https://orcid.org/0000-0001-6485-0221
Abstract
Observational studies have shown a link between elevated body mass index (BMI) and the risk of polycystic ovary syndrome (PCOS). While Mendelian randomization (MR) studies in Europeans have suggested a causal role of increased BMI in PCOS, whether the same role is suggested in Asians has yet to be investigated. We used MR studies to infer causal effects using genetic data from East Asian populations.
We performed a 2-sample bidirectional MR analysis using summary statistics from genome-wide association studies (GWAS) of BMI (with up to 173 430 individuals) and PCOS (4386 cases and 8017 controls) in East Asian populations. Seventy-eight single nucleotide polymorphisms (SNPs) correlated with BMI were selected as genetic instrumental variables to estimate the causal effect of BMI on PCOS using the inverse-variance weighted (IVW) method. To test the reliability of the results, further sensitivity analyses included MR–Egger regression, weighted median estimates, and leave-one-out analysis. The IVW analysis indicated a significant association between high BMI and the risk of PCOS (odds ratio per standard deviation higher BMI, 2.208; 95% confidence interval 1.537 to 3.168, P = 1.77 × 10–5). In contrast, the genetic risk of PCOS had no significant effect on BMI.
The results of our bidirectional MR study showed that an increase in BMI causes PCOS, while PCOS does not cause an increased BMI. This study provides further genetic support for a link between BMI and PCOS. Further research is needed to interpret the potential mechanisms of this association.
Polycystic ovary syndrome (PCOS) is the most common gynecological endocrine and metabolic disorder. The prevalence rate in childbearing age women is 3.4% to 13% (1). The cause of PCOS is not fully understood, but it is thought to be influenced by environmental and genetic factors. Women with PCOS are at significantly increased risk for impaired glucose tolerance, insulin resistance, dyslipidemia, and type 2 diabetes mellitus (2-4). PCOS seriously affects the reproductive ability of childbearing age women (5). Patients with PCOS are usually obese compared with healthy people (6), and the incidence of PCOS is increasingly common in obese people. In addition, lifestyle changes and weight loss can improve the characteristics of PCOS (7). In recent years, there have been some studies on the relationship between obesity genes and susceptibility to PCOS, but the results are different. FTO, a fat mass and obesity-related gene, is an important candidate gene for obesity identified by genome-wide association studies (GWAS) (8) and is considered to be associated with PCOS in some studies (9-11); other studies have shown that FTO is not associated with PCOS (12-14). A recent meta-analysis found that the FTO variant of women with PCOS had a greater impact on body mass index (BMI) than that of normal controls (15), which suggests that PCOS itself may change the impact of FTO on the BMI of women with PCOS (16).
Although PCOS seems to be closely related to obesity, it is not clear whether there is a causal relationship between them. Mendelian randomization (MR) may help to clarify this relationship (17). This approach uses genetic variants as instruments for making causal inferences and leverages the random assortment of alleles at the time of conception to overcome limitations inherent in observational studies, thus improving causal inferences. There are racial differences in the impact of BMI on PCOS (18). In recent years, several MR studies on the causal relationship between BMI and PCOS in European populations have yielded positive results (16,19,20), but no studies in East Asian populations have been reported so far.
In this study, we used 78 single nucleotide polymorphisms (SNPs) datasets for BMI to explore the causal relationship between BMI and PCOS in East Asian populations.
Methods
Study design and data sources
Two-sample MR analysis (21) was used to estimate the causal relationship between BMI and the risk of PCOS. This 2-sample MR study is based on a single-sample MR study (22). This analysis allows the exposure and results to come from 2 samples, and the data can be published aggregate analyses of genome-wide associations between the exposure and results for 2 diseases or traits (23). In the first set of 2-sample MR analyses, we used BMI as the exposure and PCOS as the outcome, and SNPs were used as instrumental variables.
