Our study also has several limitations. First, lifetime cannabis use was analyzed as a dichotomous measure combining experimental and regular users in a single group. Additionally, the different samples varied substantially regarding the age of the participants, the prevalence of cannabis use, and the country’s policies regarding cannabis use. All these factors may introduce heterogeneity which may reduce the power to detect genetic associations. Secondly, power of some analyses may have been limited. For example, the MR analysis from cannabis to schizophrenia was based on an instrument of only 5 SNPs, and the summary statistics of some traits used for the genetic correlation analyses in LD-score regression (e.g. cannabis dependence) were based on a small sample size. Finally, some regions identified in the SNP-based analyses did not appear in the gene-based analyses. In particular, inspection of the region around rs9773390 (in ZNF704) showed that the top SNP in this region was isolated, and that the SNP was only available in two of the three datasets (not in UK-Biobank). SNPs in LD with the top SNP that were included in all three datasets were not genome-wide significant. Thus, this result may not represent a robust association.
