The contribution of rare genetic variants to the pathogenesis of polycystic ovary syndrome

Abstract

Polycystic ovary syndrome (PCOS) is a highly heritable disorder, but only a small proportion of the heritability can be accounted for by common genetic risk variants identified to date. It is possible that variants with lower allele frequencies that cannot be detected using genome-wide association study arrays contribute to PCOS. Here, we discuss the challenges inherent to studying rare genetic variants in complex disease and review several recent studies that have used DNA sequencing techniques to investigate whether rare variants play a role in PCOS pathogenesis. We evaluate these findings in the context of the latest literature in PCOS and complex disease genetics.

PCOS is a highly heritable complex genetic disorder (1). Genome-wide association studies (GWAS) have reproducibly mapped close to 20 susceptibility loci associated with PCOS diagnosis. The first two GWAS were completed in Han Chinese PCOS case-control cohorts fulfilling the Rotterdam diagnostic criteria. These GWAS together identified eleven PCOS association signals (2,3), several of which have since been replicated in targeted genotyping studies in European ancestry PCOS (4–6) (Table 1). The first GWAS in European ancestry PCOS included only cases fulfilling NIH diagnostic criteria (7). It identified two novel loci and replicated one of the Han Chinese loci. The second European ancestry GWAS included cases fulfilling Rotterdam diagnostic criteria so that non-NIH Rotterdam phenotypes were studied, in addition to NIH phenotype cases. A large number of cases with self-reported PCOS (8) were also studied. This GWAS replicated one signal from the first European GWAS and two from the Han Chinese GWAS, in addition to identifying three novel loci. A recent meta-analysis of European ancestry PCOS GWAS (9) identified three novel loci and replicated 11 of the previously reported loci (Table 1). The meta-analysis contained cases diagnosed by NIH and Rotterdam criteria as well as those diagnosed by self-report. However, only one locus differed significantly in its association by diagnostic criteria; otherwise, the genetic architecture was similar in NIH and non-NIH Rotterdam PCOS phenotypes as well as in self-reported PCOS cases across common variants at 13 loci.

Table 1.

PCOS GWAS Susceptibility Loci

Chromosome	Han Chinese¶ Chen, 2011 (2) Shi, 2012 (3)	European 1*, Hayes, 2015 (7)	European 2*¶§ Day, 2015 (8)	European Meta*¶§ Day, 2018 (9)
2p16.3	LHCGR
2p16.3	FSHR
2p21	THADA		THADA	THADA
2q34			ERBB4	ERBB4
5q13.1			RAD50	RAD50
8p32.1		GATA4/NEIL2		GATA4/NEIL2†
9p24.1				PLGRKT
9q22.32	C9orf3	C9orf3		C9orf3
9q33.3	DENND1A			DENND1A
11p14.1		FSHB	FSHB	FSHB
11q22.1	YAP1		YAP1	YAP1
11q23.1				ZBTB16
12q13.2	RAB5B/SUOX			RAB5B/ERBB3
12q14.3	HMGA2
12q21.2			KRR1	KRR1
16q12.1	TOX3			TOX3
19q13.3	INSR
20q11.21	SUMO1P1
20q13.2				MAPRE1

^*PCOS diagnosis was based on NIH criteria,

^¶Rotterdam criteria, or

^§self-report.

^†PCOS association dependent on diagnostic criteria in bold; association with remaining loci was similar for PCOS diagnosed by self-report and PCOS diagnosed by NIH or non-NIH Rotterdam criteria

These GWAS have significantly advanced our understanding of the pathophysiology of PCOS by implicating gonadotropin secretion (FSHB) and action (LHCGR, FSHR), androgen biosynthesis (DENND1A), metabolic regulation (THADA, INSR,) and follicle development (YAP1, HMGA2) in PCOS pathogenesis. Nevertheless, GWAS have important limitations. GWAS are based on the hypothesis that common variants, typically defined by minor allele frequencies (MAFs) of 5% or greater, cause common, complex traits/diseases, such as PCOS (10). GWAS arrays measure single nucleotide polymorphisms (SNPs) that span the entire genome. The GWAS SNPs are located within segments of the genome where genetic variation is highly correlated or in ‘linkage disequilibrium’ (LD) (11). These SNPs serve as markers, so-called “tag SNPs”, for the surrounding genomic regions, or haplotypes. Therefore, trait-associated alleles identified via GWAS are often not the causal variants themselves, but rather are associated or in LD with the underlying causal variant(s) within the same region.

