Replication of Association of a Novel Insulin Receptor Gene Polymorphism with Polycystic Ovary Syndrome

Abstract

Objective

To evaluate association with polycystic ovary syndrome (PCOS) of 295 variants in 39 genes central to metabolic insulin signaling and glycogen synthase kinase-3β (GSK-3β) regulation, followed by replication efforts.

Design

Case-control association study, with discovery and replication cohorts.

Setting

Subjects were recruited from reproductive endocrinology clinics, controls were recruited from communities surrounding the University of Alabama at Birmingham and Erasmus Medical Center.

Patients

273 cases with PCOS and 173 controls in the discovery cohort; 526 cases and 3585 controls in the replication cohort. All subjects were Caucasian.

Interventions

Phenotypic and genotypic assessment.

Main Outcome Measures

295 SNPs, PCOS status.

Results

Several SNPs were associated with PCOS in the discovery cohort. Four insulin receptor (INSR) SNPs and three insulin receptor substrate 2 (IRS2) SNPs associated with PCOS (P<0.05) were genotyped in the replication cohort. One INSR SNP (rs2252673) replicated association with PCOS. The minor allele conferred increased odds of PCOS in both cohorts, independent of body mass index (BMI).

Conclusions

A pathway-based, tagging SNP approach allowed us to identify novel INSR SNPs associated with PCOS, one of which confirmed association in a large replication cohort.

Introduction

Insulin resistance and compensatory hyperinsulinemia are frequent findings in polycystic ovary syndrome (PCOS), affecting ~70% of cases (1). The fact that insulin-sensitizing agents ameliorate features of PCOS points to insulin resistance as a key pathophysiologic factor (2).

Not only is PCOS itself heritable, but within PCOS, insulin resistance is under significant genetic control (3). First degree relatives of women with PCOS exhibit an increased prevalence of insulin resistance, whether or not they have PCOS (4). For these reasons, some of the earliest candidate genes examined for PCOS were components of the insulin signaling pathway (5). However, insulin signaling, which starts with insulin binding to the insulin receptor (INSR) on the cell surface, is a complex system that interacts with other signaling pathways (6). INSR initiates metabolic effects (e.g. stimulation of glucose uptake and glycogen synthesis) via two main pathways, the phosphatidylinositol-3 (PI3) kinase cascade and the Cbl/CAP pathway (7); it also promotes cell proliferation via the mitogenic MAP kinase cascade. These are the best-understood immediate systems activated by the INSR. Given the considerable number of components of these systems, insulin signaling pathways have been incompletely examined to date. Examples of studied genes include AKT2, INSR, IRS1, IRS2, GSK3B, PTP1B, PPP1R3, and SORBS1 (8).

What has been needed is an analysis that considers all of the key components of the insulin signaling pathway, using a tagging SNP approach to encompass the majority of genetic variation across the target regions. Advances in high-throughput genotyping have made this possible. Therefore, to fully interrogate the insulin signaling pathway for susceptibility loci for PCOS, we genotyped the INSR gene plus 27 genes coding for pathway components related to metabolic effects (PI3-kinase and CAP/Cbl pathways) (Figure 1). We also genotyped 11 genes that have been described as regulators of glycogen synthase kinase-3β (9), given our prior genetic and physiologic results implicating this factor and its upstream regulator AKT2 in PCOS (10–12). The genes with the most significant SNPs associated with PCOS in the discovery phase were INSR and IRS2. Therefore, four SNPs in INSR and three SNPS in IRS2 were genotyped and analyzed in a larger replication cohort. One of the INSR SNPs replicated association with PCOS.

Figure 1.

Metabolic insulin signaling pathway genes genotyped in the discovery cohort. The protein products of the thirty nine genes that were genotyped in the discovery cohort are indicated. Names containing a slash indicate two isoforms encoded by two different genes. p = phosphate group. IP3 = phosphatidylinositol (3,4,5)-trisphosphate.

Materials and Methods

Subjects and Phenotyping

Discovery cohort

We studied 275 unrelated White PCOS patients and 173 White control women recruited at the University of Alabama at Birmingham (UAB). Study subjects were premenopausal, non-pregnant, on no hormonal therapy, including oral contraceptives or insulin-sensitizing agents, for at least three months, and met 1990 NIH criteria for PCOS (13). Parameters for defining hirsutism, hyperandrogenemia, ovulatory dysfunction, and exclusion of related disorders were previously reported (14). Supplemental Table 1 presents clinical characteristics.

