Common variants in the CYP2C19 gene are associated with susceptibility to endometriosis

Abstract

Objective

To follow-up previous studies highlighting a possible role for cytochrome P450, family 2, subfamily C, 19 (CYP2C19) in susceptibility to endometriosis by searching for additional variants in the CYP2C19 gene that may be associated with the disease.

Design

Case-control study.

Setting

Academic research.

Subject(s)

Cases = 2,271 women with surgically confirmed endometriosis; Controls = 939 women with self-report of no endometriosis and 1,770 unscreened population samples.

Intervention(s)

Sequencing of the CYP2C19 region and follow-up of 80 SNPs in two case-control samples.

Main outcome measure(s)

Allele frequency differences between cases and controls.

Results

Sequencing of the CYP2C19 gene region resulted in the detection of a large number of known and novel SNPs. Genotyping of 80 polymorphic SNPs in 901 endometriosis cases and 939 controls resulted in study-wide significant association signals for SNPs in moderate or complete LD with rs4244285, a functional SNP in exon 5 that abrogates CYP2C19 function through the creation of an alternative splice site. Evidence of association was also detected for another functional SNP in the CYP2C19 promoter, rs12248560, highlighted in our previous study.

Conclusion(s)

Functional variants in CYP2C19 may contribute to endometriosis susceptibility in both familial and sporadic cases.

Introduction

Endometriosis is a gynaecological disease that affects 6-10% of women of reproductive age and causes a variety of symptoms including severe menstrual pain, chronic pelvic pain and subfertility (1). The disease is inherited as a complex trait (1-3), with up to 52% of the variation in liability accounted for by genetic factors (4). Recent genome-wide association studies (GWAS) have revealed a number of common genetic variants associated with susceptibility to endometriosis (5-7, 8). Both linkage and association studies take advantage of linkage disequilibrium (LD), where the alleles of SNPs located within short distances of each other are correlated. As has been the case in other complex diseases, the causal variants underlying endometriosis risk may be uncommon or novel SNPs in LD with the common markers genotyped in the original studies. Such SNPs can be detected by sequencing of regions surrounding association and linkage signals, and may contribute more to disease risk in the population than the common variants originally detected (9).

We recently fine-mapped a linkage peak for endometriosis on chromosome 10 (10), finding suggestive evidence of association to the cytochrome P450 family 2, subfamily C, 19 (CYP2C19) gene (11). The CYP2C19 protein participates in the metabolism of estrogen (12, 13) and up to 10% of clinically administered medications (14) including the anti-estrogenic drug tamoxifen (15). The gene has a number of functional variants that influence drug metabolism, one of which, rs12248560, we found to be associated with endometriosis risk (11). As this common SNP (minor allele frequency = 0.21) did not fully account for the linkage signal it is possible that additional SNPs contribute to the endometriosis susceptibility associated with this chromosomal region. Interestingly, we also found nominal association to a SNP independent of rs12248560, rs4244285 (LD estimate r² = 0.04), a loss-of-function variant in CYP2C19 exon 5 which had been previously associated with endometriosis risk in a small candidate gene study (16), although this result was not replicated in an even smaller sample (17).

The previously detected fine-mapping association signal extended across 252.3 kilobases (Kb) of chromosome 10 from the 3′ end of the helicase, lymphoid-specific (HELLS) gene to the 3′ end of CYP2C19 (11). This region includes another CYP2C gene family member, CYP2C18. In the current study, we aimed to search for additional variants in this region that may be contributing to endometriosis risk. We screened endometriosis cases for such risk variants using two strategies. First, to search for variants that may have a direct effect on gene activity we screened the 5′ and 3′ untranslated regions (UTRs), exons and intron-exon boundaries of CYP2C18 and CYP2C19 in 20 unrelated endometriosis cases with a strong family history of disease. Next, to broaden the search to include inter-genic and intronic regions that may harbour variants with regulatory effects on gene activity we sequenced the entire 252.3 Kb association region in DNA pools of 384 endometriosis cases and 384 controls. Case-control association analyses were then performed on a subset of the variants we detected by sequencing, and provided further evidence that functional SNPs in CYP2C19 may contribute to endometriosis risk.

