The Inflammatory Gene Pathway Is Not a Major Contributor to Polycystic Ovary Snydrome

Abstract

Context:

Although inflammation is clearly associated with obesity, diabetes, and insulin resistance, the role of chronic inflammation in the etiology of polycystic ovary syndrome (PCOS) is unclear.

Objective:

To determine whether chronic inflammation plays a causal role in the etiology of PCOS, we tested for an association between PCOS and genetic markers mapping to 80 members of the inflammatory pathway.

Design:

This was a case-control association study.

Setting:

The setting was an academic medical center.

Patients or Participants:

A total of 905 index case patients with PCOS and 955 control women (108 intensively phenotyped subjects with normal androgen levels and regular menses and 847 minimally phenotyped subjects with regular menses and no history of PCOS).

Interventions:

Subjects were genotyped at single nucleotide polymorphisms mapping to 80 inflammatory genes. Logistic regression was used to test for an association between 822 single nucleotide polymorphisms and PCOS after adjustment for population stratification, body mass index, and/or age. In the index patients, we also tested for association with 11 quantitative traits (body mass index and testosterone, fasting insulin, fasting glucose, 2-hour postchallenge glucose, LH, FSH, total cholesterol, high-density lipoprotein, low-density lipoprotein, and triglyceride levels).

Main Outcome Measures:

The evidence for an association with PCOS and with 11 quantitative traits was investigated.

Results:

Nominally significant evidence for an association was observed with MAP3K7, IKBKG, TNFRS11A, AKT2, IL6R, and IRF1, but no results remained statistically significant after adjustment for multiple testing.

Conclusions:

Genetic variation in the inflammatory pathway is not a major contributor to the etiology of PCOS or related quantitative traits in women with PCOS.

Polycystic ovary syndrome (PCOS) is a common endocrine disorder that occurs in ∼7% of reproductive age women (1). It is characterized by hyperandrogenemia, oligomenorrhea, and reduced fertility (2). Women with PCOS are also at an increased risk for obesity, insulin resistance, and pancreatic β-cell dysfunction (for a review, see Ref. 3).

There is a well-established link between obesity, diabetes, and insulin resistance and chronic low-grade inflammation (4, 5). Given the frequent association of these abnormalities with PCOS, it has been suggested that inflammation is a critical component of PCOS (for a review, see Ref. 6). However, the role of chronic inflammation in PCOS is not clear (for a review, see Ref. 6). C-reactive protein (CRP) and TNF-α levels are elevated in women with PCOS compared with those in reproductively normal women, although many of these studies have been limited by a failure to control for the independent effect of obesity on these parameters. In contrast, studies of other markers of inflammation including IL-6 receptor, IL-18, soluble TNF receptor (TNFRSF1B), and soluble intracellular adhesion molecule-1 have found no association with PCOS or results have been inconclusive (7–11). In a meta-analysis of 31 studies, Escobar-Morreale et al (12) evaluated the role of CRP, IL-6, and TNFα in PCOS. After accounting for body mass index (BMI), only CRP levels remain significantly associated with PCOS.

Accordingly, it is not clear whether inflammatory cytokines are elevated in women with PCOS due to an intrinsic perturbation of the inflammatory pathway or the systemic proinflammatory state resulting from its associated obesity and/or insulin resistance. An approach to differentiate between parameters that are intercorrelated as opposed to causality related has been to investigate whether genetic variations in the pathways implicated are associated with the outcome of interest. For example, in Mendelian randomization, associations between gene variants regulating parameters of interest, eg, SHBG (13), and outcomes, eg, type 2 diabetes, respectively, have provided insight into causality. In a conceptually similar approach, we tested for an association between 822 single nucleotide polymorphisms (SNPs) in 80 inflammatory pathway genes and PCOS in 905 women with PCOS and 955 control women. This study provides the most comprehensive genetic evaluation of the inflammatory gene pathway in PCOS to date.

Subjects and Methods

Subjects

This study was approved by the institutional review boards of participating centers, and written informed consent was obtained from all participants (14). We studied 905 index cases (probands) with PCOS and 955 control women (108 intensively phenotyped subjects recruited in parallel to the PCOS probands and 847 minimally phenotyped subjects from a DNA repository) as described previously (14). PCOS was defined according to the classic Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) criteria and fulfilled the NICHD, Rotterdam, and Androgen Excess Society criteria for the diagnosis of PCOS (14). Intensively phenotyped control women (n = 108) had normal androgen levels, regular menses, and normal glucose tolerance in an oral glucose tolerance test (OGTT) and were of similar age and ethnicity as the women with PCOS (14). To increase the number of control subjects, we included minimally phenotyped reproductive age women (n = 847) selected from a large-scale gene bank that combines a centralized genomic DNA sample (http://www.nugene.org), as described previously (14). All participants were US Caucasians of European ancestry.

