Autocrine activation of MAPK signaling mediates intrinsic tolerance to androgen deprivation in LY6D prostate cancer cells

SUMMARY The emergence of castration-resistant prostate cancer remains an area of unmet clinical need. We recently identiﬁed a subpopulation of normal prostate progenitor cells, characterized by an intrinsic resistance to androgen deprivation and expression of LY6D. We here demonstrate that conditional deletion of PTEN in the murine prostate epithelium causes an expansion of transformed LY6D + progenitor cells without impairing stem cell properties. Transcriptomic analyses of LY6D + luminal cells identiﬁed an autocrine positive feedback loop, based on the secretion of amphiregulin (AREG)-mediated activation of mitogen-activated protein kinase (MAPK) signaling, increasing cellular ﬁtness and organoid formation. Pharmacological interference with this pathway overcomes the castration-resistant properties of LY6D + cells with a suppression of orga-noid formation and loss of LY6D + cells in vivo . Notably, LY6D + tumor cells are enriched in high-grade and androgen-resistant prostate cancer, providing clinical evidence for their contribution to advanced disease. Our data indicate that early interference with MAPK inhibitors can prevent progression of castration-resistant prostate cancer.


In brief
Steiner et al. demonstrate that PTEN deletion in luminal prostate cell causes an expansion of transformed LY6D + progenitor cells with intrinsic castrationresistant properties through autocrine AREG-mediated MAPK activation. Pharmacological interference of MAPK pathway overcomes the castrationresistance properties of LY6D + cells, providing a rationale to target preexistent castration-resistant cells at early stages.

INTRODUCTION
Primary prostate cancer (PCa) comprises a spectrum of clinically, morphologically, and molecularly heterogeneous disease phenotypes, which frequently include multiple tumor foci within the same patient, with diverse genomic and molecular profiles. [1][2][3][4] Genetic lineage-tracing studies, organoid models, and single-cell profiling have been instrumental in interrogating the cellular heterogeneity of the basal and luminal prostate compartments. [5][6][7][8][9][10][11][12][13][14] Within given tumor foci, individual cells and distinct cell lineages differ in their capacity for oncogenic transformation and tumor initiation. Notably, the contribution of diverse cellular subtypes to human PCa is still poorly defined. Recent single-cell studies have dissected the cellular anatomy of the mouse and human prostate, both in hormonally naive and hormone-treated normal prostate. 6,8,9,11,12,15,16 These studies confirmed the considerable complexity of cellular subtypes within the prostatic epithelial compartment. Notably, while the basal lineage was relatively homogeneous, the luminal lineage showed high cellular heterogeneity with multiple cell subtypes, including secretory luminal, stem-like luminal, and periurethral luminal cells. 6,8,9,11,12,16  The development of PCa is strongly associated with loss of PTEN function, 17,18 which frequency increases in metastatic PCa. [19][20][21] PTEN is a critical regulator of key cellular processes in normal and malignant epithelial cells as well as in immune cells and thereby also orchestrates tumor microenvironment responses. 22,23 Numerous studies have demonstrated that both basal and luminal PTEN-deficient cells can be the cell of origin for PCa, albeit with different tumor-initiating latency. 7,14,[24][25][26] These differential contributions of basal and luminal cells to prostate tumor dynamics illustrate the key role of the cell of origin and their unique cell-intrinsic properties to disease progression and potentially also to response to therapies.
We have previously identified LY6D as a cell-surface protein that marks intrinsically castration-resistant (CR) luminal prostate progenitors. 6 In organoid culture, untransformed LY6D + prostate cells exhibit higher organoid-forming potential compared with LY6D À cells, revealing their intrinsic stem and progenitor properties. Building upon these findings, we now define the underlying cell-intrinsic mechanism regulating CR prostate tumors initiated from LY6D + luminal progenitors. We identify that the intrinsic transcriptional regulatory program of LY6D + progenitor cells is required for survival of aggressive, treatment-resistant clones, based on an autocrine signaling pathway in CR LY6D + luminal cells, which involves amphiregulin (AREG) secretion and epidermal growth factor receptor (EGFR) signaling.

LY6D + luminal progenitor cells contribute to malignant transformation of PTEN-deficient tumors
We previously demonstrated that LY6D + luminal prostate progenitor cells are intrinsically resistant to androgen deprivation in non-tumor-prone mice. 6 To investigate whether LY6D + luminal cells also play a significant role for CR PCa, we employed a conditional mouse model, allowing for tumor initiation at any given time by systemic administration of tamoxifen. For this, we used an inducible luminal lineage-specific mouse model, where Pten deletion is induced in Keratin8 (K8 + ) luminal cells (K8-Cre ERT2 ; Pten Flox/Flox ;Rosa26-Stop Flox/Flox -EYFP, hereafter K8PY mice), as well as age-matched Pten-competent mice as controls (K8-Cre ERT2 ;Rosa26-Stop Flox/Flox -EYFP; hereafter K8Y mice) (Figure 1A). Upon tamoxifen treatment ( Figure 1B), activated CRE recombinase drives Pten deletion and expression of the enhanced yellow fluorescence protein (YFP) in K8 + luminal cells ( Figure 1C).
