Combined MYC Activation and Pten Loss Are Sufficient to Create Genomic Instability and Lethal Metastatic Prostate Cancer

Prostate cancer represents a growing problem as populations across the globe are aging and its incidence is linked tightly to age. The lifetime risk of being diagnosed with prostate cancer in the United States is approximately 1 in 6, yet death occurs in approximately 1 in 33 men. Thus, most prostate cancers are not life-threatening. Lethality most often results from metastasis of castration-resistant disease that spreads to lymph nodes, liver, lungs, and bone. Lethal prostate carcinomas tend to have large numbers of genomic alterations, particularly copy-number/structural alterations (1–4), consistent with the hypothesis that widespread genomic instability may be required for lethal disease development (5, 6). The mechanism by which genetic instability drives disease progression is likely due to the development of intratumoral heterogeneity of genetic changes (most of which, in prostate cancer, are copy number changes), which can drive disease progression as a result of an evolutionary process in which variant subclones are selected that have aggressive features (e.g., the ability to form metastatic deposits and to become castration resistant; ref. 7). Therefore, intensive efforts are focused on determining which somatic genome alterations lead to the development of genetic instability, metastasis and castration resistance. The most common genetic alterations include TMPRSS2-ERG gene fusions as well as large-scale copy number changes, with recurrent point mutations and small insertions and deletions occurring less frequently (1–3, 8–11). Frequent copy number alterations (CNA) include deletions on chromosome 8p involving NKX3-1, amplifications of chromosome 8q24 involving MYC, and deletions of PTEN on chromosome 10q23. Gain of MYC at 8q24 and loss of PTEN are associated with a high Gleason score, disease progression, and poor clinical outcome (1, 3, 12–21). Further, in a recent study by Liu and colleagues (1), gain of MYC and loss of PTEN as a combination were the only copy number changes that were associated with a markedly elevated risk of prostate cancer–specific mortality independent of other risk factors, raising the hypothesis that MYC gain and PTEN loss may cooperate to drive genomic instability and lethal disease in human prostate cancer.

Genetically engineered mouse models (GEMM) that phenocopy all stages of prostate cancer, including the development of pre-invasive prostatic intraepithelial neoplasia lesions, locally invasive disease, metastatic dissemination to relevant organs, and the progression to castration resistance in immune-competent animals, would prove invaluable. Despite two decades of effort to develop mouse prostate cancer models, all extant models have at least one of the following limitations: (i) they are driven by alterations in genes not commonly found to be genetically altered in human prostate cancer; (ii) they do not develop widespread metastatic disease; (iii) they develop prominent histologic features not commonly found in human prostate cancer (e.g., small cell neuroendocrine or sarcomatoid differentiation); or (iv) they generally do not develop significant numbers of genomic alterations/genetic instability in the absence of forced telomere dysfunction (22–24). Additionally, as most genetically engineered models of prostate cancer rely on forced androgen-driven oncogene expression, these models are limited when exploring the effects of castration/androgen deprivation, because such treatments necessarily result in direct repression of the transgene, which in turn generally leads to growth suppression (22).

Prior studies using GEMMs have shown that in the mouse prostate, loss of both Pten alleles, or activation of MYC, can each result in PIN and early invasive carcinoma, with a very low penetrance of metastases (22). Kim and colleagues developed Z-MYC mice, in which a CMV enhancer/β actin promoter–driven MYC gene is expressed in a small fraction of luminal cells upon Probasin/Cre–mediated activation (25). The lesions obtained were proliferative but were reported to be histologically normal or arrested at low-grade PIN. Deletion of one or both alleles of Pten in Z-MYC mice resulted in acceleration of PIN and early carcinoma development, showing cooperativity between MYC and Pten, although metastases were not reported (25). In a subsequent study, Z-MYC mice with single copy Pten disruption (Pten^fl/±) on a Tp53^−/− background resulted in selection for loss of the second Pten allele and the development of carcinomas and lymph node metastases (26). These findings further demonstrate cooperativity between MYC and Pten, although metastases developed only in mice lacking Tp53, distant metastases to other sites were not seen, and genomic instability was not examined (26). Cho and colleagues (27) used intraprostatic lentiviral infection to inactivate Pten and Tp53 (RapidCaP) and found that distant metastatic lesions exhibited increased Myc expression, which was required for metastatic tumor formation/maintenance. Further, using a similar strategy to activate MYC in the context of low Pten (Pten^hy/−), Cho and colleagues demonstrated cooperativity between MYC and Pten in the development of local disease spread, although distant metastases were not seen (27). To our knowledge, the only study to date to report widespread metastatic prostate carcinoma and genomic instability was in mice with targeted disruption of both copies of Tp53 and Pten in the setting of forced telomere shortening (6). However, aggressive tumors in these mice with telomere dysfunction have been reported to be largely sarcomatoid, a phenotype lacking in the vast majority of human prostatic cancers (22).

