Targeting pre-existing club-like cells in prostate cancer potentiates androgen deprivation therapy

TL;DR

This study shows that targeting specific pre-existing cells in prostate cancer can improve the effectiveness of androgen deprivation therapy.

Contribution

The paper introduces a novel therapeutic strategy targeting castration-tolerant Club-like cells to enhance treatment outcomes in prostate cancer.

Findings

Castration tolerance in LSCmed cells arises from transcriptional reprogramming, not intrinsic properties.

Dual targeting of FOSL1/AP-1 and PIM kinases suppresses the growth of aggressive prostate cancer subtypes.

FOSL1 gene expression is elevated in stem-like CRPC and is upregulated by enzalutamide treatment in high-risk patients.

Abstract

A critical knowledge gap in prostate cancer research is understanding whether castration-tolerant progenitor-like cells that reside in treatment-naïve tumors play a direct role in therapy resistance and tumor progression. Herein, we reveal that the castration tolerance of LSCmed (Lin-, Sca-1+, CD49fmed) progenitor cells, the mouse equivalent of human prostatic Club cells, arises not from intrinsic properties, but from significant transcriptional reprogramming. Utilizing single-cell RNA sequencing of LSCmed cells isolated from prostate-specific Pten-deficient (Ptenpc−/−) mice, we identify the emergence of castration-resistant LSCmed cells enriched in stem-like features, driven by the transcription factor FOSL1/AP-1. We demonstrate that cells exhibiting Ptenpc−/− LSCmed characteristics are prevalent in aggressive double-negative prostate cancer (DNPC) subtypes recently identified in human castration-resistant prostate cancer (CRPC). Furthermore, our findings show that the dual-targeting agents JQ-1 and CX-6258—focused on FOSL1/AP-1 and PIM kinases, respectively—effectively suppress both the progenitor properties and the growth of mouse and human DNPC surrogates in vitro and in vivo. Thus, early eradication of castration-tolerant Club-like cells presents a promising therapeutic strategy to mitigate prostate cancer progression toward CRPC.

Although castration is the primary treatment for advanced prostate cancer, castration-resistant prostate cancer (CRPC) eventually develops in all patients. Here, we identify a therapeutic strategy that enhances the castration efficacy by targeting castration-tolerant cells in prostate tumors.

LSCmed cells, the mouse equivalent of human prostatic Club cells, exhibit high molecular similarity with human double negative prostate cancer subtypes lacking androgen receptor signature and neuroendocrine features (MSPC, CRPC-SCL).LSCmed cells respond to castration by a transcriptomic reprogramming orchestrated by the transcription factor FOSL1/AP-1, leading to increased aggressiveness.FOSL1 and FOSL1 target gene expression is elevated in stem-like-cell CRPC and is upregulated by neo-adjuvant enzalutamide monotherapy in patients with high-risk prostate cancer.Mouse and human Club-like cell proliferation, survival and progenitor properties, and CRPC growth in mice, are inhibited by combined pharmacological inhibition of FOSL1/AP-1 and PIM kinases.

LSCmed cells, the mouse equivalent of human prostatic Club cells, exhibit high molecular similarity with human double negative prostate cancer subtypes lacking androgen receptor signature and neuroendocrine features (MSPC, CRPC-SCL).

LSCmed cells respond to castration by a transcriptomic reprogramming orchestrated by the transcription factor FOSL1/AP-1, leading to increased aggressiveness.

FOSL1 and FOSL1 target gene expression is elevated in stem-like-cell CRPC and is upregulated by neo-adjuvant enzalutamide monotherapy in patients with high-risk prostate cancer.

Mouse and human Club-like cell proliferation, survival and progenitor properties, and CRPC growth in mice, are inhibited by combined pharmacological inhibition of FOSL1/AP-1 and PIM kinases.

Although castration is the primary treatment for advanced prostate cancer, castration-resistant prostate cancer (CRPC) eventually develops in all patients. Here, we identify a therapeutic strategy that enhances the castration efficacy by targeting castration-tolerant cells in prostate tumors.

Figures

Transcriptomic heterogeneity of *Pten*^*pc−/−*^ LSC^med^ cells.(A) The UMAP analysis (unsupervised Louvain-clustering) of *Pten*^*pc−/−*^ LSC^med^ cells analyzed by scRNAseq identified three distinct LSC^med^ cell clusters named LSC^med^-0, LSC^med^-1 and LSC^med^-2. (B) Distribution of the three LSC^med^ cell subpopulations in prostates from intact *versus* castrated mice. (C) The AUCell score of the AR signature (Sackmann Sala et al, [2017](#CR81)) in the three LSC^med^ cell subpopulations is shown. p* < 0.05; p* < 0.001; *p* < 0.0001 (Wilcoxon test, Holm-method adjusted). Exact *p* values:

Transcriptional plasticity dynamics of *Pten*^*pc−/−*^ LSC^med^ cells after castration.(A) Schematic representation of positive selection (upper panel) versus transcriptional reprogramming (lower panel) as two non-mutually exclusive mechanisms that could promote amplification of LSC^med^-1 and LSC^med^-2 at the expense of LSC^med^-0 cell subpopulations after castration (see text for explanations). (B, C) Images (B) and proportion of Ki-67-positive cells (C) quantified from immunohistochemistry (scale bar: 250 μm) in prostates from intact *Pten*^*pc−/−*^ mice *versus* 5 days, 21 days and 2 mont

FOSL1/AP-1 complex is a major transcriptional hub for *Pten*^*pc−/−*^ LSC^med^ cells emerging post-castration.(A–D) Identification of the main upstream regulators of the three LSC^med^ cell subpopulations by SCENIC analysis. The top-20 (fold change) activated (red) and inactivated (blue) transcription factors in LSC^med^-0 (A), LSC^med^-1 (B) et LSC^med^-2 (D) cells are represented as waterfall plots. The red arrow in B identifies *Fosl1*. The transcription factor network of the three most pertinent regulons of LSC^med^-1 cells is represented in (C) (see Fig. [EV3](#Fig9) for LSC^med^-0 and LS

Targeting FOSL1/AP-1 and Pim family prevents the growth of LSC^med^ and human Club-like cells in vitro.(A, B) Expression per *Pten*^*pc−/−*^ LSC^med^ cell subpopulation (A) and UMAP projection (B) of *Pim1*. (C, D) Number (C) and size (D) of organoids generated from sorted *Pten*^*pc−/−*^ LSC^med^ cells after 10 days of culture in medium containing DMSO or 1 nM JQ-1 and 1 nM CX-6258, alone or combined, as indicated. Data were normalized to the control (DMSO) condition (biological replicates, *n* = 5 independent experiments). *p* < 0.01; p* < 0.001; *p* < 0.0001 (one-way ANOVA and Dunnett post

Transcriptomic heterogeneity of *Pten*^*pc−/−*^ LSC^med^ cells.(A) The UMAP analysis (unsupervised Louvain-clustering) of *Pten*^*pc−/−*^ LSC^med^ cells analyzed by scRNAseq identified three distinct LSC^med^ cell clusters named LSC^med^-0, LSC^med^-1 and LSC^med^-2. (B) Distribution of the three LSC^med^ cell subpopulations in prostates from intact *versus* castrated mice. (C) The AUCell score of the AR signature (Sackmann Sala et al, [2017](#CR81)) in the three LSC^med^ cell subpopulations is shown. p* < 0.05; p* < 0.001; *p* < 0.0001 (Wilcoxon test, Holm-method adjusted). Exact *p* values:

Transcriptional plasticity dynamics of *Pten*^*pc−/−*^ LSC^med^ cells after castration.(A) Schematic representation of positive selection (upper panel) versus transcriptional reprogramming (lower panel) as two non-mutually exclusive mechanisms that could promote amplification of LSC^med^-1 and LSC^med^-2 at the expense of LSC^med^-0 cell subpopulations after castration (see text for explanations). (B, C) Images (B) and proportion of Ki-67-positive cells (C) quantified from immunohistochemistry (scale bar: 250 μm) in prostates from intact *Pten*^*pc−/−*^ mice *versus* 5 days, 21 days and 2 mont

FOSL1/AP-1 complex is a major transcriptional hub for *Pten*^*pc−/−*^ LSC^med^ cells emerging post-castration.(A–D) Identification of the main upstream regulators of the three LSC^med^ cell subpopulations by SCENIC analysis. The top-20 (fold change) activated (red) and inactivated (blue) transcription factors in LSC^med^-0 (A), LSC^med^-1 (B) et LSC^med^-2 (D) cells are represented as waterfall plots. The red arrow in B identifies *Fosl1*. The transcription factor network of the three most pertinent regulons of LSC^med^-1 cells is represented in (C) (see Fig. [EV3](#Fig9) for LSC^med^-0 and LS

Targeting FOSL1/AP-1 and Pim family prevents the growth of LSC^med^ and human Club-like cells in vitro.(A, B) Expression per *Pten*^*pc−/−*^ LSC^med^ cell subpopulation (A) and UMAP projection (B) of *Pim1*. (C, D) Number (C) and size (D) of organoids generated from sorted *Pten*^*pc−/−*^ LSC^med^ cells after 10 days of culture in medium containing DMSO or 1 nM JQ-1 and 1 nM CX-6258, alone or combined, as indicated. Data were normalized to the control (DMSO) condition (biological replicates, *n* = 5 independent experiments). *p* < 0.01; p* < 0.001; *p* < 0.0001 (one-way ANOVA and Dunnett post

Combined JQ-1 and CX-6258 treatment reduces CRPC growth and aggressiveness in vivo.(A) Timeline representation of the experimental protocol used for *Pten*^*pc−/−*^ mouse castration and pharmacological treatments. Castrated mice received vehicle (*n* = 8) or daily JQ-1 intraperitoneal injection and biweekly CX-6258 oral administration (*n* = 8) during 28 days. (B) After 28 days of treatment, prostates were microdissected and weighed. *p* < 0.01 (unpaired two-tailed *t*-test with Welch’s correction). Exact *p* value: *p* = 0.0075. (C) For all mice, prostate histological structures were classifi

Full text

The paper explainedProblemThe major challenge in prostate cancer treatment is to overcome resistance to castration therapy that ultimately leads to patient death. The mechanisms that drive the shift of tumors from castration-sensitivity to castration-resistance remain poorly understood. Club cells are a newly identified progenitor-like cell type predicted to be castration-resistant. While these cells have been identified in prostate tumors prior to castration, their functional contribution to therapy resistance is unexplored.ResultIn this study, we used a mouse model of castration-resistant prostate cancer (prostate-specific inactivation of the tumor suppressor gene Pten) that is enriched in Club-like cells. We report that the castration tolerance of these cells is not intrinsic, but results from a transcriptional shift that enhances several hallmarks of cancer aggressiveness, such as epithelial-to-mesenchymal transition, basal cell features and stem-like programs. We identified the transcription factor FOSL1/AP-1 as a major driver of this process, called cell plasticity. Dual pharmacological targeting of FOSL1 and PIM1, a serine/threonine kinase known to promote prostate cancer, abrogates the in vitro progenitor properties of the various mouse and human Club-like models that we used, and this was correlated to a marked reduction of in vivo tumor growth after castration.ImpactOur findings show that targeting stemness and tumor-initiating capacities of Club-like prostatic cells by dual inhibition of FOSL1/AP-1 and PIM kinase is a promising strategy to potentiate androgen deprivation therapy and mitigate prostate cancer relapse.

Introduction

Prostate cancer (PCa) is the second most frequent male cancer worldwide and the fifth in terms of death (Bray et al, 2018). PCa is androgen-dependent, and the gold-standard treatment of advanced PCa is androgen deprivation therapy (ADT), i.e., chemical castration. A clinical response is initially observed in virtually all patients, reflecting the death of androgen-dependent tumoral cells. Within 1 to 3 years, however, castration-resistant PCa (CRPC) develops in most patients, leading to metastatic dissemination to distant sites such as bone and liver (Bubendorf et al, 2000). Although CRPC can be further treated with second-generation androgen receptor (AR) pathway inhibitors (ARPI, e.g., abiraterone, enzalutamide, apalutamide, and darolutamide), tumors eventually become resistant within a few months of treatment (Wang et al, 2021), which has been linked to the emergence of aggressive forms of CRPC, such as mesenchymal stem-like PCa (MSPC) and neuroendocrine PCa (NEPC) (Han et al, 2022). JAK/STAT and FGFR signaling (Chan et al, 2022), FOSL1/YAP (Tang et al, 2022), ONECUT2 (Qian et al, 2024), FOXA1 (Eyunni et al, 2025) and FOXA2 (Wang et al, 2024) were recently identified as drivers of these aggressive forms of CRPC via lineage plasticity, opening new hopes for the treatment of advanced stages of the disease. Therapeutic interventions earlier in the course of the disease to eradicate tumor cells that drive the shift from hormone-sensitive tumors to CRPC may improve the prognosis of PCa patients. Intriguingly, the potential contribution of intrinsic androgen-independent prostatic cells that preexist in naïve prostate tumors is ill-explored.

Club and Hillock cells have been recently identified by single-cell RNA sequencing (scRNAseq) as epithelial cells abundant in the prostatic urethra and collecting ducts, but rare in the healthy human prostate (Henry et al, 2018). Club-like cells have also been identified in treatment-naïve primary prostate tumors using scRNAseq and spatial transcriptomics (Chen et al, 2021; Kiviaho et al, 2024; Song et al, 2022). These two cell types exhibit very similar transcriptomic profiles and can be discriminated by the differential expression of lineage markers reflecting a more basal (Hillock) versus luminal (Club) phenotype (Henry et al, 2018). Club/Hillock cells are enriched in stem/progenitor-like programs and exhibit low androgen-signaling, hence they are predicted to be ADT-tolerant (Baures et al, 2022a; Henry et al, 2018). Accordingly, the transcriptomic signatures of Club/Hillock cells are enriched in the MSPC molecular subtype of PCa (Han et al, 2022). MSPC largely overlaps with the stem-cell-like (SCL) molecular subtype described by others (Tang et al, 2022), both of which are surrogates for double-negative PCa (DNPC) characterized by reduced AR expression signature and no neuroendocrine features (Bluemn et al, 2017). Although MSPC is frequently observed at the CRPC stage, i.e., post-castration (Han et al, 2022), analysis of The Cancer Genome Atlas (TCGA), CPC-GENE (Fraser et al, 2017) and DFKZ (Gerhauser et al, 2018) datasets reveals that 27–74% of treatment-naïve localized tumors exhibit mixed MSPC and AR-positive PCa (ARPC) features (Han et al, 2022). Moreover, up to 11% of the tumors in the early-onset (<55 years of age) DFKZ dataset were identified as MSPC subtype (Han et al, 2022). These de novo MSPC tumors are associated with enrichment for PTEN deletions, more advanced pathology (Gleason score, T/N stages) and premature patient death after treatment compared to ARPC or mixed ARPC/MSPC tumors. The link between biallelic PTEN loss and DNPC subtype (57 versus 17% frequency in other subtypes, p = 0.031) has been recently confirmed (Lundberg et al, 2023). Together, these findings converge on the question of whether Club cells preexisting in treatment-naïve tumors may also contribute to therapeutic resistance and tumor progression towards aggressive forms of CRPC. To our knowledge, however, these cells have not been functionally characterized.

