Somatic Tissue Engineering in Mouse Models Reveals an Actionable Role for WNT Pathway Alterations in Prostate Cancer Metastasis

Prostate cancer is the most frequent cancer in American men and the second leading cause of their cancer-associated death (1). Although localized disease is associated with an excellent prognosis, the 5-year survival rate drops dramatically in patients with metastatic prostate cancer, from nearly 100% to 30%. Androgen deprivation therapy (ADT) has been the therapy of choice for patients with prostate cancer for several decades (2); however, many patients who initially respond acquire resistance to ADT and eventually develop metastatic castration-resistant prostate cancer (mCRPC).

Advanced prostate cancer is extensively characterized on the molecular level. In addition to androgen receptor (AR) amplification and activation of other AR pathway genes that are induced to bypass ADT, recent sequencing studies have identified various “noncanonical drivers” such as loss of TP53, amplification of MYC and MYCN, and alterations in the PI3K, WNT, and/or DNA repair pathways to be enriched in advanced prostate cancer (3–8). Yet, the potential functional role of many of these genetic alterations in driving either castration resistance or metastasis is unclear.

One way to study the functional role of various genetic perturbations in a physiologic context is through the use of genetically engineered mouse models (GEMM). Although these models have been critical for our understanding of prostate cancer biology as well as for preclinical testing of new therapies (9), they lack the flexibility to test the impact of potential genetic drivers in a timely manner. GEMMs of prostate cancer have additional limitations: (i) many common genotypes are not represented; (ii) most prostate-specific promoters used are androgen dependent, making the study of androgen deprivation in these models difficult; (iii) extensive intercrossing is needed to produce the alleles required for lethal disease; (iv) tumor latency is long; (v) metastatic penetrance is often low; and (vi) genes are frequently altered throughout the tissue, leading to multifocal tumors and potential effects of gene mutations on normal tissue that are not reflective of the human scenario.

In an effort to overcome these limitations, we took advantage of in vivo tissue electroporation (10–13) to produce somatic alterations directly in the prostate gland of otherwise wild-type (WT) mice. We envisioned that this approach would produce focal prostate tumors of defined cancer genotypes, enabling the assessment of disease progression and/or therapy response in a physiologic context in both a cost- and time-sensitive manner compared with that required for the production of multiallelic germline strains. After validating the method in comparison with traditional GEMMs, we then used the electroporation-based GEMM (EPO-GEMM) approach to study genetic alterations associated with late-stage prostate cancer and confirmed our results with an orthogonal organoid-based approach. These models identify WNT pathway alterations as actionable events that drive prostate cancer metastasis.

As a first attempt to produce prostate carcinoma in mice using tissue electroporation, we chose to introduce alterations leading to MYC overexpression and PTEN loss, which co-occur in advanced human prostate cancer and have been previously validated as prostate cancer drivers in mice (refs. 14, 15; Supplementary Fig. S1A). To this end, we performed a survival surgery to expose the prostate and delivered a plasmid cocktail containing (i) a transposon vector expressing a human MYC cDNA, (ii) a Sleeping Beauty transposase (SB13), and (iii) a gene-editing vector coexpressing Cas9 and a single-guide RNA (sgRNA) targeting Pten into one anterior lobe of the prostate of C57BL/6 male mice via in vivo electroporation (Fig. 1A). Although introduction of the transposon vector (to overexpress MYC) or the CRISPR/Cas9 vector (to edit Pten) alone did not produce tumors for up to 1 year after electroporation, the combination was highly oncogenic, with lethal tumors arising with 83% penetrance and producing a median overall survival of 88 days (Fig. 1B).

We compared the resulting tumor features with tumors arising in a classic GEMM model harboring MYC overexpression and a conditional Pten allele: Nkx3.1^CreERT2/+;Pten^fl/fl;ARR2/Pbsn-MYC (NP^hiMYC) mice (Zou and colleagues, in preparation; ref. 14). As was noted in the germline model, MYC;sgPten (MPt) EPO-GEMM tumors harbored prostatic intraepithelial neoplasia lesions (Supplementary Fig. S1B) together with regions of well-differentiated adenocarcinoma that expressed high levels of the luminal markers AR and Cytokeratin 8 (CK8) and moderate levels of MYC and the proliferation marker Ki-67 (Fig. 1C; Supplementary Fig. S1C and S1D). Concurrently, many lesions contained adjacent poorly differentiated tumor regions with reduced to absent expression of AR and CK8, and a higher frequency of Ki-67 compared with areas of well-differentiated adenocarcinoma (Fig. 1D; Supplementary Fig. S1C and S1D). Occasionally, poorly differentiated areas also expressed the neuroendocrine (NE) marker synaptophysin (SYP; Fig. 1D). As in the germline model, MPt EPO-GEMM tumors metastasized to lymph nodes, liver, and lungs (Supplementary Fig. S1E). NP^hiMYC classic GEMM mice develop metastasis with higher penetrance than MPt EPO-GEMM mice (>80% vs. 54%), which is likely because of the accelerated rate of disease formation and death in the EPO-GEMM model (median survival >300 vs. 87.5 days; Zou and colleagues, in preparation) as well as differences in the genetic background of the mice. These results validate the EPO-GEMM platform as a fast and feasible system to model lethal, metastatic prostate cancer in mice.

Alterations in the TP53 tumor suppressor are rarely seen in early stages of prostate cancer but are among the most frequently altered genes in advanced disease (4, 16), where they portend a particularly poor prognosis (17). TP53 alterations also frequently co-occur with MYC amplifications in patients with prostate cancer (Supplementary Fig. S2A). We therefore tested whether the combination of MYC overexpression with Trp53 (hereafter simply referred to as p53) disruption could lead to advanced prostate cancer in the EPO-GEMM platform (Supplementary Fig. S2B). Although delivery of the MYC-containing transposon vector or plasmids expressing a p53 sgRNA and Cas9 (to edit p53) alone did not lead to tumor formation up to 1 year after surgery, MYC;sgp53 (MP) mice developed lethal prostate cancer as early as 64 days after electroporation, with 76% penetrance and a median survival of 114 days (Fig. 2A). Macroscopically, primary tumors were identified in the anterior lobe of the prostate, and approximately 64% of the animals showed metastatic spread to the lymph nodes, peritoneum, liver, or lungs (Fig. 2B). Of note, although disseminated tumor cells (DTC) could be detected in the bone marrow using sensitive PCR approaches (Supplementary Fig. S2C), macroscopic bone metastasis was not observed using either histologic or radiographic approaches (data not shown), suggesting that these genetic alterations might primarily lead to metastasis formation in soft tissues.

We next characterized the histologic, biological, and molecular features of MP prostate tumors. Both primary tumors and distant metastases consisted of poorly differentiated prostate cancer with low to absent expression of the luminal markers AR and CK8, basal marker Cytokeratin 5 (CK5), and neuroendocrine marker SYP, and high levels of MYC and Ki-67 expression (Fig. 2C; Supplementary Fig. S2D). As MP tumors were mostly negative for both AR and neuroendocrine markers (Fig. 2C; Supplementary Fig. S2D–S2F), they resemble “double-negative” prostate cancer (DNPC), a subtype that has recently been described in patients with castration-resistant prostate cancer (CRPC) with increasing frequency and is enriched for alterations in TP53 (18). Gene set enrichment analysis (GSEA) following RNA sequencing (RNA-seq) revealed that, despite some heterogeneity in AR activity, the majority of MP tumors displayed low expression of AR and neuroendocrine transcriptional signatures, consistent with DNPC (refs. 19, 20; Supplementary Fig. S2G). In line with these findings, histologic characterization of a human CRPC patient tumor harboring MYC and TP53 alterations revealed heterogeneous areas that had high expression of the luminal markers AR and CK8 or the neuroendocrine marker SYP, as well as others that showed low to absent expression of both AR and neuroendocrine markers (i.e., DNPC; Supplementary Fig. S2H).