Additional statistical methods (12–15) and DNA sequencing of GWAS loci (16,17) are required to identify potential pathogenic variants and candidate genes within a given region in a process known as “fine mapping”. However, causality can only be proven with functional studies (18). The majority of GWAS variants are in noncoding portions of the genome, which makes elucidating their functional significance exceptionally challenging (19). The exact mechanisms by which most GWAS susceptibility loci contribute to disease pathogenesis remain unknown (20). Indeed, while PCOS GWAS have implicated many plausible candidate genes, the corresponding pathogenic variants within these putative causal genes have yet to be identified (21).

By definition, GWAS variants are prevalent in the population and, therefore, would not be expected to have large effects on phenotype (1). Consistent with this prediction, most of the risk variants identified from GWAS have modest effect sizes (22). In many complex traits/diseases, including PCOS, GWAS variants identified to date account for less than 10% of the estimated heritability (the phenotypic variance attributable to genetic differences) of the disorder (23,24). The prevailing hypothesis for this deficit in heritability had been that variants with lower frequency that cannot be detected using GWAS arrays contribute to complex traits (23,25,26). It was anticipated that these rare variants would be in coding portions of the genome where they would have large effects on phenotype by disrupting the encoded molecule. Consequently, elucidating the biologic relevance of such coding rare variants would be amenable to traditional molecular approaches. Support for this hypothesis came from studies that found rare genetic variants with large effects on HDL (27), adiponectin (28) and triglyceride levels (29) as well as on blood pressure (30) in the general population.

Although the precise definition varies from study to study, the term “rare variants” generally refers to variants that have a MAF on the order of less than 1-2% in a given population. Testing for rare variants has only recently become feasible because of cost-reductions in DNA sequencing technologies (31–33). Rare variants can be inferred, or “imputed”, from GWAS data using large genomic databases, but this approach is only reliable down to a MAF of ~0.1% (34,35) and is not yet available for many ancestral populations (genetic variation is ancestry-dependent). Successful utilization of sequencing data to study rare variants in a complex disease framework still requires overcoming a number of unique challenges. First, rare variants occur too infrequently to perform standard case-control allele association tests without recruiting prohibitively large cohorts. For example, having 80% power to detect an association for a variant with 0.001% MAF and a phenotypic effect size of one standard deviation (β≈4) would require a sample size over one million (36). Populations can be enriched for causal rare variants by studying extreme phenotypes or by incorporating families with multiple affected individuals (26), but even in enriched cohorts, most rare variant allele frequencies are still too low for individual variant associations (37), particularly because—in contrast to Mendelian disorders—pathogenic alleles in complex traits are not individually predictive of phenotypes (38,39). Collapsing methods that aggregate sets of rare variants within the same gene or genomic region into single statistics, such as burden tests (40) or sequence kernel association tests (41), can be used to test for association with a given disease or phenotype. However, aggregating variants according to presumed functional correlation presents another significant challenge because of the complex nonlinear nature of the genome (42). For example, functional variants within an intron of a gene do not necessarily act through that gene (43,44).

Second, there is a tremendous amount of variation within the human genome. Accordingly, analyzing whole exome and whole genome sequence data for functionally relevant genetic variation is computationally arduous. Efficient processing of sequencing data requires access to high-performance computing clusters and/or specially designed bioinformatics hardware (45). Many different software tools and pipelines exist for processing sequencing data and performing corresponding statistical tests, but advanced bioinformatics training is required to perform such analyses and ensure interoperability between software versions and file formats.

