Innate immunity and the NF-κB pathway control prostate stem cell plasticity, reprogramming and tumor initiation

Abstract

Prostate epithelium develops from multipotent stem cells (SCs) which are replaced in adult life by different lineage-restricted basal and luminal unipotent SCs. Deletion of Pten re-induces multipotency in basal cells (BC). However, the molecular mechanisms regulating BC plasticity and tumor initiation are poorly understood. Here, we showed that Pten deletion in BCs led to distinct cell fate reprogramming and tumor initiation in a regionalized manner. Single-cell RNA-seq, ATAC-seq and in situ characterization revealed that following Pten deletion in anterior and dorsolateral prostates, BCs were highly plastic and reprogrammed into a Hillock-like state, progressing into a proximal-like luminal state before giving rise to invasive tumors. This BC reprogramming was associated with the activation of innate immunity. Pharmacological targeting of Il-1, JAK/STAT and Nfkb as well as genetic deletion of Nfkb inhibit Pten-induced cell plasticity and reprogramming in a cellular autonomous manner, opening new opportunities for prevention and treatment of prostate cancer.

Introduction

The Prostate epithelium is composed of basal cells (BCs), luminal cells (LCs), and rare neuroendocrine cells¹. It develops from multipotent basal stem cells (SC) which are largely replaced in adult life by distinct pools of unipotent basal and luminal stem cells^2–6. Adult unipotent basal SCs can reactivate multipotency in response to prostate inflammation, LC ablation and oncogenic mutations^4–8. The mouse prostate epithelium is composed by three lobes: the anterior prostate (AP), the dorsolateral prostate (DLP) and the ventral prostate (VP). These different prostate regions expressed different transcriptional program and are maintained by their own pool of progenitors^9–12. The proximal prostate at the junction with the urethra is sustained by Trop2/Krt4 expressing LCs whereas the distal prostate is sustained by LC expressing Nkx3.1^9–13.

Prostate cancer (PC) is one of the most frequent cancers and the second cause of death in men^14,15. Most prostate tumors are acinar adenocarcinoma expressing androgen receptor (AR)¹⁴. The most frequently mutated genes in prostate cancer consist of different gene fusions activating ETS transcription factors and deletion of the tumor suppressor gene Pten^14,15. Pten deletion in mouse prostate leads to similar tumors as found in human and that progress from prostate intraepithelial neoplasia lesions (PINs) into invasive adenocarcinoma^{4–6,13–17}. Genetic lineage tracing and transplantation experiments have demonstrated that both BC and LC can serve as cell of origin of prostate cancer^{4–6,13,18–21}. Conflicting findings have been reported to whether BC or LC led to the most aggressive prostate tumors^4–6. However, how the spatially distinct prostate SC states identified by scRNA-seq contribute to prostate tumor initiation is currently unknown. Transplantation experiments have shown the greater potential of BC over LC to serve as the cells of origin of prostate cancer^18–20. Pten deletion in BC give rise to LC before progressing into prostate cancer^4–6. The mechanisms by which Pten deletion activates cell plasticity in BC and consequently lead to tumor initiation is not well understood.

In this study, we combined mouse genetic lineage tracing, single-cell transcriptional, chromatin profiling and functional experiments in mice in vivo and in organoids in vitro to investigate the importance of the cells of origin in controlling tumor heterogeneity and the mechanisms regulating oncogene-induced SC plasticity and transcriptional reprogramming during prostate tumor initiation.

Results

Region dependent BC plasticity and tumor progression

To define the mechanisms regulating basal to luminal transition upon Pten deletion in BC, we first assessed the temporal kinetic of LC appearance by monitoring by flow cytometry (FACS) the proportion of YFP positive BC (CD49f^High EpCAM⁺) and LC (CD49f^Low EpCAM⁺)²² following tamoxifen (TAM) administration to K5CreER/Pten^fl/fl/Rosa-YFP mice (Fig. 1a) in the different prostate lobes. At 1 week following TAM administration, only BC were YFP+ demonstrating the specific targeting of BC by the K5CreER. The first YFP+ LCs appeared between 4–6 weeks and gradually increased over time. This cell fate switch occurred more rapidly in the AP and DLP as compared to VP (Fig. 1a-d and Extended Data Fig. 1a-e). Immunofluorescence of prostate with basal (K14) and luminal (K8) markers demonstrated the presence of large patches of YFP+ LCs and the existence of basal/luminal hybrid cells co-expressing basal and luminal markers within the same cells, supporting the notion that basal to luminal transition passes through a hybrid state (Fig. 1e and 1f). EdU pulse-chase experiments following Pten deletion in BC showed that symmetric BC division (BC-BC EdU doublets) and asymmetric BC division (BC-LC EdU doublets) occurred with the same frequency, suggesting that BCs balance self-renewal and LC differentiation upon Pten deletion (Extended Data Fig. 1f and g). Histological characterization of tumor initiation showed that the first signs of high-grade PINs (HGPIN) occurred concomitantly to the hybrid state and basal to luminal transition and the first signs of invasive adenocarcinoma (ADC) transition were visible at 12 weeks following Pten deletion in AP and DLP (Fig. 1g and h). In contrast, at the same time point, only atypical intraductal cribriform proliferations of the prostate (AIP) and focal intraductal carcinoma of prostate (IDC-P) with no invasive ADC were observed in the VP, suggesting that tumor progression occurs more slowly in VP compared to DLP and AP upon Pten deletion in BC (Fig. 1g and h). At 6 months, invasive ADC showed a more homogeneous pattern in AP and DLP, whereas in VP foci of invasive ADC alternate with HGPIN, AIP and IDC-P lesions (Fig. 1g and h), further showing the delay in tumor progression in BC derived tumorigenic lesions in the VP as compared to the DLP and AP.

Figure 1. Pten deletion induced BC plasticity and tumor initiation in a cell of origin and region-specific manners.

(a) Genetic strategy to lineage trace BC following Pten deletion. (b) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells from K5CreER/Pten^fl/fl/RosaYFP mice (K5-PTEN) and (c) quantification of the % of YFP+ LCs in total YFP+ cells at indicated time after TAM administration. n=3 mice (1w, 4w, 7w, 8w and 20w), n=4 mice (6w and 12w). (d) Quantification of the % of YFP+ BCs and LCs in total BCs and LCs in K5-PTEN mice. n=6 mice (1w), n=9 mice (6-8w). (e) Representative images of immunostaining and (f) quantification of the % of YFP+ K14+ in total K14 cells, hybrid cells in total YFP+ cells, YFP+ K8+ in total K8+ cells using indicated antibodies. Arrows indicate hybrid cells (YFP+ K14+ K8+). Scale bar, 50 μm (up), 10 μm (down). n=3 mice. (g) H&E-stained histological sections and (h) quantification of different types of tumorigenic lesions along prostate tumor progression in K5CreER/RosaYFP (CTL) and K5-PTEN mice at 8 weeks and at indicated time after TAM administration, respectively. Scale bar, 50 μm. Arrows indicate HGPIN, and arrowheads indicate invasive ADC. Number of mice (n) is indicated. (i) Gene ontology (GO) analysis of genes upregulated by more than 2-fold in FACS isolated YFP+ LCs arising from Pten deleted BCs from AP/DLP compared to VP in K5-PTEN mice 6 months following TAM administration. (j) Relative mRNA expression from bulk RNA-seq of FACS isolated LCs from AP/DLP versus VP (n=3 for AP/DLP, n=2 for VP). (k) H&E-stained (up) and Col1a1 immunostaining (down, red) and (l) quantification stromal histological appearance. S indicates stroma region. Scale bar, 50 μm. Number of mice (n) is indicated. (m) Representative images of immunostaining of the prostate stroma using anti-GFP (green), anti-CD45 (red) and (n) quantification of CD45 positive cells in 0.1mm² prostate section area. Scale bar, 20 μm. n=5 mice. Graphs are mean ± s.e.m. p-values are derived from two-tailed modified Fisher’s Exact Test with Benjamini–Hochberg correction (i), two-way (d) and one-way (n) ANOVA with Tukey’s test.

RNA-seq of FACS isolated LCs of different lobes at 6 months following Pten deletion in BC showed a strong upregulation of Collagen, ECM, metalloprotease and innate immunity/inflammatory signaling pathways including many chemokines and cytokines in the AP/DLP as compared to the VP in BC derived LC tumors (Fig. 1i and j). In situ characterization of the tumors showed a major difference in the stromal composition across the different lobes with a much stronger inflammatory/immune and fibroblastic/desmoplastic infiltration as well as collagen (Col1a1) deposition in the AP/DLP compared to the VP (Fig. 1k-n).

Interestingly, deletion of Pten in LC using K8CreER/Pten^fl/fl/Rosa-YFP mice did not promote luminal SC plasticity and multipotency (Fig. 2a-d and Extended Data Fig. 1h, i). In AP and DLP, tumor progression was slower in LCs derived tumors as compared to BC derived tumors as shown by the absence of high-grade lesions in K8CreER/Pten mice at 6-8 weeks and the paucity of florid IDC-P at 12 weeks. In contrast, tumors progressed faster following Pten deletion in LC compared to BC in the VP (Fig. 2e and f).

Figure 2. Pten deletion in LCs does not promote multipotency and is associated with slower tumor progression in AP/DLP compared to BC derived tumors.

(a) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells from K8-PTEN mice and (b) quantification of the % of YFP+ BCs and LCs in total YFP+ cells at indicated time after TAM administration. n=3 mice. (c) Representative prostate immunostaining and (d) quantification of the % of YFP+ K14+ in total K14 cells, YFP+ K8+ in total K8+ cells from K8-PTEN mice 4 weeks after TAM administration using indicated antibodies. Scale bar, 50 μm. n=2 mice. (e) H&E-stained histological sections and (f) quantification of different types of tumorigenic lesions along prostate tumor progression. Scale bar, 50 μm. Arrows indicate HGPIN, and arrowheads indicate invasive ADC. Number of mice (n) is indicated. (g) FACS isolated of YFP+ LC tumors in AP/DLP from BC and LC derived tumors 6-month after TAM administration. (h) GO term of genes upregulated more than 2-fold in FACS isolated YFP+ LCs arising from BC versus LC derived tumors in AP/DLP. (i) Relative mRNA expression from Bulk RNA-seq of FACS isolated YFP+ LCs arising from BC versus LC derived tumors in AP/DLP. n=3 mice. (j) H&E-stained (up) and Col1a1 immunostaining (down, red) and (k) quantification of stromal histological appearance from BC and LC derived AP tumors 6-month after TAM administration. Scale bar, 50 μm. S indicates stroma region. Number of mice (n) is indicated. (l) Representative immunostaining of the myeloid infiltration and (m) quantification of Ly6G positive cells from BC and LC derived AP tumors 6-month after TAM administration using anti-GFP, anti-Ly6G antibodies. Scale bar, 50 μm. n=3 mice. (n) Immunostainings of EdU and GFP and (o) quantification of EDU positive cells in YFP positive cells from control, BC and LC derived AP/DLP tumors 3 months after TAM administration 24 hours following EDU injection. Scale bar, 20 μm. n=3 mice. Graphs are mean ± s.e.m. p-values are derived from two-tailed modified Fisher’s Exact Test with Benjamini–Hochberg correction (h), two-sided unpaired t-test (m) and one-way ANOVA with Tukey’s test (o).

To understand the reasons for the difference in tumor progression in BC and LC derived tumors in AP/DLP, we performed RNA-seq of FACS isolated LCs at 6 months following Pten deletion in BC or in LC (Fig. 2g). Interestingly, we observed a strong decrease in collagen, ECM, cell cycle and innate immunity/inflammatory signaling pathways in AP/DLP tumors originating from LC as compared to BC (Fig. 2h and i). The difference in gene expression between BC and LC derived tumors in AP/DLP was associated with a decrease in stromal infiltration, collagen deposition and Ly6G+ myeloid cell recruitment with decrease in tumor cell proliferation (Fig. 2j-o).

Altogether these data indicate that the cells of origin of prostate tumors control SC plasticity and tumor progression upon Pten deletion.