The Pten^pc−/−^ mouse is an appropriate preclinical model to address these key questions as its developing prostatic tumors are highly populated by the mouse equivalent of human prostatic Club/Hillock cells termed LSC^med^ (Lin^-^, Sca-1^+^, CD49f^med^) cells (Baures et al, 2022a; Baures et al, 2022b; Sackmann Sala et al, 2017). In this acknowledged mouse model of CRPC, the tumor suppressor gene Pten is selectively inactivated in prostatic luminal epithelium, recapitulating one of the most frequent mutations observed in the human disease (Mulholland et al, 2011; Wang et al, 2003), especially in the DNPC subtype (Han et al, 2022; Lundberg et al, 2023). According to their low AR-signaling profile, the proliferation rate of Pten^pc−/−^ LSC^med^ cells is unaffected by castration (Sackmann Sala et al, 2017). LSC^med^ cells are enriched in stem/progenitor features (Baures et al, 2022a). Accordingly, they exhibit tumorsphere-/organoid-forming capacity in vitro (Baures et al, 2022b; Sackmann Sala et al, 2017) and tumor-initiating capacity both in situ (Guo et al, 2020) and in graft assays (Sackmann Sala et al, 2017). Together, these findings support a key role of these Club-like cells in castration resistance and progression towards CRPC (for a review, Baures et al, 2022a).

In this study, we show that Pten^pc−/−^ LSC^med^ cells are not intrinsically castration-tolerant, but undergo transcriptomic reprogramming after androgen deprivation that enhances their aggressiveness. Using scRNAseq specifically applied to Pten^pc−/−^ LSC^med^ cells sorted from intact versus castrated mice, we show that castration favors the emergence of a LSC^med^ cell subpopulation enriched in stem-like features. We identified the transcription factor FOSL1/AP-1 as the major driver of this transcriptomic phenotype switch. Combined pharmacological inhibition of FOSL1/AP-1 and PIM1 kinase abrogated the progenitor properties of Pten^pc−/−^ LSC^med^ cells and markedly reduced the growth of CRPC in castrated Pten^pc−/−^ mice. In line with the transcriptional resemblance of Pten^pc−/−^ mouse LSC^med^ cells and the human DNPC subtype, our drug combination also reduced the growth of human MSPC-like PC-3 cells xenografted in castrated host mice. These results suggest that early eradication of castration-tolerant Club-like cells presents a promising therapeutic strategy to mitigate PCa progression toward CRPC.

Results

Ptenpc−/− LSCmed cells share transcriptomic features with human MSPC/SCL subtypes

LSC^med^ cells isolated from Pten^pc−/−^ mice exhibit castration tolerance in vivo and form organoids in an androgen-independent manner (Baures et al, 2022b; Sackmann Sala et al, 2017). To investigate whether castration impacts the fate of LSC^med^ cells, we conducted Smart-seq2-based scRNAseq analysis of 384 LSC^med^ sorted cells from dissociated prostates of intact and 2-month-castrated mice (Fig. 1A; Appendix Fig. S1A). After selecting for high-quality single-cell-transcriptomes, we performed dimension reduction using uniform manifold approximation and projection (UMAP) and annotated the resulting single-cell space according to the corresponding castration profiles (Fig. 1B). The identity of the analyzed cells was assessed by the mRNA expression of the LSC^med^ marker Krt4 (Sackmann Sala et al, 2017) and of the surface markers used in cell sorting (Appendix Fig. S1B–E). The intrinsically low AR signaling of LSC^med^ cells (Sackmann Sala et al, 2017) was further reduced after castration (Fig. 1C). Differential gene expression (DEG) analysis identified 206 genes that exhibited at least a twofold (adj. pval <0.001) difference after castration (Dataset EV1). The most discriminative genes (based on adj. pval and fold change (FC)) per experimental condition were plotted as heatmap (Fig. 1D). Notably, post-castration, we observed the decreased expression of androgen-regulated genes (Fxyd3, Mme) (Mevel et al, 2020), and concomitant increased expression of Krt13, a Hillock cell marker (Henry et al, 2018), and Krt6b which is part of the squamous differentiation signature that has been associated with DNPC (Labrecque et al, 2019; Lundberg et al, 2023). Treatment-induced ARPC to DNPC-squamous conversion has been proposed to be one potential pathway to bypass AR-suppression strategies (Labrecque et al, 2019). Finally, various stem-related genes (Lgr4, Klf4, Tgfb2, Ly6d) and inflammatory genes (Ifi202b, Cxcl5) were also upregulated. This first experiment shows that Pten^pc−/−^ LSC^med^ cells are actually sensitive to mouse castration.Figure 1Pten^pc−/−^ LSC^med^ cells as a robust surrogate of human MSPC subtype.(A) Schematic representation of the LSC^med^ scRNAseq protocol. Two 6-month-old Pten^pc−/−^ mice, castrated or not at 4 months of age (Day 120), were used in each group and pooled before scRNAseq analysis. (B) UMAP space of Pten^pc−/−^ LSC^med^ cells colored according to the castration status. (C) AR signature expression intensities (AUCell score) (Sackmann Sala et al, 2017) in LSC^med^ cells per Pten^pc−/−^ mouse, stratified by castration. p = 2.18e-15 (Wilcoxon test, Holm-method adjusted), n = 166 (naïve), n = 140 (castrated). Boxplots within a violin plot display median (line), 25th–75th percentiles (box), whiskers at 1.5×IQR, and individual outliers. (D) The most discriminative (FC and adj. pval) genes of LSC^med^ cells from intact versus castrated Pten^pc−/−^ mice are shown. (E,** F**) The AUCell scores of the MSPC/ARPC/NEPC (Han et al, 2022) (E) and CRPC-WNT/-SCL/-AR/-NE (Tang et al, 2022) (F) human CRPC subtype signatures in LSC^med^ cells per Pten^pc−/−^ mouse, stratified by castration. ns not significant; **p < 0.01; ***p < 0.001; ****p < 0.0001 (Wilcoxon test, Holm-method adjusted), n = 166 (naïve), n = 140 (castrated). Exact p values: p = 0.48 (ARPC), p = 0.00066 (MSPC), p = 0.00017 (NEPC), p = 0.57 (CRPC-AR), p = 0.0038 (CRPC-SCL), p = 0.0058 (CRPC-WNT), p = 0.88 (CRPC-NE).

We then compared the signatures of Pten^pc−/−^ LSC^med^ cells in intact and castrated conditions to recently reported signatures of human PCa subtypes (Dataset EV2) (Han et al, 2022; Tang et al, 2022). This analysis revealed that MSPC (Han et al, 2022) and its counterpart in the Tang et al study (Tang et al, 2022), CRPC-SCL, are markedly enriched in typical LSC^med^ cell genes compared to other PCa subtypes (Fig. 1E,F). The enrichment was further enhanced when using post-castration LSC^med^ cell signatures. This was mainly due to the upregulation of genes such as Anxa1, Tm4sf1, Ltbp1, Msn, Lgals3, Arhgdib (for the MSPC phenotype) and Plat, Sdc4, Cstb, Cd44, Plau, Epha2, S100a14, Pttg1ip and Hif1a (for the SCL phenotype).

Together, these results highlight that Pten^pc−/−^ LSC^med^ cells are not intrinsically non-responsive to castration and exhibit high molecular similarity with human MSPC/SCL tumors, i.e., DNPC subtypes that are enriched in stemness features and associated with ADT-resistance and metastatic potential.

Castration strengthens the stem-like characteristics of one Ptenpc−/− LSCmed cell subpopulation

To better characterize the impact of castration on Pten^pc−/−^ LSC^med^ cells, we conducted unsupervised Louvain-clustering of these cells and identified three subpopulations (Fig. 2A; Dataset EV3). Hereafter, these three LSC^med^ clusters are referred to as LSC^med^-0, LSC^med^-1 and LSC^med^-2. The LSC^med^-0 subpopulation, highly predominant before castration, was drastically reduced after castration when LSC^med^-1 subpopulation became predominant (16-fold increased) (Fig. 2B). LSC^med^-2 subpopulation exhibited a more modest increase (threefold) after castration (Fig. 2B). RT-qPCR analysis of selected subpopulation markers in LSC^med^ cells sorted from intact and 2-month-castrated Pten^pc−/−^ mice confirmed the opposite regulation of LSC^med^-0 versus LSC^med^-1 and LSC^med^-2 genes. Individually, several marker genes of LSC^med^-0 (e.g., Spink5, Ly6c1) and LSC^med^-1 (e.g., Bcar1, Fosl1, Pim1, F3) cells exhibited very homogenous response amplitude, which validated the conclusions of scRNAseq experiments (Fig. EV1A,B). The amplitude of LSC^med^-2 gene upregulation was more heterogeneous (Fig. EV1C). According to the observation reported above for the bulk LSC^med^ cell pool (Fig. 1C), AR signaling was lower in LSC^med^-1 cells, and to a lesser extent, in LSC^med^-2 cells, compared to LSC^med^-0 cells (Fig. 2C).Figure 2. Transcriptomic heterogeneity of Pten^pc−/−^ LSC^med^ cells.(A) The UMAP analysis (unsupervised Louvain-clustering) of Pten^pc−/−^ LSC^med^ cells analyzed by scRNAseq identified three distinct LSC^med^ cell clusters named LSC^med^-0, LSC^med^-1 and LSC^med^-2. (B) Distribution of the three LSC^med^ cell subpopulations in prostates from intact versus castrated mice. (C) The AUCell score of the AR signature (Sackmann Sala et al, 2017) in the three LSC^med^ cell subpopulations is shown. *p < 0.05; ***p < 0.001; ****p < 0.0001 (Wilcoxon test, Holm-method adjusted). Exact p values: p = 6e-18 (0 versus 1), 0.017 (0 versus 2), p = 0.00017 (1 versus 2), n = 140 (cluster 0), n = 130 (cluster 1), n = 28 (cluster 2). (D) Discriminative marker genes (top 15) of the three LSC^med^ cell subpopulations. (E–I) Representation of the AUCell scores of the luminal (E) and basal (F) signatures (Joseph et al, 2020b) per castration status (n = 166 (naïve), n = 140 (castrated)), and of the Club (G) and Hillock (H) signatures (Henry et al, 2018), and score of early EMT (I) signature (Meyer-Schaller et al, 2019) per LSC^med^ cell subpopulations. ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 (Wilcoxon test, Holm-method adjusted). Exact p values: luminal p = 0.017; basal p = 2.9e-05; Club p = 0.49 (0 vs 1), p = 0.00085 (0 vs 2), p = 0.0072 (1 vs 2); Hillock p = 1.3e-10 (0 vs 1), p = 7.3e-09 (0 vs 2), p = 0.021 (1 vs 2); early EMT p = 5.4 e-14 (0 vs 1), p = 0.12 (0 vs 2), p = 0.00076 (1 vs 2), n = 140 (cluster 0), n = 130 (cluster 1), n = 28 (cluster 2). Boxplots within violin plots (E–H) display median (line), 25th–75th percentiles (box), whiskers at 1.5×IQR, and individual outliers. (J) StemChecker analysis in each LSC^med^ cell subpopulations (−log (adj. p values) are represented). (K) Organoid-forming capacity of LSC^med^ cells sorted from 10-month-old Pten^pc−/−^ mice, intact versus 2-month-castrated (biological replicates, n = 3 independent experiments). ns not significant (p = 0,35; unpaired t-test with Welch’s correction). Error bars represent SD. Source data are available online for this figure.

LSC^med^-0 cells express Psca, a well-described androgen-dependent marker in both murine and human luminal progenitors of the prostate (Crowell et al, 2019) (red arrow on Fig. 2D; Dataset EV3). Gene Ontology (GO) enrichment analysis highlighted hallmarks of protein secretion, oxidative phosphorylation, DNA repair and prostate development (Fig. EV1D,E). With the exception of MYC and E2F targets, most of the pathways in the Hallmark 50 gene dataset were less enriched in LSC^med^-0 cells compared to the two other subpopulations (Fig. EV1D).

