCOS and that the presence of ovaries is neither essential nor required for PCOS while obesity features prominently [63]. Genetic research is providing greater insights into the mechanisms and nature of this complex polygenic disorder.
Given the differences in hormonal and metabolic profiles, treatment strategies and outcomes between obese- and non-obese PCOS phenotypes, it has been proposed to incorporate obesity as a clinical sign for classifying PCOS phenotypes [64]. However, lean PCOS phenotypes which are less prevalent than obese or overweight PCOS phenotypes, may have different underlying genetic architectures and pathogenic mechanisms and may need to be investigated separately.
Findings of multivariable MR models and their implications
Results from multivariable MR modelling allude to the complex nature of causal associations between obesity and PCOS. As suggested by the two significant multivariable MR models which comprised of only the anthropometric markers of central obesity, it is possible that in the presence of normal-weight central obesity, direct causal effects are exerted on the onset of PCOS. However, in the presence of both central- and general- obesity, more complex causal mechanisms seem to be at play, as indicated by all corresponding, non-significant multivariable MR models. These findings strongly suggest the possible existence of intricate phenomena such as causal mediation when both central- and general obesity are present. For instance, a separate study which found total causal effects of both sex hormone-binding globulin and testosterone on coronary heart disease, but no direct causal effects, led to the discovery of causal mediations of both these exposures on coronary heart disease via subsequent causal network MR modelling [65]. Although we did not perform network MR, such downstream analyses will be useful to dissect complex causal mechanisms between obesity and PCOS. Other potential phenomena such as nonlinear causality and causal interactions should also be assessed in future studies with prudent study designs such as nonlinear MR [66] and factorial MR [67], respectively. Lastly, other caveats such as causality between traits and high polygenicity of traits should also be assessed in future studies, to fully disentangle complex causal mechanisms between obesity and PCOS.
Bidirectional associations/reverse causality
For bidirectional 2SMR analyses, sufficient number of IVs could not be retrieved from existing data sources, but, with the gradual expansion of these open-source GWAS summary data, we envisage that it will be possible to conduct bidirectional analyses more robustly, in the future. Stringent eligibility criteria to select only highly uncorrelated and independent SNPs (LD-R2 = 0.001) may also have contributed to the lack of functional variants produced for assessing reverse causality. However, concordant evidence from previous bidirectional analyses provides reassurance that there is no reverse causality i.e. PCOS would not cause obesity [25, 26].
Clinical and public health value of our findings
Our findings indicating complex causal associations of obesity and adiposity with PCOS have clinical and public health implications. This analysis reinforced findings of previous studies indicating total causal effects of general obesity on PCOS [25–28]. In addition, this study offered novel results such as the total causal effects of central obesity on PCOS, potentially direct causal effects of normal weight-central obesity on PCOS, and that complex causal mechanisms in the presence of both general- and central obesity might lead to PCOS.
Of note, although our analyses indicated possibly direct causal effects of normal weight-central obesity on PCOS, people with central obesity but of normal weight are generally ignored in clinical guidelines [68]. Therefore, women with normal weight-central obesity will require an equally important focus as attributed to general obesity phenotype, in clinical guidelines for PCOS management. To this end, Mediterranean style dietary patterns and physical activity both are evidence-based interventions to reduce central obesity and ensure weight stabilization [69].
Causality of central obesity/adiposity demarcated by WC/HC in the present study corroborate findings from mechanistic studies focused on PCOS etiology. Truncal/visceral obesity promotes insulin resistance which in turn aggravates metabolic and reproductive functions in PCOS such as hyperandrogenism, dyslipidemia, and anovulation [70]. Evidence suggests the formation of a vicious cycle by central obesity which involves hyperandrogenemia and truncal adiposity in PCOS [71].
Non-significance of WHR might perhaps be owing to its lesser validity and reliability as an anthropometric marker of central obesity, compared to WC/HC, as concluded by previous studies [72, 73]. Lack of adequate statistical power in MR models to uncover causal associations from open-source data might be another contributory factor. This may be verified in future studies using larger cohorts.