In the present work, we show that human MYC activation and Pten loss driven by androgen-independent Hoxb13 control elements in mouse prostate luminal cells (28) produces genomic instability and highly penetrant metastatic disease in the absence of induced telomere dysfunction or Tp53 loss of function. This model, simulating the highly relevant MYC activation and PTEN loss observed in aggressive human cancers, generates prostate adenocarcinomas that recapitulate many key aspects of the human disease, including widespread metastases to multiple organs such as lymph nodes, liver and lung.

The Cre recombinase coding sequence was PCR-amplified from pTurboCRE (GenBank accession no. AF334827) with the following primers: forward, 5′-CGCAGATCTGGCACCCAAGAAGAAGAGGAAG-3′; and reverse, 5′-CTGCAAGATGGCGATTAGTCTAGATCTGCG-3′. The resulting PCR product was digested with BglII and inserted into BamHI-digested pLZURA-Nkx (29). The SV40 early poly(A) signal present on a 135-bp XbaI fragment was inserted downstream of the CRE open reading frame into an SpeI site to generate pNkx-TCRE. To generate pHoxb13-TCRE, a SmaI-SpeI fragment containing the coding region of CRE was obtained from pNkx-TCRE and cloned into SmaI- and SpeI-digested pLZKAN-Hoxb13. A recombinogenic fragment of pHoxb13-TCRE was released by KpnI-SacII digestion, purified by sucrose fractionation, and recombineered into exon 1 of the Hoxb13 transcription unit at codon 8 within BAC RP23-335O22 as described (28). Escherichia coli strain DY380 transformants selected for chloramphenicol resistance were confirmed to contain the Cre coding region by RFLP analysis. A correctly recombined Hoxb13-TCRE BAC clone was prepared using Qiagen Large-Construct DNA preparation reagents according to the manufacturer's instructions, linearized by with PI-SceI digestion, purified on a 10% to 40% sucrose gradient and injected into single-cell FVB/N embryos. Potential transgenic founders were screened by Southern blot analyses of genomic tail DNA digested with EcoRI and EcoRV. A 255-bp probe within Hoxb13 exon 1 (+124 to +378 bp) was used for Southern blot analyses. Transgenic founder mice were identified by the presence of a 10-kb hybridizing component. PCR-based genotyping of F₁ offspring of founders bred to FVB mates was performed to amplify a 286-bp fragment within Cre with the following primers: forward, 5′-TCGCAAGAACCTGATGGACA-3′; and reverse, 5′-CAGCATTGCTGTCACTTGG-3′. Hoxb13-Cre activity was functionally tested using the R26R allele (30). Heterozygous R26R mice (The Jackson Laboratory, Bar Harbor, Maine) were backcrossed to FVB/N and F₁ offspring carrying the R26R locus were identified by Southern blot analyses using a lacZ probe on EcoRV digested tail DNA. R26R × FVB F₁ mice were interbred to derive an F₂ and homozygous R26R offspring were identified by Southern blot analyses. Hoxb13-TCRE founders were bred to homozygous R26R mice and double transgenic mice offspring were identified by PCR and confirmed by Southern blot analyses as described above. The Hoxb13-MYC BAC was generated using the same strategy described above for the Hoxb13-TCRE BAC. The human MYC coding sequence was PCR-amplified from pCMV-Sport6 (Open Biosystems) with the following primers: forward, 5′-GCATCCCGGGGATCCACCCCTCAACGTTAGCTTCACC-3′; and reverse, 5′-GCGGATCCACTAGTTCCTTACGCACAAGAGTTCCGTA-3′. The resulting PCR product was digested with BamHI and inserted into a BamHI-digested vector bearing Hoxb13 homologous arms for recombineering. The β-globin intron/poly(A) signal present in the pUGH17-1 vector (31) was PCR amplified and inserted downstream of the MYC open reading to generate pHoxb13-MYC-BG poly(A). A recombinogenic fragment of pHoxb13-MYC-BG poly(A) (complete sequence available upon request) was released by KpnI and NheI digestion, purified by sucrose fractionation, and recombineered into exon 1 of the Hoxb13 transcription unit at codon 8 within BAC RP23-335O22 as described (28). E. coli strain DY380 transformants selected for Kanamycin resistance were confirmed to contain the MYC coding region by RFLP analysis. A correctly recombined Hoxb13-MYC-BG Poly(A) BAC clone was prepared using Qiagen (Valencia) Large-Construct DNA preparation reagents according to the manufacturer's instructions, linearized by with PI-SceI digestion, purified on a 10% to 40% sucrose gradient and injected into single-cell FVB/N embryos. Potential transgenic founders were screened by Southern blot analyses of genomic tail DNA digested with EcoRI. A 255-bp probe within Hoxb13 exon 1 (+124 to +378 bp) was used for Southern blot analyses. Transgenic founder mice were identified by the presence of a 6-kb hybridizing component. PCR-based genotyping of F₁ offspring of founders bred to FVB mates was performed to amplify a 238-bp fragment within MYC with the following primers: forward, 5′-TCCAGCGCCTTCTCTCCGTCCTCGGATTCTC-3′; reverse, 5′-ATGCGTCGACCCTTTTGCCAGGAGCCTGCCTCTTTTCCAC-3′.