Because the EPO-GEMM approach does not a priori discriminate between cell types within a targeted tissue, we set out to confirm that the tumors we observed were derived from the epithelial compartment. One powerful feature of the method is the ability to introduce somatic alterations into hosts of different genetic backgrounds to rapidly enable improved resolution of the target cell or, in principle, the study of tumor–host interactions. In one series of experiments, a plasmid encoding Cre recombinase was electroporated into the anterior prostate lobe of mice containing a Lox-Stop-Lox (LSL)-mKate cassette, and tissues were examined for mKate fluorescence in prostate luminal and basal cells. The results indicated that both epithelial cell types (CK8⁺ luminal and p63⁺ basal cells) were targeted by prostate electroporation (Supplementary Fig. S3A). In a second experiment, a MYC transposon vector and the Sleeping Beauty transposase together with a sgRNA targeting p53 (sgp53) were delivered into the prostate of Probasin (Pb)-Cre4;LSL-Cas9 mice, where only luminal or basal cells expressing PB could undergo CRISPR/Cas9-mediated p53 editing (Supplementary Fig. S3B). Following electroporation, Pb-Cre4;LSL-Cas9 mice yielded MP tumors with similar DNPC histologic features and metastatic patterns as were observed in WT C57BL/6 mice (Supplementary Fig. S3C–S3E), implying that prostate tumors driven by MYC overexpression and p53 loss can originate from the epithelial compartment.

Human DNPC arises from prostate tumors that lose AR expression and therefore no longer display sensitivity to therapies targeting the AR signaling pathway (18, 20). Accordingly, three independent prostate cancer cell lines generated from MP EPO-GEMM tumors showed only a marginal reduction in growth at a 10 μmol/L therapeutic dose of the AR inhibitor enzalutamide as compared with the androgen-dependent Myc-CaP (generated from the MYC^hi mouse model; refs. 21, 22) and human LNCaP cell lines, which showed markedly reduced growth at lower concentrations of the drug (Fig. 2D; Supplementary Fig. S4A). Furthermore, although a modest yet statistically significant survival benefit was observed after surgical castration of a cohort of MP prostate tumor–bearing mice compared with the noncastrated cohort, all tumors continued to progress over a short 1-week observation period (Fig. 2E; Supplementary Fig. S4B and S4C). These data underscore the intratumoral heterogeneity of MP prostate tumors, with a subset of tumor cells remaining androgen-sensitive but the majority becoming resistant to androgen deprivation.

TP53-mutant human cancers often harbor complex genomes with a high rate of copy-number variations (CNV; refs. 23–25), which has been linked to metastasis formation and tumor relapse in prostate cancer (26). Similarly, murine MP tumors displayed a high rate of CNVs compared with MPt prostate tumors as assessed by sparse whole-genome sequencing (Fig. 2F). Among the gains and losses that were observed were alterations linked to late-stage disease, including recurrent amplifications in chromosome 3 (harboring Pik3ca and Sox2) and deletions in chromosome 14 (harboring the tumor suppressor Nkx3.1), as well as focal amplifications of Ar and c-Jun found in single tumors (Fig. 2F; Supplementary Fig. S4D–S4F; refs. 4, 5, 27, 28). Therefore, MP EPO-GEMM prostate tumors mirror the histologic, castration-resistant, genomic instability, and metastatic features found in human prostate cancers with TP53 mutations.

To further characterize the molecular features of MP tumors, we transcriptionally profiled a series of samples obtained from end-stage EPO-GEMM mice. RNA-seq followed by principal component analysis (PCA) revealed that MP samples clustered into two distinct groups when compared with WT normal prostate or MPt prostate tumor samples (Fig. 3A; Supplementary Fig. S5A). GSEA as well as gene ontology pathway analysis revealed that a WNT–β-catenin pathway signature was enriched in one of the MP clusters (Fig. 3B; Supplementary Fig. S5B and S5C). In agreement, unsupervised clustering based on transcriptional expression of known WNT pathway genes bifurcated the MP samples into WNT high (WNT^hi) and WNT low (WNT^lo) groups (Fig. 3C; Supplementary Fig. S5D). Interestingly, WNT pathway activity appeared to correlate with the magnitude of metastatic spread, with WNT^hi tumors displaying a trend toward an overall greater frequency of metastasis, and in particular to the liver (Fig. 3D; Supplementary Fig. S5E), suggesting that deregulation of the WNT pathway may associate with prostate cancer metastasis.

Canonical WNT signaling is triggered by the binding of WNT ligands to WNT receptors such as LRP5/6, which results in the dismantling of the β-catenin destruction complex, leading to the translocation of β-catenin to the nucleus where it associates with the TCF family of transcription factors to activate WNT target genes (29). We therefore stained MP tumor sections for the expression of β-catenin and TCF7 as well as the presence of porcupine (PORCN), an O-Acyltransferase that is required for WNT ligand secretion and activation (30). β-catenin and TCF7 were elevated in WNT^hi MP tumors and associated with high levels of PORCN expression (Fig. 3E; Supplementary Fig. S5F and S5G). In contrast, none of these factors were detectable in WNT^lo MP tumors.

WNT^hi tumors also expressed other molecular features consistent with WNT pathway activation. For example, our CNV analysis indicated that two WNT^hi MP samples harbored focal amplifications of Lrp6, a WNT receptor, and Wnt2b, a WNT ligand, which were associated with significant transcriptional upregulation of these genes (Fig. 3F; Supplementary Fig. S5G–S5I). Another tumor acquired a mutation in Apc, a negative regulator of the WNT pathway, corresponding to an event found (albeit rarely) in human tumors (Fig. 3G). Interestingly, although Apc mRNA expression was reduced in both WNT^hi and WNT^lo MP tumors compared with normal murine prostate tissue, only WNT^hi tumors had induction of Porcn and a number of canonical WNT ligands (Wnt 1, 2a, 2b, 10a, and 10b; Fig. 3C; Supplementary Fig. S5G and S5J). These data suggest that multiple factors contribute to WNT pathway activation in a subset of MP tumors.

The link between WNT pathway activation and the more aggressive and metastatic disease identified in our EPO-GEMMs is in line with data from patients with prostate cancer (20). In a dataset of patients with either localized or advanced disease, APC mutations are most tightly associated with metastatic, non–castration-resistant prostate cancer (5). Further inspection of datasets of human primary [The Cancer Genome Atlas (TCGA); ref. 31] and metastatic [Stand Up To Cancer (SU2C); refs. 3, 4] prostate cancer indicated that focal amplifications of LRP5 and LRP6 (also observed in one of our MP WNT^hi tumors) are associated with high transcriptional LRP5 and LRP6 activity and occurred at a significantly higher rate in metastatic prostate tumors than in localized disease (Fig. 4A and B; Supplementary Fig. S6A and S6B). Although alterations in TP53 as well as MYC are associated with castration-resistant disease, mutations predicted to activate WNT signaling were enriched in patients with metastatic disease independently of castration resistance status (Fig. 4C–H). Overall, patients with prostate tumors harboring WNT pathway alterations showed a significantly higher metastatic frequency and reduced overall survival (Fig. 4I and J; ref. 3). Finally, when comparing primary prostate tumor biopsies from different patient cohorts, β-catenin expression (as a readout of WNT pathway activation) was higher in tumors from patients with metastatic compared with locoregional disease (Fig. 4K and L).

