Our study possesses several limitations, and some of those are typical for many GWAS. For example, the size of the current study may have been too small to detect associations at a genome wide significance level. Phenotypes studied are genetically complex and are likely determined by many genes of modest effects, which makes them difficult to detect with genome-wide significance for the currently studied sample size. It is worth noting that the much larger Tobacco and Genetics Consortium did not identify any genome-wide significant associations for age at smoking initiation in a meta-analysis encompassing over 20,000 individuals.10 This emphasizes the need for even larger studies to study smoking behaviors in order to detect variants with presumably low effect sizes. Despite the modest effect sizes of the genetic variants implicated by GWAS, key biological pathways may be identified using such approaches. Secondly, the genotype imputation accuracy could have had an impact on our results. Many top SNPs from the analyses of age at smoking initiation were imputed in all 4 cohorts; however, the association peaks of these regions also contained SNPs
