Opposing transcriptional programs of KLF5 and AR emerge during therapy for advanced prostate cancer

Abstract

Endocrine therapies for prostate cancer inhibit the androgen receptor (AR) transcription factor. In most cases, AR activity resumes during therapy and drives progression to castration-resistant prostate cancer (CRPC). However, therapy can also promote lineage plasticity and select for AR-independent phenotypes that are uniformly lethal. Here, we demonstrate the stem cell transcription factor Krüppel-like factor 5 (KLF5) is low or absent in prostate cancers prior to endocrine therapy, but induced in a subset of CRPC, including CRPC displaying lineage plasticity. KLF5 and AR physically interact on chromatin and drive opposing transcriptional programs, with KLF5 promoting cellular migration, anchorage-independent growth, and basal epithelial cell phenotypes. We identify ERBB2 as a point of transcriptional convergence displaying activation by KLF5 and repression by AR. ERBB2 inhibitors preferentially block KLF5-driven oncogenic phenotypes. These findings implicate KLF5 as an oncogene that can be upregulated in CRPC to oppose AR activities and promote lineage plasticity.

Fig. 1. KLF5 levels are upregulated in CRPC.

a KLF5 mRNA measured using quantitative RT-PCR in AR-positive LNCaP, VCaP, 22Rv1 and AR-negative DU145, NCI-H660 and PC-3 cell lines. n = 9, mean ± 95% CI from three biological replicates in technical triplicate. b Protein measured by western blot using the same cell lines as a. Actin is a loading control. c Representative immunohistochemistry (IHC) staining of benign, localized prostate cancer (hormone naïve) and CRPC tissues from patients or patient-derived xenografts (PDX) using an antibody specific for KLF5. d Quantification of KLF5 IHC staining in tissue microarrays by a genitourinary pathologist (PM). Bounds of boxes are lower/upper quartiles with median; whiskers show range from minima to maxima; dots are the average staining intensity of an individual case or PDX. n = 50 localized prostate cancer cases (n = 50/50 cases contained benign luminal cells, n = 47/50 cases contained benign basal cells, n = 47/50 slides contained cancer cells), n = 14 localized CRPC cases, and n = 29 CRPC PDXs. P-values are from two-sided Mann–Whitney U-tests. n.s. = not significant. e Klf5 mRNA measured using quantitative RT-PCR in organoids derived from intact Pten−/− mouse prostates maintained in androgen-replete (DHT, dihydrotestosterone) conditions or castrate Pten−/− mouse prostates maintained in androgen-depleted conditions. n = 6, mean ±95% CI from two biological replicates in technical triplicate. f KLF5 mRNA measured by RT-PCR in LNCaP cells and sub-lines derived from castration-resistant (16D) or castration/enzalutamide-resistant (49F and 42D) LNCaP xenograft tumors. n = 9, mean ±195% CI from three biological replicates in technical triplicate. g Protein measured by western blot using the same cell lines as in f. Tubulin is a loading control.

Fig. 2. CRPC models display increased durability of KLF5 induction by androgens.

a KLF5 mRNA measured by RT-PCR in LNCaP cells cultured in androgen-deplete medium supplemented with 1 nM DHT for indicated time-points. n = 6, mean ± 95% CI from 2 biological replicates in technical triplicate. b KLF5 protein measured by western blot using the same conditions as a. ERK2 is a loading control. c KLF5 mRNA measured by RT-PCR in VCaP cells as in a. n = 6, mean ± 95% CI from 2 biological replicates in technical triplicate. d KLF5 protein measured by western blot using the same conditions as in c. ERK2 is a loading control. e KLF5 protein measured by western blot in VCaP cells treated with 1 nM mibolerone (Mib) as in d. One additional replicate experiment was performed that yielded a comparable result. f KLF5 protein measured by western blot in C4-2B cells treated as in b. One additional replicate experiment was performed that yielded a comparable result. g KLF5 mRNA measured by RT-PCR in LNCaP and LNCaP-derived CRPC sub-lines 16D, 49F, and 42D cells treated as in a. n = 6, mean ± 95% CI from three biological replicates in technical duplicate. h Schematic of LNCaP cell culture conditions for i. enz = enzalutamide, DMSO = vehicle control. i KLF5 mRNA measured by RT-PCR in cells cultured as in i. n = 6, mean ± 95% CI from 2 biological replicates in technical triplicate. j KLF5 protein measured in R1-AD1 cells by western blot using the same conditions as a. One additional replicate experiment was performed that yielded a comparable result. k KLF5 and AR proteins measured by western blot using LNCaP cells cultured in androgen-deplete medium (CSS), LNCaP cells stimulated 8 h with 1 nM DHT as in a, or indicated cell lines grown in their respective standard medium conditions. Actin is loading control. One additional replicate experiment was performed that yielded a comparable result.

