Genetic Architecture and Molecular Neuropathology of Human Cocaine Addiction

Abstract

We integrated genomic and bioinformatic analyses, using data from the largest genome-wide association study of cocaine dependence (CD; n = 6546; 82.37% with CD; 57.39% male) and the largest postmortem gene-expression sample of individuals with cocaine use disorder (CUD; n = 36; 51.35% with CUD; 100% male). Our genome-wide analyses identified one novel gene (NDUFB9) associated with the genetic predisposition to CD in African-Americans. The genetic architecture of CD was similar across ancestries. Individual genes associated with CD demonstrated modest overlap across European-Americans and African-Americans, but the genetic liability for CD converged on many similar tissue types (brain, heart, blood, liver) across ancestries. In a separate sample, we investigated the neuronal gene expression associated with CUD by using RNA sequencing of dorsal-lateral prefrontal cortex neurons. We identified 133 genes differentially expressed between CUD case patients and cocaine-free control subjects, including previously implicated candidates for cocaine use/addiction (FOSB, ARC, KCNJ9/GIRK3, NR4A2, JUNB, and MECP2). Differential expression analyses significantly correlated across European-Americans and African-Americans. While genes significantly associated with CD via genome-wide methods were not differentially expressed, two of these genes (NDUFB9 and C1qL2) were part of a robust gene coexpression network associated with CUD involved in neurotransmission (GABA, acetylcholine, serotonin, and dopamine) and drug addiction. We then used a "guilt-by-association" approach to unravel the biological relevance of NDUFB9 and C1qL2 in the context of CD. In sum, our study furthers the understanding of the genetic architecture and molecular neuropathology of human cocaine addiction and provides a framework for translating biological meaning into otherwise obscure genome-wide associations.SIGNIFICANCE STATEMENT Our study further clarifies the genetic and neurobiological contributions to cocaine addiction, provides a rapid approach for generating testable hypotheses for specific candidates identified by genome-wide research, and investigates the cross-ancestral biological contributions to cocaine use disorder/dependence for individuals of European-American and African-American ancestries.

Keywords: GWAS follow-up; RNA sequencing; cocaine dependence; cocaine use disorder; genome-wide association study (GWAS); multiancestry.

Figure 1.

Miami plot visualizing results from gene-based association analyses. Each dot represents an individual protein-coding gene, the x-axis denotes chromosome number, and the y-axis shows the −log₁₀ p value. African-American results are displayed on top and European-American results (Huggett and Stallings, 2020) are shown on the bottom. The dashed red line represents genome-wide significance, and the dashed brown line represents an unadjusted/nominal p value threshold <0.05. Red dots are genes nominally significant in both African-Americans and European-Americans.

Figure 2.

Miami plot showing the associations of 130 genes from candidate neurotransmitter systems. Each gene is color coded by neurotransmitter type, and the different shapes represent the different parts of the system. The x-axis denotes chromosome number, and the y-axis shows the −log₁₀ p value with African-Americans displayed on top and European-Americans shown on bottom. The red dashed line represents the Bonferroni correction for multiple testing (p < 0.05/130), and the brown dashed line represents the unadjusted/nominal p value threshold < 0.05.

Figure 3.

The implicated tissue types based on genes nominally associated with cocaine dependence (CD) separately by ancestry are shown. The x-axis shows all tissue types (GTEx) sorted alphabetically, and the y-axis represents the –log₁₀ p value. Solid boxes denote results from the African-American analysis, and dashed boxes show European-American results from the study by Huggett and Stallings (2020). Red bars show replicated tissue types that were significantly enriched (p_adj < 0.05) across both ancestries. The labels of replicated tissues are emphasized in bold text on the x-axis.

Figure 4.

Volcano plot showing genes/transcripts that are expressed differently in human PFC neurons between control subjects (n = 17) and individuals with CUD (n = 19). Each dot represents a gene/transcript. The x-axis denotes the log₂ fold change with positive values corresponding to increased expression in those with CUD. The y-axis shows the –log₁₀ FDR-adjusted p value, and all genes above the red dashed line survive correction for multiple testing (133 gene/transcripts; p_adj < 0.05). We labeled all genes significantly associated with the genetic predisposition to CD and highlighted significantly differentially expressed genes/transcripts previously implicated in cocaine use and the most abundant noncoding transcripts (pseudogenes).

Figure 5.

Heat scatter plot depicting the correlation of neuronal dlPFC gene expression associated with CUD from African-Americans (n = 13) and European-Americans (n = 10). The x-axis shows the Wald statistics from the European-American differential expression analysis, and the y-axis represents the Wald statistics from the African-American differential expression analysis. Each dot represents a specific gene/transcript, and the bright red color shows the highest frequency, whereas the light purple/pink indicates the lowest frequency of genes/transcripts. The dashed black line highlights the Pearson product correlation of gene expression across ethnicities (r = 0.174, p < 2e-16).

Figure 6.

A, The x-axis shows the twelve WGCNA gene coexpression networks. The y-axis shows the absolute value of Wald statistics (from whole-sample differential expression analysis) of all genes within a defined/discrete WGCNA network. The directions of associations were determined by assessing whether mean effect sizes for gene networks were positive or negative. All WGCNA gene networks to the right of the dashed red line are significantly associated with cocaine use disorder (*p_adj < 0.05; **p_adj < 0.01; ***p_adj < 0.001; ****p_adj < 0.0001). B, The six associated WGCNA gene networks were subsequently tested for the enrichment of the 133 differentially expressed genes in dlPFC neurons. The y-axis represents the odds ratio calculated by a two-sided Fisher's exact test. Only the blue gene network demonstrated significant enrichment and was selected for follow-up investigation. ****p < 0.0001. C, Potential functions of blue gene network via functional annotation analysis of pathways from KEGG. We picked 30 representative functions/pathways and grouped them into five domains, which are labeled by colors.

Figure 7.

The genes from the blue gene network significantly enriched for drug addiction and neurotransmission from KEGG (2019) and their relation to the genes associated with the predisposition to CD (in triangles) are shown. Coexpression patterns with NDUFB9 and C1qL2 are highlighted in red. Only coexpression patterns above a weighted r > 0.05 are shown. Genes shown in cyan represent increased expression in dlPFC neurons for those with CUD, and those in magenta represent decreased expression.