Fig. 3: NSD2 targeting reverts neuroendocrine differentiation and restores AR expression.

a,c,e, H&E staining and immunofluorescence of sections from NPPO-1NE (a), NPPO-2 (c) and MSKPCa10 (e) organoids cultured in the absence of DHT after CRISPR-mediated knockout of Nsd2 (sgNsd2) or treatment with sgCtrl. BSD, blasticidin (drug-selection marker). Scale bars, 50 µm. b,d, Density plots for VIPER-analysed scRNA-seq data from NPPO-1NE (b) and NPPO-2 (d) organoids after Nsd2 knockout or treatment with sgCtrl. Stacked colour bars at left indicate proportion of cells in each cluster; colour scales at right indicate estimated probability density. f,g, Left, gene set enrichment analysis of pseudo-bulk snRNA-seq data showing enrichment for a canonical AR target signature⁵⁷ in NPPO-1NE organoids treated with sgCtrl (f) or sgNsd2 (g). Vertical blue lines (top) indicate the position of genes from the predefined set in the genes ranked from the least expressed to the most expressed in the organoid snRNA-seq sample (bottom). The enrichment score (ES) is shown on the y axis, and the normalized enrichment score (NES) with a nominal P value was determined from 1,000 random permutations of gene labels using permutation tests (one-sided). Right, DC projection of protein activity inferred from snRNA-seq data, showing gene expression enrichment for a canonical AR target signature⁵⁷. Cluster composition is shown as a stacked bar plot on the left; NES values indicating AR target gene enrichment are shown as a colour plot on the right from blue (negative) to red (positive). h, Growth curves for the indicated organoid lines in the absence or presence of DHT. Data points indicate the mean ± s.d. (n = 16 biological replicates). Organoids were cultured for three passages without DHT before treatment with 100 nM DHT or DMSO as a control. Data were analysed using two-way ANOVA and Tukey’s multiple comparison test.