1. Genetics and Genomics
  2. Medicine
Download icon

Effects of lifelong testosterone exposure on health and disease using Mendelian randomization

  1. Pedrum Mohammadi-Shemirani
  2. Michael Chong
  3. Marie Pigeyre
  4. Robert W Morton
  5. Hertzel C Gerstein
  6. Guillaume Paré  Is a corresponding author
  1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Canada
  2. Thrombosis and Atherosclerosis Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Canada
  3. Department of Medical Sciences, McMaster University, Canada
  4. Department of Biochemistry and Biomedical Sciences, McMaster University, Canada
  5. Department of Medicine, McMaster University, Hamilton Health Sciences, Canada
  6. Department of Kinesiology, McMaster University, Canada
  7. Department of Pathology and Molecular Medicine, McMaster University, Michael G. DeGroote School of Medicine, Canada
  8. Department of Health Research Methods, Evidence, and Impact, McMaster University, Canada
Research Article
  • Cited 0
  • Views 562
  • Annotations
Cite this article as: eLife 2020;9:e58914 doi: 10.7554/eLife.58914

Abstract

Testosterone products are prescribed to males for a variety of possible health benefits, but causal effects are unclear. Evidence from randomized trials are difficult to obtain, particularly regarding effects on long-term or rare outcomes. Mendelian randomization analyses were performed to infer phenome-wide effects of free testosterone on 461 outcomes in 161,268 males from the UK Biobank study. Lifelong increased free testosterone had beneficial effects on increased bone mineral density, and decreased body fat; adverse effects on decreased HDL, and increased risks of prostate cancer, androgenic alopecia, spinal stenosis, and hypertension; and context-dependent effects on increased hematocrit and decreased C-reactive protein. No benefit was observed for type 2 diabetes, cardiovascular or cognitive outcomes. Mendelian randomization suggests benefits of long-term increased testosterone should be considered against adverse effects, notably increased prostate cancer and hypertension. Well-powered randomized trials are needed to conclusively address risks and benefits of testosterone treatment on these outcomes.

eLife digest

Men experience a gradual decline in their testosterone levels as they grow older. However, the effects of testosterone and the consequences of supplementation on the human body have been unclear.

Scientists use so-called randomized controlled trials to establish cause-and-effect and to reduce bias. In these experiments, participants are randomly assigned to a either a treatment group (that receives the intervention being tested) or a control group (that either receives an alternative intervention, a dummy or placebo, or no intervention at all).

Randomization ensures that both groups are balanced, and any resulting differences can be attributed to the treatment. However, randomized controlled trials are time-consuming and expensive, so trials of testosterone have had relatively small numbers of participants and short follow-up periods. This makes it difficult to draw conclusions about any potential effects of testosterone administration on less common diseases in men.

Now, Paré et al. investigated the effects of naturally produced testosterone using Mendelian randomization, which mimics randomized trials by exploiting the fact that parents randomly pass on their unique genetic variants to their children at conception. This random assignment of genetic variants leads to its informal namesake, “nature’s clinical trial”, and provides the ability to study cause-and-effect for any genetically determined factors, such as testosterone levels.

Paré et al. studied the long-term effects of testosterone on 22 diseases previously explored in randomized controlled trials, and hundreds of other traits and diseases that have not been investigated in any randomized controlled trials yet.

The Mendelian randomization analysis made it possible to examine the effects of lifelong naturally elevated testosterone levels on 469 traits and diseases. Paré et al. found that testosterone increased the density of bone mineral and decreased body fat. However, it also increased the risks of prostate cancer, high blood pressure, baldness and a condition affecting the spine. It also increased the number of red blood cells and decreased a marker of inflammation, which may be beneficial or detrimental depending on the context.

This shows that genetic analyses can be powerful methods to prioritize the allocation of limited resources towards investigating the most pressing clinical questions. The results of this study may help inform physicians and patients about the effects of long-term testosterone use. Ultimately, large randomized controlled trials are needed to conclusively address the cause-and-effect on these diseases.

Introduction

In developed countries, rising rates of both serum testosterone level testing and therapy initiation have been observed among older male patients (Handelsman, 2013; Layton et al., 2014). In the USA alone, it is estimated 1.5–1.7% of males are prescribed testosterone (Baillargeon et al., 2018; Jasuja et al., 2017). Randomized clinical trials (RCT) have attempted to elucidate the benefits and risks of testosterone treatment (Bhasin et al., 2018a; Gagliano-Jucá and Basaria, 2019). These studies identified short-term beneficial effects on bone mineral density (BMD), sexual function, body fat and muscle mass, and anaemia; potential adverse effects on venous thrombosis and coronary artery plaque; and no effects on cognitive function, fatigue, or hemoglobin A1c (HbA1c) (Bhasin et al., 2018a; Gagliano-Jucá and Basaria, 2019; Mohler et al., 2018; Snyder et al., 2018). However, given the logistic and financial challenges involved in a well-powered RCT with appropriate follow-up, there is unlikely to be satisfactory evidence regarding long-term effects and risks of adverse outcomes, such as myocardial infarction (MI), stroke and cancer (Gagliano-Jucá and Basaria, 2019). Given the rates of testosterone prescription, efforts to resolve the causal effects of testosterone on health outcomes have important public health implications (Bhasin et al., 2018a).

Mendelian randomization (MR) is a technique for causal inference that leverages the random allocation of genetic variants to infer the unconfounded relationship between an exposure and outcome. Similar to the random assignment of participants to experimental groups in a RCT, genetic variants are randomly allocated at meiosis (Davies et al., 2018). For instance, if individuals genetically randomized to produce higher testosterone develop different rates of cardiovascular disease (CVD), then MR analysis supports a causal effect of testosterone on risk of CVD (Figure 1—figure supplement 1). Notably, this technique has previously replicated RCT findings, among others demonstrating causal roles for LDL cholesterol and dysglycemia on CVD risk (Holmes et al., 2015; Ross et al., 2015). Earlier MR studies investigating the effects of testosterone have demonstrated harmful effects on lipid levels but inconsistent effects on CVD, and they were limited by the small number of genetic variants (Schooling et al., 2018; Zhao et al., 2014). A recent MR study using the UK Biobank identified a large number of genetic variants associated with testosterone and found evidence for harmful effects on several types of cancers but sex-specific effects on type 2 diabetes (T2D) (Ruth et al., 2020). This study highlighted the importance of performing sex-specific analyses for testosterone, but it was focused on glycemic and oncologic traits (Ruth et al., 2020). Therefore, we sought to expand the scope of prior studies by performing a comprehensive scan of the effects of free testosterone on human disease in males.

We hypothesized that MR and genetic risk score (GRS) analyses would enable estimation of the causal effects of longstanding exposure to high levels of free testosterone on health outcomes in males. We first conducted a genome-wide association study (GWAS) for calculated free testosterone (CFT) in male participants of the UK Biobank (n = 161,268) cohort to identify genetic determinants of free testosterone levels. Then, using MR, we investigated the causal effects of lifelong genetically-elevated free testosterone levels on a priori health outcomes previously investigated in RCTs of testosterone treatment, encompassing: expected clinical benefits (physical activity, strength, fat-free body mass, body fat, BMD, dementia, depression) and potential adverse effects (androgenic alopecia, heematocrit, T2D, prostate cancer, benign prostate hyperplasia, blood pressure, CVD, heart failure, ischemic stroke) (Figure 1Bhasin et al., 2018a; Gagliano-Jucá and Basaria, 2019; Mohler et al., 2018; Snyder et al., 2018). Finally, we used GRS to investigate the associations of lifelong genetically-elevated free testosterone levels on 439 health outcomes, encompassing diseases (n = 415) and biomarkers of health (n = 24) (Figure 1).

Figure 1 with 6 supplements see all
Flowchart depicting overall study design.

Free testosterone levels were calculated in males from the UK Biobank cohort. Then, genetic variants were tested for association with levels of CFT and carried forward if: genome-wide significant (p<5×10−8) and unassociated with SHBG (p<0.05). In the subset of unrelated males, these genetic variants were used to investigate the effect of genetically-predicted CFT on: (1) 22 a priori outcomes relevant to suspected effects of testosterone treatment using Mendelian randomization, and (2) 439 outcomes in a hypothesis-free approach using a weighted genetic risk score. CFT, calculated free testosterone; MR, Mendelian randomization; SHBG, sex hormone-binding globulin.

Results

Genetic determinants of CFT in males

To calculate free testosterone levels, 187,524 males in the white, British subset of the UK Biobank cohort were excluded if they had missing levels of total testosterone, SHBG and albumin, or self-reported taking androgen medications. After these exclusions, the study population consisted of 161,268 males with an average CFT of 0.210 nmol/L (Supplementary file 1 - Table 1 and Figure 1—figure supplement 2).

There were 13,338 genetic variants associated with CFT that reached genome-wide significance (p<5×10−8). After removing genetic variants associated with natural-log-transformed SHBG, there were 7048 genetic variants that comprised 93 independent signals carried forward for subsequent genetic analyses (Supplementary file 1 - Table 2 and Figure 1—figure supplement 3). Overall, chip-based heritability of CFT was estimated at 15% (95% CI = 14 to 16), while these 93 independent genetic variants associated with CFT explained 3.7% of the total variance of CFT levels in males from the UK Biobank.

