Genetically predicted androgenic profiles and adverse cardiac markers: a sex‐specific Mendelian randomization study

Abstract Aims Observational evidence suggests associations between sex hormone levels and heart failure (HF). We used sex‐specific genetic variants associated with androgenic sex hormone profiles to investigate the causal relevance of androgenic sex hormone profiles on cardiac structure and function and HF using Mendelian randomization (MR). Methods and results Sex‐specific uncorrelated genome‐wide significant (P < 5 × 10−8) variants predicting sex hormone‐binding globulin (SHBG), total testosterone, and bioavailable testosterone were extracted from summary statistics of genome‐wide association study (GWAS) on 425 097 participants in the UK Biobank. Sex‐specific gene–outcome association estimates were computed for left ventricular ejection fraction (LVEF), left ventricular end‐diastolic and end‐systolic volumes (LVEDV and LVESV, respectively), left ventricular stroke volume (LVSV), cardiac index, and cardiac output in 11 528 female and 14 356 male UK Biobank Imaging Study participants and for incident or prevalent HF in an external cohort of 47 309 cases and 930 014 controls. Inverse‐variance weighted MR was the primary analysis method. In females, higher genetically predicted bioavailable testosterone was associated with lower LVEDV [β per nmol/L = −0.11 (−0.19 to −0.03), P = 0.006], lower LVESV [β = −0.09 (−0.17 to −0.01), P = 0.022], lower LVSV [β = −0.11 (−0.18 to −0.03), P = 0.005], lower cardiac output [β = −0.08 (−0.16 to 0.00), P = 0.046], and lower cardiac index [β = −0.08 (−0.16 to −0.01), P = 0.034] and a higher risk of HF [odds ratio 1.10 (1.01–1.19), P = 0.026] on external validation analysis in larger scale, sex‐adjusted GWAS data. Higher genetically predicted SHBG was associated with higher LVEDV [β per nmol/L = 0.17 (0.08–0.25), P = 2 × 10−4], higher LVESV [β = 0.13 (0.05–0.22), P = 0.003], and higher LVSV [β = 0.18 (0.08–0.28), P = 2 × 10−4]. In males, higher genetically predicted total and bioavailable testosterone was associated with lower LVESV [β = −0.07 (−0.12 to −0.02), P = 0.007] and LVEF [β = −0.11 (−0.18 to −0.04), P = 0.003], respectively. Conclusions This study supports a causal effect of pro‐androgenic sex hormone profiles in females on adverse markers of left ventricular structure and function typically associated with HF with preserved ejection fraction and with HF. There was weaker evidence of association in males.


Introduction
Heart failure (HF) is a major global health threat, affecting 64.3 million worldwide. 1,2Trends in lifetime HF risk vary substantially between sexes, suggesting a potential role of sex hormones in its pathogenesis.
Hyperandrogenic states, characterized by increased levels of bioavailable testosterone, commonly occur in both males and females due to physiological, pathological, and iatrogenic factors.Physiologically, females are known to switch to a proportionally more pro-androgenic hormonal profile after the menopause, 3 and a transient physiological increase in testosterone levels is known to occur after exercise in both males and females. 4Pathologically, hyperandrogenic states might occur with adrenal tumours and, for females, in the much more common polycystic ovarian syndrome (PCOS). 5Finally, iatrogenic increases might occur secondary to the use of testosterone-based hormonal therapies.In males, the use of testosterone replacement therapy peaked as high as 3.2% among adults in the United States in 2011, reducing to 1.3% in 2016 following research linking its use to heightened cardiovascular risk. 6,7In birth-assigned females, the use of testosterone-based gender-affirming hormonal therapy is also on the rise.While up-to-date figures regarding prescription of these therapies are currently lacking, population-based estimates of gender incongruence for birth-assigned female persons range from 0.4% to 1.2%. 8dditionally, multiple hormone replacement therapy (HRT) preparations used during the peri-menopause or postmenopause contain testosterone. 9he cardiovascular associations of hyperandrogenism vary between males and females.In males, an inverse association between androgen levels and cardiovascular risk has been reported, 10 and concordantly, androgen deprivation therapies have been associated with improved cardiometabolic risk profiles. 11However, prior studies have reported a paradoxically higher rate of major cardiovascular events. 7Additionally, randomized studies of active testosterone administration among at-risk males have been generally null, 12,13 and some even describe harm. 14,15The relationship is more complex in females.Observational studies in pre-menopausal females describe an association between pro-androgenic hormonal profiles and cardiovascular risk, 16 and PCOS is known to be associated with adverse cardiometabolic profile and higher cardiovascular event rates. 17In post-menopausal females, results are conflicted, with some observational studies reporting a protective effect of higher androgen levels 18,19 and others a detrimental one. 20,21Cohort studies in transgender men undergoing testosterone-based hormone therapy have reported a higher risk of cardiovascular disease and adverse cardiometabolic profiles. 22,23he observational nature of currently available evidence is a major limitation to causal inference because observational associations are liable to residual confounding and bias.
Mendelian randomization (MR) is a method that leverages 'natural' randomization to high or low genetic predisposition to risk factors within an instrumental variable framework to explore the causal relevance of risk factors on an outcome.As the allocation to genetic risk is random, this is akin to randomization in a clinical trial and thus less liable to influence by observational confounding.
Improving current understanding regarding sex hormones and cardiovascular risk is a key public health priority.First, it might lead to improved life course risk stratification relating to physiological, pathological, or iatrogenic hormonal changes.Second, it may better inform care for people receiving testosterone-based hormonal therapy by improving knowledge of cardiovascular effects.
This study aims to leverage genetic variants associated with bioavailable testosterone, as well the closely related measures of sex hormone-binding globulin (SHBG) and total testosterone, to evaluate their causal relevance on multiple cardiac magnetic resonance (CMR) imaging markers of cardiac structure and function, and HF risk.