Mice carrying a floxed Pten (Pten^Fl) conditional allele (32) were purchased from The Jackson Laboratory (stock number 004597). The deletion of Pten allele(s) was confirmed using the following primers, which yielded a wild-type product of 240 bp and a floxed allele product of 320 bp: forward primer 1, 5′-TTGCACAGTATCCTTTTGAAG-3′; forward primer 2, 5′-GTCTCTGGTCCTTAC TTCC-3′; and reverse primer, 5′-ACGAGACTAGTGAGACGTGC-3′. The PCR protocol was followed according to the NCI Mouse Repository.

Hemizygous Hoxb13-Cre^+/− mice were mated to the Pten^Fl/Fl mice to generate the genotype Hoxb13-Cre^+/−∣Pten^Fl/+. These mice were interbred to generate mice with the Hoxb13-Cre^+/−∣Pten^Fl/Fl genotype. Hoxb13-MYC mice were then mated to Hoxb13-Cre^+/−∣Pten^Fl/Fl mice to generate offspring heterozygous for all three alleles (Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/+). These mice were then interbred to generate triple transgenic mice, which were Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/Fl, referred to as BMPC mice in this article. Cohorts of triple transgenic males for analyses were derived from crosses of varying genotypes, including Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/+ males mated with Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/Fl females, in addition to Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/Fl by Hoxb13-MYC^+/−∣Hoxb13-Cre^+/−∣Pten^Fl/Fl crosses.

Control and tumor tissues were dissected and fixed in 10% neutral-buffered formalin for 48 hours at room temperature and then transferred to phosphate buffered saline at 4°C until submission. Samples were processed, paraffin-embedded, sectioned, and stained with hematoxylin and eosin (H&E) according to standard protocols. Immunohistochemical analysis was performed on tissue sections using antibodies that include MYC (Rabbit Monoclonal, Clone EP121, Epitomics, 1:600), Pten (Rabbit Monoclonal, Clone D4.3, Cell Signaling Technology, 1:200), Ki67 (Rabbit Polyclonal, NCL-Ki67p, Leica Biosystems, 1:3,000), Nkx3.1 (Rabbit Polyclonal, 1:6,000; ref. 33), cytokeratin 18 (CK18; Rabbit Polyclonal, Novus Biologicals, NB100-91814, 1:8,000), androgen receptor (AR; Rabbit Polyclonal, Santa Cruz Biotechnology, sc-816, 1:400), chromogranin A (Rabbit Polyclonal, Novus Biologicals, NB120-15160, 1:8,000), FoxA1 (Rabbit Polyclonal, Santa Cruz, sc-6553, HNF-3 α/β, 1:400), FoxA2 (Rabbit Polyclonal, Abcam, HNF-3 Beta, 1:400), p63 (Mouse Monoclonal, Clone 4A4, Neomarkers, 1:600), p-AKT (Rabbit Monoclonal, Cell Signaling Technology, Ser473 D9E, 1:100), and p-S6 (Rabbit Monoclonal, Cell Signaling Technology, Ser235/236 D57.2.2E; 1:4,000).