To assess the distribution of LY6D + progenitor cells in adult male mice across the entire organ prior to tumor initiation (1 week post tamoxifen treatment), whole prostates from hor-mone-naive (HN) and castrated (CR) mice were analyzed for the expression of Ly6D mRNA by RNA in situ hybridization (ISH) in the K8 + luminal compartment in 12-week-old males. In keeping with our previous observations, LY6D + luminal progenitors are located in both proximal and luminal regions ( Figure 1D). In contrast to young males (4 weeks old), 6 adult males showed a higher proportion of LY6D + progenitors in distal prostate regions in all lobes, irrespective of whether mice were castrated (CR) or not (HN). Importantly, the frequency of LY6D + cells was significantly higher in CR mice compared with HN males (Figures 1D, S1A, and S1B).
K8-driven loss of Pten triggers formation of low-grade prostate intraepithelial neoplasia (PIN) lesions in all prostate lobes 1 month post tamoxifen treatment. 7 These lesions then progress to highgrade PIN lesions (HG-PIN) 8-16 weeks after tamoxifen-induced recombination, which transform into early prostate adenocarcinoma at 24 weeks. We then evaluated the distribution of LY6D + progenitor cells 48 weeks after Pten deletion once highgrade PIN lesions have developed. In agreement with previous observations, 6 Pten deletion in CR luminal cells led to the development of high-grade PIN lesions and adenocarcinoma with cribriform morphology at earlier time points ( Figure S1C). Our analyses indicated that the proportion of YFP + LY6D + luminal tumor cells increased in PTEN-dependent tumors initiated from both HN and CR prostate luminal cells (Figures 1E and S1D-S1F). We further validated our findings from RNA-ISH staining by immunofluorescence (IF) ( Figure S1G). In keeping with our RNA-ISH data, LY6D + cells were readily detectable in PTENdeficient cells, identified by positive staining for YFP and phosphorylated alpha serine/threonine protein kinase (pAKT) in HN and CR prostates ( Figure 1F).
In conclusion, our data indicate that LY6D + luminal cells are present in premalignant lesions and their frequency increases with PTEN-mediated tumor formation.
Organoid-forming capacity and castration resistance of LY6D + progenitor cells are not affected by PTEN loss Besides their intrinsically CR properties, LY6D + cells are characterized by increased organoid-forming capacity. 6 PTEN loss has previously been reported to modulate epithelial growth and differentiation of prostate and mammary cells in organoid culture. 5,27,28 We therefore hypothesized that PTEN deletion in the prostate luminal compartment could affect the stem properties and organoid-forming capacity of LY6D + cells. To address this question, we assessed the frequency of luminal progenitor-like cells by isolating YFP + LY6D + and YFP + LY6D À prostate cells from K8Y control and K8PY PTEN-deficient mice (Figures 2A,  2B, and S2A-S2C). Pten deletion in the HN status resulted in a Article ll OPEN ACCESS small increase in YFP + LY6D + cells compared with PTEN-proficient controls; however, the total number of cells in the prostates remained comparable with those of control K8Y mice ( Figures 2C and S2D). In contrast, after castration, Pten deletion resulted in a significant increase in the total number of prostate cells, which were also enriched for LY6D + progenitor cells, resulting in tumor lesions with a high proportion (>80%) of YFP + LY6D + cells ( Figures 2B, 2C, and S2D).
To further study the growth and proliferation potential of PTEN-deficient organoids, we assessed organoid area and symmetrical growth (roundness). YFP + LY6D + progenitor-derived organoids from PTEN-proficient cells displayed a luminal-like appearance, whereas PTEN-deficient organoids showed a more tumoroid-like morphology ( Figure 2G). Overall, PTEN-deficient organoids were characterized by a large area (>100 mm) and loss of a luminal-layer structure, indicated by filled lumens ( Figures 2G and 2I), morphological features resembling hyperplasia of in vivo PIN lesions. The multi-lobular appearance of the K8PY organoids resulted in decreased roundness compared with the K8Y organoids ( Figures S2F and S2G). Altogether, these data indicate that the YFP + LY6D + subset is enriched in progenitor-like cells, regardless of PTEN activation and androgen stimulation.
We next assessed the lineage specificity of the in vitro PTENnull organoids derived from YFP + LY6D + luminal cells. Multilineage structures composed of both basal K5 + /p63 + and luminal K8 + cells were observed in organoids derived from both YFP + LY6D + and YFP + LY6D À luminal cells ( Figures 2J and 2K). Importantly, YFP + LY6D + cell-derived organoids were predominantly bipotent as indicated by double-positive (p63 + K8 + ) cells ( Figure 2J), in contrast to YFP + LY6D À cell-derived organoids, which contained cells positive for either K8 or p63 ( Figure 2K). To validate the presence of bipotent LY6D + cells in vivo, we stained tumor tissues from HN K8PY mice for luminal and basal markers. In keeping with our findings from ex vivo organoids, LY6D + cells also maintained a bipotent phenotype in vivo, indicated by the presence of cells positive for K8, K5, and p63 ( Figure 2L).
In conclusion, our results demonstrate that LY6D distinguishes a subpopulation of CR luminal cells that are consistently enriched in stem/progenitor properties, which is maintained upon PTEN-loss-driven transformation.
