Genome‐Wide European Ancestry Study Identifies Coronary Artery Disease–Associated Loci Through Gene–Sex Hormone Interaction

Abstract

Background

Although sex differences in coronary artery disease (CAD) risk have been observed, little is known about the role of sex hormones in CAD genetics. Accounting for sex hormone levels may help identify CAD‐risk loci and extend our knowledge of its genetic architecture.

Methods and Results

A total of 365 662 individuals of European ancestry enrolled in the UK Biobank were considered. Genetic interaction of total testosterone, bioavailable testosterone, and SHBG (sex hormone‐binding globulin) were evaluated. Gene–environment interactions in millions of samples software was used to conduct sex‐stratified genome‐wide interaction analysis with prevalent CAD as the outcome. Participant age at enrollment and principal components 1 to 10 were adjusted as covariates. We identified 45 loci in men and 8 loci in women that reached genome‐wide significance (P < 5 × 10⁻⁸) for CAD. Ten of the loci identified (5 loci in both men and women) were through joint effects and would not have been picked up using a traditional genome‐wide association study. Two of the joint effect loci in women were independently identified with significant single nucleotide polymorphism‐total testosterone interactions.

Conclusions

This genome‐wide gene–sex hormone interaction study identified genomic‐risk loci that may contribute to the differential CAD risk between men and women, which otherwise would not have been discovered in a traditional genome‐wide association study solely including marginal genetic effects.

Nonstandard Abbreviations and Acronyms

BAT

bioavailable testosterone

GEM

gene–environment interactions in millions of samples

PRS

polygenic risk score

SHBG

sex hormone‐binding globulin

TT

total testosterone

Clinical Perspective.

What Is New?

• This sex‐stratified genetic association study encompassing 365 662 individuals of European ancestry enrolled identified novel loci of coronary artery disease by incorporating genetic interactions of sex hormones.

What Are the Clinical Implications?

• Incorporating genetic interactions of sex hormones could help explain interindividual differences, including sex differences in coronary artery disease risk, beyond the contribution from main genetic effects.

• Incorporating the nongenetic risk factors in genome‐wide association studies can improve our understanding of the genetic architecture of coronary artery disease.

Cardiovascular disease, and specifically coronary artery disease (CAD), remains the leading cause of death worldwide, with global estimates suggesting a near doubling of prevalent cases from 271 million in 1990 to 523 million cases in 2020. ¹ Despite advancements in treatment modalities, including newer antiplatelet and novel lipid‐lowering agents, the decline in CAD incidence rate has plateaued, both in the United States and worldwide. ¹ In 2020, nearly 207 of every 100 000 people died of either CAD or stroke, claiming more lives than cancer and lower respiratory tract diseases combined. ² These concerning trends have been associated with significant sex‐based disparities. Although CAD‐ and stroke‐related mortality rates are higher in men compared with women, the age‐adjusted decline in mortality rates over time have been stronger in men. ³ The contribution of sex hormones to the differential cardiovascular risk between men and women, specifically in relation to the genetic architecture of CAD, remains incompletely established.

Contemporary literature suggests that sex hormones have differential effects on the vasculature and atherosclerotic process in men and women. In women, estrogen exerts its effects through ER‐α (Estrogen Receptor‐alpha) and β (Estrogen Receptor‐beta) receptors that are found on the vascular endothelium and vascular smooth muscle cells. ER‐β induces vasorelaxation by enhancing the levels of endothelial NO and prostaglandin I2, decreases vascular smooth muscle proliferation, and decreases vasoconstriction in response to stressors by promoting endothelial K⁺ channel expression and function, leading to diminished extracellular Ca²⁺ influx. ⁴ It additionally supports a favorable lipid balance by increasing the levels of high‐density lipoprotein cholesterol and lowering the levels of low‐density lipoprotein cholesterol. ⁵ The impact of androgens, namely testosterone, is more complex. In men, testosterone exerts a vasodilatory effect independent of endothelial cells (NO pathway) by stimulating vascular smooth muscle relaxation, possibly via inhibition of L‐ and T‐type calcium channels. ⁶ Low levels have been associated with a higher risk of metabolic syndrome and adverse cardiovascular events in men. ⁷ In postmenopausal women, higher testosterone levels have been linked with an increased risk of metabolic syndrome, insulin resistance, as well as higher systolic blood pressure. ⁸ Given the complex interplay of sex hormones with CAD, it is important to investigate their role in the context of the genetic architecture of the disease process.

