Psychiatric disorders converge on common pathways but diverge in cellular context, spatial distribution, and directionality of genetic effects

Abstract

Psychiatric conditions share common genes, but mechanisms that differentiate diagnoses remain unclear. We present a multidimensional framework for functional analysis of rare copy number variants (CNVs) across 6 diagnostic categories, including schizophrenia (SCZ), autism (ASD), bipolar disorder (BD), depression (MDD), PTSD, and ADHD (N = 574,965). Using gene-set burden analysis (GSBA), we tested duplication (DUP) and deletion (DEL) burden across 2,645 functional gene sets defined by the intersections of pathways, cell types, and cortical regions. While diagnoses converge on shared pathways, mixed-effects modeling revealed divergence of pathway effects by cell type, brain region, and gene dosage. Factor analysis identified latent dimensions aligned with clinical axes. A primary factor (F1) captured reciprocal dose-dependent effects of DUP and DEL in SCZ reflecting positive and negative effects in excitatory versus inhibitory neurons and association versus sensory cortex. SCZ and ASD were both strongly aligned with F1 but with opposing directionalities. Orthogonal factors highlighted neuronal versus non-neuronal effects in mood disorders (F2) and differential spatial distributions of DEL effects in ADHD and MDD (F3). High-impact CNVs at 16p11.2 and 22q11.2 were enriched for combinations of cell-type-specific genes involved in pathways consistent with our broader findings. These results reveal molecular and cellular mechanisms that are broadly shared across psychiatric traits but differ between diagnostic categories in context and directionality.

Fig. 1|. Investigating association of pathways, cell types and brain regions by Gene Set Burden Analysis (GSBA).

Gene sets were derived for Pathway (from GO, KEGG, REACTOME, and BioCarta), Cell type (from single cell study, Velmeshev et al.), and Cortical regions (from Glasser parcellation of the Allen Brain Atlas). Case-control association of CNV burden collapsed across gene sets, was then tested by logistic regression and meta-analysis was performed across genotyping platforms. Functional gene set associations were tested for 6 major psychiatric conditions (ASD, ADHD, SCZ, PTSD, MDD, BD).

Fig. 2|

Rare CNVs association analysis results in molecular pathways and neuronal cell types. (a) Enrichment map showing clusters of functional modules that are significantly associated with any condition. CNV associations are color-coded as a portion with a node where red indicates a DEL association in ASD, orange indicates a DEL association in SCZ, blue indicates a DUP association in SCZ, and yellow indicates a DEL association in ADHD. Gene-sets not forming a cluster of 3 or more members were excluded. Gene set clusters are listed in Table S4. (b) The heatmap represents the results at the pathway-cluster level, with color indicating z-score from meta-analysis. (c) A UMAP plot displays cell clusters colored by prenatal (teal) and postnatal (red) periods. (d) Heatmaps show association results at the cell type level with color indicating z-score, where red represents a higher burden of CNVs in cases and blue represents a depletion of CNVs burden in cases. An asterisk indicates statistically significant associations (q-value <0.1). Summary statistics of the initial primary gene sets and for the final set of pathway clusters are in Tables S3, and S5 respectively.

Fig. 3|

Rare (a) DEL and (c) DUP association analysis results of the cortical brain regions in the 6 conditions. Color indicates the association level (z-score) with red indicating the CNV association with the cases, while blue indicates the depletion of CNVs in cases (Table S3). Correlation results between CNV associations in (b) DEL and (d) DUP against the dominant transcriptomic brain gradient (PC1 of AHBA). Each circle represents a brain region gene set. Kendall’s Tau and corresponding q-value are shown in the title of each scatterplot. Solid diagonal trend line indicates significant correlation (q_SPIN<0.05). The cortical map at the top left corner illustrates the transcriptomic gradient from PC1 AHBA.

Fig. 4|. Associations of pathways with psychiatric traits vary by cell-type and gene dosage.

(a) Schematic illustrating how gene sets were defined by intersecting pathway, cell type, and cortical region dimensions. Example intersections include Chromatin-ExNeu, Chromatin-Assoc, ExNeu-Assoc, and Chromatin-ExNeu-Assoc. (b) Full model R² estimates showing the total variance in gene-set z-scores explained by main effects and interaction terms for each diagnosis. Models included pathway, cell type, brain region, dosage, and all combinations of two-way and three-way interactions. (c) R² estimates for individual interaction terms, quantifying the contribution of each interaction to the explained variance. The pathway×celltype×dosage interaction consistently explains the largest proportion of variance across diagnoses, highlighting the importance of dosage-sensitive and cell-type-specific pathway effects (Tables S9–S10).

Fig. 5|. Differentiation of diagnostic categories based on gene-dosage effects in pathways by cell type and brain region.

(a) Genetic correlations between diagnostic categories when each diagnosis-dosage combination is treated as an independent component, see also Table S11, *p<0.05) **q<0.05). Diagnosis-dosages with factor loadings >0.25 were grouped and labeled to highlight psychiatric traits contributing to F1, F2 and F3. (b): Factor loadings of DEL and DUP for disorders reveal a distinct profile for each diagnostic category. (c, g, k) Gene set-factor scores for the three factors, cell types and pathways were ordered using a simple sign-based bi-clustering algorithm (see methods) (Table S14). (d,h,i) Factor scores are representative of dose-dependent effects of genes. Scatterplots of gene set effect sizes (z-score) are shown for the top 2 diagnosis-dosage groupings with highest absolute factor loadings for factor F1, F2, and F3, and factor score of each gene set is indicated using the same color scale as in panels c,g,k. Solid trend lines indicate significant correlation between the diagnosis-dosage pair. (e,j,m) Factor analysis of gene sets with genome-wide significant loci removed yielded results with highly concordant gene set factor scores (e,f,i,j,m,n; tau_F1=0.45, tau_F2=0.53, tau_F3=0.32, p<2.2e-16; Table S14), demonstrating that these patterns are not attributable to a select subset of major loci.

Fig. 6|. Cell-type specific expression of genes within major CNV loci 16p11.2 BP4-BP5 and 22q11.2 A-D suggests that the functional influence of a CNV in the brain may be driven by distinct pathway effects across a variety of cell types.

CNV associations displayed in (a) and (d) were obtained from Shanta et al. Colors indicate the association direction and effect size (z-score), and asterisks indicate FDR<10% results. (b) and (e) heatmaps show log2 fold-change of cell type expression of the genes within each locus. The colors indicate the differential expression level. CNV-gene-gene-set networks in (c) and (f) display the CNV genes and their participation in the pathway-cell-type stratified gene sets. Shapes represent different entities of the network where the big circle in the middle is a GWS locus, peripheral circles are genes in the locus. A gene may be linked to one or more pathways (diamond) and at the end of the pathway, a cell type (square) is connected to indicate the gene membership of one or more stratified pathways of the same cell type. The color of diamond nodes indicates the group of pathways.