Material and Methods

Samples

Case samples were taken from the set of 3,908 endometriosis cases with and without a family history of disease recruited by the QIMR Berghofer Medical Research Institute between 1995-2002 (5). All women had completed a questionnaire and provided a blood sample. A surgical diagnosis was confirmed by retrospective examination of medical records; disease severity was determined using the revised American Fertility Society classification system (18). Control samples comprised unrelated individuals originally recruited through QIMR Berghofer for either a twin study of gynaecological health who self-reported no endometriosis (discovery sample) (4) and the sample previously utilised for our endometriosis GWAS (replication sample) (5) recruited through the Brisbane Adolescent Twin Study (19, 20). Approval for this study, and to obtain medical records, for collection of blood for DNA extraction and all questionnaires and interview schedules, and for the inclusion of twin individuals recruited through the Australian Twin Registry, was obtained from the QIMR Berghofer Human Research Ethics Committee. All participants gave written informed consent.

Familial sequencing

DNA samples from 20 unrelated women from 20 of the most ‘case-dense’ endometriosis families included in our linkage study (10) were subjected to Sanger sequencing to search for potentially high-risk rare sequence variants contributing to endometriosis risk. Each family had 3 or more affected first- or second-degree relatives, with the sample chosen for sequencing having either the more severe disease (according to rAFS stages I-IV) (18), or the earliest age of onset if affected family members had been assigned the same stage.

Primers for the PCR amplification and sequencing of the promoter and 3′ UTR regions, intron-exon boundaries and all exons of the CYP2C18 and CYP2C19 gene were designed using Primer 3.0 (21) (Supplementary Table 1). PCRs were performed in 15 μl reactions containing 1X PCR buffer, 1.5 mM MgCl2, 1 U Amplitaq Gold (all Applied Biosystems, CA, USA), 200 μM each dNTP (Promega, Madison, WI, USA), 1 μM each of forward and reverse primers and 50 ng of DNA. PCR cycling conditions included initial denaturation at 95°C for 5 mins; ‘touchdown’ cycling at 95°C for 30 secs, 60°C (-0.5 per cycle) for 30 secs and 72°C for 30 secs for 20 cycles; 95°C for 30 secs, 50°C for 30 secs and 72°C for 30 secs for 15 cycles; final extension 72°C for 10 minutes. Products were verified by electrophoresis through 2% agarose gels, cleaned with Exonuclease I and Shrimp Alkaline Phosphotase (Fermentas, Burlington, Ontario,Canada) and sequenced using BigDye 3.0 terminator chemistry (Applied Biosystems).

Pooled sequencing of the HELLS-CYP2C19 region

DNA samples had previously been combined into two pools comprising 384 endometriosis cases (case pool) and 384 unrelated gynaecological health study samples (control pool) (22). Targeted re-sequencing and analysis of the case and control DNA pools were performed by deCODE Scientific Services (Reykjavik, Iceland). Briefly, for each pool, DNA within the ∼252.3 kb region extending from the 3′ end of HELLS to the 3′ end of CYP2C19 (covering chromosome 10 bases 96,361,710-96,704,010 (genome build GRCh37/hg19)) was amplified in sections by long-range PCR (average length 8,021 bases) and sequenced on an Illumina GAIIx platform (Illumina, San Diego, CA, USA). Short read sequences were then aligned and sequence variants (SVs; termed as such as many loci were yet to be verified) called by deCODE using the Illumina programs CASAVA and Elandv2. In addition, we reanalysed the sequences using programs tailored for DNA pools: sequences were aligned using the program Novoalign (http://www.novocraft.com/main/index.php) and SVs called using CRISP (23).