Effect of genetically-predicted free testosterone on 22 a priori health outcomes

In males from the UK Biobank, sample size for the quantitative risk factors ranged from 30,439 to 156,403, while number of cases for dichotomous outcomes ranged from 1003 to 70,283 (Table 1). After adjusting for the 22 outcomes tested, one-sample MR analysis using IVW regression identified significant effects of CFT on hematocrit percentage, body fat-free percentage, body fat percentage, heel BMD, androgenic alopecia, and prostate cancer (Table 1). Each 0.1 nmol/L higher CFT had beneficial effects on increased heel BMD (0.40 SD; 95% CI = 0.25 to 0.54; p=1.10×10−7), increased body fat-free percentage (1.91%; 95% CI = 1.48 to 2.35; p=9.06×10−18), and decreased body fat percentage (−1.88%; 95% CI = −2.31 to −1.45; p=1.65×10−17), but deleterious effects on increased hematocrit percentage (1.37%; 95% CI = 1.12 to 1.62; p=1.03×10−27), risk of prostate cancer (OR = 1.51; 95% CI = 1.21 to 1.88; p=2.1×10−4), and risk of androgenic alopecia (OR = 1.49; 95% CI = 1.19 to 1.86; p=5.28×10−4) (Figure 3—figure supplements 16). Leave-one-out analyses did not identify any outlying individual genetic variants responsible for the observed effects on any significant outcomes.

Table 1
Effect of calculated free testosterone on 22 health outcomes from the UK Biobank relevant to effects of testosterone treatment in males.
OutcomeEffect per 0.1 nmol/L increased CFT (95% CI)P-valueSample Size
Cases/Controls
Outcomes with Expected Clinical Benefits
Body fat-free percentage*1.91% (1.48 to 2.35)9.06E-18154254
Body fat percentage*−1.88% (−2.31 to −1.45)1.65E-17153772
Heel bone mineral density*0.40 SD (0.25 to 0.54)1.10E-0790676
DepressionOR = 1.45 (1.1 to 1.91)7.77E-034725/152485
Accelerometer-based physical activity0.89 milligravity (−0.05 to 1.82)0.0630439
All fractureOR = 0.89 (0.71 to 1.11)0.309133/148077
Handgrip strength0.29 kg (−0.31 to 0.89)0.34156400
All dementiaOR = 1.26 (0.67 to 2.34)0.471003/156207
Outcomes with Potential Adverse Effects
Hematocrit percentage*1.37% (1.12 to 1.62)1.03E-27152872
Prostate cancer*OR = 1.51 (1.21 to 1.88)2.10E-047586/149624
Androgenic alopecia*OR = 1.49 (1.19 to 1.86)5.28E-0470283/85756
Benign prostatic hyperplasiaOR = 1.36 (1.10 to 1.67)3.80E-0310894/146316
Myocardial infarctionOR = 1.23 (1 to 1.53)0.059398/147812
Glucose−0.06 mmol/L (−0.14 to 0.02)0.12138307
Hemoglobin A1c−0.34 mmol/mol (−0.82 to 0.15)0.17149828
All strokeOR = 1.18 (0.90 to 1.56)0.234569/152641
Diastolic blood pressure0.27 mmHg (−0.30 to 0.85)0.35148384
Ischemic strokeOR = 0.92 (0.61 to 1.37)0.672122/155088
Systolic blood pressure−0.12 mmHg (−1.23 to 1.00)0.84148383
Type 2 diabetesOR = 1.02 (0.81 to 1.28)0.8711079/146131
Venous thromboembolismOR = 1.02 (0.74 to 1.4)0.924127/153083
Heart failureOR = 1.01 (0.76 to 1.34)0.954288/152922
  1. * Significant adjusting for Bonferroni correction of 22 outcomes (p<2.27×10−3).

    CFT, calculated free testosterone.

Sensitivity analyses were performed to detect violations of MR assumptions. Egger regression did not detect evidence of directional pleiotropy for any outcomes (pintercept <0.05) (Supplementary file 1 - Table 3). Results using MR-RAPS were consistent with IVW regression method for all significant outcomes (Supplementary file 1 – Table 4). However, MR-PRESSO detected evidence of pleiotropic variants for hematocrit percentage, body fat-free percentage, body fat percentage, heel BMD, androgenic alopecia, whole body fat-free mass, hemoglobin A1C, glucose, handgrip strength, systolic blood pressure, diastolic blood pressure, T2D, and benign prostate hyperplasia (Supplementary file 1 - Table 5). However, removal of pleiotropic variants made no changes to the significance or interpretation of earlier results using IVW regression (Supplementary file 1 - Table 5).

Phenome-wide effects of genetically-predicted free testosterone

To discover novel effects of free testosterone, we tested for the association of a GRS for testosterone with 415 diseases and 24 biomarkers in the same subpopulation of unrelated males from the UK Biobank. Sample size for biomarkers ranged from 118,783 for lipoprotein(a) to 149,940 for total cholesterol, while number of cases for diseases ranged from 876 for ‘localized superficial swelling, mass, or lump’ to 40,960 for ‘hypertension’ (Figure 2—source data 1). After adjusting for the 439 outcomes tested, each 0.1 nmol/L increase in genetically-predicted CFT was significantly associated with beneficial effects on lowered C-reactive protein (β = −0.085 SD; 95% CI = −0.119 to −0.052; p=6.15×10−7) but adverse effects on increased creatinine (β = 0.113 SD; 95% CI = 0.079 to 0.146; p=4.78×10−11), lowered apolipoprotein A (β = −0.018 g/L; 95% CI = −0.026 to −0.01; p=1.55×10−5), lowered HDL (β = −0.074 SD; 95% CI = −0.109 to −0.039; p=3.62×10−5), and increased risks of hypertension (OR = 1.17; 95% CI = 1.08 to 1.26; p=2.83×10−5), and spinal stenosis (OR = 2.03; 95% CI = 1.51 to 2.75; p=3.82×10−6) (Table 2 and Figure 2).

Phenome-wide survey of effects of genetically-predicted calculated free testosterone on 439 health outcomes in males from the UK Biobank.

Logistic or linear regression was used to assess the association of the genetic score for free testosterone against each dichotomous or quantitative outcome, respectively. -log10(p-values) for the association of each outcome on the y-axis are stratified into subcategories on the x-axis. Labelled outcomes were statistically significant adjusting for multiple hypothesis testing (p<1.14×10−4).

Table 2
Effects of calculated free testosterone on 439 health outcomes in males from the UK Biobank significant after adjusting for multiple hypothesis testing using Bonferroni correction (p<1.14×10−4).
OutcomeEffect per 0.1 nmol/L increased CFT
(95% CI)
P-valueSample Size Cases/Controls
Creatinine0.113 SD
(0.079 to 0.146)
4.78 × 10−11149849
C-reactive protein−0.085 SD
(−0.119 to −0.052)
6.15 × 10−7149547
Spinal stenosisOR = 2.03
(1.51 to 2.75)
3.82 × 10−61917/150919
Apolipoprotein A−0.018 g/L
(−0.026 to −0.01)
1.55 × 10−5138185
HDL cholesterol−0.074 SD
(−0.109 to −0.039)
3.62 × 10−5138394
Essential hypertensionOR = 1.17
(1.08 to 1.27)
7.53 × 10−540809/115957
HypertensionOR = 1.17
(1.08 to 1.26)
1.05 × 10−440960/115957
  1. CFT, calculated free testosterone; HDL, high density lipoprotein; GRS, genetic risk score.

As confirmation, we demonstrated the GRS was indeed not associated with natural log-transformed natural log-transformed SHBG levels in males (p=0.12). For all statistically significant outcomes, associations were directionally consistent after removing participants taking blood pressure medication (Supplementary file 1 - Table 6) or cholesterol-lowering medication (Supplementary file 1 - Table 7). Further sensitivity analyses were performed by repeating the one-sample MR analysis using 52 genetic variants associated with total testosterone in males from the UK Biobank (Supplementary file 1 - Table 8). For all statistically significant outcomes, effects observed using total testosterone genetic variants were directionally consistent with CFT, and results for all outcomes are presented in Supplementary file 1 - Tables 9 and 10. Finally, most effect estimates for genetically-predicted testosterone in this study were comparable in magnitude to effect sizes reported in RCTs except bone mineral density (Figure 3).

Figure 3 with 6 supplements see all
Comparison of effect sizes reported in randomized controlled trials and Mendelian randomization analyses.

Error bars indicate 95% confidence intervals around the effect estimate. MR effect estimates are reported in terms of 0.1 nmol/L of CFT to approximate expected effect sizes after initiation of testosterone treatment (Bhasin et al., 2018b).

Discussion

We herein perform MR and GRS analyses of CFT to identify effects of endogenous free testosterone in males on 461 health outcomes. All effects are reported in terms of 0.1 nmol/L of CFT to approximate expected effect sizes after initiation of testosterone treatment (Bhasin et al., 2018b). Among 22 a priori outcomes with suspected effects based on RCTs of testosterone treatment, MR analyses demonstrated that each 0.1 nmol/L increase in CFT was associated with adverse effects on increased risk of prostate cancer, risk of androgenic alopecia, and hematocrit percentage, but beneficial effects on increased heel BMD, increased body fat-free percentage and decreased body fat percentage. Findings on body composition, hematocrit, and BMD are consistent with short-term effects in randomized trials of testosterone treatment (Bhasin et al., 2018a). Although testosterone treatment has not been conclusively shown to increase risk of prostate cancer and androgenic alopecia in RCTs, androgen suppression therapies, such as of 5α-reductase inhibitors, are used as treatment for androgenic alopecia and prostate cancer (Adil and Godwin, 2017; Andriole et al., 2010). The increased risk of prostate cancer replicates effects of testosterone observed in a previous MR analysis using independent data from the PRACTICAL consortium, and further supports the role of testosterone in development of these outcomes. As the leading cause of cancer among men, the predicted 1.5-fold increased risk as a result of changes in testosterone observed after initiation of testosterone treatment warrants further investigation in clinical trials and greater scrutiny in at-risk patient populations (American Cancer Society, 2019; Bhasin et al., 2018b). Furthermore, these results cast doubt on cardiovascular, cognitive, or metabolic benefit for increased testosterone, as we do not find evidence of a beneficial effect of CFT on hard endpoints, such as dementia, MI, stroke, fractures, or T2D (Aukrust et al., 2009). Most of the estimates from MR analyses were comparable with effect sizes from RCTs (Figure 3). There was only significant heterogeneity between the effects on BMD for MR and RCT, but it is difficult to make direct comparisons due to variable change in testosterone levels after administration of testosterone in each RCT, different methods and anatomical sites of BMD estimation, and differences between short-term effects in RCTs relative to lifelong effects in MR.

