Fig. 7: NDD gene mutants with shared behavioral phenotypes in zebrafish resolve unique and cell-type-specific gene-level convergent signatures and are rescued by predicted medications.
a, NDD risk genes uniquely cluster based on sleep–wake/visual-startle behavioral responses in zebrafish mutants. set 1: nrxn1a, mbd5, kdm5bab; set 2: phf12ab, skiab, chd2, smarcc2; set 3: kdm6bab, kmt5b, kmt2cab; set 4: wacab, arid1b, phf21aab, chd8, ash1l. b, Gene expression in human mature iGLUTs and iGABAs correlates with expression in the zebrafish brain. Cellular deconvolution of WT larval zebrafish brain expression based on adult human single-cell brain reference identifying neurons as the largest proportion of cells in the fish brain. Gene expression in WT zebrafish brain significantly positively correlates with gene expression of mature iGLUTs (two-tailed Pearson’s correlation with Holm’s multiple testing correction; R = 0.39, PHolm < 0.001) and iGABAs (R = 0.39, PHolm < 0.001). c, For each of the four behaviorally defined sets, gene-level convergence (DEGs with significant and shared direction of effect across all NDD genes within each of the four sets (P value-based DEG meta-analysis (METAL), FDR-adjusted Pmeta < 0.05, Cochran’s heterogeneity Q-test PHet > 0.05)) is largely nonoverlapping between mature iGLUTs and iGABAs, with unique enrichments for common psychiatric risk gene targets. Numbers of convergent genes that are upregulated (+) or downregulated (−) for each NDD set are indicated. d, In both iGABAs and iGLUTs, all four behavioral sets were enriched for FMRP targets. Gene targets of neurodevelopmental rare variants were significantly enriched for convergent signatures only in mature iGLUTs; behavioral set 4 uniquely significantly enriched for secondary targets of ASD LoF variants and set 3 uniquely enriched for primary targets of SCZ nonsynonymous variants. e, In iGLUTs, NDD-related behaviors were enriched only in sets 1 and 3, with enrichments for language, speech and intellectual delays in sets 1, 3 and 4. All sets were enriched for seizure and hypertonia. f, Potential ‘rescue’ drugs for these four phenotypic groups were selected from enrichment scores using CMap and filtered for drugs included in a screen of 376 compounds for behavioral effects in zebrafish. Top candidates that were significantly negatively enriched for iGLUT convergence from CMap and negatively correlated with mutant behavioral features were tested in mutant lines representative of sets 2–4. n.p. indicates that the drug repaglinide was not present in the CMap dataset. Mutant × drug combinations were as follows: chd2Δ7/Δ7 × pravastatin; kdm6bab F0 × paclitaxel; kdm6bab F0 × sirolimus; kmt5bΔ208,1i, Δ5/Δ208,1i, Δ5 × paclitaxel; kmt5bΔ208,1i, Δ5/Δ208,1i, Δ5 × sirolimus; ash1l1i, Δ60,19i/ 1i, Δ60,19i × ezetimibe; ash1l1i, Δ60,19i/ 1i, Δ60,19i × repaglinide; ash1l1i, Δ60,19i/ 1i, Δ60,19i × rosuvastatin; ash1l1i, Δ60,19i/ 1i, Δ60,19i × sunitinib; phf21aab F0 × amiodarone; phf21aab F0 × fluvoxamine. g, For behaviors that were significantly different between mutant + DMSO and WT + DMSO (P < 0.06), we characterized the effect of the mutant × drug on behavior as exacerbated (a) (significant effect mutant + drug-versus-WT > significant effect mutant-versus-WT), unchanged (b) (significant effect mutant + drug-versus-WT = significant effect mutant-versus-WT), partial rescue (c) (significant effect mutant + drug-versus-WT < effect mutant-versus-WT), rescued (d) (significant effect mutant-versus-WT, no significant effect mutant + drug-versus-WT) or over-corrected (e) (mutant + drug-versus-WT opposite direction of significant effect mutant-versus-WT). All drugs reversed at least one dysregulated behavior except for sirolimus in kmt5b. (i) Comparison of the magnitude of effect (beta, n = 24 parameters) on behavior between the mutant + DMSO compared with mutant + drug groups shows rescue of select behavioral features in kdm6b and chd2 mutants by paclitaxel (Shapiro–Wilk’s normality: W = 0.94301, P = 0.02121; two-sided Wilcoxon signed rank test statistic = 52, P = 0.0053, n = 24) and pravastatin (Shapiro–Wilk’s normality: W = 0.97587, P = 0.4203, two-sided paired Welch’s t-statistic = −3.533, P = 0.01394, n = 24), respectively. (ii) The phf21a mutant phenotype was strongly opposed by fluvoxamine (left: normality statistic = 0.93744, P = 0.01295; two-sided Wilcoxon signed rank test statistic = 196, P = 0.19; right: two-tailed Pearson’s correlation = −0.58, P = 0.0028, n = 24). In the box plots, the median (black line) and mean (red point) are shown. The lower and upper hinges correspond to 1st and 3rd quartiles. Upper and lower whiskers extend from the hinge to the largest or smallest value up to 1.5 × IQR. All data points are plotted individually. Illustrations in b created in BioRender; Townsley, K. https://biorender.com/rvk1zn2 (2026). AMIO, amiodarone; EZE, ezetimibe; FLUVO, fluvoxamine; PACLI, paclitaxel; PRA, pravastatin; REP, repaglinide; ROS, rosuvastatin; SIRO, sirolimus; SUN, sunitinib.
