Phenotypes Associated With Polycystic Ovary Syndrome Risk Variants

Abstract

Polycystic ovary syndrome (PCOS) affects approximately 10% of women of reproductive age worldwide, making it the most common endocrinopathy [1]. PCOS is diagnosed based on menstrual irregularity, hyperandrogenism, and/or polycystic ovaries [2]. Many women with PCOS have metabolic complications, although metabolic abnormalities are not a component of the defining features. PCOS has been associated with insulin resistance and obesity, and women with PCOS are at greater risk for developing diabetes, metabolic syndrome, and cardiovascular disease [3].

Despite the prevalence of PCOS, the etiology of the syndrome remains unclear. Genetic studies have established that PCOS is a complex genetic disorder. The largest genome-wide association study (GWAS) in women with PCOS, as presented in a recent abstract, has expanded the known genetic underpinnings to 29 risk variants [4]. The variants identified in GWAS are located in introns or between genes making the effect of the variants difficult to ascertain. Some of the variants are located in or near genes coding for hormones or proteins known to affect physical features. We therefore hypothesized that these 29 risk variants would be associated with variation in phenotypic traits in women with PCOS and controls.

Methods

Subjects

The analysis included 404 women with PCOS and 408 controls from a Boston cohort. Cases were diagnosed with PCOS based on the presence of both ovulatory dysfunction and clinical and/or biochemical hyperandrogenism, consistent with “classic” PCOS, as defined by the National Institutes of Health (NIH) criteria and phenotypes A and B of the Rotterdam Criteria, but excluding those with regular menses or no hyperandrogenism [2]. We previously demonstrated that 97% of these women will also have polycystic ovary morphology on ultrasound [5]. All subjects were between 18 and 40 years of age, on no hormonal medications for at least 3 months, and on no medications affecting insulin, inflammation, or lipid levels for at least 1 month [5]. All subjects were screened and were negative for other endocrine pathologies and had normal prolactin, thyrotropin (thyroid stimulating hormone), and/or free T4 index.

The study was approved by the Institutional Review Board of the Massachusetts General Hospital and the University of Utah. All recruited subjects gave written informed consent.

Data Collection

Subjects were studied > 10 days after their last menstrual period to avoid the time frame in which androgen and LH levels are suppressed after a spontaneous ovulation [6], and after a 12-hour fast [5]. Subjects underwent a history and physical examination. Measures of adiposity were obtained including body mass index (BMI), waist circumference at the level of the iliac crest, and hip circumference at the maximum diameter. Subjects were scored for hirsutism by the method of Ferriman and Gallwey [6].

Each subject underwent blood sampling for metabolic factors (fasting blood glucose, insulin); cardiovascular lipid risk factors (total cholesterol, triglycerides, low-density lipoprotein [LDL], high-density lipoprotein [HDL]), gonadotropins (luteinizing hormone [LH], follicle-stimulating hormone [FSH]); and sex steroids and binding proteins (testosterone, androstenedione, dehydroepiandrosterone sulfate [DHEAS], sex hormone binding globulin [SHBG]), as described previously [5]. Additional blood samples were drawn at 10 and 20 minutes to calculate an average LH and FSH concentration [6]. Finally, pelvic ultrasound was performed for evaluation of follicle number and ovarian volume (Phillips, 5-MHz convex array transducer). Assays were performed as previously described, with LH and FSH levels expressed as international units per liter as equivalents of the Second International Reference Preparation 71/223 of human menopausal gonadotropins [5].

Genotyping and Association Analyses

Genotyping was performed using the OmniHumanExpress Bead Chip (Illumina, San Diego, CA, USA). Single nucleotide polymorphisms (SNPs) that were present at a frequency of ≥ 1% in the population were included in the initial analysis. For the SNPs not directly genotyped, imputation was performed in the software package Minimac4 (GitHub—statgen/Minimac4) using the Haplotype Reference Consortium Data [7].

Statistical Analysis

Data were analyzed using linear regression with the genotype at each risk locus and log-transformed traits in the combined sample of cases and controls. Multiple linear regression was also performed with age and BMI as covariates. We used the false discovery rate to determine significance for 29 variants and 15 independent traits (BMI, waist circumference, hip circumference, LDL, HDL, testosterone, DHEAS, androstenedione, SHBG, LH, FSH, ovarian volume, follicle number, fasting glucose, and fasting insulin). Data are also shown without log normalization in the results.