Among the remaining outcomes without well-established effects from RCTs, we identified evidence of novel associations between an increased GRS for CFT with adverse effects on creatinine, HDL, apolipoprotein A, hypertension, and spinal stenosis, but beneficial effects on C-reactive protein. Higher genetically-predicted free testosterone was associated with increased creatinine (β = 0.113 SD; 95% CI = 0.079 to 0.146; p=4.78×10−11). Mechanistically, effects of testosterone on renal function are unclear, but this effect may be mediated through the known effect of testosterone on increased muscle mass which is tightly related to serum creatinine (Carrero et al., 2009; Filler et al., 2016; Schutte et al., 1981). HDL cholesterol (β = −0.074 SD; 95% CI = −0.109 to −0.039; p=3.62×10−5) and its main protein component, apolipoprotein A (β = −0.018 g/L; 95% CI = −0.026 to −0.01; p=1.55×10−5), were both decreased with higher genetically-predicted free testosterone. Likewise, the Testosterone Trials found male participants over 65 years of age randomized to testosterone experienced mildly lowered HDL cholesterol levels after 12 months (Mohler et al., 2018; Snyder et al., 2018). Higher free testosterone was associated with decreased C-reactive protein (CRP) (β = −0.085 SD; 95% CI = −0.119 to −0.052; p=6.15×10−7). Although the Testosterone Trials did not find any change in CRP in its testosterone arm, testosterone is widely-believed to have suppressive effects on the immune system which may extend to markers of inflammation such as CRP (Trigunaite et al., 2015). Furthermore, despite no effect on SBP or DBP, our analyses suggest 0.1 mol/L higher free testosterone is associated with increased risk of hypertension (OR = 1.17; 95% CI = 1.08 to 1.27; p=1.05×10−4). Given the multifactorial nature of this disease, the apparent discrepancy between blood pressure and hypertension may be explained by an effect on other risk factors that develop into hypertension. Moreover, both human and animal studies suggest a role of testosterone on hypertension. A randomized controlled trial found testosterone administration increased levels of NT-proBNP, and studies of both transgender men and anabolic steroid users have found testosterone increased arterial stiffness and blood pressure (Bachmann et al., 2019; Hartgens and Kuipers, 2004; Velho et al., 2017). Meanwhile, animal models have shown testosterone may aggravate hypertension and exacerbate increased production of reactive oxygen species specifically in hypertensive but not normotensive rat vascular endothelial tissue (Chignalia et al., 2012; Reckelhoff et al., 1998). Testosterone is widely-believed to have anti-inflammatory and osteogenic effects, but our analyses showed an association with higher risk of spinal stenosis (OR = 2.03; 95% CI = 1.51 to 2.75; p=3.82×10−6). However, the literature shows some evidence that higher testosterone is associated with greater loss of cartilage in healthy older males, and evidence from mouse models suggest testosterone has a sex-specific role in worsening osteoarthritis, a common risk factor for spinal stenosis (Hanna, 2005; Hl et al., 2007).

In comparison to previous MR studies, our results broaden the scope of the existing literature by comprehensively assessing the effects of testosterone on 461 health outcomes including hard endpoints and intermediate biomarkers. Moreover, a key strength of this study was the stringent attempt to control for pleiotropic effects of SHBG on free testosterone by conservatively removing any genetic variants in the GRS that were associated with SHBG (p<0.05). The apparent difference between protective effects of testosterone observed in a previous MR analysis of testosterone and lack of protective effect in our study might be a result of less stringent control for pleiotropic effects of SHBG in the previous study. Given studies have identified associations between SHBG and risk of T2D independent of testosterone and a direct role of SHBG in mediating signalling on target cells, insufficient controls for SHBG may lead to residual pleiotropic effects (Lakshman et al., 2010; Rosner et al., 2010; Vikan et al., 2010). Other reasons may include genetic variants explaining less variation in testosterone levels in our study, fewer cases of T2D leading to inadequate statistical power to detect weaker effects in our study, or other differences between the populations of the UK Biobank in our study and DIAGRAM consortium used by Ruth et al., 2020.

There are several limitations of this study. First, an assumption of the MR analysis is that the effect of the genetic variant on the outcome occurs only through free testosterone levels, such that there are no pleiotropic effects through other proteins or mechanisms (Davies et al., 2018). This concern was minimized by the use of multiple genetic variants, which limited the likelihood of a common alternative pathway confounding our observation. Moreover, we performed several sensitivity analyses and excluded genetic variants associated with SHBG levels, which is a potential source of pleiotropy through its effects on other hormones. Although a stringent p-value threshold was selected for genetic variants, the winner’s curse phenomenon may still bias genetic effect sizes due to the same sample being used to select genetic variants and estimate effect sizes on testosterone. Additionally, one-sample MR may be susceptible to bias towards the confounded estimate if the genetic variants are ‘weak instruments’, which can occur if the genetic variants don’t explain enough of the variance in free testosterone levels (Davies et al., 2018). To address this concern, we confirmed the selected genetic variants were strong instruments using a common threshold in MR literature (F-statistic >10) (Davies et al., 2018). Next, the UK Biobank is generally healthier and higher socioeconomic status than the general population, so there are insufficient cases to detect effects on certain rarer outcomes, such as Alzheimer’s disease, and inadequate power to identify weaker effects of free testosterone on common outcomes. Relatedly, an inherent limitation for outcomes ascertained using linked electronic medical records is a lack of adjudication and consistent application of codes in clinical practice. In the UK Biobank, CFT levels were below the reference ranges for young healthy individuals, which may be attributable to the older age of the cohort and inherent inaccuracy of immunoassays at lower levels of total testosterone. Total testosterone levels are similarly low relative to reference ranges and comparable to previous studies in the UK Biobank (Peila et al., 2020; Petermann-Rocha et al., 2020). Additional sources of variability introduced into the total testosterone measurements include differences in fasting times, diets, and time of day at which blood was drawn from participants. Nevertheless, genetic variants associated with testosterone consistently replicated known effects of testosterone on established outcomes, such as body fat, body fat-free mass, and hematocrit (Table 1). Furthermore, although the free hormone hypothesis is still debated by experts, we found largely consistent effects on outcomes using genetically-predicted free testosterone and total testosterone (Handelsman, 2017). The only significant outcomes from MR analyses with free testosterone that showed no significant effect with total testosterone across all MR methods were HDL (p=0.55) and apolipoprotein A (p=0.45). Finally, these results represent lifelong effects of endogenous free testosterone and may not necessarily reflect effects of exogenous testosterone treatment, which can vary in duration, age of initiation, and dosage.

Taken altogether, the decision to initiate long-term testosterone use warrants careful consideration of benefits and risk. Beneficial effects on body composition, sexual function, hematocrit, and BMD should be weighed against detrimental effects on androgenic alopecia, prostate cancer, hypertension and spinal stenosis, and no detectable beneficial effects on other major clinical endpoints. Ultimately, well-designed and appropriately powered RCTs, such as the ongoing TRAVERSE trials (clinicaltrials.gov, NCT03518034), are necessary to conclusively address questions of safety and effectiveness of testosterone treatment. However, as demonstrated in this study, genetically-informed analyses can be powerful tools to aid health professionals in prioritizing allocation of limited resources towards investigating the most pressing questions.

Materials and methods

Study population - UK Biobank

Request a detailed protocol

The UK Biobank is a large-scale longitudinal cohort study that recruited over 500,000 people between the ages of 37–73 across the United Kingdom from 2006 to 2010 (Sudlow et al., 2015) (RRID:SCR_012815). UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee (REC reference: 11/NW/0382). This research was conducted using the UK Biobank under Application Number 15255. For this study, UK Biobank participants were included if white British ancestry, and no self-reported androgen medication at recruitment based on field ID 20003.

Measurement of testosterone and sex hormone-binding globulin in UK Biobank

Request a detailed protocol

In the UK Biobank, total testosterone and sex hormone-binding globulin (SHBG) were measured on a Beckman Coulter Unicel DXI 800 using a one-step competitive analysis and two-step sandwich immunoassay, respectively. Analytical range for the immunoassays of total testosterone and SHBG were 0.35 to 55.52 and 0.33 to (226-242) nmol/L, respectively. For total testosterone, within-laboratory CV for high, medium, and low concentration quality control samples were 4.15, 3.66, and 8.34%. For SHBG, within-laboratory CV for high, medium, and low concentration quality control samples were 5.22, 5.25, and 5.67%. For each blood sample drawn at recruitment, testosterone, SHBG, and albumin were each measured only once. Testosterone and SHBG measurements were flagged if they fell outside the manufacturer’s observed reportable range, or samples reported high levels of bilirubin, hemoglobin or lipids/turbidity that might interfere with the assay. Testosterone measurements were flagged if levels of total protein (<55 or>85 g/L) or triglycerides (>20 mmol/L) could interfere with the assay measurements. To monitor assay consistency, all samples were run with internal quality control samples between batches and operations used external quality assurance schemes against the ISO 17025:2005 standard.