Controls were healthy women, with regular menstrual cycles and no evidence of hirsutism, acne, alopecia, or endocrine dysfunction and had not taken hormonal therapy (including oral contraceptives) for at least three months. Controls were recruited by word of mouth and advertisements calling for “healthy women.”

All subjects gave written informed consent. The study was performed according to the guidelines of the Institutional Review Boards of UAB and Cedars-Sinai Medical Center.

Replication cohort

Replication efforts were carried out in an independent cohort consisting of 526 Caucasian PCOS women and 3585 unselected controls from the general population. The replication cohort included patients with oligo- (menstrual cycle ≥35 days) or amenorrhea (absence of vaginal bleeding for at least 6 months), serum FSH levels within the normal limits (1–10 U/l), i.e. normogonadotropic anovulation (classification according to the World Health Organization), and PCOS. PCOS was diagnosed according to the 2003 revised Rotterdam criteria (15). Patients with non-Caucasian ethnic background were excluded.

Controls were drawn from the Rotterdam Study, the design of which was described previously (16). In brief, this is a large population-based study of elderly subjects from a specific area near Rotterdam (Ommoord). All women with age of menopause >45 years and available DNA were included in the present analysis, providing a reference group of allele frequencies reflective of the local general Caucasian population (rather than a control group wherein PCOS was specifically excluded).

The study was approved by the institutional review board at the Erasmus Medical Center and informed consent was obtained from all patients as well as controls.

Genotyping

SNPs were selected using data from the Caucasian (CEU) subjects in the International HapMap database (release 24, http://hapmap.ncbi.nlm.nih.gov/) with the aim of exploiting linkage disequilibrium (LD) for the study of each gene. For all 39 genes, we selected SNPs to tag (r²>0.8) the majority of variation in each gene plus 10 kb upstream and 3 kb downstream or 5% upstream and 5% downstream (depending on gene size), resulting in a total of 341 SNPs. Supplemental Table 2 displays gene statistics; Supplemental Table 3 displays SNP statistics. The average SNP coverage was 74%. 338 of the 341 SNPs were genotyped using the oligo ligation assay (GoldenGate chemistry) on an Illumina (San Diego, CA) BeadStation (17). Six of the SNPs were not polymorphic, and one failed Hardy-Weinberg equilibrium (P<0.001); these SNPs were not considered further. SNPs were clustered in a single project that included all samples and utilized Illumina’s BeadStudio clustering algorithm. Each SNP was then manually reviewed to remove individual samples from each cluster plot that were not clearly assigned to one cluster (major or minor allele homozygotes or heterozygotes). SNPs that did not clearly form three clusters for each genotype group were removed from the analysis. Thirty-nine SNPs were removed because they failed this quality control. Thus, 292 SNPs were ultimately analyzed of the 338 genotyped on this platform. The genotyping success rate for these 292 SNPs was 93.6%. Duplicate genotyping of 10 subjects yielded 100% concordance.

The remaining 3 of the 341 SNPs (from the IRS1 gene) were genotyped using Applied Biosystems Taqman Assays-On-Demand (Applied Biosystems, Foster City, CA) according to manufacturer’s instructions. The overall genotyping success rate was 93.4%.

Four INSR and three IRS2 SNPs were genotyped in the replication study. For the PCOS cases, Taqman Allelic Discrimination (Applied Biosystems) was used for the INSR SNPs. For the Rotterdam Study controls, genotypes from two INSR SNPs (rs12459488, rs12971499) were extracted from the imputed genotypes GWAS database, while INSR SNPs rs2252673 and rs10401628 were genotyped using Taqman Allelic Discrimination. The IRS2 SNPs (rs7997595, rs7987237, rs1865434) were extracted from the imputed genotypes GWAS database.

We also genotyped 127 ancestry informative markers (18) (listed in Supplemental Table 4) in the discovery cohort and examined the cohort for population structure by generating principal components in Eigenstrat implemented in Golden Helix (Bozeman, MT). Plotting the first and second most informative principal components revealed that the cases and controls were of similar ancestry (Supplemental Figure 1), with the exception of two control subjects, who were excluded from analysis. None of the principal components were associated with PCOS status.

Statistical Analysis

Unpaired t-tests and chi-square tests were used to compare clinical characteristics between cases and controls; quantitative traits were log- or square-root-transformed as appropriate to reduce non-normality. Data are presented as median (interquartile range).