Ethics and data access
The study is reported using the methodological framework from the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guidelines 24 and is in line with recommendations outlined in the Sex and Gender Equity in Research guidelines. 25For the purpose of sex-specific analyses, sex was defined as that which was genetically inferred.
Summary-level data from publicly available genome-wide association study (GWAS) summary statistics were used and are available to download freely from the cited publications.Relevant ethical approval and participant consent were obtained in each individual study.Individual-level data in the UK Biobank were accessed under Application 13784.Participants provided written informed consent for data release for research purposes within approved applications.][28] All statistical analyses were performed using R Version 4.1.1(2021-02-15) 29 using the MendelianRandomization 30 and TwoSampleMR packages. 30

Instrumental variable selection
Sex-specific instrumental variants were extracted from Ruth et al.'s GWAS on 425 097 UK Biobank participants of predominantly European ancestry. 31The primary exposure considered in this study was bioavailable testosterone (males n = 178 782, females n = 188 507, unit = nmol/L).As testosterone exists in a physiological equilibrium between its bioavailable form and its bound form to SHBG, we additionally studied the secondary exposures of SHBG (males n = 180 726, females n = 189 473, unit = nmol/L) and total testosterone (males n = 194 453, females n = 230 454, unit = nmol/L).Given their equilibrium state, it would be expected that any direct association observed for the exposure of bioavailable testosterone would be reflected by an inverse association for SHBG and vice versa.Sex-specific, uncorrelated (r 2 < 0.05) single nucleotide polymorphisms (SNPs) associated with each exposure at genome-wide significance (P < 5 × 10 À8 ) were selected as instrumental variants.The list of instrumental variants for each exposure, with respective effect sizes and effect alleles, is provided in Supporting Information, Tables S1-S6.A summary of the study methodology and data sources is presented in Figure 1.
The methods used to derive measures of cardiac structure and function in the UK Biobank Imaging Study have been previously described. 32The definition of HF in the UK Biobank cohort is presented in Supporting Information, Table S7, which includes diagnoses recorded across all hospital inpatient records.Sex-specific gene-outcome association estimates were extracted from the UK Biobank and UK Biobank Imaging Study.In females, sample sizes were as follows: 11 528 participants for LVEF, LVEDV, LVESV, LVSV, and CO and 11 525 participants for CI.In males, sample sizes were as follows: 14 358 participants for LVEF, LVEDV, LVESV, LVSV, and CO and 14 356 participants for CI.Genetic association estimates for the outcome of HF were extracted from the results of Shah et al.'s GWAS on 47 309 HF cases and 930 014 controls. 33This GWAS was not sex-specific, but the association estimates were adjusted for sex.Validation analysis was performed on sex-specific data from the UK Biobank to look for consistency of estimates, including 4128 cases and 194 687 controls for females and 7713 cases and 161 014 controls for males.