More recently, efforts have been focused at improving our understanding of the genetic basis of CAD pathogenesis to facilitate disease prevention and developing novel pharmacological agents. ⁹ Contemporary multiancestry genome‐wide association studies (GWASs) have identified >250 risk loci for CAD that have shed new light on the putative mechanistic pathways for CAD. ⁹ , ¹⁰ Research on CAD risk loci has facilitated the development of polygenic risk scores (PRSs) to gauge overall genetic susceptibility. However, there is a scarcity of studies examining the interplay between genetic factors and CAD risk factors in large cohorts, particularly on sexual disparities in CAD. Gene–environment interaction (G × E) studies explore whether environmental risk exposures exert varying effects across distinct genotypic groups, potentially elucidating CAD risk components beyond additive genetic effects^{. 11} G × E studies have helped improve our understanding of the mechanistic role of lifestyle risk factors, such as smoking, on cardiovascular health. A multiancestry gene‐smoking interaction study of 387 272 individuals identified 13 new loci that explained the differential impact of smoking status on lipid biomarkers. ¹² Another gene‐smoking interaction study of 60 919 CAD cases and 80 243 controls identified a novel allelic variation in the expression of rs7178051 upstream of the gene ADAMTS7 that was associated with reduced ADAMTS7 expression and conferred stronger CAD protection in never‐smokers than in ever‐smokers. ¹³

The genetic interaction of sex hormones, namely SHBG (sex hormone‐binding globulin), total testosterone (TT), bioavailable testosterone (BAT), and estradiol have been evaluated in PRS studies to identify novel single nucleotide polymorphisms (SNPs) associated with the development of intelligence and addictive behaviors such as alcohol consumption. ¹⁴ However, their interaction with disease loci in the context of CAD risk has not been evaluated. Consequently, in the present report, we used a large cohort of European ancestry to carry out a genome‐wide sex‐stratified study of CAD by incorporating genetic interactions of sex hormone levels, namely SHBG, TT, and BAT.

METHODS

UK Biobank

The UK Biobank (UKB) is a cohort of >500 000 individuals from the United Kingdom for whom detailed phenotypic and genotypic data were collected. ¹³ All participants provided written informed consent for their data to be used for health‐related research at the time of recruitment, and the UKB received ethical approval from the National Information Governance Board for Health and Social Care and the National Health Service North West Center for Research Ethics Committee. ¹⁵ Details of the study protocol and data collection are available online (https://www.ukbiobank.ac.uk/media/gnkeyh2q/study‐rationale.pdf). This research has been conducted using the UKB Resource under application number 34 031.

Study Cohort

Participants in the UKB were queried for the presence of CAD using International Classification of Diseases, Tenth Revision (ICD‐10) codes I20‐I25 (covering ischemic heart diseases) and Office of Population Censuses and Surveys Classification of Interventions and Procedures, Version 4 (OPCS4) codes K40‐K46, K49‐K50, and K75 (which include replacement, transluminal balloon angioplasty, and 2 other therapeutic transluminal operations on coronary artery and percutaneous transluminal balloon angioplasty and insertion of stents) in accordance with Huang et al. ¹⁶ A participant was defined as a CAD case if they had at least 1 occurrence of any of the previous codes.

TT, SHBG, and albumin measurements were attempted in all UKB participants at enrollment. ¹⁷ SHBG (UKB field identification number 30832) and testosterone (UKB field identification number 30852) were measured by chemiluminescent immunoassays (Beckman Coulter; AU5800), whereas albumin (UKB field identification number 30602) was measured by a colorimetric assay (Beckman Coulter; AU5800). BAT was calculated according to the Vermeulen equation ¹⁸ (Figure S1). Participants with hormone levels outside of the reportable range for the assay, with no blood measurement data available or who had biomarker measurements that did not pass quality control, were not considered. These criteria excluded 69 024 individuals with missing TT data, 69 544 with a missing SHBG study, and 65 150 individuals with missing albumin data. The majority of individuals in the UKB (91%) had estradiol measurements below the limit of detection for the assay; thus, estradiol was not included as an interaction term in this study. The missing data of testosterone and SHBG may not be at random, which may lead to biased estimates of the associations. If the missingness is mostly due to measurement below the assay's detectable threshold, removing individuals with missing data could skew the results toward the mean. Future studies with more sensitive assays and more complete measurements of sex hormone data will provide unbiased estimates of the identified associations.