For four BMPC mice, tail, primary prostate cancer, and metastatic tumor tissues were obtained and DNA was extracted using the Qiagen AllPrep DNA/RNA/Protein Mini Kit. Mouse male reference DNA (The Jackson Laboratory) was used as a common reference for all samples. Similarly, tail and prostate DNA were prepared from three Hoxb13-MYC^+/− and three Hoxb13-Cre^+/−∣Pten^Fl/Fl mice. For Hoxb13-MYC^+/− mice, the ventral lobes, which develop the most extensive PIN lesions, were used from mice that were between 10 and 13 weeks of age and similarly for the Hoxb13-Cre^+/−∣Pten^Fl/Fl mice, the anterior lobe was used from mice that were between 17 and 20 weeks of age. Array comparative genomic hybridization (aCGH) was performed with SurePrint G3 Mouse CGH 4 × 180 K Microarrays essentially as recommended by the manufacturer (Agilent Technologies). Briefly, 1.5 μg of genomic DNA was digested for 2 hours at 37°C with 1 unit/mL of AluI and RsaI restriction endonucleases (Invitrogen). The reaction was terminated by incubating at 65°C for 20 minutes. Digested DNA was subsequently labeled with Cy3 dye (for experimental samples: tail DNA, primary prostate cancer DNA, or metastatic prostate cancer DNA samples) or with Cy5 for the reference male DNA. Each Cy3-labeled sample was combined with the Cy5-labeled reference DNA and diluted in aCGH buffer and mouse Cot-1 DNA (Invitrogen). These combined labeled DNA samples were hybridized on the Mouse CGH 4 × 180 K microarrays in a hybridization rotator at 20 rpm for 40 hours at 65°C before being washed in Oligo aCGH Wash buffers according to the manufacturer's instructions (Agilent). aCGH slides were scanned on an Agilent DNA Microarray Scanner at a resolution of 2 μm. The resulting Agilent Feature Extraction data were analyzed using the Partek Genomics Suite v. 6.6 software (Partek). The Cy5/Cy3 log-ratio data were imported from the feature extraction file, transformed to log base 2 and multiplied by −1 to generate log2ratio(Cy3/Cy5). To facilitate identification of somatic CNAs in the cancer DNA not present in the tail DNA, the logratio data from the tail DNA for each mouse were subtracted from the matched primary and metastatic cancer logratio data. The resulting baseline subtracted logratio data were subjected to copy number detection using the genomic segmentation pipeline under default settings. All significantly altered regions with a false discovery rate of 0.05 were identified for each sample. The Partek copy number pipeline was used under default conditions to summarize regions of CNAs across samples, annotate with overlapping genes, and carry out gene ontology gene set enrichment analysis on genes overlapping with regions showing gains or losses in any sample. Enriched gene ontology terms with FDR = 0.05 were identified.

FVB/N mice carrying a recombineered transgene (34), allowing Hoxb13-driven MYC overexpression in the prostate epithelium (Hoxb13-MYC mice), developed PIN lesions that were characterized by marked nuclear and nucleolar enlargement within cells that maintained a columnar cellular polarity (Supplementary Fig. S1), similar to changes seen in Hi-MYC and Lo-MYC mice and to changes seen in human high-grade PIN (35, 36). MYC overexpression was observed in luminal epithelial cell nuclei (Supplementary Fig. S2C). Further phenotypic analyses showed robust androgen receptor (AR) and keratin 18 expression (CK18) in the atypical luminal cells, and positive Pten staining in all epithelial cells (Supplementary Fig. S2). To disrupt Pten alleles within the prostate, Pten^Fl∣Fl mice (The Jackson Laboratory; ref. 37) were crossed to FVB/N transgenic mice with Hoxb13-driven Cre recombinase expression (Hoxb13-Cre∣Pten^Fl/Fl mice). Prostate lobes of Hoxb13-MYC and Hoxb13-Cre∣Pten^Fl/Fl were dissected at multiple time points and only PIN lesions, with no invasive carcinomas, were present up to one year of age.