The BMI data used in our study came from the Biobank Japan (BBJ) GWAS analyses with additional replications using the Japan public health center–based prospective study and Tohoku Medical Megabank Project datasets (which consist of up to 173 430 individuals), which identified 85 new BMI loci in the Japanese population (24,25). The 85 loci included 2 male-specific loci and 5 loci on the X chromosome. We excluded these 7 loci; thus, 78 BMI loci were ultimately examined. The PCOS data were from GWAS of Chinese populations, and the cohort included 4386 PCOS cases and 8017 controls (25-27). If a BMI-associated index SNP was absent in the PCOS data set, we used a proxy SNP based on the BBJ GWAS (all supplementary material and figures are located in a digital research materials repository (28)). We selected the top significant variant associated with BMI in the BBJ GWAS, which was also present in the PCOS data set, using the PLINK “clump” program (29) with disequilibrium information based on the 1000 Genomes Project East Asian data (30). Finally, the 78 obtained SNPs were used as a genetic tool variable (28). Among the 78 SNPs, 27 were from the same region (defined as within 500 kb of each other) of the BMI SNP found in the European BMI GWAS, and most of them are in linkage disequilibrium (r2 > 0.2) with each other (28).
The second group of MR analyses regarded PCOS as the exposure and BMI as the outcome. The summary results of the GWAS meta-analysis of PCOS for the Chinese population (25,26) were used as the data source for constructing PCOS instrument variables. In this study, the instrumental variables for PCOS were ultimately composed of 11 SNPs (28).
The Chinese and Japanese populations account for the majority of the total population of East Asia. Previous genetic studies of complex diseases or traits for East Asian populations usually included both Chinese and Japanese populations. For many BMI-associated loci, the associations were similar between Chinese and Japanese populations (31-33).
All original studies passed ethics review and obtained informed consent.
Mendelian randomization
The inverse-variance weighted (IVW) method is a weighted linear regression model, which aggregates and minimizes the sum of the variances of 2 or more random variables, and each random variable is inversely proportional to its variance (34). The premise for using this method is that all genetic variations are valid instrumental variables (35). We weighted the estimated impact of each SNP on PCOS and on BMI. These estimates were then aggregated using fixed or random effect meta-analysis models to provide a comprehensive assessment of the impact of genetic BMI on the risk of PCOS (36,37). R (version 3.4.0, R Foundation) and its associated MR software packages (TwoSampleMR (17,21,35,38) and MendelianRandomization (39)) were used for all statistical analyses. The MR input function in the MR packages for R software can use the summary data for the combination of genetic variation, the effect value and standard error associated with the exposure, and genetic variation and outcome (35).
Genetic pleiotropic assessment
To assess whether the test violated the MR hypothesis, we used MR–Egger regression, which is often employed in meta-analyses to check for publication bias, to detect genetic directional pleiotropic effects (35,40). Because MR–Egger regression has a strong ability to resist directional multiplicity under the assumption of instrument strength independent of direct effect (InSIDE), the intercept is an index of directional multiplicity (35).
In addition, to obtain a more reliable MR estimate, the weighted median method (41) was used to supplement the MR–Egger regression. In this method, MR estimates are ordered by their inverse variance from smallest to largest. The weighted median estimate is a 50% weighted percentile. If more than 50% of the weight comes from valid SNPs, then this method can be considered to produce an unbiased estimate of the MR causal effect. Studies have confirmed that the weighted median method has more obvious advantages than the MR–Egger method in improving causality detection efficiency and reducing class I error, and it is robust to the InSIDE assumption. Therefore, in this study, we used these 2 methods to better estimate the causal effect and the potential bias in the evaluation results.
Funnel plots were used to perform a visual examination of symmetry, and any deviation indicated potential pleiotropy of the genetic instrumental variables, similar to the funnel plots used in meta-analyses (42). In addition, the causal effect of pleiotropic validity was provided by the estimation of the slope from the MR–Egger regression. However, if the selected SNPs did not explain the large proportional difference in exposure, then this estimate was considered insufficient (35). InSIDE indicated that the SNP estimate of the impact of exposure must be independent of its direct impact on the outcome. Nevertheless, the MR–Egger method still tends to provide reliable causal estimation, even if the selected SNP is a weak tool (35).
