Genotyping in Cohort A was carried out using the Applied Biosystems Assays by Design (ABD) or Assays on Demand (AOD) 5′nuclease Taqman technology (Applied Biosystems, Foster City, CA) as recommended by the manufacturer on the 7900HT DNA analysis system (Applied Biosystems). Duplicate genotyping of a HapMap CEU trio yielded a 99.7% concordance rate. The genotyping success rate was 97%. All SNPs were in Hardy-Weinberg equilibrium. We genotyped seven markers, consisting of the SNPs most associated with PCOS in the Chinese GWAS or proxies for these SNPs (). Rs13405728 maps to 2p13.6 (LHCGR
), rs12468394, rs13429458, rs6544661, and rs11891936 map to 2p21 (THADA)
, and rs2479106 and rs10818854 map to 19q33.3 (DENND1A).
Rs6544661 was used as a proxy for rs12478601 with which it is in complete LD (r2
=1) in the CEU population of the HapMap database (release 24, http://hapmap.ncbi.nlm.nih.gov/
). A third SNP, rs10986105, associated with PCOS in the Chinese GWAS was not genotyped herein, because, unlike in Asian populations, in European populations it is in high LD with rs10818854 (r2
=0.83 in HapMap CEU). In some cases, technical reasons necessitated genotyping of proxy SNPs in linkage disequilibrium with the SNPs of interest; r2
values in are linkage disequilibrium (HapMap CEU) of the genotyped SNP with the Chinese GWAS SNP. Seven SNPs genotyped in each cohort captured the variation of the eight SNPs of interest.
Genotyping of SNPs most significantly associated with PCOS in the Chinese GWAS
In Cohort B, genotyping was carried out using iSelect Infinium technology, following the manufacturer’s protocol (Illumina, San Diego, CA) [16
]. SNPs were excluded if the genotyping failure rate was >10% or if the minor allele frequency was <3%. Duplicate genotyping of 12 samples yielded a 100% concordance rate. The genotyping success rate was 99.97%. A total of seven SNPs were examined (), all of which were in Hardy-Weinberg equilibrium. In the 2p13.6 locus (LHCGR
), we genotyped rs6732721, a proxy for rs13405728 (r2
= 0.87 in HapMap CEU). In the 2p21 locus (THADA
), we genotyped rs12468394, rs13429458, rs11891936, and rs6544661. In the 9q33.3 (DENND1A
) locus, we genotyped rs2479106 and rs12337273, a proxy for two of the Chinese GWAS SNPs (r2
=0.83 with rs10818854 and r2
=1 with rs10986105 in HapMap CEU).
Unpaired t-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. Quantitative trait data are presented as median (interquartile range).
Separately in Cohorts A and B, association analyses were conducted using logistic regression; the dependent variable was PCOS status, and the independent variable was genotype (additive model). To assess whether the effects of the seven SNPs were independent of BMI, adjusted analyses were conducted with inclusion of BMI as an additional independent variable. 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 Cohort A and Cohort B using inverse variance weighting.
We used the Genetic Power Calculator package to calculate the power to detect an association between SNPs tested and PCOS in our cohort [18
]. The allele frequencies in the power analysis are those for the PCOS associated allele (or corresponding proxy SNP alleles) at each SNP in the Hapmap CEU cohort (rs13405728 Allele G = 0.058; rs6732721 Allele C = 0.067; rs12468394 Allele A = 0.508; rs13429458 Allele C = 0.096; rs12478601 Allele T = 0.600; rs11891936 Allele A =0.200; rs6544661 Allele G = 0.600; rs2479106 Allele G = 0.300; rs10818854 Allele A = 0.050; rs10986105 Allele C = 0.042; rs12337273 Allele G = 0.034). Other parameters used for this analyses were: 957 controls, 939 cases for cohort A, 845 controls, 535 cases for cohort B, 1802 controls, 1474 cases for the complete cohort, genotype relative risk of 1.5 and 2.0 under an additive model. Assuming these parameters, we had 88% power to detect an effect at p< 1×10−4
in the complete cohort for at least one SNP mapping to DENND1A
with genotype relative risks of 1.5 and 2.0. For LHCGR
we had 57% power to detect an effect at p = 0.01 in the complete cohort for a genotype relative risk of 1.5 and >80% power detect an effect at p = 0.01 in the complete sample for a genotype relative risk of 2.0. We therefore had modest to sufficient power to detect a relevant effect in our cohort.
Quantitative trait association analyses
Exploratory association analyses of the seven SNPs against BMI, total testosterone, DHEAS, fasting insulin, and fasting glucose were conducted within each cohort. Because these analyses are not in replication of results from the Chinese GWAS, we applied a multiple testing corrected p value of 0.001 (=0.05/35; accounting for seven SNPs against five traits).
Allele frequencies and genotype frequencies for each SNP are shown in . In Cohort A, of the seven SNPs genotyped, four variants (three in THADA, one in DENND1A) were associated with PCOS (). These associations remained significant after adjustment for BMI (). The highest level of significance and greatest effect size were observed for the DENND1A SNP rs10818854. This was the only SNP associated with PCOS in Cohort B, with a similar odds ratio; adjustment for BMI did not materially alter this association ( and ). None of the seven SNPs was associated with BMI itself.
Allele and genotype frequencies
Unadjusted genetic association analyses
BMI-adjusted genetic association analyses
Meta-analysis results of the two cohorts are shown in and . The DENND1A SNP rs10818854 was highly associated with PCOS (unadjusted P=9.8 × 10−8, BMI-adjusted P=6.5 × 10−8). Three SNPs in the THADA locus, rs12468394, rs6544661, and rs11891936, were significantly associated with PCOS at lower levels of significance ( and ). The magnitude and direction of effects of the European PCOS associated DENND1A and THADA SNPs were similar to those observed in the Chinese GWAS ().