On the whole, findings of this study indicating complex causality of central- and general obesity on PCOS, highlight the need to address the prevention of both forms of obesity at a population level through successful fiscal policies and statutory regulations [74]. Our findings are also congruent with recommendations from current international guidelines on PCOS, which underscore the prevention of excess weight gain in all women with PCOS, alongside lifestyle interventions focused on diet, weight loss, and physical activity for women with PCOS with overweight and obesity. As in the standard clinical management of obesity, pharmacotherapy and other treatments may also be appropriate [19].
Conclusions
In this study, we revealed that both central- and general obesity yield total causal effects on PCOS, via 2SMR analyses using female-only, large-sample cohorts of European ancestry. Furthermore, it demonstrated potential direct causal effects of normal weight-central obesity on PCOS and complex causal mechanisms associated with PCOS in the presence of both central- and general obesity. Findings warrant further studies to explore genetic mechanisms underpinning causal associations of obesity with this common and complex condition. They also support international guidelines and underscore the importance of addressing both central- and general obesity phenotypes and adiposity for the prevention and management of PCOS.
Supporting information
S1 Fig. Scatter plot illustrating the distribution of individual ratio estimates of height as the exposure, with polycystic ovarian syndrome as the outcome.
Trend lines from the four different two-sample Mendelian randomization methods employed, are also included in each scatter plot.
https://doi.org/10.1371/journal.pone.0269191.s001
(TIF)
S2 Fig. Scatter plot illustrating the distribution of individual ratio estimates of waist-to-hip ratio as the exposure, with polycystic ovarian syndrome as the outcome.
Trend lines from the four different two-sample Mendelian randomization methods employed, are also included in each scatter plot.
https://doi.org/10.1371/journal.pone.0269191.s002
(TIF)
S3 Fig. Forest plot of height as the exposure, against polycystic ovarian syndrome as the outcome.
Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
https://doi.org/10.1371/journal.pone.0269191.s003
(TIF)
S4 Fig. Forest plot of waist-to-hip ratio as the exposure, against polycystic ovarian syndrome as the outcome.
Effects of individual SNPs and pooled estimates from MR-Egger- and inverse variance weighted methods are visualized.
https://doi.org/10.1371/journal.pone.0269191.s004
(TIF)
S5 Fig. Leave-one-out sensitivity analysis plot of height as the exposure, against polycystic ovarian syndrome as the outcome.
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs.
https://doi.org/10.1371/journal.pone.0269191.s005
(TIF)
S6 Fig. Leave-one-out sensitivity analysis of the plot of waist-to-hip ratio as the exposure, against polycystic ovarian syndrome as the outcome.
A given dark point indicates the effect measure from inverse variance weighted Mendelian randomization analysis excluding that specific SNP. The red lines indicate pooled analyses encompassing all SNPs.
https://doi.org/10.1371/journal.pone.0269191.s006
(TIF)
S7 Fig. Funnel plot of height as the exposure against polycystic ovarian syndrome as the outcome.
https://doi.org/10.1371/journal.pone.0269191.s007
(TIF)
S8 Fig. Funnel plot of waist-to-hip ratio as the exposure against polycystic ovarian syndrome as the outcome.
https://doi.org/10.1371/journal.pone.0269191.s008
(TIF)
S9 Fig. Radial plot of height as the exposure.
No significant outliers were detected.
https://doi.org/10.1371/journal.pone.0269191.s009
(TIF)
S10 Fig. Radial plot as waist-to-hip ratio as the exposure.
No significant outliers were detected.
https://doi.org/10.1371/journal.pone.0269191.s010
(TIF)
S6 Table. Congruent evidence from other studies supporting for putative mechanistic roles of the shared and non-shared genes underlying the causal associations between anthropometric markers of central—and/or general obesity and PCOS.
