Investigating heterogeneous genetic effects contributing to smoking behaviors

Abstract

AbstractThe genetic architecture of numerous smoking behaviors is highly polygenic, but these genetic effects are heterogeneous and depend on moderating factors. Here, we used common SNPs from the UK Biobank to investigate heterogeneous genetic effects for smoking heaviness using cigarettes per day (CPD) records, smoking initiation (SI), and smoking cessation (SC). We assessed heterogeneous effects across levels of sex, age of smoking initiation, major depressive disorder (MDD) DSM-V-like diagnosis, generalized anxiety disorder (GAD) DSM-V-like diagnosis, and whether an individual has seen a psychiatrist for nerves, anxiety, tension, or depression. We observed suggestive evidence of heterogeneous genetic effects for CPD and SC, moderated by MDD and GAD, respectively [ (SE = 0.15) between MDD cases and controls, and (SE = 0.28) between GAD cases and controls, p < 0.05]. We detected 5 SNPs with genome-wide significant evidence of heterogeneous effects moderated by either MDD or GAD (p-value < 5×10-8) for CPD and SC. We also observed strong evidence for heterogeneous genetic effects for SI between sexes (between-sex , SE = 0.02). While we detected no individual SNPs that were moderated by sex at genome-wide significance (all p-value > 5×10-8), we observed evidence of novel genome-wide significant SI-SNP associations using sex-stratified GWAS; six loci were discovered in either men or women separately that were not identified in a previous smoking meta-analysis that had a 6-fold larger, sex-combined sample. Furthermore, using several independent testing samples, there was suggestive evidence that the prediction ability of a polygenic risk score (PRS) for smoking initiation improved through the utilization of sex-specific SNP effects. This work suggests that a more nuanced approach to GWAS analyses is warranted, as potential heterogeneous effects can complicate variant discovery and polygenic risk score accuracy.Author SummarySmoking imposes a heavy health burden and is highly polygenic in architecture. Consistent with most complex traits, many causal loci have yet to be identified, even when using the largest available samples. One possible reason for the difficulty in inferring genetic variants associated with such complex traits is that common genetic variants possess context-dependent (or heterogeneous) effects. Utilizing the UK Biobank, we find evidence for heterogeneous SNP effects on smoking initiation, heaviness, and cessation among psychiatric disorder cases and controls and between sexes. Failure to model such heterogeneity (when accounting for sample size) resulted in lower independent sample predictive ability. This work encourages a more nuanced approach to GWAS and polygenic risk prediction. The assumption that all genetic effects are homogeneous limits our understanding of complex traits when heterogeneous effects are present.