Mendelian randomization
Inverse-variance weighted (IVW) MR with multiplicative random effects 34 was used for primary analysis for all models. 35esults are presented as odds ratios (ORs) with 95% confidence intervals for each unit increment in genetically predicted exposure on HF risk.For the CMR outcomes, results are presented as beta coefficients (β) and 95% confidence intervals, indicating the expected unit change in CMR param- Genetically predicted androgenic profiles and adverse cardiac markers: a sex-specific Mendelian randomization study eter for each nmol/L increment in the genetically predicted exposure.

Sensitivity analyses
Sensitivity analyses were carried out to ascertain potential violations of the instrumental variable assumptions.These include that: (i) the instruments associate with exposures of interest (relevance); (ii) do not have confounders associated with the outcome (independence); and (iii) no horizontal/direct pleiotropic effects exist (exclusion restriction).
To assess the relevance assumption, instrument strength was quantified using F-statistics, reported in Supporting Information, Table S8.To assess the exclusion restriction assumption, sensitivity analysis was carried out using weighted median MR and MR-Egger. 36The weighted mean estimate has been shown to provide consistent results in simulations when up to 50% of instrumental variants included in the analysis violate this assumption. 37The MR-Egger test detects evidence of directional pleiotropy under a weaker assumption that the instrument strength is independent of direct effects (In-SIDE assumption) 36 through the introduction of an intercept term.
Where evidence of potential pleiotropy was discovered, instrumental variants were examined in more detail through a phenome-wide association search using PhenoScanner (Supporting Information, Table S9).Among phenotypes associated with instrumental variants, candidate phenotypes to adjust for in multivariable MR (MVMR 38 ) analyses were chosen based on prior knowledge of causal associations of the phenotypes with HF.Association estimates between genetically predicted sex hormones and the outcome in question were adjusted by the associations of the instrumental SNPs with body mass index (BMI) and systolic blood pressure (SBP), extracted respectively from Pulit et al.'s 39 (n = 806 834 participants) and Evangelou et al.'s 40 (n = 738 168 participants) GWAS summary statistics.The strength of instruments for the sex hormone trait in the modified multivariable analyses was quantified using conditional F-statistics. 41

Measures of cardiac function
In females, higher genetically predicted bioavailable testosterone was associated with a lower CO [β = À0.08 (À0.16 to 0.00), P = 0.046] and a lower CI [β = À0.08 (À0.16 to À0.01), P = 0.034].No associations were evident for the secondary exposures of genetically predicted SHBG and genetically predicted total testosterone with either outcome, and none of the exposures were associated with LVEF.The results are reported in Figure 2 and Table 1.
Genetically predicted androgenic profiles and adverse cardiac markers: a sex-specific Mendelian randomization study  There was no significant association with HF for genetically predicted SHBG and genetically predicted total testosterone in both the primary and validation analyses, as reported in Tables 1 and 3 and Figure 3.

Measures of cardiac structure
Higher genetically predicted total testosterone was associated with a lower LVESV in males [β = À0.07 (À0.12 to À0.02), P = 0.007], but no associations were observed for genetically predicted bioavailable testosterone and SHBG, as reported in Figure 4. Sensitivity analyses were consistent with the primary analyses, as displayed in Table 2.In males, none of the exposures were associated with either LVEDV or LVSV.
The results are reported in Figure 3.

Measures of cardiac function
In males, higher genetically predicted bioavailable testosterone was associated with a lower LVEF [β = À0.11(À0.18 to À0.04), P = 0.003].No associations with LVEF were observed for genetically predicted SHBG and total testosterone.Additionally, as displayed in Figure 4, none of the genetically predicted exposures were associated with CO or CI.

Heart failure
In the analysis on the external cohort, higher genetically predicted bioavailable testosterone was not associated with HF [OR 1.04 (0.96-1.11),P = 0.359], and neither were genetically predicted total testosterone or SHBG (Table 3 and Figure 4).The null results were consistent in the validation analysis in the UK Biobank, as reported in Table 1.

