FGFR1–WNT–TGF-β Signaling in Prostate Cancer Mouse Models Recapitulates Human Reactive Stroma

Cancer growth and metastasis require proliferation, cell migration, and stromal remodeling—functions also critical during organogenesis and wound repair. The fibroblast growth factors (FGF) and their receptors (FGFR) play pivotal roles in development, wound healing, and tumorigenesis (1). FGFR1 is a receptor tyrosine kinase that signals through the RAF–mitogen-activated protein/extracellular signal–regulated kinase (MEK)–mitogen-activated protein kinase (MAPK)–extracellular signal–regulated kinase-1/2 (ERK1/2) kinase cascade and the phosphoinositide 3-kinase (PI3K)–AKT axis, both of which are well-described oncogenic pathways, shown to promote androgen independence in prostate cancer (2, 3). WNT signaling is vital for embryogenesis, homeostasis of adult tissues, and wound repair, as well as being associated with malignancy (4–11). In the absence of WNT ligands, the GSK-3β–containing “destruction complex” phosphorylates β-catenin, targeting it for proteasomal degradation (4). Canonical WNT signaling involves WNT ligands binding to the coreceptors Frizzled and lipoprotein receptor-related protein (LRP)-5/6, facilitating disruption of the destruction complex, allowing β-catenin to translocate to the nucleus.

There are many examples of FGF and WNT signaling cooperating in development and tumorigenesis (12–14). In mammary tissue, FGFR1 rapidly accelerates Wnt-1–induced carcinomas (15), whereas the FGFR1 targets, MAPK–pERK, are known to modulate WNT signaling and colorectal tumorigenicity (16). In addition, WNT signaling cooperates with activated K-RAS, and is upregulated in FGFR1-driven prostate cancer (10, 17). Moreover, recent searches of two independent databases revealed a modulation of either FGF and/or canonical Wnt signaling members in 81% and 92% of completely characterized prostate adenocarcinoma cases, respectively. Modulation of both pathways within the same case occurred in 38% and 72% of the cases, respectively (18). Furthermore, tumor-associated reactive stroma is similar to hyperproliferative stroma found during wound repair. Appropriately, mechanical injury modulates the FGF and WNT, as well as TGF-β, signaling pathways (19–24). Finally, the expression of FGF, WNT, and TGF-β signaling primes bone metastatic niches. These intricate developmental and wound-repair interactions support the synergistic cooperation of these signaling pathways in carcinogenesis that includes a poorly understood, complementary stromal component.

We hypothesized that FGFR1 and canonical Wnt signaling, within the same epithelial cells (intracellular), should accelerate tumor progression, given the previously described data (discussed below). However, it was unclear whether (intra/intercellular) paracrine signals derived from associated stromal tissue would further accelerate tumor progression, producing a second potential target for therapeutic intervention. In order to study epithelial FGFR1 signaling, we previously generated the synthetic ligand-inducible FGFR1 transgenic mouse model, JOCK1, via use of the prostate epithelium-specific, composite probasin promoter, ARR₂PB (25). The addition of a lipid-permeable chemical inducer of dimerization (CID), AP20187, causes the intracellular signaling domains of FGFR1 to oligomerize, inducing downstream signaling (26, 27). We previously demonstrated that prolonged induction (∼60 weeks) of FGFR1 results in a step-wise progression to invasive prostate cancer with distant metastasis (17). Furthermore, we recently demonstrated that induced aggregation of the WNT coreceptor LRP5, is sufficient to induce β-catenin stabilization, similar to canonical WNT signaling (28). We subsequently developed two novel transgenic mouse models wherein WNT signaling can be specifically activated in the prostate epithelium, yielding “Pro-Cat” mice (Prostatic Activator of β-catenin; ARR₂PB promoter) or ubiquitously, in “Ubi-Cat” mice (Ubiquitous Activator of β-catenin; H-2K^b promoter; ref.28). Interestingly, Pro-Cat mice do not progress beyond prostatic hyperplasia; however, after a year of induction, by which time murine prostatic epithelia can undergo natural wounding (29), two out of six Ubi-Cat mice developed adenocarcinoma (28). Furthermore, WNT signaling derived from Ubi-Cat adult stroma enhanced prostate tissue reconstitutions, further supporting an additional role for stromal WNT signaling in tumorigenesis (28).