Combined oncogenic hits increase BC plasticity

Activating mutations in PIK3CA have been found in human prostate cancers and Pten deletion together with oncogenic Pik3ca expression accelerate prostate cancer formation in mice²³. To assess whether increasing the strength of PI3K signaling accelerates basal to luminal transition, we generated a double mutant mouse expressing Pik3ca^H1047R together with Pten deletion (K5CreER/Pten^fl/fl/Pik3ca^H1047R/Rosa-YFP) (Fig. 3a). We found that basal to luminal transition was dramatically accelerated in these double mutant mice (Fig. 3b-d). Immunostaining analysis revealed the presence of large patches of YFP+ LCs and hybrid basal/luminal cells at 3 weeks following TAM administration in AP and DLP (Fig. 3e and f). This major acceleration in basal to luminal transition was accompanied by a dramatic acceleration of tumorigenesis in DLP and AP (Fig. 3g and h).

Figure 3. Pten Deletion together with oncogenic Pik3ca expression or p53 deletion accelerate BC plasticity.

(a) Genetic strategy to lineage trace BC with Pten deletion and oncogenic Pik3ca expression. (b) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells and quantification of the % of (c) YFP+ BCs in total BCs and (d) YFP+ LCs in total LCs in K5-PTEN (n=3) and K5-PTEN-PIK mice (n=5) 3 weeks after TAM administration. (e) Representative prostate immunostainings of K5-PTEN-PIK 3W mice and (f) quantification of the % of hybrid cells in total YFP+ cells, YFP+ K8+ in total K8+ cells using indicated antibodies. Arrows indicate hybrid cells (YFP+ K14+ K8+). Scale bar, 50 μm (up), 20 μm (down). n=3 mice. (g) H&E-stained histological sections and (h) quantification of different types of tumorigenic lesions along prostate tumor progression. Scale bar, 50 μm. n=5 mice. (i) Genetic strategy to lineage trace BC with Pten and P53 deletion. (j) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells and quantification of the % of (k) YFP+ BCs in total BCs and (l) YFP+ LCs in total LCs in K5-PTEN (n=3) and K5-PTEN-P53 mice (n=5) 4 weeks after TAM administration. (m) Representative prostate immunostainings of K5-PTEN-P53 4W mice and (n) quantification of the % of hybrid cells in total YFP+ cells and YFP+ K8+ cells in total K8+ cells using indicated antibodies. Arrows indicate hybrid cells. Scale bar, 20 μm. n=3 mice. (o) Representative immunostaining 1 week after AdeK5Cre injection in AP and (p) quantification of YFP+ K14+ and YFP+K8+ in total YFP+ cells using anti-GFP, anti-K8 and anti-K14 antibodies. Scale bar, 20 μm. n=3 mice. Representative immunostaining of (q) Pten^fl/fl/Pik3ca^H1047R/RosaYFP and (s) Pten^fl/fl/P53^fl/fl/RosaYFP mice 2 months after AdeK5Cre injection. Scale bar, 50 μm. Quantification of the % of hybrid and YFP+ K8+ cells in total YFP+ cells in (r) Pten/Pik (n=3) and (t) Pten/P53 mice (n=2). Graphs are mean ± s.e.m. p-values are derived from two-way (d, l) and one-way (f, n) ANOVA with Tukey’s test.

Deletion of Trp53 and Pten is found in human prostate cancers and accelerate prostate tumorigenesis in mice^24,25. The co-deletion of Trp53 and Pten in BC and LC using Probasin-CRE lead to the occurrence of hybrid basal and luminal cells found mainly in early PIN lesions and progression of tumors with epithelial-mesenchymal transition (EMT) features²⁶. Trp53 deletion together with Pten deletion greatly accelerated BC plasticity and the differentiation of BC into LCs in the DLP and AP and not in the VP (Fig. 3i-l). Immunostaining analysis showed the presence of large patches of YFP+ LCs and hybrid basal/luminal cells at 4 weeks following TAM administration in AP and DLP (Fig. 3m and n).

These two mouse models exhibited severe hyperplastic growth in the facial skin and eyelids, leading to poor general health that requires to terminate the experiments 6 weeks after TAM administration, preventing studying the long-term consequence of these double mutant mice.

To exclude that the phenotypes observed using K5CreER is not due to a non-cellular mechanism related to the expression of the K5CreER in other epithelia such as the skin and the esophagus, we performed co-deletion of Pten/Pik3ca and Pten/Trp53 using intra-prostatic injection of adenovirus expressing CRE under the K5 promoter (Ade K5Cre) in Pten^fl/fl/Pik3ca^H1047R/RosaYFP and Pten^fl/fl/P53^fl/fl/RosaYFP mice. Immunostaining revealed that one week after intraprostatic injection of Ade K5Cre only BC were YFP labelled, showing the specificity of the initial BC targeting by the adenovirus-Cre (Fig. 3o and p). Two months following oncogenic recombination in BC, BCs massively differentiated into LC with the presence of basal/luminal hybrid cells in both mouse models (Fig. 3q-t), demonstrating that prostate SC plasticity is not mediated by inflammation induced in other organs targeted by the K5CreER.

Altogether, these results demonstrate that BC plasticity and the generation of hybrid cells following oncogenic hits in prostate BCs is associated with tumorigenesis in a region specific and in an oncogenic dosage-dependent manner.

mTOR inhibition blocks BC plasticity and tumor initiation

Pten deletion and oncogenic Pik3ca have been shown to activate Akt, which in turn activates mTOR signaling²⁷. Inhibition of Akt and mTOR signaling has been used in combination with anti-androgen therapy to treat castration resistant prostate cancer in preclinical models^27–29. However, the role of mTOR signaling in the regulation of cell plasticity and the activation of BC multipotency and prostate tumor initiation is currently unknown. To assess whether mTOR signaling can regulate BC multipotency and the early step of prostate tumor initiation, we treated K5CreER/Pten^fl/fl/Rosa-YFP mice with Rapamycin, the first developed mTOR inhibitor³⁰, after oncogenic recombination and assessed basal to luminal transition 6 weeks after TAM administration. Rapamycin administration dramatically decreased the proportion of YFP+ LCs arising from the differentiation of BC following Pten deletion (Extended Data Fig. 2a-c). Rapamycin treatment also strongly inhibited BC multipotency, basal-luminal hybrid state and the differentiation of BC into LCs following combined Pten deletion and Pik3ca activation (Extended Data Fig. 2d-f).

Importantly, the strong inhibition of BC multipotency following Pten deletion was accompanied by a major reduction in tumorigenesis (Extended Data Fig. 2g and 2h). Altogether, these results show that mTOR signaling is critical for the activation of BC multipotency following oncogenic hits and inhibition of mTOR signaling by Rapamycin blocks the early step of prostate tumor initiation.

Regionalization of BC reprogramming upon Pten deletion

To decipher the molecular mechanisms associated with the basal to luminal transition that accompanies prostate tumorigenesis, we performed single-cell RNAseq (scRNA-seq) of FACS isolated epithelial cells 6-weeks following Pten deletion in K5CreER/Pten^fl/fl/Rosa-YFP. Unsupervised clustering of WT prostate epithelial cells showed the presence of well separated BC and LC clusters (Fig. 4a), corresponding to the different prostate lobes (Dorsal, lateral and ventral) and different regions within the prostate (proximal and distal), the anterior and dorsal distal lobes give rise to one cluster called antero-dorsal (AD), whereas the proximal, lateral and ventral lobes give rise to three additional clusters as previously reported (Extended Data Fig. 3a-g)^9–11. In contrast, upon Pten deletion, we found a continuum of cells between the BC clusters and LC clusters with a major expansion of the proliferative BC cluster and the proximal prostate cluster (Fig. 4b and Extended Data Fig. 3h-o). Lineage trajectory analysis using Slingshot revealed that upon Pten deletion, there was a stepwise reprogramming of BC into Hillock-like cells characterized by the expression of Krt13, previously described in the human healthy prostate and in human prostate cancers^9,31–34, then hybrid BC and Proximal LCs (Ly6d, Krt6a) that finally gave rise to proximal LCs (Krt4, Psca, Clu, Wfdc2, Pigr, Ltf) (Fig. 4c-f and Extended Data Fig. 3h-k). Analysis of the proportion of expressed basal and luminal marker genes during the lineage trajectory showed a progressive decrease of BC genes and a progressive increase of LC genes along the differentiation pathway (Extended Data Fig. 4a-d). By analyzing the proportion of BC and LC marker expression across the different cluster cell populations^8,35, we found that the hybrid cell cluster co-expressed a high proportion of basal and luminal genes within the same cells, confirming their hybrid state (Fig. 4g-i). To visualize the spatial localization of the hillock, hybrid and proximal-like LC that appears following Pten deletion, Pten/p53 deletion and Pten/Pik3ca^H1047R, we performed co-immunofluorescence using representative markers of these three states: Hillock (Krt14, Krt13, Aqp3), Hybrid (Krt14, Krt4, Aqp3) and LC proximal state (Krt8, Krt4, Trop2) (Extended Data Fig. 4e-h). Interestingly, this first cellular reprogramming into Hillock and proximal like states first occurred in the AP and DLP and not in the VP and the spatial localization of the different cell states identified by in situ characterization was consistent with the pseudo-time ordering predicted by the computational lineage trajectory analysis (Fig. 4j-o).

Figure 4. Region dependent BC reprograming following Pten deletion.

UMAP dimensionality reduction of scRNA-seq data shows unsupervised clustering of FACS-isolated prostate epithelial cells from (a) wildtype (CTL) and (b) YFP+ epithelial cells from K5-PTEN mice, 6 weeks after TAM administration (K5-PTEN 6w). Black arrows indicate two possible trajectories of basal-to-luminal cell reprogramming. Expression of (c) Krt13 and (d) Aqp3 and (e) Krt4 in CTL and K5-PTEN 6w. Color bar indicates gene expression level. (f) Slingshot pseudotime trajectory analysis illustrating the lineage trajectory from BC to Proximal like LC. Quantitative assessment of LC and BC marker gene expression for (g) all clusters, (h) hybrid BC proximal-like (HY BC Prox) and (i) HY Nkx3.1 in K5-PTEN 6w: Scatterplot with the x-axis representing the adjusted proportion of BC-specific marker genes and the y-axis representing the adjusted proportion of LC-specific markers. Representative immunostaining of the VP, DLP, AP proximal (AP prox) and AP distal of CTL and K5-PTEN 6w using anti-GFP, (j) anti-K13, (l) anti-Aqp3 and (n) anti-K4 antibodies. Scale bar, 50 μm. Quantification of the % of (k) K13+ cells, (m) Aqp3+ cells, (o) K4+ cells in total YFP+ cells in CTL and K5-PTEN 6w mice. n=3 mice. (p) UMAP plots colored by normalized gene expression values for Nkx3.1 gene expression in CTL and K5-PTEN 6w. Red outlines show BC expressing Nkx3.1 in CTL, whereas it highlights BCs expressing Nkx3.1 that pass through a hybrid cluster and end in ventral LCs in K5-PTEN 6w. (q) Slingshot pseudotime trajectory analysis for the trajectory from BC to Ventral LC. (r) Representative immunostainings of GFP, Nkx3.1 and K14 in VP, DLP, AP prox and AP distal of CTL and K5-PTEN prostate 6w after Pten deletion. Scale bar, 20 μm. (s) Quantification of the % of Nkx3.1+ cells in total YFP+ cells in CTL and K5-PTEN 6w mice. n=3 mice. (t) Cell plasticity, lineage infidelity and tumor progression following Pten deletion in BC occurs in a region-specific manner during the early stage of prostate cancer initiation. Graphs are mean ± s.e.m. p-values are derived from two-way ANOVA with Tukey’s test.

We found a second continuum of cells spanning from BC expressing high level of Nkx3.1 toward Ventral LCs expressing Spink1, Sbp or Sbpl passed through a hybrid state, co-expressing both basal and luminal markers (Fig. 4i, 4p and Extended Data Fig. 3m). Slingshot lineage trajectory inferred a second differentiation trajectory from BC expressing Nkx3.1 toward VP LCs ( (Fig. 4q and Extended Data Fig. 5a-d). In situ characterization of this second lineage trajectory showed that this cellular reprogramming and differentiation into Nkx3.1 LCs occurred in the VP (Fig. 4r and 4s). As these lesions progressed into adenocarcinoma, Nkx3.1 was progressively lost (Extended Data Fig. 5e). Like the cancers arising from VP BCs, LCs did not give rise to Hillock-like cells and less Krt4 expressing proximal states following Pten deletion in LCs (Extended Data Fig. 5f).