Sensitivity analyses
Sensitivity analysis was performed after the exclusion of alternative SNPs to ensure that the MR studies were not affected by these SNPs. We also performed a leave-one-out analysis to determine if there were any SNP-driven associations, meaning that MR was performed to delete different SNPs in each iteration.
Results
We compared the standard IVW analysis with the MR–Egger analysis for potential horizontal multidirectional correction. We drew a scatter diagram (Fig. 1) to prove the relationship between the influence of SNPs on exposure factors (BMI) and on outcome factors (PCOS), which further confirmed the result of zero multiplicity.
Scatter plot with BMI as the exposure and PCOS as the outcome. The red line shows the results of the standard MR analysis (inverse-variance weighted), the green line shows the pleiotropy-adjusted MR–Egger regression line, the blue line shows the results for the simple median, and the purple line shows the results for the weighted median.
The standard IVW MR results showed that every kilogram increase in BMI was associated with an increase in PCOS (odds ratio [OR] 2.21, 95% confidence interval [CI] 1.537 to 3.168, P = 1.77 × 10 –5) (Table 1 and see (28)). The MR–Egger regression method (OR 3.00, 95% CI 1.033 to 8.715, P = .043) was used to test the hypothesis of direct pleiotropy suggested by the IVW analysis. Whether there is statistical significance between the intercept and zero in the regression equation indicates whether the IVW model has pleiotropy. Statistical analysis showed that the intercept = –0.009 (95% CI –0.040 to 0.021; P = .548), indicating no statistical significance, which means that genetic pleiotropy did not cause bias to the causal effect estimates. These findings are supported by a funnel plot that is visually symmetric (28).
Mendelian randomization results for causality for BMI causing PCOS
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 2.416 | 1.426-4.096 | .001 |
| Weighted median | 3.245 | 1.960-5.366 | 4.80 × 10 –6 |
| IVW | 2.208 | 1.537-3.168 | 1.77 × 10 –5 |
| MR–Egger | 3.001 | 1.033-8.715 | .043 |
| (intercept) | 1.000 | –0.040 to 0.021 | .548 |
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 2.416 | 1.426-4.096 | .001 |
| Weighted median | 3.245 | 1.960-5.366 | 4.80 × 10 –6 |
| IVW | 2.208 | 1.537-3.168 | 1.77 × 10 –5 |
| MR–Egger | 3.001 | 1.033-8.715 | .043 |
| (intercept) | 1.000 | –0.040 to 0.021 | .548 |
Abbreviations: BMI, body mass index; CI, confidence interval; IVW, inverse-variance weighted; MR, Mendelian randomization; OR, odds ratio; PCOS, polycystic ovary syndrome.
Mendelian randomization results for causality for BMI causing PCOS
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 2.416 | 1.426-4.096 | .001 |
| Weighted median | 3.245 | 1.960-5.366 | 4.80 × 10 –6 |
| IVW | 2.208 | 1.537-3.168 | 1.77 × 10 –5 |
| MR–Egger | 3.001 | 1.033-8.715 | .043 |
| (intercept) | 1.000 | –0.040 to 0.021 | .548 |
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 2.416 | 1.426-4.096 | .001 |
| Weighted median | 3.245 | 1.960-5.366 | 4.80 × 10 –6 |
| IVW | 2.208 | 1.537-3.168 | 1.77 × 10 –5 |
| MR–Egger | 3.001 | 1.033-8.715 | .043 |
| (intercept) | 1.000 | –0.040 to 0.021 | .548 |
Abbreviations: BMI, body mass index; CI, confidence interval; IVW, inverse-variance weighted; MR, Mendelian randomization; OR, odds ratio; PCOS, polycystic ovary syndrome.
The leave-one-out analysis showed that most of the associated signals were not driven by a single genetic marker (28).
None of the 78 BMI genetic risk variables evaluated were individually associated with PCOS after multiple tests (P > 6.41 × 10-4, .05/78) (28).
In the bidirectional MR analysis, the standard IVW MR results showed that the genetic determinants of PCOS had no contribution to increased BMI (OR 0.998, 95% CI 0.998 to 1.009, P = .77). The MR–Egger analysis also suggested no evidence of the existence of pleiotropy (intercept: 1.000, 95% CI 0.923 to 1.006, P = .712) (Table 2 and Fig. 2).