In the discovery cohort, allelic analyses were conducted using chi-square tests to compare allele frequencies between cases and controls for 295 SNPs. The same allelic analyses were carried out in the replication cohort (7 SNPs). We targeted for replication 7 SNPs in the genes with the most SNPs associated with PCOS at P<0.05. In the replication stage, SNPs were considered significantly associated with PCOS if the P-value was below 0.007 (0.05/7 ≈ 0.007).

To assess whether the effects of the seven associated SNPs were independent of BMI, adjusted analyses were conducted with logistic regression; the dependent variable was PCOS status, and the independent variables were genotype (dominant model) and BMI. Furthermore, we directly evaluated the seven SNPs for an effect on BMI by conducting linear regression wherein BMI was the dependent variable. Meta-analyses were conducted on the logistic regression results of the discovery and replication cohorts using the program METAL (http://www.sph.umich.edu/csg/abecasis/metal/).

The discovery cohort sample size has good power (≥80%) to detect association of risk alleles of frequency ≥0.2 with PCOS at odds ratio ≥1.75 and moderate power (40–60%) to detect association at odds ratio 1.5; the power to detect association of rare risk alleles (frequency ≤0.1) with PCOS at odds ratios of ≤1.5 is limited (19).

Results

In the discovery cohort (275 cases and 171 controls [2 controls excluded based on principal component analysis]), we analyzed 295 SNPs for association with PCOS (Supplemental Table 3). SNPs with allelic association with PCOS (P<0.05) are listed in Table 1. The gene with the largest number of SNPs significantly associated with PCOS in the discovery phase was INSR. IRS2 also contained several associated SNPs at the top of the list. These observations, and the key importance of INSR and IRS2 in the initial steps of insulin signaling, prompted us to attempt replication of the four INSR SNPs (rs12459488, rs12971499, rs2252673, rs10401628) and three IRS2 SNPs (rs7997595, rs7987237, rs1865434) most associated with PCOS in the discovery cohort. These seven SNPs were not associated with BMI, either in the entire discovery cohort (linear regression, adjusted for age and diagnosis) or in the PCOS women only (adjusted for age).

Table 1.

Allelic association results in the discovery cohort

Name	Gene	Chr	Alleles (major/minor)	Associated Allele	Case Frequency	Control Frequency	Chi square	P value
rs7997595	IRS2	13	C:G	C	0.881	0.795	11.281	0.001
rs10401628	INSR	19	G:A	G	0.919	0.856	8.298	0.004
rs2274490	SORBS1	10	T:C	C	0.430	0.334	7.428	0.006
rs7987237	IRS2	13	C:T	C	0.884	0.817	7.398	0.007
rs12459488	INSR	19	C:G	C	0.760	0.674	7.335	0.007
rs12971499	INSR	19	T:C	T	0.635	0.543	6.932	0.009
rs2252673	INSR	19	C:G	G	0.178	0.112	6.479	0.011
rs1865434	IRS2	13	A:G	A	0.875	0.814	5.875	0.015
rs4952834	RHOQ	2	G:C	C	0.338	0.261	5.500	0.019
rs3753242	PRKCZ	1	C:T	C	0.953	0.913	5.432	0.020
rs2267922	PIK3R2	19	C:G	C	0.539	0.460	5.002	0.025
rs3736328	PIK3R2	19	C:G	G	0.523	0.444	4.978	0.026
rs7207345	PRKCA	17	T:C	C	0.281	0.212	4.902	0.027
rs6510949	INSR	19	G:T	T	0.070	0.034	4.858	0.028
rs7641983	PIK3CA	3	C:T	C	0.832	0.772	4.610	0.032
rs7646409	PIK3CA	3	T:C	T	0.833	0.774	4.527	0.033
rs3094127	FLOT1	6	T:C	C	0.262	0.199	4.328	0.038
rs6020608	PTPN1	20	C:T	C	0.764	0.702	3.923	0.048
rs13420857	RHOQ	2	C:G	G	0.391	0.323	3.905	0.048
rs6443624	PIK3CA	3	C:A	C	0.771	0.711	3.848	0.050
rs7651265	PIK3CA	3	A:G	A	0.906	0.862	3.832	0.050

SNPs are displayed in order of significance. P values are for allelic association with PCOS (no covariate adjustment). INSR and IRS2 SNPs that were selected for genotyping in the replication cohort are indicated in bold.