LSC^med^-1 cells represent the most amplified subpopulation after castration. It expresses Ly6d (red arrow on Fig. 2D; Dataset EV3), a marker of castration-resistant luminal progenitors that is detected in primary PCa, amplified in mCRPC and associated with biochemical relapse and decreased overall survival (Barros-Silva et al, 2018; Steiner et al, 2023). Interestingly, we also noted the upregulation of Krt5 (red arrow on Fig. 2D; Dataset EV3), associated with an increase in the global basal signature and a slight decrease in the luminal score in response to castration (Fig. 2E,F), as previously observed in human MSPC (Han et al, 2022). The comparison of the transcriptomic signature (top 50 genes) of Pten^pc−/−^ LSC^med^-1 cells with scRNAseq data of benign epithelial cells (Data Ref: Crowley et al, 2020a; Crowley et al, 2020b; Data ref: Joseph et al, 2020a; Joseph et al, 2020b) confirmed the enrichment of basal markers at the expense of luminal markers, concomitant with the strengthening of the identity of luminal progenitors (LumP) (Fig. EV1H). According to the enrichment in basal features, LSC^med^-1 cells showed a stronger Hillock cell profile, while the Club cell profile was unchanged (Fig. 2G,H). LSC^med^-1 cells were also associated with enrichment in basal breast cancer signature, as well as in other oncogenic pathways (e.g., thyroid, pancreatic and lung cancer) (Fig. EV1F). Typical features of human MSPC/SCL subtypes (Han et al, 2022; Tang et al, 2022), including EMT (Fig. 2I), migration and stem signatures (Fig. EV1D,F), were also enriched in LSC^med^-1 cells. The enrichment of stemness features was further highlighted by StemChecker analysis (Fig. 2J). This analysis allows the association of any input gene list with a curated stemness signature database by calculating enrichment values based on shared genes (hypergeometric test-based-log adj. p value) (Pinto et al, 2015). The StemChecker analysis indicated a higher stemness-associated transcriptional identity of LSC^med^-1 cells, as suggested by the identification of various stemness-associated genes such as Klf4 or Cd44 (Dataset EV4). Accordingly, GO enrichment analysis highlighted several stemness-associated pathways, e.g., Wnt, Tgfβ, Notch, and Hedgehog signaling (green arrows in Fig. EV1D). The trend towards the higher organoid-forming capacity of LSC^med^ cells sorted from the prostates of 2-month-castrated mice (enriched in LSC^med^-1 cells) versus intact mice (enriched in LSC^med^-0 cells) agrees with stemness enrichment in LSC^med^ cells post-castration (Fig. 2K; Appendix Fig. S2). Together, the molecular features of LSC^med^-1 cells suggest increased aggressiveness of this particular subpopulation amplified following castration.

Finally, the most differentially expressed gene of LSC^med^-2 cells is Onecut2 (red arrow on Fig. 2D). Onecut2 is known to be a driver of neuroendocrine differentiation in PCa (Chan et al, 2022; Guo et al, 2019; Qian et al, 2024; Rotinen et al, 2018). In LSC^med^-2 cells, Onecut2 expression was associated with low enrichment (AUCell score values <0.05) of a neuroendocrine signature (Han et al, 2022) (Fig. EV1I), but typical neuroendocrine markers such as Chromogranin A and Synaptophysin were not detected in the LSC^med^-2 cell transcriptome. In contrast to LSC^med^-1 cells, LSC^med^-2 cells were not enriched in stemness-associated programs (Fig. 2J) and exhibited less basal features (Fig. EV1H). GO enrichment analysis highlighted the increased oxidative stress and hypoxia in LSC^med^-2 cells. Of note, the hypoxia-associated enzyme transglutaminase 2 (Tgm2), which is a marker of LSC^med^-2 cells (Dataset EV3), was recently identified as a marker of malignant progression (Abu El Maaty et al, 2022). This suggests that the LSC^med^-2 subpopulation could have the potential to promote cancer progression, which could be facilitated by various oncogenic pathways associated with tumor aggressiveness (Fig. EV1G).

Together, these results demonstrate that castration of Pten^pc−/−^ mice promotes the emergence of LSC^med^ cells presenting with a transcriptomic profile enriched in basal, EMT and stemness features and lineage plasticity drivers. All these features are known to correlate with cancer cell aggressiveness.

Transcriptomic plasticity of Ptenpc−/− LSCmed cells as an adaptive mechanism to castration

The transcriptomic analysis described above identified three LSC^med^ cell subpopulations whose ratios considerably evolve in response to castration (Fig. 2B). We hypothesized that this could result from a phenomenon of (i) positive selection, i.e., robust proliferation of LSC^med^-1 cells possibly associated to death of LSC^med^-0 cells, and/or (ii) cell plasticity, i.e., a transcriptional switch from LSC^med^-0 to LSC^med^-1 and LSC^med^-2 profiles (Fig. 3A). To address these hypotheses, we first quantified Ki-67 positive epithelial cells in various fields of prostates harvested from Pten^pc−/−^ mice sacrificed prior to castration or at 5, 21, or 60 days after castration (Appendix Fig. S3A). Confirming our previous findings (Sackmann Sala et al, 2017), cell proliferation was not altered following castration (Fig. 3B,C). The pattern of dead cells (TUNEL assay) was focal, and the global level remained particularly low (∼2.5%), with no significant variation after castration (Fig. 3D; Appendix Fig. S3B,C). In contrast, we showed by RT-qPCR that the expression of DEGs associated with each LSC^med^ cell subpopulation quickly evolved after castration. Downregulation of LSC^med^-0 markers (e.g., Hoxb13 and Spink5) and concomitant upregulation of LSC^med^-1 markers (e.g., Ifi202b and Tm4sf1) were detected as early as 5 days after castration (Fig. 3E,F). Although the pattern of LSC^med^-0 (down) versus LSC^med^-1 (up) markers displayed a clear time-dependent response, the variability observed between the three biological replicates was consistent with a dynamic process. For LSC^med^-2 cell markers (Fig. 3G), fold-changes were overall more modest (with the exception of Reg1), suggesting that a very low number of cells exhibited these features within 21 days post-castration.Figure 3. Transcriptional plasticity dynamics of Pten^pc−/−^ LSC^med^ cells after castration.(A) Schematic representation of positive selection (upper panel) versus transcriptional reprogramming (lower panel) as two non-mutually exclusive mechanisms that could promote amplification of LSC^med^-1 and LSC^med^-2 at the expense of LSC^med^-0 cell subpopulations after castration (see text for explanations). (B,** C**) Images (B) and proportion of Ki-67-positive cells (C) quantified from immunohistochemistry (scale bar: 250 μm) in prostates from intact Pten^pc−/−^ mice versus 5 days, 21 days and 2 months after castration, two mice per condition, 5 to 10 fields counted per animal (See Appendix Table S3A for total cell counts per animal). (D) TUNEL-positive cells in prostates from intact Pten^pc−/−^ mice versus 5 days, 21 days, and 2 months after castration, 3 to 4 mice per condition, 9 to 23 fields counted per animal (See Appendix Table S3B for total cell counts per animal). ns, not significant (Welch’s ANOVA with Brown–Forsythe correction, Dunnett’s T3 post hoc test). Exact p value: p = 0, 0.9469 (naive vs 5 days), p = 0.8017 (naive vs 21 days), p = 0.7613 (naive vs 2 months). (E–G) The expression of selected marker genes of LSC^med^-0 (E), LSC^med^-1 (F), and LSC^med^-2 (G) cells was measured by RT-qPCR in bulk LSC^med^ cells sorted from intact Pten^pc−/−^ mice and 5 and 21 days after castration. N = 3 independent experiments, each biological replicate corresponds to one to three pooled animals for a total of n = 5 (naive) and n = 7 (5 and 21 days). Data were normalized to the expression in intact mice. *p < 0.05; **p < 0.01; ***p < 0.001 (unpaired t-test with Welch’s correction). Exact p values are reported in Appendix Table S1. (H) RNA velocity analysis (scVelo) in Pten^pc−/−^ LSC^med^ cells analyzed by scRNAseq, based on the inference of directed dynamic information by leveraging splicing kinetics. (I–K) Pseudotime-ordering analysis (Monocle): projection of cells per cluster (I), castration status (J), and pseudotime (K). All error bars in this figure represent SD. Source data are available online for this figure.

The concomitance of early transcriptional changes with the unaltered rates of cell proliferation and cell death favors a transcriptional reprogramming trajectory from LSC^med^-0 towards LSC^med^-1 cells shortly after castration, while the emergence of LSC^med^-2 cells may occur at a later stage. RNA velocity analysis (scVelo), based on the inference of directed dynamic information by leveraging splicing kinetics, predicted LSC^med^-0 cells as a likely root cell population (Fig. 3H). This prediction was further supported by pseudotime-ordering analysis (Monocle) (Fig. 3I–K). The transcriptional heterogeneity of LSC^med^-0 and LSC^med^-1 cells was evident from the presence of “intermediate” cells distributed along the lower trajectory before castration (Fig. 3I–K). In contrast, the grouping of all LSC^med^-2 cells at the upper branch extremity suggests higher transcriptional homogeneity. Trajectory analyses predicted that post-castration LSC^med^-1 cells evolve from pre-castration LSC^med^-0 cells, while post-castration LSC^med^-2 cells may originate from LSC^med^-0 and/or LSC^med^-1 cells.

To support the bioinformatic and RT-qPCR evidence of the rapid emergence of LSC^med^-1 cells after castration, we used CD44 as a marker of this LSC^med^ subpopulation (Dataset EV3), and we analyzed its expression by immunohistochemistry. We observed that CD44 was significantly upregulated shortly (5 days) after castration (Fig. EV2A,B). Consistent with the luminal profile of CD44 expression (Fig. EV2B), co-immunostaining of CD44 and the LSC^med^ marker KRT4 was frequently observed (Fig. EV2C). CD44 is considered a general marker of stem/progenitor cells present in the prostate epithelium, including basal cells (Hurt et al, 2008; Tran et al, 2002; Zhou et al, 2024). Although CD44 and p63 co-immunostaining was occasionally observed (Fig. EV2D), the low abundance of basal cells relative to LSC^med^ cells in Pten^pc−/−^ mouse prostate (Fig. EV2C,D) supports the LSC^med^-1 identity of the majority of CD44+ epithelial cells.

These results collectively support the hypothesis that cellular plasticity underlies an adaptive resistance mechanism of LSC^med^ cells, which is initiated shortly after castration.

AP-1 complex is a major regulator of Ptenpc−/− LSCmed cells emerging post-castration

We next performed gene regulatory inference using SCENIC (Aibar et al, 2017) to identify transcriptional upstream regulators of the three LSC^med^ cell subpopulations. This analysis predicted Hoxb13 as a major regulator of LSC^med^-0 cells (Figs. 4A and EV3A). Hoxb13 was described as a key driver of the prostatic luminal cell differentiation (Huang et al, 2007). Other highlighted transcription factors regulating LSC^med^-0 cells included Ddit3, Bmyc, Nr2f2, Atf3, Grhl2, and Ehf (Figs. 4A and EV3A). DDIT3 was recently shown to be upregulated in CRPC patients and proposed to be associated with progression to CRPC (Jung et al, 2024). EHF is a transcription factor of the ETS family that controls epithelial cell differentiation, and its loss promotes prostate tumors enriched in EMT signature and mesenchymal/stem-like features (Albino et al, 2012). Finally, while Bmyc may be correlated to the upregulation of MYC targets (Fig. EV1D), Nr2f2 has been shown to cooperate with Pten deletion to promote malignant progression (Qin et al, 2013).Figure 4FOSL1/AP-1 complex is a major transcriptional hub for Pten^pc−/−^ LSC^med^ cells emerging post-castration.(A–D) Identification of the main upstream regulators of the three LSC^med^ cell subpopulations by SCENIC analysis. The top-20 (fold change) activated (red) and inactivated (blue) transcription factors in LSC^med^-0 (A), LSC^med^-1 (B) et LSC^med^-2 (D) cells are represented as waterfall plots. The red arrow in B identifies Fosl1. The transcription factor network of the three most pertinent regulons of LSC^med^-1 cells is represented in (C) (see Fig. EV3 for LSC^med^-0 and LSC^med^-2 cell transcription factor networks). (E,** F**) Fosl1 expression (E) and AUCell Score of FOSL1 regulon (F) in the three LSC^med^ cell subpopulations. ns, not significant; *p < 0.05; p < 0.0001 (Wilcoxon test, Holm-method adjusted). Exact p values: Fosl1: p = 3.7e-21 (0 vs 1), 1.3e-06 (0 vs 2), p = 0.11 (1 vs 2); Fosl1 regulon: p = 4.8e-26 (0 vs 1), 1e-07 (0 vs 2) and p = 0.031 (1 vs 2), n = 140 (cluster 0), n = 130 (cluster 1), n = 28 (cluster 2). (G) Correlation of Fosl1 expression and FOSL1 regulon in the three LSC^med^ cell subpopulations. (H) Projection of Fosl1 expression during pseudotime (pval = 5.01e-21, qval = 6.37e-19). (I, J) Expression of FOSL1 (I) and FOSL1 regulon (J) in SU2C patient cohort (Data Ref: Abida et al, 2019a), according to the tumor molecular subtype (n = 125 ARPC, n = 40 NEPC, n = 76 SCL, n = 25 WNT). Boxplots display median (line), 25th–75th percentiles (box), whiskers at 1.5×IQR, and individual outliers. ns, not significant; **p < 0.01; ***p < 0.001; ****p < 0.0001 (Kruskal–Wallis with Holm-corrected Wilcoxon post hoc tests). See Appendix Table S1 for exact p values.

Remarkably, the transcription factor profile of LSC^med^-1 cells (Fig. 4B) was the near-perfect inverse mirror image of that of LSC^med^-0 cells (Fig. 4A). This observation further supports that castration triggers cell plasticity of LSC^med^-0 towards LSC^med^-1 cells by switching on/off selected transcription factors. We identified Fosl1 as the most pertinent (based on −log10(pval)) regulator of the LSC^med^-1 cell signature (Fig. 4B). FOSL1 is a transcription factor that acts in the AP-1 complex mainly as a heterodimer involving Jun proteins (c-Jun, JunB, JunD) (Eferl and Wagner, 2003). Other typical transcription factors of LSC^med^-1 cells included Mafg, Klf13, Foxq1, Nfκb1, Pbx1, Ets1, and Fosb (Fig. 4B). Of note, both Mafg and Klf13 share several common target genes with  Fosl1 (Fig. 4C).

Finally, we identified Hnf4a as a master regulator of LSC^med^-2 cell signature (Figs. 4D and EV3B). HNF4a is a nuclear receptor known for its role as a tumor suppressor in the prostate by its ability to promote senescence in prostatic cells (Wang et al, 2020). HNF4a shares several target genes with Onecut2 (Fig. EV3B), suggesting a potential role of HNF4a in the cellular plasticity leading to the LSC^med^-2 signature. Nr1d2, another transcriptional regulator promoting lineage plasticity towards NEPC (He et al, 2021), was also identified as a top gene of LSC^med^-2 cells (Fig. 4D). The expression of Gata2 and Gata3 (Fig. 4D) suggests that LSC^med^-2 cells present a certain degree of luminal differentiation (Xiao et al, 2016).