In this report, we delineate synergism between FGF and WNT pathways across different subtissue layers in prostate carcinogenesis by crossing Pro-Cat and Ubi-Cat lines onto the JOCK1 background. We observe that WNT signaling accelerates FGFR1-driven adenocarcinoma, through both intracellular cross-talk (Pro-Cat × JOCK1) and intra/intercellular cross-talk (Ubi-Cat × JOCK1) by significantly accelerating adenocarcinoma initiation, whereas intra/intercellular cross-talk accelerates tumor progression. In addition, we observed that WNT signaling drives an inflammatory, fibroblastic reactive stroma in both endogenous tumor and reconstitutions assays, in contrast to the primarily myofibroblastic reactive stroma seen in human prostate cancer (Table 1). Finally, we have identified increased stromal TGF-β signaling in both human and murine tumors as a potential mechanism for accelerated tumorigenesis.

Mice were housed in a pathogen-free facility by following approved Institutional Animal Care and Use Committee protocols. Generation and genotyping of transgenic mice have been previously described (27, 28). Mice were treated biweekly, starting at 6 weeks of age, by intraperitoneal (i.p.) injections of AP20187 (Ariad Pharmaceuticals) at 2 mg/kg in drug diluent (1,2-propanediol, 22.5% PEG400, 1.25% Tween 80). The 8-week-old nu/nu mice were procured from Charles River Laboratories. See Table 2 for N values.

Tissue microarrays have been described previously (30). Radical prostatectomy tissues were collected with informed consent, as approved by the Baylor College of Medicine Institutional Review Board, by the Prostate Cancer Specialized Program of Research Excellence (SPORE). All patients underwent surgery for clinically localized prostate cancer and had received no prior treatment.

Tissues were fixed in either 10% buffered formalin (JOCK1, Ubi-Cat, and Ubi-Cat × JOCK1) or 4% paraformaldehyde (Pro-Cat and Pro-Cat × JOCK1) overnight, embedded in paraffin, and 5-μm sections were stained with hematoxylin and eosin (H&E) or the Accustain Trichrome Stain system (Sigma-Aldrich).

Immunohistochemical (IHC) analysis of 5-μm sections was performed, as described, for each antibody using standard light microscopy. SMA: no antigen retrieval; 1:400 overnight, 4°C mouse monoclonal smooth muscle-α actin (SMA; Sigma A2547); secondary—BioCare's MM AP Polymer 30 minutes at room temperature; Vector red substrate for 20 minutes at room temperature. Vimentin: 20 minutes heated EDTA buffer pH 8.0 antigen retrieval; 1:50 overnight, 4°C goat polyclonal vimentin (sc-7557); secondary—BioCare's goat HRP detection kit for 15 minutes at room temperature; substrate—DAB, enhanced by BioCare's DAB Sparkle. Tenascin C: heated EDTA buffer, pH 8.0 antigen retrieval; 1:10 overnight, 4°C rabbit monoclonal tenascin C (ab108930); secondary—1:500 Anti-rabbit, 1 hour at room temperature; Vector red substrate, 30 minutes at room temperature. Phospho-Smad2: heated citrate buffer, pH 6.0 (Diagnostic BioSystems), 20 minutes antigen retrieval; 1:100 overnight, 4°C, pSmad2 (Millipore, AB3849); secondary—Rabbit on Rodent (BioCare Medical), for 30 minutes at room temperature; substrate—DAB, enhanced by BioCare's DAB Sparkle. Ki-67: heated citrate buffer, pH 6.0 (Diagnostic BioSystems), 20 minutes antigen retrieval; 1:100 overnight, 4°C, anti-Ki-67 (SP6 Neomarkers RM-9106-SO); secondary—Rabbit on Rodent (BioCare Medical), for 30 minutes at room temperature; substrate—DAB, enhanced by BioCare's DAB Sparkle.