Fig. 3. The KLF5 and KLK2/3 super-enhancers display an inverse pattern of H3K27ac activation marks.

a, b Gene track views of H3K27ac and AR ChIP-seq data from AR-positive LNCaP and VCaP cell lines and the AR-negative PC-3 cell line at genomic loci for KLF5 and KLK2/KLK3. c, d Gene track views of H3K27ac from four clinical CRPC specimens at genomic loci for KLF5 and KLK2/KLK3. H3K27ac track heights were set using H3K27ac density at housekeeping gene enhancers shown in Supplementary Fig. 3.

Fig. 4. KLF5 promotes oncogenic phenotypes in prostate cancer cells.

a KLF5 protein measured by western blot using R1-AD1 cells infected with control shRNA lentivirus (shC) or two independent KLF5-targeted shRNA lentiviruses (shK1 and shK2). ERK2 is a loading control. b R1-AD1 cells as in a were analyzed by 3D soft agar colony formation assays in androgen-deplete medium supplemented with 1 nM DHT or vehicle control (0.1% ethanol). n = 8, mean ±95% CI from 2 independent experiments in biological quadruplicate (n = 8). c R1-AD1 cells transfected with KLF5-targeted siRNAs (siK1 and siK2) or control siRNA (siC) were analyzed by chemotactic migration assays. n = 6, mean ±95% CI, 2 independent experiments in biological triplicate. d KLF5 protein measured by western blot using R1-AD1 cells, LNCaP cells, and LNCaP cells infected with empty lentivirus or lentivirus encoding KLF5. ERK2 is a loading control. e LNCaP cells as in d were analyzed by 3D soft agar colony formation assays in an androgen-replete medium supplemented with enzalutamide or vehicle control (DMSO). n = 6, mean ±95% CI, 2 independent experiments in biological triplicate. f LNCaP cells as in d were analyzed by chemotactic migration assays. n = 4 or 6, mean ±95% CI, 2 independent experiments in biological duplicate or triplicate. P-values are from 2-sided, 2-tailed t-tests.

Fig. 5. KLF5 and AR display opposing regulation of oncogenic gene sets and genes defining basal cell identity.

a KLF5 peptides (green) identified in biological replicate AR-targeted RIME experiments using R1-AD1 cells cultured with 1 nM DHT or vehicle (0.1% ethanol) control. b RNA-seq experimental design for knockdown of KLF5 with two independent shRNAs targeting KLF5 (shKLF5-1 and −2) or control shRNA (shCTRL), cultured in an androgen-deplete medium, and treatment with 1 nM DHT or vehicle control (0.1% v/v ethanol, ETH) for 24 h. c Normalized enrichment scores for all 189 MSigDB oncogenic signatures derived from gene set enrichment analysis (GSEA) using R1-AD1 gene expression data reflecting AR activity (differential expression under DHT vs. ETH). d GSEA-derived normalized enrichment scores as in c using R1-AD1 gene expression data reflecting KLF5 activity (differential expression in shCTRL vs. shKLF5). Oncogenic Signatures are colored blue or red based on whether they were positively or negatively enriched in c with FDR < 0.01. e GSEA-derived normalized enrichment scores (NES) of luminal, basal, or neuroendocrine (NE) cell gene sets using R1-AD1 gene expression data reflecting KLF5 activity (differential expression in shCTRL vs. shKLF5, top) or AR activity (differential expression in DHT vs. ETH, bottom). f Schematic of longitudinal gene expression time-points during adenocarcinoma to neuroendocrine CRPC (NEPC) progression of LTL-331 patient-derived xenografts (PDX). g Z-scores of individual gene expression at time-points as in f. h Activity scores derived from the expression of gene sets distinguishing prostate epithelial cell types (same gene sets as e) assessed at time-points as in f. i Activity scores derived from the expression of gene sets reflecting KLF5 activity, AR activity, and cell cycle progression assessed at time-points as in f.