Although Fosl1 was identified as the top gene of LSC^med^-1 cells (Fig. 4B), it was expressed at a similar level in LSC^med^-2 cells, and at consistent, though lower, level in LSC^med^-0 cells (Fig. 4E). The FOSL1-regulated network expectedly followed the same pattern (Fig. 4F,G). FOSL1/AP-1 is known to be involved in various processes, including stemness, EMT and cell plasticity (Eferl and Wagner, 2003; Feldker et al, 2020; Marques et al, 2021). In the mouse prostate, FOSL1/AP-1 was recently identified as one of the actors of differentiated (androgen-dependent) luminal cell plasticity in both healthy (Kirk et al, 2024) and cancer (Tang et al, 2022) contexts. To investigate further the potential role of FOSL1/AP-1 complex in Club-like cell plasticity upon castration, we plotted the Fosl1 expression in function of pseudotime. As illustrated in Fig. 4H, we observed that Fosl1 expression progressively increased post-castration. To support the bioinformatic (Fig. 4) and RT-qPCR (Fig. EV1B) evidence of FOSL1/AP-1 upregulation after castration, we analyzed FOSL1 expression by immunohistochemistry. In line with these data, the number of epithelial cells exhibiting FOSL1 nuclear staining gradually increased following castration (Fig. EV2E,F). The intensity of nuclear staining also tended to be higher after castration than before (Fig. EV2G). FOSL1 nuclear immunostaining was mainly localized in the luminal epithelium (Fig. EV2E). Accordingly, FOSL1+/KRT4+ cells (Fig. EV2H) and FOSL1+/CD44+ cells (Fig. EV2I) were frequently observed by co-immunostaining, whereas FOSL1+/p63+ cells were much less abundant (Fig. EV2J). Western blot analysis of FOSL1 expression in sorted LSC^med^ cells corroborated the trend toward upregulation of FOSL1 in this particular cell population after castration (Fig. EV2K). Together, these data confirmed the LSC^med^-1 identity of most FOSL1+ epithelial cells in Pten^pc−/−^ mouse prostate tumors, further suggesting that the FOSL1/AP-1 regulon is a key driver of post-castration transcriptional switch of LSC^med^ cells. Other AP-1 factors could contribute as their transcriptional networks were also highlighted by SCENIC analysis, albeit at lower significance (Atf3 in LSC^med^-0 cells, Fosl2 and Fosb in LSC^med^-1 cells, Fos in LSC^med^-2 cells) (Fig. 4A,B,D, and Dataset EV3). Strikingly, in human PCa, FOSL1 and its regulon are particularly associated with the MSPC/SCL subtype (Fig. 4I,J) (Data Ref: Abida et al, 2019a; Abida et al, 2019b). In fact, we noticed that the majority of AP-1 complex members (FOSL1, FOSL2, FOS, JUN, JUNB, ATF3, and* BATF*) are expressed at higher levels in MSPC/SCL than in the ARPC tumors (Figs. 4I,J and EV4A–F). Accordingly, consistent expression of several AP-1 complex members is observed in all three LSC^med^ cell subpopulations (Fig. EV4G–L). This data further highlights the relevance of Pten^pc−/−^ LSC^med^ cells as a surrogate of the aggressive DNPC subtype.

Together, these analyses show that members of the AP-1 complex, and in particular FOSL1, are predicted to be major drivers of the transcriptional switch occurring post-castration in LSC^med^/Club-like cells.

Pharmacological targeting of LSCmed and Club/Hillock-like cancer cells in vitro

We next used Pten^pc−/−^ LSC^med^ cells as a DNPC surrogate to identify relevant therapeutic targets for this incurable cancer subtype. Considering that LSC^med^-1 cells are the most amplified after castration (Fig. 2B), the least androgen-sensitive (Fig. 2C) and the most enriched in stemness features (Fig. 2J), we reasoned that targeting this particular cell subpopulation was the most relevant strategy to deviate LSC^med^ cells from progression towards CRPC in our mouse model. This hypothesis was strengthened by the observation that, in the recent DARANA study (Linder et al, 2022a; Data ref: Linder et al, 2022b), 3-month enzalutamide treatment of naïve high-risk PCa patients led to a marked enrichment of prostate tumors in LSC^med^-1 signature at the expense of LSC^med^-0 signature (Fig. EV5A), mimicking what we observed in Pten^pc−/−^ mice after castration (Fig. 2B). Based on the results shown in the previous section, FOSL1/AP-1 was identified as the most obvious therapeutic target to interfere with LSC^med^-1 cell emergence. The high expression of various members of the AP-1 family, including Fosl1, in the two other LSC^med^ subpopulations (Fig. EV4G–L) suggested a wide spectrum of targeted LSC^med^ cells. The markedly increased expression of FOSL1 and FOSL1 target genes after enzalutamide treatment of PCa patients in the DARANA study (Linder et al, 2022a; Data ref: Linder et al, 2022b) (Fig. EV5B,C) further supports the relevance of FOSL1 as a therapeutic target in the human disease.

We noticed that Pim1 is part of the Fosl1 gene network in the mouse (Fig. 4C). PIM1 is a member of the PIM serine/threonine kinase family that phosphorylates substrates controlling various tumorigenic phenotypes, including cell survival and proliferation (Rout et al, 2024; Wang et al, 2001). Of note, PIM1 has also been shown to promote transcript elongation of the FOSL1 gene through phosphorylation of serine 10 at histone H3 at the FOSL1 enhancer (Zippo et al, 2009). In PCa, it was shown to promote cell invasion and migration (Rebello et al, 2016; Santio et al, 2015) and ligand-independent androgen receptor phosphorylation associated with hormone-refractory PCa (Ha et al, 2013). PIM1 pharmacological targeting was shown to reduce prostate tumor growth in mouse and human preclinical settings (Hu et al, 2009; Rebello et al, 2016). We found that Pim1 is drastically enriched in Pten^pc−/−^ LSC^med^-1 and LSC^med^-2 cell subpopulations, which are predominant post-castration (Fig. 5A,B). Of the two other paralogs of the Pim kinase family, Pim3, but not Pim2, is highly expressed in all Pten^pc−/−^ LSC^med^ cell subpopulations (Fig. EV4M,N). Of note, PIM1 expression was also increased by enzalutamide treatment in naïve PCa patients (Fig. EV5D), albeit less markedly than FOSL1 (Fig. EV5B).Figure 5. Targeting FOSL1/AP-1 and Pim family prevents the growth of LSC^med^ and human Club-like cells in vitro.(A,** B**) Expression per Pten^pc−/−^ LSC^med^ cell subpopulation (A) and UMAP projection (B) of Pim1. (C,** D**) Number (C) and size (D) of organoids generated from sorted Pten^pc−/−^ LSC^med^ cells after 10 days of culture in medium containing DMSO or 1 nM JQ-1 and 1 nM CX-6258, alone or combined, as indicated. Data were normalized to the control (DMSO) condition (biological replicates, n = 5 independent experiments). **p < 0.01; ***p < 0.001; p < 0.0001 (one-way ANOVA and Dunnett post hoc test). Exact p values are reported in Appendix Table S1. (E, F) Enrichment of various epithelial cell signatures (E) and LSC^med^ subpopulation signatures (F) in HPV-10, PC-3, DU145, LNCaP, and 22Rv1 cell lines (Data Ref: Wang et al, 2007a). (G) Enrichment of CRPC-AR, CRPC-WNT, CRPC-NE, and CRPC-SCL human tumoral subtypes signatures (Dataset EV2) in HPV-10, PC-3, DU145, LNCaP, and 22Rv1 cell lines. (H) Human HPV-10 cells were treated for 72 h with 1 µM JQ-1 and 1 µM CX-6258, alone or combined (as indicated), then the number of adherent cells was counted (biological replicates, n = 4 independent experiments). The data were normalized to the control condition (DMSO). **p < 0.01; ***p < 0.001 (one-way ANOVA and Dunnett post hoc test). Exact p values are reported in Appendix Table S1. (I) Images of HPV-10 cells after 72 h of treatment. (J) The viability of HPV-10 cells (adherent + in suspension) was determined by trypan blue staining (biological replicates, n = 4 independent experiments). ns not significant; *p < 0.05 (one-way ANOVA and Dunnett post hoc test). Exact p values are reported in Appendix Table S1. (K, L) Same as (H, J) with PC-3 cells (biological replicates, n = 4 independent experiments). ns not significant; *p < 0.05, **p < 0.01; *p < 0.001 (one-way ANOVA and Dunnett post hoc test). Exact p values are reported in Appendix Table S1. (M) Tumorsphere-forming capacity of HPV-10 cells in medium containing DMSO or 1 µM JQ-1 and 1 µM CX-**6258, alone or combined as indicated. Data were normalized to the DMSO condition (biological replicates, n = 3 independent experiments). ***p < 0.001 (one-way ANOVA and Dunnett post hoc test). Exact p values are reported in Appendix Table S1. All error bars in this figure represent SD. Source data are available online for this figure.

We tested JQ-1 and CX-6258 as pharmacological inhibitors of these two selected targets in functional assays. JQ-1 is a global BET/AP-1 inhibitor that has been shown to mainly target FOSL1 in lung adenocarcinomas (Casalino et al, 2022; Lockwood et al, 2012) and to phenocopy FOSL1 knockdown in other types of tumors (Baker et al, 2015; Bid et al, 2016). Importantly, JQ-1 had been validated in preclinical in vivo settings prior to this study (Shimamura et al, 2013; Shu et al, 2016). CX-6258 is a pan-PIM kinase inhibitor (Haddach et al, 2012), which prevents compensation by any other expressed Pim paralog (e.g., Pim3 in our case) as reported in other studies (Mikkers et al, 2002; van der Lugt et al, 1995). CX-6258 was recently shown to interfere with PCa progression in preclinical models (Rebello et al, 2016).

We used the organoid assay to determine the functional impact of these pharmacological inhibitors on Pten^pc−/−^ LSC^med^ cell properties. We first determined the dose-response of each drug on the organoid-forming capacity of these cells. Both JQ-1 and CX-6258 were very active in the nanomolar range (Fig. EV6A,B), indicating high sensitivity of LSC^med^ cells to these drugs. At 1 nM concentration, each drug decreased by half both the number and the size of the organoids generated from sorted Pten^pc−/−^ LSC^med^ cells, indicating inhibition of their progenitor and proliferation capacities, respectively. These effects were further potentiated when both drugs were combined (Fig. 5C,D). These data indicate that targeting FOSL1/AP-1 and/or PIM kinases markedly alters the growth and progenitor properties of Pten^pc−/−^ LSC^med^ cells.

We then aimed to validate this therapeutic approach using human Club/Hillock cancer cells. In contrast to Pten^pc−/−^ mouse LSC^med^ cells, there are currently no established procedures to sort, culture and expand Club/Hillock cells from fresh specimens of human prostate cancer. Therefore, we sought a relevant cell line matching the Club/Hillock phenotype. By comparing human PCa cell line signatures (Data Ref: Wang et al, 2007a; Wang et al, 2007b) to mouse LSC^med^ (Sackmann Sala et al, 2017) and human Club, Hillock (Henry et al, 2018), basal and luminal (Pitzen et al, 2025) cell signatures (Dataset EV5), we found that HPV-10 cells are highly enriched in LSC^med^/Club/Hillock features (Fig. 5E). In comparison, androgen-independent PC-3 and DU145 cell lines exhibited an intermediate enrichment, while androgen-dependent LNCaP and 22Rv1, classically described as luminal prostate cancer cells, were the least enriched in LSC^med^/Club/Hillock features among the cell panel analyzed (Fig. 5E).

The HPV-10 cell line is not commonly used in the field. In contrast to PC-3 and DU145 cells, which were derived from PCa metastatic sites, HPV-10 cells were isolated from a high-grade primary prostate adenocarcinoma. They grow in culture in the absence of androgens, which indicates their intrinsic castration tolerance. Accordingly, no PSCA expression was detected by RT-qPCR (Appendix Fig. S4A). Otherwise, we confirmed the expression of various LSC^med^ markers (e.g., KRT4 and KRT7) in HPV-10 cells compared to the luminal LNCaP cell line (Appendix Fig. S4A). Immunofluorescence analysis showed the co-expression of KRT4 (Club cell marker) and KRT13 (Hillock cell marker) (Appendix Fig. S4B), identifying HPV-10 as a mixed Club/Hillock cell line. This is reminiscent of murine LSC^med^ cells that also match both Club and Hillock signatures (Baures et al, 2022a; Baures et al, 2022b). HPV-10 cells are particularly enriched in the LSC^med^-0 signature (Fig. 5F), and they express FOSL1 and PIM1 at a much higher level than LNCaP cells, further supporting their Club-like cell profile (Fig. EV6C; Appendix Fig. S4C). While HPV-10 cells are not tumorigenic, the accumulation of chromosomal alterations, including amplification of c-MYC, characteristic of tumorigenic PC-3 cells, makes them tumorigenic (Hukku et al, 2000). Altogether these data identify the HPV-10 cell line as a relevant model of Club/Hillock PCa cells preexisting in prostate tumors, i.e., at a less aggressive stage than metastatic PC-3 cells that are exclusively enriched in LSC^med^-1 signature (Fig. 5F). Finally, HPV-10 cells, but not LNCaP and 22Rv1 cells, expressed the genes of the CRPC-SCL signature to a similar degree as PC-3 and DU145 cells, previously identified as MSPC/SCL models (Han et al, 2022; Tang et al, 2022) (Fig. 5G; Dataset EV2). This identifies HPV-10 cells as another surrogate of the DNPC subtype.