PTEN-deficient LY6D + luminal tumor cells maintain a luminal progenitor-specific transcriptome Our data establish LY6D + progenitor cells as intrinsically CR cells present not only in normal but also in transformed prostate tissues. To define how LY6D + luminal progenitor cells contribute to PCa, we performed gene expression profiling by RNA sequencing (RNA-seq) on fluorescence-activated cell sorting (FACS)-isolated YFP + LY6D + and YFP + LY6D À prostate luminal subpopulations, derived from PTEN-deficient mice 4 months post tamoxifen treatment ( Figure 3A).
Principal-component analysis (PCA) of RNA-seq data revealed a clear distinction between LY6D + and LY6D À luminal cells derived from PTEN-deficient mice, in both HN and CR conditions ( Figure 3B). The transcriptional variance across the LY6D + and LY6D À luminal cells from castrated mice was lower compared with their HN counterparts, which was expected as castration predominantly affects the transcriptional profile within the resistant subpopulations. 11 We first analyzed the transcriptional profile of LY6D + luminal cells from both HN and CR PTEN-deficient mice ( Figure 3C). Differential expression analysis and gene set enrichment analysis (GSEA) revealed that the LY6D + luminal subsets only differ (false discovery rate [FDR] <0.05) in genes involved in the androgen receptor (AR) response pathway between HN and CR groups ( Figures 3C and 3D). Notably, LY6D + luminal subsets from HN and CR mice had comparable Ar expression levels, while the expression levels of AR target genes significantly decreased upon castration ( Figure 3E), indicating that these transcriptional differences were not due to variations in   To further define the characteristics of PTEN-deficient LY6D + luminal tumor cells, we compared their transcriptional profile with YFP + LY6D À tumor cells from treatment-naive or castrated K8PY mice. Differential gene expression analysis (FDR < 0.05) revealed a conserved transcriptional program of LY6D + luminal tumor cells in both HN and CR conditions compared with their LY6D À luminal counterparts ( Figures 3I and S3A). Importantly, and in keeping with our IF data ( Figure 2J To assess the effect of PTEN deletion on the transcriptional profile of YFP + LY6D + cells, we analyzed genes differentially expressed between LY6D + cells derived from Pten wild-type (K8Y) and Pten-null (K8PY) prostates. GSEA indicated upregulation of multiple oncogenic pathways (FDR < 0.05) by PTEN deficiency in LY6D + as well as LY6D À luminal cells, including the upregulation of AKT and mammalian target of rapamycin (mTOR) pathways ( Figures S3D and S3E).
Collectively, these results indicated that PTEN deficiency activated pro-tumorigenic genes in YFP + LY6D + cells, which maintain their pre-existing transcriptional profile of intrinsically CR LY6D + luminal progenitors.
Distinct regulation of growth factor pathways in PTENdeficient LY6D + luminal cells Differential expression analysis between LY6D + and LY6D À luminal cells revealed a transcriptional pattern characteristic for each of these cell populations. To define signaling pathways orchestrating the differential gene expression, we performed ingenuity pathway analysis (IPA) of the transcriptomic profiles of luminal tumor cells. These analyses indicated a significant (FDR < 0.05) upregulation of ''growth factor signaling pathways,'' ''inflammatory response,'' and ''integrin signaling'' present in YFP + LY6D + cells, conserved between HN and CR states (Fig-ure 4A). Intriguingly, the ''regulation of epithelial-mesenchymal transition (EMT) by growth factor pathway'' appeared to be of particular interest to us, as both the ''upregulation of the genes involved in EMT'' and ''activation of epidermal growth factor receptor family (ERBB)-extracellular signal-regulated kinases (ERK) pathways'' have previously been associated with stemness and castration resistance. 29,30 Additionally, ERK5, STAT3, and Phospholipase C signaling, implicated in tumor growth and chemoresistance, [31][32][33] were upregulated in YFP + LY6D + tumor cells, suggesting high ERK activity in these cells ( Figures 4A, S4A, and S4B).
To further define canonical growth factors driving these signaling pathways, we applied GSEA to the transcriptomic data from PTEN-deficient LY6D + and LY6Dluminal cells. ERK and EGFR-PI3K signaling pathways were significantly enriched in PTEN-deficient YFP + LY6D + tumor cells compared with their YFP + LY6D À counterparts, in both HN and CR mice ( Figure 4B). These data suggest that not only mitogen-activated protein kinase (MAPK) but also PI3K signaling may sustain LY6D + tumor cell survival in vivo compared with PTEN-deficient LY6D À tumor cells.