Mendelian randomization results for causality for PCOS influencing BMI
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 1.000 | 0.984-1.017 | .949 |
| Weighted median | 1.000 | 0.987-1.016 | .880 |
| IVW | 0.998 | 0.998-1.009 | .770 |
| MR-Egger | 1.000 | 0.975-1.031 | .819 |
| (intercept) | 1.000 | 0.923-1.006 | .712 |
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 1.000 | 0.984-1.017 | .949 |
| Weighted median | 1.000 | 0.987-1.016 | .880 |
| IVW | 0.998 | 0.998-1.009 | .770 |
| MR-Egger | 1.000 | 0.975-1.031 | .819 |
| (intercept) | 1.000 | 0.923-1.006 | .712 |
Abbreviations: BMI, body mass index; CI, confidence interval; IVW, inverse-variance weighted; MR, Mendelian randomization; OR, odds ratio; PCOS, polycystic ovary syndrome.
Mendelian randomization results for causality for PCOS influencing BMI
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 1.000 | 0.984-1.017 | .949 |
| Weighted median | 1.000 | 0.987-1.016 | .880 |
| IVW | 0.998 | 0.998-1.009 | .770 |
| MR-Egger | 1.000 | 0.975-1.031 | .819 |
| (intercept) | 1.000 | 0.923-1.006 | .712 |
| Methods | OR | 95% CI | P value |
|---|---|---|---|
| Simple median | 1.000 | 0.984-1.017 | .949 |
| Weighted median | 1.000 | 0.987-1.016 | .880 |
| IVW | 0.998 | 0.998-1.009 | .770 |
| MR-Egger | 1.000 | 0.975-1.031 | .819 |
| (intercept) | 1.000 | 0.923-1.006 | .712 |
Abbreviations: BMI, body mass index; CI, confidence interval; IVW, inverse-variance weighted; MR, Mendelian randomization; OR, odds ratio; PCOS, polycystic ovary syndrome.
Scatter plot with PCOS as the exposure and BMI as the outcome. The red line shows the results of the standard MR analysis (inverse-variance weighted), the green line shows the pleiotropy-adjusted MR-Egger regression line, the blue line shows the results for the simple median, and the purple line shows the results for the weighted median.
Discussion
This study used 2-sample bidirectional MR analyses to study the relationship between BMI and PCOS with the help of large-scale GWAS, and the results showed a causal association between BMI and PCOS, with increased BMI increasing the risk of PCOS. We also conducted a reverse MR analysis, and the results showed that there was no causal relationship between PCOS and BMI. To date, our research was the first MR analysis to explore the causal relationship between BMI and PCOS in an East Asian population.
As a common and complex genetic disease with multiple etiologies, PCOS is caused by the interaction of multiple genes and environmental factors (43). Many metabolic abnormalities of PCOS overlap with those that compose metabolic syndrome, such as central obesity, elevated fasting blood glucose levels, and hypertriglyceridemia (44). Based on the existing public GWAS data, this study used a 2-sample MR method and conducted a bidirectional MR analysis. The IVW method has more power to detect effects and is often used as the primary Mendelian randomization method (45,46). Although MR–Egger regression has the disadvantage of having less power, it is also included in the sensitivity analysis because it resists pleiotropic effects and provides a test for detecting pleiotropic effects (47). If the same results can be obtained from various MR methods, it will increase the stability of our results. The results of this MR analysis may provide some good evidence for assessing the causal role of BMI in the etiology of PCOS, because, compared to traditional observational epidemiological studies, the results of this analysis are less affected by a variety of factors (48). In addition, since genetic variation is stable throughout a person’s lifetime (49), our results represent a risk of PCOS due to elevated BMI. The results obtained using our MR study provide evidence that elevated BMI genes are closely associated with an increased risk of PCOS. This also provides a further theoretical basis for addressing the rising obesity rate and investigating whether lifestyle interventions can help alleviate PCOS risk.