https://doi.org/10.1371/journal.pone.0269191.s016
(DOCX)
References
- 1. Anagnostis P, Tarlatzis BC, Kauffman RP. Polycystic ovarian syndrome (PCOS): Long-term metabolic consequences. Metabolism. 2018;86:33–43. https://doi.org/10.1016/j.metabol.2017.09.016 pmid:29024702
- 2. Raperport C, Homburg R. The Source of Polycystic Ovarian Syndrome. Clin Med Insights Reprod Health. 2019;13:1179558119871467. https://doi.org/10.1177/1179558119871467 pmid:31516315
- 3. Moore AM, Prescott M, Marshall CJ, Yip SH, Campbell RE. Enhancement of a robust arcuate GABAergic input to gonadotropin-releasing hormone neurons in a model of polycystic ovarian syndrome. Proc Natl Acad Sci U S A. 2015;112(2):596–601. https://doi.org/10.1073/pnas.1415038112 pmid:25550522
- 4. Palomba S, Santagni S, Falbo A, La Sala GB. Complications and challenges associated with polycystic ovary syndrome: current perspectives. Int J Womens Health. 2015;7:745–63. https://doi.org/10.2147/IJWH.S70314 pmid:26261426
- 5. Zeng X, Xie YJ, Liu YT, Long SL, Mo ZC. Polycystic ovarian syndrome: Correlation between hyperandrogenism, insulin resistance and obesity. Clin Chim Acta. 2020;502:214–221. https://doi.org/10.1016/j.cca.2019.11.003 pmid:31733195
- 6. Hayes MG, Urbanek M, Ehrmann DA, Armstrong LL, Lee JY, Sisk R, et al. Genome-wide association of polycystic ovary syndrome implicates alterations in gonadotropin secretion in European ancestry populations. Nat Commun. 2015;6:7502. https://doi.org/10.1038/ncomms8502 pmid:26284813
- 7. Gur EB, Karadeniz M, Turan GA. Fetal programming of polycystic ovary syndrome. World J Diabetes. 2015;6(7):936–42. pmid:26185601
- 8. Azziz R. Introduction: Determinants of polycystic ovary syndrome. Fertil Steril. 2016;106(1):4–5. https://doi.org/10.1016/j.fertnstert.2016.05.009 pmid:27238627
- 9. Merkin SS, Phy JL, Sites CK, Yang D. Environmental determinants of polycystic ovary syndrome. Fertil Steril. 2016;106(1):16–24. https://doi.org/10.1016/j.fertnstert.2016.05.011 pmid:27240194
- 10. Abraham Gnanadass S, Divakar Prabhu Y, Valsala Gopalakrishnan A. Association of metabolic and inflammatory markers with polycystic ovarian syndrome (PCOS): an update. Arch Gynecol Obstet. 2021;303(3):631–643. https://doi.org/10.1007/s00404-020-05951-2 pmid:33439300
- 11. Bates GW, Legro RS. Longterm management of Polycystic Ovarian Syndrome (PCOS). Mol Cell Endocrinol. 2013;373(1–2):91–7. https://doi.org/10.1016/j.mce.2012.10.029 pmid:23261983
- 12. Peña AS, Teede H, Hewawasam E, Hull ML, Gibson-Helm M. Diagnosis experiences of adolescents with polycystic ovary syndrome: Cross-sectional study. Clin Endocrinol (Oxf). 2022;96(1):62–69. https://doi.org/10.1111/cen.14604 pmid:34585425
- 13. Neumeyer S, Hemani G, Zeggini E. Strengthening Causal Inference for Complex Disease Using Molecular Quantitative Trait Loci. Trends Mol Med. 2020;26(2):232–241. https://doi.org/10.1016/j.molmed.2019.10.004 pmid:31718940
- 14. Davies NM, Holmes MV, Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ. 2018;362:k601. https://doi.org/10.1136/bmj.k601 pmid:30002074
- 15. Skrivankova VW, Richmond RC, Woolf BAR, Yarmolinsky J, Davies NM, Swanson SA, et al. Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization: The STROBE-MR Statement. JAMA. 2021;326(16):1614–1621. pmid:34698778
