Genetics of PCOS: What’s New?

Genetic variants confer risk for PCOS

PCOS remains the most common endocrinopathy in reproductive age women, with lifetime costs for diagnosis and treatment estimated in the billions of dollars.¹ The etiology has been difficult to identify despite years of physiologic studies. Further, the diagnosis in adolescents may be masked by normal developmental changes.² Thus, early intervention and prevention strategies are not available. In adults, the broad diagnostic criteria encompass a heterogeneous patient population. While some patients have increased risk for type 2 diabetes and metabolic consequences, others appear to suffer mainly from infertility related to irregular menses.^3–5 Therefore, future health implications are difficult to predict in an individual patient.

Genetic variants could provide the critical predictive risk information for the development of PCOS and PCOS comorbidities. PCOS is highly heritable.⁶ The tetrachoric correlation for PCOS, a correlation used for binary data, is twice as high in monozygotic twins (0.71 [0.43 to 0.88]) compared to dizygotic twins (0.38 [0.00 to 0.66]), suggesting PCOS is approximately 79% influenced by genetic variance.⁶ While twin studies are the strongest to estimate genetic risk, many additional studies have demonstrated familial clustering of PCOS, with approximately 50% of sisters sharing PCOS features.⁷ While this pattern of inheritance in sisters suggested a dominant mode of inheritance, it is now clear that PCOS is a complex, polygenic disorder in which multiple risk loci with small effects contribute to disease, similar to the genetics of type 2 diabetes.^8–10

Great leaps in genetics came with the mapping of the human genome, the realization that the majority of DNA variation is shared at high rates across all humans, and the understanding that multiple DNA variants are inherited together in blocks (linkage disequilibrium or LD). Scientists used the new knowledge to develop genotype arrays, with representative DNA variants found in 1% or more of the population, covering a significant portion of the LD blocks across the human genome.^11,12 Genome-wide association studies (GWAS) used these arrays to identify risk variants that were over represented in disease cases compared to controls. Because the number of DNA variants tested was so large, approaching one million, the p value indicating a significant association must be very low (<5×10⁻⁸).¹² In addition, the effect sizes of these common variants are small, requiring tens to hundreds of thousands of participants to demonstrate effects. Thus, the use of these approaches to understand PCOS has been a challenge.

The initial GWASs were performed in Han Chinese subjects, demonstrating 11 gene loci associated with PCOS (Table 1).^9,10 Our group and others found that several of these loci also confer risk for PCOS in European women.^13–19 Additional loci emerged from meta-analyses and replication in large groups of European women.^17,18 Together, these studies culminated in the largest PCOS GWAS from the International PCOS Consortium, which included a meta-analysis of 10,074 PCOS cases and 103,164 controls.²⁰ Together, there were a total of 19 loci that confer risk for PCOS in Han Chinese and European women (Table 1).

Table 1.

Single Nucleotide Polymorphisms Associated with Polycystic Ovary Syndrome at Genome-Wide Significance.

	Chromosome : Position	Risk Variant/Reference Variant	Nearest Gene	Associated Phenotypes*	References
Han Chinese Only	2:48978159	A/G	LHCGR		9
	12:66224461	A/C	HMGA2		10
	19:7166109	A/G	INSR		10
	20:52447303	A/G	SUMO1P1		10
Both	2:43721508 2:43561780	T/C A/G	THADA	PCOM, Increased T	9,20,36
	2:49247832	T/C	FSHR	Increased FSH, OD	9,20
	2:213391766 2:212291772	A/G A/G	ERRB4	OD, PCOM	20,22,52
	9:97648587 9:97723266	A/G A/G	C9orf3	HA, OD, PCOM	9,20
	9:126525212 9:126619233	A/G A/G	DENND1A	HA, OD, PCOM	9,20,22
	11:30226356	T/C	ARL14EP/FSHB	Decreased FSH, Increased LH, OD	20,53
	11:102070639 11:102043240	A/G A/G	YAP1	OD, PCOM	10,20
	12:56390636 12:56477694	A/G T/A	ERBB3/RAB5	HA, OD, PCOM	10,20
	16:52347819 16:52375777	T/G	TOX3	HA	10,20
European	3:149319873	C/T	WWTR1	OD	22
	5:131813204	T/C	IRF1/RAD50	Increased testosterone	20
	6:159898261	C/T	SOD2	HA, PCOM	22
	8:11623889	A/T	GATA4/NEIL2	OD, PCOM	20
	9:5440589	A/C	PLGRKT		20
	11:113949232	T/C	ZBTB16	OD, PCOM	20
	12:75941042	C/T	KRR1	OD	20
	20:31420757	T/A	MAPRE1		20