To gain further insights into the molecular mechanisms that promote BC reprogramming into the Hillock-like/Hybrid/Proximal Like LC differentiation path or BC Nkx3.1 toward the Ventral LC differentiation path, we performed differential gene expression analysis between the hybrid cells and the BC or their respective LC population. As compared to BC, the hybrid basal/luminal proximal population expressed higher levels of Krt4, Psca, Wfdc2 or Clu (Extended Data Fig.5g), in concordance with the high expression of these genes within the proximal prostate cluster^9–11,16. When compared to the proximal LCs, this hybrid cluster presented an increased expression of many genes of the BC identity such as Krt5, Krt14, Krt15, Krt17 or Col17a1 (Extended Data Fig. 5h). The overlapping upregulated genes, including Aqp3, Ly6d and Krt6a, exhibit a distinct hybrid gene signature (Extended Data Fig. 3k, 5g, 5h). In the other trajectory, when comparing the gene expression of the hybrid Nkx3.1 with BCs, we found a higher level of expression of C1rb, Nkx3.1, Fgl1, Tgm4 (Extended Data Fig. 5i). Conversely, when compared with ventral LC, this second hybrid population expressed higher level of basal genes such as Krt14 or Krt5 (Extended Data Fig. 5j). Altogether, these data show that the two different BC populations targeted by Pten deletion will undergo distinct reprogramming giving rise to two distinct lineage differentiation trajectories toward different LC lineages in a region-specific manner (Fig. 4t).

BC plasticity is associated with innate immunity activation

To gain further insights into the gene regulatory networks controlling the two differentiation paths, we performed regulatory network analysis using SCENIC³⁶, which infers the transcription factors (TFs) and their downstream target genes (regulons) that are active in the different cell clusters corresponding to the different differentiation paths. In the first differentiation path, we found that many TFs relaying inflammation and innate immunity pathways such as Irf6, Irf7, Stat1, Stat2, Nfkb2 or Relb were active in the hillock and hybrid cells (Fig. 5a). In addition, we found that lineage determinant TFs such as Elf3, Creb5 or Grhl3 were more active in these two cell lineages as well as in the proximal LCs (Extended Data Fig. 5k-m).

Figure 5. BC plasticity following Pten deletion is associated with the activation of interferon and TNF pathways in epithelial cells.

(a) Regulatory network analysis using SCENIC: UMAP plots representing regulon activity (top) and corresponding transcription factor expression (bottom) in K5-PTEN 6w mice. Color scales represent SCENIC AUC scores for regulon activity and normalized gene expression levels of each transcription factor. Exp, expression. (b) GO analysis of genes upregulated more than 2-fold on bulk RNA-seq of FACS isolated YFP+ BCs from K5-PTEN mice 5-weeks after TAM administration compared (BC PTEN) to WT BCs (BC CTL). (c) GO analysis of genes upregulated more than 2-fold in FACS isolated YFP+ LCs arising from Pten deleted BCs from K5-PTEN mice 5-weeks after TAM injection (LC PTEN) compared to WT LCs (LC CTL). (d) Relative mRNA expression (bulk RNA-seq) of BC CTL, LC CTL, BC PTEN and LC PTEN (n=2 samples). (e) Representative examples of ATAC-seq peaks of inflammatory-related genes from FACS isolated BC CTL, LC CTL and BC PTEN from K5-PTEN mice 6 weeks after TAM administration. 100k cells pooled from at least three mice. Scale for visualization: Il1r1 (0-21), Cd52 (0-13), Cd55 (0-24), Cd83 (0-13), Ifit1 (0-13) and Cxcl5 (0-18). (f) TF motif enrichment analysis of peaks upregulated in BC PTEN compared to BC CTL. P-values were calculated using a binomial test. (g) Western blot of FACS isolated LIN- epithelial cells from CD1 mice (CTL epi) and Lin- YFP+ epithelial cells of K5-PTEN mice 6 weeks after TAM injection (PTEN YFP+) using the indicated antibodies. n=3 mice. (h) Representative immunostaining of myeloid cells in the prostate and (i) quantification of Ly6G positive cells from K5CreER/RosaYFP (n=3 mice) and K5CreER/Pten^fl/fl/RosaYFP mice (n=4 mice) 6-weeks after TAM administration using anti-GFP and anti-Ly6G antibodies. Scale bar, 20 μm. Graphs are mean ± s.e.m. p-values are derived from two-sided unpaired t-test. For b and c, P-values were derived from two-tailed modified Fisher’s Exact Test with Benjamini–Hochberg correction.

To investigate further the molecular mechanisms that lead to the cell fate changes upon Pten deletion in BC, we performed bulk RNA-seq of FACS isolated BC and LC in WT mice and 5 weeks following Pten deletion in BC. Bulk RNA-seq is more sensitive than scRNA-seq to detect low abundant RNA and changes in gene expression. We found that 1067 genes were upregulated more than 2-fold between WT and Pten deleted BCs and 2844 genes were upregulated in LCs following Pten deletion. Gene set enrichment analysis revealed the enrichment of genes regulating extracellular matrix (ECM), EGF like domain, neutrophil chemotaxis, inflammatory response and innate immunity in BCs, as well as angiogenesis, mitosis and cell cycle in LCs upon Pten deletion (Fig. 5b and 5c). Genes regulating inflammation such as Il1a, Il1r1, Oasl1 and Cd74, interferon responsive genes such as Ifit1 and Tnfaip2, chemokines such as Cxcl2, Cxcl13, Cxcl15 as well as genes associated with the cell fate change such as Krt4, Krt6a, Grhl3, Ly6d, Wfdc2 and Clu were upregulated in BC and even further in LC upon Pten deletion (Fig. 5d). Proliferation genes were moderately upregulated in BCs but more strongly in LCs upon Pten deletion (Fig. 5d). To investigate the onset of innate immune gene activation, we FACS isolated BCs at different time points after Pten deletion in BCs and performed quantitative reverse transcription PCR (qRT-PCR) analysis of inflammatory genes. Our results showed that Il1a and Cxcl2 started to be upregulated 4 weeks after Pten deletion at the same time that basal to luminal transition began to be observed (Extended Data Fig. 6a).

To unravel the chromatin remodeling and gene regulatory networks associated with oncogenic hits induced BC multipotency in the prostate, we performed bulk ATAC-seq of FACS isolated BCs and LCs from WT mice and BCs 6 weeks following Pten deletion in BC. We found that 1974 peaks were more accessible, and 1942 peaks were less accessible in BCs upon Pten deletion (Fig. 5e). Motif discovery analysis of the peaks upregulated in BC following Pten deletion revealed the increase of AP-1 TF motif, P63 and Nfkb/p65/Rel motif (Fig. 5f), in good accordance with the SCENIC analysis and the activation of Nfkb/innate inflammation transcription program found by bulk and single-cell RNA-seq. The expression and the active phosphorylation of Stat1 and p65 protein were increased after Pten deletion in BCs (Fig. 5g).

To assess whether the lack of expression of Hillock like state following Pten deletion in BC from VP is the consequence of a decrease of chromatin accessibility of the regulatory regions of the genes associated with Hillock like state, we performed ATAC-seq on YFP+ FACS isolated Pten deleted BCs specifically from the VP, DLP and AP lobes. Our data show that BCs from VP presented a decrease in the chromatin accessibility at the enhancers remodeled following Pten deletion of Hillock associated genes such as Krt13 and Aqp3 as compared to BCs of AP/DLP (Extended Data Fig. 6b and 6c), supporting the notion that Hillock associated genes are epigenetically primed to get upregulated following Pten deletion in AP/DLP compared to VP. Motif discovery of these different upregulated peaks in AP and DLP versus VP showed enrichment in AP-1, KLF, and ETS family of TFs (Extended Data Fig. 6d).

It has been shown that ETS transcription factor ERG (ERG) acts as a master regulator of luminal differentiation in Pten deleted tumors³⁷. ETS motif was enriched in ATAC-seq peaks upregulated in BC following Pten deletion in AP/DLP versus VP, and ERG mRNA expression was upregulated in BC and even more strongly in LC arising from BC from AP/DLP versus VP following Pten deletion (Extended Data Fig. 6e and 6f), supporting the notion that ERG can promote LC fate differentiation from BC following Pten deletion in AP and DLP.

As Rapamycin treatment strongly inhibits the activation of BC multipotency and the initiation of prostate tumorigenesis, we performed bulk-RNAseq in BCs 3 weeks following Pten deletion and Pik3ca activation. The deletion of Pten and expression of oncogenic Pik3ca induced the upregulation of 1594 genes in BCs (Extended Data Fig. 7a). Gene set enrichment analysis of these upregulated genes revealed the strong upregulation of genes regulating DNA replication and cell cycle progression as well as genes involved in innate immunity and inflammation (Extended Data Fig. 7b). Rapamycin administration inhibited the upregulation of about 50% of the genes upregulated by Pten deletion and expression of oncogenic Pik3ca and prevented the upregulation of many genes regulating cell proliferation, cell reprogramming and innate immunity (Extended Data Fig. 7a-d). The secretion of leukocyte chemoattractant such as Cxcl2, Cxcl13 and Cxcl15 following Pten deletion in BCs was accompanied by the recruitment of neutrophils and myeloid-derived suppressor cells (MSDC) surrounding the BCs and LCs arising from the multipotent BCs, which was prevented by Rapamycin administration (Fig. 5h, 5i and Extended Data Fig. 7e and 7f). Altogether, these data show that Pten deletion in BC activates the innate immune pathway in a mTOR dependent manner.

Regional reprogramming in mouse prostate cancers

To assess whether the regional reprogramming and the activation of the innate immunity pathway observed during the early stages of tumor initiation persist in more advanced prostate tumor, we performed scRNA-seq 10 months following Pten deletion in BC when almost the whole prostate presents signs of tumorigenesis. Unsupervised clustering analysis revealed the presence of different basal, hybrid and luminal cell clusters (Fig. 6a and Extended Data Fig. 8a-e), reminiscent of the different clusters found 6 weeks following Pten deletion. Strong activation of innate immunity was observed in some BC, hybrid and LC clusters with the upregulation of Irf6, Irf7, Irf9, Stat1 and Rela expression and regulon activity (Fig. 6b and Extended Data Fig. 8f-h). Markers of Hillock/Hybrid-like cells (Krt13, Aqp3) and proximal LC differentiation such as Krt4, Clu, Wfdc2, Pigr, and Ppp1r1b, were expressed in hybrid and LC at 10 months following Pten deletion (Fig. 6c-e). Immunostaining of Krt13, Aqp3, and Krt4 revealed that these markers were still expressed in more advanced tumors in AP and DLP and much less in VP (Extended Data Fig. 8i-o). Similarly, adenovirus mediated Pten/Pik3ca or Pten/P53 recombination in BC presented increased expression of Krt13, Aqp3, and Krt4 and myeloid cell infiltration in prostate tumors (Extended Data Fig. 8p-r). Altogether, these data indicate that the region-specific differentiation into Hillock and LC proximal like states as well as the activation of the innate immunity pathway persist at the later stages of mouse prostate tumorigenesis.

Figure 6. Regional reprogramming and activation of innate immunity persist in mouse prostate tumors and human prostate cancers.

(a) Cell populations identified by scRNA-seq of FACS isolated LIN- YFP+ epithelial cells from K5-PTEN mice 10 months after TAM administration: UMAP plots with different colors representing unsupervised clustering. (b) Regulatory network analysis using SCENIC: UMAP dimensionality reduction plots with color scaling representing the SCENIC AUC values for transcription factor activation (top) and normalized gene expression (bottom). UMAP plots colored by normalized gene expression values of canonical cell markers of BC (c) (Krt14), (d) Hillock and hybrid-like (Krt13, Aqp3) and (e) Proximal LC (Krt4, Clu, Wfdc2, Pigr and Ppp1r1b). (f) Cell populations on ERG-driven human prostate cancer cells³⁸ : basal epithelial cells (BE), non-malignant luminal epithelial cells (LE), ERG-positive tumor cells (ERG+T) and ERG-negative tumor cells (ERG-T); human prostate cancer atlas³⁹: basal epithelial cells (BE), hillock epithelial cells (HE), club epithelial cells (CE); KLK3-high luminal epithelial cells (LE-KLK3), and KLK4-high luminal epithelial cells (LE-KLK4); treatment-naïve prostate adenocarcinoma⁴⁰: basal epithelial cells (BE), hillock epithelial cells (HE), club epithelial cells (CE) and luminal epithelial cells (LE); invasive cribriform carcinoma and intraductal carcinoma data⁴¹: basal and hillock cells (BC/Hillock), club cells and ductal cells (Club/Ductal), non-malignant luminal epithelial with low expression of AR (LC_ARlow) and non-malignant luminal epithelial with high expression of AR (LC_ARhigh). (g) UMAP dimensionality reduction plots with color scaling representing enrichment score for the reprogramming markers. (h) UMAP dimensionality reduction plots for AQP3 and PIGR with color scaling representing the level of gene expression. (i) IHC of PIGR, AQP3 and STAT1 and IHC score of (j) PIGR, (k) AQP3 and (l) STAT1 in PC with different GG. Scale bar, 50 μm. Number of patients (n) is indicated. p-values are derived from two-sided Fisher’s exact test. Pos, positive; Neg, negative.

