Unraveling the genetic architecture of smoking: pathways of genetic risk through addiction and socioeconomics

Background

Smoking behaviours are heritable. Currently, geneticists are trying to unravel the genetic architecture of these behaviours with genome-wide association studies (GWASs). In a GWAS, the associations between a trait (e.g. smoking initiation) and millions of genetic loci across the genome are estimated. Whilst the available traits that represent smoking behaviours contain data from millions of individuals, they are generally questionnaire items that capture a particular aspect of smoking (e.g. age of smoking initiation), limiting the study of smoking as a whole. In addition, GWASs do not account for important confounders and correlates. For instance, the genes underlying smoking and use of other substances are highly correlated and this is most likely captured in a smoking GWAS, introducing noise. By depicting the genetic relationships between smoking and other highly related traits (use of other substances, and socioeconomic status), we aim to obtain ‘cleaner’ smoking-specific genome-wide genetic estimates that encompass a wide range of smoking traits and are more strongly related to smoking-specific vulnerability (e.g. related to nicotine metabolism).

Methods

We use a multivariate statistical framework based on structure equation modelling (SEM) called Genomic SEM, that uses GWAS summary statistics from different traits. With Genomic SEM we will fit various explorative and theory-driven models, to capture the joint genetic architecture of smoking and related traits. The latent factor structure will yield insights on the joint genetic structure of said traits. Once the architecture is clear, a GWAS will be run on the smoking-specific latent factor.

Results

Preliminary results show genetic associations between smoking, substance use, and socioeconomic status traits. We infer the existence of a latent genetic factor corresponding to a general vulnerability to addiction, as well as separate latent factors for each individual substance (cannabis, tobacco, alcohol), which emphasizes the value of obtaining a clean signal as well as grouping the smoking variables together. Furthermore, we estimate genome-wide genetic effects for the latent smoking factor, which are more directly related to the biology of smoking and can be used to better elucidate which genes influence smoking and through which pathways.

Conclusions

We conclude that despite the widely accepted impression that genetic research accuracy will improve naturally with the gradual increase of the available datasets, developing methods that refine the genetic signals is highly beneficial. Firstly, it allows us to understand the underlying structure between genetic factors, better assess risk factors for smoking and biological mechanisms that may provide future targets for (biological) interventions. In addition, it provides more specific (smoking) genetic instruments that can be used for causal association studies.