Results

Of 29 risk variants, 7 were associated with variation in measurable phenotypic traits. Two variants were associated with gonadotropin levels (Table 1). LH levels were higher at the PCOS risk alleles in loci containing the gene FSHB (rs11031005-C: TT 18.7 ± 16.5, TC 24.3 ± 24.8, CC 24.7 ± 15.0 IU/L). The variant at the PRSS23 locus was associated with higher FSH levels (rs4944653-G: AA 9.0 ± 3.1, AG 9.5 ± 3.2, GG 10.7 ± 4.6 IU/L) (Table 1 and Supplementary Table S1) [8]. The LH to FSH ratio was also higher at the locus containing FSHB (rs11031005-C: TT 1.8 ± 1.1, TC 2.3 ± 1.6, CC 2.5 ± 1.5).

Of note, women who carried 1 or 2 rs13004711-C alleles at the FSHR locus had higher FSH levels (Supplementary Table S1) [8]. However, the relationship was not significant based on the false discovery rate (β = 0.022 ± 0.090; P = .020), nor when controlling for age and BMI as covariates (β = 0.021 ± 0.089; P = .019).

We next assessed the relationship between the PCOS risk variants and circulating androgen levels. The PCOS risk alleles at the locus containing the genes NEIL2/GATA4 (rs804263-C: TT 2.8 ± 1.2, TC 3.3 ± 1.4, CC 3.5 ± 1.6 ng/mL) was associated with higher androstenedione levels (Table 2 and Supplementary Table S1) [8]. The PCOS risk allele at the CYP3 complex locus was associated with higher DHEAS levels (rs148982377-T: CT 164.2 ± 102.3, TT 221.5 ± 113.5 μg/dL). The PCOS risk allele at the SHBG locus was nominally associated with lower SHBG levels (rs1641518-A: GG 57.7 ± 30.1, GA 51.3 ± 29.1, AA 43.3 ± 24.2 nmol/L; P = .003), and was less significant after controlling for BMI (P = .02), consistent with mediation by BMI.

We also assessed the association between PCOS risk variants and ovarian morphology on ultrasound. PCOS risk alleles at loci containing SHBG (rs1641518-A: GG 11.6 ± 5.8, GA 13.2 ± 6.4, AA 13.7 ± 3.1 mL) and ZBTB16 (rs1784692-T: CC 7.8 ± 2.2, CT 11.0 ± 5.1, TT 12.5 ± 6.3 mL) were associated with higher ovarian volume, although the ZBTB16 relationship was partially mediated by age and BMI (Table 3 and Supplementary Table S1) [8]. The PCOS risk allele at the CYP3 locus was associated with lower maximum follicle number (rs148982377-T: CT 12.1 ± 4.1, TT 11.4 ± 4.0).

We finally assessed the effect of PCOS risk variants on metabolic traits. The PCOS risk allele at the FTO locus was associated with higher BMI (rs8047587-T: GG 24.9 ± 6.6, GT 27.3 ± 7.6, TT 28.8 ± 8.9 kg/m²) and greater hip circumference (rs8047587-T: GG 101.2 ± 12.6, GT 103.5 ± 13.7, TT 106.5 ± 17.1 cm). As expected, the association between the PCOS risk allele and hip circumference was no longer significant when controlled for BMI.

We evaluated the same gene variants in the subjects with PCOS, alone (Supplementary Table S2) [8]. In the subset of subjects with PCOS, the PCOS risk variants that remained significant were those at the FSHB locus with higher LH levels (β ± SE 0.015 ± 0.0028; P < .001) and PRSS23 with higher FSH levels (0.067 ± 0.018; P < .001). The other variants did not reach significance.

Discussion

We demonstrate associations between PCOS genetic risk variants and PCOS phenotypes. Of the 29 risk loci, 7 were associated with variation in measurable PCOS traits, including gonadotropin and androgen levels, ovarian morphology, and anthropomorphic features. These relationships illuminate the role of the variants in the defining features related to PCOS pathophysiology. Conversely, they help to delineate that these hormone levels or traits play an elemental role in the clinical presentation and etiology of PCOS.