Discussion
In this study, we explored associations between androgenic sex hormone profiles and multiple markers of cardiac structure and function.A pro-androgenic profile was associated with adverse LV remodelling with features of greater cardiac ageing broadly reflective of an HF with preserved ejection fraction (HFpEF) phenotype.These relationships appeared more prominent in females, in whom higher bioavailable testosterone (and lower SHBG) was associated with a smaller structure (lower LVEDV, LVESV, and LVSV) and function (lower LVSV, CO, and CI).A pro-androgenic profile was associated with a greater likelihood of clinical HF in females, but not in males.Thus, our results demonstrate androgenic hormones as a proponent of poorer cardiovascular health, in particular HF, which appears more important in females than males.
Several observational studies have explored the association between androgenic sex hormone profiles and cardiovascular risk in females, linking to both menopausal and pathological conditions.Post-menopausal females exhibit proportionally more androgenic sex hormone profiles, which has been, albeit inconsistently, 21,42,43 suggested to associate with increased cardiovascular risk. 19,20This pattern is reflected in females with PCOS, who exhibit higher absolute androgen levels and have also been reported to suffer higher risk of cardiovascular disease.However, drawing causal inference from observational studies is limited due to the residual confounding.The present study provides genetic evidence to support a causal role of higher androgen levels in females on adverse cardiac remodelling.This study adds to current literature as the first to employ MR to explore this question, limiting the influence of observational confounding and therefore providing more reliable evidence of causal relationships.
There are several potential mechanisms behind the findings of this study.Females have a 20-fold to 25-fold lower circulating androgen concentration compared with males. 44ong-term androgen receptor activation in females has been shown to predispose to insulin resistance, combined with up-regulation of genes involved in HDL catabolism leading to dyslipidaemia. 16Indeed, Ruth et al. have previously described an association between higher androgen levels and Genetically predicted androgenic profiles and adverse cardiac markers: a sex-specific Mendelian randomization study worse cardiometabolic profiles in females using MR. 31 Additionally, hormonal changes occurring in the peri-menopause, when androgenic transient hormones increase and then decline to 15% of pre-menopausal levels post-menopause, 44 have been shown to accelerate adverse arterial remodelling through proliferation of vascular smooth muscle cells, 45,46 leading to greater arterial stiffness and thus accelerated age-related hypertension, contributing to augmented cardiovascular risk.Physiological levels of testosterone increase nitric oxide synthesis through gene induction, which is not seen with high levels of testosterone. 46Finally, the menopause transition and its concurrent proportionally androgenic state have been associated with significant changes in body composition, 47 including increased deposition of visceral fat.Increased visceral fat, especially in the pericardial compartment 48 and abdominal area, 46 might further augment cardiovascular risk through both endocrine and inflammatory paracrine mechanisms.It is important to note that the above mechanisms form a complex picture and existing research is contradictory alluding to multifactorial explanations linking androgens to cardiovascular risk.
In males, higher androgen levels were associated with reduced LVEF and LVESV.These results are contradictory with a prior MR analysis 49 and also with available observational evidence.In the prior MR analysis, genetically predicted testosterone was associated with HF. 49 The different findings could be attributed to the different health statuses in their study population compared with the UK Biobank, comprising a relatively healthy cohort.Furthermore, there were low numbers of participants with HF used in the prior study 49 and survival bias could have played a part in the UK Biobank.In observational studies, androgen deprivation use as a treatment for prostate cancer 50 or due to hypogonadism, 7,51,52 have been associated with cardiovascular risk.A potential explanation for this apparent equivocality might relate to non-linear effects, whereby those with both extremely low and high androgen levels suffer increased risk.Overall, given the important inconsistencies in available evidence, cardiac pharmacovigilance in the context of both androgen-depriving and androgen-enhancing therapies in males remains an important priority, and further research is needed to explore potential non-linear effects or establish effect modifiers.
In this study, we identified potential pleiotropic effects in the association between bioavailable testosterone and cardiac function in females.This finding requires further discussion.First, this only involves one of the exposures, bioavailable testosterone, and the specific outcomes of LVSV, CO, and CI (not LVESV, LVEDV, and HF).Given the biologically plausible and consistent results observed across other outcomes and exposures, the potential pleiotropy discovered in some specific associations does not substantially alter the broader interpretation of study findings.Second, we further explored this by highlighting potential culprit phenotypes of BMI and SBP and demonstrated a more likely role of BMI in the pleio-tropic effects.This suggests that BMI might either directly or indirectly be responsible for at least part of the observed associations, highlighting it as a key target for surveillance and risk modification.
These results carry important clinical implications.First, the results support a potentially adverse mechanistic role of androgens in the cardiovascular association of multiple pathological states characterized by higher androgen levels in females, such as PCOS, and cardiac disease.The results therefore encourage further research to elucidate whether androgens are an important or primary mediator of this association, as they represent therapeutic targets amenable to intervention.Second, they support further research exploring the long-term cardiovascular health of individuals taking testosterone-based hormone treatments.These include post-menopausal females taking HRT preparations that contain testosterone, as well as individuals undergoing gender-affirming hormone treatments.Gender-affirming hormone treatment use has increased significantly in recent years, and creating a high-quality evidence base regarding long-term cardiovascular health should be a key priority to ensure high standards of care.Evidence among existing studies is conflicting, with some suggesting an increased risk of cardiovascular disease in transgender males undergoing hormone therapy 53,54 yet others suggesting no effect. 22,55,56mportantly, higher risk of hypertension has been reported, suggesting mechanistic effects on the vascular system consistent with the HFpEF imaging phenotype observed in this study.
There are several limitations to discuss.First, magnitude of effect estimates for reported associations cannot be interpreted intuitively as real-life unit changes in outcome expected per unit change in the exposure through pharmacological modification.Measures of exposures in MR studies reflect a lifetime exposure, rather than a time-limited modification in the exposure secondary to pharmacotherapy.Second, we were unable to expand the analysis to other sex hormones (e.g.oestradiol and progesterone) and additional measures of cardiac structure and function (e.g.LV mass) due to the lack of sufficiently powered data available.This should be an important target for future research.Finally, in this study, we utilized two-sample MR methods in a onesample setting as both gene-exposure and gene-outcome association data were derived from the UK Biobank.However, this has been shown to be unlikely to significantly bias analyses in large-scale biobanks. 57