Embryonic stroma: prostates from 7- to 13-week-old untreated JOCK1, Pro-Cat × JOCK1, or Ubi-Cat × JOCK1 mice were isolated and sorted for basal progenitors (LSC), as previously described (31). In total, 100,000 LSC cells were mixed with 100,000 rat urogenital sinus mesenchyme (UGSM) in 100 μL Matrigel and transplanted subcutaneously into nu/nu mice and treated ± CID for 8 weeks or 30 weeks (JOCK1 grafts).

Adult stroma: prostates from adult untreated, wild-type (WT), JOCK1, or Ubi-Cat mice were isolated, and sorted for two populations: LSC (Lin⁻, Sca1⁺, CD49f⁺) and stroma (Lin⁻, Sca1⁺, CD49f⁻). In total, 100,000 cells were mixed from each population from different genotypes in 100 μL Matrigel, transplanted subcutaneously into nu/nu mice and treated ± CID (16 weeks).

Frozen samples were collected in ornithine carbamyl transferase (OCT) medium and stored at −80°C. mPIN lesions and adjacent reactive stroma were captured from 5-μm sections on the Pixcell Laser Capture machine. Total RNA was isolated using the CapSure Macro LCM caps (Applied Biosystems LCM0211) and amplified using Arcturus RiboAmp PLUS (Applied Biosystems KIT0501). Amplified RNA was labeled using the Invitrogen cDNA labeling system (L1015-06) and hybridized to the Agilent Whole Mouse Genome Oligo (4 × 44) Microarray (G4122) chip for analysis.

Histologic analysis was performed by a board-certified pathologist (M.M. Ittmann) in a blinded fashion. The penetrance of each histologic phenotype throughout the prostate (whole anterior, dorsal/lateral, and ventral lobes) was graded on a scale of 0 to 3 (0, none; 1, ≤25%; 2, ∼50%; 3, ≥75%). Scores for each phenotype were averaged across all mice in a particular group, giving the mean group score. Group phenotype scores were then weighted summed [(2 × PIN score) + (3 × adenocarcinoma score) + (4 × sarcomatoid carcinoma score) + (5 × sarcoma score)] to derive the total pathologic score for each group (plotted with SEM). In addition, tissues were scored for reactive stroma penetrance (disorganized stroma, characterized by fibroblasts, neovasculature, and inflammation) on a scale of 0 to 3 (0, none; 1, ≤25%; 2, ∼50%; 3, ≥75%). Finally, the score for each mouse was averaged to get the mean score for the group (reported with SEM). See Table 2 for the number of mice in each group. Statistical significance was determined by two-way ANOVA with Bonferroni multiple comparisons within the interaction effect. Two-tailed post-hoc power analysis for an α of 0.05 was also performed.

To determine the ratio of p-Smad2 in reactive stroma adjacent to increasing epithelia grades, 20 random high-power (×60 optic) fields of each mouse group and 387 normal, 272 PIN, and 343 tumor fields of patient samples were examined. Positive cells were defined as any stromal cells (outside the basement membrane: includes fibroblasts, muscle cells, and endothelial cells; excludes immune cells and epithelial cells) with clear nuclear staining (brown IHC over a faint blue counterstain). The total number of p-Smad2⁺ stromal nuclei were divided by the total number of stromal cells per field, and these values were averaged to determine the mean of each group with significance determined by two-way ANOVA with Bonferroni multiple comparisons. The significance of specific human subsets was determined by a t test or Mann–Whitney Rank sum test, as needed.

The percentage of Ki-67⁺ stromal cells was quantified by Inform software analysis of 10 random ×10 micrographs imaged by a Nuance camera. Percentages for each micrograph were calculated, and averaged to give a mean for each mouse. The mean score for each mouse was averaged to give the mean score for the group (reported value).