- 16. Dadachanji R, Patil A, Joshi B, Mukherjee S. Elucidating the impact of obesity on hormonal and metabolic perturbations in polycystic ovary syndrome phenotypes in Indian women. PLoS One. 2021;16(2):e0246862. https://doi.org/10.1371/journal.pone.0246862 pmid:33635862
- 17. Barber TM, Hanson P, Weickert MO, Franks S. Obesity and Polycystic Ovary Syndrome: Implications for Pathogenesis and Novel Management Strategies. Clin Med Insights Reprod Health. 2019;13:1179558119874042. https://doi.org/10.1177/1179558119874042 pmid:31523137
- 18. de Zegher F, López-Bermejo A, Ibáñez L. Central Obesity, Faster Maturation, and ’PCOS’ in Girls. Trends Endocrinol Metab. 2018;29(12):815–818. https://doi.org/10.1016/j.tem.2018.09.005 pmid:30297320
- 19. Glueck CJ, Goldenberg N. Characteristics of obesity in polycystic ovary syndrome: Etiology, treatment, and genetics. Metabolism. 2019;92:108–120. https://doi.org/10.1016/j.metabol.2018.11.002 pmid:30445140
- 20. Khan MJ, Ullah A, Basit S. Genetic Basis of Polycystic Ovary Syndrome (PCOS): Current Perspectives. Appl Clin Genet. 2019;12:249–260. https://doi.org/10.2147/TACG.S200341 pmid:31920361
- 21. Kokosar M, Benrick A, Perfilyev A, Fornes R, Nilsson E, Maliqueo M, et al. Epigenetic and Transcriptional Alterations in Human Adipose Tissue of Polycystic Ovary Syndrome. Sci Rep. 2016;6:22883. https://doi.org/10.1038/srep22883 pmid:26975253
- 22. Liu AL, Xie HJ, Xie HY, Liu J, Yin J, Hu JS, et al. Association between fat mass and obesity associated (FTO) gene rs9939609 A/T polymorphism and polycystic ovary syndrome: a systematic review and meta-analysis. BMC Med Genet. 2017;18(1):89. https://doi.org/10.1186/s12881-017-0452-1 pmid:28826396
- 23. Wang X, Wang K, Yan J, Wu M. A meta-analysis on associations of FTO, MTHFR and TCF7L2 polymorphisms with polycystic ovary syndrome. Genomics. 2020;112(2):1516–1521. https://doi.org/10.1016/j.ygeno.2019.08.023 pmid:31470081
- 24. Toosy S, Sodi R, Pappachan JM. Lean polycystic ovary syndrome (PCOS): an evidence-based practical approach. J Diabetes Metab Disord. 2018;17(2):277–285. https://doi.org/10.1007/s40200-018-0371-5 pmid:30918863
- 25. Brower MA, Hai Y, Jones MR, Guo X, Chen YI, Rotter JI, et al. Bidirectional Mendelian randomization to explore the causal relationships between body mass index and polycystic ovary syndrome. Hum Reprod. 2019;34(1):127–136. https://doi.org/10.1093/humrep/dey343 pmid:30496407
- 26. Zhao Y, Xu Y, Wang X, Xu L, Chen J, Gao C, et al. Body Mass Index and Polycystic Ovary Syndrome: A 2-Sample Bidirectional Mendelian Randomization Study. J Clin Endocrinol Metab. 2020;105(6):dgaa125. https://doi.org/10.1210/clinem/dgaa125 pmid:32163573
- 27. Day FR, Hinds DA, Tung JY, Stolk L, Styrkarsdottir U, Saxena R, et al. Causal mechanisms and balancing selection inferred from genetic associations with polycystic ovary syndrome. Nat Commun. 2015;6:8464. https://doi.org/10.1038/ncomms9464 pmid:26416764
- 28. Day F, Karaderi T, Jones MR, Meun C, He C, Drong A, et al. Large-scale genome-wide meta-analysis of polycystic ovary syndrome suggests shared genetic architecture for different diagnosis criteria. PLoS Genet. 2018;14(12):e1007813. https://doi.org/10.1371/journal.pgen.1007813 pmid:30566500
- 29. Ross R, Neeland IJ, Yamashita S, Shai I, Seidell J, Magni P, et al. Waist circumference as a vital sign in clinical practice: a Consensus Statement from the IAS and ICCR Working Group on Visceral Obesity. Nat Rev Endocrinol. 2020;16(3):177–189. https://doi.org/10.1038/s41574-019-0310-7 pmid:32020062