To define the underlying mechanisms contributing to the activation of EGFR/ERK pathways observed in our GSEA, we analyzed the expression of both Erbb family receptors and ligands in PTEN-deficient tumor cells (LY6D + versus LY6D À luminal cells). Comparative expression analysis across all conditions (HN and CR) revealed high levels of the Erbb family, low-affinity ligand amphiregulin (Areg) in YFP + LY6D + luminal cells compared with LY6D À luminal cells ( Figure 4C). As diverse ERBB ligands can drive differential cellular responses, 34,35 we next assessed the expression of all Erbb family members (Figure 4D). Importantly, we observed a significant upregulation of the EGFR ligand Areg and the simultaneous downregulation of Egf in LY6D + luminal tumor cells in all conditions. In addition, Egfr was minimally upregulated (p < 0.001) in LY6D + luminal cells from HN animals ( Figure 4D). In contrast, minor or no significant changes were observed in the expression of other Erbb receptors (Erbb2 and Erbb3) or ERBB ligands (Tgfa, Btc, Hb-egf), and Erbb4 and Ereg were also not detected. LY6D + cells from K8Y control mice showed a concordant expression pattern of Erbb family members ( Figures S4C-S4F). A positive correlation between Egfr and Areg expression was further demonstrated by gene expression analysis of control and PTEN-deficient  Figures 4E and S4C). Notably, AREG has previously been shown to inhibit epidermal growth factor (EGF) expression during tissue repair, 36 suggesting a direct inhibitory effect of AREG on EGF expression in LY6D + luminal cells. This hypothesis was further supported by our previous single-cell RNA-seq (scRNA-seq) dataset ( Figure S4G 12 ). Last, we assessed Areg and Egf expression in publicly available datasets from defined stages of murine prostate tumor development. 37 Notably, Areg levels increased with disease progression from low-grade PIN to high-grade PIN stage, while Egf expression levels dropped from early transformation onward ( Figures S4H  and S4I). We therefore hypothesize that an expansion of LY6D + tumor cells during tumor progression contributed to the observed shift between Areg and Egf ligands.
To identify transcription factors (TFs) that could govern the gene expression profiles of luminal YFP + LY6D + versus YFP + LY6D À subpopulations, we performed discriminant regulon expression analysis (DoRothEA). 38,39 The analysis of highest-confidence regulons (i.e., levels A and B) revealed an enrichment in the activity of 44 and 38 TFs in the HN and CR groups, respectively. Since YFP + LY6D + cells are intrinsically CR cells, we focused on the common TF activity in both HN and CR luminal cells (Table S1). The TFs positively enriched (FDR < 0.05) in both HN and CR K8PY YFP + LY6D + cells were mainly involved in stress and growth responses ( Figure S5A), in agreement with the GSEA and IPA analysis. We then assessed whether any of those TFs may be involved in the transcriptional regulation of Areg, which is uniquely upregulated in PTEN-deficient YFP + LY6D + cells. Further evaluation indeed revealed that Areg is a transcriptional target of early growth response 1 (EGR1). 40 Importantly, Egr1 was also found to be significantly enriched in PTEN-intact K8Y luminal YFP + LY6D + cells (both HN and CR; Figures S5B and S5C). Notably, silencing of Egr1 resulted in a pronounced downregulation of Areg expression in prostate luminal cancer cells ( Figure 4F), demonstrating that EGR1 is required for Areg expression. To provide corroborative evidence for this idea, we generated an AREG gene signature from a recent study (GSE116864 41 ), defining the genes differentially expressed (À1 % Log 2 FC % 1.5; FDR > 0.05) on PC3 and DU145 PCa cells upon stimulation with conditioned medium (CM) from AREG-expressing stroma cells. GSEA showed the presence of an AREG-driven gene signature in PTEN-deficient YFP + LY6D + luminal cells compared with LY6D À luminal cells from all conditions (HN and CR K8PY prostates) ( Figure 4G). We also identified an upregulation of EGR1 in AREG-stimulated PC3 cells ( Figure S5D), suggesting a positive feedback loop between AREG and EGR1. Together, these analyses provide evidence that EGR1 is involved in AREG-mediated growth response.
Cell autonomous secretion of AREG activates MAPK signaling in LY6D + prostate cells To test the hypothesis that AREG is an autocrine factor promoting survival and growth of LY6D + PCa cells, we first assessed AREG protein levels in vivo and in vitro. Notably, AREG can be anchored to the membrane, where it can trigger EGFR signaling in a juxtracrine manner. 42 However, membrane-anchored AREG (pro-AREG) can also be cleaved proteolytically to produce mature AREG, 43-45 a soluble form that activates EGFR through autocrine and paracrine signaling. 46,47 In keeping with our transcriptional data, we observed significantly high AREG expression levels in Ly6d-expressing cells in transformed prostates from K8PY mice ( Figures 5A-5C). Importantly, the expression pattern of AREG in LY6D + prostate cells varied from ''nuclear'' to ''cytoplasmic'' and ''membrane,'' suggesting distinct molecular functions of AREG. 48,49 To assess whether AREG was actively secreted by YFP + LY6D + luminal cells, FACS-sorted YFP + LY6D + and YFP + LY6D À cells (i.e., K8PY mice) were seeded in organoid cultures, and subsequently CM collected to evaluate AREG content ( Figure S5E). After 7 days in complete mouse prostate organoid culture medium (mPOM), organoids were washed and incubated for further 24 h in growth-factor-free medium. The secreted form of AREG was measured in organoid-derived CM. Notably, we observed high levels of secreted AREG in the CM derived from YFP + LY6D + organoids, in both HN and CR tumor K8PY prostate cells ( Figures 5D and 5E), demonstrating that AREG is actively secreted by prostate YFP + LY6D + luminal cells. IF of organoids grown in mPOM indeed confirmed that YFP + LY6D + expresses AREG, which was undetectable in organoids derived from YFP + LY6D À cells. Similarly, EGR1 was expressed in LY6D +derived organoids but not their negative counterparts ( Figure 5F).
