Fig. 5: Gene signatures derived from single-cell RNA sequencing capture important features of prostate cancer models and stratify patients with advanced disease.

a Heatmap of single-cell gene signature gene set variation analysis (GSVA) enrichment scores in bulk RNA-sequencing of LNCaP treated with dihydrotestosterone (DHT) or enzalutamide (ENZ), and either sensitive or resistant to ENZ. b Heatmap of gene signature GSVA enrichment scores in bulk RNA sequencing from VCaP subline derivatives VCaP-T (long-term cultured with 10 µM testosterone), VCaP-CT (VCaP-T long term cultured with 0.1 nM testosterone), VCaP-CT-ET (VCaP-CT cultured long-term with 10 µM ENZ), VCaP-CT-BR (VCaP-CT cultured long term with bicalutamide), and VCaP-CT-BR-ER (VCaP-CT-BR long-term treated with ENZ upon reaching bicalutamide insensitivity). c Kaplan–Meier progression-free survival curves for Alumkal et al.² patients stratified into two groups based on median GSVA score for the Persist gene signature. The two-sided log-rank p-value is shown above the curves. d Kaplan–Meier progression-free survival curves for Alumkal et al. patients stratified into two groups based on median GSVA score for the NEPC-upregulated gene signature. The two-sided log-rank p-value is shown above the curves. e Kaplan–Meier overall survival curve for abiraterone (ABI)- and ENZ-naive patients from the Stand Up 2 Cancer (SU2C) CRPC cohort⁵ stratified into two groups based on median GSVA score for the Persist gene signature. The two-sided log-rank p-value is shown above the curve. f Summary table of gene signature GSVA score associations with progression-free survival (PFS) or overall survival (OS) in the clinical datasets. Only gene signatures significantly associated with PFS or OS in one or more datasets are shown. Good indicates that a higher score for the signature (a score higher than the median) is associated with better survival outcome, while poor indicates that a higher signature score is associated with worse survival outcome. Two-sided log-rank p-values are shown in parentheses. For each dataset, the header indicates the number of samples included, along with other qualifying information of the dataset. We used ABI/ENZ-naive patients from the capture-based SU2C CRPC RNA-seq dataset. See also Supplementary Fig. 5.