Individual genotyping and association analyses

Genotyping of SVs detected by individual and pooled sequencing was conducted in a ‘discovery’ sample of of 958 endometriosis cases and 959 unrelated gynaecological health study controls examined in previous genetic association studies for endometriosis conducted at QIMR Berghofer (24-29). This sample includes all 768 individuals (384 cases and 384 controls) included in the DNA pools described above. SVs were genotyped in multiplex assays using the Sequenom MassARRAY Genomics Platform (Sequenom, San Diego, CA, USA). We selected 197 SVs for individual genotyping including two SNPs detected only by familial-case sequencing, 60 SVs called by deCODE only, 60 SVs called by QIMR only and 75 SVs called by both analyses. SVs detected by targeted sequencing were included according to the following criteria: 1) Nominally significant (P < 0.05) allele frequency differences between the case and control pools (83 SVs across the 3 groups), and 2) Location in potentially functional areas (e.g. all exonic SVs in addition to SVs located in 5′ and 3′ UTRs, regions of conserved sequence or within putative transcription factor binding sites) as determined by the program ANNOVAR (112 SVs) (30).

Genotyping quality control was performed for each multiplex separately, where SVs with >5% missing genotypes and Hardy-Weinberg P-values <1×10^-4 were excluded from further analyses. In addition, individuals missing >5% of data, or for whom we have recent genetic evidence of non-Caucasian ancestry detected during our GWAS analysis (5), were also removed. Single SNP and haplotype association analyses were performed including 901 of the 958 endometriosis cases and 939 of the 959 gynaecological health study controls passing all quality control metrics using PLINK (31). Study-wide significance at the discovery phase (P_adjusted <0.05) was determined by 10,000 permutations of the dataset, where case-control status was randomly swapped to break the genotype-phenotype relationship while preserving the LD structure between the SNPs, to generate empirical significance levels corrected for multiple testing resulting from the large number of SNPs included in the study.

Replication

As only Australian samples were genotyped in the ‘discovery’ phase, we performed in silico replication on Australian cases and controls only, using a sample comprising 1,370 unrelated endometriosis cases and 1,770 unrelated Brisbane Adolescent Twin study controls drawn from our recent GWAS for endometriosis (5). Briefly, all Australian GWAS individuals not included as, or related to, an individual from the 958 case sample were utilised as cases or controls for the replication sample. As the best SNPs from the discovery phase analysis were not present on the Illumina 610K genotyping chips used to genotype the GWAS samples, we analysed genotype data and dosage scores for SNPs previously imputed to HapMap2 (32). All imputed SNPs had genotype concordance values of between 97.1-99.7% for 560 endometriosis cases for whom both actual and imputed genotypes were available, with imputation quality scores >0.9 indicating these SNPs were imputed with a high degree of accuracy.

Association analyses were performed in the replication dataset for three SNPs, rs1326837, rs4244285 and rs12248560, as described above. Since only associations in the same direction as in the discovery sample can be considered to be replicated, one-sided P-values were obtained by halving the standard (two-sided) P-values (8). Unadjusted results obtained for both the discovery and replication sets were then included in a meta-analysis performed using METAL (8, 33), which converts P-values to Z-scores that are then weighted by the square-root of the sample size for each sample set, to determine the total evidence of association for each SNP.

Results

Sequencing familial endometriosis cases

Sequencing of 20 cases from 20 unrelated case-dense endometriosis families for both the CYP2C18 and CYP2C19 genes revealed the presence of a number of known SNPs in addition to two novel intronic variants, one in intron 7 of CYP2C18 (chr10:96492949) and the second in intron 7 of CYP2C19 (chr10:96602941). As both SNPs were located at least 50 bases from the respective intron-exon boundaries neither were predicted to have an effect on gene-splicing (ESEfinder3.0: http://rulai.cshl.edu/cgi-bin/tools/ESE3/esefinder.cgi?process=home). A number of affected women carried the known functional CYP2C19 SNPS rs12248560 (CYP2C19*17, see http://www.cypalleles.ki.se/cyp2c19.htm for allele nomenclature) in the promoter region (associated with endometriosis risk in our previous fine-mapping study (11)), rs17878459 (p.Glu92Asp: CYP2C19*3) in exon 2, and rs4244285 (p.Pro271Pro: CYP2C19*2) in exon 5.