To directly test whether WNT pathway alterations produce more aggressive and metastatic prostate cancers, we took advantage of the flexible EPO-GEMM platform to engineer tumors with constitutive WNT pathway activation. Specifically, we combined the MYC transposon vector with a dual CRISPR vector targeting p53 as well as Apc at codon 892 (creating an N-terminal truncated protein) to generate MPApc EPO-GEMM mice (Supplementary Fig. S7A and S7B). As expected, the resulting tumors displayed evidence of both p53 and Apc alterations, WNT pathway activation as assessed by abundant nuclear β-catenin and TCF7 expression, and all of the features of metastatic DNPC (Fig. 5A; Supplementary Fig. S7C–S7H).

Apc disruption also increased metastatic spread; 100% of MPApc mice developed distant metastases (19/19) as compared with approximately 64% of mice harboring MP prostate tumors (9/14; Fig. 5B and C). Furthermore, MPApc mice displayed a significantly reduced survival compared with MP cohorts (median survival 114 vs. 47 days; Fig. 5D). Moreover, the resulting MPApc tumors contained tumor regions with loss of E-cadherin and gain in Vimentin expression, indicative of an invasive, epithelial-to-mesenchymal transition (EMT)–like phenotype that has been previously associated with prostate cancer metastasis (Fig. 5E; ref. 32, 33). Of note, a second Apc-targeting sgRNA predicted to produce a more central Apc truncation (at codon 1529) that recapitulates the most common Apc mutations found in human prostate cancer (Supplementary Fig. S8A) also activated WNT signaling and produced metastatic disease in 100% of the mice (Supplementary Fig. S8B–S8E). Furthermore, EPO-GEMM models produced with MYC, Pten, and Apc alterations (at codon 892; MPtApc) displayed an increased frequency of metastasis compared with those with MYC and Pten alterations alone (100% vs. 54%; Fig. 5F–I; Supplementary Fig. S8F). Therefore, distinct Apc truncations produce WNT pathway activation and promote metastasis in prostate cancers driven by multiple genetic configurations.

To validate the link between WNT pathway activation and prostate cancer metastasis in a well-defined orthogonal system, we created a series of mouse prostate organoids engineered to contain different genetic alterations. Starting from either sgp53-edited (hereafter referred to as p53^−/−) or Pten^−/− (derived from Pb-Cre; Pten^flox/flox mice) murine organoids, we engineered WNT pathway activation by CRISPR/Cas9-mediated genome editing of the Apc locus (hereafter referred to as Apc^−/−). Interestingly, constitutive WNT pathway activation through Apc disruption (at codon 892) in combination with Pten or p53 deletion led to increased organoid growth as well as changes in organoid morphology in vitro (Supplementary Fig. S9A–S9D).

Orthotopic transplantation of Pten^−/−;Apc^−/− and p53^−/−;Apc^−/− organoids into immunodeficient NOD-scid IL2Rγ^null (NSG) mice led to a higher penetrance of tumor formation and enhanced tumor growth after 15 weeks compared with organoids harboring single-gene alterations (Fig. 6A; Supplementary Fig. S9E). The resulting p53^−/−;Apc^−/− and Pten^−/−;Apc^−/− prostate tumors were histologically similar to MPApc and MPtApc EPO-GEMM tumors, displaying poorly differentiated tumor regions that were low for AR and the luminal markers CK8 and CK18, which was independently confirmed by immunoblotting for AR and the AR targets FKBP5 and NKX3.1 (Fig. 6B; Supplementary Fig. S9F). Phenocopying the results in the EPO-GEMMs, these p53^−/−;Apc^−/− and Pten^−/−;Apc^−/− tumors also displayed an EMT-like phenotype containing regions with loss of E-cadherin and gain in Vimentin expression (Fig. 6B).

Metastatic spread could not be assessed in the orthotopic transplantation model, as mice succumb to tumor-induced urethral obstruction early on during tumor development. However, in a tail vein injection–based assay, prostate organoids harboring an Apc mutation (p53^−/−;Apc^−/− and Pten^−/−;Apc^−/−) produced observable lung metastasis 4 weeks after injection in contrast to their corresponding Pten^−/− or p53^−/− controls (Fig. 6C). Similar results were obtained in p53^−/− prostate organoids harboring a mutation in the central portion of the Apc gene at codon 1405 (Apc¹⁴⁰⁵; Supplementary Fig. S9G–S9I). The increased metastatic capacity of Apc-mutant tumors involved canonical WNT signaling, as short hairpin RNA (shRNA)–mediated Ctnnb1 knockdown in organoids harboring Apc mutations completely ablated metastasis formation (Fig. 6D). Together, these studies validate Apc mutations and WNT pathway activation as a driver of disease progression, invasion, and metastasis in prostate cancer.

Owing to its deregulation in a plethora of other cancers (34, 35), a number of pharmacologic agents have been developed to block different components of the WNT signaling pathway. Small-molecule inhibitors of the PARP family members tankyrase 1 and 2 stabilize Axin1 and increase phosphorylation and degradation of β-catenin, and have yielded promising preclinical results (36–39). Indeed, treatment with the tankyrase inhibitor G007-LK showed markedly decreased growth in all tested prostate cancer cell lines with high WNT activity compared with those without WNT pathway induction, including normal mouse fibroblasts (Fig. 7A). Similar to results reported in a recent study exploring G007-LK efficacy in Apc-mutant colon cancer (39), we found that prostate tumor cells harboring the central Apc¹⁵²⁹ truncating mutation, which retains more of the β-catenin binding domain compared with the shorter N-terminal Apc⁸⁹² mutant, were even more sensitive to the inhibitory effects of G007-LK treatment (Fig. 7A). The drug produced molecular changes consistent with an on-target effect, as short-term G007-LK treatment of MPApc prostate cancer lines led to stabilization of Axin1 and increased phosphorylation of β-catenin; it also decreased WNT activity as measured by a decrease in the mRNA levels of Axin2, a WNT transcriptional target, and reduced TCF7 reporter activity (Supplementary Fig. S10A–S10C). These effects could be partially rescued by overexpression of the constitutively active S45P β-catenin mutant (Supplementary Fig. S10D and S10E).

The EPO-GEMM approach makes it feasible to generate large cohorts of mice that develop relatively synchronous tumors of a defined genotype, thereby providing a powerful system to test new treatment approaches in the preclinical setting. Therefore, we generated a series of MPApc EPO-GEMM mice (with Apc truncations at codon 892) and, upon tumor manifestation as assessed by ultrasound, randomized animals and treated cohorts with vehicle or G007-LK. Tankyrase inhibition produced a significant reduction in primary tumor growth and, importantly, reduced the occurrence of macrometastatic disease (Fig. 7B; Supplementary Fig. S10F). These effects produced a near doubling in life span of G007-LK–treated prostate tumor–bearing mice as compared with those in the vehicle-treated cohort (Fig. 7C). Of note, although the reduction in metastasis observed following G007-LK treatment could arise indirectly from its inhibitory effects on primary tumor growth, G007-LK also prevented metastatic colonization in a tail vein injection assay (Fig. 7D–F).

