Variants in DENND1A Are Associated with Polycystic Ovary Syndrome in Women of European Ancestry

Abstract

Context:

A genome-wide association study has identified three loci (five independent signals) that confer risk for polycystic ovary syndrome (PCOS) in Han Chinese women. Replication is necessary to determine whether the same variants confer risk for PCOS in women of European ancestry.

Objective:

The objective of the study was to test whether these PCOS risk variants in Han Chinese women confer risk for PCOS in women of European ancestry.

Design:

This was a case-control study.

Setting:

The study was conducted at deCODE Genetics in Iceland and two academic medical centers in the United States.

Patients:

Cases were 376 Icelandic women and 565 and 203 women from Boston, MA, and Chicago, IL, respectively, all diagnosed with PCOS by the National Institutes of Health criteria. Controls were 16,947, 483, and 189 women not known to have PCOS from Iceland, Boston, and Chicago, respectively.

Intervention:

There were no interventions.

Main Outcomes:

Main outcomes were allele frequencies for seven variants in PCOS cases and controls.

Results:

Two strongly correlated Han Chinese PCOS risk variants on chromosome 9q33.3, rs10986105[C], and rs10818854[A], were replicated in samples of European ancestry with odds ratio of 1.68 (P = 0.00033) and odds ratio of 1.53 (P = 0.0019), respectively. Other risk variants at 2p16.3 (rs13405728), 2p21 (rs12468394, rs12478601, and rs13429458), and 9q33.3 (rs2479106), or variants correlated with them, did not associate with PCOS. The same allele of rs10986105 that increased the risk of PCOS also increased the risk of hyperandrogenism in women without PCOS from Iceland and demonstrated a stronger risk for PCOS defined by the National Institutes of Health criteria than the Rotterdam criteria.

Conclusions:

We replicated one of the five Chinese PCOS association signals, represented by rs10986105 and rs10818854 on 9q33, in individuals of European ancestry. Examination of the subjects meeting at least one of the Rotterdam criteria for PCOS suggests that the variant may be involved in the hyperandrogenism and possibly the irregular menses of PCOS.

Polycystic ovary syndrome (PCOS) affects 7–10% of reproductive-age women, making it the most common endocrinopathy in this age group. Irregular menstrual cycles, hyperandrogenism, and polycystic ovarian morphology are cardinal features (1, 2). Obesity and insulin resistance are also common along with increased risk of diabetes, metabolic syndrome, and other cardiovascular diseases (3, 4). Despite the detrimental impact of the disorder on women's health, the etiology is poorly understood.

Twin studies suggest that genetic influences explain more than 70% of PCOS pathogenesis (5). Previous studies have demonstrated association between variants in more than 70 candidate genes and risk for PCOS, although the majority of these have not been replicated. A PCOS genome-wide association study in Han Chinese women identified three susceptibility loci for PCOS; at 2p16.3, 2p21, and 9q33.3 (6). Although one group recently replicated two of these loci (7), another did not (8). We examined the same susceptibility variants in additional PCOS case control sets of European ancestry. We also investigated variants that correlate strongly with the risk variants in the Chinese population for association with PCOS in the Icelandic sample set.

Materials and Methods

Study subjects

Subjects with PCOS in Iceland and Boston, MA, were recruited in tandem starting in 2003 (9). An additional sample set was recruited from Chicago, IL (Supplemental Table 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org) (10). All subjects were women of European ancestry. All PCOS subjects were aged 18–45 yr. Subjects with other disorders resulting in the same symptoms were excluded (9, 10). The study was approved by the Data Protection Commission of Iceland, the National Bioethics Committee of Iceland, and the Institutional Review Boards of the Massachusetts General Hospital and the University of Chicago. All subjects gave written informed consent.