The UKB includes genotype information for 438 427 participants. Genotyping was done using the Applied Biosystems UKB Axiom Array, and quality control was performed using batch effect, plate effect, Hardy‐Weinberg equilibrium, sex effect, and array effect tests, which are described in detail in the UKB genotyping documentation. ¹⁹

Ethnicity was self‐reported during the patients' initial assessment center visit, whereas genetic data were used to estimate kinship coefficients, as described in the UKB genotyping documentation. ¹⁹ The final data set included participants with complete information of age at enrollment, sex, the first 10 principal components, hormone level, and CAD status. Principal components were computed by the UKB, first on a subset of high‐quality unrelated samples before the loadings were projected onto all samples. The full process is described in the UKB genotyping documentation. ¹⁹ Only White European subjects with kinship <0.0884 were included in the genetic interaction analyses stratified by sex.

Standard PRSs were generated from 3 GWAS data sets external to the UKB, as described in the UKB PRS documentation (standard PRS for CAD: UKB field identification number 26227). ²⁰ Briefly, genetic variants used to generate PRS weights were required to have an information metric score >0.8 in the UKB, not display large differences in allele frequency between UKB genetically inferred ancestry groups and either gnomAD or 1000 Genomes Project, and not display evidence of departures from Hardy‐Weinberg equilibrium (P>10⁻¹⁰ in any ancestry group). Indels, pseudo autosomal regions, and any variants with minor allele frequency <0.05 in the 1000 Genomes Project data set were excluded. PRS algorithms were built from trait‐specific meta‐analyses using a Bayesian approach, where appropriate combining data across multiple ancestries and related traits. Per‐individual PRS values were calculated as the genome‐wide sum of per‐variant posterior effect size multiplied by allele dosage. Further details on the methodology applied to develop the PRS can be found in the UKB PRS documentation. ²⁰

Statistical Analysis

Sex‐stratified association of CAD with overall genetics and hormone levels was first investigated in logistic regression models using CAD as the outcome and participant age (at enrollment), PRS, and the first 10 principal components as model covariates. All association, genomic, and bioinformatic analyses were performed in a sex‐stratified manner to account for the differences in sex hormone levels in men and women. Interaction terms between PRS and hormone levels were also added to regression models. Only patients with outcome, hormone, and PRS data available were included. The glm function of R version 4.0.3 was used to calculate raw and standardized estimates, which were then converted to odd ratios (ORs).

To better understand how ORs varied by PRS at specific hormone levels, and potential nonlinear PRS effect, men and women were categorized into 2 PRS groups. Within each sex group, we defined high and low PRS groups using the median PRS value as the cutoff. We then tested the interactions between PRS category and each of the sex hormone‐related traits with the CAD outcome using logistic regression models adjusted for age and top 10 principal components.

GWAS Analysis

Gene–environment interaction analysis for millions of samples (GEM) enables GWASs in millions of samples while allowing for multiple exposures, control for genotype‐covariate interactions, and robust inference. ²¹ GEM considers a generalized linear model with interaction terms and, in our case, the model may be written as CAD ~ β₀ + β_CC + β_GG + β_EE + β_GE(G × E), where β₀ is the intercept for the model; β_C and C represent the coefficients and values, respectively, for the covariates (age and first 10 principal components); β_G and G represent the coefficients and values for the genotype, respectively; β_E and E represent the coefficients and values for the exposure (sex hormone), respectively; and β_GE represents the coefficient for the interaction term G × E. Based on this equation, GEM calculates coefficient estimates and standard errors for interaction effects (G × E) and genetic main effects (G), and also conducts joint tests (2 df test) of the genetic main and interaction effects (G + G × E) and reports the P value. The null hypothesis for the test of genetic main effects is H₀: β_G = 0. The null hypothesis for the test of interaction effects is H₀: β_GE = 0. The null hypothesis for the joint test is H₀: β_G = β_GE = 0.

Within the same analysis, GEM also considers a standard GWAS model with no interaction terms (CAD ~ β₀ + β_CC + β_GG + β_EE) and reports coefficient estimates, standard errors, and P values for the marginal genetic effects (G) of this model. It should be noted that within this article, marginal effects will refer to genetic effects from a model with no interaction terms, and genetic main effects will refer to genetic effects from a model with interaction terms. GEM version 1.4.2 was used to conduct sex‐stratified gene–environment interaction analysis with CAD as the outcome and sex hormones as exposures under default settings.