Regional reprogramming in human prostate cancers

To investigate whether similar reprogramming also occurs in human prostate cancers, we assessed the expression of markers defining Hillock, Hybrid and proximal-like states identified from our K5-PTEN 6W scRNA-seq data across four different types of human prostate cancer scRNA-seq data^38–41 (Fig. 6f).

Using these four human prostate cancer single-cell datasets, we quantitatively assessed the enrichment of mouse reprogramming gene signatures in human samples. In the ERG-driven prostate cancer dataset³⁸, we observed a progressive enrichment of the reprograming signature, increasing from basal epithelial cells to the club cells, followed by a decrease in signature gene expression toward luminal epithelial cells. In the prostate cell atlas dataset³⁹, the reprogramming signature genes showed the highest enrichment in hillock and club epithelial cells. A similar pattern was observed in a dataset derived from treatment-naïve prostate adenocarcinoma patients⁴⁰, where hillock and club epithelial cells also showed the highest enrichment score. Finally, in a dataset composed of invasive cribriform carcinoma and intraductal carcinoma samples⁴¹, reprogramming signature enrichment was observed in both BC/Hillock and Club/Ductal cell populations (Fig. 6g). Furthermore, the levels of the markers associated with the different cell states during BC reprograming, including AQP3, PIGR, KRT13, and WFDC2, were higher in hillock and club epithelial cells compared to other cell types across all the different human prostate cancer scRNA-seq datasets, as found for the combined signature (Fig. 6h and Extended Data Fig. 9a-f). Pseudotime lineage trajectory in human prostate cancer datasets showed that BCs differentiated into hillock or club-like cells before differentiating into luminal tumor cells (Extended Data Fig. 9g-i). These results show that the expression of the gene signature associated with the reprogramming of BC into LC following Pten deletion in mouse prostate tumor initiation is also found in human prostate cancers.

To further strengthen our analysis on human prostate cancer scRNA-seq datasets, we performed immunohistochemistry staining of the selected reprogramming markers on 136 human prostate cancer (PCs) specimens across various Gleason groups (GG), including low-grade (GG1), intermediate-grade (GG2-3), and high-grade (GG4-5) prostate cancers. The Hillock/Hybrid marker AQP3, the proximal marker PIGR, and the inflammatory marker STAT1 were expressed at higher level and in more tumor cells in patients with high-grade prostate cancers (Fig. 6i-l and Extended Data Fig. 9j). Altogether, these findings support the notion that cell fate reprogramming occurs in a subset of human prostate cancers with the most invasive features and that lineage reprogramming could serve as a predictive marker for aggressive disease.

Targeting innate immunity inhibits BC plasticity

To assess the functional consequences of the activation of innate immunity occurring following Pten deletion on SC plasticity, we pharmacologically inhibited these pathways and assessed the impact of their inhibition on basal to luminal transition (Fig. 7a). Administration of JSH-23, a small molecule inhibitor of NF-kb, Ruxolitinib, a Janus kinase (JAK1 and JAK2) inhibitor 6 weeks following Pten deletion in BC and 3 weeks after targeting Pten-Pik3ca in BC decreased the differentiation of BC into LC (Fig. 7b and 7c). To assess whether the increase expression of Il1 following Pten deletion in BC promotes the activation of innate immunity and cell plasticity in prostate epithelial cells, we treated the mice with Anakinra, an Il1R inhibitor⁴². Targeting Il1 decreased Pten induced cell plasticity and basal to luminal cell transition (Fig. 7d and 7e). In addition, these treatments also prevented the reprogramming of BC into Hillock and proximal like states in AP and DLP (Fig. 7f-j). Interestingly, targeting the Il1, Nfkb and JAK/STAT signaling also decreased myeloid cell infiltration following Pten deletion and Pik3ca activation (Fig. 7k-m). These data demonstrate that targeting innate immunity inhibits BC plasticity and reprogramming that occur following Pten deletion.

Figure 7. Pharmacological inhibition of Nfkb, JAK/STAT and Il1 inhibits SC plasticity induced by Pten deletion.

(a) Experimental design. (b) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells from DMSO, JSH-23 and Ruxolitinib treated K5-PTEN 6w and K5-PTEN-PIK 3w mice. (c) Quantification of % of YFP+ LCs in total YFP+ cells of whole prostate in DMSO, JSH-23 and Ruxolitinib treated K5-PTEN 6w mice, n=14 mice (DMSO), n=12 mice (JSH-23), n=13 mice (Ruxolitinib), and K5-PTEN-PIK 3w mice, n=8 mice (DMSO), n=7 mice (JSH-23), n=7 mice (Ruxolitinib). (d) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells from Saline (i.p.) and Anakinra (i.p., 10mg/kg, 3 injections per week) treated K5-PTEN 6w mice and K5-PTEN-PIK 3w mice. (e) Quantification of % of Lin- YFP+ LCs in total YFP+ cells of whole prostate in K5-PTEN 6w, n=5 mice (Saline), n=6 mice (Anakinra), and K5-PTEN-PIK 3w mice, n=4 mice (Saline), n=6 mice (Anakinra). (f) Representative immunostainings of AP of DMSO, JSH-23 and Ruxolitinib treated K5-PTEN-PIK 3w mice using anti-GFP (green), anti-K8 (red), anti-K13 (red) and anti-K4 (red) antibodies. Scale bar, 50 μm. n=3 mice. Quantification of the % of YFP+ K8+ in total K8+ cells, YFP+ K13+ and YFP+ K4+ in total YFP+ cells of DLP/AP in DMSO, JSH-23 and Ruxolitinib treated (g) K5-PTEN 6w mice and (h) K5-PTEN-PIK 3w mice. n=3 mice. Quantification of YFP+ K8+ in total K8+ cells, YFP+ K13+ and YFP+ K4+ in total YFP+ cells of DLP/AP in Saline and Anakinra treated (i) K5-PTEN 6w mice and (j) K5-PTEN-PIK 3w mice. n=3 mice. (k) Representative immunostainings of myeloid cell infiltration of AP in DMSO, JSH-23 and Ruxolitinib treated K5-PTEN-PIK 3w mice using anti-GFP (green), anti-Ly6G and anti-K14 antibodies. Scale bar, 50 μm. n=3 mice. Quantification of Ly6G positive cells in the DLP/AP from (l) DMSO, JSH-23 and Ruxolitinib treated, (m) Saline and Anakinra treated K5-PTEN-PIK 3w mice. n=3 mice. Graphs are mean ± s.e.m. p-values are derived from two-sided unpaired t-test (e, i, j, m) and one-way ANOVA with Dunnett test (c, g, h, l). Ruxo, ruxolitinib; Sal, saline; Ana, anakinra.

Cell autonomous role of innate immunity in BC plasticity

As Pten deletion and Pten/Pik3ca oncogenic hits promote the activation of innate immunity in prostate epithelial cells and the recruitment of innate immune cells in the prostate during prostate tumor initiation, we assessed whether these oncogenic stimuli promote cell plasticity and oncogenic induced reprogramming in vitro in the absence of stromal cells. To this end, we generated epithelial prostate organoids and assessed the impact of Pten or Pten/Pik3ca recombination on prostate epithelial cell states in vitro. Whereas in the absence of oncogenic hits, prostate organoids were composed of crowned of basal and luminal layers surrounding a lumen (hollow organoid), following Pten deletion and even faster following Pten/Pik3ca recombination in BC, these organoids became hyperplastic with their lumens filled of cells (opaque organoid) (Fig. 8a and 8b). Immunostaining for YFP and K8 revealed that the hyperplastic organoids observed following Pten and Pten/Pik3ca recombination were filled with LC co-expressing YFP and K8 (Fig. 8c and 8d). Treatment of these organoids with JSH-23 and Ruxolitinib decreased the proportion of hyperplastic organoids and basal to luminal transition in vitro (Fig. 8e-g). These data demonstrate that BC SC plasticity following Pten deletion occurs in prostate organoid in vitro in the absence of stromal and immune cells and can be blocked by pharmacological inhibition of Nfkb and JAK/STAT pathways.

Figure 8. Cellular autonomous activation of Nfkb mediate BC plasticity and regional reprogramming following Pten deletion.

(a) Representative images and H&E staining of prostate organoids and (b) quantification of hyperplastic organoids after 5 days with or without TAM administration. Organoids are established from indicated mice. Scale bar, 50 μm. n=4 independent experiments. (c) Representative images of immunostaining of indicated prostate organoids and (d) quantification of % of YFP+ K8+ in total YFP+ cells after TAM 5 days administration using indicated antibodies. Scale bar, 50 μm (up), 20 μm (down). n=3 independent experiments. (e) Representative images and immunostaining of DMSO, JSH-23 (1uM) and Ruxolitinib (10uM) treated prostate organoids from K5-PTEN mice 5 days following TAM administration and quantification of (f) hyperplastic organoids and (g) YFP+K8+ in total YFP+ cells using indicated antibodies. Scale bar, 500 μm (up), 50 μm (down). n=3 independent experiments. (h) Representative FACS plot of CD49f and EpCAM expression in YFP+ prostate epithelial cells (i) quantification of the % of YFP+ LCs in total LCs in K5-PTEN and K5-PTEN-P65 mice 6 weeks after TAM administration. n=7 mice. (j) Representative prostate immunostainings of K5-PTEN 6W and K5-PTEN-P65 6W mice using indicated antibodies. Scale bar, 50 μm. n=3 mice. (k) Quantification of the % of YFP+ K8+ in total K8+ cells, YFP+ K13+, YFP+ Aqp3+ and YFP+ K4+ in total YFP+ cells (n=3), and Ly6G+ cells (n=4) in DLP/AP. (l) GO analysis of genes downregulated more than 2-fold in RNA-seq from FACS isolated YFP+ BCs from K5-PTEN-P65 6W mice compared to BCs from K5-PTEN 6W mice. (m) Relative mRNA expression from bulk RNA-seq of FACS isolated YFP+ BCs from K5-PTEN 6W and K5-PTEN-P65 6W mice. n=2. (n) H&E-stained histological sections of DLP and (o) quantification of the different histological lesions of prostate tumorigenesis from K5-PTEN 6W and K5-PTEN-P65 6W mice. Scale bar, 50 μm. Number of mice (n) is indicated. Graphs are mean ± s.e.m. p-values are derived from two-way (b) ANOVA with Tukey’s test, one-way ANOVA with Dunnett test (d, f, g), two-sided unpaired t-test (i, k) and two-tailed modified Fisher’s Exact Test with Benjamini–Hochberg correction (l).

To assess that the cellular reprogramming of BC into Hillock cells and proximal like luminal cells following Pten deletion in vivo is mediated by the activation of NfkB in a cellular autonomous manner as suggested by the RNA-seq and ATAC-seq in vivo and by the organoid in vitro data, we performed the co-deletion of Rela/p65 subunit of the NfkB complex together with Pten in BC (K5CreER/PTEN^fl/fl/P65^fl/fl/RosaYFP) and assess its impact on the BC plasticity and cellular reprogramming. The deletion of P65 in BC together with Pten deletion strongly inhibited BC plasticity and the differentiation of BC into LC, the reprogramming of BC into Hillock and proximal LC like states, as well as the recruitment of myeloid cells (Fig. 8h-k).