Our research also has some limitations. First, the 2-sample MR analysis used summary data from GWAS, and it was not possible to detect the nonlinear relationship between exposure factors and disease outcomes or to conduct a female subgroup analysis. Second, the 2-sample analysis cannot explain the pathogenesis of PCOS, so it is necessary to further explore the pathogenic mechanism of BMI in PCOS.
Finally, these findings need to be carried out in the context of other evidence with specific associations, but the Mendelian randomization method we used in this study can provide strong support for clarifying the direction of causality. In this way, we believe that this study can contribute to clinical treatment and population prevention efforts to improve health by correcting the causal pathways that lead to the disease.
In summary, our study provides evidence that an increased BMI is associated with an increased risk of PCOS in East Asian populations. These findings provide a theoretical basis for further research to increase the potential therapeutic benefit of lowering BMI to prevent the onset and progression of PCOS.
Abbreviations
- BBJ
Biobank Japan
- BMI
body mass index
- CI
confidence interval
- GWAS
genome-wide association studies
- IVW
inverse-variance weighted
- MR
Mendelian randomization
- OR
odds ratio
- SD
standard deviation
- SNP
single nucleotide polymorphism
- PCOS
polycystic ovary syndrome
Acknowledgments
We would like to thank BBJ for opening up access to their data.
Financial Support. This work was supported by the 973 Program (2015CB559100), the National Key R&D Program of China (2016YFC1306903, 2017YFC0908105, 2019YFA0905400), Shanghai Municipal Science and Technology Major Project (2017SHZDZX01), the Natural Science Foundation of China (U1804284, 81701321, 81871051, 81871055, 81421061, 31571012, 81501154, 81571408), Taishan Scholars Program of Shandong Province (tsqn201812153), Natural Science Foundation of Shandong Province (ZR2019YQ14), the Program of Shanghai Subject Chief Scientist (15XD1502200), the National Program for Support of Top-Notch Young Professionals, Shanghai Key Laboratory of Psychotic Disorders (13dz2260500), the National Program for Support of Top-Notch Young Professionals to Y.S., Shanghai Hospital Development Center (SHDC12016115), Shanghai Youth Top-notch Talent Support Program, Shanghai Mental Health Center (2016-fx-02), Shanghai Municipal Commission of Science and Technology (17JC1402900, 17490712200, 18DZ2260200), and Shanghai Municipal Health Commission (ZK2015B01, 201540114), Scientific Research and Development Fund of Shanghai Jiao Tong University (19X150010012).
Author Contributions: Conceived and designed the experiments: Zhiqiang Li, Yongyong Shi. Performed the experiments: Zhiqiang Li, Yalin Zhao, Yuping Xu. Analyzed the data: Zhiqiang Li, Yalin Zhao, Xiaomeng Wang, Yuping Xu. Contributed reagents/materials/analysis tools: Han Zhao, Chengwen Gao, Chaunhong Wu, Jianhua Chen, Ruirui Chen, Qian Zhang, Juan Zhou, Dun Pan, Zhou Wang, Li You, Yunxia Cao. Wrote the first draft of the manuscript: Yalin Zhao, Yuping Xu, Xiaomeng Wang. Contributed to the writing of the manuscript: Zhiqiang Li, Yalin Zhao. Enrolled patients: Xiaomeng Wang, Lin Xu. Agree with the manuscript’s results and conclusions: Zhiqiang Li, Yongyong Shi, Li You, Yalin Zhao, Xiaomeng Wang, Lin Xu, Jianhua Chen, Han Zhao, Chengwen Gao, Chuanhong Wu, Dun Pan, Qian Zhang, Juan Zhou, Yuping Xu, Ruirui Chen, Zhuo Wang. All authors have read, and confirm that they meet, ICMJE criteria for authorship.
Additional Information
Disclosure Summary: The authors declare no conflict of interest.
Data Availability: All data generated or analyzed during this study are included in this published article or in the data repositories listed in References.
References
Author notes
These authors contributed equally to this work.
These authors jointly supervised this work: L.Y., Y.C., Z.L. and Y.S.