We also produced prostate tumors in mice following orthotopic transplantation of MPApc⁸⁹² cells transduced with a WNT pathway reporter construct (7TCF-luciferase) and initiated treatment with G007-LK or vehicle control. In both the EPO-GEMM and orthotopic transplantation setting, tankyrase inhibition significantly decreased WNT pathway transcriptional activity in vivo as measured by a reduction in Axin2 mRNA levels and a decrease in 7TCF-luciferase activity following bioluminescence imaging (BLI; Supplementary Fig. S10G and S10H). In addition, none of the G007-LK–treated mice had detectable metastases at the endpoint, whereas half of the mice in the vehicle-treated cohort did (Supplementary Fig. S10I and S10J).

Finally, to rule out the possibility that the antitumor effects of tankyrase inhibition were unique to tumors with engineered WNT pathway mutations, we produced orthotopic tumors from a 7TCF-luciferase–expressing MP WNT^hi cell line and, upon tumor manifestation, mice were treated with vehicle or G007-LK (Fig. 7G). As predicted from the above findings, G007-LK treatment reduced WNT transcriptional activity (Supplementary Fig. S10K and S10L) and reduced primary tumor growth as well as the frequency and number of lung and liver metastases (Fig. 7H and I; Supplementary Fig. S10M–S10Q). Together, these data demonstrate that prostate tumors harboring WNT pathway alterations can acquire a dependency on WNT signaling that can be targeted therapeutically.

In this study we engineered and validated a nongermline GEMM (EPO-GEMM) that is flexible, efficient, and allows for the functional characterization of potential genetic drivers in prostate cancer. Although traditional prostate cancer GEMMs are extraordinarily powerful, they are time-consuming andexpensive, and typically require intercrossing of several germline strains to produce tumors that arise sporadically over long time periods. In contrast, the EPO-GEMM platform allows for the production of synchronized cohorts of mice harboring genetically defined tumors at a much greater speed and scale. For example, the Nkx3.1^CreERT2/+;Pten^fl/fl;ARR2/Pbsn-MYC GEMM requires intercrossing of least four germline alleles, producing only a small number of animals with the correct genotype that develop tumors after a much longer latency than the MPt EPO-GEMM model (319 vs. 87.5 day median survival; Arriaga and colleagues, manuscript under review). Still, as with traditional GEMMs, EPO-GEMM models can recapitulate features of the human disease, and, as shown here, MP tumors show features of “double-negative” castration-resistant prostate cancer.

Beyond the increased speed associated with the approach, EPO-GEMMs have additional advantages compared with the traditional GEMMs. Similar to other somatic engineering approaches (40, 41), the costs of EPO-GEMM models are minimal, requiring only the purchase of as many WT mice as are needed to harbor the desired genotypes. As in human patients, tumors are focal and surrounded by normal tissue. Moreover, the models enable production of mice bearing tumors across a range of different genetic configurations in parallel, thereby facilitating their comparison. The relatively synchronous nature of tumor onset also enables the production of cohorts for preclinical studies. Tumors can also be engineered in any strain of mice, including those with alterations in particular stromal cell components, thereby enabling valuable studies of tumor–host interactions. Finally, the system is extremely portable, and in principle requires only sending plasmids and protocols to other laboratories. Some of the advantages of the EPO-GEMM system overlap with those achieved using the RapidCaP model (42), another strategy for somatic engineering of the prostate, although restrictions on the size and variety of vectors that can be used in that system limit its potential for producing genotypic diversity.

In this study, we took advantage of the unique features of the prostate EPO-GEMM approach to rapidly produce new genotypic configurations suggested from analysis of the original tumors, target different germline strains to establish the epithelial origin of the tumors, and generate synchronized cohorts of mice harboring genetically defined tumors for preclinical studies. Supporting the broad utility of this approach, others have used electroporation to produce cancers in the prostate and other organs (10–13). Other nongermline mouse modeling approaches, for example, organoid transplantation (40, 43, 44), stem cell manipulation/retransplantation (45, 46), and other somatic tissue engineering approaches (47, 48), are continuing to make mouse models more accessible to the broader research community.

Using the EPO-GEMM approach, we identified WNT pathway activation as a potent contributor to aggressive and metastatic prostate cancer. Hence, a subset of MP prostate tumors spontaneously acquired signatures of WNT pathway activation. These tumors showed either genetic alterations or upregulation of genes capable of activating WNT signaling. Data from genomic studies of human prostate cancer identified APC mutations as prominently associated with metastatic prostate cancer, and, consistent with our findings, further analyses linked previously undescribed amplifications of the WNT coreceptors LRP5 and LRP6 to metastatic disease in patients. Leveraging both the EPO-GEMM platform and an orthogonal organoid transplantation approach, we showed that disruption of the Apc tumor suppressor by modeling human-relevant mutations could drive aggressive and metastatic disease, a link that has not previously been confirmed in traditional GEMMs (49, 50). Finally, we demonstrated that a tankyrase inhibitor could repress WNT signaling, reduce metastasis, and improve overall survival.

Whereas a recent study demonstrated that p53 loss directly leads to the secretion of WNT ligands that contribute to breast cancer metastasis by driving systemic inflammation (51), the contribution of p53 loss and WNT pathway activation to prostate cancer progression in our model is distinct; here, p53 loss drives progression to aggressive CRPC and, perhaps by conferring inherent genomic instability, enables the acquisition of genetic and transcriptional alterations in the WNT pathway that converge to drive metastasis. Although we did not identify a specific WNT ligand or receptor necessary for pathway activation, we found that PORCN, which is required for proper WNT ligand processing and secretion, was commonly induced only in WNT-activated tumors, suggesting that an increase in autocrine/paracrine WNT signaling is driving pathway activity. Differences in cell-of-origin may also contribute to these heterogeneous mechanisms of WNT pathway activation, as well as possibly explain why we see WNT activity in some tumors but not others.

Importantly, alterations in Apc are also sufficient to promote metastasis irrespective of the p53 (or MYC) genomic status, because MPtApc EPO-GEMMs and Pten^−/−;Apc^−/− organoid transplant prostate tumors also displayed an increased rate of metastasis. Moreover, Apc-altered prostate organoids displayed enhanced metastatic progression in immunodeficient hosts, indicating that WNT pathway activation can drive metastasis formation in prostate cancer independently of its role in mediating systemic inflammation. Notably, we identified markers associated with EMT, a phenotype which has previously been linked to the acquisition of invasive and stemness characteristics and metastasis formation in prostate cancer (32, 33, 52), in tumors following WNT pathway activation, thus providing a potential mechanism of how WNT pathway activation stimulates metastatic spread. Collectively, our results functionally validate clinical associations between WNT pathway alterations, metastasis, and poor survival in patients with CRPC (ref. 3, 6; Fig. 4).

Our results also demonstrate that WNT pathway activation can confer an actionable vulnerability in the setting of mCRPC. Indeed, tumor cells harboring WNT pathway activation displayed enhanced sensitivity to tankyrase inhibition in vitro and in vivo, and our data support the notion that those tumors harboring central Apc truncations or alterations in WNT ligands/receptors, which are collectively the most common WNT pathway alterations found in human CRPC, will be the most sensitive to its effects. Although we cannot rule out the possibility that some of the antitumor effects produced by G007-LK result from off-target effects on stromal cells, the ability of tankyrase inhibition to impede WNT signaling in tumor cells in vivo, and its selective effects on WNT-altered tumor cells in vitro, support the notion that these effects arise, at least in part, from inhibiting WNT signaling in tumor cells themselves. Beyond this, our observation that LRP5/6 amplifications are common in human metastatic prostate cancer raises the possibility that these tumors will also be sensitive to PORCN inhibition or other upstream WNT pathway antagonists. Although dose-limiting toxicities may preclude sufficient tankyrase inhibition in patients, our results provide proof-of-concept that WNT pathway alterations can produce actionable dependencies in prostate cancer. As therapeutic options are limited in this advanced stage of the disease, inhibition of the WNT pathway might be a valid strategy to treat patients with mCRPC.