Genome-wide association study of CFT

Request a detailed protocol

Individual-level genetic data was available for 488,317 participants that consented to blood collection and genotyping. Genotyping was performed with the Applied Biosystems UK Biobank Lung Exome Variant Evaluation (UK BiLEVE) and UK Biobank Axiom arrays (Affymetrix Research Services Laboratory, Santa Clara, California, USA). Description of quality control has been previously described in detail (Bycroft et al., 2017). Genetic variants located in the human leukocyte antigen gene complex were excluded due to extensive pleiotropic effects.

For genome-wide association testing, samples were restricted to a subset of 161,268 males with white British ancestry, no androgen medication (n = 2,137), and no missing values of testosterone, SHBG, or albumin at recruitment. Free testosterone at recruitment was calculated using the Vermeulen equation (Vermeulen et al., 1999). CFT levels were winsorized such that outlying values greater or less than four standard deviations (SD) away from the mean in males were set to 4 SD.

This study was restricted to genetic variants from ‘v3’ release of the UK Biobank data including those present in the Haplotype Reference Consortium and 1000 Genomes panels with imputation imputation quality greater than 0.7, no deviation from Hardy-Weinberg equilibrium (p>1×10−10) and minor allele frequency greater than 1% (McCarthy et al., 2016). To allow for genetic relatedness between participants, linear mixed models in BOLT-LMM were used to test for associations of genetic variants (Loh et al., 2015). The model was adjusted for age, age2, chip type, assessment center, and the first 20 genetic principal components. Genetic variants near the SHBG gene may alter binding affinity for testosterone thereby violating assumptions of the Vermeulen equation, or risk having pleiotropic effects through binding of other sex hormones (Ohlsson et al., 2011). Therefore, any genetic variants associated with CFT reaching genome-wide significance (p≤5×10−8) were excluded if associated with natural log-transformed SHBG levels at a stringent threshold (p<0.05) in the same subset of the UK Biobank (Figure 1—figure supplement 4). To arrive at an independent set of genetic variants, variants associated with CFT but not SHBG were pruned based on linkage disequilibrium (LD) at a threshold of r2 <0.01 using Europeans from 1000 Genomes phase three as reference panel (Abecasis et al., 2012) (RRID:SCR_006828).

Genomic inflation factor (λ) was 1.2 and calculated as the ratio of the median test statistic from the GWAS relative to the expected median test statistic under a null model (Figure 1—figure supplement 5). To distinguish between an inflated λ due to population stratification or polygenic inheritance of the trait, the intercept of an LD score regression line was determined to be 1.03 indicating the observed inflation could be attributed to polygenicity rather than uncontrolled population stratification. LD score regression was performed and intercept was calculated with LDSC software (Bulik-Sullivan et al., 2015) using 1000 Genomes Europeans phase three data as the LD reference panel (Abecasis et al., 2012).

Definition of health-related UK Biobank outcomes

Request a detailed protocol

For MR analyses, 22 health outcomes were selected a priori based on relevance with known or suspected effects of testosterone treatment and categorized based on expected beneficial or adverse effects from RCT data. Outcomes with expected beneficial effects were fractures at any site, heel BMD, body fat percentage, body fat-free percentage, dementia, depression, handgrip strength, and physical activity level measured by wrist-worn accelerometer. Outcomes with potential adverse effects were stroke, androgenic alopecia, benign prostate hyperplasia (BPH), blood pressure, glucose, hematocrit percentage, hemoglobin A1c, heart failure, prostate cancer, MI, type 2 diabetes (T2D), and venous thromboembolism. Depression was coded using a ‘broad’ definition as previously described, which included self-reported depressive symptoms with associated impairment, or having sought help for ‘nerves, anxiety, tensions or depression’ (Howard et al., 2018). Androgenic alopecia was defined based on participants’ responses to the question, ‘Which of the following best describes your hair/balding pattern?’ (field ID 2395). Available options were four pictures of hair patterns (Supplementary file 1Figure 1). Individuals with pattern 3 or four were cases, pattern 1 and 2 were controls, and ‘do not know’ or ‘prefer not to answer’ responses were excluded. Physical activity was assessed using the overall acceleration average from wrist-worn accelerometer devices over the course of approximately 7 days. Following UK Biobank recommendations, individuals were excluded from the analysis based on poorly calibrated data (field ID: 90016) or having worn the device for insufficient time to get a stable measure of physical activity (field ID: 90015) (Doherty et al., 2017). Blood pressure measures were coded as the average of two automated measurements of blood pressure taken a few moments apart by a registered nurse using an Omron 705 IT electronic blood pressure monitor. Body fat percentage and whole body fat-free mass were estimated based on impedance measurements from a Tanita BC418MA body composition analyser. Heel BMD was estimated as a T-score based on quantitative ultrasound index through the calcaneus relative to that expected in someone of the same sex. Handgrip strength was calculated as the average of right and left hands measured using a Jamar J00105 hydraulic hand dynamometer. hemoglobin A1C was measured using high performance liquid chromatography analysis on a Bio-Rad VARIANT II Turbo. Glucose was measured using hexokinase analysis on a Beckman Coulter AU5800. Hematocrit percentage was measured using a Coulter LH750 and calculated as the relative volume of packed erythrocytes to whole blood, computed by the formula: redbloodcellsmeancorpuscularvolume10. Detailed descriptions of all 22 outcomes are shown in Supplementary file 1 – Table 11.

For hypothesis-free GRS analyses, we included 24 blood biomarkers measured at recruitment and 415 diseases derived from linked electronic medical records (Supplementary file 1 - Table 12; Brion et al., 2013; Denny et al., 2013; Wu et al., 2019). Disease outcomes were defined using the previously published ‘PheCode’ scheme to aggregate ICD-10 codes from hospital episodes (field ID 41270), death registry (field ID 40001 and 40002), and cancer registry (field ID 40006) records (Denny et al., 2013; Wu et al., 2019). Given the small number of cases for many disease outcomes, any outcomes with detectable odds ratios less than 0.5 or greater than 2 per 0.1 nmol/L at 80% power were excluded (ncases < 871) based on approximate changes in response to testosterone supplementation (Bhasin et al., 2018b; Brion et al., 2013; Traustadóttir et al., 2018). After these exclusions, there were 415 diseases that remained for subsequent analyses in this study. Furthermore, all blood biomarkers measured by the UK Biobank at recruitment were included except estradiol and rheumatoid factor, which were complicated by majority missing values below the limit of detection of the assay (nbiomarkers = 24). Detailed descriptions of all 439 outcomes (415 diseases and 24 biomarkers) are shown in Supplementary file 1 – Table 12.

Mendelian randomization analysis

Request a detailed protocol

In a subset of unrelated males with White British ancestry, the association of all independent genetic variants associated with CFT were determined for each of the 22 a priori outcomes using additive genetic models in BGENIE v1.2 and adjusted for the same covariates as the model for CFT (Bycroft et al., 2017). For each of the 22 outcomes, one-sample MR analysis was used to combine the effect of each independent genetic variant on CFT with its effect on the outcome using the inverse variance-weighted (IVW) method (Burgess et al., 2016). Effect estimates were reported per 0.1 nmol/L increase in CFT levels based on approximate changes in response to testosterone treatment (Bhasin et al., 2018b). For dichotomous outcomes, odds ratios were approximated as previously described (Adams et al., 2018) by converting linear effect estimates from BGENIE to log-odds scale using:

logOR= k(1-k), where k is the proportion of cases for the given outcome.

Given the polygenic nature of testosterone and potential for pleiotropy, for outcomes with statistically significant effects using the IVW method, standard sensitivity analyses were conducted to correct for pleiotropic effects, such as MR-Egger, MR-RAPS, and MR-PRESSO (Bowden et al., 2015; Verbanck et al., 2018). To investigate and correct for directional pleiotropy on each outcome, we performed Egger regression. For outcomes with y-intercept of the regression line significantly different from 0 (p<0.05), there was evidence of directional pleiotropy and the causal estimate from MR Egger was reported to attempt to control for pleiotropic effects (Bowden et al., 2015). As a sensitivity analysis robust to idiosyncratic pleiotropy and weak instrument bias, MR-RAPS (Robust Adjusted Profile Score) was conducted using overdispersion and Tukey’s loss function (Zhao et al., 2018). To detect and correct for potential bias from invalid variants with pleiotropic effects, we performed the MR-PRESSO (Mendelian Randomization Pleiotropy RESidual Sum and Outlier) test with 10,000 simulations (Verbanck et al., 2018). The global test p-value evaluated whether there was any overall horizontal pleiotropy among all genetic variants. For outcomes with significant p-values (p<0.05), outlying genetic variants with predicted pleiotropic effects were removed and MR analysis repeated to correct for horizontal pleiotropy. The distortion test evaluated whether removal of the pleiotropic variants resulted in a significantly different causal estimate (p<0.05). Leave-one-out analysis was performed such that the IVW MR analysis was repeated after each genetic variant was excluded to identify effects on an outcome that are driven by a single outlying genetic variant. Furthermore, the set of genetic variants used in MR analysis were assessed for ‘weak instrument bias’, which can result in biased estimates if genetic variants don’t explain enough variance in exposure (e.g., CFT) levels (Pierce et al., 2011). Lastly, as a sensitivity analysis, all MR and GRS analyses were repeated using genetic variants associated with total testosterone. Finally, for significant outcomes, we compared estimated effect sizes from this MR study with reported effect sizes from random controlled trials of testosterone therapy, where possible, in Figure 3 (Cui et al., 2014; Fernández-Balsells et al., 2010; Ng Tang Fui et al., 2016; Zhang et al., 2020).