None of the seven SNPs were associated with total testosterone, DHEAS, fasting insulin, or fasting glucose (data not shown).
In this study, we performed a meta-analysis of two case control cohorts examining whether variants recently identified in a GWAS for PCOS in Chinese Han subjects would be associated with PCOS in non-Hispanic whites. Of the seven SNPs tested, one, rs10818854 in the DENND1A
gene, was highly associated with PCOS in the present analysis; three SNPs in THADA
were associated with PCOS in the meta-analysis. Finding variants in two of three genes associated with PCOS in both Chinese and Europeans is not unexpected, because the remarkably similar prevalence of PCOS around the globe suggests it might be an ancient disorder, for which the existence of common susceptibility genes and alleles in different races would be predicted [19
]. Indeed, among the four SNPs associated with PCOS in both Chinese and Europeans, the same alleles were associated with PCOS with similar odds ratios.
Until recently, the field of PCOS genetics was dominated by candidate gene studies, which examined over 100 genes of which only a few have been replicated [2
]. Given the modest success of candidate gene studies in PCOS, GWAS for this condition have been highly anticipated. The first was conducted in a cohort of cases and controls from China, with two levels of replication [3
]. That study identified SNPs in three loci, at chromosomes 2p13.1, 2p21, and 9q33.3, as replicated loci for PCOS. Genes found at these loci include LHCGR, THADA
, and DENND1A
, respectively. The three DENND1A
SNPs associated with PCOS in the Chinese GWAS, rs10818854, rs2479106, rs10986105, are independent (r2
< 0.7) of each other in Asian populations [3
]. Since in white populations, rs10818854 and rs10986105 are highly correlated (r2
= 0.83 in HapMap CEU), we examined only the former in this study and found it to be associated with PCOS.
(also known as connecdenn) encodes a protein involved in endosomal membrane trafficking [25
]. Its N-terminus contains a DENN (differentially expressed in neoplastic versus normal cells) motif, which is found in many proteins but whose function is uncertain [26
]. At the plasma membrane, DENND1A interacts with clathrin and the clathrin adaptor protein AP-1, via residues in the C-terminal end [27
]. DENND1A also functions as a guanine nucleotide exchange factor (GEF) for the small GTPase Rab35, serving to link Rab35 with clathrin-mediated endocytosis [28
]. DENND1A has been found in neuronal clathrin-coated vesicles, where it plays a role in synaptic vesicle endocytosis [27
]. DENND1A is ubiquitously expressed, with highest levels in kidney and brain [29
]. Genetic variation in DENND1A
has been associated with personality traits (rs7852296, P=9 × 10−6
]) and weakly with cleft lip/palate (rs1928482, P=0.03, in only one of multiple cohorts studied [31
]). Because a protein such as this would be expected to affect diverse processes, it is not surprising to find it associated with PCOS, a syndrome characterized by dysfunction in multiple organ systems (ovary, adrenal, hypothalamus, pituitary, insulin-responsive tissues). It has been speculated that DENND1A
may affect the development of PCOS via altered activity of endoplasmic reticulum aminopeptidase 1 [3
While it is possible that DENND1A
contributes directly to the PCOS phenotype, alternatively the positive association signal may be related to variation in another gene in linkage disequilibrium with the positive SNP. One such candidate gene encodes a microRNA, mIR60, which is co-localized with DENND1A
. mIR601 up-regulates expression of actin cytoskeleton, down-regulates the Fas-induced apoptosis pathway, and represses nuclear factor-kappaB transcription factor-dependent reporter expression [32
]. The control of these signalling pathways might directly or indirectly contribute to the PCOS phenotype.
DENND1A has a paralog, DENND1C, located on chromosome 19p13, a region that we previously suggested harbors a risk allele for PCOS [21
]; however, SNPs in the DENND1C
gene were not associated with PCOS (data not shown).
Multiple SNPs in the THADA
locus were associated with PCOS. Of interest, rs7578597, a missense variant in THADA
, was associated with type 2 diabetes in a large GWAS meta-analysis [4
]. However, the THADA
SNPs associated with PCOS in our data are not in LD with the diabetes variant in the HapMap and 1000 Genomes Project databases [33
]. This resembles the situation with the gene TCF7L2
, wherein different variants affect diabetes susceptibility and PCOS susceptibility [34
Of the three genes considered, only LHCGR
did not show association with PCOS in our European-derived cohorts. The SNP genotyped is rare in Europeans (minor allele frequency 0.05 in HapMap CEU) but common in Han Chinese (frequency 0.23 in HapMap) resulting in reduced power to detect association of this SNP with PCOS in our cohorts, as our meta-analysis effect size (OR 0.8) was similar to that observed in the Chinese study (OR 0.7) [3
might contain susceptibility SNPs for PCOS in whites not tagged by SNP rs13405728 in Chinese. A similar finding was observed in the gene TCF7L2
, wherein different variants were associated with type 2 diabetes in whites and Chinese [35
]. Alternatively, the LD pattern in the Caucasian populations may be sufficiently different from that in the Chinese so that the genotyped SNP is not detecting the same causal variant in the two populations. Comprehensive fine mapping is needed to evaluate this possibility in PCOS.
In conclusion, the GWAS era has finally arrived in PCOS genetics. By discovering loci that would not otherwise be considered in traditional candidate gene approaches, GWAS will open new avenues in genetic and physiologic research in PCOS. DENND1A and THADA appear to affect PCOS risk in at least two different racial groups.