Uncovering the complex relationship between balding, testosterone and skin cancers in men

Abstract

Male-pattern baldness (MPB) is related to dysregulation of androgens such as testosterone. A previously observed relationship between MPB and skin cancer may be due to greater exposure to ultraviolet radiation or indicate a role for androgenic pathways in the pathogenesis of skin cancers. We dissected this relationship via Mendelian randomization (MR) analyses, using genetic data from recent male-only meta-analyses of cutaneous melanoma (12,232 cases; 20,566 controls) and keratinocyte cancers (KCs) (up to 17,512 cases; >100,000 controls), followed by stratified MR analysis by body-sites. We found strong associations between MPB and the risk of KC, but not with androgens, and multivariable models revealed that this relationship was heavily confounded by MPB single nucleotide polymorphisms involved in pigmentation pathways. Site-stratified MR analyses revealed strong associations between MPB with head and neck squamous cell carcinoma and melanoma, suggesting that sun exposure on the scalp, rather than androgens, is the main driver. Men with less hair covering likely explains, at least in part, the higher incidence of melanoma in men residing in countries with high ambient UV.

Fig. 1. Schematic diagram outlining the overall study approach of modelling genetic outliers via MVMR.

Each panel (a), (b), (c) and (d) is listed in chronological order of the analysis procedure. a Schematic MR diagram. b MR scatter plot. c Selection of candidate traits for inclusion into MVMR via PheWAS findings. d Modelling the candidate traits into the MVMR analysis to obtain the marginal effect of MPB on skin cancer risk, by conditioning on endogenous testosterone levels and other candidate traits.

Fig. 2. Comparison of MR effect sizes for MPB on skin cancer risk.

beta_IVW refers to the (fixed effect) inverse variance weighted MR effect estimate; p= corresponding two-sided z-test P-value of the beta_IVW estimate; MPB Male pattern baldness. Each panel illustrates the MR scatter plot for the association between genetically predicted MPB score and the log(OR) on various types of skin cancer. Each point represents a single MPB SNP instrument, with the corresponding horizontal and vertical error bars reflecting its standard error on the genetic association with MPB and skin cancers, respectively. Points in light blue are MPB variants that showed stronger evidence of association in the UKB subset independent of those used to derive the SNP-skin cancer association. The annotated SNPs on each panel refers to the SNPs identified as outliers via the MR-PRESSO outlier test. Apart from melanoma (right-bottom panel; which shows predominantly null findings), the IVW effect estimates derived using MPB variants after excluding the genetic outliers show strong attenuation of effect sizes towards the null. Source data are provided as a Source Data file.

Fig. 3. Comparison of MR-derived association between MPB and various univariable and MVMR models.

UniMR − MPB SNPs with p < 1e − 5 only: Univariable MR model for MPB on skin cancer risks using on MPB SNPs with association (z-score) two-tailed P value < 1e-5 on MPB in the independent UK Biobank subset. UniMR − excl. outliers: Univariable MR model for MPB on skin cancer risks using all MPB SNPs excluding pleiotropic SNPs detected by MR-PRESSO. MVMR Model 1: MVMR model incorporating MPB, freeT and totalT. MVMR model 2: MVMR model incorporating MPB, totalT, skin colour and hair colour. The error bars reflect the 95% confidence intervals around each OR estimate. For all MVMR models, the reported OR estimates are the marginal OR estimates for skin cancer per 1 SD increase in MPB upon conditioning on the genetic effect sizes from other traits included in the model. Note that for MVMR model 2, including freeT into the model resulted in the weakening of the combined instrument for MVMR (conditional F-statistics<10 for some traits), which might result in weak instrument bias and hence were not included in the main analysis. However, these findings can be accessed in Supplementary Tables. Source data is provided as a Source Data file.