^*HA=hyperandrogenism, OD=ovulatory dysfunction, PCOM=polycystic ovary morphology Data from Refs 9, 10, 20, 22, 36, 52, 53.

The next wave of GWAS will capitalize on large DNA collections agnostic to disease and linked to electronic medical records to identify cases and controls using International Classification of Disease (ICD) codes or more complicated algorithms. Using the Partners Healthcare electronic medical records, ICD codes performed equivalently to an algorithm validated to identify PCOS cases using natural language processing.²¹ Supporting the use of ICD codes for PCOS identification, the first GWAS using electronic medical records to pull PCOS cases replicated two previously identified loci from the International PCOS consortium and located two additional loci (Table 1).²² The study also identified the first locus in African American women.²² The stage is now set for further studies employing larger numbers and expanding the ethnicities examined.

Is Genetic Architecture the Same for All PCOS Subtypes?

A robust discussion always ensues when the criteria for PCOS are discussed (see 2. Diagnosis of Polycystic Ovarian Syndrome: Which Criteria to Use When?). While there was concern that using National Institutes of Health (NIH), Rotterdam and self-reported PCOS would dilute the ability to identify genetic risk for PCOS, the problem did not materialize. In fact, the genetic architecture was similar across the diagnostic subtypes, with the exception of one locus, GATA4/NEIL2, which demonstrated heterogeneity of effect.²⁰ At this locus, the most significant association was demonstrated in subjects defined using the NIH criteria (odds ratio for risk [OR] 1.33; 1.26–1.41 95% confidence intervals), and the lowest for self-report (OR 1.08; 1.03–1.13), although all were risk loci. Taken together with validation of PCOS cases from the electronic record, large studies using electronic datasets to isolate all subtypes of PCOS will be valid. In addition to studying larger datasets, it is time to determine what the risk means and how to move our expanding scientific understanding to clinically applicable diagnostic and treatment options.

Moving from genotype to functional mechanism – How does a risk variant cause PCOS?

Identifying the underlying function of the GWAS risk variants has been difficult because the majority are non-coding and reside in introns and between genes where a mechanism is not apparent. It has also been estimated that only one-third of the associated variants affect the nearest gene.¹² Nevertheless, approximately 60% of the common variants associated with disease map to regions of DNA hypersensitivity and regulatory regions.^23,24 Investigators have capitalized on these properties in other diseases to determine the mechanism through which variants associated with LDL cholesterol levels, autoimmune disorders and Crohn’s disease cause risk.^23,25,26 Elucidating the underlying mechanism that ties a common variant to disease risk requires examining regulatory influence and the associated gene expression.