Groups were compared for statistical significance with one-tailed Student t tests.

Expression arrays were processed and loess-normalized using BioConductor, and heat maps were generated using JavaTreeView (32). Array data have been deposited in the Gene Expression Omnibus (GEO; accession GSE41344).

Both murine and human prostate tumorigenesis advances through well-characterized pathologic “hallmarks”: epithelial hyperplasia, mPIN, adenocarcinoma, and transitional sarcomatoid. Transitional sarcomatoid lesions have been recently reclassified as sarcomatoid carcinoma and are considered a histologic manifestation of epithelial–mesenchymal transition (EMT; Supplementary Fig. S3; ref. 33). Our previous work demonstrated that JOCK1 mice steadily progress through these hallmarks, similar to human prostate cancer, and reach an adenocarcinoma state by 40 weeks of CID treatment (Fig. 1; ref. 17). To determine whether canonical WNT signaling accelerates JOCK1 progression, thereby modeling more aggressive human cancer, we crossed Pro-Cat (inducible WNT signaling in prostate epithelium) and Ubi-Cat mice (ubiquitous expression of inducible WNT signaling) with JOCK1 animals (Supplementary Figs S1 and S2). Both strains were treated with CID biweekly for 12, 24, 40, or 60 weeks. Tissues were collected and analyzed for the presence and penetrance of progression hallmarks. Seminal vesicle occlusion correlated with prostate transformation (Supplementary Fig. S5); all other tissues were histologically normal (data not shown). Pro-Cat mice did not progress beyond minor hyperplasia. When combined with FGFR1 signaling, all Pro-Cat × JOCK1 mice developed adenocarcinoma significantly faster than monogenic JOCK1 mice (24 weeks vs. 40 weeks). However, tumors in Pro-Cat × JOCK1 mice did not progress until 60 weeks of treatment, similar to JOCK1, consistent with an age-dependent and self-limiting progression threshold or “brake”. Similar to Pro-Cat animals, most Ubi-Cat mice did not progress beyond hyperplasia; however, after 1 year of CID treatment, two out of six mice developed adenocarcinoma (28), also suggesting an age-dependent, self-limiting step. When crossed to JOCK1, all Ubi-Cat × JOCK1 mice displayed the same significant acceleration in initiation as Pro-Cat × JOCK1 mice; however, Ubi-Cat × JOCK1 mice continued to progress, resulting in tissues containing 50% sarcomatoid carcinoma by 40 weeks, 20 weeks earlier than JOCK1 or Pro-Cat × JOCK1 mice. Prostate tumors phenocopied the JOCK1 histopathology with heterogeneous, invasive hyperchromatic epithelial cells and loss of the basal cell layer, as previously described. These data suggest that the addition of WNT signaling changed the rate of progression: intracellular cross-talk (Pro-Cat × JOCK1) accelerated FGFR1-driven adenocarcinoma initiation, whereas intra- and intercellular cross-talk (Ubi-Cat × JOCK1) accelerated both progression and initiation.

We next investigated the ability of basal progenitor cells, known as LSC (Lin⁻, Sca1⁺, CD49f⁺), to reconstitute prostate pathology seen in the endogenous tumors. LSC cells from JOCK1, Pro-Cat × JOCK1, and Ubi-Cat × JOCK1 mice were isolated, transplanted subcutaneously along with inductive rat urogenital sinus mesenchyme in Matrigel into recipient mice, and treated ± CID for 8 weeks (28). Consistent with the autochthonous tumor progression, all transgenic LSC grafts displayed hyperplasia within 8 weeks. Interestingly, JOCK1 grafts also displayed mPIN, like the double-transgenic lines (Fig. 2A), indicating loss of a tumor-suppressing function, typically distinguishing WNT-accelerated FGFR1-driven pathology. To test whether JOCK1 grafts also achieved accelerated adenocarcinoma, we extended the CID treatment and observed reconstitution of adenocarcinoma-like lesions by 30 weeks, 10 weeks earlier than autochthonous JOCK1 tumors (Fig. 2B). This accelerated tumorigenesis could be attributable to the use of pro-inductive embryonic stroma, which, when combined with oncogenic initiation (FGFR1), recapitulated the pronounced, tumor-associated reactive stroma observed in double-transgenic mice. Thus, in the presence of inductive stroma, reconstituted grafts, containing FGFR1 and FGFR1/WNT signaling-activated progenitor cells, can progress to similar levels of pathology as in autochthonous transgenic tissues.