Primary prostate cells depend on growth factors to proliferate in vitro and in vivo, and EGF has been shown to be a key regulator for prostate organoids. 10,50 Our data suggested that LY6D + luminal cells may be less dependent on exogenous growth factors due to their capacity to endogenously secrete growth factors such as AREG. To test this hypothesis, sorted LY6D + and LY6D À luminal cells were seeded in either complete mPOM (also containing EGF) or in reduced growth factor (rGF) medium without EGF (rGF medium). Strikingly, YFP + LY6D + luminal cells derived from HN K8PY mice were able to grow organoids under reduced growth conditions. Although a lower organoid-forming capacity in rGF conditions, YFP + LY6D + luminal cells maintained a growth advantage compared with YFP + LY6D À cells ( Figures 5G and 5H). A similar result was observed in CR K8PY-derived organoids (Figures S5F and S5G). Importantly, AREG supplementation to rGF medium further promoted organoid outgrowth from YFP + LY6D + cells, irrespective of androgen stimulation ( Figures 5I and 5J). These results demonstrate that YFP + LY6D + -derived AREG secretion promotes organoid formation, which is maintained even in the absence of exogenous growth factors.
Targeting AREG/EGFR/ERK signaling impairs growth of CR LY6D + prostate cells Our data indicate the presence of a constitutive activation of the AREG/EGFR/EGR1 signaling pathway in LY6D + luminal tumor cells. To mechanistically delineate how AREG promotes growth and survival, we investigated AREG signaling in DVL3 cells, a recently established mouse cell line from a treatment-naive PTEN/TP53 prostate adenocarcinoma. 51 Stimulation of cells with exogenous AREG resulted in the rapid phosphorylation of EGFR at Tyr 1086 and Tyr 845 sites ( Figures S6A and S6B), but not at Tyr 992, which is known to be activated by EGF. 35 Further downstream, ERK1/2 is phosphorylated by AREG ( Figures S6A and  S6C), while activation of AKT was unaffected. Notably, protein levels of EGR1 also increased shortly after AREG stimulation in DVL3 tumor cells ( Figures S6A and S6D), suggesting posttranslational stabilization of EGR1 by AREG. AREG-mediated ERK1/2 activation and EGR1 upregulation could be suppressed by pretreatment of cells with the MEK inhibitor trametinib ( Figures 6A,  and S6E). Similarly, erlotinib also inhibited AREG-mediated ERK activation and EGR1 upregulation ( Figures S6F-S6I).
To test whether autocrine AREG promotes cell survival, DVL3 were plated in BME2 matrix in the presence or absence of androgen stimulation (i.e., ±dihydrotestosterone [DHT]) in three conditions: rGF mPOM (lacking EGF), rGF with EGF (equivalent to complete mPOM), and rGF with AREG ( Figure S6J). Upon 3D culture of DVL3 in mPOM, cells upregulated AREG and LY6D, further supporting the association between these genes ( Figure S6K). We then assessed the levels of secreted AREG in CM from LY6D + -enriched DVL3 cells, in the absence or presence of EGF or AREG. Both EGF and AREG further increased secretion of AREG ( Figure S6L). Notably, the levels of secreted AREG were much higher upon AREG stimulation irrespective of androgen stimulation ( Figure S6L). To test whether induced AREG levels were transcriptionally regulated, we assessed their RNA expression levels ( Figure S6M). In keeping with our ELISA data, Areg transcript levels increased by stimulating cells with EGF or AREG, while Egf levels remained unchanged (Figure S6M). IF analysis further confirmed the expression of AREG in LY6D + DVL3 prostate tumor cells ( Figure S6N). Similarly, Egr1 mRNA levels increased upon AREG stimulation in the presence and absence of androgens, while its levels decreased by EGF stimulation in HN conditions ( Figure S6M). Collectively, these data demonstrate that AREG stimulates its own expression, establishing an autonomous, self-amplifying signaling loop involving activated MAPK, irrespective of androgen stimulation.
To determine a functional role for AREG-induced MAPK signaling for cell growth, we treated LY6D-enriched DVL3 cells with MAPK inhibitors and then assessed their organotypic growth in 3D cultures. Blockage of MAPK signaling significantly inhibited organotypic outgrowth of DVL3 cells (Figures 6B, 6C, and S6O). Notably, under androgen-deprived conditions, inhibition of EGFR signaling further compromised outgrowth of LY6Denriched DVL3 cells (Figures 6B, 6C, and S6O). To evaluate whether the effects of MAPK inhibition extend to intrinsically CR primary tumor cells, we analyzed outgrowth of organoids in isolated, PTEN-deficient YFP + LY6D + and YFP + LY6Dtumor cells. In keeping with our cell-line data, blockade of MAPK signaling with trametinib almost completely abrogated cell growth of both LY6D + and LY6D À primary prostate cells, but these effects were significantly more pronounced in LY6D + tumor cells due to their intrinsic growth advantage in the absence of androgen (Figures 6D-6F). These data provide direct evidence Article ll that EGFR signaling is augmented by autocrine AREG secretion, regulating cell growth of CR LY6D + prostate tumor cells.
