Supplementary Material for: Association of Predicted Expression and Multimodel Association Analysis of Substance Abuse Traits

Introduction
Genome-wide association studies(GWAS) have played a critical role in identifying many thousands of loci associated with complex phenotypes and diseases. This has led to several translations of novel disease susceptibility genes into drug targets and care. This, however, has not been the case for analyses where sample sizes are small, which suffer from multiple comparison testing. The present study examined the statistical impact of combining a burden test methodology, PrediXcan, with a multimodel meta-analysis, Cross Phenotype Association (CPASSOC).
Methods
The analysis was conducted on 5 addiction traits: family alcoholism, cannabis craving, alcohol, nicotine, and cannabis dependence, and 10 brain tissues: anterior cingulate cortex BA24, cerebellar hemisphere, cortex, hippocampus, nucleus accumbens basal ganglia, caudate basal ganglia, cerebellum, frontal cortex BA9, hypothalamus, and putamen basal ganglia. Our sample consisted of 1,862 participants from the University of California, San Francisco (UCSF) Family Alcoholism Study.
Results
The post PrediXcan, gene-phenotype association without aggregation resulted in 2 significant results, HCG27 and SPPL2B. Aggregating across phenotypes resulted no significant findings. in 1 significant and 1 suggestive association, COA5 and HLA-B, respectively. Aggregating across tissues resulted in 15 significant and 5 suggestive associations: PPIE, RPL36AL, FOXN2, MTERF4, SEPTIN2, CIAO3, RPL36AL, ZNF304, CCDC66, SSPOP, SLC7A9, LY75, MTRF1L, COA5, and RRP7A; RPS23, GNMT, ERV3-1, APIP, HLA-B, respectively.
Discussion
Given the relatively small size of the cohort, this multimodel approach was able to find over a dozen significant associations between predicted gene expression and addiction traits. With the onset of improved transcriptome data, this approach should increase in efficacy.