LNCaP and Myc-CAP cells were provided by P.A. Watson. Primary murine fibroblasts from C57BL/6 mice were purchased from Cell Biologics, Inc. and grown in complete fibroblast media (M2267). MPt, MP, and MPApc murine prostate cancer cell lines were derived from EPO-GEMM prostate tumors with these genotypes. To generate these cell lines, prostate tumors were minced, digested in DMEM media containing 3 mg/mL Dispase II (Gibco) and 1 mg/mL Collagenase IV (C5138;Sigma) for 1 hour at 37°C, and plated on 10-cm culture dishes coated with 100 μg/mL collagen (PureCol; 5005; Advanced Biomatrix). Primary cultures were passaged at least three times to remove fibroblast contamination. All prostate cancer cell lines were maintained in a humidified incubator at 37°C with 5% CO₂, and grown in RPMI1640 or DMEM supplemented with 10% FBS and 100 IU/mL penicillin/streptomycin. All cell lines used were negative for Mycoplasma.

Enzalutamide (S1250) and G007-LK (S7239) were purchased from Selleck Chemicals for in vitro studies. Drugs for in vitro studies were dissolved in DMSO (vehicle) to yield 10 mmol/L stock solutions and stored at −80°C. For in vitro studies, growth medium with or without drugs was changed every 3 days. For in vivo studies, G007-LK (B5830) was purchased from APExBIO. G007-LK was dissolved in 10% DMSO and then reconstitued in 20% Cremophor EL (Sigma-Aldrich) in saline.

For visualizing WNT pathway activity in vitro and in vivo, MP and MPApc cell lines were engineered to express a 7TCF-Luciferase construct (a gift from Roel Nusse, Stanford University; Addgene plasmid # 24308). Lentiviruses were packaged by cotransfection of Gag-Pol–expressing 293 T cells with expression constructs and envelope vectors (VSV-G) using the calcium phosphate method. Following transduction, cells were selected with 4 μg/mL puromycin for 1 week.

MiRE-based shRNAs targeting Apc, Ctnnb1, and Renilla were cloned into MSCV-based vectors as described previously (53, 54). Retroviruses were packaged by cotransfection of Gag-Pol–expressing 293 T cells with expression constructs and envelope vectors (VSV-G) using polyethylenimine (PEI; Sigma-Aldrich). Following transduction with shRNA retroviral constructs, cell selection was performed with 4 μg/mL puromycin for 1 week. Perturbation of WNT pathway activity following Apc or Ctnnb1 knockdown was confirmed by qRT-PCR or readout of TCF activity through TOPFLASH assays.

Plasmids containing a mutant form of β-catenin (βcat^S45P) were provided by L.E. Dow. To engineer MPApc cell lines to express βcat^S45P, retroviruses were packaged by cotransfection of Gag-Pol–expressing 293T cells with expression constructs and envelope vectors (VSV-G) using PEI. Following transduction, cells were selected with 4 μg/mL puromycin for 1 week.

Mouse prostate organoids were established and cultured as described previously (55). Pten^−/− organoids were established from Pb-Cre; Pten^flox/flox mice. WT or Pten^−/− organoids were transduced with lentiCas9-Blast and the bulk population was selected in blasticidin for 3 days. WT organoids were then transduced with LentiCRISPRv2-sgp53 and bulk selected in puromycin for 3 days to generate p53^−/− organoids. Apc-mutant organoid lines were generated using Retro-sgApc-tdTomato constructs targeting codons 884 and 1405 (provided by T. Han and L.E. Dow) and were bulk-sorted to enrich for transduced cells as described previously (56). Ctnnb1 knockdown was achieved using MiRE-based shRNAs targeting Ctnnb1 as described above.

Organoid growth analysis was carried out as described previously (57). A thousand cells per 50 μL Matrigel dome were seeded in EGF withdrawal medium, and each timepoint consisted of three domes in a 24-well plate. Cell viability was assessed using the CellTiter-Glo Viability Assay (Promega) according to the manufacturer's protocol. All values for each timepoint were normalized to day 1 readings.

All mouse experiments were approved by the Memorial Sloan Kettering Cancer Center (MSKCC; New York, NY) Internal Animal Care and Use Committee. Mice were maintained under specific pathogen-free conditions, and food and water were provided ad libitum. Mice were purchased from Jackson Laboratory. Pb-Cre4 (58) male mice were crossed with LSL-Cas9 female mice to produce Pb-Cre4;LSL-Cas9 male mice for generation of EPO-GEMMs.

Eight- to 12-week old WT C57BL/6 or transgenic Pb-Cre;LSL-Cas9 and Rosa26-CAGs-LSL-RIK (59) male mice were anesthetized with isoflurane and the surgical site (pelvic region) scrubbed with a povidone-iodine scrub (Betadine) and rinsed with 70% alcohol. After opening the peritoneal cavity, the left seminal vesicle was used as a landmark and the left anterior lobe of the prostate was pulled out. Plasmid mix (50 μL; see specifications below) was injected into the left anterior lobe of the prostate using a 27.5 gauge syringe, and tweezer electrodes were tightly placed around the injection bubble. Two pulses of electrical current (60 V) given for 35-millisecond lengths at 500-millisecond intervals were then applied using an in vivo electroporator (Nepa Gene NEPA21 Type II Electroporator). After electroporation, the peritoneal cavity was rinsed with 0.5 mL of prewarmed saline. After the procedure, the peritoneal cavity was sutured and the skin closed with skin staples. The mice were kept at 37°C until they awoke, and postsurgery pain management was done with injections of buprenorphine and/or meloxicam for the 3 following days. Tumor formation was assessed by ultrasound imaging, and mice were sacrificed following early tumor development or at endpoint. Genome editing in EPO-GEMM tumors was confirmed by Sanger sequencing.

To generate EPO-GEMM tumors in C57BL/6 WT mice, the following vectors and concentrations were used: a pT3-MYC transposon vector (5 μg), a Sleeping Beauty transposase (SB13; 1 μg), and/or a pX330 CRISPR/Cas9 vector (20 μg; Addgene #42230) targeting the respective tumor suppressor genes. For generation of tumors in Pb-Cre;LSL-Cas9 mice, a pT3-MYC transposon vector (10 μg; Addgene #92046), pT3-sgp53 transposon vector (20 μg), and SB13 (6 μg) were used. For assessment of tissue recombination in Rosa26-CAGs-LSL-RIK mice, a PGK-Cre vector (10 μg) was used. The Sleeping Beauty transposase (SB13) and the pT3 transposon vector were a generous gift from Dr. Xin Chen (University of California, San Francisco, San Francisco, CA). The pX330 vector was a gift from Feng Zhang of Broad Institute (Addgene plasmid # 42230).

The following sgRNAs were used to target the respective tumor suppressor gene locus:

• Pten: GTTTGTGGTCTGCCAGCTAA

• p53: ACCCTGTCACCGAGACCCC

• Apc⁸⁹²: CAGGAACCTCATCAAAACG

• Apc¹⁵²⁹: CAGTTCAGGAAAACGACAA

• Apc¹⁴⁰⁵: GTTCAGAGTGAGCCATGTAG

To generate the pX330 vector containing two sgRNAs, the vector was opened using the XbaI cloning site and the sgRNA-casette containing the second guide was PCR cloned into the vector using the following primers: XbaI U6 forward: ATGCTTCTAGAGAGGGCCTATTTCCCATGATT and NheI gRNA scaffold reverse: ATGTCGCTAAGCTCTAGCTCTAAAACAAAAAAGC.