To determine the effects of JQ-1 and CX-6258 on human Club/Hillock-like cells, we first performed dose-response assays using the HPV-10 cell line (Fig. EV6D). For both drugs, we observed a significant inhibition of the number of adherent cells in the micromolar range, indicating lower drug sensitivity compared to Pten^pc−/−^ LSC^med^ cells (Fig. EV6A,B). Based on that data, we used 1 µM of each drug for subsequent experiments. As observed above with Pten^pc−/−^ LSC^med^ cells, the combination of JQ-1 and CX-6258 potentiated the inhibition of HPV-10 cell growth in two-dimensional cultures (Fig. 5H,I). The moderate (15–25%) ratio of dead cells at 72 h of treatment (Fig. 5J) suggested that the drugs mainly prevented cell proliferation, as reported in former studies using JQ-1 (Li et al, 2019; Zhang et al, 2020; Zhang et al, 2021). A similar pattern of drug response was observed with PC-3 cells (Fig. 5K,L). In addition, we found that the tumorsphere formation by HPV-10 cells was significantly altered by the drugs, again with almost a total inhibition when using the drug combination (Fig. 5M). This assay could not be performed with PC-3 cells as these cells failed to generate tumorspheres in our hands.

To confirm the FOSL1-dependence of the effects observed with JQ-1, we looked for an alternative FOSL1 inhibitor. T5224 is an AP-1 inhibitor that was recently shown to directly bind to FOSL1 (Zaman et al, 2024) and to inhibit tumorigenesis with low/moderate potency in various cancer models (Kamide et al, 2016; Tang et al, 2022). A PROTAC version of T5224 was reported to selectively degrade FOSL1 in head and neck squamous cell carcinoma, leading to increased potency compared to the parental compound (Zaman et al, 2024). Treatment of PC-3 cells with T5224-PROTAC led to similar effects compared to JQ-1 treatment, alone or combined with CX-6258 (Fig. EV6E,F). HPV-10 cells exhibited even higher drug sensitivity compared to PC-3 cells (Fig. EV6G,H). These data strengthen the role of FOSL1 in Club-like cell growth and survival.

To definitely establish this conclusion, we aimed to silence FOSL1 expression in PC-3 and HPV-10 cells. HPV-10 cells proved extremely sensitive to lipofectamine, which per se was detrimental to cell survival at lipofectamine concentrations required for siRNA entry into cells. In contrast, efficient FOSL1 knockdown in PC-3 cells was obtained with three different siRNAs (Fig. EV6I). Using the most potent of them (siFOSL1(1)), we observed that FOSL1 silencing significantly altered the growth and viability of PC-3 cells (Fig. EV6I–L). These effects were quantitatively similar to those obtained with 15 µM T5224-PROTAC (Fig. EV6E,F).

Together, these results indicate that FOSL1 is a relevant target, and JQ-1/CX-6258 combination is an effective therapy, to abolish the progenitor and growth properties of mouse and human Club-like prostate cancer cells in vitro.

Targeting the FOSL1/AP-1 and Pim family markedly reduces CRPC growth in vivo

We then aimed to validate this drug combination strategy in a CRPC context in vivo. Based on the progressive upregulation of Fosl1 and Pim1 expression in LSC^med^ cells soon after Pten^pc−/−^ mouse castration (Figs. 3F and EV1B and EV2E–G), we reasoned that castration-induced reprogramming of LSC^med^-0 towards LSC^med^-1 transcriptional profiles should provide a window of opportunity for combination therapy involving JQ-1 and CX-6258.

Two groups of experimental mice (n = 8 each) were castrated, and after 4-day recovery, we started treatment with JQ-1 (daily) and CX-6258 (biweekly) versus vehicle (Fig. 6A). After 28 days, mice were euthanized, then prostates were harvested, weighed, and processed for histopathological analyses or functional assays. The kidneys, lungs, and livers of five animals from both the control and treated groups were also harvested and examined by two pathologists blinded to group assignments to assess drug toxicity. Only minor pathological lesions were observed (Appendix Table S2), with no significant differences between the control and treated groups (Appendix Fig. S5A,B). We concluded that these are incidental lesions, and the administered drug combination had no toxic effect at that dosage. The absence of treatment-induced cell death as determined by the TUNEL assay fully supported this conclusion (Appendix Fig. S5C).Figure 6. Combined JQ-1 and CX-6258 treatment reduces CRPC growth and aggressiveness in vivo.(A) Timeline representation of the experimental protocol used for Pten^pc−/−^ mouse castration and pharmacological treatments. Castrated mice received vehicle (n = 8) or daily JQ-1 intraperitoneal injection and biweekly CX-6258 oral administration (n = 8) during 28 days. (B) After 28 days of treatment, prostates were microdissected and weighed. p < 0.01 (unpaired two-tailed t-test with Welch’s correction). Exact p value: p = 0.0075. (C) For all mice, prostate histological structures were classified as benign (normal glands, PINs) or tumoral (cribriform, adenocarcinoma) as described in the Materials and methods (scale bar = 100 μm). (D) Quantification of the effects of pharmacological treatments on prostate histopathology illustrated in (C). p < 0.01 (unpaired two-tailed t-test with Welch’s correction). Exact p values: p = 0.0087 (benign), p = 0.0087 (tumoral). (E, F) After 28 days of treatment, LSC^med^ cells were sorted from the prostate of vehicle- or drug-treated animals and cultured in organoid medium in the absence of pharmacological inhibitors. Images were taken (E) and the number of organoids was measured (F) after 10 days of culture (biological replicates, n = 3 experiments). ***p < 0.001 (unpaired two-tailed t-test with Welch’s correction). Exact p value: p = 0.0023. (G) Timeline representation of the experimental protocol used for PC-3 cell xenograft assay and pharmacological treatments. 1.10^6^ PC-3 cells were injected into each flank of castrated immunodeficient NSG mice. Mice were randomized to receive vehicle (n = 12) or daily JQ-1 intraperitoneal injection and biweekly CX-6258 oral administration (n = 7) during 24 days. (H–J) During the protocol, tumor size was measured manually at days 15, 19, 22, and 24 of treatment (I). **p < 0.01; ***p < 0.001; ****p < 0.0001 (unpaired two-tailed t-test with Welch’s correction comparing vehicle and JQ-1 + CX-6258 at each time). Exact p values: p = 7.1e-05 (D15), p = 4.3e-04 (D19), p = 1.5e-05 (D22), p = 7.8e-07 (D24). At sacrifice (D24), tumors were microdissected (H) and weighed (J). ***p < 0.001 (unpaired two-tailed t-test with Welch’s correction). P value: p = 0.000055. All error bars in this figure represent SD. Source data are available online for this figure.

We observed that the prostate weight in the drug-treated group was reduced by 41% compared to the vehicle-treated group (Fig. 6B). This was associated with a decreased proportion of advanced histologically-classified structures at the benefit of benign phenotypes (i.e., normal glands and prostate intraepithelial neoplasia [PINs]) (Fig. 6C,D). In parallel to these studies, LSC^med^ cells sorted from a set of fresh prostates of vehicle- or drug-treated mice (each n = 3) were cultured in organoid medium for 10 days in the absence of JQ-1 and CX-6258. Remarkably, we observed that LSC^med^ cells sorted from drug-treated Pten^pc−/−^ mice had virtually lost any capacity to generate organoids compared to the vehicle-treated group (Fig. 6E,F).

To determine whether JQ-1/CX-6258 treatment interfered with the emergence of the LSC^med^-1 cell subpopulation, as expected, the whole transcriptome of EpCAM+ cells in prostate sections from vehicle- versus drug-treated castrated Pten^pc−/−^ mice was analyzed by digital spatial profiling (Dataset EV6). Prostate sections from naïve (non-castrated) Pten^pc−/−^ mice were used as a control in this experiment. This analysis confirmed the marked enrichment of the LSC^med^-1 signature score in castrated Pten^pc−/−^ mice (Fig. EV7A). No significant variation in the LSC^med^-0 and LSC^med^-2 signature scores could be identified in the glands of the three cohorts analyzed (Appendix Fig. S6), possibly reflecting the lower sensitivity of this bulk transcriptomic profiling technology compared to scRNAseq (Fig. 2B). Importantly, the LSC^med^-1 signature score was reduced by JQ-1/CX-6258 treatment, although it did not reach the score of naïve mice (Fig. EV7A). In total, 21 genes were significantly deregulated (p value <0.05 and a log2 fold change >0.5) in castrated mice after the treatment (C3, Lgals3bp, Srrm2, Srgn, Ifi203, Ifi27l2a, Ifitm3, Ctsl, Oas2, Apobr, Pglyrp1, Hsd11b2, Atp2a3, H2-D1, H2-Q6, Basp1, S100a6, Cd74, H2-K1, Psmb8, and* Hspa1a*). All top genes affected by the treatment in both LSC^med^-1 (C3, Cd74, Hspa1a, Ifi203, Ifi27l2a, and Ifitm3) and LSC^med^-2 (S100a6 and Srgn) signatures were downregulated (Fig. EV7B). These genes are mainly involved in inflammatory pathways, such as genes involved in the complement system (i.e., C3), antigen presentation (e.g., Cd74), and IFN-α and -γ pathways (e.g., Ifitm3). Note that for some transcripts, the treatment normalized their levels (Hspa1a, S100a6, and Srgn). No specific histological features could be correlated to any signature enrichment (Fig. EV7C). Taken together, these results confirm that the LSC^med^-1 signature is amplified in Pten^pc−/−^ mice after castration and that the combination of FOSL1 and PIM kinase inhibitors targets this cell subpopulation.

Next, we set up a xenograft assay to assess the in vivo efficacy of our therapeutic strategy on human DNPC cells. We used the PC-3 cell line as HPV-10 cells have been characterized as non-tumorigenic in vivo (Hukku et al, 2000; Weijerman et al, 1994). Immunodeficient NSG mice were castrated, and PC-3 cells were then injected subcutaneously into both flanks of the animals before the start of treatment four days later (Fig. 6G). Follow-up of tumor volume showed that tumor growth was markedly delayed in the drug-treated group compared to the vehicle-treated group (Fig. 6H,I; Appendix Fig. S7). On day 24, tumors were excised and weighed (Fig. 6H,J). Both the volume (−62%) and the weight (−59%) of tumors were drastically reduced in the drug-treated group.

Together, these data demonstrate that pharmacological targeting of FOSL1/AP-1 and PIM kinases using the combination of JQ-1 and CX-6258 inhibitors markedly reduces the growth in vivo of androgen-independent MSPC/SCL-like cells in a castration context.

Discussion

A critical knowledge gap in prostate cancer research is to understand whether castration-tolerant progenitor-like cells that reside in treatment-naive tumors play a direct role in castration resistance, making them relevant therapeutic targets. This study uncovers the unexpected finding that preexisting Club-like cells are not intrinsically non-responsive to androgen deprivation but actually respond to castration by a transcriptional switch that increases their aggressiveness. This transcriptional plasticity is mainly orchestrated by the FOSL1/AP-1 complex. This finding opens new therapeutic avenues to eradicate treatment-resistant PCa cells early in the course of the disease for suppressing progression towards CRPC. Supporting this promising therapeutic approach, combined pharmacological targeting of FOSL1/AP-1 and PIM kinases markedly reduced the in vivo tumor growth capacities of preclinical DNPC models, which was correlated to the suppression of their progenitor properties in vitro.

The concept of preexisting CRPC cells in treatment-naïve tumors was pioneered by the group of Risbridger using patient-derived xenografts of treatment-naïve early-stage tumors (Toivanen et al, 2013). These authors showed that, 4 weeks after mouse castration, residual tumor foci contained stem-like tumor cells able to regenerate proliferating tumors upon androgen replenishment. However, the co-expression of stem-like (NANOG, ALDH1, and CD44) and differentiated (NKX3.1) cell markers cannot rule out that these CRPC cells actually arose from castration-induced luminal cell plasticity, a mechanism of castration resistance recently documented in various mouse models (Chan et al, 2022; Karthaus et al, 2020; Kirk et al, 2024). More recently, the existence of stem-like/EMT-enriched CRPC cells in treatment-naïve prostate tumors has been revealed by scRNAseq approaches. These cells have been correlated with biochemical recurrence (rising PSA levels) and distant metastasis (Cheng et al, 2022). Other authors (Chen et al, 2021) identified one basal/intermediate epithelial cell population (referred to as cluster 10) that we subsequently showed to be enriched in typical Club/Hillock cell markers (Baures et al, 2022b). This cluster was identified in each individual patient, albeit at a highly variable ratio (1 to 25% of the epithelial cell pool). Song and colleagues classified PCa-associated Club-like cells into six transcriptomic states, the largest of which exhibited the highest luminal and AR signaling characteristics (Song et al, 2022). This is also what we observed for the Pten^pc−/−^ mouse LSC^med^-0 subpopulation predominant before castration. All these scRNAseq studies corroborate the report by Han and colleagues, who showed on large cohorts that the MSPC subtype, enriched in Club cell features, is present de novo in a significant proportion of treatment-naïve PCa (Han et al, 2022).

According to the molecular similarity of mouse LSC^med^ and human Club/Hillock cells, we established that Pten^pc−/−^ mouse LSC^med^ cells are a robust surrogate of human DNPC/MSPC/SCL subtype. Based on this finding, the Pten^pc−/−^ mouse model was attractive as it shows a highly consistent enrichment of LSC^med^ cells in tumors prior to castration (Sackmann Sala et al, 2017), mimicking in an amplified way human de novo MSPCs. LSC^med^-like cells have been classically considered as an end-stage of androgen-independent cells. Our study shows that this is not the case as naïve Pten^pc−/−^ LSC^med^ cells, despite their intrinsically low androgen-signaling, are sensitive to castration and respond to androgen deprivation by a transcriptional switch. Before castration, Pten^pc−/−^ LSC^med^ cells mainly exhibit the LSC^med^-0 profile enriched in developmental signaling pathways, as previously observed for wild-type LSC^med^-like luminal progenitor cells that participate in the early steps of prostate morphogenesis (Baures et al, 2022a; Mevel et al, 2020). In contrast, post-castration Pten^pc−/−^ LSC^med^ cells mainly exhibit the LSC^med^-1 profile, which is comparatively enriched in stemness and basal/Hillock features. This finding echoes a recent report showing that DNPC, and in particular the CRPC-SCL subtype, is enriched in cells exhibiting a mixed basal, Club, and Hillock identity (Pitzen et al, 2025). This profile, which is typical of the human MSPC subtype resistant to ARPIs (Han et al, 2022), further increases the transcriptomic similarity between post-castration LSC^med^ cells and the MSPC/SCL subtypes. The acquisition of an intermediate basal/luminal phenotype by cell plasticity has also been reported in triple-negative breast cancer, and it was shown to promote resistance to chemotherapy (Marsolier et al, 2022). In the prostate context, the acquisition of a basal/stem-cell signature identifies aggressive PCa phenotypes (Smith et al, 2015). This signature was also enriched in hormone-sensitive metastases compared to organ-confined prostate adenocarcinomas, suggesting that the enrichment of basal features parallels PCa progression. Therefore, the acquisition of basal features by LSC^med^-like cells post-castration could be pivotal for their aggressiveness and the emergence of MSPC-CRPC tumors in response to treatment.