Anthropometric measurements

Blood pressure, waist circumference, weight, and height were measured as reported previously (15).

OGTT

In 426 women with PCOS and 103 intensively phenotyped control women, a 75-g OGTT was performed after a 300-g carbohydrate diet and overnight fast as described previously (16). Morning fasting blood samples were obtained for baseline insulin and glucose levels and the 2-hour postglucose challenge.

Hormone and Lipid Assays

A morning fasting blood sample was obtained for fasting reproductive and metabolic hormones, as listed below. Circulating levels of glucose, insulin, total testosterone (T), non-SHBG–bound testosterone, LH, FSH, total cholesterol, high-density lipoprotein (HDL), low-density lipoprotein (LDL), and triglycerides (TGs) were determined as reported previously (15, 17).

Genotyping

We submitted 911 SNPs mapping to 80 genes belonging to the inflammatory pathway for genotyping using the iSelect custom genotyping system (Illumina). SNPs were selected as described previously and included haplotype tagging SNPs (SNPs) and known coding SNPs (SNPs). Genotypes were called using Illumina GenomeStudio Genotyping Module software (http://www.illumina.com/software/genomestudio_software.ilmn) as described previously (14).

Data analysis

Principal component analysis (PCA)

As described previously (14), we used genotype data from 253 ancestry informative markers and EIGENSTRAT/SmartPCA in the EIGENSOFT software package (http://genepath.med.harvard.edu/∼reich/Software.htm), implemented in the BC/SNPmax platform (http://www.bcplatforms.com/solutions/solutions_1_1.html) to calculate principal components (PCs) (18). SmartPCA identified 5 genetic outliers who were removed from subsequent analyses. For association studies, the first 2 PCs derived from the PCA were used as covariates to adjust for population stratification.

Association testing

We tested for association between SNPs and the dichotomous trait, PCOS, with 4 logistic regression models. Model 1 adjusted for population stratification with the first 2 PCs from the PCA. Model 2 adjusted for BMI and population stratification. Model 3 adjusted for age, BMI, and population stratification. Model 4 adjusted for age and population stratification. These analyses were implemented using PLINK version 1.06 within the BC/SNPmax platform (http://www.bcplatforms.com/solutions/solutions_1_1.html) (19).

We further assessed the impact of genetic variation at the SNPs on the distribution of 11 quantitative traits in the subjects with PCOS. The traits tested were BMI (n = 905), total T (n = 885), fasting insulin (n = 811), fasting glucose (n = 861), 2-hour glucose (n = 426), LH (n = 817), FSH (n = 816), total cholesterol (n = 741), HDL (n = 778), LDL (n = 756), and TGs (778). We tested for association with quantitative traits using the 4 models described above. For BMI, the second and third models were not evaluated, because those models use BMI as a covariate. Analyses were performed using PLINK version 1.06 (19).

Correction for multiple testing

Given that this study was a targeted analysis of a subset of the genome, we adjusted the level of α for the number of independent loci tested (the number of haplotype blocks tagged by the genotyped SNPs plus singleton SNPs that did not map to a haplotype block). The 822 successfully genotyped SNPs mapped to 236 independent loci, and we, therefore, considered a value of P <2.1 × 10⁻⁴ as significant.

Power analysis

We used the CaTS package to calculate the power to detect an association with PCOS in our cohort (20). The parameters were 905 control women, 955 case patients, and allele frequency of 0.25 and assumed an additive model.

Results

Study participant characteristics

Phenotypic characteristics of the study participants are shown in Supplemental Table 1 published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org. PCOS case patients were significantly younger than the minimally phenotyped control women and had significantly higher BMI than the intensively and minimally phenotyped control women. Total T, non-SHBG–bound testosterone, fasting insulin, 2-hour glucose, LH, FSH, total cholesterol, LDL, and triglyceride levels were significantly elevated in the women with PCOS compared with the intensively phenotyped control women, whereas HDL was lower in women with PCOS than in the intensively phenotyped control women.