High-contrast ultrasound imaging was performed on a Vevo 2100 System with a MS250 13- to 24-MHz Scanhead (VisualSonics) to stage and quantify prostate tumor burden. Tumor volume was analyzed using Vevo LAB software.

BLI was used to track luciferase expression in orthotopically transplanted MPApc or MP WNT^hi prostate cell line tumors expressing a 7TCF-Luciferase reporter, as well as orthotopically and intravenously transplanted organoids expressing a Luciferase reporter. Mice were injected intraperitoneally with luciferin (5 mg/mouse; Gold Technologies) and then imaged on a Xenogen IVIS Spectrum Imager (PerkinElmer) 10–15 minutes later for 60 seconds. Quantification of luciferase signaling was analyzed using Living Image Software (Caliper Life Sciences).

Fifty thousand MPApc or MP WNT^hi prostate tumor cells expressing a 7TCF-Luciferase reporter were resuspended in 25 μL of a 50% Matrigel (BD Biosciences)/50% PBS solution and injected into the right anterior prostate lobe of 8- to 10-week old male C57BL/6 mice using a Hamilton Syringe as described previously (52). BLI was used to assess tumor formation, and mice were subsequently randomized and enrolled into treatment groups. The impact on metastatic burden was assessed after 4 weeks of treatment.

Five hundred thousand MP or MPApc prostate tumor cells were resuspended in 400 μL of PBS and tail vein injected into 8- to 10-week-old Nu/Nu (Nude) male mice.

A LentiLuciferase-Neo construct was transduced into all organoid lines and bulk selected for 3 days in neomycin. 3 × 10⁶ cells per mouse were used for orthotopic injection. Organoids were dissociated into single cells and resuspended in 50% Matrigel and 50% medium before injection into male NSG mice. In vivo luciferase signals were measured once a week on an IVIS spectrum imager. Mouse prostate tissues were collected after 13 weeks for histologic analysis.

Twenty-five thousand dissociated LentiLuciferase-Neo transduced organoid cells were resuspended in 400 μL of PBS and tail vein injected into NSG mice. In vivo luciferase signals were measured once a week on an IVIS spectrum imager. Mouse lung tissues were collected 40 days after injection.

Castration was performed as described previously (60). EPO-GEMM mice were monitored for prostate tumor development by ultrasound, and enrolled and randomized into treatment groups once tumors reached 500 mm³. Ultrasound imaging was repeated every week following castration to assess changes in prostate tumor burden. Upon sacrifice, prostate tumor tissue was allocated for 10% formalin fixation and Optimal cutting temperature frozen blocks.

EPO-GEMM mice were monitored for prostate tumor development by ultrasound, and enrolled and randomized into treatment groups once tumors reached 500 mm³. C57BL/6 mice orthotopically transplanted with MP and MPApc prostate tumors cells expressing a 7TCF-Luciferase reporter were evaluated by BLI to verify tumor development before being randomized into various study cohorts. Nude mice tail vein injected with MPApc prostate tumor cells were randomized and treated with either G007-LK or vehicle control the day before injection to assess metastasis prevention.

Mice were treated with vehicle or G007-LK (30 mg/kg body weight) by intraperitoneal injection for 5 consecutive days followed by 2 days off treatment. Ultrasound and/or BLI were repeated every week during treatment to assess changes in prostate tumor burden. No obvious toxicities were observed in vehicle- or drug-treated animals as assessed by changes in body weight. Upon sacrifice, prostate tumor tissue was allocated for 10% formalin fixation and snap-frozen tissue for DNA/RNA analysis.

The presence of peritoneal, lymph node, thorax, lung, and liver metastases was determined at survival or experimental endpoint by gross examination under a dissecting scope. Metastasis burden and the total number of individual metastases was further quantified from hematoxylin and eosin (H&E)–stained sections. The presence of DTCs in the bone marrow of EPO-GEMM mice was assessed following PCR genotyping for the presence of the human MYC allele in the bone marrow flushes from the hind limbs of these mice. PCR genotyping of MP EPO-GEMM prostate tumors and normal WT prostate tissue was used as positive and negative controls, respectively.

Tissues were fixed overnight in 10% formalin, embedded in paraffin, and cut into 5-μm sections. H&E and IHC/immunofluorescence staining was performed using standard protocols. The following primary antibodies were used: AR (Sc-816), p63 (Sc-8431), and CK8 (Sc-8020; Santa Cruz Biotechnology); Porcupine (PORCN; AB105543), MYC (AB32072), Ki-67 (AB16667), and LRP6 (AB24386; Abcam); Cytokeratin 5 (CK5; 905501) and Cytokeratin 8 (CK8; 904801; BioLegend); β-catenin (BD610153), E-cadherin (BD610181), and ASCL1 (MASH1;BD556604; BD Biosciences); Synaptophysin (SYP; 1485-1; Epitomics); mKate2 (AB233; Evrogen); p63 (4A4, Ventana); Vimentin (5741), and TCF1/TCF7 (2203; Cell Signaling Technology). Histopathologic features in EPO-GEMM primary prostate tumors and metastases were assessed by a trained veterinary pathologist (J.E. Wilkinson).

For RNA-seq analysis of the transcriptional profiles of MPt and MP EPO-GEMM prostate tumors, as well as normal anterior lobe tissue from prostates of WT C57BL/6, total RNA was extracted from bulk tissue using the RNeasy Mini Kit (Qiagen). Purified polyA mRNA was subsequently fragmented, and first- and second-strand cDNA synthesis was performed using standard Illumina mRNA TruSeq library preparation protocols. Double-stranded cDNA was subsequently processed for TruSeq dual-index Illumina library generation. For sequencing, pooled multiplexed libraries were run on a HiSeq 2500 machine on RAPID mode. Approximately 10 × 10⁶ 76-bp single-end reads were retrieved per replicate condition. Resulting RNA-seq data were analyzed by removing adaptor sequences using Trimmomatic (61), aligning sequencing data to GRCm38.91(mm10) with STAR (62), and genome wide transcript counting using featureCounts (63) to generate a TPM matrix of transcript counts. Genes were identified as differentially expressed using R package DESeq2 with a cutoff of absolute log₂FoldChange ≥ 1 and P_adj < 0.05 between experimental conditions (64). Functional enrichments of these differentially expressed genes were performed with enrichment analysis tool Enrichr (65) and the retrieved combined score [log(P value) × z-score) was displayed.

PCA was performed using the DESeq2 package in R. Gene expressions of RNA-seq data were clustered using hierarchical clustering based on one minus Pearson correlation test. For pathway enrichment analysis, the weighted GSEA Preranked mode was used on a set of curated signatures in the molecular signatures database (MSigDB v7.0; http://www.broadinstitute.org/gsea/msigdb/index.jsp). From 22,596 signatures, only signatures with 15–500 genes were considered for further analyses. From the results, enriched signatures with a P_adj value less than 0.05 were considered as statistically significant.