The power of aggregate-level tests to detect causal variants is highly dependent on a priori variant screening (46). Ideally, only functional alleles would be included in association testing, as inclusion of inconsequential variants introduces statistical noise. Variant effect prediction (VEP) tools exist that model either evolutionary conservation (47,48) or protein changes (49,50), but substantial differences in their design and output metrics complicate integration of their results (51). Further, thresholds based on VEP results are typically subjective (52). Functional annotation of the genome has expanded significantly in recent years (53–56) but not for all cell types and genomic regions (57). Despite the difficulties associated with studying rare variants, mounting evidence indicates that they do contribute to complex traits (58,59), and corresponding technologies and methods for studying rare variants continue to evolve at a rapid rate (60,61).

We investigated the potential role rare variants play in the pathogenesis of PCOS. We used two approaches: case-control candidate gene sequencing studies and family-based analyses. Anti-Müllerian hormone (AMH) is a highly plausible PCOS candidate gene. It is an important regulator of folliculogenesis (62), steroidogenesis (63) and neuroendocrine signaling (64). Circulating levels of AMH are elevated in women with PCOS (65,66). We performed targeted sequencing of AMH in a cohort of 643 PCOS cases and 153 controls and identified 18 rare (MAF≤1%) coding variants that were present in PCOS cases but not in controls (PCOS-specific variants) (67). We then measured the functional impact of these variants using AMH-mediated luciferase assays in transfected COS7 cells. Of these 18 PCOS-specific variants, 17 demonstrated a significant reduction in AMH-mediated signaling capacity. Further, using a gene-based burden test approach (40), the functional AMH variants were significantly associated with PCOS using population-based controls. There were also rare coding variants that were present in both cases and controls. However, these variants had no impact on AMH signaling. This finding emphasizes the importance of functional assays to confirm the biologic relevance of genetic variants.

We subsequently investigated rare variation in the regulatory regions of AMH and its specific type 2 receptor, AMHR2 (68). We identified 20 additional PCOS-specific variants in or near AMH and AMHR2 that resulted in significantly reduced AMH signaling activity and were also significantly associated with PCOS. These variants included one missense variant in AMHR2, 16 noncoding/splicing variants in or upstream of AMHR2, and 3 noncoding variants upstream of AMH. AMH levels in cases with functional variants were significantly higher than control women (67). However, PCOS women with functional AMH/AMHR2 variants had significantly lower AMH levels compared to other women with PCOS (68). Five of the AMH variants have also been found in men with persistent Müllerian duct syndrome, a rare disorder in which males retain internal Müllerian duct structures (69). AMH levels in these men are typically normal or undetectable (70). These findings suggest that AMH signaling capacity does not correlate positively with circulating AMH levels. Reduced AMH signaling could contribute to PCOS, however, by increasing androgen production due to loss of CYP17 inhibition by AMH. CYP17 is a key enzyme in androgen biosynthesis (71). The AMH variants with reduced AMH signaling showed a significant reduction in CYP17α1 expression inhibition compared to wild-type AMH (68). Collectively, about 6.7% of PCOS-affected women from these cohorts had one or more of the AMH/AMHR2 rare variants (67,68). Further, these studies illustrate the power of limiting sequencing to a few highly plausible candidate genes to avoid the statistical penalty of correcting for multiple testing in genome-wide analyses.

We performed whole-genome sequencing on DNA from 261 individuals from 62 families with one or more daughters with PCOS. We tested for gene-level associations of rare variants (MAF≤2%) that were predicted-to-be-deleterious (48,51) with PCOS and its concomitant hormonal traits using a quantitative trait meta-analysis (72). We found rare, primarily noncoding variants in DENND1A that were significantly associated with the following reproductive and metabolic traits: testosterone, dehydroepiandrosterone sulfate, sex hormone-binding globulin, insulin, luteinizing hormone (LH), and follicle-stimulating hormone (FSH). Common variants in DENND1A were associated with PCOS diagnosis in GWAS (2,3,9). Subsequent studies demonstrated that DENND1A is an important regulator of human ovarian androgen biosynthesis (73,74). However, none of the common GWAS variants were in known functionally important regions of the gene. Further, sequencing DENND1A in a limited number of PCOS cases did not identify pathogenic variants (73). Most of the rare variants we found were not in LD with the DENND1A GWAS variants (75). However, nine of the DENND1A variants were predicted to significantly disrupt transcription factor binding, while 23 were predicted to significantly affect RNA-binding protein motifs. Nevertheless, functional assays are needed to confirm these predictions.