Conclusions
In conclusion, these data support a causal association between bioavailable testosterone in females and adverse measures of cardiac structure and function in a pattern broadly consistent with HFpEF, as well as a higher risk of HF.In males, higher androgen levels were associated with reduced LVEF and LVESV.The results call for further investigation of the prognostic value of sex-specific risk factors typically associated with pro-androgenic states (e.g.PCOS) for cardiovascular risk stratification of females.By highlighting a likely causal role of testosterone, they also encourage exploration of whether this might be a modifiable mediator of the increased cardiovascular risk that is known to occur in PCOS.Additionally, they encourage pharmacovigilance and ongoing research to characterize the long-term cardiovascular health of individuals undergoing testosterone-based therapies for both gender affirmation and hormone replacement.

Figure 2
Figure 2Cardiovascular magnetic resonance in females.Mendelian randomization estimates for the effects of bioavailable testosterone, sex hormone-binding globulin, and total testosterone on cardiac magnetic resonance outcomes in females.CI, confidence interval; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume; LVSV, left ventricular stroke volume.

Figure 3
Figure 3 Heart failure in females and males.Mendelian randomization estimates for the effects of bioavailable testosterone, total testosterone, and sex hormone-binding globulin (SHBG) on the risk of heart failure in females and males.CI, confidence interval.

Figure 4
Figure 4 Cardiovascular magnetic resonance in males.Mendelian randomization estimates for the effects of bioavailable testosterone, sex hormone-binding globulin, and total testosterone on cardiac magnetic resonance outcomes in males.CI, confidence interval; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume; LVSV, left ventricular stroke volume.

Table 1
Mendelian randomization estimates for the effects of androgenic sex hormones on cardiovascular outcomes, using an inverse-variance weighted model CI, confidence interval; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume; LVSV, left ventricular stroke volume; OR, odds ratio; SNPs, single nucleotide polymorphisms.a

Table 2
Mendelian randomization (MR) sensitivity analyses for the effects of reproductive factors on cardiovascular outcomes, using weighted median MR and MR-Egger models

Table 3
Mendelian randomization estimates for the effects of reproductive factors on heart failure, using an inverse-variance weighted model CI, confidence interval; SNPs, single nucleotide polymorphisms.