Sequencing DNA pools of endometriosis cases and controls

Sequencing of the 252.3 Kb between the end of the HELLS gene to the end of CYP2C19 in our two DNA pools of 384 endometriosis cases and 384 controls resulted in adequate sequence data (>500-fold coverage) for 96.8% and 94.0% of the region in each pool, respectively. Average sequence coverage was 1,261-fold for the case pool and 1,372-fold for the control pool. The largest gap of missing sequence (∼20 Kb) was in the region of 96.38-96.40 Mb for both pools (Supplementary Fig. 1).

deCODE Genetics called 3,051 SVs using their analysis pipeline, of which 2,418 (79%) were novel (not present in SNP databases such as dbSNP, HapMap or the July 2010 (Phase2) release of the 1000 Genomes data). Most of the SVs were rare: 2,319 (76%) had pool minor allele frequencies (MAFs) of <1% (range <0.0001-0.499; Table 1). At QIMR we called 1,447 SVs using the CRISP program, of which 958 (66%) were novel, and 706 (49%) had pool MAFs <1% (range <0.0001-0.497; Table 1). A total of 629 SVs were called by both deCODE and QIMR using the two variant-calling approaches. Of these, 148 (23.5%) were novel and 158 (25%) had pool MAFs <1% (range 0.0004-0.50).

Table 1.

Average minor allele frequencies in the endometriosis case and control DNA pools by SNP calling method.

MAF range	deCODE	QIMR
< 0.001	26 (0.9%)	58 (0.04%)
0.001-0.01	2281 (74.7%)	651 (45%)
0.01-0.05	441 (14.5%)	323 (22.3%)
0.05-0.1	109 (3.5%)	152 (10.5%)
0.1-0.5	194 (6.4%)	263 (18.2%)

Sequence variant polymorphism in individual genotyping

We selected 197 SVs for individual genotyping (of which 63 were known SNPs), including 2 SVs detected only by familial-case sequencing, 60 SVs called by deCODE only, 60 SVs called by QIMR only, and 75 SVs called by both analyses. Genotyping of the 197 SVs in the discovery sample revealed that 119 were in fact monomorphic (e.g. no alternative allele was detected) in the 384 case and 384 control samples used to make up the DNA pools, although two of these SVs were polymorphic in the larger sample of 901 cases and 939 controls passing our quality control. Neither genomic location (inter-genic or intronic vs. exonic, potentially functional or conserved areas) or MAF in the pools were reliable predictors of whether SVs would be polymorphic or monomorphic when individually genotyped: SVs monomorphic in the individual genotyping had pool MAFs ranging from 0.001-0.40 in either the case or control pools, while SVs polymorphic in individual genotyping had pool MAFs ranging from 0.001-0.17.

SVs were instead more likely to be polymorphic if they were known SNPs (i.e. they had previously been assigned an ‘rs’ number) and had been called by both variant-calling programs utilised in this study. Only 17/134 (12.7%) novel SVs were polymorphic in individual genotyping. Of the 63 known SNPs, 61 (97%) were polymorphic when individually genotyped, although these had pool MAFs within the range of the SVs shown to be monomorphic upon individual genotyping (0.001-0.5).

Fine-mapping across the CYP2C19 region

Following quality control, 80 SNPs were polymorphic in the discovery sample of 901 endometriosis cases and 939 controls included in the association analysis. A nominal association signal (P ≤ 0.05) was detected for 35 SNPs, 19 of which remained significant (P_adjusted ≤ 0.05) following permutation to correct for multiple testing (Table 2; Supplementary Table 2). Association signal was distributed across the sequenced region (Fig. 1), with the best signal seen for rs1326837 (P = 4.6 × 10^-5, OR = 1.38; P_adjusted = 1.2 × 10^-3). All 19 significant SNPs were in moderate to high LD with each other, with analysis conditioning on rs1326837 removing all signal from SNPs in high LD (r² > 0.8) and reducing but not eliminating the significance of SNPs in moderate LD (r² 0.5-0.8), indicating that SNPs in both high and moderate LD with rs1326837 are contributing to the endometriosis risk conferred by this haploptype.

Table 2.

Association signal to endometriosis across the region between the HELLS and CYP2C19 genes on chromosome 10q26. Signal across the region is accounted for by SNPs in 2 blocks of linkage disequilbrium (LD), with the best SNP in each block indicated by *. rs4244285 is included as a known functional SNP in moderate LD (r² 0.64) with rs1326837.