To account for the effect of menopause on hormone levels, we performed a sensitivity analysis stratified by menopause status for any significant interactions that were discovered in women. Menopause status was collected via patient questionnaire in the initial assessment center visit (https://biobank.ndph.ox.ac.uk/showcase/exinfo.cgi?src = baseline_data). In a separate analysis, patients' menopause status was also included as a covariate in the GEM model along with age at enrollment, first 10 principal components, terms for the main effect of SNP and sex hormone, and the SNP × sex hormone interaction term; correlation coefficients were evaluated comparing the models with and without menopause as a covariate. We also performed sensitivity analyses for body mass index (BMI) because of the known association of BMI with sex hormone levels and for hormone replacement therapy, in which individuals known to have taken hormone replacement therapy were excluded.

Bioinformatic Analysis

Functional Mapping and Annotation of Genome‐Wide Association Studies (FUMA) is an online platform that can be used to annotate, prioritize, visualize, and interpret GWAS results. ²² The SNP2GENE function of FUMA version 1.5.1 was used, under default settings, to analyze the robust marginal, joint, and interaction effects summary statistics output from GEM. For each analysis, FUMA identified genomic‐risk loci, lead SNPs, independently significant SNPs, and candidate GWAS‐tagged SNPs. The definitions of these terms can be found in the FUMA documentation and in Table S1. ²² FUMA also conducts gene‐property analysis to identify tissue specificity of the phenotype.

Nonoverlapping marginal and joint risk loci (ie, SNPs with a 250 kb window and in linkage disequilibrium of r ² ≥ 0.6 with 1 of the independent SNPs) were identified using the GenomicRanges package from Bioconductor. ²³ This software package allows for identifying and defining genomic ranges, genomic positions, and groups of genomic ranges from GWAS results to streamline the process of identifying nonoverlapping loci.

We also performed expression quantitative trait loci analysis in addition to our main G × E analysis to identify genetic variants that may affect gene expression levels and help explain the biological mechanism involved in the genetic association of sex hormones with CAD risk (further details are provided in Data S1). Colocalization analysis was performed to investigate the association of gene expression with SNPs identified in the main and interaction effects in our analysis (further details are provided in Data S1).

RESULTS

Study Cohort

There were 486 162 UKB participants with CAD and CAD PRS information available. Of these, 364 573 had TT measurements, 364 897 had SHBG measurements, and 295 480 had TT, SHBG, and albumin measurements allowing for the calculation of BAT. Summary statistics of the TT study cohort, which is representative of all 3 cohorts, can be found in Table 1. The number of women and men in the GEM analyses for TT, SHBG, and BAT were 181 182, 195 271, and 144 840, and 184 128, 170 391, and 151 233, respectively.

Table 1.

Characteristics of the Cohort Used During the Total Testosterone × Polygenic Risk Score Interaction Analysis

				Men controls		All men
N (%)	169 189 (55.84)	11 675 (60.92)	180 864 (100)	154 641 (56.26)	29 068 (61.06)	183 709 (100)
Age, y	55.84±7.94	60.92±6.53	56.17±7.96	56.26±8.11	61.06±6.45	57.02±8.06
BMI, kg/m2	26.97±5.09	29.26±5.83	27.12±5.18	27.63±4.14	29.00±4.54	27.85±4.23
Cholesterol, nmol/L	5.91±1.11	5.68±1.29	5.90±1.12	5.58±1.08	5.04±1.24	5.49±1.13
PRSCAD	−0.19±0.95	0.12±0.98	−0.17±0.95	−0.24±0.94	0.19±0.95	−0.17±0.95
TT, nmol/L	1.12±0.63	1.09±0.67	1.12±0.63	12.09±3.71	11.43±3.69	11.98±3.71
SHBG, nmol/L	62.47±30.31	55.2±28.63	62.00±30.26	39.83±16.75	40.26±16.84	39.9±16.76
BAT, nmol/L	0.36±0.25	0.37±0.28	0.36±0.26	5.27±1.56	4.88±1.53	5.21±1.56
T2D (%)	8786 (5.19)	2449 (20.98)	11 235 (6.21)	13 434 (8.69)	7308 (25.14)	20 742 (11.29)
Current smoker (%)	15 120 (8.36)	1557 (0.86)	16 677 (9.22)	18 374 (10.00)	3911 (2.13)	22 285 (12.13)
Former smoker (%)	53 022 (29.31)	4336 (2.40)	57 358 (31.71)	57 813 (31.47)	14 285 (7.78)	72 098 (39.25)
Nonsmoker (%)	100 499 (55.56)	5712 (3.16)	106 211 (58.72)	77 967 (42.44)	10 720 (5.84)	88 687 (48.28)
Postmenopause (%)	100 893 (59.63)	8528 (73.04)	109 421 (60.49)	NA	NA	NA