We identified FOSL1/AP-1 as a major driver of LSC^med^ cell plasticity in response to castration. While the AP-1 complex is known to mediate cellular stress, the detection of FOSL1 upregulation by various technological approaches using tissue samples immediately processed after animal sacrifice (e.g., IHC, RT-qPCR) eliminates any technical bias due to cellular stress potentially induced by methodologies involving cellular dissociation of prostatic tissue (e.g., cell sorting, scRNAseq). The AP-1 complex, and in particular FOSL1, is involved in the development of many adenocarcinomas (Casalino et al, 2022). Acting as a transcription factor, facilitator of chromatin accessibility/opening and/or super enhancer (Bi et al, 2020; Dong et al, 2021; Kadur Lakshminarasimha Murthy et al, 2022), FOSL1 regulates various tumor-associated processes e.g., cell proliferation (Zanconato et al, 2015), EMT (Feldker et al, 2020; Marques et al, 2021), invasion and metastases (Iskit et al, 2015), and stemness (Marques et al, 2021). Via its association with partners such as YAP and TAZ (Feldker et al, 2020; Zanconato et al, 2015), FOSL1 also promotes cell plasticity towards increased stemness states (Marques et al, 2021) and resistance to treatment (Bi et al, 2020). FOSL1 has been recently shown to be involved in the reprogramming of mouse prostatic luminal cells towards a castration-tolerant progenitor-like state following androgen depletion (Tang et al, 2022). We here identify FOSL1 as a master regulator of LSC^med^-1 cells, i.e., the predominant LSC^med^ population after castration. The consistent expression levels of Fosl1 and of various members of the AP-1 complex in both LSC^med^-1 and LSC^med^-2 cells strengthen the key role of this transcriptional complex in the molecular switch of preexisting LSC^med^ cells in response to castration. In humans, FOSL1 has been identified as the top key transcription factor in CRPC-SCL (Tang et al, 2022). It is positively associated with chromatin remodeling that promotes CRPC-SCL subtype in the castrated context by enhancing CRPC-SCL-associated enhancer accessibility, in cooperation with the YAP-TAZ complex (Tang et al, 2022). As observed for post-castration LSC^med^ cells, human CRPC-SCL is enriched in various AP-1 family members. Furthermore, we found that both FOSL1 and FOSL1-target genes are upregulated by enzalutamide treatment of naïve PCa patients (Fig. EV5 and Linder et al, 2022a). Together, these data support the key role of this transcriptional complex in the adaptation of treatment-naïve PCa luminal cells—both differentiated (Tang et al, 2022) and progenitor-like (this study)—to androgen deprivation.

Based on our findings, we reasoned that inhibition of FOSL1/AP-1 pathway should suppress the castration-driven mechanisms of resistance of LSC^med^ progenitor cells promoting CRPC. This therapeutic strategy was strengthened by coupling pharmacological inhibition of PIM kinase, previously shown to promote cell stemness (Gao et al, 2019; Jimenez-Garcia et al, 2016) and to affect PCa growth (Hu et al, 2009; Rebello et al, 2016). The combination of castration and JQ-1/CX-6258 treatment of Pten^pc−/−^ mice markedly reduced in situ CRPC growth and histopathology after only 28 days compared to castration alone. Similar tumor growth reduction was observed with xenografted PC-3 cells. One remarkable effect of this pharmacological treatment was to abrogate the progenitor properties of Pten^pc−/−^ LSC^med^ cells as determined from their organoid/tumorsphere-forming capacity. Remarkably, this effect was also observed with LSC^med^ cells harvested from JQ-1/CX-6258-treated Pten^pc−/−^ mice, without adding drugs in the organoid assay (Fig. 6F). Such a radical inhibition of their ability to form organoids cannot be explained solely by the reduction in the LSC^med^-1 signature revealed by spatial transcriptomic analysis. Although it is at present unknown whether this treatment suppresses the progenitor properties of some LSC^med^ cells, or eradicates the progenitor cells per se, our data establish a correlation between in situ tumor growth and in vitro progenitor capacity.

In the absence of established procedures to enrich human Club/Hillock cells from prostate specimens, the progenitor-like properties of the latter cells have never been experimentally assessed. We here identified the HPV-10 cell line as the best surrogate of tumoral Club/Hillock cells. According to their prostatic adenocarcinoma origin (Hukku et al, 2000) and their enrichment in the LSC^med^-0 signature, HPV-10 cells appear to be a relevant model of Club-like cells preexisting in naïve tumors. We show that these cells are able to form tumorspheres in vitro, which is consistent with the in silico-predicted progenitor-like capacities of Club/Hillock cells (Henry et al, 2018). As observed with Pten^pc−/−^ LSC^med^ cells, combined JQ-1/CX-6258 treatment completely abrogated the tumorsphere-forming capacity of HPV-10 cells. Their growth was also markedly reduced while their viability was mildly affected. Taken together, the results of our in vitro and in vivo assays suggest that combined pharmacological inhibition of the FOSL1/AP-1 complex and PIM1 interferes with the growth, survival, transcriptomic reprogramming and progenitor capacities of mouse and human Club-like cells. Further studies are needed to understand how all these effects work together to abrogate the promotion of CRPC growth by Club-type cells.

Neuroendocrine differentiation (NED) is a mechanism of late-stage therapeutic resistance, increasingly observed since the generalization of ARPI treatment of advanced prostate cancer (Yamada and Beltran, 2021). Several observations suggest that LSC^med^/Club/Hillock cells may be predisposed to NED under therapeutic pressure. Wfdc2 and Sox2 are part of the LSC^med^ cell signature (Baures et al, 2022a); both genes are positively associated with the neuroendocrine signature in human PCa (Dong et al, 2020), and Sox2 is essential for NED in mice (Kwon et al, 2021). Similarly, human Club/Hillock cells are enriched in neuronal stem programs (Yan et al, 2022). In Pten^pc−/−^ mice, our scRNAseq analysis showed that LSC^med^-2 cells, which are moderately amplified after castration, are characterized by the expression of Onecut2. This transcription factor is a major inducer of early NED and biochemical recurrence via the inhibition of AR and FoxA1 transcriptional programs (Rotinen et al, 2018). Still, LSC^med^-2 cells showed only low enrichment in NEPC signature and no expression of late NED markers 2 months post-castration. This is consistent with the unaltered LSC^med^-2 signature observed by spatial transcriptomics analysis. The heterogeneity of LSC^med^-2 marker appearance after castration (RT-qPCR data) is informative as it suggests that the kinetics of progression towards the LSC^med^-2 state is not only delayed, but also animal-dependent, compared to the emergence of the LSC^med^-1 state. Together, these data suggest that a longer delay than the 2 months post-castration analyzed in this study may be required to alter the expression level of factors known to positively regulate NED (e.g., Ascl1 or Ezh2) or negatively (Klf5). Further genetic stress, e.g., loss of Rb1 and/or Tp53, two key negative regulators of the neuroendocrine signature (Qian et al, 2022), may also be required to drive full NED of some LSC^med^-2 cells. Of interest, Onecut2 was recently shown to be a broadly acting lineage plasticity facilitator as it also supports the appearance of treatment-resistant adenocarcinoma through multiple mechanisms (Qian et al, 2024). In this context, strategies targeting progenitor cells presenting with the LSC^med^-2 profile may also be relevant to prevent progression towards aggressive forms of CRPC. The consistent expression of AP-1 family members and PIM kinases in LSC^med^-2 cells suggests that our targeting strategy using JQ-1/CX-6258, primarily designed to target LSC^med^-1 cells, may also counteract LSC^med^-2 cell amplification post-castration. Alternatively, the recently reported small molecule inhibitor of ONECUT2 (Qian et al, 2024) should also be considered. Additional studies are needed to elucidate the actual kinetics of LSC^med^-2 cell emergence in Pten^pc−/−^ mice in order to challenge these therapeutic strategies.

This study presents some limitations. While FOSL1 silencing in vitro confirmed the relevance of targeting this transcription factor in Club-like cells, additional genetically-engineered models, including Fosl1-deficient Pten^pc−/−^ mice, are needed to better delineate the actual effects of FOSL1 inhibition on LSC^med^ cell growth, survival, plasticity and progenitor properties. Such models will also help identify any functional redundancy involving other AP-1 members, which may promote therapeutic resistance to FOSL1 inhibition. Another hurdle relates to the specificity of the drugs used to target FOSL1. Although JQ-1 was shown to mainly target FOSL1 and to phenocopy FOSL1 knockdown in various tumors (Baker et al, 2015; Bid et al, 2016; Casalino et al, 2022; Lockwood et al, 2012), other genes may also be altered by this BET/AP-1 inhibitor. In this matter, exhaustive in vitro and in vivo characterization of the effects of T5224-PROTAC, recently shown to directly bind to, and degrade, FOSL1 (Zaman et al, 2024), is necessary to evaluate its therapeutic superiority to JQ-1 in terms of specificity and efficacy.

In conclusion, our study uncovers the critical role of transcriptional reprogramming in preexisting Club-like cells as a novel mechanism driving castration resistance. We demonstrate that FOSL1/AP-1 is a key regulator of this molecular transformation, resulting in the emergence of aggressive progenitor cells enriched in basal characteristics. Furthermore, our findings align with recent research indicating that the shift from androgen-dependent luminal cells to androgen-independent progenitor-like cells during castration is similarly mediated by this transcriptional complex. This establishes FOSL1/AP-1 as a pivotal factor in the mechanisms underlying resistance to anti-androgen therapy. Therefore, our dual therapeutic approach targeting both FOSL1/AP-1 and PIM kinases combined to castration presents a promising strategy to counteract these early mechanisms of castration resistance and disrupt the progression towards CRPC.

Methods

Reagents and tools tableReagent/resourceReference or sourceIdentifier or Catalog Number Experimental models HPV-10ATCCCRL-2220 RRID: CVCL_3495PC-3ATCCCRL-1435 RRID: CVCL_0035LNCaPATCCCRL-1740 RRID: CVCL_0395Pten^loxP/loxP^ (M. musculus)Sackmann Sala et al, 2017RRID: MGI:5292551Pb-Cre4 (M. musculus)Sackmann Sala et al, 2017RRID: IMSR_NCIMR:01XF5NSG (NOD.Cg-Prkdc^scid^ Il2rg^tm1Wjl^/SzJ) (M. musculus)Charles River LaboratoriesRRID: SCR_003792 Recombinant DNA N/A Antibodies Rat anti-CD31 (clone 390) coupled to fluorescein isothiocyanateeBioscience11-0311-82 RRID: AB_465012Rat anti-CD45 (clone 30-F11) coupled to fluorescein isothiocyanateeBioscience11-0451-81 RRID: AB_465049Rat anti-TER-119 coupled to fluorescein isothiocyanateeBioscience11-5921-85 RRID: AB_465312Rat anti-EpCAM (clone G8.8) coupled to phosphatidylethanolamine-Cyanine7eBioscience25-5791-80 RRID: AB_1724047Rat anti-Sca-1 (clone D7) coupled to allophycocyanineBioscience17-5981-81 RRID: AB_469486Rat anti-CD49f (clone GoH3) coupled to phosphatidylethanolamineeBioscience12-0495-81 RRID: AB_891478Rabbit anti-Ki-67 (clone SP6)Zytomed SystemsRBK027-05Rabbit anti-KRT4 (clone EP1599Y)Abcamab51-599 RRID: AB_869888Rabbit anti-KRT13 (polyclonal)Sigma-AldrichHPA030877 RRID: AB_2673641Rat anti-CD44 (IHC)Biolegend103001 RRID: AB_312952Anti-p63 (4A4)BioSBBSB3605Anti-FRA1 (Fosl1), clone F2C9L (IHC/IF)Cell Signaling Technologies28801S RRID: AB_3697251Mouse anti-FRA1 (Fosl1), clone C-12 (WB)Santa-Cruz Biotechnologiessc-28310 RRID: AB_627632Rabbit anti-GAPDH (WB)Cell Signaling Technologies5174 RRID: AB_10622025Anti-rabbit IgG, HRP-linked AntibodyCell Signaling Technologies7074 RRID: AB_2099233Anti-mouse-IgG, HRP-linked AntibodyCell Signaling Technologies7076 RRID: AB_330924Alexa Fluor 488 anti-mouse CD236 (EpCAM)AbcamAB237384Rabbit Anti-Rat IgG Antibody, mouse adsorbed (H + L), BiotinylatedVector laboratoriesBA-4001-.5Mouse-IgGκ BP-HRPSanta-Cruz Biotechnologiessc-516102 RRID: AB_2687626Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor™ 488InvitrogenA-11008 Oligonucleotides and other sequence-based reagents PCR primers (this study)Sigma-AldrichAppendix Table S3siRNA FOSL1Horizon DiscoveryJ-004341-05-0010J-004341-08-0010Pool siRNA scrambleHorizon DiscoveryD-001810-10-05 Chemicals, Enzymes and other reagents JQ-1MedChemExpressHY-13030CX-6258MedChemExpressHY-18095FOSL1 degrader 1 (T5224-PROTAC)MedChemExpressHY-162723CremophorMedChemExpressHY-Y1890PEG300MedChemExpressHY-Y0873Tween80MedChemExpressHY-Y1891Hoechst 33342InvitrogenH3570Tween-20Thermo Scientific ChemicalsJ20605.APPFAThermo Scientific Chemicals047377.9 LLipofectamineThermo Fisher Scientific18324012PuromicyneInvivogenant-pr-1Doxycycline hyclateSigma-AldrichD9891Fetal bovine serumEurobioCVFSVF00-01 (#lot S80515)Dulbecco’s modified Eagle medium (DMEM)Thermo Fisher Scientific31966-021Keratinocyte Serum Free Medium (K-SFM)Thermo Fisher Scientific17005-034Opti-MEM™ I Reduced-SerumThermo Fisher Scientific31985070Advanced DMEM/F-12 (DMEM/F-12)Thermo Fisher Scientific12634010Low-growth factor-containing MatrigelCorning354230Bovine Pituitary Extract (BPE)Thermo Fisher Scientific13028-014Penicillin /StreptomycinThermo Fisher Scientific15140122Collagenase Type IThermo Fisher Scientific17018029DNase IRoche05952077103EGFPeproTechAF-100-15-500UGHistopaque-1119Sigma-Aldrich11191Trypan blueThermo Fisher Scientific15250061SYTOX BlueLife TechnologiesS34857SYTO™ 83 Orange Fluorescent Nucleic Acid StainThermo Fisher ScientificS1136496-well plateSarstedt83.392596-well plates (DNase, RNase free)VWR732-2387HBSSThermo Fisher Scientific14175095Phosphate-buffered saline (PBS, 1X)Thermo Fisher ScientificJ61196.APHalt™ Protease and Phosphatase Inhibitor Cocktail (100X)Thermo Fisher Scientific78442Pierce™ BCA Protein Assay KitsThermo Fisher Scientific23225NuPAGE™ LDS Sample Buffer (4X)Thermo Fisher ScientificWG1402BX10Immobilon Forte Western HRP substrateMilliporeWBLUF0500Nucleospin RNA XSMacherey-Nagel740902Nucleospin RNAMacherey-Nagel740984SuperScript™ VILO™ cDNA Synthesis KitInvitrogen11754050GoScript™ Reverse TranscriptasePromegaA5003GoTaq(R) qPCR Master MixPromegaA6002In Situ Cell Death Detection Kit (TUNEL)Roche11 684 809 910VECTASTAIN® Elite® ABC-HRP Kit, PeroxidaseVector laboratoriesPK-6100DAB Substrate Kit, Peroxidase (HRP)Vector laboratoriesSK-4100Single-cell lysis bufferTakara635013AmpureXP beadsBeckman CoulterCat# A 63881High-sensitivity DNA chipAgilentCat# 5067-4626Nextera XT kitIlluminaCat# FC-131-1096)Nextera XT index Kit V2IlluminaCat# FC-131-1001 Software Seurat3 package in R https://satijalab.org/seurat/get_started.html RRID:SCR_016341Stemchecker tool http://stemchecker.sysbiolab.eu RRID:SCR_025014Monocle2 http://cole-trapnell-lab.github.io/monocle-release/docs/ RRID:SCR_016339Velocyto http://velocyto.org/ RRID:SCR_018167ScVelo https://github.com/theislab/scvelo RRID:SCR_018168Python http://python-xy.github.io/ RRID:SCR_006903MSigDB http://software.broadinstitute.org/gsea/msigdb/index.jsp RRID:SCR_016863SCENIC https://github.com/aertslab/SCENIC RRID:SCR_017247Cytoscape http://cytoscape.org RRID: SCR_003032Fiji http://fiji.sc RRID: SCR_002285NDP.view 2 https://www.hamamatsu.com/eu/en/product/life-science-and-medical-systems/digital-slide-scanner/U12388-01.html RRID:SCR_025177QuPath https://qupath.github.io/ RRID:SCR_018257GraphPad Prism version 9.00 for Windows http://www.graphpad.com/ RRID:SCR_002798FlowJo https://www.flowjo.com/ RRID:SCR_008520Bio-Rad Image Lab SoftwareRRID:SCR_014210 Other