The risk variants identified near FSHβ (rs11031006 and s11031005) illustrate a regulatory mechanism for genome-wide associated variants. The rs11031006 and rs11031005 variants not only confer risk for PCOS, but are also associated with lower FSH and higher LH levels.^17,18,27,28 The association between rs11031006 and PCOS risk disappears when the relationship is controlled for LH levels, suggesting that PCOS risk at the FSH locus is driven by gonadotropin dysregulation.¹⁸ The rs11031006 and s11031005 risk variants are located 26 kilobases proximal to the FSHβ promoter.^17,18,27 The region of interest is located approximately 500 bases downstream from a histone H3 mono-methylation of the 4^th lysine residue (H3K4Me1) enhancer site and 2,000 bases upstream from open chromatin sites (identified by DNase-Seq). H3K4Me1 and histone H3 acetylation of the 27^th lysine residue (H3K27Ac) enhancer sites were identified at rs11031006 in a lymphoblastoid cell line, along with a H3K4me1 enhancer site in 2 stem cell lines and induced pluripotent stem cells, pointing to the role of the region in gene regulation.^29,30 Co-accessibility analysis demonstrated an open, putative cis-regulatory region at the site and disruption of the conserved locus in a mouse gonadotrope cell line resulted in increased Fshb expression and FSH protein secretion.³¹ Expression of an additional candidate gene located within the same linkage disequilibrium block, ADP ribosylation factor like GTPase 14 effector protein (ARL14EP), did not change.^31–33 Taken together, the human DNA features, co-accessibility study and mouse experimental cell line data provide evidence that this risk locus affects a regulatory region for FSHβ. The additional effect of the risk variants on LH levels may be the result of decreased estradiol or the relative excess of Cga, because in mice with Fshb deleted, Lhb expression and LH secretion is increased.^34,35 Ongoing research will determine how the region regulates FSH secretion, focusing on the stimulatory hormones involved, transcription factors affected and three-dimensional interactions with the FSH promoter.³¹

There is growing evidence that some of the genetic variants play a role in PCOS risk through testosterone regulation. The IRF1/RAD50 locus was the only locus associated with testosterone levels in the European GWAS.²⁰ However, the association was not found in a GWAS of testosterone levels,³⁶ and may therefore play a role through indirect effects on testosterone or follicle number. Instead, the GWAS of testosterone levels found THADA and FSHβ loci are important testosterone regulators in women.³⁶ The relationship between the FSHβ locus and testosterone may occur through the elevated LH levels, whereas the THADA PCOS risk locus is found approximately 150bp from a H3K27Ac enhancer site and may, therefore, also mark a regulatory site.³⁰

Other work has examined the genetic locus containing DENND1A, a risk locus that has been replicated in Han Chinese and European women.^9,20,22 The gene variant within the intron of DENND1A is associated with hyperandrogenism, defined using both clinical and biochemical criteria.^19,20 When DENND1A.v2 splice variant found at increased levels in theca cells from women with PCOS was over expressed in a theca cell line, androgen biosynthesis and CYP17A1expression increased, whereas knockdown of the transcript in PCOS theca cells reduced androgen production and CYP17A1 mRNA.³⁷ However, the heterozygote in a mouse model in which DENND1A was deleted had no fertility or adult phenotype.³⁸ The homozygous deletion was embryonic lethal, but did result in abnormal brain development and a decrease in primordial germ cell number and differentiation in the fetus.³⁸ In another mouse model expressing human DENND1A.v2 mRNA, protein was not measurable, but CYP17A1 mRNA expression increased, as did androstenedione secretion from ex vivo theca cells.³⁹ However, DENND1A variants have not been associated with testosterone levels, per se, in any study. Taken together, the variants in the intron of DENND1A affect the degree of hyperandrogenism, but it remains to be determined whether the PCOS variants are acting through DENND1A mRNA levels or whether the variant might have a role in regulating another nearby gene.

Mendelian Randomization – What does it tell us?

Mendelian randomization is a statistical approach used in genetic studies to determine whether an exposure, measured by a genetically influenced trait such as obesity, has a causal effect on a disease of interest (PCOS).⁴⁰ The approach quantifies whether the genetic risk for the exposure influences the presence of the disease. The approach assumes that the trait or exposure (obesity) has a completely independent genetic architecture from PCOS and that any genetic risk for obesity should be randomly distributed in genomes from women with PCOS because the risk genes are not linked, i.e. located in very close in proximity on the chromosomes. In a simplified manner, any obesity gene risk variant would be found at equal frequency in PCOS cases and controls if there was no causal relationship between obesity and PCOS.