Each of the rare variants in DENND1A was observed in only one or two families. Collectively, however, the rare variants were present in 50% of families. This finding is of considerable interest in light of our original family studies that demonstrated a unimodal distribution for testosterone levels in control women but a bimodal distribution in sisters of women with PCOS (76) suggesting that elevated testosterone levels in PCOS were affected by specific autosomal alleles within families. Rare noncoding variants in DENND1A that altered gene regulation could result in increased testosterone production. Testosterone antagonizes the effects of estradiol and progesterone to slow hypothalamic GnRH contributing to enhanced LH release characteristic of PCOS (1,77). Consistent with this hypothesis, LH:FSH ratios were significantly higher in PCOS women with one or more of the DENND1A rare variants compared to PCOS women without any of the DENND1A variants (72). While no other gene associations in the family-based sequencing reached genome-wide significance, two of the top-5 associated genes were highly plausible PCOS candidate genes: C9orf3, which is a PCOS GWAS gene (2,3,9,78), and BMP6, which is a regulator of folliculogenesis in granulosa cells (79), and was more highly expressed in PCOS women compared with reproductively normal control women (80).

These studies support a model of PCOS in which pathogenic variants, while individually rare, tend to occur in genes regulating relevant disease pathways. The DENND1A findings support our studies implicating hyperandrogenemia as a core biologic pathway in PCOS (76). Hyperandrogenemia is a consistent reproductive phenotype in male as well as female PCOS relatives (76,81), including premenarchal daughters (82). Further, one of the earliest biomarkers in at-risk children, daughters of affected women, is evidence for global increases in 5α-reductase activity, which would enhance the conversion of T to its more potent metabolite, dihydrotestosterone (83,84).

The emerging consensus is that complex traits are primarily driven by noncoding variation (85,86), both common and rare (87,88). The idea that rare variants account for most of the heritability of complex traits was largely disproved by a series of studies that modeled cumulative contributions from all common SNPs within a population (not just those with significant trait associations), and found that common SNPs with small effect sizes collectively account for a much more significant portion of complex disease heritability than GWAS SNPs alone (59,89,90). Accordingly, as GWAS sample sizes have increased, the number of significant associations discovered per study has increased proportionally (91). However, the distribution of variant effects sizes and allele frequencies varies between different complex traits and diseases (92). Additionally, recent large-scale sequencing studies have demonstrated that remaining unexplained heritability in complex traits and diseases can be accounted for by rare variants that are not in LD with any common variation (93,94). Our studies indicate that both common and rare noncoding genetic variants contribute to PCOS pathogenesis, analogous to other complex traits/diseases. However, we also identified rare, pathogenic coding variants in AMH in about 3% of affected European ancestry cases (67). It is highly likely that as more candidate genes are sequenced, additional rare coding variants contributing to PCOS will be identified.

In conclusion, although the study of rare variants in PCOS remains extremely limited, results from our initial studies indicate that these variants contribute to PCOS. Candidate gene studies have indicated that rare coding and noncoding variants affecting the bioactivity of AMH or its receptor are present in a substantial minority of PCOS cases (67,68). Rare, primarily noncoding, variants in DENND1A appear to play a role in the pathogenesis of familial PCOS in ~50% of families (72). These variants have implicated several plausible causal pathways, in particular androgen biosynthesis, consistent with our family studies of PCOS phenotypes (76,81,82). Functional studies are needed to determine the specific mechanisms by which these noncoding variants contribute to disease pathogenesis. More broadly studying the potential effects of rare variants represents a next step in the progression towards understanding of the genetic architecture of PCOS.

Highlights.

• Studying rare variants in complex disease presents unique challenges

• Rare variants affecting AMH or AMHR2 are present in many PCOS cases

• Rare variants in DENND1A associated with altered hormone levels in PCOS families