Multi-trait genome-wide association study of opioid addiction: OPRM1 and beyond.

Genomic SEM model and Manhattan plot. (a) A common factor (pg) gSEM model (using GenomicSEM) is fit with summary statistics from GENOA, MVP12-YP-SAGE, PGC, and Partners Health cohorts. Standardized estimates and standard errors are shown for each free parameter. Model fit is shown by a non-significant chi-square test, high Akaike information criterion (AIC, higher is better) and comparative fit index (CFI) equal to 1.0, and low standardized root mean squared root (SRMR) values (ideally < 0.05). (b) Manhattan plot for gSEM results with summary statistics from GWAS from each cohort. Bonferroni correction was used to correct for multiple comparisons; associations with P < 2 × 10–8 (indicated by horizontal black bar) were genome-wide significant (top SNP highlighted in red).

Association of major haplotypes for genome-wide significant OPRM1 variants with OA. (a) The 3 major haplotypes for genome-wide significant OPRM1 variants. Haplotype A is the predominant haplotype (frequency ~ 0.69 among contributing cohorts) and consists of major alleles for all variants. Haplotype B (frequency ~ 0.13 among contributing cohorts) consists of the minor allele for rs1799971 and the major allele for all other variants. Haplotype C (frequency ~ 0.16 among contributing cohorts) consists of the major allele for rs1799971 and minor allele for all other variants. The cohorts for whom we had the raw data to conduct the haplotype analyses were: UHS, VIDUS, ODB, Yale-Penn, CATS and Kreek (Supplementary Table 1). (b) Association of OPRM1 haplotypes with OA. Haplotype C is associated with increased risk of OA when compared to Haplotype A or Haplotype B, whereas Haplotype B does not have a significant impact on OA relative to Haplotype A. The single variant results using the cohorts contributing to the haplotype analyses were: rs1799971 beta = -– 0.058, p = 0.135; rs9478500 beta = 0.205, p = 2.43 × 10–9.

Genetic correlations of opioid addiction (OA) with 38 other brain-related phenotypes. Correlations were calculated using linkage disequilibrium (LD) score regression with the gSEM OA GWAS meta-analysis results, compared with results made available via LD Hub or study investigators (see Supplementary Table 24 for original references). Phenotypes were grouped by disease/trait or measurement category, as indicated by different colorings. Dots indicate the mean values for genetic correlation (rg); error bars show the 95% confidence intervals; the dashed vertical black line corresponds to rg = 0 (no correlation with OA), and the solid vertical black line corresponds to rg = 1.0 (complete correlation with OA). Phenotypes with significant correlation with OA are bolded (1 degree of freedom Chi-square test; Bonferroni adjusted p-value < 0.05 after accounting for 38 independent tests). Exact p-values are provided in Supplementary Table 10). CUD, Cannabis use disorder; DPW, drinks per week; FTND, Fagerström test for nicotine dependence; HSI, heaviness of smoking index; CPD, cigarettes per day; ADHD, attention deficit hyperactivity disorder; PTSD, post-traumatic stress disorder; MDD, major depressive disorder; ASD, autism spectrum disorder; ICV, intracranial volume.

Gene-level Manhattan Plot. GWAS results were summarized at the gene-level using MAGMA. Bonferroni correction was used to correct for multiple comparisons; associations with P < 3 × 10–6 (indicated by horizontal red dotted line) were genome-wide significant.

Colocalization of GWAS loci and QTLs for selected genes across 10 brain tissues. Posterior probabilities of supporting hypotheses regarding the association of each trait with SNPs in a region were calculated using coloc. For OPRM1, SNP-gene cis-eQTL associations were reported in GTEx Analysis v8 for only 6 of the 10 tissues.