The Landscape of Shared and Divergent Genetic Influences across 14 Psychiatric Disorders

Abstract

Psychiatric disorders display high levels of comorbidity and genetic overlap^1,2. Genomic methods have shown that even for schizophrenia and bipolar disorder, two disorders long-thought to be etiologically distinct³, the majority of genetic signal is shared⁴. Furthermore, recent cross-disorder analyses have uncovered over a hundred pleiotropic loci shared across eight disorders⁵. However, the full scope of shared and disorder-specific genetic basis of psychopathology remains largely uncharted. Here, we address this gap by triangulating across a suite of cutting-edge statistical genetic and functional genomic analyses applied to 14 childhood- and adult-onset psychiatric disorders (1,056,201 cases). Our analyses identify and characterize five underlying genomic factors⁶ that explain the majority of the genetic variance of the individual disorders (~66% on average) and are associated with 268 pleiotropic loci. We observed particularly high levels of polygenic overlap⁷ and local genetic correlation⁸ and very few disorder-specific loci⁹ for two factors defined by: (i) schizophrenia and bipolar disorder ("SB factor"), and by (ii) major depression, PTSD, and anxiety ("internalizing factor"). At the functional level, we applied multiple methods^10-12 which demonstrated that the shared genetic signal across the SB factor was substantially enriched in genes expressed in excitatory neurons, whereas the internalizing factor was associated with oligodendrocyte biology. By comparison, the genetic signal shared across all 14 disorders was enriched for broad biological processes (e.g., transcriptional regulation). These results indicate increasing differentiation of biological function at different levels of shared cross-disorder risk, from quite general vulnerability to more specific pathways associated with subsets of disorders. These observations may inform a more neurobiologically valid psychiatric nosology and implicate novel targets for therapeutic developments designed to treat commonly occurring comorbid presentations.

Figure 1.. Genome-wide structural models.

Panel A depicts the heatmap of r_g’s across the 14 disorders as estimated using LD-score regression (LDSC) on the lower diagonal and the correlations among the psychiatric genomic factors as estimated using GenomicSEM above the diagonal. Disorders that load on the same factor are depicted in the same color. LDSC estimates were used as input to Genomic SEM to produce the results in the remaining two panels. Panel B depicts estimates from the five-factor model along with standard errors in parentheses. Estimates are standardized relative to the SNP-based heritabilities, where the total variance in the disorders is equal to the sum of the squared factor loading (the single-headed arrow(s) from the factor to the disorder) and the residual variance (the values on the double-headed arrows on the single-color circles). Disorders are shown as pie charts with the proportion of unique (residual) variance shaded in purple and the variance explained by the psychiatric factors in the color of the corresponding factor. Panel C displays the standardized estimates from the p-factor model. The disorders are again color coded in blue and orange as in Panel B and the first-order factors (F1-F5) additionally color coded to show variance explained by the second-order, p-factor in yellow. ADHD = attention-deficit hyperactivity disorder; ASD = autism spectrum disorder; OCD = obsessive compulsive disorder; SCZ = schizophrenia; TS = Tourette’s syndrome; MD = major depression; AN = anorexia nervosa; BIP = bipolar disorder; ANX = anxiety disorder; PTSD = post-traumatic stress disorder; AUD = alcohol use disorder; OUD = opioid use disorder; NIC = nicotine dependence; CUD = cannabis use disorder.

Figure 2.. LAVA and MiXeR results for select disorders.

Results are shown for LAVA (volcano plots) and MiXeR (Venn diagrams) for four of the most well-powered disorders: schizophrenia (SCZ), bipolar disorder (BIP), major depression (MD), and attention-deficit/hyperactivity disorder (ADHD). LAVA results are shown as a volcano plot of the local genetic correlations (r_g) on the x-axis and -log10(p) values on the y-axis. Stronger negative r_g’s are depicted by darker shades of blue, while stronger positive r_g’s are shown in darker shades of red. The points for local r_g’s that were significant at a Bonferroni corrected threshold are depicted with black outlines. MiXeR results are shown as adjacent Venn diagrams that convey unique and shared polygenic components at the causal level. The numbers within the Venn diagrams indicate the estimated quantity of causal variants (in thousands) per component, explaining 90% of SNP heritability in each phenotype, followed by the standard error. The size of the circles reflects the degree of polygenicity. r_g is represented in the horizontal bars beneath the Venn diagrams.

Figure 3.. Local genetic correlations.

