Fig. 4: Distinct transcriptional signatures underlying sex-dependent rsFC effects in depression.

a Schematic representing partial least squares regression (PLS-R) analysis for uncovering spatial relationships between regional gene expression patterns and rsFC abnormalities in five transcriptomic ROIs (BA25, NAc, BA11, BA8/9, aIns). b–e PLS-R identified multivariate spatial correlations between gene expression and rsFC effects that were statistically significant and reproducible in cross-validation in four models: female BA25 (b), female NAc (c), male BA25 (d), and male BA8/9 (e). In each scatterplot, individual data points represent gene expression and rsFC effects for a given brain region, projected into a space defined by the first PLS component. The correlation for the first PLS component (ρ) and the mean correlation in held-out data across 10 iterations of 10-fold cross validation (ρ_ho) are depicted at the top. Statistical significance was evaluated compared to shuffled predictor data (spin test + FDR correction; see Online Methods). * = FDR q < 0.05. f, g Heatmaps depicting the loading weight (LW) Z-scores for the top 5 genes (in rows) with the strongest positive LWs (f) and the strongest negative LWs (g) in the PLS models depicted in (b–e), shown in columns. For comparison purposes, we also plot the corresponding models for the same three ROIs in the opposite sex, which were nominally significant (P_rand < 0.05, see Supplementary Fig. 5). Black tiles in the gray tables to the right of each heatmap denote membership of corresponding genes (labeled in rows) in selected gene sets relevant to MDD, as defined by a recent large-scale depression GWAS⁵² (“MDD GWAS”), the DisGeNet platform, or an RNA-seq study identifying differentially expressed genes in brain tissue in MDD patients (“MDD DEX”)⁸, or implicated in neurodevelopment, synapse function, or immune signaling as per the corresponding Gene Ontology Biological Process terms. h fGSEA results for differentially expressed genes in depression. In all four PLS models, genes predicting the spatial distribution of connectivity abnormalities in MDD are enriched for genes that show increased (green) or decreased (red) expression in a previously published RNA-seq analysis of brain tissue donated by MDD subjects of the corresponding sex⁸. fGSEA-generated normalized enrichment scores (x-axis), p-values (plotted in each bar), and adjusted p-values (darkened bar color if FDR q < 0.05) are plotted. Negative enrichment scores denote enrichment among genes with negative LWs in the PLS regression model, and positive enrichment scores denote enrichment among genes with positive LWs. i Histograms depicting the number of genes (y-axis and qualitative color gradient) with increased (left, green) or decreased (right, red) expression in males with MDD in each loading weight decile of our PLS-R ranked gene list for the male BA8/9 model. Error bars signify the 95% confidence interval of expected number of genes per decile in 10,000 simulations using random gene sets. j, k fGSEA results for depression-related risk genes as defined in DisGeNet database (j) and genes whose expression in nervous system tissue is modulated by significant SNPs from the most recent large-scale depression GWAS⁵² (k). l, m Gene Ontology (GO) analysis identified enrichments for GO Biological Process terms containing the corresponding phrases (listed in rows) at the (l) top and (m) bottom of LW-ranked gene lists in the four PLS models listed in (b–e). Significance of GO enrichment (i.e. -log(FDRq)) depicted in color scale.