- 30. Wang SH, Su MH, Chen CY, Lin YF, Feng YA, Hsiao PC, et al. Causality of abdominal obesity on cognition: a trans-ethnic Mendelian randomization study. Int J Obes (Lond). 2022. https://doi.org/10.1038/s41366-022-01138-8 pmid:35538205
- 31. Du D, Jing Z, Zhang G, Dang X, Liu R, Song J. The relationship between central obesity and bone mineral density: a Mendelian randomization study. Diabetol Metab Syndr. 2022;14(1):63. https://doi.org/10.1186/s13098-022-00840-x pmid:35501835
- 32. Li D, Zou Y. Causal effects of life course adiposity on chronic kidney disease: a Mendelian randomization study. Ann Palliat Med. 2021;10(10):10861–10869. pmid:34763447
- 33. Randall JC, Winkler TW, Kutalik Z, Berndt SI, Jackson AU, Monda KL, et al. Sex-stratified genome-wide association studies including 270,000 individuals show sexual dimorphism in genetic loci for anthropometric traits. PLoS Genet. 2013;9(6):e1003500. https://doi.org/10.1371/journal.pgen.1003500 pmid:23754948
- 34. Locke AE, Kahali B, Berndt SI, Justice AE, Pers TH, Day FR, et al. Genetic studies of body mass index yield new insights for obesity biology. Nature. 2015;518(7538):197–206. https://doi.org/10.1038/nature14177 pmid:25673413
- 35. Shungin D, Winkler TW, Croteau-Chonka DC, Ferreira T, Locke AE, Mägi R, et al. New genetic loci link adipose and insulin biology to body fat distribution. Nature. 2015;518(7538):187–196. https://doi.org/10.1038/nature14132 pmid:25673412
- 36. Gill D, Efstathiadou A, Cawood K, Tzoulaki I, Dehghan A. Education protects against coronary heart disease and stroke independently of cognitive function: evidence from Mendelian randomization. Int J Epidemiol. 2019;48(5):1468–1477. https://doi.org/10.1093/ije/dyz200 pmid:31562522
- 37. Burgess S, Small DS, Thompson SG. A review of instrumental variable estimators for Mendelian randomization. Stat Methods Med Res. 2017;26(5):2333–2355. https://doi.org/10.1177/0962280215597579 pmid:26282889
- 38. Burgess S, Davey Smith G, Davies NM, Dudbridge F, Gill D, Glymour MM, et al. Guidelines for performing Mendelian randomization investigations. Wellcome Open Res. 2020;4:186. https://doi.org/10.12688/wellcomeopenres.15555.2 pmid:32760811
- 39. Burgess S, Thompson SG. Interpreting findings from Mendelian randomization using the MR-Egger method. Eur J Epidemiol. 2017;32(5):377–389. https://doi.org/10.1007/s10654-017-0255-x pmid:28527048
- 40. Burgess S, Bowden J, Fall T, Ingelsson E, Thompson SG. Sensitivity Analyses for Robust Causal Inference from Mendelian Randomization Analyses with Multiple Genetic Variants. Epidemiology. 2017;28(1):30–42. pmid:27749700
- 41. Hwang LD, Lawlor DA, Freathy RM, Evans DM, Warrington NM. Using a two-sample Mendelian randomization design to investigate a possible causal effect of maternal lipid concentrations on offspring birth weight. Int J Epidemiol. 2019;48(5):1457–1467. https://doi.org/10.1093/ije/dyz160 pmid:31335958
- 42. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693–698. https://doi.org/10.1038/s41588-018-0099-7 pmid:29686387
- 43. Bowden J, Spiller W, Del Greco M F, Sheehan N, Thompson J, Minelli C, et al. Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression. Int J Epidemiol. 2018;47(4):1264–1278. https://doi.org/10.1093/ije/dyy101 pmid:29961852
- 44.