In consideration of ‘weak instrument bias’, the F-statistic was 66 for the genetic variants associated with CFT, which was considered a strong instrument based on the recommended threshold of greater than 10 (Davies et al., 2018). MR-PRESSO was performed using the MR-PRESSO package and all other MR analyses were implemented using the TwoSampleMR package (Hemani et al., 2018; Verbanck et al., 2018) (RRID:SCR_019010).

Genetic risk score analysis

Request a detailed protocol

A genetically-predicted value of CFT was determined for each individual by constructing weighted GRS in the unrelated White British subset of UK Biobank males (n = 157,252). Weighted GRS were calculated by multiplying the effect of each CFT-associated genetic variant by the number of effect-corresponding alleles and summing this value for each individual. The GRS was tested for association with outcomes using logistic or linear regression models for case-control or quantitative outcomes, respectively, and adjusted for the same covariates as the GWAS for CFT. Effect estimates were reported per 0.1 nmol/L increase in CFT levels based on approximate changes in response to testosterone treatment (Bhasin et al., 2018b). As sensitivity analyses, we repeated GRS analyses after excluding males that self-reported taking blood pressure (n = 38,676) or cholesterol medication (n = 35,737) at recruitment based on field ID 6177.

Genetic determinants and effects of total testosterone in males

Request a detailed protocol

As a set of sensitivity checks, we repeated all GWAS, MR, and GRS analyses using total testosterone. In the White British subset of the UK Biobank, there were 175,421 males with total testosterone measured with an average 11.9 nmol/L (Figure 1—figure supplement 6). In this population, a genome-wide association study was conducted for total testosterone as described herein for CFT. After removing genetic variants associated with natural-log-transformed SHBG and LD pruning for independent SNPs (r2 <0.01), there were 52 independent genetic variants associated (p<5×10−8) with total testosterone in males from the UK Biobank (Supplementary file 1 – Table 8).

All statistical analyses were performed under R version 3.6.0, unless otherwise specified (RRID:SCR_001905). A two-sided p-value less than 5 × 10−8 for GWAS, 2.27 × 10−3 (0.05/22 outcomes) for a priori MR analyses, and 1.14 × 10−4 (0.05/439 outcomes) for hypothesis-free GRS analyses was considered statistically significant.

References

  1. 1
  2. 2
  3. 3
  4. 4
    Cancer Facts & Figures 2019
    1. American Cancer Society
    (2019)
    Cancer Facts & Figures 2019.
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61

Decision letter

  1. Dolores Shoback
    Reviewing Editor; University of California, San Francisco, United States
  2. Eduardo Franco
    Senior Editor; McGill University, Canada
  3. Shalender Bhasin
    Reviewer
  4. Mathis Grossman
    Reviewer
  5. Qingyuan Zhao
    Reviewer; University of Cambridge, United Kingdom

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

We appreciate the key role of Mendelian randomization analyses in assessing the wide spectrum of testosterone related outcomes in men that you explore in this paper. The approach well complements rigorous randomized trials and provides important information that such large data sets can inform on. Given the increasing use of testosterone in older men in many countries, these data are important in both confirming and extending what trials have shown and in highlighting the lack of effects of lifelong testosterone levels on cardiovascular, cognitive and metabolic outcomes.

Decision letter after peer review:

Thank you for submitting your article "Effects of lifelong testosterone exposure on health and disease: a Mendelian randomization study" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Eduardo Franco as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, when editors judge that a submitted work as a whole belongs in eLife but that some conclusions require a modest amount of additional new data, as they do with your paper, we are asking that the manuscript be revised to either limit claims to those supported by data in hand, or to explicitly state that the relevant conclusions require additional supporting data.

Our expectation is that the authors will eventually carry out the additional experiments and report on how they affect the relevant conclusions either in a preprint on bioRxiv or medRxiv, or if appropriate, as a Research Advance in eLife, either of which would be linked to the original paper.

This paper provides a large amount of new data on the role of testosterone levels in metabolic outcomes. It provides support for insights gained from trials as well as new insights in especially controversial areas.

Please direct your revisions to the matters discussed by the reviewers as below and consider the additional analyses that are requested as the editors believe addressing these issues will strengthen your paper.

Reviewer #1:

The long-term benefits and risks of testosterone are incompletely understood. Recent randomized trials, especially the TTrials, have taught us a great deal about the efficacy of testosterone replacement therapy, the long-term risks of MACE, prostate cancer, and diabetes remain unclear. Although randomized controlled trials remain the gold standard, MR studies, such as these, can provide useful complementary data.

In this manuscript, the authors performed Mendelian randomization analyses to infer phenome-wide effects of free testosterone on large number of outcomes in male participants of the UK Biobank study. The analyses yielded several very interesting findings, including the positive association of genetically-determined free testosterone levels with increased bone mineral density, and decreased body fat; decreased HDL, and increased risks of prostate cancer, androgenic alopecia, hypertension; and some other phenotypes. The analyses included individual-level data from a large sample from the UK Biobank. Some of the outcomes that were analyzed are especially important: diabetes, prostate cancer, venous thrombosis, myocardial infarction, and stroke because testosterone's effects on these outcomes have remain unclear from the RCTs published so far. The authors performed several sensitivity analyses to ensure that the effects were not driven by any single variant. Additional, Egger regression and MR-PRESSO were performed to detect and correct for potential pleiotropy. For prostate cancer, a two-sample MR analysis was performed using data from the PRACTICAL Consortium and the UK Biobank.

A small number of MR studies have previously investigated the effects of total testosterone on lipids, bone mineral density, and CVD risk; many of them have included relatively smaller samples. This study represents a comprehensive effort to estimate the relation of genetic loci associated with calculated free testosterone levels with a large number of outcomes.

Another recently published MR study by Ruth et al. in Nature Medicine (cited in this paper) reported that genetically determined testosterone levels were associated with sexually dimorphic effects on diabetes risk (Ruth et al., 2020). Although some of the conditions and phenotypes analyzed by Ruth et al. are similar to those reported in this manuscript, the current manuscript includes additional analyses that were not evaluated by Ruth et al. Furthermore, some of the findings differ from those reported by Ruth et al. Therefore, the manuscript includes some important novel information beyond that which was reported by Ruth et al.

A very large body of data are presented, representing a huge amount of work. Thus, the information presented in this manuscript represents an important addition to the extant literature on this topic.

Specific comments:

1) The ascertainment of outcomes and diagnoses using electronic medical records has some inherent problems. These problems are greater for ascertaining some outcomes such as dementia, Alzheimer's dementia, depression, BPH, etc because of the lack of rigorously defined pre-specified diagnostic criteria in clinical practice, lack of prospective adjudication, and nonuniform application of diagnostic codes by clinicians in practice. Many types of lower urinary tract symptoms get coded as BPH. These limitations should be acknowledged in the Discussion.

2) A substantial fraction of adult men (1.5 to 1.7% in the US) are being treated with testosterone. Were testosterone-treated men excluded from the analyses? This is important because testosterone treatment. could confound the analyses. Also, men with genetically-determined low testosterone level are at increased risk of getting treated with testosterone. Therefore, it would be important to know what fraction of people in the UK Biobank data were treated with testosterone and whether they were excluded from the analyses.

3) Total testosterone levels were measured using a platform-based immunoassay that are well known to lack accuracy, especially in the low range. More importantly, free testosterone levels were calculated using an equation that is based on a linear model of testosterone binding to SHBG, which has been shown to be erroneous. The measurement problems are perhaps reflected in the fact that the average calculated free testosterone level in the sample (0.21 nmol/L) is substantially lower than the mean free testosterone levels determined by equilibrium dialysis, the reference method. The reviewer recognizes that the authors had no choice but to use the data that were available in the UK Biobank. But acknowledging this limitation in the Discussion would be important.

4) Also, the basic characteristics (LLOQ, precision and accuracy, specificity, analytical range) of the assays should be provided.

5) How much of the variation in calculated free T levels was explained by the genetic loci that were associated with CFT levels? The Manhattan plot in Figure 1—figure supplement 3 shows the distribution of p-values from genome-wide association study of calculated free testosterone after exclusion of SHBG-associated variants based on chromosomal location. This figure contains really important data. Although the GWAS of total and free T levels have been published, it would be very useful to include the information on these loci and whether any new loci were discovered.

6) Figure 1—figure supplement 1. The units for SHBG are in log units which would be difficult for the readers to comprehend; changing the units to nmol/L would make it easier to get a sense of the distribution of values.

7) The authors found significant associations with some really clinically important outcomes, such as prostate cancer, prostate cancer, androgenic alopecia, and hypertension. Some discussion of the effect size and meaningfulness of the observed effect would be valuable in putting these observations in clinical context.

8) Some of the findings of the analyses, especially on diabetes and prostate cancer risk, differ from those reported by Ruth et al. The authors should comment on why the findings differ in the two sets of analyses that used the same body of UK Biobank data.

9) It is stated that the methods for outcome ascertainment are included in a table in a supplementary file. Criteria for some of the outcomes are provided (e.g., alopecia, depression); I may have missed it, but I did not find the criteria for outcome ascertainment that were used in the definition of a number of other outcomes (e.g., dementia, diabetes, BPH, prostate cancer, etc.).

Reviewer #2:

This is a very interesting and valuable study using a mendelian randomization approach to infer (with the appropriate caveats) causal effects of genetically determined serum testosterone on a variety of phenotypes considered to be androgen sensitive. Strengths include the large cohort (albeit limited to white UK men), and the careful analyses conducted. Some outcomes are expected, others perhaps less so, and may represent chance finding.

Comments to the authors:

1) Given that one of the view aspects that all testosterone guidelines agree on is that total testosterone is the principal measurement to confirm a clinical diagnosis of androgen deficiency, it would be interesting to present results according to total testosterone-or at least defend decision to not do so; while the “free hormone hypothesis” is supported by some studies, not all experts agree on this, as the evidence is not definitive.