To define the role of p65 in Pten deletion mediated BC plasticity, we performed bulk RNA seq of FACS isolated BCs after Pten or Pten/p65 deletion in BC. Gene set enrichment analysis revealed the enrichment of genes regulating innate immunity, cytokine, chemokine, interferon and TNF/Nfkb signaling pathways in genes downregulated following Pten/p65 deletion as well as genes associated with the Hillock/Hybrid/proximal like reprogramming such as Krt13, Aqp3, Krt4, Krt6a, Ly6d, Wfdc2, Pigr and Clu (Fig. 8l and 8m). Interestingly, genes of the distal luminal like state such as Nkx3.1, C1rb, Tgm4 were upregulated upon Pten/p65 deletion (Fig. 8m). Importantly, the strong inhibition of BC reprogramming and activation of inflammation/innate immunity following p65 deletion was accompanied by a major reduction in prostate tumorigenesis (Fig. 8n and 8o).

Altogether these data demonstrate that the Pten induced BC plasticity is mediated by the activation of innate immunity and Nfkb signaling pathway in prostate epithelial cells in a cellular autonomous manner.

Discussion

In this study, we uncovered the importance of the cells of origin and regionalization in regulating prostate tumor heterogeneity and the molecular mechanisms controlling SC plasticity and lineage infidelity following oncogenic hits during the early stage of prostate tumorigenesis (Extended Data Fig. 10).

Our results reconcile conflicting findings regarding the differential tumor phenotypes arising from basal cells (BCs) and luminal cells (LCs)^4–6. We demonstrate that BCs from the AP and DLP give rise to more invasive tumors compared to BCs from the VPs and LCs. However, tumorigenesis is more rapid when initiated in VP-derived LCs. The observation that BC-derived tumors are more aggressive in AP and DLP aligns with the findings of Lu et al⁴. Our findings emphasize that tumor progression is strongly influenced by the prostate lobe from which the tumor originates, an aspect often overlooked in previous studies potentially explaining the discrepancies between the different studies^5,6. Interestingly, this variation in tumor phenotype correlates with different patterns of cellular reprogramming following Pten deletion.

We identify a previously unexplored reprogramming trajectory in BCs from AP and DLP upon Pten deletion, which undergo a stepwise transition through Hillock cells and a hybrid basal-luminal state before differentiating into proximal-like LCs. Hillock cells, marked by Krt13 expression, have been reported in the human proximal prostate near the prostate-urethra junction^9,31–33, but their role in tumor initiation remains unexplored. These different cell states that arise following Pten deletion in BCs from AP and DLP persist in more advanced tumors and maintain their proximal luminal differentiation features. As proximal LCs were proposed to present increased self-renewing potential^{10–12,43–46}, it is possible that this proximal luminal state confers stemness and multi-lineage differentiation to the cells of origin of prostate cancer that boost their tumorigenic potential. In contrast, VP BCs follow a distinct differentiation pathway, transitioning directly into distal-like LCs with lower plasticity and slower tumor progression. Unlike BCs, LCs fail to reactivate multipotent programs following Pten deletion, and their tumors exhibit reduced immune infiltration and collagen deposition. Due to health issues linked to the expression of K5CreER in other tissues, the long-term consequences of these histological differences in prostate tumor progression, such as metastasis or animal survival, could not be assessed in our model.

Our findings have important implications for human prostate cancer. Like the mouse prostate, the human prostate consists of three distinct zones: the peripheral, transition, and central zones^47,48. Notably, approximately 70% of prostate cancers arise in the peripheral zone, where tumors exhibit more aggressive features and worse clinical outcomes compared to transition zone derived tumors^47–49. Transcriptomic analyses indicate that the human peripheral zone resembles the mouse DLP^11,50, suggesting conserved regional influences on prostate tumorigenesis.

The difference in BC plasticity and cellular reprogramming across the different lobes are accompanied by different remodeling of the tumor microenvironment (TME). BC-Hillock-Hybrid-Proximal reprogramming in AP/DLP is associated with extensive ECM remodeling, desmoplastic stroma formation, and enhanced myeloid cell recruitment. This correlates with elevated expression of collagens, ECM components, metalloproteases, and inflammatory mediators. These results indicate that the cellular origin and regional context of oncogene-driven reprogramming control the remodeling of TME, reinforcing cell plasticity and accelerating tumor progression.

At the mechanistic level, our transcriptomic and chromatin profiling show that upon oncogenic hits, BC activates innate immunity including NFkB, JAK/STAT and interferon. Previous studies have linked Pten deletion in prostatic epithelium to the activation of senescence-associated secretory phenotype^51–53, that has been shown to recruit myeloid cells that, in turn, inhibits senescence, promotes tumor progression and the development of castration-resistant tumors^53–57. Our data shows that inhibition of JAK/STAT and NFkB decreases SC plasticity and basal to luminal transition following Pten deletion in BC. Increased JAK/STAT inflammatory signals have been observed during the development of androgen resistant prostate cancer and the switch from prostate adenocarcinoma to neuroendocrine cancer, suggesting that the emergence of tumor plasticity at the later state of prostate cancer also depends on an increased inflammatory state^35,58. JAK/STAT signals have recently been found to sustain Nkx3.1-expressing basal cells during homeostasis⁵⁹. Pten deletion induces the upregulation of Il1a and its receptor Il1r1 in BC, activating the innate immune pathway in hybrid cells. Inhibition of Il1R decreases basal SC plasticity, supporting the notion that Il1/IL1R expression in BC following Pten deletion mediates an autocrine loop that sustains BC plasticity, and the cellular reprogramming associated with tumor initiation. Pten deletion in vitro in prostate organoid BCs promotes BC plasticity and luminal differentiation and deletion of p65/Rela in BC demonstrate that this activation of the immune immunity pathway in Pten targeted epithelial cells is mediated by a cellular autonomous mechanism. Taken together, our study demonstrates that innate immunity activation in BC is essential for cell plasticity, inflammation, and tumor initiation, revealing key molecular mechanisms linking Pten-driven inflammation to prostate cancer initiation and progression. These mechanisms that promote SC plasticity may also be relevant to the initiation of other cancers.

Our study establishes that reprogramming signatures and the Hillock like state identified in Pten-deleted mouse BCs are also present cells in human prostate cancer across four independent single-cell RNA-seq datasets^38–41. These cells display transcriptomic profiles resembling castration-resistant prostate cancer (CRPC)^9,12,33,60 and are linked to inflammation and myeloid infiltration^61,62. The presence of Hillock and Club-like cells in later stages of human prostate may reflect either their progressive accumulation over time during tumorigenesis as the lineage trajectory analysis suggests or indicate the activation of lineage plasticity at later stages of tumor progression. Furthermore, immunohistochemistry analyses of a large human prostate cancer cohort reveal that the expression of Hillock marker AQP3, Club-like marker PIGR, and inflammatory marker STAT1 correlates with high-grade tumors. This suggests that BC plasticity and differentiation trajectories are not uniform across all prostate cancers, and that lineage reprogramming could serve as a predictive biomarker for aggressive disease.

In conclusion, our study uncovers a fundamental link between the cells of origin, regional influences, and tumor plasticity in prostate cancer initiation. We demonstrate that innate immunity activation in BCs is crucial for SC plasticity, inflammation, and tumor progression. By identifying reprogramming-associated immune pathways as key drivers of tumor initiation, our findings open new avenues for therapeutic interventions targeting inflammation-induced plasticity in prostate cancer and potentially other malignancies.

Methods

Ethical Statement

Mice colonies were housed in a certified animal facility in compliance with European guidelines. The room temperature was maintained between 20 and 24 °C, with relative humidity kept at 55 ± 10%. Food, water and two types of nesting material were provided in each cage. A semi-natural light 12-hour light/dark cycle was implemented. All animal experiments were approved by the ethical committee (Commission d’Ethique et du Bien Être Animal, CEBEA) of the Faculty of Medicine, Université Libre de Bruxelles under protocols #673N, #854N, #914N. CEBEA follows the European Convention for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes. As prostate tumors grow intra-abdominally, humane endpoints based on general health and a 20% body weight loss limit, approved by CEBEA, were not exceeded in this study. Male mice were used at adult age (over 8 weeks) for the prostate analysis.

Mouse genetically lineage tracing

The generation of K5CreER and K8-CreERT2 was previously described⁶³. Rosa26-YFP and Pten^fl/fl mice were obtained from the Jackson laboratory. Pik3caH1047R knock-in mice, in which wild-type exon 20 is replaced by H1047R mutant exon 20 upon Cre recombination, were described previously⁶⁴. P53^fl/fl mice⁶⁵ were obtained from the National Cancer Institute at Frederick. P65^fl/fl mice⁶⁶ were imported from Dr. Manolis Pasparakis lab. Male mice with a mixed genetic background were used in this study. Mice were induced with 15 mg of tamoxifen (TAM) (T5648; diluted in sunflower seed oil, Sigma) by intraperitoneal injection (3 injections of 5 mg every second days). For long trace (over 12 weeks) of K5CreER/Pten^fl/fl/RosaYFP mice, the TAM injection reduced to 6-10mg. For K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice and K5CreER/Pten^fl/fl/P53^fl/fl/RosaYFP mice, the TAM injection reduced to 10mg (3 injections of 3.3mg every 4 days).

Intraprostatic adenoviral infection

Intraprostatic adenoviral infection was performed as previously described^67–69. Briefly, AdeK5Cre (Ad5-bK5-Cre) (Dr. Anton Berns, Netherlands Cancer Institute) were obtained at high titer (8E+10 to 2E+10 pfu/ml) from Viral Vector Core at University of Iowa. 1 µl of AdeK5Cre was mixed with 9 µl of organoid medium containing 16 µg/ml Polybrene (Millipore Sigma). A total of 10 µl of this adenoviral solution was injected into the AP of transgenic mice as described in the figure legends.

Immunostaining

Prostate tissue from adult mice was micro-dissected under a stereoscope to isolate the different prostate lobes. The dissected lobes were fixed in 4% PFA at RT for 2h. Tissue were washed with PBS and then incubated overnight at 4 °C in 30% sucrose. Samples were subsequently embedded in OCT and stored at −80 °C. Prostate organoids were collected in cold basal medium, washed with cold PBS and then fixed in 4% PFA for 30 minutes. After washing, organoids were embedded in OCT and stored at -80°C.

Cryosections of 5 μm were cut using an HM560 Microm cryostat (Mikron Instrument). Sections were blocked in buffer containing 1% BSA, 5% HS, 0.2% TritonX-100 in PBS for 1h at RT. Primary antibodies, diluted in blocking buffer, were incubated overnight at 4 °C. The sections were then washed three times in PBS and incubated with secondary antibodies diluted in blocking buffer for 1h at RT. Detailed antibody information were provided in Supplementary Table. Nuclei (Blue) were stained using Hoechst solution (1:1000 dilution), and slides were mounted with DAKO mounting medium (Sigma). Images were acquired at RT using a ZEISS Axioscan 7, an LSM780 confocal microscope fitted on an Axiovert M200 inverted microscope equipped with a C-Apochromat (40X, A.=1.2) water immersion objective and using a Zeiss Axio Imager M2 fluorescence microscope with a Zeiss Axiocam MR3 camera using Axiovision release 4.8 software. Brightness, contrast, picture size and analyses were performed using ZEN software and Adobe Photoshop CS6.

EdU experiments

Male mice were injected with a single intraperitoneal injection of EdU (2.5 mg/ml in PBS) 24 h before euthanasia. EdU staining was performed according to the manufacturer’s instructions (Thermo Fisher Scientific, c10340). For co-expression with EdU, the K14 or K8 primary antibody staining was performed first, and then the EdU protocol was followed.

H&E staining and prostate tumor classification

Hematoxylin/Eosin (H&E) staining was performed on paraffin-embedded sections using standard laboratory protocols. Histological assessments of mice prostate samples were performed by an expert uropathologist by assimilating the histological criteria observed in human prostate according to the 2022 WHO Classification (WHO Classification of Tumours Editorial Board. WHO Classification of Tumours, 5^th Ed.. Urinary and Male Genital Tumours. Chapter 4: Tumours of the prostate p193-233. International Agency for Research on Cancer, Lyon 2022, France). Lesions were categorized as normal, high-grade prostatic intraepithelial neoplasia (HGPIN), atypical intraductal proliferation (AIP), intraductal carcinoma of the prostate (IDC-P) and invasive adenocarcinoma (ADC). As defined in the 2022 WHO Classification, HGPIN is considered to be the earliest histologically recognizable precursor of ADC (exhibiting only tufting, micropapillary or flat pattern); whereas AIP is characterized by atypical cribriform proliferations that do not reach the threshold for IDC-P. To approach tumor formation heterogeneity across samples, IDC-P were categorized as focal or florid based on the percentage of IDC-P observed (i.e. < 50% or > 50%). Morphological appearance of the tumor stroma was characterized as normal, slightly inflammatory (Oedematous), highly inflammatory, desmoplastic.