Data are presented as count (percentage) or mean±SD, as appropriate. Cases refer to participants with CAD; controls refer to participants with no CAD.

BAT indicates bioavailable testosterone; BMI, body mass index; CAD, coronary artery disease; PRS_CAD, polygenic risk score for CAD; SHBG, sex hormone‐binding globulin; T2D, type 2 diabetes; and TT, total testosterone.

Association of Sex Hormones and Overall Genetics (PRS) With CAD

A logistic regression model was built to examine whether interactions between overall genetics (PRS) and the sex hormones (TT, SHBG, and BAT) were associated with CAD in men and women separately (Table 2). Holding sex hormone levels at 0, the PRS_CAD was consistently associated with increased odds of CAD in women (OR, 1.40–1.43; P<2E‐16) and men (OR, 1.57–1.73; P<2E‐16). Holding PRS_CAD at 0, TT was associated with lower odds of CAD in men (OR, 0.87 [95% CI, 0.85–0.88]) with no association in women (OR, 1.01 [95% CI, 0.99–1.03]). SHBG was associated with lower odds of CAD in both men (OR, 0.87 [95% CI, 0.86–0.89]) and women (OR, 0.80 [95% CI, 0.79–0.82]). BAT behaved inversely between the sexes, with lower odds of CAD in men (OR, 0.93 [95% CI, 0.91–0.94]) and higher odds of CAD in women (OR, 1.08 [95% CI, 1.06–1.10]). Only SHBG in men showed a significant interaction with the PRS (OR, 0.98 [95% CI, 0.97–0.99]; P=4.57E‐03), indicating that every unit higher in SHBG decreases the OR of PRS_CAD by approximately 2%.

Table 2.

Logistic Regression Model Results for Hormone × PRS (PRS for CAD) Interaction Analysis With Age and Principal Components 1 to 10 as Covariates

		Women		Men
	Predictor	OR (95% CI)	P‐value	OR (95% CI)	P value
TT model	PRSCAD	1.41 (1.36–1.46)	<2E‐16	1.68 (1.60–1.76)	<2E‐16
	TT	1.01 (0.98–1.03)	0.43	0.87 (0.85–0.88)	<2E‐16
	PRSCAD × TT	1.00 (0.98–1.02)	0.94	0.99 (0.99–1.01)	0.31
SHBG model	PRSCAD	1.43 (1.37–1.49)	<2E‐16	1.73 (1.66–1.79)	<2E‐16
	SHBG	0.80 (0.79–0.82)	<2E‐16	0.87 (0.86–0.89)	<2E‐16
	PRSCAD × SHBG	0.98 (0.96–1.00)	0.13	0.98 (0.97–0.99)	4.57E‐03
BAT model	PRSCAD	1.40 (1.36–1.45)	<2E‐16	1.57 (1.48–1.65)	<2E‐16
	BAT	1.08 (1.06–1.10)	<2E‐16	0.93 (0.91–0.94)	<2E‐16
	PRSCAD × BAT	1.00 (0.98–1.02)	0.90	1.01 (1.00–1.03)	0.08

Hormone refers to either TT, SHBG, or BAT. The interaction variable in the table refers to the interaction between the PRS and hormone in question. The ORs shown are per change in SD. BAT indicates bioavailable testosterone; CAD, coronary artery disease; OR, odds ratio; PRS, polygenic risk score; PRS_CAD, polygenic risk score for CAD; SHBG, sex hormone‐binding globulin; and TT, total testosterone.

We also tested the PRS‐by‐sex hormone interaction associated with CAD using a binary PRS variable with a cutoff of median PRS. Similar to the interaction analysis results using continuous PRS, only the PRS × SHBG interaction term in men was significant (P=0.02) (Table S2).