We next generated compound mutant mice with Hoxb13-driven overexpression of MYC and with Hoxb13-mediated conditional disruption of Pten alleles (Hoxb13-MYC⁺∣Hoxb13-Cre⁺∣Pten^Fl/Fl), referred to as BMPC mice. Cohorts of BMPC males were necropsied at various intervals from 4 weeks onward (Supplementary Tables S1 and S2). For controls, Hoxb13-MYC⁺∣Hoxb13-Cre⁺∣Pten^FL/+ males as well as wild-type FVB/N males were used. All BMPC mice developed PIN lesions in all four prostate lobes by 4 to 8 weeks. Immunohistochemical analysis confirmed overexpression of MYC and variable loss of Pten in PIN cells, which also phenocopied human PIN for strong positive staining for AR and Foxa1, with variable loss of Nkx3.1 (Fig. 1 and Supplementary Figs. S3 and S4). MYC staining occurred in virtually all cells that showed morphologic PIN by H&E staining, yet Pten loss appeared quite heterogeneous (Fig. 1C, Supplementary Fig. S4), despite Cre expression being driven by the same Hoxb13 regulatory elements. This focal loss of Pten is reminiscent of human prostate cancer in that primary tumors often lose PTEN as a subclonal process subsequent to TMPRSS2-ERG gene fusion events (38). Increased staining for p-AKT and p-S6 ribosomal protein was evident in areas of Pten loss (Supplementary Fig. S4) in PIN lesions and this pattern was similar in large invasive and metastatic lesions (not shown). Stromal invasion was first evident at 12–14 weeks (Supplementary Fig. S5) and as early as 16 weeks, large invasive adenocarcinoma lesions involving multiple prostate lobes were present (Fig. 2). Tumor cells in invasive lesions contained markedly enlarged nucleoli and nuclei (unlike small cell neuroendocrine carcinomas) and showed more pleomorphism than their intraepithelial counterparts (Figs. 1K and 3A,a, e). Although there was occasional acinar/glandular formation, these tumors generally resembled very high-grade primary prostatic adenocarcinomas in humans (Gleason score 5 + 5 = 10). Also like their human counterparts, invasive prostatic carcinomas lacked basal cells as shown by the absence of staining for p63 and keratin 5 (CK5; not shown). Further, invasive carcinomas remained positive for CK18 and FoxA1 (not shown), but were either negative for neuroendocrine markers (chromogranin A, FoxA2) or showed only variable, occasional tumor cell staining (<1%; Supplementary Fig. S6). All overtly invasive lesions, typically showing very high volume disease, were negative for AR (Fig. 1N). Among animals 16 weeks of age or older, 97% (31/32) developed gross metastases to pelvic lymph nodes (Fig. 2A). Grossly visible and microscopic metastases were also common to liver (Fig. 3A, a–d), and lung, and one mouse had a single bone metastasis in a thoracic vertebra (Fig. 3A, e–h; Supplementary Table S2). Tumor cells within metastatic lesions were similar to invasive primary lesions in that they were strongly positive for MYC (Fig. 3A, c and g), completely negative for Pten protein (Fig. 3A, d and h), positive for CK18 (Fig. 3A, b and f) and Foxa1 (not shown), and negative or focally positive for neuroendocrine markers including chromagranin A, and Foxa2 (Supplementary Fig. S6). They were also negative for nuclear AR and basal cell markers (CK5 and p63; data not shown). All BMPC mice (N = 32) reached criteria for euthanasia by 40 weeks (Fig. 3B). By contrast, none of the control Hoxb13-MYC⁺∣Hoxb13-Cre⁺∣Pten^FL/+ mice developed large primary invasive adenocarcinomas or metastatic lesions (Supplementary Table S3) by 52 weeks of age, indicating that complete Pten loss was required for the development of aggressive/lethal carcinomas. No sarcomatoid or neuroendocrine differentiation was observed in either primary or metastatic lesions.