			Fine-mapping				Replication		Meta-analysis
LD block	SNP	Location	MAF# Ca/Co	P-value	Padjusted	OR (95% CIs)	P-value	OR (95% CIs)	P-value	OR (95% CIs)
1	rs1326837*	HELLS-CYP2C18	0.25/0.19	4.6×10-5	1.2×10-3	1.38 (1.18-1.61)	3.1×10-3	1.19 (1.07-1.30)	1.3×10-6	1.26 (1.07-1.42)
1	rs4244285	CYP2C19 exon 5	0.17/0.13	9.7×10-4	2.5×10-2	1.35 (1.12-1.61)	8.0×10-3	1.19 (1.05-1.33)	9.4×10-5	1.25 (1.06-1.41)
2	rs12248560*	CYP2C19 promoter	0.80/0.76	6.4×10-3	1.6×10-1	0.80 (0.69-0.94)	2.5×10-2	0.88 (0.78-0.97)	5.8×10-4	0.85 (0.75-0.95)

^#MAF = minor allele frequency

Figure 1.

Association signal across the 252.3 Kb region between the HELLS and CYP2C19 genes on chromosome 10. The genomic positions of the top SNP rs1326837 and the two functional SNPs rs4244285 and rs12248560 are indicated by dashes next to the SNP names.

The corresponding P-value for the 19-SNP haplotype was 8.19 × 10^-5, and there was no evidence of an independent effect for rs1326837. These 19 SNPs include rs4244285, representing the subgroup of SNPs in moderate LD with rs1326837, which achieved an unadjusted P = 9.2 × 10^-4 (P_adjusted = 2.5 × 10^-2), and retained an unadjusted P = 2.7 × 10^-2 in the conditional analyses.

The association signal for both rs1326837 and rs4244285 was replicated in the independent dataset of 1,370 mostly non-familial endometriosis cases and 1,770 population controls, with a one-sided P-value for rs1326837 of 3.1 × 10^-3 (OR = 1.19), and rs4244285 P = 8.0 × 10^-3 (OR=1.18). Meta-analysis of the discovery and replication datasets produced a combined P-value for rs1326837 of 1.3 × 10^-6, and for rs4244285 of 9.4 × 10^-5, surpassing our threshold of total study-wide significance calculated as 6.2 × 10^-4 (where P = 0.05/80 SNPs).

Evidence of association was also detected for rs12248560 (P = 6.4 × 10^-3, OR = 0.80), the functional promoter SNP highlighted in our previous fine-mapping study (Table 2; Supplementary Table 2). LD between rs12248560 and SNPs of the rs1326837 haplotype is low (average r² = 0.04), hence the association signal was not affected by the conditional analysis described above, indicating that this SNP represents an independent effect on endometriosis risk. rs12248560 was, however, no longer significant following permutation of the 901 case and 939 control dataset, indicating the signal for this SNP was not study-wide significant in the discovery set (P_adjusted = 0.16). However, there was evidence of association in the replication sample (1-sided P = 2.5 × 10^-2, OR = 0.88), and meta-analysis of the unadjusted discovery and replication data-sets produced a combined P-value for rs12248560 of 5.8 × 10^-4, indicating this SNP to be significant over the total study sample.

Discussion

We have found genetic variants in the CYP2C19 gene to be associated with susceptibility to endometriosis amongst women with and without a family history of the disease. Previously, we found the rs12248560 functional promoter SNP to be associated with a decreased risk of endometriosis following our initial fine-mapping of a linkage peak on chromosome 10 (11). While study-wide significant, the rs12248560 signal is somewhat reduced in the current study (fine-mapping sample minor allele P=4.9×10^-4, OR = 0.78 (11) versus current discovery sample P=6.4×10^-3, OR = 0.80), this appears mostly due to the smaller sample size as only Australian cases and controls were included here. Although the effect of rs12248560 on gene transcription requires clarification (CYP2C19*17 was initially suggested to cause an ultra-rapid drug metaboliser phenotype but this has recently been questioned (34)), this SNP has also been associated with a decreased risk of breast cancer, particularly in women treated with hormone replacement therapy for 10 years or more (35), indicating that functional CYP2C19 variants can influence the risk of estrogen-dependent conditions.