R Core Team (2022). R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://www.R-project.org/
- 45. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife. 2018;7:e34408. https://doi.org/10.7554/eLife.34408.001 pmid:29846171
- 46. Harbaum L, Hennigs JK, Simon M, Oqueka T, Watz H, Klose H. Genetic evidence for a causative effect of airflow obstruction on left ventricular filling: a Mendelian randomisation study. Respir Res. 2021;22(1):199. https://doi.org/10.1186/s12931-021-01795-9 pmid:34233669
- 47. Pan W, Sun W, Yang S, Zhuang H, Jiang H, Ju H, et al. LDL-C plays a causal role on T2DM: a Mendelian randomization analysis. Aging (Albany NY). 2020;12(3):2584–2594. https://doi.org/10.18632/aging.102763
- 48. Ghoussaini M, Mountjoy E, Carmona M, Peat G, Schmidt EM, Hercules A, et al. Open Targets Genetics: systematic identification of trait-associated genes using large-scale genetics and functional genomics. Nucleic Acids Res. 2021;49(D1):D1311–D1320. https://doi.org/10.1093/nar/gkaa840 pmid:33045747
- 49. Sanderson E, Davey Smith G, Windmeijer F, Bowden J. An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. Int J Epidemiol. 2019;48(3):713–727. https://doi.org/10.1093/ije/dyy262 pmid:30535378
- 50. Burgess S, Thompson SG. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181(4):251–60. https://doi.org/10.1093/aje/kwu283 pmid:25632051
- 51. Broadbent JR, Foley CN, Grant AJ, Mason AM, Staley JR, Burgess S. MendelianRandomization v0.5.0: updates to an R package for performing Mendelian randomization analyses using summarized data. Wellcome Open Res. 2020;5:252. https://doi.org/10.12688/wellcomeopenres.16374.2 pmid:33381656
- 52. Yavorska OO, Burgess S. MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data. Int J Epidemiol. 2017;46(6):1734–1739. https://doi.org/10.1093/ije/dyx034 pmid:28398548
- 53. Yoshikawa M, Asaba K. Educational Attainment Decreases the Risk of COVID-19 Severity in the European Population: A Two-Sample Mendelian Randomization Study. Front Public Health. 2021;9:673451. https://doi.org/10.3389/fpubh.2021.673451 pmid:34150709
- 54. Bogari NM. Genetic construction between polycystic ovarian syndrome and type 2 diabetes. Saudi J Biol Sci. 2020;27(10):2539–2543. https://doi.org/10.1016/j.sjbs.2020.05.004 pmid:32994709
- 55. Fawcett KA, Barroso I. The genetics of obesity: FTO leads the way. Trends Genet. 2010;26(6):266–74. https://doi.org/10.1016/j.tig.2010.02.006 pmid:20381893
- 56. Cai X, Liu C, Mou S. Association between fat mass- and obesity-associated (FTO) gene polymorphism and polycystic ovary syndrome: a meta-analysis. PLoS One. 2014;9(1):e86972. https://doi.org/10.1371/journal.pone.0086972 pmid:24466303
- 57. Ewens KG, Jones MR, Ankener W, Stewart DR, Urbanek M, Dunaif A, et al. FTO and MC4R gene variants are associated with obesity in polycystic ovary syndrome. PLoS One. 2011;6(1):e16390. https://doi.org/10.1371/journal.pone.0016390 pmid:21283731
- 58. Yuan H, Zhu G, Wang F, Wang X, Guo H, Shen M. Interaction between common variants of FTO and MC4R is associated with risk of PCOS. Reprod Biol Endocrinol. 2015;13:55. https://doi.org/10.1186/s12958-015-0050-z pmid:26032905
- 59. Sathyapalan T, Atkin SL. Mediators of inflammation in polycystic ovary syndrome in relation to adiposity. Mediators Inflamm. 2010;2010:758656. https://doi.org/10.1155/2010/758656 pmid:20396393