Pharmacological treatments in vivo

Mice over 8 weeks old were injected with TAM as described previously. After that, Rapamycin (i.p., 6mg/kg body weight, MedChemExpress, HY-10219), JSH-23 (i.p., 6mg/kg body weight, MedChemExpress, HY-13982), Ruxolitinib (i.p., 10mg/kg body weight, MedChemExpress, HY-50856) and Anakinra (i.p., 10mg/kg body weight, Sobi, Kineret) were injected into mice. Rapamycin was firstly dissolved in EtOH to get 40mg/ml and then diluted in buffer (10% PEG400 and 10% Tweeen80) to reach 1mg/ml Rapamycin. As control, same amount EtOH dilute in buffer was injected to the mice. JSH-23 and Ruxolitinib were first dissolved in DMSO to get 40mg/ml and then diluted in buffer (50% Saline, 45% PEG400 and 5% Tweeen80). As control, same amount DMSO diluted in buffer and then injected to the mice. Anakinra (100mg/0.67ml) was diluted in Saline to get 2mg/ml. As control, same amount Saline was injected to the mice. For K5CreER/Pten^fl/fl/RosaYFP mice, the Rapamycin, JSH-23 and Ruxolitinib injection start after 1-week TAM injection for continuous 5 weeks (3 injection per week, every second day). Anakinra was injected 4-week after TAM injection for continuous 2 weeks (3 injection per week, every second day). For K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice, the Rapamycin, JSH-23, Ruxolitinib and Anakinra injection start right after the first TAM injection for continuous 3 weeks (3 injections per week, every second day).

Cell labelling, flow cytometry and sorting

Mouse prostate single cell preparation as previously described²². Briefly, prostate tissue including ventral, dorso-lateral and anterior prostate were isolated from indicated mouse lines at indicated time points after TAM injection. Tissues were minced in a 6-cm culture plate and digested in 5mg/ml collagenase type II (Life Technologies, cat. no. 17101-015) with 10 μM Y-27632 dihydrochloride (Abmole Bioscience, cat. no. M1817) for 1.5-2 h at 37°C on a shaking platform. Glandular structures were then washed with Advanced DMEM/F12 and centrifuged at 150g for 5 mins at 4°C. Structures were further digested in Trypsin for 10-20 min at 37°C and Trypsin activity was quenched using 2% FBS in PBS. Cells were passed through a 40μm cell strainer and incubated with fluorochrome-conjugated primary antibodies for 30 mins on ice with shaking every 10 min. Detailed antibody information was provided in Supplementary Table. Cells were washed with 2% FBS/PBS and resuspended in DAPI or Hoechst before analysis. Data analysis and cell sorting were performed on a FACSAria sorter or LSRFortessa using the FACS DiVa software (BD Biosciences). Dead cells (DAPI+) and Lin+ (CD45+, CD31+ and CD140a+) cells were excluded before analysis. Due to technical challenges in obtaining enough BC-derived LCs from adult wild-type mice, LCs isolated from adult CD1 mice were used to compare with LCs arising from Pten-deleted BCs. The following populations were analyzed and sorted BCs: DAPI- LIN- YFP+ CD49f^high EpCAM⁺ and LCs: DAPI- LIN- YFP+ CD49f^low EpCAM⁺.

Western Blot

Prostate epithelial cells were FACS isolated as described in the section “Cell labelling, flow cytometry and sorting”. To obtain sufficient protein, around 500k DAPI- LIN- epithelial cells from CTL mice and DAPI- LIN- YFP+ epithelial cells from K5CreER/Pten^fl/fl/RosaYFP 6 week after TAM administration were FACS isolated from the whole prostate of at least three mice. Cells were lysed in cell lysis buffer (Cell Signaling, Cat#9803) supplemented with a phosphatase inhibitor cocktail (Cell Signaling, cat#5870) and 1mM PMSF (Sigma, Cat#P7626) on ice for 5 minutes. The lysates were sonicated (5 X 10 seconds) and centrifuged at 14.000 g for 10 min at 4°C. Cell lysate mixed with loading buffer were heated at 99°C for 5 mins, then loaded in NuPage 10% Bis-Tris gel (Invitrogen, Cat#NP0315BOX) and separated by electrophoresis. Proteins were transferred to PVDF membranes, blocked in 5% milk for 1 hour and incubated overnight with primary antibodies (details in Supplementary Table). Membrane was re-probed after incubation in stripping buffer (Invitrogen, Cat#46430) for 15 minutes. Anti-rabbit IgG conjugated with horseradish peroxidase (HRP) (1:5000, Sigma Aldrich, Gena9340) was used as the secondary antibody. Blots were developed using an iBright 1500 (Invitrogen).

Prostate organoids experiments

Prostate organoids experiments were performed as previously described^70,71. Briefly, whole prostate tissues were collected from indicated mouse lines. Tissues were minced in a 6-cm culture plate and digested in 5mg/ml collagenase type II with 10 μM Y-27632 1 to 1.5 hours at 37°C on a shaker. Glandular structures were then washed with Advanced DMEM/F12 and centrifuged at 150g for 5 mins at 4°C. After that, structures were digested in 1 ml TrypLE (Life Technologies, cat. no. 12605-010) with Y-27632 10 uM at 37°C for 10 min. Structures were washed and mixed in Matrigel (Corning, cat. no. 356231) at the dilution 2:1 (Matrigel: basal medium). A 40 ul drop of this mixture was plated in the center of 24-well plate. The plate was turned upside down in CO₂ incubator (5% CO₂, 37 °C) for 15 mins to allow the Matrigel to solidify. Organoids were cultured in Advanced DMEM/F12 supplemented with B27 (Life Technologies, cat. no. 17504-044), 10mM HEPES (Life Technologies, cat. no. 15630-056), Glutamax (Life Technologies, cat. no. 35050-068), Penicillin/Streptomycin (Life Technologies, cat. no. 15140-122), 1.25mM N-acetylcysteine (Sigma-Aldrich, cat. no. A9165), 100 ng/ml recombinant Noggin (Peprotech, cat. no. 250-38), 100 ng/ml recombinant R-spondin1 (R&D Systems, cat. no. 3474-RS-250), 50 ng/ml EGF (PeproTech, cat. no. AF-100-15), 1nM dihydrotestosterone (Sigma, cat. no. A8380), 200nM A83-01 (Tocris Bioscience, cat. no. 2939) and 10 μM ROCK inhibitor Y-27632 (only for the first week). To activate Cre recombinase and trace BCs, fully established organoids (after 5-7 days) were then treated with 1 uM of 4-OH-TAM for 48h. Afterwards, the medium was refreshed every 2–3 days without TAM. DMSO (0.1%), JSH-23 (1uM) and Ruxolitinib (10uM) were maintained in the medium before collecting for analysis.

RNA-sequencing

Mouse prostate BCs and LCs were FACS isolated as described in ‘Cell labelling, flow cytometry and sorting’. RNA was extracted from FACS-isolated cells using the RNeasy Micro Kit (QIAGEN), following the manufacturer’s instructions. Before sequencing, the quality of RNA was evaluated by Bioanalyzer 2100 (Agilent). Indexed complementary DNA libraries were prepared using the Ovation Solo RNA-seq System (NuGEN) according to the manufacturer’s instructions. Multiplexed libraries were loaded onto flow cells and sequenced using the NovaSeq 6000 S2 Reagent Kit (200 cycles) on a NovaSeq 6000 System (Illumina). Sequencing generated approximately 25 million paired-end reads per sample.

RNA-seq analysis

Quality of each raw dataset was assessed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The adaptor sequences and low-quality regions were trimmed by TrimmomaticPE (ver.0.39)⁷². Trimmed reads were aligned to the mouse reference genome (Grcm38.87) using the STAR aligner (v2.7.5a)⁷³. Genome annotations for Grcm38.87 were obtained from Ensembl (ftp.Ensembl.org). Duplicate reads were removed using the Picard (v2.1.1; http://broadinstitute.github.io/picard/) MarkDuplicates. Following transcript assembly, gene-level counts were generated with HTSeq (v.0.11.1)⁷⁴ and normalized to count per 20 million. The average gene expression of each gene was calculated for each cell population using at least two biological replicates and fold changes were computed between subpopulations.

For the differentially regulated genes across the conditions, genes with a fold change ≥ 2 were classified as upregulated, while those with a fold change ≤ 0.5 were classified as downregulated. Genes with fewer than 10 counts in any individual sample were excluded from the differential expression analysis. Genes upregulated in each signature were tested for enrichment in each Gene Ontology (GO) class using the DAVID⁷⁵ v.2023q4 web server.

ATAC sequencing

For ATAC-seq, 100,000 BCs and LCs were FACS isolated from pooled mouse prostate lobes (> 3 mice). For lobe-specific analysis, BCs were separately FACS isolated from the VP, DLP and AP. Sorted cells were collected in 1 ml PBS with 3% FBS on ice. Cells were centrifuged and cell pellets were resuspended in 100 μl lysis buffer (TrisHCl 10 mM, NaCl 10 mM, MgCl2 3 mM, Igepal 0.1%). Following centrifugation at 500g for 25 min at 4 °C, nuclei were resuspended in 50μl tagmentation mix (Nextera DNA sample preparation kit, Illumina). The reaction was performed at 37 °C for 30 min and DNA was purified using the MiniElute purification kit (QIAGEN) following the manufacturer’s protocol. DNA libraries were PCR amplified (Nextera DNA Sample Preparation Kit, Illumina) and size selected from 200 to 800 bp (BluePippin, Sage Sciences), following the manufacturer’s recommendations.

ATAC-seq analysis

Adaptor sequences were trimmed using TrimmomaticPE (ver.0.39)⁷² with the following parameters: HEADCROP:10 CROP:70 ILLUMINACLIP:adaptor_file:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36. ATAC–seq paired-end reads were then aligned to mouse reference genome (Grcm38) using Bowtie2 (v.2.2.6)⁷⁶ with the parameters: X 2000 --fr --very-sensitive --no-discordant --no-unal --no-mixed --non-deterministic. Mitochondrial reads, reads mapped to unmapped or random contigs and those with a mapping quality below 20 were removed using SAMtools⁷⁷. Duplicate reads were suppressed by Picard (ver. 2.1.1) MarkDuplicates module. Peak on each individual sample were called using MACS2 (v.2.1.0.20151222)⁷⁸ with the parameters: -f BAMPE -g mm --nomodel --shift 0 -q 0.01. Peaks from the different subpopulations were merged for downstream analysis. Reads counts for each merged peak in individual samples were calculated by HTSeq-count⁷⁴ using options ‘-f bam -r pos -m intersection-nonempty’. These counts were normalized to one million mapped reads in merged peaks and fold-change was calculated compared to control. Peaks were associated to genes with GREAT software (v.4.0.4)⁷⁹ with the following parameters: 5.0 kb in proximal upstream, 1.0 kb in proximal downstream and 100.0 kb in distal. For further analysis, the peaks which are not annotated to any of genes were excluded. Differential peaks are defined as peaks having at least a two-fold change compared to control and being called peak in the expanded condition.

De novo motif search was performed using findMotifsGenome.pl tool in HOMER suite⁸⁰, searching for motifs 6–12 bp within ±250 bp from the peak center. For motif prediction, differently regulated peaks are given as targets, while non-regulated peaks from the merged peak set, identified using the intersect function of BEDTools (v2.27.0)⁸¹, were used as background.

Single-cell RNA sequencing

For CD1 mice, all epithelial cells including BCs and LCs (LIN-, EpCAM+) were FACS-isolated. For K5CreER/Pten^fl/fl/RosaYFP mice after 6-week TAM induction, BCs (LIN-YFP+ CD49f^high EpCAM⁺), LCs (LIN- YFP+ CD49f^low EpCAM⁺) and intermediated cells were FACS-isolated. For K5CreER/Pten^fl/fl/RosaYFP mice after 10-month TAM induction, LIN- YFP+ EpCAM+ cells were FACS-isolated.

