Comparative transcriptomics reveals a mixed basal, club, and hillock epithelial cell identity in castration-resistant prostate cancer

Abstract

Inhibiting the androgen receptor (AR) is effective for treatment of advanced prostate cancers because of their AR-dependent luminal epithelial cell identity. Tumors progress during therapy to castration-resistant prostate cancer (CRPC) by restoring AR signaling and maintaining luminal identity or by converting through lineage plasticity to a neuroendocrine (NE) identity or double-negative CRPC (DNPC) lacking luminal or NE identities. Here, we show that DNPC cells express genes defining basal, club, and hillock epithelial cells from benign prostate. We identified KLF5 as a regulator of genes defining this mixed basal, club, and hillock cell identity in DNPC models. KLF5-mediated upregulation of RARG uncovered a DNPC sensitivity to growth inhibition by retinoic acid receptor agonists, which down-regulated KLF5 and up-regulated AR. These findings offer CRPC classifications based on prostate epithelial cell identities and nominate KLF5 and RARG as therapeutic targets for CRPC displaying a mixed basal, club, and hillock identity.

Keywords: KLF5; castration-resistant prostate cancer; cell identity; lineage plasticity.

Fig. 1.

Gene signatures defining basal, club, and hillock epithelial cell identities are coexpressed in CRPC. (A) Strategy used to derive scores for five BPECTs and four CRPCSTs by determining average expression of BPECT and CRPCST-defining genes in four bulk RNA-seq datasets generated from CRPC specimens. The number of CRPC specimens, their anatomic locations, and tissue collection method are indicated. (B) Correlation plots comparing BPECT scores in four CRPC bulk RNA-seq datasets. (C) Correlation plots comparing BPECT and CRPCST scores in the CRPC bulk RNA-seq datasets from (A). Indicated scores are plotted for each tumor within a dataset. P-values were adjusted using Bonferroni correction.

Fig. 2.

Gene signatures defining basal, club, and hillock cell identities are coexpressed in individual CRPC cells. (A) Strategy used to derive scores for five BPECTs and four CRPCSTs in a scRNA-seq dataset consisting of 27,338 tumor cells from 14 patient biopsies. (B) UMAP plots of tumor cells colored by sample ID. (C) UMAP plots of tumor cells as in (B) colored by histology of the biopsy. (D) UMAP plots of tumor cells as in (B) colored by the score for indicated BPECTs. (E) Correlation plots comparing BPECT scores in scRNA-seq data from 27,338 tumor cells. (F) UMAP plots of tumor cells as in (B) colored by the score for indicated CRPCSTs. (G) Correlation plots comparing BPECT and CRPCST scores in scRNA-seq data from 27,338 tumor cells. P-values were adjusted using Bonferroni correction.

Fig. 3.

A set of 23 transcription factors display gene expression correlating with a mixed basal, club, and hillock cell identity in CRPC. (A) Correlation plots showing the average Pearson correlation of transcription factors having a significantly positive correlation with one BPECT cell identity in at least 2 of the 4 CRPC datasets. The number of transcription factors with a significantly positive correlation with each BPECT identity is shown on the Left. (B) Upset plot showing overlap in transcription factors identified from (A). (C) Average Pearson correlation of 23 BCH TFs from (B) with BPECTs.

Fig. 4.

Cell line models of DNPC display a mixed basal, club, and hillock cell identity. (A) Violin plots illustrating z-scored expression of genes defining BPECTs using RNA-seq data from LNCaP Parental and APIPC cell lines treated with the synthetic androgen R1881 (1 nM) or vehicle control. Green lines represent median. (B) RT-qPCR data showing relative expression of epithelial cell identity markers in LNCaP Parental and APIPC cell lines. Bars represent mean ± 95% CI of three biological replicates. (C) Western blot of AR and indicated cell identity markers. Caret (^) indicates undetectable expression relative to NCI-H660 (positive control, see SI Appendix Fig. S6D). Tubulin is a loading control. (D) Models derived from the LTL331 PDX. (E) Violin plots illustrating z-scored expression of genes defining BPECTs using RNA-seq data from cells and tissues illustrated in (D). Green lines represent median. (F) RT-qPCR data showing relative expression of cell identity markers. Bars represent mean ± 95% CI of three biological replicates. (G) Western blot of AR and indicated cell identity markers. Tubulin is a loading control. For all statistical comparisons shown, ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, and ns = not significant.

Fig. 5.