To determine the effects of androgen deprivation, we castrated animals at different time points. Castration at 4 weeks of age led to a marked delay in the development of aggressive cancers, demonstrating the early androgen dependence of disease (Supplementary Table S4; Supplementary Fig. S7). Nevertheless, histologic analyses revealed residual PIN at times, and, in one case, a microscopic invasive carcinoma at 38 weeks of age, suggesting that these castration-resistant residual lesions have potential to progress, albeit at a markedly reduced rate. Castration at either 8 or 14 weeks also had a marked effect by delaying tumor progression, especially metastases (Supplementary Table S4). Nevertheless, lesions often progressed to invasive disease, with occasional metastatic spread, indicating the spontaneous emergence of castration-resistant disease (Supplementary Table S4).

We performed genome-wide microarray-based comparative genomic hybridization (aCGH) analyses on DNA from matched frozen primary tumor tissue, one site of metastasis (either lymph node or lung), and tail (germline) DNA from each of four mice. Every primary and metastatic tumor analyzed showed evidence of significant somatic CNAs, with each tumor harboring on average approximately 20 regions of copy number gains or losses spanning significant portions of the genome (range: 7–43 regions, spanning 1.27–118.95 Mbp of sequence; Supplementary Tables S6 and S7; Fig. 4). Among these alterations, 10 regions showed evidence of recurrent somatic CNAs, occurring in at least 1 tumor from at least 2 BMPC mice (Fig. 4A, Supplementary Table S8), and 8 regions showed evidence of high copy number gain (logratio > 1.2; 3 regions) or low copy number loss representing putative homozygous loss (logratio < −1.2; 5 regions; Supplementary Table S9). Interestingly, genes within regions of copy number gain were enriched for chromatin modulation and hormone signaling pathways, whereas genes within regions of copy number loss were enriched for cytokine/chemokine signaling, immune system, and development/differentiation pathways (Supplementary Fig. S7); these pathway categories are often altered at a genetic or epigenetic level in human prostate cancers (2, 39). Representative CNAs from chromosome 8, including a large deletion encompassing a smaller region of putative homozygous loss in both the primary and metastasis lesions from one mouse, is shown in Fig. 4B. Clustering of copy number data revealed that primary and metastatic tumors from the same mouse were more similar to each other than they were to tumors from other mice (Fig. 4B, dendrogram), suggesting that the primary and metastatic lesions within mice had clonal evolutionary relationships, similar to those observed for human prostate cancer (39–41). In contrast, CNA analyses of Hoxb13-MYC^+/− and Hoxb13-Cre^+/−∣Pten^Fl/Fl mice with PIN lesions revealed no CNAs. No correlations were identified between the number of CNAs and the extent or location of metastases.

Although late-stage human prostate adenocarcinomas frequently show loss of heterozygosity at the TP53 locus, copy number variation at the mouse Tp53 locus was not observed in BMPC tumors. To explore the possibility that point mutation of the Tp53 gene may have occurred, exons 3 to 8 (data not shown) were PCR-amplified and sequenced from tumor DNA extracted from the same cases. No mutations were observed.

Here we show that conditional activation of MYC along with loss of Pten in mouse prostatic luminal epithelial cells, simulating the genetic activation of MYC and loss of PTEN observed in aggressive human prostate cancers, creates the full spectrum of prostate cancer initiation and progression in a manner that strongly phenocopies human prostate cancer. This disease progression appears to occur through the induction of genomic instability evidenced by widespread CNAs, similar to human prostate cancer progression. Thus, although activation of MYC alone or disruption of Pten alone result in very early disease, the combination of these two events synergize in creating profound changes leading to highly aggressive disease that progresses from PIN precursor lesions, to invasive adenocarcinoma, to metastatic adenocarcinoma, and ultimately to castration-resistant metastatic adenocarcinoma. As these primary tumors develop highly prevalent CNAs, and neither alteration alone results in significant numbers of CNAs (24, 35), we show the first evidence in vivo that the combination of Pten loss and MYC activation in prostatic luminal cells is sufficient to induce the acquisition and tolerance of genetic instability, independent of forced telomere dysfunction or loss/mutation of Tp53. MYC overexpression is known to induce genetic instability (42), and our results suggest that PTEN serves to repress this process in prostate cells. Recent findings indicating that PTEN is required for DNA repair and that loss of PTEN can result in high levels of DNA damage in cells that simultaneously repress apoptosis due to increased PI3K signaling, are consistent with this hypothesis (43). Future studies are required to address this question directly. It is interesting to note that a subset of the CNAs detected in the BMPC tumors overlap with regions of known copy number variants in mouse strains. These alterations are likely to be additional somatic alterations at these regions even beyond the baseline CNV state in the mouse strain, since we controlled for the constitutional copy number state with the tail DNA. In addition, the single gene models did not show these similar CNAs, suggesting that these regions were not simply identified due to undercorrection.