Using the Mendelian randomization approach, genetic determinants for traits including BMI, fasting insulin, depression, male pattern balding and age at menopause are found at increased frequency in PCOS.²⁰ These traits are therefore considered causal for PCOS, reinforcing previous observational studies. The male pattern balding thus presents itself as a male phenotype for PCOS, as has been hypothesized in the past.⁴¹

Mendelian randomization has been approached in the reverse fashion, examining whether PCOS causes obesity. In the reverse sense, the exposure is taken as PCOS and the outcome obesity.⁴² However, the PCOS exposure is based on the relatively small GWAS, to date. More accurately, one could look at testosterone levels, which are heritable and a critical component of PCOS.³⁶ A large Mendelian randomization study of genetic variants determining testosterone levels confirmed the causal relationship between higher testosterone and PCOS.³⁶ The study also demonstrated a causal relationship between bioavailable testosterone and type 2 diabetes in women, likely driven by SHBG levels, which are known to be inversely related to insulin resistance.³⁶ In men, genetic determinants of testosterone were associated with decreased type 2 diabetes risk in the same study. These data point to the possibility that PCOS risk through intrinsic increases in testosterone may not have the same metabolic risk in men in the same families.

Genetic Risk Scores – Are they clinically useful for PCOS diagnosis?

Ideally, clinicians need a test that will predict risk of PCOS before the onset of symptoms, so that they can intervene early. Polygenic (genetic) risk scores for PCOS could provide the needed predictive information. A polygenic risk score is a construct of the probability of disease posed by common genetic risk variants, with each variant weighted based on the relative influence of the variant on the disease. The area under the curve of the receiver operator characteristic (ROC) measures the discriminatory power of a polygenic risk score to detect disease. The closer the area under the ROC curve to 1, the better the disease discrimination. As an example, the predictive value of the polygenic risk score for coronary artery disease using common genetic variants is 0.79–0.8 and identifies 8% of the population at 3-fold risk for coronary artery disease.⁴³ The polygenic risk score of common variants identifies up to 20 times greater numbers than familial hypercholesterolemia rare variants for coronary artery disease and outperformed risk detection from clinical risk assessment (family history, cholesterol levels hypertension).⁴³ Although a genetic risk score can be used at any age, one could add historical or phenotypic risk factors such as BMI to improve its predictive ability.

These models depend on the availability of large GWAS studies, ensuring that the known genetic risk variants capture a sufficient proportion of the genetic risk for PCOS. Unfortunately to date, PCOS GWASs have not reached these threshold levels in women of European and Han Chinese ancestry and have not begun to examine genetic risk in other ethnic populations to any significant degree. A genetic risk score using the most up to date GWAS resulted in an area under the ROC curve of 0.54 to 0.72 for detecting PCOS in women of European ancestry.^20,44 The best model did not perform as well in women of African American ancestry with the ROC area under the curve only 0.54.⁴⁴ Therefore, the discriminatory value of these calculated genetic risk scores are not yet clinically useful.

Understanding the discriminatory limitations in PCOS, the polygenic risk score was used in a phenome-wide association study to identify comorbid PCOS disease from electronic medical records using ICD codes.⁴⁴ The phenotypes are hypothesized to be associated with PCOS risk as they are identified using the polygenic risk score marking predisposition for PCOS. When the polygenic risk score was used to identify associated phenotypes across an electronic dataset of over 120,000 subjects, the score was associated with polycystic ovaries, as expected. The score was also associated with morbid obesity and type 2 diabetes in men and women, along with hypercholesterolemia, disorders of lipid metabolism, hypertension, and sleep apnea in women.⁴⁴ In a second phenome-wide study, an association was found between three individual genome-wide significant risk variants and mental health disorders, providing support for the previous Mendelian Randomization studies.²²

Even with future development of an accurate polygenic risk score for PCOS, it is not clear that a predictive test will be sufficient to change behavior of patients and physicians to prevent disease. For example, obesity is a causal factor for PCOS risk.²⁰ Nevertheless, previous studies examining the possibility that genetic risk will change behavior determined that behavior was not changed by predicted risk.⁴⁵

