Fig. 2: NSD2 and H3K36me2 are upregulated in CRPC-NE and correlate with poor patient outcomes.

a, Immunostaining of H3K36me2 in NPPO-1 organoids. Images are shown with and without co-staining for VIM. Scale bars, 50 µm. b, Scatter plot comparing H3K36me2 immunostaining fluorescence intensity between neuroendocrine (NE) and non-neuroendocrine (Non-NE) cells in three replicate experiments. Data points indicate the mean ± s.d. (n = 55 (NE) or n = 26 (non-NE) cells). Mean fluorescence intensities were compared using unpaired t-tests (two-tailed). c, Dot plot of NSD1, NSD2 and NSD3 expression in a published scRNA-seq dataset²⁵. d,e, Western blots of the indicated histone marks and NSD2 and EZH2 proteins in neuroendocrine and non-neuroendocrine organoid lines (for source data, see Supplementary Fig. 1). f, Genome browser view of CUT&Tag signals for H3K36me2, H3K27me3 and H3K27ac together with bulk RNA-seq reads at Chga, Foxa2, Onecut2, Ascl1 and Mycn loci in the indicated organoid lines. Genomic position annotations are shown at the top. g,h, Analysis of NSD2 and H3K36me2 levels in a prostate cancer (PCa) TMA. Violin plots show the percentage of NSD2⁺ cells (g) and H3K36me2^high cells (h) in CHGA⁺ neuroendocrine tumour cells or AR⁺CHGA^– tumour cells in each patient. Data are expressed as median and interquartile (IQR) ranges (primary PCa CHGA^–, n = 33; de novo NEPC CHGA⁺, n = 6; mCRPC CHGA^–, n = 18; CRPC-NE CHGA⁺, n = 6). Welch’s analysis of variance (ANOVA) followed by Dunnett’s multiple comparison test in g; unpaired t-test (two-tailed) in h. i,j, Kaplan–Meier plots of overall survival from the time of CRPC biopsy based on NSD2 gene expression in bulk transcriptomes from independent mCRPC patient cohorts: RMH (i) (n = 28 out of 94) and PCF–SU2C (j) (n = 27 out of 141). The gene expression cut-off was determined using the optimized Maxstat method; P values were calculated using the log-rank test (two-sided).