Panel A displays an overview of the average patterns of local genetic correlations (r_g’s) across the genome for all pairs of disorders in the form of a heatmap (below diagonal) and a network plot (above diagonal). The colors of the heatmap represent the average local r_g’s across all evaluated loci, with dot size reflecting the strengths of average associations, and numbers indicating how many of the local r_g’s were significant. These results are mirrored in the network plot, where the width or the edges reflect the number of significant associations, meaning that only disorders with at least one significant local r_g are connected, and the edge opacity the strength of the average local r_g across tested loci. Note that label colors are concordant with the Genomic SEM factor structure from Fig. 1 and, as shown, disorders of similar colors also tend to be proximally located within the network. Panel B displays the local r_g structure within the top r_g hotspot on chromosome 11 (112,755,447–114,742,317 [GRCh37]), i.e. the locus where the greatest number of significant r_g’s were found across all disorder pairs. Here, the network plot illustrates all significant r_g’s detected in this locus, with both edge width and opacity reflecting the strength of the association. The region plot in the middle displays the genes contained within the hotspot, and the table below shows the genetic correlation estimates (Rho), 95% confidence intervals (CI_lower, CI_upper), variance explained (R²) and p-values (P) for all significant pairwise local r_g’s in this locus. Label colors are concordant with those used for the Genomic SEM factor structure in Fig. 1.

Figure 4.. Locus level results.

Panel A depicts a heatmap of CC-GWAS loci below the diagonal across pairwise combinations of disorders, where darker orange shading indicates a higher number of CC-GWAS hits. CC-GWAS results are not shown for the Internalizing disorders as their genetic correlations were too high or for nicotine dependence as this is a continuously measured trait. Genomic SEM results (number of hits and mean chi-square for each factor and factor-specific Q_SNP estimate) are reported above the diagonal. Results for the p-factor are depicted above the plot in yellow, along with a Venn diagram of overlap between p-factor, p-factor Q_SNP, and overall CC-GWAS hits. The disorders are ordered and colored according to the Genomic SEM factor structure from Fig. 1. Panels B and C depict the Miami and QQ-plots for the p-factor and SBs factors, respectively. These panels depict results for the -log10(p-values) for the factor on the top half of the Miami plot and the log10(p-values) for Q_SNP on the bottom half. Per the legend at the top of this set of panels, factors hits that were within 100 kb of univariate hits are shown in black triangles, novel hits for the factors that were not within 100 kb of a univariate or Q_SNP hit are depicted as red triangles, and Q_SNP hits as purple diamonds. The gray dashed line indicates the Bonferroni corrected, genome-wide significance threshold. Panel D depicts the Manhattan plot for the CC-GWAS comparison across major depression (MD) and schizophrenia (SCZ), which produced the most hits (depicted as orange diamonds), as well as the scatterplot of standardized case-control effect sizes of MD (x-axis) vs SCZ (y-axis), with CC-GWAS significant SNPs labeled in red.

Figure 5.. Functional genomics annotation of multi-diagnostic variants.

(A) Gene ontology enrichment analysis of predicted target genes with transdiagnostic associations (i.e., variants associated with the p-factor), or those target genes associated with the SB factor that were not overlapping with p-factor target genes. (B) Averaged and normalized expression levels of target genes of the indicated classes along the temporal trajectory of human brain development. Shading around the lines reflect 95% confidence intervals. (C and D) Enrichment by expression of CDG3 variants target genes associated with the indicated factors in (C) fetal brain cell types using two independent scRNA-Seq data sets^, or (D) adult brain cell types using two independent snRNA-Seq data sets.^, SB_Q = SB factor Q_SNP; Int_Q = Internalizing disorders Q_SNP; Int = Internalizing disorders factor.

Figure 6.. Stratified Genomic SEM results.

Panel A depicts the enrichment point estimates for the four evolutionarily conserved annotations for the five-factor and p-factor models. The remaining two panels display only results for the SB, Internalizing, and p-factor due to the limited signal for the other factors. The two Conserved Mammals annotations reflect two sets of evolutionarily conserved genetic variants created by the original S-LDSC developers^, from data from Lindblad-Toh et al. (denoted with a ^{^}) and genomic evolutionary rate profiling scores from Davydov et al. (denoted with ^{^^}). Panel B displays patterns of enrichment across 9 minor allele frequency (MAF) bins (note that one of the 10 initial bins was removed following QC procedures described in the Method). Within the MAF range of 5% to 50%, these results reveal increasing enrichment across higher levels of MAF, suggesting that the genetic signal is most concentrated in the most common variants. Panel C shows the enrichment results for different brain cell types, protein-truncating variant intolerant (PI) genes, and the intersection across PI genes and brain cell types. All point estimates are depicted with 95% confidence intervals. Enrichment estimates that were significant at a strict Bonferroni significance threshold are shown with a *.