To further assess the relevance of MAPK signaling for the propagation of LY6D + cancer cells, we treated tumor-bearing PTEN-deficient K8PY mice for 4 weeks with trametinib. Mice were castrated 1 week prior to trametinib treatment to test whether MAPK inhibition is therapeutically meaningful in addition to androgen deprivation ( Figure 6G). Assessment of LY6D + cells in K8PY mice showed that MAPK inhibition significantly reduced the number of LY6D + cells in castrated animals ( Figures 6H and  6I), associated with a marked reduction of AREG and pERK ( Figures 6H and 6J-6L), indicating the presence of an autocrine stimulation of LY6D + cells in vivo, which depends on MAPK signaling. Altogether these data provide evidence for the presence of an AREG-dependent autocrine feedback loop, required for the survival of LY6D + cells.
Human LY6D + prostate tumor cells are present post androgen-deprived therapy Our PCa preclinical data suggested that LY6D + luminal cells contribute to androgen resistance in human PCa. Accordingly, we previously demonstrated the presence of LY6D + cells in human normal and tumor prostate tissues, 6 but their abundance in malignant tissues so far remained unknown. We speculated that the high level of intra-and interpatient heterogeneity of PCa 1,3,4 could possibly affect the abundancy of LY6D + cells in PCa tumor samples. To account for this, we built a tissue microarray (TMA) capturing the multifocality of prostate tumors ( Figures S7A and S7B; Table S2). Multiplex IF analyses identified expression of LY6D in prostate epithelial cells marked by cytokeratin (CK) and prostate-specific membrane antigen (PSMA) in most core samples ( Figures 7A and S7A). Notably, the expression pattern of LY6D in the normal tissue was sparse and focal ( Figure 7A, and 6 ). Importantly, we observed a strong correlation between LY6D and AREG staining ( Figure 7B). We next extended our analyses to lymph node metastasis derived from patients undergoing extended prostatectomy and pelvic lymphadenectomy. In nine out of 10 patient samples, LY6D + prostate cells were also present in the lymph node metastasis with variations in their frequencies compared with the primary tissue ( Figures 7B-7D). Notably, since these patients were HN, no evolutionary pressure existed for the selection of CR cells. Further analyses of LY6D expression and Gleason scores of the respected tissue area indicated a correlation between LY6D and higher Gleason scores ( Figure 7D), supporting the hypothesis that LY6D + cells significantly contribute to PCa progression.
Based on our previous data and the autocrine survival loop of LY6D + luminal cells described here, we hypothesized that castration of PCa patients further selects for survival of intrinsically CR, LY6D + luminal cells. To test this, we obtained matching longitudinal biopsies from three PCa patients at the time of their diagnosis and after androgen deprivation therpay (ADT) ( Table S3). Similar to our results from treatment-naive tissues from prostatectomies ( Figures 7A-7C), we identified LY6D + cells in all biopsies ( Figures 7E and S7C). Importantly, 6 months after ADT, LY6D + tumor cells were the dominant cell population in two out of three samples, indicative of their in vivo selection by ADT (Figures 7F and S7E).
In conclusion, our data demonstrate that a substantial fraction of tumor cells in HN patients express LY6D, and their frequency correlates with higher Gleason grades. Furthermore, the selective pressure imposed by ADT favors the outgrowth of LY6D + cells in vivo and in keeping with our preclinical data.

DISCUSSION
Resistance to androgen deprivation remains a clinical challenge for patients with PCa and is the main cause for death of patients with advanced PCa. We have previously defined that LY6D marks a population of intrinsically CR luminal stem/progenitor cells present in normal, non-malignant tissues. We here identified an autocrine signaling loop, present in LY6D + luminal progenitor cells, which provides survival signals through AREGmediated MAPK signaling. The presence of different subsets of prostate progenitor cells in the adult prostate, their spatial distribution (proximal versus distal), and their contribution to human PCa remain controversial. Recent studies indeed suggest a differential function of co-existent luminal progenitors. 8,9,11,12,24,[52][53][54] Notably, the frequency of distal progenitors increases with age, 52 as does the content of somatic mutations in the peripheral zone, 55 suggesting a role also of distal environment in tumor initiation. From early stages of development 12 to adulthood, we observed LY6D + progenitor cells scattered in both proximal and distal regions. Notably, in adult male mice (>12 weeks old) the intrinsically CR LY6D + luminal population was preferentially located in distal regions. Interestingly, some studies have reported high Ly6d expression only in the proximal and periurethral regions of the prostate, 8,11,16 which may in part be due to the higher number of LY6D + progenitors in the proximal area in young mice (1-3 months). We here describe the presence of LY6D + prostate cells in the human androgen-ablated prostate epithelium, further validating our findings in our luminal PCa mouse model. This finding strongly supports the idea that preferentially distal LY6D + luminal cells initiate and drive aggressive prostate tumors.