Chr, chromosome

These seven SNPs were genotyped in the replication cohort consisting of 526 Caucasian PCOS women and 3585 controls. Association results are presented in Table 2. SNP rs2252673 replicated association with PCOS (P=0.002). Logistic regression analyses revealed that the association of rs2252673 was independent of BMI (Table 3). Carriers of the minor G allele at rs2252673 had an increased odds of PCOS in both the discovery cohort (BMI-adjusted odds ratio (OR)=2.06, 95%CI: 1.21–3.52, P=0.008) and the replication cohort (BMI-adjusted OR=1.32, 95%CI: 1.08–1.60, P=0.006). P-value meta-analysis of the two cohorts for this SNP revealed a combined P=0.00058 (Table 3). The significant associations utilizing the dominant model (Table 3) exhibited similar significance in the additive model.

Table 2.

Allelic association of four INSR and three IRS2 SNPs with PCOS in the replication cohort

SNP	Gene	Case MAF	Control MAF	Chi Square	P value
rs12459488	INSR	0.292	0.288	0.58	0.81
rs12971499	INSR	0.432	0.412	1.38	0.24
rs2252673	INSR	0.209	0.170	9.56	0.002
rs10401628	INSR	0.127	0.125	0.02	0.89
rs7997595	IRS2	0.173	0.166	0.186	0.666
rs7987237	IRS2	0.151	0.153	0.039	0.842
rs1865434	IRS2	0.153	0.153	0.014	0.907

Allelic P values are derived from chi-square testing.

MAF, minor allele frequency

Table 3.

Logistic regression association of four INSR and three IRS2 SNPs in the discovery and replication cohorts

SNP	Gene	Discovery cohort			Replication cohort			Meta-analysis
		OR	95% CI	P value	OR	95% CI	P value	P value
rs12459488	INSR	0.62	0.39–0.98	0.04	1.11	0.92–1.34	0.30	0.71
rs12971499	INSR	0.68	0.42–1.11	0.12	1.05	0.86–1.28	0.067	0.21
rs2252673	INSR	2.06	1.21–3.52	0.008	1.32	1.08–1.60	0.006	0.00058
rs10401628	INSR	0.45	0.25–0.79	0.006	1.00	0.80–1.25	0.99	0.40
rs7997595	IRS2	0.54	0.33–0.89	0.015	1.09	0.84–1.41	0.54	0.84
rs7987237	IRS2	0.54	0.33–0.90	0.019	1.03	0.79–1.35	0.84	0.58
rs1865434	IRS2	0.58	0.35–0.95	0.032	1.05	0.80–1.37	0.73	0.72

Logistic regression models included PCOS status as the dependent variable, and genotype (dominant model) and BMI as independent variables. The P value meta-analysis was conducted on logistic regression models in the discovery and replication cohorts.

OR, odds ratio; 95% CI, 95% confidence interval

We also analyzed the association of the INSR SNP in the replication cohort with inclusion of only the 286 cases meeting the NIH 1990 definition of PCOS. With the halving of the number of cases in this analysis, statistical significance was diminished, but the odds ratio of association with PCOS for rs2252673 (BMI-adjusted OR=1.34, 95%CI: 1.03–1.74, P=0.03) was identical to that in the entire replication cohort.

Discussion

We have replicated association with PCOS of an INSR variant that predisposes to the syndrome. This result sheds light on the origins of PCOS as an inherited condition characterized by insulin resistance in many affected individuals.

Genetic variation in INSR may affect PCOS risk via an effect on insulin signaling. This was recognized as a possibility early in the field of PCOS genetics. Initial studies screening exons of the INSR gene (typically by polymerase chain reaction and single-stranded conformational polymorphisms, PCR-SSCP) found missense/nonsense mutations only in isolated cases of women with PCOS-like phenotypes (5). Such screens commonly identified only a silent C/T variation in codon His1058, leading many to conclude that INSR variation was not a major factor in PCOS.

In subsequent years, several genetic association studies of INSR in PCOS case/control cohorts were published, most of which focused on His1058 C/T variant; a recent meta-analysis of these studies revealed a non-significant trend (OR 1.28, 95% CI 0.88–1.85) for association of this silent variant with PCOS (5). In Chinese subjects, a Cys1008Arg variant in exon 17 was associated with PCOS and lower insulin sensitivity index (20); however, the frequency of this variant in other populations has not been determined (not available in dbSNP, http://www.ncbi.nlm.nih.gov/SNP/). A Korean study that genotyped nine variants (initially found by sequencing) found association of a novel intron 21 variant with PCOS; this is another variant not yet listed in dbSNP (21). Thus, we cannot determine whether our replicated INSR SNP is in LD with either of these variants. The microsatellite D19S884, a replicated PCOS locus, was initially selected based on proximity to INSR; however, it resides in a different gene (FBN3) over 850 kb from INSR and therefore likely represents an independent susceptibility locus (22).