CNVs were inferred from sparse whole-genome sequencing data as described previously (66, 67). In brief, 1 μg of bulk genomic DNA was extracted from prostate tumors and tissue using the DNeasy Blood and Tissue Kit (Qiagen) and sonicated using the Covaris instrument. Sonicated DNA was subsequently end-repaired/A-tailed, followed by ligation of TruSeq dual indexed adaptors. Indexed libraries were enriched via PCR and sequenced in multiplex fashion using the Illumina HiSeq2500 Instrument to achieve roughly 1 × 10⁶ uniquely mappable reads per sample, a read count sufficient to allow copy-number inference to a resolution of approximately 400 kb. For data analysis, uniquely mapped reads were counted in genomic bins corrected for mappability. Read counts were subsequently corrected for guanine cytosine content, normalized, and segmented using circular binary segmentation. Segmented copy-number calls are illustrated as relative gains and losses to the median copy number of the entire genome. Broad events (chromosome wide and several megabase sized events) are discernible in a genome-wide manner as illustrated in Fig. 2F. Focal events, namely chromosomal amplifications, are discernible in zoom-in views of chromosomes as depicted in Fig. 3F; Supplementary Fig. S4D and S4E.

Tumors were profiled for genomic alterations in M-IMPACT_v1key cancer-associated genes using our custom, deep sequencing MSK-IMPACT assay that surveys 468 known cancer driver genes. Custom DNA probes were designed for targeted sequencing of all exons and selected introns of oncogenes, tumor suppressor genes, and members of pathways deemed actionable by targeted therapies. Genomic DNA from tumor and matched normal WT prostate anterior lobe tissue samples were subjected to sequence library preparation and exon capture (NimbleGen). Up to 30 barcoded sequence libraries were pooled at equimolar concentrations and input into a single exon capture reaction, as described previously (68). Pooled libraries containing captured DNA fragments were subsequently sequenced on the Illumina HiSeq System.

Sequence data were demultiplexed using BCL2FASTQv1.8.3 (Illumina), and vesitigial adapter sequences were removed from the 3′ end of sequence reads. Reads were aligned in paired-end mode to the hg19 b37 version of the genome using BWA-MEM (Burrows-Wheeler Alignment tool). Local realignment and quality score recalibration were performed using Genome Analysis Toolkit (GATK) according to GATK best practices (69). Paired-sample variant calling was performed on tumor samples and their respective matched normals to identify point mutations/single-nucleotide variants (SNV) and small insertions/deletion (indel). MuTect (version 1.1.4; ref. 70) was used for SNV calling and SomaticIndelDetector, a tool in GATKv.2.3.9, was used for detecting indel events. Variants were subsequently annotated using Annovar, and annotations relative to the canonical transcript for each gene (derived from a list of known canonical transcripts obtained from the UCSC genome browser) were reported.

Tissue microarrays (purchased from US Biolab) containing a total of 126 prostate tumor specimens from 66 patients with localized and metastatic disease were stained for β-catenin expression by immunofluorescence through the Molecular Cytology Core Facility at MSKCC using a Discovery XT Processor (Ventana Medical Systems). Briefly, tissue sections were deparaffinized with EZPrep Buffer (Ventana Medical Systems) and antigen retrieval was performed with CC1 Buffer (Ventana Medical Systems). Sections were blocked for 30 minutes with Background Buster Solution (Innovex), followed by avidin-biotin blocking for 8 minutes (Ventana Medical Systems). Sections were incubated with a β-catenin antibody (8814; Cell Signaling Technology) for 5 hours, followed by a 60-minuteincubation with biotinylated goat anti-rabbit IgG (PK6101; Vector Laboratories) at a 1:200 dilution. Detection was performed with Streptavidin-HRP D (part of DABMap Kit, Ventana Medical Systems), followed by incubation with Tyramide Alexa 488 (B40953; Invitrogen) prepared according to the manufacturer's instructions. After staining, slides were counterstained with DAPI (D9542; Sigma Aldrich) for 10 minutes and cover slipped with Mowiol. Tissues were then scored on a 0–3 scale for β-catenin expression, with scores of 0 and 1 as “negative” and 2 and 3 as “positive” for β-catenin.

CBioPortal.org was used to plot the frequency of mutations, amplifications, and/or deletions in genes of interest in patients with prostate cancer from various datasets. TP53 alterations included deep deletions (homozygous loss) as well as missense, inframe, and truncating mutations. A Kaplan–Meier survival curve of patients with prostate cancer with or without WNT pathway alterations was generated using part of the SU2C dataset (3), which included 47 patients in the WNT-activated group and 81 patients in the non–WNT activated group. Patients were randomized into the two groups based on WNT pathway activating alterations in the following genes: CTNNB1, APC, AXIN2, WIF1, SFRP1, DKK1, RNF43, ZNRF3, GSK3B, TCF7, TLE1, LRP5, LRP6, and WNT2B (71). The percentage of WNT pathway–altered prostate tumor specimens from patients with locoregional versus metastatic disease was determined from an MSK-IMPACT dataset (5), which included 194 patients with locoregional prostate cancer, 135 patients with metastatic prostate cancer, and 147 patients with mCRPC. Locoregional disease in this setting indicated disease without distant clinical or pathologic spread, including lymph node involvement in the pelvis only. LRP5 and LRP6 amplification frequency was determined from a dataset containing samples obtained from primary tumors where CNV analysis was performed by Affymetrix SNP 6.0 (31), or two datasets containing samples obtained from metastatic sites where CNV analysis was performed by whole-exome sequencing (3, 4). LRP5 and LRP6 expression levels in amplified (AMP or GAIN) or nonamplified tumors were determined in mCRPC patients samples from the SU2C dataset (3) using normalized fpkm values and CNV calls.

Histopathologic analysis was performed on a primary prostate tumor tissue biopsy from a patient with mCRPC treated at MSKCC harboring a MYC amplification and p53 alteration (L114Ffs*33) as part of the MSK-IMPACT cohort (5, 72). Clinical sequencing analysis (MSK-IMPACT) was completed on this and other samples and collected using a web-based electronic data capture. IHC and sequencing analysis on human tissue samples was performed under MSKCC Institutional Review Board approval. All samples and clinical data were deidentified.

We adhered to the AR⁺/neuroendocrine-positive (NE⁺)/DN prostate cancer subtype classification as proposed in ref. 18. Briefly, AR and neuroendocrine scores were calculated according to the expression of the mRNA z-scores of 10 AR activity genes (KLK3, KLK2, TMPRSS2, FKBP5, NKX3-1, PLPP1, PMEPA1, PART1, ALDH1A3, and STEAP4) and 10 neuroendocrine signature genes (SYP, CHGA, CHGB, ENO2, CHRNB2, SCG3, SCN3A, PCSK1, ELAVL4, and NKX2-1) for mouse and human prostate samples (19). Subsequently, samples for each dataset were normalized from 1 (highest expression of either neuroendocrine or AR score, respectively) to 0 (lowest expression of either neuroendocrine or AR score, respectively) as displayed in the scattered plot. IHC staining and quantification of AR and SYP/ASCL1 (neuroendocrine) marker expression was also used for subtype classification in some mouse and human tumors. DNPCs were defined as those that lacked expression of both AR⁺ and NE⁺ markers.

Ten thousand cells were plated in 100 μL of media containing 10% FBS per well of a black-walled 96-well plate (Perkin Elmer). After 24 hours, cells were transfected using PEI with 170 ng of TOPFLASH Firefly reporter and 30 ng of pRL-SV40P Renilla constructs provided by T. Tamella (MSKCC). In initial experiments, the WNT-insensitive FOPFLASH Firefly reporter (also provided by T. Tamella) was used to rule out signal background (not shown). Thirty-six hours after transfection, Firefly and Renilla signals were detected using Dual-Glo Luciferase Detection Reagents (Promega) according to the manufacturer's instructions. A Varioskan Flash Plate Reader (Thermo Fischer Scientific) was used to detect luminescence. To control for transfection efficiency, Firefly luciferase levels were normalized to Renilla luciferase levels to generate the measure of relative 7TCF activity.