Genotyping

Of the 911 SNPs, 89 were excluded for poor genotyping quality (P₁₀ quality score = <0.40 or P₅₀ quality score = <0.60), whereas 822 SNPs (799 haplotype tagging SNPs and 42 coding SNPs) were successfully genotyped (Supplemental Table 2). The average genotype call rate was >99.9%, minor allele frequency (MAF) was 0.253, and genotype concordance in DNA replicates was >99.9%.

Association testing

The strongest evidence for association was observed with SNPs mapping to mitogen-activated protein kinase kinase kinase 7 (MAPK37), inhibitor of kappa light polypeptide gene enhancer in B-cells, kinase gamma (IKBKG), TNF receptor superfamily, member 11a, NFKB [nuclear factor κB] activator (TNFRSF11A), v-akt murine thymoma viral oncogene homolog 2 (AKT2), and interleukin 6 receptor (IL6R) (Table 1). After adjustment for 236 independent genomic regions, none of the loci reached statistical significance, although the association of rs7754169 within MAP3K7 almost reaches statistical significance (P_nominal = 2.2 × 10⁻⁴ vs P_corrected = .052).

Table 1.

SNPs With Strongest Evidence for Association With PCOS

Gene	SNP	MAF	Allele	Model 1			Model 2			Model 3			Model 4
				OR	95% CI	P Value	OR	95% CI	P Value	OR	95% CI	P Value	OR	95% CI	P Value
MAP3K7	rs7754169	0.366	A	0.83	0.73–0.95	.0085	0.84	0.71–0.99	.033	0.72	0.56–0.88	.0012	0.73	0.62–0.86	2.2 × 10−4
MAP3K7	rs2325302	0.252	G	0.81	0.69–0.95	.0087	0.80	0.66–0.97	.023	0.67	0.52–0.85	.0011	0.70	0.58–0.86	5.1 × 10−4
MAP3K7	rs1999090	0.272	C	0.83	0.72–0.96	.014	0.68	0.68–0.97	.020	0.70	0.57–0.88	.0016	0.75	0.63–0.90	.0015
IKBKG	rs2472394	0.101	A	0.69	0.55–0.87	.0017	0.66	0.50–0.87	.0027	0.64	0.46–0.90	.011	0.68	0.51–0.90	.0067
TNFRSF11A	rs7236060	0.354	G	1.17	1.01–1.35	.036	1.20	1.01–1.42	.037	1.40	1.13–1.73	.0019	1.29	1.08–1.53	.0045
TNFRSF11A	rs2981004	0.265	G	0.78	0.67–0.92	.0020	0.80	0.67–0.96	.017	0.78	0.62–0.98	.035	0.79	0.76–0.95	.011
AKT2	rs11671439	0.089	C	1.47	1.15–1.89	.0023	1.50	1.12–2.00	.0069	1.43	0.99–2.07	.056	1.52	1.12–2.06	.0067
IL6R	rs1386821	0.167	C	1.19	1.01–1.40	.036	1.13	0.93–1.37	.22	1.24	0.98–1.58	.080	1.36	1.12–1.66	.0025
IRF1	rs9282763	0.372	G	0.84	0.73–0.96	.0131	0.83	0.72–0.96	.011	0.77	0.65–0.92	.0028	0.79	0.64–0.97	.026

Abbreviations: CI, confidence interval; MAF, minor allele frequency in HapMap CEU; OR, odds ratio. Model 1, PCA-adjusted; model 2, BMI- and PCI-adjusted; model 3, age-, BMI-, and PCA-adjusted; model 4, age- and PCA-adjusted.

Quantitative trait analysis

The strongest evidence for association with a quantitative trait was between HDL levels and the gene for endothelin (EDN1) in women with PCOS (rs9369217, P value range = 3.8 × 10⁻⁵ to 7.4 × 10⁻⁵, depending on the model tested), whereas a second variant within EDN1, rs9296344, was associated with LH levels (P value range = 6.2–8.1 × 10⁻⁵). rs8072050 within the gene for ribosomal protein S6 kinase, 70 kDa, polypeptide 1 (RPS6KB1), was associated with fasting insulin levels (P value range = 4.5–12 × 10⁻⁵) (Table 2).

Table 2.