SNP × Sex Hormone Interactions Associated With CAD

There were significantly more marginal and joint genomic‐risk loci associated with CAD in men compared with women in each hormone analysis (Table 3). However, of the male TT and SHBG loci, only 1 joint locus (per hormone) did not overlap with any marginal loci (Table 4). Three of the male BAT joint loci did not overlap with any marginal loci. In women, there were 5 joint loci, all in the TT analysis, that did not overlap with any marginal loci; 2 of these loci (rs141224235 and rs71021448) were also identified via interaction effects. Notably, these loci correspond to the THADA and TRDN genes, respectively (Table 5). The Miami plots depicting the results of the marginal and joint effects for each hormone group in men and women, as well as the corresponding locus plots for the THADA and TRDN regions, are provided in Figures S2–S11. Venn diagrams of overlapping loci identified by including each interaction term are also provided in Figures S12 and S13. It is noteworthy that the detected associations were also tested in the Asian and African ancestry cohorts from the UKB, where they did not reach genome‐wide significance (data provided in Table S3).

Table 3.

Number of Genomic‐Risk Loci Identified by Functional Mapping and Annotation of Genome‐Wide Association Studies for Each Hormone and Effect

Hormone	Effect	Women	Men
		Number of genomic loci	Number of genomic loci
TT	Interaction	2	0
TT	Marginal	3	38
TT	Joint	8	33
TT	Joint, not marginal	5	1
SHBG	Interaction	0	0
SHBG	Marginal	3	32
SHBG	Joint	3	26
SHBG	Joint, not marginal	0	1
BAT	Interaction	0	0
BAT	Marginal	3	28
BAT	Joint	2	25
BAT	Joint, not marginal	0	3

BAT indicates bioavailable testosterone; SHBG, sex hormone‐binding globulin; and TT, total testosterone.

Table 4.

Genomic‐Risk Loci Identified Through Joint Effects That Do Not Overlap Genomic‐Risk Loci Identified Through Marginal Effects in Men and Women

Hormone	SNPID	Chromosome	Position	EA	NEA	EAF	P value joint	Beta (SE) interaction	P value interaction	Beta (SE) genetic main effects	P value genetic main effects	Nearest gene
Women
TT	rs141224235	2	43 810 346	C	T	1.27E‐02	1.68E‐09	−0.528 (0.09)	1.52E‐08	−0.093 (0.06)	0.123	THADA
TT	rs142382396	6	93 677 410	C	T	1.04E‐02	3.11E‐09	−0.241 (0.07)	2.21E‐04	−0.268 (0.06)	1.81E‐05	ATF1P1
TT	rs71021448	6	123 774 735	TA	T	0.09	4.39E‐08	−0.176 (0.03)	6.25E‐09	3.24E‐04 (0.02)	0.989	TRDN
TT	rs117950365	9	76 027 162	A	G	1.01E‐02	1.46E‐08	−0.274 (0.06)	1.14E‐05	−0.197 (0.07)	0.003	RP11‐404E6.1
TT	rs142077273	13	60 488 354	C	T	1.24E‐02	2.09E‐08	−0.146 (0.06)	0.009	−0.278 (0.06)	8.47E‐07	DIAPH3
Men
TT	rs560408926	1	151 721 083	T	G	0.09	1.09E‐08	−0.014 (4.5E‐03)	0.002	−0.088 (0.02)	3.69E‐08	RNU6‐662P
SHBG	rs73029806	6	159 846 076	C	A	0.08	4.40E‐08	0.003 (1.1E‐03)	0.004	−0.088 (0.02)	5.03E‐07	RP11‐125D12.2
BAT	rs2630759	5	176 915 669	C	A	0.49	1.46E‐08	0.036 (0.01)	1.12E‐05	−0.033 (0.01)	0.002	PDLIM7
BAT	rs111532669	9	98 208 964	T	C	0.03	1.76E‐08	−0.057 (0.02)	0.007	−0.171 (0.03)	1.87E‐08	PTCH1
BAT	17:62398954_CA_C	17	62 398 954	C	CA	0.54	4.11E‐09	0.032 (0.01)	7.91E‐05	0.058 (0.01)	3.09 E‐08	RPL31P57

BAT indicates bioavailable testosterone; EA, effect allele; EAF, effect allele frequency; NEA, noneffect allele; SHBG, sex hormone‐binding globulin; SNP ID, single nucleotide polymorphism identification number; and TT, total testosterone.