We replicated the relationship between the FSHB locus (rs11031005-C) and LH levels and the LH to FSH ratio, previously associated with a variant in complete linkage disequilibrium (rs11031006, r² = 1, P < .001) [9]. Importantly, these PCOS risk alleles at the FSHB locus are associated with lower FSH levels in larger datasets, but they appear to have a more subtle influence on FSH than the influence on LH levels. The rs11031005 variant is located in a highly conserved region containing transcription factors and H3K27Ac marks [10]. The region has open chromatin specifically in gonadotropes [11]. Further, deletion of a portion of the region increases FSH expression and secretion from a gonadotropin cell line model [11]. Therefore, the region appears to suppress FSH levels. The lower FSH beta subunit (FSHB) may result in excess alpha subunit, which can then associate with LHB to increase LH levels and the LH to FSH ratio, contributing to the cardinal features of PCOS [12].

Unlike FSHB, the relationship between FSH levels and the rs13004711 PCOS risk variant at the FSHR locus did not reach the Bonferroni-corrected threshold of significance. The associated risk variant is located in a region with candidate cis regulatory elements and DNase hypersensitivity peaks [10]. We previously demonstrated a stronger relationship between FSH levels and a variant in modest linkage disequilibrium (rs2268361, r² = 0.288, P < .001) [13]. The rs2268361-T risk variant is also associated with higher FSHR mRNA levels in testes [14], while rs13004711 has no expression quantitative trait locus (eQTL) in reproductive tissues. Taken together, variants at the locus could result in FSH receptor protein that is less abundant or less functional with FSH levels higher to maintain normal stimulation. Additional fine mapping will be needed to characterize the FSHR locus.

A variant (rs4944653-G) in a locus containing PRSS23 was associated with higher FSH levels. PRSS23 is a serine protease that is downregulated by gonadotropins and highly expressed in theca tissues near the time of ovulation [15]. It is also expressed in atretic follicles [15]. Increased LH levels in PCOS could downregulate PRSS23, which would impair the process of follicular atresia leading to increasing ovarian follicle number and greater feedback on FSH. Of note, the variant is located within a lncRNA, FZD4-DT, which is an eQTL in pituitary tissue and is also near FZD4, which is an eQTL in testicular tissue [14]. Further, the region has many cis regulatory elements and histone marks that could be important for regulation of other genes [10]. Therefore, the lncRNA or FZD4 may be an alternate risk-inducing feature at the locus.

The variant at the CYP3 locus was previously associated with DHEAS genetic determinants and we now demonstrate that it is associated with higher DHEAS levels [16-18]. The CYP3 complex contains several cytochrome p450 C family enzymes involved in steroidogenesis [19], including the conversion of 17-OH pregnenolone to DHEA and CYP3A7 metabolism of DHEA and DHEAS [20]. An insertion variant in the promotor of CYP3A7 was also found to be significantly associated with DHEAS levels [21], providing further support for the relationship between the locus and androgen levels. Interestingly, the Cyp3a knockout mice do not have a fertility phenotype and had no differences in estradiol or testosterone, although DHEAS and androstenedione were not assessed [22].

The PCOS risk variant at the CYP3 complex was also associated with lower maximum follicle number. It is not clear why follicle count was lower in the presence of the variant, but follicle number was counted in a single plane and could underestimate total follicle number [23, 24]. An alternative consideration is that this variant affects the transcription factor ZNF789 because it is located in the ZNF789 intron. Although the function of the transcription factor is not known, it is highly expressed in the ovary and could be playing a role in the regulation of androgen levels and follicle number at this locus [14].

The rs1641518-A variant at the SHBG locus was nominally associated with lower SHBG levels, as has also been demonstrated for many variants in the region [16, 25]. The association appeared to be mediated by BMI because the association was no longer significant after controlling for BMI. SHBG is known to be decreased in PCOS, leading to increased free testosterone, which is a hallmark feature [5]. The rs1641518-A variant at the SHBG locus was also associated with increased ovarian volume. Developing follicles express androgen receptors which act to promote follicular growth [26]. Previous studies suggest that increased testosterone may increase follicle number. Taken together, the lower SHBG and higher free testosterone could increase ovarian volume through increased number of follicles [26] or increased theca cell volume.