Several previous attempts have been made to model Pten loss or PI3K pathway activation in combination with MYC overexpression. Prostate-epithelial restricted co-expression of human AKT1 and MYC leads to accelerated disease progression, stromal changes, and immune cell infiltration but does not affect metastasis (44). In addition, conditional Pten loss in the Hi-MYC background gave rise to large primary tumors (44), although metastatic lesions were not reported upon. Highly focal epithelial MYC expression (Z-MYC mice) combined with conditional Pten loss also fostered the appearance of more aggressive pathology with increased proliferation and development of early invasive carcinomas apparently without affecting metastasis (25, 26). When combined with loss of one allele of Tp53, Z-MYC/Pten mice developed adenocarcinoma that metastasized to lymph nodes (26). Given the well-documented dose-dependent biologic effects of MYC overexpression (45), we submit that the lethal phenotype that invariably develops in the BMPC mice is likely a consequence of a higher level or broader expression of MYC throughout the prostatic epithelium driven by Hoxb13 regulatory elements in our system, at least as compared with the Z-MYC model. Interestingly, Kim and colleagues (25) speculated that increased MYC expression could lead to genetic instability, and the data reported here provide evidence that this is indeed the case.

It is of interest that other studies that have obtained macroscopic metastatic disease in mice by disruption of both copies of Pten have reported that disease progression required loss of both copies of Tp53 (27, 46). Thus, the current study, in which we used mice that were wild type for Tp53 and found no evidence of Tp53 alterations in aggressive lesions from BMPC mice, demonstrates the novel finding that in the setting of MYC overexpression and Pten loss, emergence of aggressive distant metastatic disease can occur without Tp53 loss.

Invasive and metastatic lesions in BMPC mice do not express AR, yet also do not develop neuroendocrine nor sarcomatoid differentiation. Logothetis and colleagues have described highly aggressive prostate cancers that lack AR both with and without overt neuroendocrine features (47), and it is expected that these phenotypes will increase as androgen suppression/AR inhibitor therapies continue to improve and are used in combination. Thus, in addition to providing a model to test novel therapeutics, including those based on immunotherapies, the BMPC model provides a robust system to dissect the molecular bases underlying the evolution of AR-loss in non-neuroendocrine tumors, as well as AR-independent mechanisms of disease progression and growth.

Although the AR-negative phenotype of advanced cancers in BMPC mice does not reflect the AR status of the majority of late-stage human cancers, it is also not completely unexpected. It is well established that conditional loss of Pten in the mouse prostate impairs mouse AR (mAR) signaling in the cancers that develop (48). Reduced mAR expression at both the mRNA and protein levels has been reported, and this diminution can be partially reversed by PI3K pathway inhibition (49). Given that mAR is known to be positively autoregulated in the mouse prostate, the reduction of AR is not, a priori, surprising (50). In human prostate cancer cases with PTEN loss, human AR (hAR) expression typically persists. Although some in vitro evidence suggests that PTEN loss may lead to AR stabilization or transcriptional activation (51, 52), transcriptomic analyses of prostate cancer cases clearly demonstrate that androgen-responsive gene signaling is profoundly diminished when PTEN is lost (48, 49). We speculate that the differences in AR expression in mice versus human prostate epithelial cells in the context of PTEN loss may be related to evolutionary divergence in the extent of AR gene autoregulation between primates and rodents. The mAR gene may require a higher degree of autostimulation to maintain transcriptional activity than hAR and results in a threshold-based on/off switch that follows second order kinetics. In contrast, the hAR gene may respond in a linear fashion, wherein a reduction in autostimulation would result in reduced transcriptional output but not extinction of expression. Metastatic lesions obtained in RapidCaP mice also do not express AR (27).