Single cells were dissociated and loaded onto Chromium Single Cell 3′ microfluidic chips (V2-chemistry, PN-120232, 10X Genomics) and barcoded with a 10X Chromium controller according to the manufacturer’s recommendations. RNA was reverse transcribed, amplified and processed for 5′ adaptor ligation and index attachment. Libraries were generated with the Chromium Single Cell 3′ Library Kit (V3-chemistry, PN-120233, 10X Genomics) and sequenced on an Illumina Novaseq 6000 (paired-end, 100bp reads).

Single cell transcriptomic data analysis

Sequencing reads were aligned and annotated with the mm10-2020-A reference dataset as provided by 10X Genomics and demultiplexed using CellRanger (v.6.0.0)⁸² using default parameters.

Quality control and downstream analysis were performed using the Seurat R package (v.4.2.0)⁸³. For each sample, cells were retained if they expressed between 2500 and 7500 unique genes and had less than 15% of total UMI counts derived from mitochondrial transcripts. Read counts were normalized with the NormalizeData() function of Seurat, with parameter ‘normalization.method = "LogNormalize" and scale.factor=10000’. Principal component analysis (PCA)⁸⁴ were performed for each sample was calculated using the scaled expression data of the 2000 most variable genes, identified as outliers on a mean–variability plot via the FindVariableGenes() function. Uniform manifold approximation and projection (UMAP)⁸⁵ and graph-based clustering were carried out using Seurat with default parameters, using the first 30 principal components as input.

Clusters showing high expression of stromal (Vim, Zeb2, Fscn), fibroblast (Pdgfra, Fn1, Col1a1, Fbn1) and macrophage markers (Ptprc, Cd86, Cd68, Ccr1) were excluded. Dimensionality reduction was recalculated after removal of these non-epithelial cell populations. The final clustering resolution was selected based on epithelial heterogeneity, including BC (Krt14, Krt5, Trp63), Proximal prostate (Krt4, Psca, Wfdc2, Clu, Ppp1r1b, Ltf), Anterior-Dorsal (Nkx3.1, Tgm, Gsdma), Ventral (Sbp, Sbpl, Spink1) and Lateral lobes (Msmb, Cldn10). Upon Pten-deletion, additional cell populations composed of Hillock (Krt13), Hybrid (Krt6a, Ly6d, C1rb, Ren1), chemokine-enriched LC (Cxcl1, Cxcl2, Cxcl5, Cx3cr1), Interferon-enriched LC (Irf7, Ifit1, Ifitm1, Ifitm2, Ifi202b, Irf6) and MHC Class II antigen-enriched LCs (Cd74, H2-Aa, H2-Ab1, H2-DMb1) had emerged. Cell cycle phases were inferred using the CellCycleScoring() in Seurat, based on S-phase and G2/M gene expression.

Marker genes for each cluster were identified using the Wilcoxon rank-sum test implemented in the FindAllMarkers() function in Seurat. P-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR)⁸⁶ method in R. Genes expressed in ≥ 25% of cells within a cluster and with an average log2 fold change ≥ 0.25 were retained. Differentially expressed genes in the hybrid and hillock populations were identified using the FindMarkers() function in Seurat (parameters: logfc.threshold = 0.25, min.pct = 0.25, only.pos=T). Genes with FDR-adjusted p-value < 0.01 were considered significant.

BC/LC-specific markers were defined from the WT dataset as genes with an average log□ fold-change > 0.5, adjusted p-value < 0.01, and expression in ≥ 35% of the respective population. For each cell, the proportion of expressed markers was adjusted by modelling its linear relationship with the total number of genes detected, to correct the differences in sensitivity due to the different sequencing depth per sample.

Gene regulatory network analysis was performed using pySCENIC (ver.0.11.2)³⁶ with default parameters. To correct for stochastic variation, the pipeline was run 10 times for dataset and the average AUC (Area Under the Curve) scores were calculated for downstream analysis. Differentially activated regulons for each cluster were identified by Wilcoxon rank-sum test of AUC values across clusters. Regulons with an adjusted p-value < 0.01 were considered significantly activated.

Lineage trajectory inference was performed using slingshot (v.2.0.0)⁸⁷. To reduce pseudotime distortions from proliferative states, cells and clusters with high expression of proliferation or metabolism-related genes were excluded. Trajectory robustness was confirmed by consistent results across PCA- and UMAP-based embeddings and multiple permutations.

Quantitative RT-PCR

Total RNA of FACS isolated BCs was extracted using the Direct-zol RNA Microprep Kit (Zymo) following the manufacturer’s instructions. Genomic DNA was removed by on-column DNase treatment. cDNA was synthesized using the RevertAid First Strand cDNA Synthesis Kit (ThermoFisher). Quantitative reverse transcription PCR (qRT-PCR) was performed using SYBR Green Supermix (Applied Bioscience) on a Light Cycler 96 system (Roche). Relative gene expression was normalized to the housekeeping gene Gapdh. The following probes from Eurogentec were used: Gapdh forward: AGGTCGGTGTGAACGGATTTG, Gapdh reverse: TGTAGACCATGTAGTTGAGGTCA, Ila1 forward: CGAAGACTACAGTTCTGCCATT, Ila1 reverse: GACGTTTCAGAGGTTCTCAGAG, Cxcl2 forward: CCAACCACCAGGCTACAGG, Cxcl2 reverse: GCGTCACACTCAAGCTCTG.

Human Public Single-cell data analysis

Processed human prostate cancer datasets were downloaded from various sources: GitHub repository of the author³⁸, the Prostate Cell Atlas³⁹, and the GEO database (GSE181294⁴⁰; GSE185344⁴¹). For datasets not originally provided as Seurat objects, data were imported and converted using the Seurat R package (v.4.2.0)⁸³ to enable downstream analysis.

Reprogramming gene signatures were derived from our PTEN 6W scRNA-seq dataset using a Wilcoxon rank-sum test via FindAllMarkers() function in Seurat. P-values were adjusted with Benjamini–Hochberg FDR method⁸⁶ using p.adjust() function in R. Genes expressed in BC Hillock, hybrid BC proximal-like (HY BC Prox), or Proximal populations in ≥ 35% of cells, with average log2 fold-change ≥ 0.85 and adjusted p-value < 0.01 were retained. The top 16 differentially expressed genes per cluster were selected for downstream analysis. Mouse-to-human orthologs were identified using Ensembl BioMart by aligning Grcm38.87 (mouse) to GRCh38 (human).

Geneset enrichment analysis across four public human datasets were performed using AddModuleScore() in Seurat (default parameters). Module scores and marker gene expression were visualized with ggplot2 (v. 3.5.1) and viridis R package (version 0.6.5).

IHC Staining on human prostate cancer sections

Formalin-fixed, paraffin-embedded human prostate cancer samples representing different Gleason Grade Groups (GG) were collected from the Centre Universitaire inter Regional d’Expertise en Anatomie Pathologique Hospitalière (CurePath), Jumet, Belgium. Histopathological diagnoses were reviewed according to the 2022 WHO Classification. For each patient, one representative paraffin was used for immunohistochemical analyses. Immunohistochemistry for STAT1 and PIGR was performed on 4 µm formalin-fixed, paraffin-embedded sections using the Ventana Discovery ULTRA (Ventana Medical Systems, Inc., Tucson, AZ, USA) and the ChromoMap detection system according to the manufacturer’s protocols. Tissue sections were deparaffinized, rehydrated and subject to antigen retrieval using Discovery Cell Conditionner 1 (Tris-EDTA, pH7.8; Roche) for 40 min at 95°C. Slides were incubated with peroxidase blocking solution for 16 min. Primary antibodies were diluted in Discovery Antibody Diluent and incubated as follows: anti-PlGR at 60°C for 60 min, anti-STAT1 at 37°C for 60 min. Tissues were then incubated EnVision Flex-HRP (Agilent, cat. No. K400311-2) at 37°C for 2h. AQP-3 immunostaining was performed on the Autostainer Link 48 (Agilent Technologies, Belgium). Tissue sections were deparaffinized, rehydrated and subject to antigen retrieval using EnVision FLEX Target Retrieval High pH Solution (EDTA, pH9; Agilent, cat. No. K800421-2), for 30 min at 97°C on the PT Link (Agilent Technologies, Belgium). Peroxidase blocking was performed for 5 min, followed by incubation with Anti-AQP3 at RT for 30 min. The slides were washed and incubated with the EnVision+System-HRP labelled Polymer anti-rabbit Ig antibody for 20min (Agilent, cat. No. K400311-2). Signal was developed using diaminobenzidine and hydrogen peroxidase, followed by hematoxylin staining, dehydration and mounting. Whole-slide imaging was performed using a NanoZoomer S360 (x20 magnification, 0.46µm/pixel, Hamamatsu, Hamamatsu-City, Japan)

Statistical analysis and reproducibility

Statistical analyses based on biological replicates or independent experiments were performed using GraphPad Prism v.8.00, R (v.4.2.0) and Excel, with the methods and exact P values indicated in each figure and legend. Two-sided unpaired t-tests were used for two-group comparisons, while ANOVA followed by post-hoc tests was used for comparisons involving more than two groups. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications^8,71. Mice with low induction efficiency were excluded from the analysis and no other data were excluded. Individual data points were shown and were assumed to be normal, but this was not formally tested. Data collection and analysis were not performed blind to the conditions of the experiments. For the drug treatment, male mice were randomly divided into different treatment groups at the same age and genotype.

Extended Data

Extended Data Fig. 1. Pten deletion induced SC plasticity and tumor initiation in a cell of origin and region-specific manners.

(a) FACS gating strategy to analyze and isolate BCs and LCs from K5CreER/Pten^fl/fl/RosaYFP mice 6 weeks after TAM injection. Same strategy for all lineage tracing mice in this study. (b) Quantification of the % of Lin⁻ YFP+ LCs in total YFP+ cells in VP and DLP at different time points after TAM injection in K5CreER/Pten^fl/fl/RosaYFP mice. n=3 mice for 1w, 4w, 7w, 8w and 20w, n=4 mice for 6w and 12w. (c) Representative FACS plot of CD49f and EpCAM expression in Lin- YFP+ prostate epithelial cells from K5CreER/RosaYFP mice 6 week after TAM administration. (d) Quantification of the % of Lin⁻ YFP+ BCs and LCs in total BCs and LCs in VP, DLP and AP in K5CreER/RosaYFP mice after 6-8 weeks TAM administration. n=6 mice. (e) Quantification of the % of YFP+ K14+ and YFP+ K8+ in total K14+ and K8+ in VP, DLP and AP in K5CreER/RosaYFP mice after 4 weeks TAM administration. n=2 mice. (f) Immunostainings of EdU (Red), GFP (Green), K14 (Grey) and K8 (Grey) in the prostate of K5CreER/Pten^fl/fl/RosaYFP at 4w, 6w and 12-week after TAM administration 24 hours after EdU injection. Scale bar, 10 μm. n=3 mice. (g) Quantification of percentage of EdU doublet following basal cell division giving rise BC-BC, BC-LC and LC-LC doublets. n=152 EdU doublets (4w), n=65 EdU doublets (6w) and n= 142 EdU doublets pooled from 3 mice. (h) Genetic strategy to lineage trace LC following Pten deletion (i) Quantification of Lin- YFP+ BCs and LCs in total BCs and LCs in VP, DLP and AP at different time points after TAM administration in K8CreER/Pten^fl/fl/RosaYFP mice. n=3 mice. Graphs are mean ± s.e.m.

Extended Data Fig. 2. mTOR inhibition blocks multipotency and tumor initiation.

Representative immunostaining of the prostate from (a) K5CreER/Pten^fl/fl/RosaYFP mice 6 weeks after TAM administration (K5-PTEN 6w) and (d) K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice 3 weeks after TAM administration (K5-PTEN-PIK 3w) treated with EtOH or Rapamycin (6mg/kg, three injections per week) using anti-GFP (green), anti-K8 (red) and anti-K14 (grey) antibodies. Arrows indicate hybrid cells (YFP+ K14+ K8+). Scale bar, 20 μm (upper row), 10 μm (lower row). n=3 mice. Representative FACS plot of CD49f and EpCAM expression in Lin- YFP+ prostate epithelial cells of (b) K5-PTEN 6w mice and (e) K5-PTEN-PIK 3w mice treated with EtOH or Rapamycin. Quantification of % of LIN- YFP+ LCs in total YFP+ cells in the whole prostate of (c) K5-PTEN 6w mice and (f) K5-PTEN-PIK 3w mice treated with EtOH or Rapamycin. n=6 for K5-PTEN 6w mice, n=4 for K5-PTEN-PIK 3w mice. (g) H&E-stained histological sections and (h) quantification of different types of tumorigenic lesions along prostate tumor progression in DLP of K5-PTEN 6w mice treated with EtOH (n=7 mice) or Rapamycin (n=8 mice). Scale bar, 250 μm (upper row), 50 μm (lower row). Graphs are mean ± s.e.m. p-values are derived from two-sided unpaired t-test.