The variant at the ZBTB16 locus (rs1784692-A) was associated with increased ovarian volume, which is accounted for by stroma and follicles. Expression studies in ovarian tissue demonstrate that ZBTB16 is present in human fetal ovaries and increases in bovine fetal ovaries in conjunction with genes involved in stromal expansion and steroidogenesis [27]. It is also associated with anti-Müllerian hormone (AMH) levels, which points to a relationship with small antral follicles [27, 28]. Previous studies in prostate cancer cells suggest that ZBTB16 is an androgen-responsive gene [29]. It also mediates GATA4 gene transcription. Finally, proteomics in older women with PCOS in the UK Biobank suggest that the variant is associated with ZBTB16 protein levels [4]. Taken together, a role for ZBTB16 in oocyte development, as suggested by its relationship to ovarian volume, would be a unique pathway affected in PCOS risk.

Interestingly, there were very few gene variants associated with metabolic risk. The FTO locus (rs8047587-T) is known to be associated with fat mass and obesity and is the most influential genetic risk factor for BMI [30]. In this study, an FTO variant was associated with BMI and hip circumference and remained significant when subjects with PCOS, only, were assessed (Table 4). Deletion of FTO disrupts adiposity, eating and energy metabolism in a developmental fashion [31]. Based on its previously known effects, it is understandable that the FTO variant would be a key driver for the metabolic effects of PCOS. The relationship between obesity and PCOS has been determined as causal [4, 32], and the identification of the FTO risk gene may represent a common pathway to both PCOS and obesity.

The strengths of the study include our well-phenotyped cohort with available laboratory results obtained under controlled conditions that included fasting, appropriate follicular phase timing, and exclusion of patients taking medications. These measurements have decreased variability compared to large datasets with imprecise patient identification and randomly drawn lab results. We only examined women with PCOS resulting from irregular menses and hyperandrogenism [2, 5]. However, we previously demonstrated that there was no genetic difference in women with PCOS diagnosed using the breadth of the Rotterdam criteria [32]. Further, the subset tends to have the most substantial anthropomorphic and laboratory results, providing a greater range for better detection of genetic effects [5].

Limitations include small numbers resulting from our rigorous clinical study, which may result in missed associations. For example, the variant at the rs1641518 was associated with lower SHBG levels in our large genome-wide association study but only reached nominal significance with our smaller numbers [4]. We did correct for multiple testing using a false discovery rate rather than Bonferroni correction to avoid excessive stringency. While it is possible that we missed traits based on the size of the cohort, a power analysis suggests that we had 80% power to detect a relationship between a genetic variant and LH levels between cases and controls with only 98 women with PCOS and 98 controls, for example. Other variants required even lower numbers for 80% power. We also note that genetic risk loci may not result in phenotypically measurable traits.

The relationship between gene variants identified in GWAS and disease states has been difficult to elucidate based on the location of these variants outside coding regions. The recent GWAS in women with PCOS uncovered a number of variants that are in or near loci containing genes important for gonadotropin levels, steroidogenesis, or follicular development or function. The relationship between these variants and measured hormone levels and phenotypes supports the role of these risk variants in influencing the cardinal features of PCOS. These genetic loci support the importance of gonadotropins, androgens, and the polycystic ovary morphology components as integral to PCOS.

Funding

The work in this publication was supported by R01HD065029 and ADA 1-10-CT-57 (C.K.W.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Disclosures

The authors have nothing to disclose.

Data Availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

References

Abbreviations

 
• AMH

  anti-Müllerian hormone

 
• BMI

  body mass index

 
• DHEAS

  dehydroepiandrosterone sulfate

 
• eQTL

  expression quantitative trait locus

 
• FSH

  follicle-stimulating hormone

 
• GWAS

  genome-wide association study

 
• HDL

  high-density lipoprotein

 
• LDL

  low-density lipoprotein

 
• LH

  luteinizing hormone

 
• PCOS

  polycystic ovary syndrome

 
• SHBG

  sex hormone binding globulin

 
• SNP

  single nucleotide polymorphism