The Icelandic samples were comprised of 376 women diagnosed with PCOS as defined by the National Institutes of Health (NIH) criteria (2) and 16,947 female control subjects who participated in other genetic studies at deCODE Genetics (Reykjavík, Iceland) and who had no known features associated with PCOS such as hyperandrogenism, irregular menses or a history of polycystic ovary morphology or PCOS (Supplemental Table 1). Subjects in Iceland with at least one feature of PCOS as defined by the Rotterdam criteria were included in the analysis of PCOS-related phenotypes. This included 1348 with hyperandrogenism, 404 with irregular menses, 554 with polycystic ovary morphology on ultrasound, and 684 who fulfilled the Rotterdam criteria for PCOS (Supplemental Table 2).

The 565 subjects from Boston and 203 from Chicago had PCOS defined by the NIH criteria (2, 9, 11, 12). Control subjects from Boston consisted of 483 women aged 18–45 yr with regular menses between 21 and 35 d and no hyperandrogenism (11). Control subjects from Chicago consisted of 189 healthy reproductive-age women, aged 18 yr and older (13). Additional subjects in Boston presented with one or more PCOS criteria without fulfilling the NIH criteria for a total of 608 with hyperandrogenism, 565 with irregular menses, 575 with polycystic ovary morphology on ultrasound, and 591 who fulfilled the Rotterdam criteria for PCOS.

Single-nucleotide polymorphism (SNP) selection

Seven SNP from the three loci reported by Chen et al. (6) and listed in Table 1 were selected for analysis in all sample sets. Additional SNP that correlated strongly [(r²) > 0.5] with any of those seven SNP in 197 Han Chinese individuals included in the October 2010 release of the 1000 Genomes Project (14) were studied in the Icelandic sample set.

Table 1.

Odds ratios and P values for the association of PCOS with seven SNP identified in a genome-wide association study of Chinese women

SNP allele	OR Chinese	Frq Chinese	Samples	P	OR	Ncases	Frqcases	Ncontrols	Frqcontrols	Pcombined	OR (95% CI)	PHet
rs13405728-A	1.41	0.726	Iceland	0.55	1.12	376	0.958	16,947	0.954	0.34	1.15 (0.86, 1.52)	0.98
2p16.3			Boston	0.50	1.20	554	0.949	483	0.940
			Chicago	0.69	1.16	201	0.963	186	0.957
rs12468394-C	1.39	0.695	Iceland	0.32	1.08	376	0.517	16,947	0.497	0.077	1.10 (0.99–1.23)	0.75
2p21			Boston	0.36	1.09	554	0.542	483	0.521
			Chicago	0.17	1.22	193	0.544	180	0.494
rs13429458-A	1.49	0.793	Iceland	0.27	1.15	376	0.892	16,947	0.879	0.60	1.05 (0.87, 1.27)	0.46
2p21			Bostona	0.96	1.01	490	0.885	297	0.884
			Chicago	0.43	0.84	188	0.867	180	0.886
rs12478601-C	1.39	0.686	Iceland	0.028	1.19	376	0.504	16,947	0.470	0.18	1.09 (0.96, 1.23)	0.19
2p21			Boston	0.67	0.95	544	0.436	481	0.450
			Chicago	0.71	0.94	201	0.410	184	0.424
rs10818854-A	1.51	0.079	Iceland	0.0099	1.73	376	0.057	16,947	0.037	0.0019	1.53 (1.17, 2.00)	0.65
9q33.3			Boston	0.10	1.40	565	0.060	479	0.044
			Chicago	0.36	1.41	203	0.044	189	0.032
rs2479106-G	1.34	0.216	Iceland	0.34	1.09	376	0.243	16,947	0.227	0.45	1.05 (0.93, 1.18)	0.49
9q33.3			Boston	0.50	1.07	559	0.310	477	0.297
			Chicago	0.42	0.88	201	0.306	188	0.332
rs10986105-C	1.47	0.069	Iceland	0.0043	1.87	376	0.055	16,947	0.034	0.00033	1.68 (1.27, 2.23)	0.39
9q33.3			Boston	0.011	1.76	554	0.056	482	0.032
			Chicago	0.96	1.02	202	0.037	178	0.036

For each SNP the table includes the OR and frequency of the risk allele in the Chinese study, P value, OR, number and frequency in cases and controls for the three sample sets, and P value and OR in the samples sets combined using a Mantel-Haenszel model (18) together with a P value, P_Het, for the test of heterogeneity in the effect estimates between the sample sets. The association P values are two sided and adjusted for genomic controls in Iceland (λ = 1.16). Freq, Frequency; CI, confidence interval.