Table 5.

Genomic‐Risk Loci Identified Through Interaction Effects in Women

Hormone	SNPID	Chromosome	POS	EA	NEA	P value interaction		Beta (SE) interaction		P value genetic main effect		Beta (SE) genetic main effect		Nearest gene
						Women	Men	Women	Men	Women	Men	Women	Men
TT	rs141224235	2	43 810 346	C	T	1.52E‐08	0.992	−0.528 (0.09)	1.18E‐04 (1.18E‐02)	0.123	9.77E‐03	−0.093 (0.06)	−0.126 (0.04)	THADA
TT	rs71021448	6	123 774 735	TA	T	6.25E‐09	0.313	−0.176 (0.03)	−4.65E‐03 (4.61E‐03)	0.989	0.529	3.24E‐04 (0.02)	6.03E‐03 (0.02)	TRDN

The estimates and corresponding P values for the same loci in men have also been provided. EA indicates effect allele; NEA, noneffect allele; POS, position; SNP ID, single nucleotide polymorphism identification number; and TT, total testosterone.

In the sensitivity analysis stratified by menopause status, we did not find any genome‐wide significant SNPs in the premenopausal women, whereas we discovered 2 interaction loci in postmenopausal women: rs141224235 (2:43810346:C:T, nearest gene THADA) and rs117051048 (9:12625500:C:T, nearest gene TYRP1; Figure S14). The interaction effect β coefficients for the models with and without menopause as a covariate were highly correlated (r = 0.885), and the correlation between genome‐wide significant SNPs was even higher (r = 0.994). The correlation plot is shown in Figure S15.

In the sensitivity analysis adjusted for BMI, we found another significant interaction with SNP rs55880969 and TT in women, with the nearest gene being TMTC2 (OR, 0.64 [95% CI, 0.55–0.74]); P=5.74 E‐9; Figure S16). The correlation coefficient for the β effects in models with and without BMI as a variable was high (correlation coefficient for genome‐wide significant SNPs = 0.995; Figure S17).

In the sensitivity analysis in which women who had undergone hormone replacement therapy were excluded, we found significant interactions with rs10946167 (nearest gene RPS6KA2; OR, 0.66 [95% CI, 0.52–0.81]; P=1.67 E‐8) and rs201573028 (nearest gene SLC9B1P3; OR, 0.80 [95% CI, 0.72–0.88]; P=2.07 E‐8; Figure S18). The correlation coefficient for the β effects in models including and excluding women who have undergone hormone replacement therapy was high (correlation coefficient for genome‐wide significant SNPs = 0.993; Figure S19).

Expression quantitative trait loci analysis identified 1 significant cis‐expression quantitative trait loci of the SNP rs141224235 that also mapped to the THADA gene (further details are provided in Table S4). Transcriptome‐wide association study and colocalization analysis did not reveal any evidence of colocalization between gene expression profile of testosterone and THADA or TRDN genes (further details are provided in Data S1 and Tables S5 and S6).

DISCUSSION

In this large‐scale genome‐wide study of CAD, we systematically investigated how sex hormone levels influence the identification of CAD loci through SNP‐sex hormone interactions among men and women. We performed a genome‐wide analysis of SNP‐sex hormone interactions in 180 864 women and 183 709 men of European ancestry enrolled in the UKB. We identified 45 loci in men and 8 loci in women that reached genome‐wide significance for CAD risk. Among men, 3 genomic‐risk loci were identified for BAT, 1 for TT, and 1 for SHBG using 2 df joint effects that did not overlap those identified with marginal effects. Among women, 5 genomic‐risk loci were identified, all for TT, using joint effects that did overlap with those identified with marginal effects. Of these 5 loci, 2 loci were independently identified with 1 df interaction analysis.