2) It is not clear whether testosterone (and SHGB) were measured only once, and if so whether they were drawn in the morning in the fasted state. This is important given the diurnal variability of testosterone measurements, effects of food intake and day to day variability. Moreover, immunoassay for testosterone can be imprecise, especially at the lower range. All these factors may have limited the precision of the GWAS.

3) Interestingly, the average CFT was 0.21 nmol/L in the population (Results first paragraph) is, in the context of sexual symptoms, below the cutoff for diagnosing "Late onset hypogonadism" DOI: 10.1056/NEJMoa0911101. It is not clear whether serum testosterone was measured across the population or only in men in whom it was clinically indicated; either way the low average is surprising and requires further explanation.

4) Discussion paragraph one: "the predicted 1.5-fold increase...observed after initiation of testosterone supplementations", please clarify where these data are from.

5) Discussion paragraph five: the dichotomy between “lifestyle” and “clinical” perspective is a little forced-please rephrase. The clinical approach to testosterone treatment involves weighing benefits (e.g. body composition that may be metabolically favourable or on BMD that may (or may not) reduce fracture risk) against risks. As a matter of course while testosterone replacement in men with organic hypogonadism is undisputed, the role of testosterone treatment for symptomatic men with age-related decline in testosterone remains uncertain. I suggest to avoid the term “supplementation” as it infers correcting a clear hormone deficiency state and instead use the more neutral term “treatment” which acknowledges the possibility that treatment may be pharmacological instead of replacement.

6) Abstract: "MR suggests lifestyle benefits" this is not clear please rephrase.

Reviewer #3:

This manuscript applies Mendelian randomization to investigate potential causal effects of testosterone on biomarkers and health outcomes. Although the statistical methods used by the authors are generally appropriate, there is still some room for improvement (see the comments below). I hope the authors can address them in a revision.

1) I think the Mendelian randomization results will become a lot stronger if the authors can compare the estimated effects on the 22 a priori outcomes with the existing results from RCT (for example, using a scatterplot of MR effects versus RCT effects, with standard error bars in both directions). This will not only reveal whether there is any systematic bias of the MR design/method for testosterone but also how much the "lifelong" effect estimated by MR is larger than short term effect estimated by RCT.

2) Given all the methodological developments for MR, I am surprised to see that the authors chose to report the results of inverse variance weighting (IVW) estimator instead of the other more robust methods. IVW is only valid in the ideal theoretical setting, which is rarely the case for empirical applications. For example, in Figure 3—figure supplement 3 it is very clear that there are a few negative outliers and the IVW slope (or MR-Egger slope) seems to underestimate the positive effect suggested by the majority of the SNPs. This issue can be addressed by MR-PRESSO, but an even better alternative is MR-RAPS that handles outliers, overdispersion, and the many weak instrument asymptotic variance. Related software resources and discussion can be found in the links below:

https://github.com/qingyuanzhao/mr.raps

https://doi.org/10.1093/ije/dyz142

3) A statistical issue unaddressed by the authors is the winner's curse in selecting the genetic instruments. This happens if the same GWAS is used to both select instruments and make statistical inference. In general, the winner's curse biases the point estimator towards 0 in two-sample MR, but that bias can be more complicated when compounded with other issues like outliers. The winner's curse can be eliminated by using a three-sample MR design, in which a separate dataset is used to select instruments; see the paper in the second link above. If this is not possible, the best alternative I know is to use a very strict significance threshold for instrument selection (which the authors have already done) and acknowledge the potential bias from winner's curse in the discussion.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Effects of lifelong testosterone exposure on health and disease using Mendelian randomization" for further consideration by eLife. Your revised article has been evaluated by the editors after consultation with the original reviewers.

The manuscript has been improved but there are two remaining issues that need to be addressed before full acceptance is made, as outlined below:

One of the reviewers, who is enthusiastic about this paper, strongly requests that you make two additional analyses and/or explanations. That reviewer, an expert in the testosterone field, states:

1) One issue that I am still puzzled about is the difference in the findings from the data reported by Ruth et al. with respect to the association between genetically determined free testosterone and diabetes risk. The authors list "fewer cases of T2D leading to inadequate statistical power or other differences in the populations used for T2D analysis." If the same database of UK Biobank was used in both the analyses, why would there be a difference in the number of T2D cases or in the study population?

2) I also continue to be concerned about the calculated free T concentrations that are substantially lower than those described previously in community-dwelling men. I recognize that these are the numbers that the UK Biobank has provided, but it would be worth re-checking the calculations to make sure there is no inadvertent systematic error in computation.

Comment: The authors have been very diligent in providing a very large body of data in the supplementary tables and figures. I do not recommend inclusion of any additional data beyond what is included in the current manuscript.

https://doi.org/10.7554/eLife.58914.sa1

Author response

Reviewer #1:

[…]

Specific comments

1) The ascertainment of outcomes and diagnoses using electronic medical records has some inherent problems. These problems are greater for ascertaining some outcomes such as dementia, Alzheimer's dementia, depression, BPH, etc because of the lack of rigorously defined pre-specified diagnostic criteria in clinical practice, lack of prospective adjudication, and nonuniform application of diagnostic codes by clinicians in practice. Many types of lower urinary tract symptoms get coded as BPH. These limitations should be acknowledged in the Discussion.

We agree with the reviewer’s comments regarding the limitations of using data from electronic medical records. Unfortunately, these issues are unavoidable with the data at our disposal, and as such, we have explicitly highlighted the potential issues in the Discussion, “Relatedly, an inherent limitation for outcomes ascertained using linked electronic medical records is a lack of adjudication and consistent application of codes in clinical practice.”.

2) A substantial fraction of adult men (1.5 to 1.7% in the US) are being treated with testosterone. Were testosterone-treated men excluded from the analyses? This is important because testosterone treatment. could confound the analyses. Also, men with genetically-determined low testosterone level are at increased risk of getting treated with testosterone. Therefore, it would be important to know what fraction of people in the UK Biobank data were treated with testosterone and whether they were excluded from the analyses.

This is an important consideration for a cohort study such as the UK Biobank. For our analyses, we did exclude participants that self-reported taking androgen medication based on field ID 20003. This exclusion was previously described for the genome-wide association study, but we have now clarified the exclusion applied to the entire study by moving the sentence earlier in the revised Materials and methods. The fraction of participants excluded may have been less than 1.5-1.7% due to differences in rates of testosterone prescription between the United Kingdom and United States of America (Handelsman DJ, 2013).

3) Total testosterone levels were measured using a platform-based immunoassay that are well known to lack accuracy, especially in the low range. More importantly, free testosterone levels were calculated using an equation that is based on a linear model of testosterone binding to SHBG, which has been shown to be erroneous. The measurement problems are perhaps reflected in the fact that the average calculated free testosterone level in the sample (0.21 nmol/L) is substantially lower than the mean free testosterone levels determined by equilibrium dialysis, the reference method. The reviewer recognizes that the authors had no choice but to use the data that were available in the UK Biobank. But acknowledging this limitation in the Discussion would be important.

The reviewer brings up an important point regarding the accuracy of immunoassays, and its limitations relative to the gold-standard of equilibrium dialysis. Since this is inherent to the UK Biobank, we have elaborated in the Discussion, “In the UK Biobank, calculated free testosterone levels were below the reference ranges for young healthy individuals, which may be attributable to the older age of the cohort and inherent inaccuracy of immunoassays at lower levels of total testosterone.”.

4) Also, the basic characteristics (LLOQ, precision and accuracy, specificity, analytical range) of the assays should be provided.

In line with the previous comment, we have included the analytical characteristics of the assay as provided by the UK Biobank in the Materials and methods, “Analytical range for the immunoassays of total testosterone and SHBG were 0.35 to 55.52 and 0.33 to (226-242) nmol/L, respectively. For total testosterone, within-laboratory CV for high, medium, and low concentration quality control samples were 4.15, 3.66, and 8.34%. For SHBG, within-laboratory CV for high, medium, and low concentration quality control samples were 5.22, 5.25, and 5.67%.”.

5) How much of the variation in calculated free T levels was explained by the genetic loci that were associated with CFT levels? The Manhattan plot in Figure 3—figure supplement 3 shows the distribution of p-values from genome-wide association study of calculated free testosterone after exclusion of SHBG-associated variants based on chromosomal location. This figure contains really important data. Although the GWAS of total and free T levels have been published, it would be very useful to include the information on these loci and whether any new loci were discovered.

As per the reviewer’s suggestions, we have clarified the amount of variation explained by the loci associated with CFT at genome-wide significant loci, “Overall chip-based heritability of CFT was estimated at 15 % (95%CI = 14 to 16), while these 93 independent genetic variants associated with CFT explained 3.7% of the total variance of CFT levels in males from the UK Biobank.” . Moreover, the nearest gene(s) to independent GWS loci pictured in Figure 1—figure supplement 3 are annotated in Supplementary file 1—table 2 alongside additional details regarding each genetic variant.

6) Figure 1—figure supplement 1. The units for SHBG are in log units which would be difficult for the readers to comprehend; changing the units to nmol/L would make it easier to get a sense of the distribution of values.

Although the log-transformed values reflect the units and distribution used in our subsequent analyses, we understand that units in nmol/L are more interpretable for readers. As a result, we have added Figure 1—figure supplement 3A to reflect raw values as nmol/L while Figure 1—figure supplement 3B reflects log-transformed values.

7) The authors found significant associations with some really clinically important outcomes, such as prostate cancer, prostate cancer, androgenic alopecia, and hypertension. Some discussion of the effect size and meaningfulness of the observed effect would be valuable in putting these observations in clinical context.