Quantitative Trait Analysis Results

Trait	Gene	SNP	Allele	MAF	Model 1		Model 2		Model 3		Model 4
					β	P Value	β	P Value	β	P Value	β	P Value
HDL	EDN1	rs9369217	T	0.168	3.21	3.8 × 10−5	2.96	7.1 × 10−5	2.95	7.4 × 10−5	3.18	4.4 × 10−5
Fasting insulin	RPS6KB1	rs8072050	A	0.084	5.92	5.5 × 10−5	5.04	9.5 × 10−5	4.96	1.2 × 10−4	5.97	4.5 × 10−5
Total TGs	NFKB1A	rs8013309	C	0.150	34.5	4.8 × 10−5	33.5	6.6 × 10−5	32.8	9.2 × 10−5	33.4	8.1 × 10−5
LH	EDN1	rs9296344	C	0.080	4.91	6.2 × 10−5	4.53	6.9 × 10−5	4.52	7.4 × 10−5	4.82	8.1 × 10−5

Abbreviations: CI, confidence interval; MAF, minor allele frequency in HapMap CEU; OR, odds ratio. Model 1, PCA-adjusted; model 2, BMI- and PCI-adjusted; model 3, age-, BMI-, and PCA-adjusted; model 4, age- and PCA-adjusted.

Power analysis

For a genotype relative risk (GRR) of 1.4, we had >80% power to detect an effect at P < 2 × 10⁻⁴ and >90% power to detect an effect at P < 2 × 10⁻³, whereas for a GRR of 1.3, we had >40% power to detect an effect at P < 2 × 10⁻⁴ and >70% power to detect an effect at P < 2 × 10⁻³. We, therefore, had sufficient power to detect a variant of moderate effect size in our cohort.

Discussion

Given the potentially confounding impact that obesity and insulin resistance may have on the inflammatory response in PCOS, we used a genetic approach to determine whether the increased inflammatory state observed in PCOS is an intrinsic feature of the syndrome or simply a reflection of the systemic proinflammatory state due to obesity and insulin resistance. We hypothesized that if the inflammatory gene pathway is critical to the development of PCOS, there should be significant evidence for association with PCOS in genes belonging to the inflammatory pathway. We, therefore, evaluated genetic variation in 80 inflammatory response genes using a case-control study design. We found nominal evidence for association with SNPs located in or near MAP3K7, IKBKG, TNFRSF11A, AKT2, and IL6; however, no genes showed significant evidence for association with PCOS after adjustment for multiple testing. Among the 11 PCOS endophenotypes tested (BMI, total T, fasting insulin, 2-hour glucose, LH, FSH, total cholesterol, HDL, LDL, and TGs), the strongest evidence for association was observed between SNPs mapping to endothelin and HDL and LH levels, SNPs mapping to RPS6KB1 and fasting insulin levels, and SNPs mapping to NFKB1A and TG levels. However, these results do not remain significant after correction for multiple testing.

This is the largest and most systematic genetic evaluation of inflammatory genes in PCOS published to date (21). The study had power to detect moderate-size genetic effects (GRR >1.3). PCOS susceptibility loci identified by genome-wide association studies (GWAS) have genetic effect sizes ranging from 1.24 to 1.51 (22, 23), whereas those in large-scale meta-analyses (n > 100 000) of type 2 diabetes can be as small as 1.05 (24). Because rare coding variants as well as common tagging SNPs were included in the analysis, our study provided more comprehensive coverage of the genome than GWAS. GWAS are designed to assay common genetic variation, and it is not clear how much of the rare genetic variation is captured, even after genotype imputation (25).

The limitations of the current study are that technological restrictions did not allow us to analyze very rare variants (minor allele frequency <1%) or unique variants (variations found in a single individual), our sample size precludes the detection of variants with small effect sizes, and we did not assess the impact of epigenetic variation on expression of inflammatory genes. Given the extensive nature of the inflammatory gene pathway, we focused our analysis on inflammatory markers that are elevated in PCOS and related phenotypes, their receptors, and second messenger molecules including members of the NFKB signaling pathway. Salicylates have been shown to ameliorate the inflammatory state observed in diabetes and insulin resistance (26, 27), and this effect is believed to be mediated through the NFKB signaling pathway (28–31). Although this strategy encompasses the aspects of the inflammatory gene pathway most likely to be relevant to PCOS based on existing molecular evidence, it does exclude other parts of the inflammatory pathway, including the Toll receptor signaling cascade that could be relevant to PCOS, albeit in a less clearly obvious role.

Although several SNPs within inflammatory genes were found to have nominal evidence for association with PCOS, none remained significant after adjustment for multiple testing. Although these findings may reach statistical significance in larger data sets because of increased power to detect association, the effect sizes for these variants are expected to remain small (<GRR of 1.3) and the impact on the etiology of PCOS will be modest. From the results of these analyses, we can conclude that the bulk of the inflammatory state observed in PCOS results from the associated obesity and/or insulin resistance and is not an independent feature of the syndrome.