KLF5 is nominated as a regulator of a mixed basal, club, and hillock cell identity in CRPC. (A) Relative mRNA expression of basal (KRT14), club (PI3), and hillock (KRT13) markers in LTL331CL cells transfected with control siRNA or pooled siRNAs targeting candidate BCH transcription factors. Bars represent mean of two biological replicates. (B) Growth of transfected LTL331CL cells from (A), measured by crystal violet staining 4 d posttransfection and plotted relative to growth of cells transfected with control siRNA. Bars represent mean ± 95% CI of nine biological replicates. (C) Violin plot showing AR and KLF5 activity scores derived from scRNA-seq data from 24,450 epithelial cells from benign prostate tissue. Statistics calculated from Bonferroni-Hochberg adjusted Kruskal–Wallis tests with Dunn’s post hoc analysis. (D) Violin plots illustrating z-scored expression of genes reflecting AR activity, KLF5 expression, and genes reflecting KLF5 activity in RNA-seq data from prostate cancer tissues collected in the DARANA trial. Green lines represent median. Statistics are from Student’s t tests. (E and F) Correlation plots comparing AR and KLF5 activity scores with BPECTs in E and CRPCSTs in F from a CRPC bulk RNA-seq dataset. P-values were adjusted using Bonferroni correction. For all statistical comparisons shown, ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, and ns = not significant.

Fig. 6.

KLF5 regulates epithelial cell identity markers in models of DNPC. (A) KLF5 and markers of cell identity were measured by western blot in LNCaP APIPC cells infected with control (empty) lentivirus or lentivirus expressing KLF5. Caret (^) indicates undetectable expression relative to NCI-H660 (positive control, see SI Appendix, Fig. S12B). Tubulin is loading control. (B) Expression of cell identity markers measured by RT-qPCR in cells infected as in A. P-values are from Student’s t tests. (C) KLF5 and cell identity markers were measured by western blot in LTL331CL cells transfected with control siRNA (C) or two independent siRNAs targeting KLF5 (K1 and K2). Caret (^) indicates undetectable expression relative to NCI-H660 (positive control, see SI Appendix, Fig. S13E). Tubulin is loading control. (D) mRNA expression levels of cell identity markers measured by RT-qPCR in cells transfected as in C. P-values are from Student’s t tests. (E) Violin plots illustrating z-scored expression of genes reflecting BPECTs in RNA-seq data from cells transfected as in C. Green lines represent median. P-values are from Student’s t tests. For all statistical comparisons shown, ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, and ns = not significant.

Fig. 7.

RAR signaling is a therapeutic target in models of DNPC (A) Correlation plots comparing AR and KLF5 activity with RARG and RARA mRNA expression in a CRPC bulk RNA-seq dataset. (B) Violin plots illustrating z-scored expression of RARA and RARG in RNA-seq data from the DARANA trial. Green lines represent median. (C–F) RAR-γ and KLF5 protein levels measured by western blot in indicated cell lines. APIPC cells were infected with control (empty) lentivirus or lentivirus expressing KLF5 in E. LTL331 cells were transfected with control siRNA (C) or two independent siRNAs targeting KLF5 (K1 and K2) in F. Tubulin is a loading control. (G) Pearson correlation of average IHC staining scores of KLF5 and RAR-γ in 40 LuCaP PDX tumors. (H) Growth curves from LNCaP Parental and APIPC cells treated with ATRA for 4 d. Dark lines represent mean of light lines for each biological replicate (n = 9). Black error bars representing 95% CI are present but smaller than data points at each dose. P-values are from Student’s t test comparing IC50 values. (I) Treatment of APIPC cells for 16 d in indicated cell culture medium. (J) Western blot of AR and KLF5 protein levels in APIPC cells maintained as in I. (K) Western blot of BPECT markers in APIPC cells treated with 0.5 or 10 µM ATRA (or DMSO control) for 16 d. Caret (^) indicates undetectable expression relative to NCI-H660 (positive control, see SI Appendix, Fig. S18A). (L) Maintenance of LTL331CL cells for 16 d in indicated cell culture medium before 6-d growth assay. (M) Cell growth measured by crystal violet staining of LTL331CL cells maintained as in L. Data are mean ± 95% CI of nine biological replicates. P-values were calculated using Student’s t test with Bonferroni adjustment. (N) Western blot of AR, KLF5, and BPECT markers in LTL331CL cells treated with 0.5 or 10 µM ATRA (or DMSO control) for 16 d. Caret (^) indicates undetectable expression relative to NCI-H660 (positive control, see SI Appendix, Fig. S18D). Tubulin is loading control. For all statistical comparisons shown, ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, and ns = not significant.