^ars13429458 failed genotyping in the MALDI-TOF mass spectrometry Sequenom platform. Data are presented for the Centaurus (Nanogen) platform only.

Genotyping

The Icelandic samples were genotyped with the Illumina HumanHap300 or HumanHapCNV370 bead chips (Illumina, San Diego, CA) that included two of the seven risk variants, rs2479106 and rs10986105. Genotypes for the other five risk variants, and for correlated variants, were imputed into the Icelandic samples. For the imputation, only SNP present on both chips were used. SNP with a yield less than 95%, population minor allele frequency less than 1%, or deviation from Hardy-Weinberg equilibrium (P < 10⁻⁶) were excluded. Samples with call rates below 98% were excluded. As a reference for imputation, we used phased haplotypes of 1094 individuals of various ethnicities from the October 2010 release of the 1000 Genomes Project (14). The imputation was done using methods previously described (15).

Genotyping of the Boston and Chicago samples was performed on a Centaurus (Nanogen, San Diego, CA) platform (16). Each SNP assay was evaluated by genotyping the HapMap-CEU samples and comparing the results with the HapMap data. Mismatches were less than 1%. Positive and negative controls were present on all the genotyping plates to ensure correct genotyping. The Boston samples were also genotyped for confirmation by primer extension of multiplex products with detection by matrix-assisted laser desorption/ ionization-time of flight (MALDI-TOF) mass spectrometry using a Sequenom (San Diego, CA) platform. An additional 69 PCOS samples and 86 controls were also genotyped. All SNP had a call rate greater than 90% and Hardy-Weinberg equilibrium P > 0.05 in cases and controls. There was mismatch between Centaurus and MALDI-TOF mass spectrometry for five PCOS subjects and two control subjects, which were removed from analysis.

Statistical analysis

The NIH criteria for PCOS were used in the primary analysis for all three cohorts (2). For the Icelandic sample set, the association between SNP and PCOS was tested using SNPTEST version 2.2 (17), with an additive model and a likelihood score test to account for uncertainty in the imputation. To test the association conditional on each of the seven SNP, estimated genotype counts were included as covariates. Likewise, when adjusting for body mass index (BMI) and age, those variables were included as covariates. The P values were adjusted for relatedness of the Icelandic cases and controls using genomic control adjustment estimated using a genome-wide set of SNP present on the Illumina chips. For the Boston and Chicago sample sets, we tested the association using logistic regression. Results for the three sample sets were combined using a Mantel-Haenzel model (18), assuming the same effect but different population frequencies for the SNP. Heterogeneity was tested assuming a log-normal distribution for the effect estimates and using a likelihood ratio χ² test with degrees of freedom equal to the number of groups compared minus one. The combined effective sample size was estimated as the sum of the geometric mean of the number of cases and controls for each sample set, with the Icelandic contribution adjusted by dividing it by the genomic control adjustment factor.