Fig. 6. Transcriptional opposition of AR and KLF5 converges at ErbB pathway genes and basal-luminal cell markers.

a Heatmaps of AR and KLF5 ChIP-seq signals ±3 kb around KLF5/AR common, KLF5 only, or AR only peaks in R1-AD1 cells cultured in an androgen-deplete medium treated 4 h with 1 nM DHT or vehicle control (ethanol, EtOH). b Heatmap of RNA-seq gene expression data for a set of 831 genes located ±50 kb from a KLF5/AR common peak and differentially expressed in R1-AD1 cells based on comparisons between shRNA control (shCTRL) vs. shKLF5 conditions and/or DHT vs. vehicle conditions. The heatmap was generated by unsupervised clustering, revealing eight main clusters. c Proteins measured by western blot in R1-AD1 cells transfected with KLF5-targeted siRNAs (siK1 and siK2) or control siRNA (siC) were cultured in androgen-deplete medium supplemented with 1 nM DHT or vehicle control (EtOH). Tubulin is a loading control. One additional replicate experiment was performed that yielded a comparable result. d Cytokeratin-5 (CK5), cytokeratin 8/18 (CK8/18), and KLF5 proteins measured by western blot in R1-AD1 cells transfected with siRNAs as in c. Tubulin is a loading control. One additional replicate experiment was performed that yielded a comparable result. e Proteins measured by western blot in LNCaP cells infected with lentivirus encoding KLF5 or empty vector and cultured in an androgen-deplete medium supplemented with 1 nM DHT or vehicle control (EtOH). Tubulin is a loading control. One additional replicate experiment was performed that yielded a comparable result.

Fig. 7. KLF5 retains oncogenic function and activates ERBB2 in an enzalutamide-resistant CRPC cell line model of AR-V activity.

a KLF5 protein measured by western blot using R1-D567 cells infected with control shRNA lentivirus (shC) or two independent KLF5-targeted shRNA lentiviruses (shK1 and shK2). ERK2 is a loading control. b R1-D567 cells as in a analyzed by 3D soft agar colony formation assays. n = 14, mean ± 95% CI from four independent experiments, two performed in biological triplicate and two performed in biological quadruplicate. P-values are unadjusted from 2-sided, 2-tailed t-tests. c R1-D567 cells transfected with KLF5-targeted siRNAs (siK1 and siK2) or control siRNA (siC) analyzed by chemotactic migration assays. n = 6, mean ±95% CI, two independent experiments in biological triplicate. P-values are unadjusted from 2-sided, 2-tailed t-tests. d Heatmaps of AR variant (AR-V) and KLF5 ChIP-seq signals ±3 kb around KLF5/AR common, KLF5 only, or AR only peaks in R1-D567 cells cultured in androgen-deplete medium. e Heatmap of RNA-seq gene expression data for a set of 79 genes located ±50 kb from a KLF5/AR common peak and differentially expressed in R1-D567 cells based on comparisons between shRNA control (shCTRL) vs. shKLF5 conditions and/or DHT vs. vehicle conditions. The heatmap was generated by unsupervised clustering, revealing two main clusters. f GSEA testing enrichment of the signature ERBB2_UP.V1_UP in R1-D567 gene expression data reflecting active KLF5 (differential expression in shCTRL vs. shKLF5).

Fig. 8. Targeting ERBB2 inhibits oncogenic effects of KLF5.

a Viability assays of patient-derived organoids developed from neuroendocrine CRPC (OWCM154 and 155) or CRPC adenocarcinoma (MSK-PCA3) treated with neratinib. Data are mean ±95% CI from three independent experiments in biological triplicate (n = 9). b, Lapatinib sensitivity scores predicted using ridge regression models trained on high-throughput cancer cell line drug screens in metastatic CRPC tumors (SU2C-EC study), plotted vs. KLF5 expression (FPKM = fragments per kb per million fragments mapped). Pearson correlation coefficients (R) and 2-tailed p-values for r ≠ 0 are shown. c Lapatinib sensitivity scores as in b plotted vs. KLF5 activity score. d–g 2-dimensional growth assays for R1-AD1 cells, R1-D567 cells, LNCaP cells infected with empty lentivirus (LNCaP-Vector), and LNCaP cells infected with lentivirus encoding KLF5 (LNCaP-KLF5) cultured in androgen-replete medium and treated for 5 days with mubritinib. Gray lines are biological replicates (n = 6), black lines are mean ±95% CI, IC50 values are mean ±95% CI. h, i LNCaP cells as in f and g were analyzed by 3D soft agar colony formation assays in androgen-replete medium supplemented with lapatinib or mubritinib. n = 4, mean ±95% CI, two independent experiments in biological duplicate. j LNCaP cells as in f and g analyzed by chemotactic migration assays with the medium in the top chamber containing 10 nM mubritinib or vehicle control (DMSO). n = 6, mean ±95% CI, two independent experiments in biological triplicate. P-values are from unpaired 2-sided t-tests.