Our RNA-seq analysis to define the underlying mechanisms of CR revealed that CR LY6D + tumor cells maintain the expression of Ar; however, downstream signaling was ablated. In contrast, the EGFR-MAPK pathway was uniquely upregulated in LY6D + luminal progenitors in the absence of androgen, supporting the hypothesis that MAPK activation is an intrinsic mechanism of resistance to androgen deprivation. Our transcriptomic data provide evidence that LY6D + luminal PCa progenitors express high levels of AREG and simultaneously downregulate Egf. AREG secretion by LY6D + luminal cells leads to an autocrine stimulation, promoting cell growth and survival in the absence of androgen and adding to the complexity of Erbb ligand dynamics for the regulation of adult stem and progenitor cells. 56 Similar to the role of AREG in regulating LY6D + luminal cells described here, AREG is essential for the development of the mammary gland through an interaction with estrogen receptor (ER) signaling 57 and progression of ovarian cancer. 58 In keeping with our data, ERBB receptors and their ligands have previously been proposed to compensate for the absence of androgen signaling in PCa. 59,60 During PCa tumorigenesis, increased expression of AREG has been observed with disease progression. 61 Furthermore, an upregulation of AREG was observed following androgen deprivation in the PCa xenografts, 62 suggesting an increased presence of CR cells in the developing PCa. Our finding that the autocrine AREG stimulatory loop, present in LY6D + tumor cells, is regulated by the EGR1 TF is consistent with previous studies showing that aberrant expression of AREG in keratinocytes and breast cancer cells can stimulate EGFR and trigger the activation of ERK and EGR1. 63,64 Similarly, the upregulation of EGR1 has been associated with increased Gleason grade and progression to metastasis. 65,66 Notably, EGR1 expression can be modulated by mutated TP53, adding further complexity to its role in cancer cells. 67 While these studies clearly support the idea of AREG as a driver of castration resistance and that the AREG-EGR1 axis may be part of a more complex mechanism, it hitherto remained unknown which subset of prostate cells predominantly contribute to this. Our data demonstrating that LY6D + cells express high levels of AREG provide further details on the source of AREG in PCa and indicate a therapeutic angle to interfere with the expansion of AREG + cells.
Multiple efforts to characterize the biology associated with aggressive PCa have aimed at developing targeted therapies. Notably, the clinical and molecular heterogeneity of PCa and the lack of recurrent driver mutations in a substantial proportion of patients are obvious hurdles for stratified, randomized clinical trials, exemplified by a recently published phase III trial combining an AKT inhibitor and ADT: while preclinical and early clinical data suggested that this combination is active in PTENdeficient PCa, 68 the placebo-controlled phase III trial did not observe a benefit for the intention-to-treat population. 69 Our results support an alternative approach by using MAPK inhibitors in combination with anti-androgen therapy to prevent the emergence of CR PCa (CRPC). Targeting of the AREG/EGFR axis could be a meaningful way to eradicate intrinsically CR cells, driving disease progression. While directly antagonizing AREG seems more difficult due to its secretion, interference with its activated signaling pathway is possible through the use of EGFR and MAPK inhibitors. 70 In line with this hypothesis, EGFR has previously been identified as a potential target for the treatment of PCa. 71,72 However, small-molecule tyrosine kinase inhibitors, including erlotinib, lapatinib, and gefitinib, have shown limited effectiveness in treating CRPC as a single agent or in combination with chemotherapy. [73][74][75][76][77][78][79] Again, these poor responses likely reflect the molecular complexity of advanced metastatic CRPC. Importantly, patients entering these clinical trials were not stratified based on the expression levels of EGFR or receptor ligands, which could have affected the overall efficacy of such therapies. More encouraging data have recently emerged from a phase I trial in refractory, metastatic CRPC patients treated with trametinib, demonstrating improved overall survival. 80 Additional phase II trials are underway testing the efficacy of trametinib for the treatment of advanced PCa (NCT01990196, NCT02881242). Our in vivo data indicate that inhibiting MAPK signaling synergizes with androgen deprivation by interrupting an AREG-driven autocrine feedback loop and suggest that early treatment with MAPK inhibitors may substantially delay or even suppress the emergence of CRPC. Furthermore, our data indicate that intrinsically CR cells can be detected in biopsies by staining for LY6D, long before the initiation of androgen deprivation therapies. This may potentially allow for a preemptive treatment before the emergence of a clinically apparent disease resistant to androgen deprivation. For a clinical trial, LY6D expression may indeed be a meaningful biomarker for the clinical development of a combination therapy.
Together, our study highlights a key mechanism underlying intrinsic castration resistance in prostate cells. Our findings provide a rationale to target pre-existent CR localized tumors before the emergence of resistant disease by combining androgendeprivation therapies with emerging approaches to eliminate LY6D + cells, either with anti-EGFR therapy or the potential use of monoclonal antibodies against LY6D. Clinical trials are warranted to address whether these findings can be translated into improved patient care.