- 60. Chen H, Zhang Y, Li S, Tao Y, Gao R, Xu W, et al. The Association Between Genetically Predicted Systemic Inflammatory Regulators and Polycystic Ovary Syndrome: A Mendelian Randomization Study. Front Endocrinol (Lausanne). 2021;12:731569. https://doi.org/10.3389/fendo.2021.731569 pmid:34646235
- 61. Barber TM, Franks S. Obesity and polycystic ovary syndrome. Clin Endocrinol (Oxf). 2021;95(4):531–541. https://doi.org/10.1111/cen.14421 pmid:33460482
- 62. Thong EP, Codner E, Laven JSE, Teede H. Diabetes: a metabolic and reproductive disorder in women. Lancet Diabetes Endocrinol. 2020;8(2):134–149. https://doi.org/10.1016/S2213-8587(19)30345-6 pmid:31635966
- 63. Zhu J, Pujol-Gualdo N, Wittemans LBL, Lindgren CM, Laisk T, Hirschhorn JN, et al. Evidence From Men for Ovary-independent Effects of Genetic Risk Factors for Polycystic Ovary Syndrome. J Clin Endocrinol Metab. 2021:dgab838. https://doi.org/10.1210/clinem/dgab838
- 64. Moran C, Arriaga M, Rodriguez G, Moran S. Obesity differentially affects phenotypes of polycystic ovary syndrome. Int J Endocrinol. 2012;2012:317241. https://doi.org/10.1155/2012/317241 pmid:22829818
- 65. Li Y, Si S, Hou L, Yuan T, Chen X, Liu C, et al. Causal effect of sex hormone-binding globulin and testosterone on coronary heart disease: A multivariable and network Mendelian randomization analysis. Int J Cardiol. 2021;339:179–184. https://doi.org/10.1016/j.ijcard.2021.06.037 pmid:34181993
- 66. Staley JR, Burgess S. Semiparametric methods for estimation of a nonlinear exposure-outcome relationship using instrumental variables with application to Mendelian randomization. Genet Epidemiol. 2017;41(4):341–352. https://doi.org/10.1002/gepi.22041 pmid:28317167
- 67. Rees JMB, Foley CN, Burgess S. Factorial Mendelian randomization: using genetic variants to assess interactions. Int J Epidemiol. 2020;49(4):1147–1158. https://doi.org/10.1093/ije/dyz161 pmid:31369124
- 68. Sun Y, Liu B, Snetselaar LG, Wallace RB, Caan BJ, Rohan TE, et al. Association of Normal-Weight Central Obesity With All-Cause and Cause-Specific Mortality Among Postmenopausal Women. JAMA Netw Open. 2019;2(7):e197337. pmid:31339542
- 69. Bosomworth NJ. Normal-weight central obesity: Unique hazard of the toxic waist. Can Fam Physician. 2019;65(6):399–408. pmid:31189627
- 70. Lim SS, Davies MJ, Norman RJ, Moran LJ. Overweight, obesity and central obesity in women with polycystic ovary syndrome: a systematic review and meta-analysis. Hum Reprod Update. 2012;18(6):618–37. https://doi.org/10.1093/humupd/dms030 pmid:22767467
- 71. Escobar-Morreale HF, San Millán JL. Abdominal adiposity and the polycystic ovary syndrome. Trends Endocrinol Metab. 2007;18(7):266–72. https://doi.org/10.1016/j.tem.2007.07.003 pmid:17693095
- 72. Ahmad N, Adam SI, Nawi AM, Hassan MR, Ghazi HF. Abdominal Obesity Indicators: Waist Circumference or Waist-to-hip Ratio in Malaysian Adults Population. Int J Prev Med. 2016;7:82. pmid:27330688
- 73. Dobbelsteyn CJ, Joffres MR, MacLean DR, Flowerdew G. A comparative evaluation of waist circumference, waist-to-hip ratio and body mass index as indicators of cardiovascular risk factors. The Canadian Heart Health Surveys. Int J Obes Relat Metab Disord. 2001;25(5):652–61. https://doi.org/10.1038/sj.ijo.0801582 pmid:11360147
- 74. Roberto CA, Swinburn B, Hawkes C, Huang TT, Costa SA, Ashe M, et al. Patchy progress on obesity prevention: emerging examples, entrenched barriers, and new thinking. Lancet. 2015;385(9985):2400–9. https://doi.org/10.1016/S0140-6736(14)61744-X pmid:25703111