We feel this is an important comment as it allowed us to reflect on better communicating our findings. For all outcomes, we have represented the effect sizes in terms of 0.1 nmol/L increase of free testosterone, which reflects approximate changes observed during initiation of testosterone supplementation. We have emphasized this decision in the Discussion, “All effects are reported in terms of 0.1 nmol/L of CFT to approximate expected effect sizes after initiation of testosterone treatment.” and Figure 3 legend.

8) Some of the findings of the analyses, especially on diabetes and prostate cancer risk, differ from those reported by Ruth et al. The authors should comment on why the findings differ in the two sets of analyses that used the same body of UK Biobank data.

We briefly touched on the differences between our analyses in the Introduction. Ruth et al. similarly found a risk-conferring effect of testosterone on prostate cancer (OR = 1.23 per 1 SD bioavailable testosterone; 95% CI = 1.13 to 1.33). However, we have revised the Discussion to further elaborate on methodological differences that might explain the divergent findings regarding diabetes, “The apparent difference between protective effects of testosterone observed in a previous MR analysis of testosterone and lack of protective effect in our study might be a result of less stringent control for pleiotropic effects of SHBG in the previous study. Given studies have identified associations between SHBG and risk of T2D independent of testosterone and a direct role of SHBG in mediating signaling on target cells, insufficient control for SHBG may lead to residual pleiotropic effects (Vikan et al., 2010) (Lakshman, Bhasin and Araujo, 2010) (Rosner et al., 2010). Other reasons may include genetic variants explaining less variation in testosterone levels in our study, fewer cases of T2D leading to inadequate statistical power to detect weaker effects in our study, or other differences in the populations used for T2D analysis.”.

9) It is stated that the methods for outcome ascertainment are included in a table in a supplementary file. Criteria for some of the outcomes are provided (e.g., alopecia, depression); I may have missed it, but I did not find the criteria for outcome ascertainment that were used in the definition of a number of other outcomes (e.g., dementia, diabetes, BPH, prostate cancer, etc.)

We apologize for any confusion. The definitions for 22 a priori outcomes were listed in Supplementary file 1—table 11. These included the field IDs and ICD-10 codes, if applicable, for dementia (row 7), BPH (row 16), prostate cancer (row 22), and type 2 diabetes (row 25). The definitions for 439 phenome-wide outcomes were listed in Supplementary file 1—table 12. We further explain this distinction, “Detailed descriptions and selection criteria are available for all a priori outcomes in Supplementary file 1—table 11, and phenome-wide outcomes and biomarkers in Supplementary file 1—table 12.”, and now provide more details on the definition of outcomes in the revised Materials and methods.

Reviewer #2:

[…]

Comments to the authors:

1) Given that one of the view aspects that all testosterone guidelines agree on is that total testosterone is the principal measurement to confirm a clinical diagnosis of androgen deficiency, it would be interesting to present results according to total testosterone-or at least defend decision to not do so; while the “free hormone hypothesis” is supported by some studies, not all experts agree on this, as the evidence is not definitive.

Although we agree the “free hormone hypothesis” is not definitive, we felt this dataset presented a unique opportunity to explore the effects of free testosterone specifically. Importantly, we repeated our analyses using genetically-predicted total testosterone (Supplementary file 1—tables 9 and 10) and found largely consistent effects with the significant results using genetically-predicted free testosterone. We have added a comment to this effect in the Discussion, “Furthermore, although the free hormone hypothesis is still debated by experts, we found largely consistent effects on outcomes using genetically-predicted free testosterone and total testosterone. […] Indeed, one of the pleiotropic outliers identified by MR-PRESSO was rs9986829, which is located near DGKB – a gene associated with glucose homeostasis and type 2 diabetes in multiple cohorts (10.1371/journal.pone.0015542) (10.1038/ng.520).

2) It is not clear whether testosterone (and SHGB) were measured only once, and if so whether they were drawn in the morning in the fasted state. This is important given the diurnal variability of testosterone measurements, effects of food intake and day to day variability. Moreover, immunoassay for testosterone can be imprecise, especially at the lower range. All these factors may have limited the precision of the GWAS.

We thank the reviewer for bringing this to our attention. We have clarified that testosterone was measured only once in the Materials and methods and acknowledge the limitations associated with this source of variability in the Discussion, “Additional sources of variability introduced into the total testosterone measurements include differences in fasting times, diets, and time of day at which blood was drawn from participants. Nevertheless, genetic variants associated with testosterone consistently replicated known effects of testosterone on established outcomes, such as body fat, body fat-free mass, and haematocrit.”. Likewise, we have acknowledged the inherent limitations of the immunoassay in the revised Discussion.

3) Interestingly, the average CFT was 0.21 nmol/L in the population (Results first paragraph) is, in the context of sexual symptoms, below the cutoff for diagnosing "Late onset hypogonadism" DOI: 10.1056/NEJMoa0911101. It is not clear whether serum testosterone was measured across the population or only in men in whom it was clinically indicated; either way the low average is surprising and requires further explanation.

Testosterone was measured across the population, but the older average age (57 years) in the UK Biobank may explain the lower calculated free testosterone levels relative to reference ranges from healthy adult populations. Indeed, mean total testosterone levels were similarly low relative to reference ranges, and previous studies utilizing the UK Biobank have reported comparable levels for both free and total testosterone (Peila, Arthur and Rohan, 2020) (Petermann-Rocha et al., 2020). This is commented on in the Discussion, “In the UK Biobank, calculated free testosterone levels were below the reference ranges for young healthy individuals, which may be attributable to the older age of the cohort and inherent inaccuracy of immunoassays at lower levels of total testosterone. Total testosterone levels are similarly low relative to reference ranges and comparable to previous studies in the UK Biobank.”.

4) Discussion paragraph one: "the predicted 1.5-fold increase...observed after initiation of testosterone supplementations", please clarify where these data are from.

5) Discussion paragraph five: the dichotomy between “lifestyle” and “clinical” perspective is a little forced-please rephrase. The clinical approach to testosterone treatment involves weighing benefits (e.g. body composition that may be metabolically favourable or on BMD that may (or may not) reduce fracture risk) against risks. As a matter of course while testosterone replacement in men with organic hypogonadism is undisputed, the role of testosterone treatment for symptomatic men with age-related decline in testosterone remains uncertain. I suggest to avoid the term “supplementation” as it infers correcting a clear hormone deficiency state and instead use the more neutral term “treatment” which acknowledges the possibility that treatment may be pharmacological instead of replacement.

The intent behind the “lifestyle” and “clinical” distinction was to classify and improve interpretability of our findings, but we understand the reviewer’s concerns. Consequently, we replaced all instances of “testosterone supplementation” with “testosterone treatment” and reworded our statement to include all benefits and adverse effects as a whole, “Beneficial effects on body composition, sexual function, hematocrit, and BMD should be weighed against detrimental effects on androgenic alopecia, prostate cancer, hypertension and spinal stenosis, and no detectable beneficial effects on other major clinical endpoints.”.

6) Abstract: "MR suggests lifestyle benefits" this is not clear please rephrase.

In the same manner as the previous comment, we have removed any reference to “lifestyle” from the manuscript.

Reviewer #3:

[…]

1) I think the Mendelian randomization results will become a lot stronger if the authors can compare the estimated effects on the 22 a priori outcomes with the existing results from RCT (for example, using a scatterplot of MR effects versus RCT effects, with standard error bars in both directions). This will not only reveal whether there is any systematic bias of the MR design/method for testosterone but also how much the "lifelong" effect estimated by MR is larger than short term effect estimated by RCT.

We thank the reviewer for this insight. We weren’t able to find RCTs of testosterone therapy for all significant outcomes in the MR analyses, but we have included a plot comparing RCT versus MR effect estimates for available outcomes in Figure 3. Data sources are referenced in the Materials and methods, “Finally, for significant outcomes, we compared estimated effect sizes from this MR study with reported effect sizes from random controlled trials of testosterone therapy, where possible…”, and we have expanded on results in the Discussion, “Most of the estimates from MR analyses were comparable with effect sizes from RCTs (Figure 3). There was only significant heterogeneity between effects on BMD for MR and RCT, but it is difficult to make direct comparisons due to variable change in testosterone levels after administration of testosterone in each RCT, different methods and anatomical sites of BMD estimation, and differences between short-term effects in RCTs relative to lifelong effects in MR.”.

2) Given all the methodological developments for MR, I am surprised to see that the authors chose to report the results of inverse variance weighting (IVW) estimator instead of the other more robust methods. IVW is only valid in the ideal theoretical setting, which is rarely the case for empirical applications. For example, in Figure 3—figure supplement 3 it is very clear that there are a few negative outliers and the IVW slope (or MR-Egger slope) seems to underestimate the positive effect suggested by the majority of the SNPs. This issue can be addressed by MR-PRESSO, but an even better alternative is MR-RAPS that handles outliers, overdispersion, and the many weak instrument asymptotic variance. Related software resources and discussion can be found in the links below:

https://github.com/qingyuanzhao/mr.raps

https://doi.org/10.1093/ije/dyz142

We thank the reviewer for bringing this to our attention. As a sensitivity analysis, MR-RAPS has been conducted for all 22 a priori outcomes as described in the revised Materials and methods “As a sensitivity analysis robust to idiosyncratic pleiotropy and weak instrument bias, MR-RAPS (Robust Adjusted Profile Score) was conducted using overdispersion and Tukey’s loss function.”, Results, “Results using MR-RAPS were consistent with IVW regression method for all significant outcomes.”, and Supplementary file 1—table 4 and table 10.