We used the Bonferroni adjustment to account for the seven SNP tested and consider P < 0.05/7 = 0.007 significant. Because the additional markers investigated at each of the three loci are highly correlated through linkage disequilibrium, we used permutation of the cases and controls to estimate the significance threshold. Ten thousand permutation sets of the cases and controls were generated and the association for all the markers was repeated for each of those sets. From each set the most significant P value was recorded and the distribution of those P values used to estimate the empirical significance threshold. Linear regression using an additive genetic model was performed to test for association of PCOS risk variants with 22 log-transformed quantitative traits in PCOS cases and controls. Results were adjusted for BMI and/or age and were meta-analyzed using a fixed effects inverse-variance model. A P < 3.2 × 10⁻⁴ was considered significant after Bonferroni correction for 22 traits (testosterone, androstenedione, dehydroepiandrosterone sulfate, 17-hydroxyprogesterone, progesterone, SHBG, LH, FSH, LH to FSH ratio, prolactin, ovarian volume, follicle number, waist and hip circumference, glycosylated hemoglobin, systolic and diastolic blood pressures, fasting glucose and insulin, triglycerides, high-density lipoprotein, and low-density lipoprotein cholesterol) and seven SNP.

Results

We tested the seven SNP shown to associate with PCOS in Han-Chinese women (6) in three PCOS case-control samples of European ancestry. In the combined samples sets, all seven markers showed the same direction of effect as in the Chinese samples (Table 1), although only two of the variants, rs10986105 and rs10818854 at 9q33, associated with PCOS in the European samples with a combined odds ratio (OR) of 1.68 (P = 0.00033) and 1.53 (0.0019), respectively. Those two variants are strongly correlated both in the Chinese and European populations [(r²) > 0.65]. The results did not change with adjustment for BMI and age (Supplemental Table 3). Testosterone levels were higher with each rs12468394-A allele (β 0.067 ± 0.015; P = 9.9 × 10⁻⁶; Supplemental Table 4).

For the three variants on 2p21 and rs2479106 on 9q33, this study has more than 99% power to detect association, assuming seven SNP tested, the ORs observed in the Chinese population, the observed allele frequencies of the SNP in Europeans and the combined effective sample sizes of 1349 cases and 1349 controls and 90% power to detect association of ORs in the range 1.24 (for rs12468394) to 1.37 (for rs13429458) (Supplemental Table 5). For rs13405728, which has 5% minor allele frequency in Europeans, the study had only 61% power to detect association with the OR reported (6), and an OR of 1.57 is needed to have 90% power to detect association with this variant.

Because of potential differences in linkage disequilibrium between the Chinese and European populations, we investigated an additional 520 SNP at the three loci that strongly correlated with any of the seven SNP in the Chinese population. Using imputed genotypes for those SNP, we tested for association with PCOS in the Icelandic samples. None of these correlated SNP associated with PCOS given the empirical significance threshold of P < 0.0008, estimated based on 10,000 simulations (Supplemental Table 6).

We further examined association of rs10986105[C] with PCOS-related phenotypes in the Icelandic and Boston sample sets (Table 2). In the combined sample sets, the variant associated with hyperandrogenism (OR 1.47, P = 0.001) and irregular menses (OR 1.78, P = 0.00018) but not with polycystic ovary morphology (OR 1.07, P = 0.71). Therefore, there was a stronger association between the C allele and PCOS using the NIH criteria compared with the Rotterdam criteria. Within the group of Icelandic individuals with hyperandrogenism, the risk for the 689 subjects that did not fulfill the Rotterdam criteria for PCOS was similar to the risk for the 679 subjects that did fulfill the criteria (OR 1.37, P = 0.083) compared with an OR of 1.37 (P = 0.064).

Table 2.

Odds ratios and P values for association of PCOS-related phenotypes with rs10986105[C]. Results are shown for women from Boston and Iceland separately and combined