The Korean study genotyped rs2252673, and found that carriers of the G allele (major allele in Koreans, minor allele in Caucasians) exhibited a trend to increased PCOS (OR 1.18, 95% CI 0.53–2.26) (21), consistent with our findings for this SNP. This suggests that this variant might influence PCOS risk across three continents (North America, Europe, Asia). If PCOS is an ancient disorder, as recently proposed, one would predict shared susceptibility variants across the globe (23). Further studies of this variant are needed to establish whether rs2252673 is such a universal risk factor for PCOS.

How the replicated variant affects INSR function and/or expression is unknown. SNP rs2252673 is not in LD (r² > 0.8) with any other SNP in the INSR gene in the HapMap Caucasian database (24). As an intronic variant, it may be in LD with unknown functional (or not yet genotyped in HapMap) variants elsewhere in the gene. Sequencing may identify coding variants in LD with the associated variant. Alternatively, the variant itself may influence INSR transcription or mRNA splicing. In the TESS database (Transcription Element Search System, http://www.cbil.upenn.edu/tess), rs2252673 results in alternative possible transcription factor binding sites. Two forms of the INSR arise from alternative splicing of exon 11 (excluded from the A isoform and included in the B isoform). The A isoform, expressed mainly during fetal development and the adult brain, has low affinity for insulin but high affinity for insulin like growth factor 2; the B isoform has high affinity for insulin and is expressed in insulin responsive tissues such as muscle, liver and adipose (25). Whether the replicated INSR SNP affects INSR splicing and/or transcription must be determined by biochemical studies. Tissue-specific effects (e.g. muscle, liver, ovary) could also be evaluated.

The large size of the INSR gene (>180 kb) has challenged its investigation as a candidate gene. Prior studies only looked at one or a few variants in the gene, leading to incomplete coverage of its genetic variation. This is the first study that tagged the entire INSR region using information from HapMap (24). This was critical to allow the discovery phase to detect INSR variants potentially associated with PCOS.

Efforts to replicate genetic association results have been infrequent in the field of PCOS genetics. Examples include replication studies of HSD17B5 (26), HSD17B6 (27), CYP11A1 (28), the insulin gene (29), and others (22, 30). A key factor in our successful replication was that the replication cohort was substantially larger than the discovery cohort. Genetic effect size is typically overestimated in initial studies, a phenomenon known as the “winner’s curse” (31). Therefore, replication cohorts should be larger than discovery cohorts to have sufficient power to detect the association at a lower effect size. This proved to be true in our study, wherein the odds ratio for association with PCOS was higher in the discovery cohort than the replication cohort. Another possible explanation for the different effect sizes is that PCOS was excluded from the discovery cohort controls (increasing power for discovery), whereas the replication cohort controls consisted of an unselected sample representative of the general population. The latter large (>3500 subjects) population-based sample may indicate broad generalizability of our findings.

Ideally, both the discovery cohort and replication cohort PCOS cases would have been diagnosed using the same criteria. This was not the case because these are pre-existing cohorts. Fortunately, this did not hinder our ability to replicate at least one SNP association with PCOS. We ruled out BMI disparity between cases and controls as a cause of spurious SNP association with PCOS by directly ruling out SNP association with BMI and by adjusting for BMI as a covariate.

We did not adjust the results in the discovery cohort for multiple testing (e.g. Bonferroni correction) because we planned to follow-up with a replication effort. Replication is the optimal solution to multiple testing. We corrected for multiple testing in the replication phase. Given their central roles in insulin signaling and preponderance of significant SNPs in the discovery phase (Table 1), we first attempted to replicate variants in INSR and IRS2. In the future, we will attempt to replicate SNPs in other loci associated with PCOS in the discovery cohort. We hope that this publication inspires other investigators to also examine these loci in their cohorts.

In conclusion, in a survey of 39 genes coding for the central components of metabolic insulin signaling pathways, we identified SNPs in the INSR and IRS2 genes that were associated with PCOS. One of these SNPs replicated association with PCOS in a larger, independent cohort. These data should reinvigorate interest in INSR as a genetic susceptibility factor for PCOS.