Cell lysis was performed using RIPA Buffer (Cell signaling Technology) supplemented with phosphatase inhibitors (5 mmol/L sodium fluoride, 1 mmol/L sodium orthovanadate, 1 mmol/L sodium pyrophosphate, and 1 mmol/L β-glycerophosphate) and protease inhibitors (Protease Inhibitor Cocktail Tablets, Roche). Protein concentration was determined using a Bradford Protein Assay Kit (Bio-Rad). Proteins were separated by SDS-PAGE and transferred to polyvinyl difluoride membranes (Millipore) according to the standard protocols. Membranes were immunoblotted with antibodies against Axin1 (2087), phospho-β-catenin S33/S37/T41 (9561), PTEN (9188), P53 (2524), and FKBP5 (12210) from Cell Signaling Technology, AR (ab108341), cyclophilin B (ab16045), and NKX3.1 (ab196020) from Abcam, APC (OP44) from Millipore, and p21 (sc-6246) from Santa Cruz Biotechnology in 5% BSA in TBS-blocking buffer. After primary antibody incubation, membranes were probed with an electrochemiluminescence anti-rabbit IgG or anti-mouse IgG secondary antibody (1:10,000) from GE Healthcare Life Science and imaged using a FluorChem M system (Protein Simple). Protein loading was measured using a monoclonal β-actin antibody directly conjugated to horseradish peroxidase (1:20,000) from Sigma-Aldrich and imaged as above.

Total RNA was isolated using the RNeasy Mini Kit (Qiagen), and cDNA was obtained using the TaqMan Reverse Transcription Reagents (Applied Biosystems). Real-time PCR was performed in triplicate using SYBR Green PCR Master Mix (Applied Biosystems) on the ViiA 7 Real-Time PCR System (Invitrogen). GAPDH and mRn18s served as endogenous normalization controls.

Five thousand cells were plated in 100 μL of media containing 10% FBS per well of a black-walled 96-well plate (Perkin Elmer). The next day, the media were changed and cells were treated with G007-LK or enzalutamide for 72 hours. Following treatment, cell viability was assessed using the CellTiter-Glo Viability Assay (Promega) according to the manufacturer's protocol. IC₅₀ calculations were made using Prism 6 Software (GraphPad Software).

Statistical analyses were performed as described in the figure legend for each experiment. Group size was determined on the basis of the results of preliminary experiments, and no statistical method was used to predetermine sample size. The indicated sample size (n) represents biological replicates. Group allocation and outcome assessment were not performed in a blinded manner. All samples that met proper experimental conditions were included in the analysis. Survival was measured using the Kaplan–Meier method. Statistical significance was determined by one- and two-way ANOVA, Fisher exact test, Student t test, log-rank test, Mann–Whitney test, and Pearson correlation using Prism 6 Software (GraphPad Software) as indicated. Significance was set at P < 0.05.

RNA-seq data generated in this study are deposited in the Gene Expression Omnibus database under accession number GSE139340. Mouse IMPACT sequencing data presented in this study are deposited in the NCBI BioProject database under accession number PRJNA610252.

Figures were prepared using BioRender.com for scientific illustrations and Illustrator CC 2020 (Adobe).

L.E. Dow is a scientific advisor at Mirimus Inc. C.L Sawyers is member of the board of directors at Novartis and a member of the scientific advisory board at ORIC, Arsenal, Column Group, Nextech, Agios, Beigene, Blueprint, Housey, KSQ, PMV, Petra, and Foghorn, has ownership interest in enzalutamide royalties and apalutamide royalties, and is a Cold Spring Harbor Laboratories Science Trustee. S.W. Lowe is a founder and SAB member at Oric Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Leibold, M. Ruscetti, Z. Cao, C.L. Sawyers, S.W. Lowe

Development of methodology: J. Leibold, M. Ruscetti, Z. Cao, F.J. Sánchez-Rivera, P.A. Watson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Leibold, M. Ruscetti, Z. Cao, T. Baslan, M. Zou, W. Abida, T. Han, K.M. Tsanov, L. Zamechek, A. Kulick, C. Amor, S. Tian, K. Rybczyk, N.R. Salgado, E. de Stanchina, J.E. Wilkinson, C. Abate-Shen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Leibold, M. Ruscetti, Z. Cao, Y.-J. Ho, T. Baslan, W. Abida, J. Feucht, K.M. Tsanov, A. Kulick, C. Amor, K. Rybczyk, J.E. Wilkinson, L.E. Dow, C. Abate-Shen

Writing, review, and/or revision of the manuscript: J. Leibold, M. Ruscetti, Z. Cao, Y.-J. Ho, W. Abida, F.M. Barriga, N.R. Salgado, L.E. Dow, C. Abate-Shen, C.L. Sawyers, S.W. Lowe

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Leibold, M. Ruscetti, Z. Cao, L. Zamechek, K. Rybczyk

Study supervision: L.E. Dow, C. Abate-Shen, C.L. Sawyers, S.W. Lowe

We thank T. Tammela and O. Grbovic-Huezo for sharing tissue samples; E.M. Schatoff for sharing reagents, plasmids, and experimental protocols; N. Socci and the Bioinformatics Core for assistance with Mouse IMPACT sequencing analysis; F. Sanchez-Vega and N. Schultz for assistance with genomic analysis of human datasets; A. Tehuitzil, A. Wuest, A. Kim, C.J. Hagen, D. Choi, A.C. Agudelo Rivera, and A. Gaglio for technical assistance; C.J. Sherr and H. Wu for helpful comments on the manuscript; and other members of the Lowe laboratory for insightful discussions. We thank William H. and Alice Goodwin and the Commonwealth Foundation for Cancer Research for research support. This work was also supported by a P50 CA092629 Prostate Cancer Spore grant (to S.W. Lowe), a Memorial Sloan Kettering Cancer Center Support grant (P30 CA008748), NIH grants (R01 CA155169, CA193837, CA224079, CA092629, and CA160001 to C.L. Sawyers; CA233944 and CA087497 to S.W. Lowe; U54 OD020355 to S.W. Lowe and E. de Stanchina; and R01 CA183929 and CA173481 to C. Abate-Shen), Starr Cancer Consortium grants I10-0062 and I12-0007 (to C.L. Sawyers), and an Agilent Thought Leader Award (to S.W. Lowe). S.W. Lowe is the Geoffrey Beene Chair for Cancer Biology and S.W. Lowe and C.L. Sawyers are investigators at the Howard Hughes Medical Institute (HHMI). We also acknowledge support to J. Leibold from the German Research Foundation (DFG) and the Shulamit Katzman Endowed Postdoctoral Research Fellowship, to M. Ruscetti from the American Cancer Society (PF-16-115-01-TBG) and the NCI (K99 CA241110), to T. Baslan from the William C. and Joyce C. O'Neil Charitable Trust and Memorial Sloan Kettering Single Cell Sequencing Initiative, to J. Feucht from the Care-for-Rare Foundation and the German Research Foundation (DFG), to C. Amor from La Caixa foundation and a Harold E. Varmus graduate student fellowship from the Gerstner Sloan Kettering graduate school, to F.M. Barriga from a GMTEC Postdoctoral Research Fellowship and a MSKCC TROT Postdoctoral Fellowship (NIH T32CA160001), to K.M. Tsanov from The Jane Coffin Childs Memorial Fund for Medical Research, and to F.J. Sánchez-Rivera from HHMI (Hanna Gray Fellowship).