In our study, despite the lack of independent replication of rs115455482 in COGA, when we meta-analyzed Yale-Penn and COGA samples, the association of rs115455482 genotype with Ca-HL improved. The lack of significant replication in COGA may be due to the differential distributions of cannabis-dependence (CAD) criterion counts in Ca-HL cases and controls between Yale-Penn and COGA samples (Supplementary Fig. S6). We found that Ca-HL was significantly associated with CAD criterion counts both in Yale-Penn and COGA samples (Supplementary Fig. S7), which raises the question of potential involvement of the interaction between rs115455482 genotype and CAD severity in Ca-HL. As shown in Supplementary Fig. S7, we observed that the interaction between rs115455482*T and CAD criterion count was significantly associated with Ca-HL in Yale-Penn 1 EAs (P = 1.25 × 10−5) but not in Yale-Penn 2 EAs (P = 8.94 × 10−2) or COGA EAs (P = 0.72). This may be a potential reason for the lack of replication of rs115455482 in COGA, as the interaction between rs115455482*T and CAD criterion counts were different among Yale-Penn and COGA samples.