BD Biosciences FACSAria III Cell Sorter BD BiosciencesRRID:SCR_016695BioanalyzerAgilentCat# G2938CNextSeq 500 platformIlluminaRRID:SCR_014983qTower 2.0Analytik JenaRRID:SCR_027122Nanozoomer 2.0HamamatsuRRID:SCR_021658Olympus Slideview VS200EvidentRRID:SCR_024783Apotome 2 microscopeZeissRRID:SCR_024706M5000 EVOS inverted microscopeThermo Fisher ScientificRRID:SCR_023650GeoMx Digital Spatial ProfilerBruker (formerly nanostring)RRID:SCR_021660ChemiDoc Touch Imaging SystemBio-RadRRID:SCR_021693

Methods and protocols

Animals

Mouse colonies were housed in controlled conditions, on a 12/12-h light/dark cycle with normal food and water provided ad libitum. Pten^pc−/−^ male mice were generated by breeding Pten^loxP/loxP^ female mice with Pb-Cre4/Pten^loxP/loxP^ transgenic males and maintained on a mixed C57BL/6 and Sv/129 genetic background as described previously (Baures et al, 2022b; Sackmann Sala et al, 2017). Experiments were performed using 6- to 11-month-old mice, i.e., when aggressive malignant phenotypes were well established. As indicated, mice were surgically castrated, and the prostates were analyzed between 5 days and 2 months following castration. Prostate samples were obtained by microdissection immediately after sacrifice by cervical dislocation. Under a dissection microscope, adipose tissues were removed from the urogenital tract. The bladder, the ampullary gland and the urethra were removed to isolate the four prostate lobes. All animal procedures have been extensively described in previous publications (Baures et al, 2022b; Sackmann Sala et al, 2017).

Breeding and maintenance of mice were carried out in the accredited animal facility of the Necker campus in compliance with French and European Union regulations on the use of laboratory animals for research. Animal experiments were approved by the local ethical committee for animal experimentation (APAFIS authorizations #40276 and #49662).

JQ-1 and CX-6258 treatments

JQ-1 (Cat# HY-13030) and CX-6258 (Cat# HY-18095) were purchased from MedChemExpress. For in vivo treatments, 7-month-old Pten^pc−/−^ mice were castrated, then after 4-day recovery, they were randomized in two groups (n = 8 mice each) to receive either vehicle or a combination of JQ-1 (daily intraperitoneal injection, 50 mg/kg solubilized in 5% DMSO, 40% PEG300, 5% Tween80, and 50% H_2_O) and CX-6258 (biweekly oral administration, 100 mg/kg solubilized in 15% Cremophor and PBS), for 4 weeks. After euthanasia, the prostates were immediately microdissected and weighed, then fixed or processed for cell sorting, protein or RNA extraction, as described below.

For human PCa cell xenograft experiments, we used 12-week-old NSG male mice (NOD.Cg-Prkdc^scid^ Il2rg^tm1Wjl^/SzJ) purchased from Charles River Laboratories International Inc.. Animals were castrated and injected subcutaneously with 1.10^6^ PC-3 (ATCC, Cat# CRL-1435, RRID: CVCL_0035) in Matrigel on both flanks to fulfill the 3R rule. Four days later, mice were randomized into two groups of six mice each (n = 12 tumors) to receive vehicle or combined CX-6258 and JQ-1 treatment as described above. Before tumors were measurable, two mice in the treated group experienced issues with the gavage procedure that prevented continuation of oral treatment for ethical reasons. From day 15 of treatment, tumors were measured twice a week using a digital caliper. Tumor size was calculated using the formula (a × b2)/2, where a = length and b = width of the tumor. One tumor in the treated group initially grew abnormally fast (outlier) and regressed at the last timepoint (Appendix Fig. S7), therefore it was not included in the analysis. After euthanasia, tumors were dissected, weighed and photographed. Tumors were then embedded in paraffin for further analysis. Investigators who performed mouse analyses were blinded to mouse treatments (vehicle versus drugs).

Histological classification post-treatment and treatment toxicity

To evaluate the effect of treatments on prostate histology, prostates from Pten^pc−/−^ mice were fixed in 4% PFA, paraffin wax-embedded, and sections were stained with hematoxylin and eosin (H&E). Analysis was performed using the QuPath Software. Glands and lumen were manually detected and outlined. For each gland, we measured the number of lumens and the proportion of lumens to the total surface of the gland. Based on these parameters, glands were classified in four different categories: (1) normal gland (a single lumen representing >70% of the gland), (2) PIN structure (a single lumen representing <70% of the gland), (3) cribriform structure (two or more lumens representing >80% of the gland), and (4) malignant gland (two or more lumens representing <80% of the gland or absence of any lumen). Following this analysis, normal and PIN structures were classified as “benign” (versus “tumoral”). Investigators who performed mouse analyses were blinded to mouse treatments (vehicle versus drugs).

To evaluate the potential toxic effects of the administered drug combination, we analyzed organ samples stained with H&E. The kidneys, lungs, and livers from five Pten^pc−/−^ mice randomly chosen in both the control and treated groups were examined. Two pathologists blinded to group assignments independently performed the histological analysis. Degenerative and inflammatory lesions were rated on a five-point scale (0 = no lesion, 1 = scattered, 2 = mild, 3 =  moderate, 4 = marked). These analyses were complemented by quantification of cell death using the TUNEL assay described below.

Spatial transcriptomics

Digital spatial profiling (GeoMx™, Bruker) was performed on prostate sections from two naïve Pten^pc−/−^ mice, two castrated and vehicle-treated Pten^pc−/−^ mice, and two castrated and JQ-1 + CX-6258-treated Pten^pc−/−^ mice. Five µm sections were baked for 30 min at 60 °C. The nuclei and the epithelial cells were stained using SYTO™ 83 Orange Fluorescent Nucleic Acid Stain (Cat# S11364, Thermo Fisher) and the Alexa Fluor 488 anti-mouse CD236 (EpCAM) antibody (Cat# AB237384, dilution 1/100, Abcam), respectively. In situ hybridization of RNA-directed DNA oligo probes (Nanostring Mouse Whole Transcriptome Atlas) was performed according to the manufacturer’s protocol. For each mouse, the EpCAM+ (i.e., epithelial) cells in five to seven regions of interest (ROI) in the DLP and five in the AP were selected.

Raw digital count conversion (DCC) files were processed using the GeomxTools package. Quality control included Grubb’s outlier detection, in which five probes were identified as outliers in at least 20% of the segments and were removed before downstream processing. All areas of interest (AOIs) displayed a gene detection rate greater than 5%, so all AOIs were retained. Gene-level filtering retained genes detected in at least 10% of segments, resulting in a final set of 13,641 genes. Data were normalized using Q3 normalization (see Dataset EV6). Differential expression analysis was performed using a linear mixed-effects model of the form ~ Treatment + (1 + Treatment|MOUSE) to incorporate fixed treatment effects and mouse-level random effects. The deregulated genes were considered for a p value <0.05 and a log2 fold change of 0.5. The LSC-med signature score genes were filtered with diff_pct > 0, p_val_adj <0.05 and by removing genes not specific to a single signature. Signature scores were calculated by computing the mean expression of the genes. Genes that were found to be differentially expressed (p value <0.05 and absolute value of log2 fold change estimate >0.5) in naïve vs castrated and to be signature genes of LSC-med 1 or 2 were selected.

Prostate dissociation

Prostates were minced using razor blades and digested in a solution of Dulbecco’s modified eagle medium (DMEM) (Thermo Fisher Scientific) containing 10% FBS (Eurobio), 1% penicillin/streptomycin (Pen/Strep) (Thermo Fisher Scientific), and 1 mg/mL Collagenase Type I solution (Thermo Fisher Scientific) for 1 h at 37 °C, followed by 5-min incubation at 37 °C in 0.05% Trypsin (Thermo Fisher Scientific). The digestion was stopped with a solution of DMEM containing DNase I (Roche). Cells were passed ten times through a 20-G syringe, then filtered through a 40-µm cell strainer to generate a single-cell suspension. Cells were subjected to differential centrifugation using Histopaque-1119 (Sigma-Aldrich) to reduce the overall level of secretion in the sample. An aliquot of cells was stained with Trypan blue (Thermo Fisher Scientific) and counted using a hemocytometer to assess the cell viability.

FACS-sorting of LSCmed cells

The procedure for cell sorting was performed as previously described (Lukacs et al, 2010; Sackmann-Sala et al, 2014; Sackmann Sala et al, 2017). Isolated prostatic cells were stained for FACS on ice for 30 min using the following rat antibodies (all from eBioscience): fluorescein isothiocyanate-coupled lineage (Lin) antibodies (anti-CD31, -CD45, and -TER-119; dilution 1/500), phosphatidylethanolamine-Cyanine7-coupled anti-EpCAM (1/500), phosphatidylethanolamine-coupled anti-CD49f (integrin alpha-6; dilution 1/50) and allophycocyanin-coupled anti-Sca-1 (lymphocyte antigen 6A-2/6E-1; dilution 1/150). Dead cells were stained using SYTOX Blue (Life Technologies). Cell sorting was performed on a BD FACS Aria III (BD Biosciences) in DMEM containing 2% FBS and 1% Pen/Strep. Lineage antibodies (CD31, CD45, and TER-119) were used to deplete hematopoietic, endothelial, and immune cells and the EpCAM antibody to distinguish epithelial from stromal cells. CD49f and SCA-1 markers were used to select LSC^med^ cells from the other epithelial cell types. Sorted cells were collected in DMEM medium supplemented with 50% FBS and 1% Pen/Strep.

Single-cell RNA sequencing and analyses

LSC^med^ cancer cells sorted from two pooled intact and two pooled castrated Pten^pc−/−^ mice were dispensed (BD FACS Aria III) in 96-well plates (VWR, DNase, RNase free) containing 2 μL of lysis buffer (0.2% Triton X-100, 4U of RNase inhibitor, Takara) per well. Plates were properly sealed and spun down at 2000×g for 1 min before storing at −80 °C. Whole transcriptome amplification was performed with a modified SMART-seq2 protocol as described previously (Picelli et al, 2014) using 23 instead of 18 cycles of cDNA amplification. PCR purification was realized with a 0.8:1 ratio (ampureXP beads:DNA). Amplified cDNA quality was monitored with a high-sensitivity DNA chip (Agilent) using the Bioanalyzer (Agilent). Sequencing libraries were performed using the Nextera XT kit (Illumina) as described previously (Picelli et al, 2014) using 1/4th of the recommended reagent volumes and 1/5th of input DNA with a tagmentation time of 9 min. Library quality was monitored with a high-sensitivity DNA chip (Agilent) using the Bioanalyzer (Agilent). Indexing was performed with the Nextera XT index Kit V2 (A–D). Up to 4 × 96 single cells were pooled per sequencing lane. Samples were sequenced on the Illumina NextSeq 500 platform using 75 bp single-end reads.