Limitations of the study
We here report an inherent mechanism of resistance to androgen deprivation of LY6D luminal prostate cells using an in vivo mouse tumor-tracing model. While our findings demonstrate activation of the MAPK-signaling pathway by an autocrine AREG-mediated feedback loop in LY6D + luminal tumor cells as the key mechanism underlying cell survival, other mechanisms sustaining CR LY6D + tumor cells cannot be ruled out. Results from our mouse model experiments translate to patients as LY6D expression is associated with higher Gleason scores and poorer clinical outcome. However, the area biopsied may limit the identification of LY6D-resistant cells in PCa patients due to the high heterogeneity underlying PCa specimens. In addition, we have used a limited number of patients in our analyses, and therefore bigger cohorts of treatment-naive metastatic patients need to be analyzed to fully assess the contribution of LY6D + cells for advanced PCa.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:   , of which 9 patients had localised disease without the nodal involvement (N0) and 11 patients presented with the pelvic LN invasion (N1) ( Table S2). The H&E stained tissue sections were evaluated by the clinical pathologist Dr Pedro Oliveira, who selected the normal versus tumor tissue for the TMA build. At least 3 cores per tissue type (normal or tumor) per patient were selected. The cores were randomised throughout the 3 blocks. Dr Oliveira performed the Gleason scoring for each tumor area of the TMA cores. ADT-treated cohort Prostate cancer specimens were obtained from the AToM Clinical Study (Table S3) which is running at The Christie Hospital. PCa patients have received a standard diagnostic biopsy (Pre-ADT samples) followed by 6 months of ADT as per routine clinical practice (Post-ADT samples). The tissue specimens from consented patients were preserved in FFPE blocks.

Surgical castration
Male mice of at least 8 weeks of age were used for the surgery. Prior to the surgery, the animals were given analgesia Rymadyl (2 mL/g, in saline) subcutaneously in the scruff and were anesthetized using inhalation agent isolfurane (Teva) at 2.5-3% in 100% Oxygen. Buprecare (Buprenorphine) analgesia solution (0.05 mg/g, in saline) was additionally given subcutaneously in the scruff, before performing the invasive surgical procedures which are invading the body wall. All procedures were performed under surgically sterile conditions. K8Y and K8PY experimental mice were castrated using standard techniques. Briefly, a small midline ventral incision was created in the skin. Epididymis was exposed and the testes, and the testes were removed by cauterisation. The remaining tissue was gently placed back under the skin and the technique was repeated on the other side. Skin incision was closed with the 6/0 absorbable surgical suture (Vicryl) in an interrupted pattern. The induction of Cre-recombinase was achieved by daily tamoxifen injections at least four weeks post-castration.  Table S5. Gene silencing by small interference RNAs (siRNAs) Prostate luminal cancer cells (DVL3) were transfected with 50 nM siRNA (siEgr1 and non-targeting control) from Dharmacon (Table S5) using DharmaFECT 3 solution (Dharmacon, #T-2003-01) according to the manufacturer instructions. Transfected cells were embedded in matrix BME (AMS Biotechnology (Europe) Limited, #3533-010-02) and cultivated in complete mPOM media at 37 C, 5% CO 2 for 7 days.
Data analysis and visualization RNA-seq data visualization Differentially expressed genes were visualized in R by the implementation of heatmaps using the pheatmap package, or volcano plots using the EnhancedVolcano package. 4-way scatterplot was created by the set of functions within the ggplot2/tidyverse package. PCA Principal Component Analysis (PCA) was performed using the PCAtools Bioconductor package (Blighe & Lun, 2019). Data was visualised with the biplot function. Log 2 -transformed normalised counts of the K8PY RNAseq dataset were used for the analysis. GSEA The molecular pathways enrichment was evaluated with the Gene Set Enrichment Analysis (GSEA) software. 92,93 For the analysis, normalised counts of the K8Y and K8PY bulk RNAseq dataset was used. The Hallmark and Gene Ontology (GO) gene set collections were evaluated using the GSEA software.
Evaluation of PC3 and DU145 datasets were obtained from the study by Xu et al., GEO accession: GSE116864. 41 The RNA-seq datasets of the PC3 and DU145 cells treated with the AREG-conditioned media, and their controls were selected for the analysis. Analysis was made in R using the package DESeq2 in order to identify differentially expressed genes. Data was normalised and extracted for highly differentially expressed genes according to the following threshold: padj_cutoff = 0.05 and LogFC_cutoff = 1.5. GSEA was run to evaluate these gene sets in the K8Y and K8PY RNA-seq data.
Additional growth factor associated pathways were assessed using the MSigDB canonical pathways of the curated gene sets (C2). For the analysis we used the gene set signatures from the BioCarta, KEGG and Reactome datasets (Table S7). IPA The Ingenuity Pathway Analysis (IPA, Qiagen) was used to identify the canonical pathways activated in YFP + LY6D + versus YFP + LY6D À prostate cells isolated from K8PY HN or K8PY CR mice. Differentially expressed genes were filtered based on the À1 R logFC R1.

DoRothEA
The transcription factor (TF) activity was inferred from the K8PY gene expression data by the DoRothEA package. 39 Data visualization was done using the pheatmap package. Correlation analysis Spearman correlation analysis was performed on the K8Y and K8PY RNA-seq dataset in R using the packages ggplot2 and ggpubr and function ggscatter. Heatmaps were created with the packages pheatmap package. Genesets used for this study: GSE116864, 41 GSE94574. 37

QUANTIFICATION AND STATISTICAL ANALYSIS
All statistical analyses were performed using GraphPad Prism v9.0.0 or in the R environment v4.0.2 (R Core Team). Data points from the individual technical and biological replicates are represented on the plots together with the means ± s.d. The statistical significance was assessed by t test or ANOVA based on the number of compared groups, two, or >2 groups, respectively. The p value is represented on the plots and each comparison is annotated. The number of independent experiments or biologically replicates are indicated in figure legends.