Precision Medicine for PCOS

Precision medicine integrates genetic, environment and lifestyle risks to stratify patients and provide individualized diagnosis and treatment plans. These individualized diagnosis and treatment plans are targeted to the underlying cause of the disease and should improve patient outcome. Ideally, precision medicine will break PCOS down into tractable subsets based on disease risk. While no striking sub phenotypes emerged when examining the components of the PCOS diagnosis,²⁰ risk categories are beginning to stand out. The FSHβ locus points to a gonadotropin etiology.^18,20 The DENND1A, THADA and IRF1/RAD50 loci are associated with hyperandrogenism or testosterone levels.^20,36 The ERBB4, YAP1, and ZBTB16 loci are associated with ovulatory dysfunction and polycystic ovary morphology, perhaps alluding to a role in follicular function and fertility.²⁰ Whereas the polygenic risk score measures the contribution of genetic risk from all potential pathways combined, it is possible that splitting risk up into individually contributing pathways may help target treatment and symptoms. Larger studies are needed to perform such an analysis.

Clinically, precision medicine should improve treatment. When we examined the LH and FSH response to GnRH stimulation (n=14), women with PCOS who carried the FSHβ rs11031006 risk allele had increased mean LH levels (28.3±0.04 vs. 22.1±5.5 IU/L; p<0.05) and an increased GnRH-stimulated LH response (LH area under the curve 19,624±2,336 vs. 7,362±2,179 IU/L min-1; p<0.05).⁴⁶ In addition, carriers of the rs11031006 risk variant demonstrated a decreased ovulatory response to clomiphene citrate (X²=7.3; p=0.007).⁴⁷ Together, these data point to an altered gonadotropin response to GnRH stimulation and estrogen receptor blockade that may be detrimental for fertility treatment in women with PCOS who already have high LH levels. Developing specific treatments to precisely manipulate the FSHβ locus could target therapy to increase FSH and decrease LH secretion.

Two lines of evidence point to obesity as a critical causal factor for PCOS. The Mendelian randomization studies demonstrate obesity as a causal factor.²⁰ The phenome-wide association study demonstrates it as a comorbid association.²² Obesity also worsens the hyperandrogenism and menstrual irregularity.^4,48 Therefore, targeting obesity in those at particular genetic or environmental risk could prevent the full PCOS phenotype or prevent disease altogether.

Implementing precision medicine to diagnose and treat PCOS will depend on a number of factors. First, we need to capture a greater component of genetic variation in GWAS. The size of the current PCOS GWAS cohorts are relatively small compared to GWASs that have been performed for CAD and type 2 diabetes.^20,43,49 PCOS GWAS are also limited in ethnic diversity and depth.⁴⁹ We will need to incorporate lifestyle factors such as weight and weight trajectory, which can influence the clinical manifestations of PCOS.^44,50 We also need to determine whether environmental factors play a role, as has been suggested by exposures to endocrine disrupting chemicals and changes in the individual’s microbiome.⁵¹ A comprehensive look will need to capture electronic health data, genomics, proteomics, metabolomics, environmental exposures and tracked or self-reported data.

Key Points.

• Using available genetic data, the current polygenic risk scores do not reach sufficient diagnostic ability for clinical use.

• Subtypes of PCOS may be identified through analysis of genetic risk variants.

• Precision medicine for PCOS requires larger studies and the ability to integrate many variables into diagnostic decision making and treatment.

Synopsis.

PCOS is a complex genetic disorder with many genetic loci contributing small risk for disease. Large genome-wide association studies (GWAS) have identified 21 genetic risk loci for PCOS in European and Han Chinese women. The genetic architecture is similar across PCOS diagnostic categories. The next wave of analysis will incorporate large genotyped datasets linked to medical records, increasing the numbers and incorporating additional ethnic subsets. The resulting genetic risk loci can then be used to create genetic risk scores to identify PCOS before its onset. The genetic data can be further incorporated with clinical information, environmental and lifestyle data for a precision medicine approach to PCOS diagnosis and treatment.