Phenotype	Samples	P	OR (95% CI)	Ncases	Frqcasesa	Ncontrols	Frqcontrols	PHet
Hyperandrogenism	Iceland	0.025	1.37	1,348	0.045	16,947	0.034
	Boston	0.013	1.73	608	0.053	482	0.032
	Combined	0.001	1.47 (1.17–1.85)					0.36
Irregular menses	Iceland	0.0064	1.79	404	0.054	16,947	0.034
	Boston	0.011	1.76	554	0.056	482	0.032
	Combined	0.00018	1.78 (1.31–2.40)					0.96
Polycystic ovary	Iceland	0.72	1.07	554	0.037	16,947	0.034
Morphology	Boston	0.86	1.05	575	0.044	259	0.042
	Combined	0.71	1.06 (0.78–1.44)					0.97
Rotterdam PCOS	Iceland	0.067	1.37	684	0.045	16,947	0.034
	Boston	0.01	1.72	591	0.053	482	0.032
	Combined	0.0023	1.50 (1.16–1.95)					0.40

The association P values are two sided and adjusted for genomic controls in Iceland (λ = 1.63, 1.20, 1.19, and 1.26 for hyperandrogenism, irregular menses, polycystic ovary morphology, and Rotterdam criteria, respectively). The table also includes the number of cases (referring to subjects with the trait indicated) and controls used, the frequency of rs10986105[C] in cases and controls, and a P value, P_Het, for the test of heterogeneity in the effect estimates between the samples from Boston and Iceland. Freq, Frequency; CI, confidence interval.

^aCases refer to subjects with the trait indicated in phenotype.

Discussion

We show that rs10986105 and rs10818854 on 9q33.3 (in linkage disequilibrium in DENND1A), previously shown to associate with PCOS in Han Chinese women, also associate with PCOS in women of European ancestry. Although previous studies did not examine the variant with the strongest association directly (rs10986105), it is of note that our second strongest risk variant (rs10818854) was the strongest risk variant identified previously (7). The phenotypic features associated with rs10986105 provide clues to its role in the etiology of PCOS because the variant was associated with hyperandrogenism, even in women without PCOS from Iceland, and was associated with irregular menses in women from Iceland and Boston. The variant was not related to a specific androgen level, perhaps because it reflects a more complex relationship with serum androgen levels or the clinical features of hyperandrogenism. It was not associated with polycystic ovary morphology and so demonstrated a stronger risk for PCOS defined by the NIH criteria than the Rotterdam criteria. Therefore, it may play a role in hyperandrogenism and possibly reproductive function but does not appear to be related to ovarian size or follicle number.

The variant with the strongest association, rs10986105, is located within the third intron of the DENND1A gene, which regulates Rab GTPases (19). The Rab proteins are important for calcium regulated exocytosis in pituitary cells and Rab3b is essential for basal and GnRH-induced gonadotropin release (20). Thus, it is possible that the variant in DENND1A is involved in exocytosis of gonadotropins.

The current data did not replicate the association at the 2p16.3 (in linkage disequilibrium with LHCGR) and the 2p21 (THADA) loci or the second signal at 9q33.3 (rs2479106 in DENND1A). It is possible that the effects of the risk variants are much lower in the European population compared with the Chinese population and hence not detected in this study. It is also possible that the linkage disequilibrium between the associated variants and possible functional variants may be disrupted while moving from Han Chinese to Europeans, although our investigation of correlated variants in those regions did not identify other risk variants associated with PCOS. Finally, it is also possible that some of the association with PCOS identified in the Han Chinese population is ethnicity specific, although this was not the case for at least one of the variants at 2p21 in THADA, rs12468394-C, which was found to associate with increased risk of PCOS in a European population (7). We note that our result for this variant, although not significant, is consistent: OR 1.10 (P = 0.077) compared with OR 1.19 (P = 0.0006). For the two studies, the combined OR of 1.15 (P = 0.0002, P_het = 0.29) (7).

The data presented here provide additional replication of a Han Chinese PCOS risk variant in women of European ancestry. The gene in which the variant is located is an interesting candidate gene for PCOS, particularly in relation to gonadotropin release and hyperandrogenism. Further studies will be needed to determine the role of the variant in the pathogenesis of PCOS. The lack of demonstrable association with PCOS for the other four Chinese signals suggests that the underlying genetic risk may differ between these two ethnicities.