The finding of additional independent association signals provides evidence for the possibility that multiple variants contributed to the original linkage peak in this region (10). Of particular interest is the top SNP from this study rs1326837. Located in the inter-genic region between HELLS and CYP2C18, there is currently no evidence for a regulatory role for this particular SNP. However, previous GWAS to find variants influencing pharmacological drug response have detected SNPs in this inter-genic region (e.g. rs12772169 and rs12777823) that are in complete and moderate LD with rs1326837, and are also linked to functional SNPs in various CYP2C genes (36, 37). Such SNPs may be contributing to the risk of endometriosis detected here.

SNPs in moderate LD with rs1326837 include the functional SNP rs4244285 (r² = 0.64). This synonymous SNP in CYP2C19 exon 5 causes no change to the amino acid (p.Pro227Pro), but creates an alternative splice site 40 bases downstream resulting in a truncated, non-functional protein (38) and a poor metaboliser phenotype (CYP2C19*2). In our previous fine-mapping study including cases from Australia and the UK we found only nominal association to rs4244285 (OR = 1.23, P = 1.1×10^-2) in 1,158 familial and sporadic cases (11), indicating that association with this SNP was not driving the association signal detected in that sample. Given that the signal for rs4244285 is more significant in the current study (discovery sample OR = 1.35, P = 9.7×10^-4; meta-analysis OR = 1.35, P = 9.4×10^-5) despite the smaller sample size may indicate that the effect of this SNP on endometriosis risk is larger in the Australian than in the UK population.

CYP2C19 participates in the conversion of 17β-estradiol (E2) to estrone (E1) (12), and in the production of 2α-hydroxy estrogen (2α-OHE) (12, 13). This gene may exert its disease–risk altering effects through two mechanisms, either independently or in concert. That SNPs causing the increased metaboliser phenotypes (rs12248560) are protective while SNPs causing the poor metaboliser phenotype (rs4244285) increase the risk of disease supports the hypothesis that CYP2C19 SNPs may alter endometriosis risk through an effect on local and/or central estrogen metabolism. Altered CYP2C19 activity may affect localised, tissue-specific rates of conversion of E2 to E1, and/or result in altered levels of the anti-estrogenic 2α-OHE metabolite, with activity either increased or decreased as compared to the wild-type isoform depending on the true causal variant(s). Evidence for the role of other genes involved in sex-steroid metabolism in endometriosis is currently equivocal at best, due mostly to small sample sizes typically analysed in candidate gene studies (39, 40).

Alternatively, CYP2C19 may affect endometriosis risk through altered metabolism of exogenous substances, which may increase the risk of developing this disease (41, 42). CYP2C19 expression and activity is altered in response to xenobiotic as well as steroid hormone exposure (43, 44). As seen for drug metabolism, it may be that CYP2C19 SNPs influence the rates at which other substances metabolised by this gene are cleared, either by the liver or locally at extra-hepatic sites, and this may moderate endometriosis risk. As the use of plastics and associated chemicals has increased over the past decades (45), we examined effect sizes for rs12248560 and rs4244285 by period of endometriosis diagnosis among our cases (pre- and from 1994, the median year of diagnosis in our sample), and while larger effects were seen for both SNPs for cases diagnosed from 1994 these results were not significantly different (Supplementary Table 3).

In this study we sought to detect rare and possibly novel, sequence variants associated with an increased risk of developing endometriosis through DNA sequencing of women with a family history, and subsequent association analyses including case samples with and without a family history of disease. Instead, the best association signals detected were for common variants linked to known functional SNPs in the CYP2C19 gene. The effect sizes of these CYP2C19 variants are small, but similar to variants associated with endometriosis in recent GWAS conducted by us (5) and others (6, 7). Although our current results are not significant at a genome-wide level, they are significant at a study-wide level and replicated in a large independent sample. Further studies to validate our findings in additional case and control datasets or functional experiments such as determining the association of SNPs with gene expression in relevant tissues will help to clarify the role of CYP2C19 in estrogen-dependent diseases. CYP2C19 is an important candidate due to its role in the metabolism of estrogen, and possibly also other substances that might be associated with the risk of developing endometriosis, and our study indicates that multiple SNPs in this gene region are likely to be contributing to this risk.