Extended Data Fig. 3. Marker genes for each cell type in control and in K5-PTEN 6w prostate epithelial cells.

Cell populations found by scRNA-seq in CTL prostate epithelial cells: (a) UMAP dimensionality reduction plots of CTL prostate epithelial cells with colors representing unsupervised clustering. (b-g) UMAP plots colored by normalized gene expression for (b) BC marker genes, (c) Proximal marker genes, (d) Ventral marker genes, (e) Lateral marker genes, (f) Antero-Dorsal marker genes and (g) Proliferative marker genes. (h) UMAP dimensionality reduction plots for K5-PTEN 6w using Seurat: UMAP dimensionality reduction plots with colors representing unsupervised clustering. (i-o) UMAP plots colored by normalized gene expression for (i) BC marker genes, (j) Proximal marker genes, (k) HY BC Prox marker genes, (l) HY Nkx3.1 marker genes, (m) Ventral marker genes, (n) Antero-Dorsal marker genes and (o) Proliferative marker genes.

Extended Data Fig. 4. Cellular trajectory and in situ characterization of Hillock, Hybrid and proximal-like LCs.

(a) Cell populations of K5-PTEN 6w, which are used for trajectory inference: UMAP dimensionality reduction plots with different colors representing unsupervised clustering. (b) UMAP plot for the K5-PTEN 6w dataset showing the trajectory from BC to proximal LCs. (c-d) UMAP plot colored by the adjusted proportion of (c) BC-specific marker genes and (d) LC-specific marker genes. (e-h) Representative immunostaining of AP/DLP of CTL, K5CreER/Pten^fl/fl/RosaYFP mice 6 weeks after TAM administration, K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice 3 weeks after TAM administration and K5CreER/Pten^fl/fl/P53^fl/fl/RosaYFP mice 4 weeks after TAM administration using (e) anti-K13, (f) anti-Aqp3, (g) anti-K4, (h) anti-Trop2 and anti-GFP (green), anti-K14 (grey) and anti-K8 (grey) antibodies. Arrows indicate Hillock cells, Hybrid cells and Proximal LCs as indicated markers. n=3 mice. Dash lines marked the expanded LCs. L: prostate lumen. Scale bar, 20 μm.

Extended Data Fig. 5. scRNA- seq characterize hybrid basal–luminal cell states and lineage trajectory in K5-PTEN 6w prostate epithelial cells.

(a) Cell populations of K5-PTEN 6w, which are used for trajectory inference: UMAP dimensionality reduction plots with different colors representing unsupervised clustering. (b) UMAP plot for the K5-PTEN 6w dataset showing the trajectory from BC to ventral LCs. (c-d) UMAP plot colored by the adjusted proportion of (c) BC-specific marker genes and (d) LC-specific marker genes. (e) Representative images of immunostaining of the VP of K5CreER/Pten^fl/fl/RosaYFP mice 6 month after TAM administration using anti-GFP (green), anti-Nkx3.1 (red). Scale bar, 50 μm. n=3 mice. (f) Representative images of immunostaining of the VP, DLP and AP of K8CreER/Pten^fl/fl/RosaYFP mice 6 months after TAM administration using anti-GFP (green), anti-K13 (red), anti-K4 (red) and anti-K14 (grey) antibodies. Scale bar, 50 μm. n=3 mice. (g) Genes significantly up-regulated on HY BC Prox compared to BC p63 low and BC p63 high in K5-PTEN 6w scRNA dataset. (h) Genes significantly up-regulated in HY BC Prox compared to proximal luminal cells. (i) Genes significantly up-regulated on HY Nkx3.1 compared to the BC Nkx3.1. (j) Genes significantly up-regulated in HY Nkx3.1 compared to the ventral prostate. (k-m) UMAP plots colored by (Left) SCENIC AUCs for regulon activation and (Right) normalized gene expression of the TF for (k) Elf3, (l) Grhl3 and (m) Creb5. For g to j, red dots are significantly differentially expressed genes (FDR adjusted P-value < 0.01) and blue-labeled genes are uniquely expressed on hybrid population. P-values were calculated using the Wilcoxon rank-sum test and adjusted using the Benjamini–Hochberg FDR method.

Extended Data Fig. 6. Temporal analysis of the activation of innate immunity and ATAC-seq and bulk RNA-seq on different robes after Pten deletion in BC.

(a) Relative mRNA expression levels of Il1a and Cxcl2 were determined by quantitative RT-PCR in FACS isolated BCs of AP/DLP from CD1 mice and YFP+ BCs from K5CreER/Pten^fl/fl/RosaYFP mice at indicated time after TAM injection. mean ± s.e.m. n=3 mice. p-values are derived from one-way ANOVA with Dunnett test. ATAC-seq peaks of (b) Krt13 and (c) Aqp3 genes from FACS isolated BCs of VP, DLP and AP of K5CreER/Pten^fl/fl/RosaYFP mice 6 weeks after TAM administration. Scale for visualization: 0-52 (Krt13) and 0-30 (Aqp3). Peaks which are up-regulated at least 2-fold on AP and DLP compared to VP are highlighted with orange box. (d) TF motif enrichment analysis of peaks upregulated in BCs of AP/DLP compared to BCs of VP in K5-PTEN 6w. P-values were calculated using a binomial test. (e) Relative Erg expression from bulk RNA-seq in FACS isolated CTL BC, CTL LC from CTL mice and YFP+ BCs (BC PTEN) and YFP+ LCs arising from Pten deleted BCs from K5CreER/Pten^fl/fl/RosaYFP mice 5-weeks after TAM injection (LC PTEN). n=2. (f) Relative Erg expression from bulk RNA-seq of FACS isolated LCs from AP/DLP versus VP in K5CreER/Pten^fl/fl/RosaYFP mice 6 months following TAM administration. (n=3 for AP/DLP, n=2 for VP).

Extended Data Fig. 7. Activation of innate immunity is a mTOR dependent.

(a) Venn diagram illustrating the number of 2-fold upregulated genes in FACS isolated LIN- YFP+ BCs of K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice 3-weeks after TAM injection (vehicle treated, BC_PTEN-PIK) compared to FACS isolated LIN- BCs of CD1 mice (vehicle treated, BC_CTL) and the number of 2-fold downregulated genes in FACS isolated LIN- YFP+ BCs of K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice 3-weeks after TAM injection treated with Rapamycin (BC_PTEN-PIK Rap) compared to BC_PTEN-PIK. (b) GO analysis of genes upregulated more than 2-fold in FACS isolated BC PTEN-PIK compared to FACS isolated BC CTL. (c) GO analysis of genes downregulated more than 2-fold in FACS isolated BC PTEN-PIK Rap compared to FACS isolated BC PTEN-PIK. (d) Expression of genes on bulkRNA-seq in FACS isolated BC CTL, BC PTEN-PIK and BC PTEN-PIK Rap. n=2 samples. (e) Representative immunostaining of myeloid cells within AP from K5CreER/Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice treated with EtOH or Rapamycin 3-weeks after TAM injection using anti-GFP (green), anti-Ly6G (red) antibodies. Scale bar, 20 μm. (f) Representative immunostaining of myeloid cells within the DLP from K5CreER/Pten^fl/fl/RosaYFP mice and K5CreER/Pten^fl/fl/P53^fl/fl/RosaYFP mice 4 weeks after TAM injection using anti-GFP (green), anti-Ly6G (red) antibodies. Scale bar, 50 μm. For b and c, P-values were derived from two-tailed modified Fisher’s Exact Test and adjusted using the Benjamini–Hochberg method.

Extended Data Fig. 8. Regional reprogramming and activation of innate immunity in BC derived mouse prostate tumors.

(a) UMAP dimensionality reduction plots with different colors representing unsupervised clustering of scRNA-seq of FACS isolated LIN- YFP+ epithelial cells from K5CreER/Pten^fl/fl/RosaYFP mice 10 months after TAM administration: (b-h) UMAP plots colored by normalized gene expression for (b) Proximal marker genes, (c) Proliferative marker genes, (d) HY BC Prox marker genes, (e) Distal LC marker genes, (f) Chemokine marker genes, (g) Interferons marker genes, (h) MHC Class II antigens marker genes. (i-n) Representative immunostaining and quantification of (i, j) K13, (k, l) Aqp3 and (m, n) K4 in VP, DLP and AP of K5CreER/Pten^fl/fl/RosaYFP mice 6 months following TAM administration. Scale bar, 50 μm. n=4 mice for K13 and Aqp3, n=3 mice for K4. (o) Relative mRNA expression from bulk RNA-seq of the indicated genes in FACS isolated YFP+ LCs from AP/DLP and VP in K5CreER/Pten^fl/fl/RosaYFP mice 6 months following TAM administration. (n=3 for AP/DLP, n=2 for VP). Representative immunostainings of AdeK5Cre infected AP of (p) Pten^fl/fl/P53/RosaYFP mice and (q) Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice 2 months after virus injection using the indicated antibodies. Scale bar, 50 μm. (r) Quantification of the % of K13+, Aqp3+ and K4+ in total YFP+ and Ly6G+ cells in AdeK5Cre injected AP of Pten^fl/fl/Pik3ca^H1047R/RosaYFP mice. n=3 infections. Graphs are mean ± s.e.m. p-values are derived from one-way ANOVA with Tukey’s test.

Extended Data Fig. 9. Regional reprogramming and activation of innate immunity in human prostate tumors.

(a) Cell populations on ERG-driven human prostate cancer cells, human prostate cancer atlas, treatment-naïve prostate adenocarcinoma and invasive cribriform carcinoma and intraductal carcinoma data. (b) UMAP dimensionality reduction plots for KRT13 and WFDC2 with color scaling representing the level of gene expression. (c-f) Heatmap of reprogramming marker genes defined on K5-PTEN 6W scRNA-seq data on (c) ERG-driven human prostate cancer cells, (d) human prostate cancer atlas, (e) treatment-naïve prostate adenocarcinoma and (f) invasive cribriform carcinoma and intraductal carcinoma data. (g-i) Slingshot pseudotime trajectory analysis illustrating the lineage trajectory on (g) ERG-driven human prostate cancer cells, (h) human prostate cancer atlas and (i) invasive cribriform carcinoma and intraductal carcinoma data. (j) Summary of IHC staining results for PIGR and AQP3 in prostate tumor samples from 136 patients, categorized by GG groups of prostate cancer. Green boxes indicate co-staining of PIGR and AQP3.

Extended Data Fig. 10. Model of the early step of prostate tumor initiation.

Cell plasticity, lineage infidelity and tumor progression following Pten deletion occurs in a region-specific manner during the early stage of prostate cancer initiation and are mediated by the activation of innate immunity in prostate basal stem cells.

Data and materials availability

All raw sequencing datasets that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession number GSE270187 (scRNA-seq), GSE270189 (bulkRNA-seq, rapamycin treated), GSE270190 (bulkRNA-seq, cell fate upon PTEN deletion in BC), GSE286018 (bulkRNA-seq of P65/Pten knock BCs), GSE286019 (bulkRNA-seq of basal-derived and luminal-derived luminal tumors), GSE270191 (ATAC-seq of pooled BC/LC of WT and PTEN-deleted cells) and GSE288787 (ATAC-seq of BCs from different lobes upon PTEN deletion). Previously published human prostate cancer datasets re-analyzed in this study were available from GitHub repository of the author (https://github.com/franklinhuanglab/scRNA-seq-Analysis-of-Prostate-Cancer-Samples)³⁸, the Prostate Cell Atlas (https://www.prostatecellatlas.org)³⁹ and GSE181294⁴⁰, GSE185344⁴¹. The data that support the findings of this study are available from the corresponding author upon request. Source data are provided with this paper.

Code availability

The codes used for data process, downstream analysis and plotting for this paper is available at https://github.com/yurasong/PTEN_codes.