Year: 2014
Type: editorial
Abstract: Vink et al. 1 generate a polygenic risk score by summing across more than 1 000 000 single nucleotide polymorphisms (SNPs) that were associated with tobacco smoking phenotypes in the large (n up to 68 000) genome-wide association study meta-analysis of the Tobacco and Genetics Consortium (TAG) 2. The resulting score was used to predict variance in identical tobacco smoking measures (e.g. cigarettes per day) in a smaller sample (n = 1583) consisting of members of the Netherlands Twin Registry (NTR). Surprisingly, while the polygenic score did not predict a significant proportion of variance in the corresponding tobacco phenotypes, it was a statistically significant predictor of modest proportions of variance (ranging from 0.4 to 1.5%) in alcohol and cannabis-related phenotypes. Several key features of this study are noteworthy. First, instead of relying on the single, most commonly implicated and most significantly associated variant for cigarettes per day (CPD) (rs16969968, or its proxy, rs1051730), the authors aggregate additively across several SNPs. This is more consistent with the view that each variant has only a modest effect size and further, that even SNPs not meeting the punishing threshold of genome-wide significance (P < 5 × 10−8) contribute to phenotypic variance. However, the extent to which rs1051730 drives the effects of the CPD polygenic score would be worth exploring (i.e. by examining variance reduced by its exclusion). Secondly, a variety of tobacco (and alcohol, cannabis) measures are studied. As substance involvement is a multi-stage process with genetic influences probably playing different roles at different stages (e.g. age at onset of use versus CPD), such a comprehensive examination is appealing. In fact, in TAG, SNPs on chromosome 15 that were associated with CPD were, in fact, not significantly genome-wide for smoking initiation. Thirdly, capitalizing on the frequent co-occurrence of alcohol and cannabis use with tobacco smoking, and their shared genetic underpinnings, the authors find that SNPs associated with tobacco smoking are associated with variance in these phenotypes. While they posit this shared liability as the rationale for this intriguing finding, one wonders how this might be possible, given that the within-trait variance prediction was not significant. How might genes associated with tobacco smoking, some of which include variants in cholinergic nicotinic receptors, predict variance in alcohol and cannabis use but not tobacco smoking? One possibility is that the lack of findings reflects reduced power due to the smaller sample size of the NTR. Might other moderators of heritable variation be at play? The study notes that the birth year of NTR participants ranged from 1915 to 1994, with data collected from 1991 to 2009. This raises the possibility that developmental and societal (i.e. cohort) effects may have attenuated genetic effects on smoking. For instance, the median ages reported in the TAG study range from 44 to 72 years; however, it is possible that a proportion of the NTR respondents in this study were considerably younger (born 1990 or later). These individuals may not have had the opportunity to start smoking (or using cannabis, which might relate to the somewhat low rate of cannabis use, 15%, in the sample) despite a genetic susceptibility. While the authors examine age of onset only in those who ever tried the substance, whether all participants were old enough to have had the opportunity to try the substance (i.e. not censored) is unclear. There is, in fact, evidence that rs16969968 has differential associations with smoking in early- versus late-onset smokers 3. These developmental effects may be more pronounced for some SNPs but not others—for instance, one study showed that the odds of heavy smoking in carriers of the risk allele in rs1051730 was the same in 14- versus 31-year-olds 4. Here, the additive nature of such scores poses a challenge, as developmental effects of all contributing variants is considered to be consistent and unidirectional. Another possibility is that the oldest individuals in the sample may have differed in their patterns of tobacco smoking due to cohort effects 5. At least one twin study indicates that heritable variation in smoking desistence increased in Australia with the implementation of more restrictive legislature in the mid-1970s 6. Even though age and cohort were accounted for as covariates, interactions between the polygenic score and these contextual factors may be implicated in the interesting pattern of findings that emerge. Whether these findings are false positives or indicative of a complex interplay between genetic susceptibility and patterns of tobacco, alcohol and cannabis use, one important finding emerges—despite the rather large aggregation of genetic variants, less than 2% of variance is explained. In contrast, models including age and sex explained up to 20% of the variance in these measures, and other studies have begun to document that family history is a better predictor of addictive behaviours than such polygenic scores 7. This leads us to speculate that prediction of heritable variation in substance use, particularly its earliest stages, may require the consideration of environmental and contextual factors. None.
Source: Addiction
DOI: 10.1111/add.12583
Authors: Michael Lynskey, Arpana Agrawal
Institutions: King's College London, Washington University in St. Louis
Language: English
Subfield: Genetics
Domain: Life Sciences
Open Access status: bronze
Funder: National Institute on Drug Abuse
Award: G2988863959
OpenAlex ID: W1607646894