3) A statistical issue unaddressed by the authors is the winner's curse in selecting the genetic instruments. This happens if the same GWAS is used to both select instruments and make statistical inference. In general, the winner's curse biases the point estimator towards 0 in two-sample MR, but that bias can be more complicated when compounded with other issues like outliers. The winner's curse can be eliminated by using a three-sample MR design, in which a separate dataset is used to select instruments; see the paper in the second link above. If this is not possible, the best alternative I know is to use a very strict significance threshold for instrument selection (which the authors have already done) and acknowledge the potential bias from winner's curse in the discussion.

We agree with the reviewer’s remarks regarding bias due to the winner’s curse phenomenon. Due to limited data availability, employing a three-sample MR design would be challenging as we don’t have access to a sufficiently well-powered third study to select instruments. As a result, we have acknowledged this source of bias in the Discussion, “Although a stringent p-value threshold was selected for genetic variants, the winner’s curse phenomenon may still bias effect sizes due to the same sample being used to select genetic variants and estimate effect sizes on testosterone.”.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

One of the reviewers, who is enthusiastic about this paper, strongly requests that you make two additional analyses and/or explanations. That reviewer, an expert in the testosterone field, states:

1) One issue that I am still puzzled about is the difference in the findings from the data reported by Ruth et al. with respect to the association between genetically determined free testosterone and diabetes risk. The authors list "fewer cases of T2D leading to inadequate statistical power or other differences in the populations used for T2D analysis." If the same database of UK Biobank was used in both the analyses, why would there be a difference in the number of T2D cases or in the study population?

We understand the confusion and hope to clarify. Although both studies used the UK Biobank as the source of testosterone data, there were differences in the source of genetic estimates for outcomes, such as type 2 diabetes. Our study employed a one-sample Mendelian randomization design deriving estimates from the UK Biobank, whereas Ruth et al. employed a two-sample Mendelian randomization design deriving estimates from male-specific GWAS summary statistics from the DIAGRAM consortium. The DIAGRAM consortium contains 34,990 cases of type 2 diabetes, whereas UK Biobank contains 11,079 cases. However, sex-specific results of the DIAGRAM consortium are unpublished and unavailable to the public at this time. Despite constraints on data availability, we attempted to approximate analyses performed by Ruth et al. in the context of our study design by combining SNPs associated with bioavailable testosterone from their study with the same T2D cases from UK Biobank used in our analysis. Results are consistent with our study showing no effect on type 2 diabetes across IVW (OR=0.97 (95%CI=0.86 to 1.08); p=0.55), MR-PRESSO (OR=1.02 (95%CI=0.93 to 1.12); p=0.65), and MR-RAPS (OR=0.99 (95%CI=0.89 to 1.10); p=0.88). Therefore, in addition to the other noted differences, such as more robust control for pleiotropic effects of SHBG in our study, this could explain the difference in the association with type 2 diabetes.

We’ve amended the manuscript to better clarify this distinction in the Discussion, “The apparent difference between protective effects of testosterone observed in a previous MR analysis of testosterone and lack of protective effect in our study might be a result of less stringent control for pleiotropic effects of SHBG in the previous study. […] Other reasons may include genetic variants explaining less variation in testosterone levels in our study, fewer cases of T2D leading to inadequate statistical power to detect weaker effects in our study, or other differences between the populations of the UK Biobank in our study and DIAGRAM consortium used by Ruth et al.”.

2) I also continue to be concerned about the calculated free T concentrations that are substantially lower than those described previously in community-dwelling men. I recognize that these are the numbers that the UK Biobank has provided, but it would be worth re-checking the calculations to make sure there is no inadvertent systematic error in computation.

We appreciate the reviewer’s insight into this abnormality. We share the curiosity regarding the apparently lower levels of testosterone in the UK Biobank population, so we rechecked our calculations against the Vermeulen equation and found no error. As we comment on in the Discussion, the low levels may be attributable to the older average age in the UK Biobank and/or use of immunoassays in the measurement of sex hormones. Indeed, average total testosterone is similarly low (11.9 nmol/L or 343 ng/dL) in the UK Biobank relative to reference ranges (Bhasin et al., 2011) (Travison et al., 2017). Most importantly, the free testosterone levels calculated by our group are unlikely to represent a computational error as they are comparable to levels reported by other independent groups using UK Biobank data in both peer-reviewed literature (Peila, Arthur and Rohan, 2020) (Yeap et al., 2020) and preprint (Watts et al., 2020) (Fan et al., 2020).

https://doi.org/10.7554/eLife.58914.sa2

Article and author information

Author details

  1. Pedrum Mohammadi-Shemirani

    1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    2. Thrombosis and Atherosclerosis Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    3. Department of Medical Sciences, McMaster University, Hamilton, Canada
    Contribution
    Data curation, Software, Formal analysis, Investigation, Visualization, Writing - original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6740-7858
  2. Michael Chong

    1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    2. Thrombosis and Atherosclerosis Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    3. Department of Biochemistry and Biomedical Sciences, McMaster University, Hamilton, Canada
    Contribution
    Data curation, Software, Formal analysis, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Marie Pigeyre

    1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    2. Department of Medicine, McMaster University, Hamilton Health Sciences, Hamilton, Canada
    Contribution
    Conceptualization, Writing - review and editing, Analysis and interpretation of data
    Competing interests
    No competing interests declared
  4. Robert W Morton

    Department of Kinesiology, McMaster University, Hamilton, Canada
    Contribution
    Writing - review and editing, Analysis and interpretation of data
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0099-4167
  5. Hertzel C Gerstein

    1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    2. Department of Medicine, McMaster University, Hamilton Health Sciences, Hamilton, Canada
    Contribution
    Conceptualization, Writing - review and editing, Analysis and interpretation of data
    Competing interests
    HCG reports research grants from Eli Lilly, AstraZeneca, Merck, Novo Nordisk, and Sanofi; honoraria for speaking from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Novo Nordisk, and Sanofi; and consulting fees from Abbott, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Merck, Novo Nordisk, Janssen, Sanofi, Kowa, and Cirius.
  6. Guillaume Paré

    1. Population Health Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    2. Thrombosis and Atherosclerosis Research Institute, David Braley Cardiac, Vascular and Stroke Research Institute, Hamilton, Canada
    3. Department of Pathology and Molecular Medicine, McMaster University, Michael G. DeGroote School of Medicine, Hamilton, Canada
    4. Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Canada
    Contribution
    Conceptualization, Data curation, Supervision, Funding acquisition, Methodology, Project administration, Writing - review and editing
    For correspondence
    pareg@mcmaster.ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6795-4760

Funding

Canadian Institutes of Health Research (Frederick Banting and Charles Best Canada Graduate Scholarships Doctoral Award)

  • Michael Chong

Canadian Institutes of Health Research (Post-Doctoral Fellowship)

  • Robert W Morton

McMaster University (E.J. Moran Campbell Internal Career Research Award)

  • Marie Pigeyre

McMaster University (McMaster-Sanofi Population Health Institute Chair in Diabetes Research and Care)

  • Hertzel C Gerstein

Cisco Systems (Professorship in Integrated Health Biosystems)

  • Guillaume Paré

Canada Research Chairs (Canada Research Chair in Genetic and Molecular Epidemiology)

  • Guillaume Paré

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors are thankful for all the participants that contributed to the UK Biobank study.

Ethics

Human subjects: UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee (REC reference: 11/NW/0382). This research was conducted using the UK Biobank under Application Number 15255.

Senior Editor

  1. Eduardo Franco, McGill University, Canada

Reviewing Editor

  1. Dolores Shoback, University of California, San Francisco, United States

Reviewers

  1. Shalender Bhasin
  2. Mathis Grossman
  3. Qingyuan Zhao, University of Cambridge, United Kingdom

Publication history

  1. Received: May 14, 2020
  2. Accepted: October 13, 2020
  3. Accepted Manuscript published: October 16, 2020 (version 1)
  4. Version of Record published: October 27, 2020 (version 2)

Copyright

© 2020, Mohammadi-Shemirani et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 562
    Page views
  • 56
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Genetics and Genomics
    2. Microbiology and Infectious Disease
    Anne M Stringer et al.
    Research Article

    A hallmark of CRISPR-Cas immunity systems is the CRISPR array, a genomic locus consisting of short, repeated sequences ('repeats') interspersed with short, variable sequences ('spacers'). CRISPR arrays are transcribed and processed into individual CRISPR RNAs that each include a single spacer, and direct Cas proteins to complementary sequence in invading nucleic acid. Most bacterial CRISPR array transcripts are unusually long for untranslated RNA, suggesting the existence of mechanisms to prevent premature transcription termination by Rho, a conserved bacterial transcription termination factor that rapidly terminates untranslated RNA. We show that Rho can prematurely terminate transcription of bacterial CRISPR arrays, and we identify a widespread antitermination mechanism that antagonizes Rho to facilitate complete transcription of CRISPR arrays. Thus, our data highlight the importance of transcription termination and antitermination in the evolution of bacterial CRISPR-Cas systems.

    1. Genetics and Genomics
    Xinhai Ye et al.
    Research Article Updated

    Insects utilize diverse food resources which can affect the evolution of their genomic repertoire, including leading to gene losses in different nutrient pathways. Here, we investigate gene loss in amino acid synthesis pathways, with special attention to hymenopterans and parasitoid wasps. Using comparative genomics, we find that synthesis capability for tryptophan, phenylalanine, tyrosine, and histidine was lost in holometabolous insects prior to hymenopteran divergence, while valine, leucine, and isoleucine were lost in the common ancestor of Hymenoptera. Subsequently, multiple loss events of lysine synthesis occurred independently in the Parasitoida and Aculeata. Experiments in the parasitoid Cotesia chilonis confirm that it has lost the ability to synthesize eight amino acids. Our findings provide insights into amino acid synthesis evolution, and specifically can be used to inform the design of parasitoid artificial diets for pest control.