The strong parallels with the spectrum of human prostate cancer initiation and progression in this model are likely a result of several important design features in the model. First, this genetically engineered model is driven by concurrent activation of MYC and loss of Pten, the combination of which has been strongly associated with human prostate cancer aggressiveness (1). Second, MYC activation and Pten loss were driven by Hoxb13 regulatory elements, allowing AR-independent genetic modulation. Avoidance of AR-dependent control elements (e.g., the commonly used probasin or the derivative ARR2 elements) likely allowed development of castration-resistant disease. In humans, the HOXB13G84E allele has recently been demonstrated to be highly associated with increased prostate cancer risk (53–55). Given this, it seems likely that it is expressed in a population of cells that contribute directly to disease development. In mice, Hoxb13 transcriptional activity is present broadly within the secretory epithelium and in a subset of basal cells in all prostate lobes (28). Interestingly, the tumors that developed in the BMPC animals had a luminal phenotype and loss of Pten was confined to luminal cells in PIN lesions (Supplementary Fig. S3), analogous to human disease. Another potential advantage of the Hoxb13 regulatory elements for driving prostate expression of MYC and loss of Pten is that we have not observed any leakiness of expression of Hoxb13-based reporter genes, MYC overexpression or Pten loss in any of these animals, whereas leakiness of the rat probasin promoter into stromal elements has been observed (56). This feature should facilitate the study of relevant cancer–stromal interactions because the genetic modulations would not occur in the stromal cells of the BMPC model. These design features have allowed us to avoid the pitfalls of other recently developed prostate cancer GEMMs, including the common occurrence of sarcomatoid or overt neuroendocrine differentiation, lack of progression through early and advanced stages of human prostate cancer in tractable time scales, and/or lack of development of genetic instability (22, 57, 58). These features should facilitate the use of this model to study prostate cancer initiation and disease progression, study mechanisms underlying the cooperativity and synergy of MYC activation and PTEN loss in the development of genomic instability and aggressive human prostate cancers, and test relevant prostate cancer therapeutic strategies in an immunocompetent model.

No potential conflicts of interest were disclosed.

Conception and design: G.K. Hubbard, M. Khalili, R.P. McMullin, P.S. Nelson, S. Yegnasubramanian, A.M. De Marzo, C.J. Bieberich

Development of methodology: G.K. Hubbard, L.N. Mutton, M. Khalili, R.P. McMullin, L.A. Horn, S. Yegnasubramanian, C.J. Bieberich

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.K. Hubbard, L.N. Mutton, M. Khalili, D. Bianchi-Frias, L.A. Horn, I. Kulac, M.S. Moubarek, P.S. Nelson, A.M. De Marzo, C.J. Bieberich

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.K. Hubbard, M. Khalili, D. Bianchi-Frias, L.A. Horn, P.S. Nelson, S. Yegnasubramanian, A.M. De Marzo, C.J. Bieberich

Writing, review, and/or revision of the manuscript: G.K. Hubbard, P.S. Nelson, S. Yegnasubramanian, A.M. De Marzo, C.J. Bieberich

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.K. Hubbard, L.N. Mutton, R.P. McMullin, M.S. Moubarek, S. Yegnasubramanian

Study supervision: G.K. Hubbard, A.M. De Marzo, C.J. Bieberich

This work was supported by the following grants: DAMD17-98-1-8477 and R21CA105342 to C.J. Bieberich; a Patrick C. Walsh Prostate Cancer Research Fund award to A.M. De Marzo and C.J. Bieberich; a 2013 Movember-Prostate Cancer Foundation Challenge Award to A.M. De Marzo, S. Yegnasubramanian, and C.J. Bieberich. A PCF Challenge Grant, R01CA183965 and P30CA006973 (NIH/NCI), and a Commonwealth Foundation Grant to S. Yegnasubramanian.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.