Single-cell transcriptome analysis

The merged gene expression matrix (raw counts) containing all samples was analyzed with the Seurat3 package and Seurat_Extend (Hua et al, 2025). Cells were filtered by nFeature_RNA (genes detected) >3000 and <7000, percent.mt (percentage of mitochondrial genes) <10. This resulted in n = 306 valid cells for downstream analysis. After filtering cells, log-normalization was performed using the default NormalizeData function in Seurat. The scaled data were regressed for cell cycle phases (S.Score and G2M.Score) and number of genes per cell (nFeature_RNA) before performing UMAP-based dimension reduction (dims = 1:20, resolution = 0.4).

StemChecker analysis

Seurat cluster (LSC^med^-0, LSC^med^-1, LSC^med^-2) characteristic murine genes (adj.pval >0.05) were analyzed for “stemness” enrichment using the stemchecker tool (RRID:SCR_025014), while masking cell cycle genes.

Pseudotime-ordering analysis

The raw count matrix (n = 306 cells) was used as input for Monocle analysis (2.14.0) (RRID:SCR_016339) and normalized by the M3Drop function. Dimension reduction was based on the variable genes (n = 1483) (max_components = 2, method =  “DDRTree”). Ordering of the cell was performed by the f orderCells (HSMM_myo, num_paths = 2, reverse = T) function and the resulting trajectory was colored by pseudotime, cell type and Seurat cluster information. Differential gene expression analysis between groups through a likelihood ratio test for comparing a full generalized linear model with additional effects to a reduced generalized linear model based on negative binomial distributions.

Differentiation trajectory analysis using velocyto and scVelo

To infer the directionality of differentiation, we employed the Velocyto pipeline (La Manno et al, 2018). Utilizing the *.bam files and cell metadata, we computed the spliced and unspliced RNA matrices. Differentiation trajectories were visualized using UMAP embeddings, which were generated with Seurat in R (RRID:SCR_016341). The trajectories, represented by arrows, were plotted using ScVelo (Bergen et al, 2020) in Python.

Gene set enrichment analyses

Gene enrichment analyses were conducted for each Seurat cluster by interrogating the C2 Chemical and Genetic Perturbations category of MsigDB via the hypeR (2.2.0) package and the GeneSetAnalysis function from the SeuratExtend package.

We employed the Hallmark 50 gene set, a collection of well-defined, biologically relevant gene sets from the Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/gsea/msigdb/human/genesets.jsp?collection=H). This analysis was performed to identify significantly enriched pathways across the LSC^med^-0, LSC^med^-1, and LSC^med^-2 clusters. The results were visualized using a heatmap, with color intensity representing z-score normalized enrichment scores.

SCENIC analysis

To investigate regulon networks, we employed a comprehensive approach combining computational analysis and visualization techniques. We utilized the SCENIC algorithm (https://github.com/aertslab/SCENIC) to infer gene regulatory networks and regulon activities. The analysis pipeline was implemented using the SeuratExtend package for data processing and visualization. We generated waterfall plots to highlight differential TF regulon activities between selected clusters and other cells, with the top ten TFs labeled for clarity. Gene regulatory networks were visualized in Cytoscape (http://cytoscape.org), with nodes representing genes (round nodes) and TFs (square nodes). Node colors were mapped to relative gene expression or regulon activity using z-score normalization, enabling comparison of expression and activity patterns among LSCmed-0, LSCmed-1, and LSCmed-2 clusters. To manage the visualization of large regulons, we capped the number of target genes per transcription factor (TF) at 200, prioritizing the most variable genes based on their ranking from Seurat’s FindVariableFeatures function. This approach ensured a balance between capturing key regulatory relationships and maintaining analytical tractability.

Cell lines and treatments

HPV-10 (ATCC, Cat# CRL-2220), PC-3 (ATCC, Cat# CRL-1435) and LNCaP (ATCC, Cat# CRL-1740) PCa cell lines were purchased from ATCC (Manassas, USA) at the beginning of this study. Frozen stocks were generated between passages 2 to 5 of initial cell cultures, then aliquots were thawed for each experiment and used before passage 20. Therefore, no cell authentication was performed. Cells were not tested for mycoplasma. None of the cells used in this study are listed in the ICLAC database of commonly misidentified lines (https://iclac.org/databases/cross-contaminations/).

Cells were grown at 37 °C and 5% CO_2_. HPV-10 cells were cultured in Keratinocyte Serum Free Medium (K-SFM) (Thermo Fisher Scientific, Cat# 17005-034) supplemented with 0.05 mg/mL bovine pituitary extract (BPE) (Thermo Fisher Scientific, Cat# 13028-014), 5 ng/mL EGF (PeproTech, Cat# AF-100-15-500UG), and 1% Pen/Strep. PC-3 and LNCaP cells were cultured in DMEM (Thermo Fisher Scientific, Cat# 31966-021) supplemented with 10% FBS (Eurobio Scientific, Cat# CVFSVF00-01, #lot S80515) and 1% Pen/Strep. Medium was changed every other day.

For drug treatments, HPV-10 and PC-3 cells were plated in six-well plates at 50,000 and 30,000 cells/well, respectively. After 3 days, cells were treated with DMSO, JQ-1 and CX-6258, alone or combined, and analyzed after 72 h, according to the protocols described below. Alternatively, HPV-10 and PC-3 cells were plated in 24-well plates at 50,000 cells/well. The next day, cells were treated with DMSO or T5224-PROTAC (5 or 15 µM) at 0, 8, and 24 h, and analyzed at 48 h.

For siRNA transfections, PC-3 cells were plated in 24-well plates at 30,000 cells/well. The next day, cells were treated with Lipofectamine (Thermo Fisher Scientific, Cat# 18324-012) with siScrambled or siFOSL1 in OPTIMEM (Thermo Fisher Scientific, Cat# 31985-070) for 6 h before changing the medium back to complete medium. Cells were analyzed at 72 h.

For cell number analysis, wells were rinsed in PBS, and adherent cells were trypsinized before being manually counted using a Malassez counting cell. For cell viability analysis, culture medium (containing floating cells) and adherent cells (trypsinized) were collected, centrifuged (300 g) and resuspended in HBSS before being counted with trypan blue as above. Cell viability was quantified as the ratio of unstained versus total (unstained + stained) cells.

Reverse transcription-quantitative PCR (RT-qPCR)

RNA extractions from Pten^pc−/−^ LSC^med^ cells and human PCa HPV-10 cells were performed with the Nucleospin RNA XS and Nucleospin RNA (Macherey-Nagel), respectively, following the manufacturer’s protocols. Reverse transcription was carried out using the SuperScript™ VILO™ cDNA Synthesis Kit (Invitrogen) for murine cells and the GoScript™ Reverse Transcriptase (Promega) for the human cell line, according to the manufacturer’s instructions.

For qPCR, iTaq Universal SYBR Green Supermix (Promega) was used, and reactions were run on a qTower 2.0 real-time thermal cycler (Analytik Jena). Expression data were normalized to Cyclophilin A for murine samples and Actin B for human samples.

Primer sequences (Sigma-Aldrich) are listed in Appendix Table S3.

Immunoblotting

Cell lysates (cell lines and sorted LSC^med^ cells) were prepared on ice in an appropriate lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 0.5% NP-40, 10% glycerol, 1% protease, and a Phosphatase Inhibitor Cocktail (Cat# 78442, Thermo Fisher Scientific)). Protein concentrations were determined with the Pierce BCA protein assay kit (Cat# 23225, Thermo Fisher Scientific). Equal protein amounts (30–40 µg) diluted in a 4× Laemmli buffer were denatured by heating at 95 °C for 5 min and separated by electrophoresis on 4–12% NuPAGE Bis-Tris Gel, and then transferred onto a 0.45-μm PVDF membrane. Membranes were blocked with 5% non-fat dry milk in PBS-T (PBS with 0.1% Tween-20) for 1 h at room temperature and then incubated overnight at 4 °C with anti-FRA1 (FOSL1) antibody (Santa-Cruz Biotechnologies, sc-28310, dilution 1/1000) or anti-GAPDH antibody (Cell Signaling Technologies, clone 5174, dilution 1/1000). Membranes were then washed with PBS-T, and incubated with the appropriate HRP-coupled secondary antibody for 1 h 30 min at RT. Membranes were then washed with PBS-T, and bound antibodies were detected using an ECL detection kit (Immobilon Western ECL, Millipore) and ChemiDoc Imaging Systems (Bio-Rad) using the CCD camera for light capture according to the manufacturer’s protocol. Signals were quantified using Image Lab Software (Bio-Rad) and normalized to GAPDH.

Organoid and tumorsphere culture and treatment

We used the reference protocol for prostate organoid culture described by Clevers’ lab (Drost et al, 2016) with modifications. LSC^med^ cells sorted from Pten^pc−/−^ mice and human PCa HPV-10 cells were plated in triplicate on a low-growth factor-containing Matrigel (Corning) layer in a 96-well plate (Sarstedt) at concentrations of 3000 or 1000 cells/well, respectively, and were cultured at 37 °C and 5% CO_2_. After one day of incubation, the culture medium was removed, and the cells were covered by a new layer of Matrigel to generate 3D cell structures. The organoid-forming capacity did not differ from the efficacy obtained using 3D droplet culture, earlier described (Drost et al, 2016). Drugs were added at this stage, and the medium (+drugs) was changed every other day. After 10 days of culture, organoids were fixed in 4% PFA, and images were taken with a 4x objective on an M5000 EVOS inverted microscope (Invitrogen) to capture the entire surface of the well. Organoid counting and surface area measurements were performed using Fiji software (http://fiji.sc) by manually outlining the organoid surface. Only structures with a size superior to 2000 μm^2^ were considered as an organoid. The number and size of organoids in the various experimental conditions were normalized to the cognate mean value obtained in the non-treated condition.

Immunohistochemistry (IHC) and immunofluorescence (IF)

For IHC, murine prostate samples were fixed in 4% PFA, paraffin wax-embedded, and 4 µm sections underwent heat-induced antigen retrieval in citrate buffer at pH 6 (95 °C, 30 min). IHC was performed as previously described (Baures et al, 2022b) using antibodies directed against Ki-67 (Zytomed Systems, Cat# RBK027-05; dilution 1/100), CD44 (Biolegend, Cat# 103001, dilution 1/200), and FOSL1 (FRA1, Cell signaling, Cat# 28801S, 1/50). Signal was amplified and detected using the Vector Elite ABC-HRP kit with DAB substrate (Vector Laboratories), and nuclei were counterstained with hematoxylin. Slides were scanned with a Nanozoomer 2.0 (Hamamatsu) or Olympus Slideview VS200 (Evident) and analyzed using NDP.view 2 software (Hamamatsu). CD44-positive cells and Ki-67-positive nuclei were quantified in ten random fields per slide using QuPath Software (https://qupath.github.io/). FOSL1-positive nuclei were quantified with QuPath after manual selection of epithelial regions. Classification of positive nuclei according to staining intensity was performed using predefined threshold values in QuPath.

For IF, prostate samples were fixed in 4% PFA, paraffin wax-embedded, and 4 µm sections underwent heat-induced antigen retrieval in citrate buffer at pH 6 (95 °C, 30 min) before permeabilization in PBS supplemented with 0.2% Tween-20 and 0.1% Triton. Antibodies directed against KRT4 (Abcam, Cat# ab51-599, dilution 1/150), CD44 (Biolegend, Cat# 103001, dilution 1/150), FOSL1 (Cell Signaling, Cat# 28801S, dilution 1/20), and P63 (BioSB, Cat# BSB3605, dilution 1/150) were incubated overnight at 4 °C and secondary antibodies were incubated during 1 h at room temperature. Nuclei were stained with Hoechst dye. Slides were scanned with Olympus Slideview VS200.

For IF, HPV-10 cells were fixed in PFA 4% for 15 min at room temperature and permeabilized for 15 min in PBS supplemented with 0.2% Tween-20. Rabbit anti-KRT4 (Abcam, ab51-599, dilution 1/100) and anti-KRT13 (Sigma-Aldrich, HPA030877, dilution 1/200) antibodies were incubated overnight at 4 °C and secondary antibodies were incubated for 1 h at room temperature. Nuclei were stained with Hoechst dye. Samples were analyzed using a 40× objective under an Apotome 2 microscope (Zeiss).

Terminal deoxynucleotidyl transferase dUTP nick end (TUNEL) labeling

The TUNEL assay was performed with the In Situ Cell Death Detection Kit (Roche, Cat# 11684809910) according to the manufacturer’s instructions. Briefly, samples underwent dewaxation and rehydration. Slides were heated in citrate buffer for 5 min at 350 W, and then incubated for 1 h at 37 °C, with the TUNEL reaction mixture. Fluorescence was observed, and images were taken using a 20x objective under an M5000 EVOS inverted microscope.

Statistical analyses

Unless otherwise stated, the data describe biological replicates, and the number of independent experiments is indicated in Figure captions. Error bars represent SD. The statistical tests performed are indicated in figure captions for each experiment (ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). A value of p < 0.05 was used as a significance cutoff for all tests. Statistical analyses were performed using GraphPad Prism version 9.00 for Windows (http://www.graphpad.com/). Statistical parameters used for transcriptomic data analyses are described in the corresponding Methods section. Animals were randomized into experimental groups. Investigators who performed histopathological analyses of mouse tissues were blinded to mouse treatments (vehicle versus drugs).

Graphics

Figures 1A, 3A, 6A,G were created with BioRender.com

Supplementary information

Appendix Dataset EV1 Dataset EV2 Dataset EV3 Dataset EV4 Dataset EV5 Dataset EV6 Source data Fig. 2 Source data Fig. 3 Source data Fig. 5 Source data Fig. 6 Expanded View Figures