Identification of new loci in this G × E analysis demonstrates the importance of incorporating sex hormones in GWASs evaluating CAD risk. Our interaction analysis identified rs141224235, located near THADA, as a possible mediator of the association of sex hormone levels, particularly TT levels, with CAD risk among women but not in men. The THADA gene has been identified to play a key role in transfer RNA processing and modification in the nucleus and cytosol. THADA was initially identified in thyroid adenomas, where chromosomal rearrangements result in the disruption of THADA and its fusion to an intron of PPARG (peroxisome proliferator‐activated receptor gamma), thereby promoting tumor genesis. ²⁴ THADA SNP rs7578597 was recently reported to be associated with type 2 diabetes in a large European GWAS. ²⁵ More recently, THADA has been shown to play an important role in the pathogenesis of polycystic ovarian syndrome in a GWAS of Han Chinese women. ²⁶ The newly discovered risk loci with THADA provides biological plausibility for the role of androgens, especially testosterone levels in the pathogenesis of CAD. Polycystic ovarian syndrome is one of the most common endocrine disorders in women of reproductive age that is characterized by elevated androgen levels. The disease can manifest across a spectrum of heterogeneous phenotypes, but it has been observed that patients with hyperandrogenism carry the highest cardiovascular disease risk. Polycystic ovarian syndrome has been observed to be associated with deranged metabolic health (dyslipidemia, insulin resistance, hypertension, diabetes) and vascular dysfunction (increased coronary calcium, increased carotid intima‐media thickness, and endothelial dysfunction). ²⁷

Our gene–sex hormone interaction analysis also identified rs71021448 on the TRDN gene as a key player of the association of total testosterone levels with CAD risk in women. TRDN gene codes a membrane bound protein triadin that plays a role in skeletal muscle and cardiac muscle excitation‐contraction coupling by regulating Ca⁺² release from the intracellular sarcoplasmic reticulum. ²⁸ Mutations in the TRDN gene have been implicated in catecholaminergic polymorphic ventricular tachycardia, an inherited fatal cardiac arrhythmia that can cause sudden death in juveniles and young adults as well as an autosomal recessive form of long‐QT syndrome, known as triadin knockout syndrome. ²⁹ Given its expression on skeletal muscle membrane, patients with homozygous TRDN gene mutations have also been observed to develop myopathy. ³⁰

In the secondary analyses adjusted for BMI, we found a significant interaction with TT in women, with the nearest gene being the TMTC2 gene. The TMTC2 gene (transmembrane O‐mannosyltransferase targeting cadherins 2) codes for a protein that is an integral membrane protein localized to the endoplasmic reticulum. This protein binds to the calcium uptake pump SERCA2B (Sarco/endoplasmic reticulum Ca²⁺ ATPase) and the carbohydrate‐binding chaperone calnexin, and has been reported to play a role in calcium homeostasis in the endoplasmic reticulum. This gene has previously been linked with Kawasaki disease, cardiac troponin I level, and high‐density lipoprotein cholesterol levels in patients with HIV and with hereditary nonsyndromic sensorineural hearing loss.

There are several limitations in this large‐scale genome‐wide investigation incorporating gene–sex hormone interactions. The UKB is primarily a European ancestry‐based repository with limited representation of African, Asian, and American Indian ancestry that limits the applicability of our findings. We were limited by the availability of data sets with well‐genotyped individuals as well as hormone data to validate our findings. The implementation of multiple hypothesis tests, multiple phenotypes, and multiple exposures may contribute to inflation at some level, and striking a balance between false‐positive and false‐negative rates in genomic interaction analysis remains challenging. There is also potential for residual confounding, particularly in the PRS_CAD analyses, where few potential confounders are accounted for. Bioavailable testosterone levels were not measured directly and were derived using total testosterone levels and albumin levels, as previously described. ¹⁸ Estradiol levels were missing for a majority of the individuals, which prevented us from evaluating their genetic interactions with CAD. Despite these limitations, the present report is bolstered by its large sample size and its incorporation of sex‐stratified analysis and sex hormone interactions.

In conclusion, this large‐scale gene–sex hormone interaction study of CAD genetics identified 45 loci in men and 8 loci in women, 2 of which were only identified by evaluating the interaction of TT levels with genotypes. These findings underscore the importance of future G × E studies evaluating the interaction of sex hormones and other environmental exposures (such as smoking and air pollution) to identify new risk loci for CAD, as well as other complex cardiovascular traits including arrythmias and inherited cardiomyopathies, among others. Equal importance should also be given to the future study of genetic interaction of sex hormones with other complex cardiometabolic traits such as obesity, diabetes, hypertension, and hyperlipidemia.

Sources of Funding

This research has been conducted using the UKB Resource under application number 34 031. It is supported in part by funding from the National Institutes of Health grants P01‐HL154996 and R01‐HL156991.

Disclosures

None.

Supporting information

Data S1 Supplemental Methods

Tables S1–S6

Figures S1–S19

References 31–36