To further characterize the reactive stroma, we performed immunohistochemical analysis of several reactive stroma markers in histologically similar tumors from JOCK1 (i.e., following 40 weeks of CID treatment) and Pro-Cat × JOCK1 and Ubi-Cat × JOCK1 (24 weeks of treatment). The reactive stroma was very collagen-rich (Fig. 3A and Supplementary Fig. S4; Trichrome, blue staining) with Ubi-Cat × JOCK1 being the most dense. The SMA layer surrounding the acini in normal stroma is greatly reduced in reactive stroma. This correlates with an overall decrease in SMA; however, JOCK1 tissues still displayed focal areas of dense SMA staining (pictured). In addition, we observed dense expression of the fibroblastic marker, vimentin, in all reactive stroma (Fig. 3A; Supplementary Fig. S4). Furthermore, focal expression of the reactive stroma marker, Tenascin C, closely correlated with invasive lesions, as previously observed in human prostate cancer (30). Finally, overall proliferation across genotypes (Fig. 3A; Ki-67) correlated with the rate of progression. Interestingly, the reactive stroma of Ubi-Cat × JOCK1 tumors revealed considerably more Ki-67⁺ cells in the stroma. Thus, β-catenin stabilization accelerated FGFR1-driven prostate cancer, producing a heterogeneous, proliferative reactive stroma that is predominantly fibroblastic, as opposed to a myofibroblastic phenotype reported in human prostate cancer.

To further understand the relationship between the reactive stroma and cancer progression, we quantified the stromal pathologic score across each group (Fig. 3B). The development of reactive stroma was CID-dependent and consistently arose in conjunction with mPIN, prior to adenocarcinoma. This observation suggests the presence of reactive stroma could be driving prostate cancer development, as previously suggested in humans (30). Finally, there was significantly more reactive stroma in the double-transgenic mice over time with the most pronounced difference seen in the Ubi-Cat × JOCK1 animals (Fig. 3B). Interestingly, in the two Ubi-Cat mice that developed CID-dependent tumors, reactive stroma was observed only at focal sites of adenocarcinoma and was not observed in normal or mPIN tissue, suggesting canonical WNT signaling in autochthonous epithelium and stroma (without oncogenic stimuli) is not sufficient to drive a reactive stroma. Taken together, these data indicate that WNT signaling drives stromal expansion and, when accompanied by appropriate oncogenic signaling (e.g., FGFR1), correlates with aggressive prostate cancer.

Although we observed cooperative effects of FGFR1 and WNT signaling, limitations of autochthonous tumor models and the lack of an all-inclusive stromal promoter precluded a full assessment of whether WNT signaling in the stroma alone is sufficient to synergize with FGFR1 signaling in the epithelium. To address this, we utilized the ubiquitous iLRP5 expression in Ubi-Cat mice for prostate reconstitution assays. LSC cells and stromal cells (Lin⁻, Sca1⁺, CD49f⁻) from adult WT, JOCK1, or Ubi-Cat mice were isolated. Cell populations from different genotypes were mixed 1:1 in Matrigel and transplanted subcutaneously into athymic mice. Ubi-Cat stroma mixed with WT LSC cells in the absence of CID had very poor engraftment (in reference to very small acini size and number and not “take” rate), and only a small, single-cell layer acini (28). However, in the presence of canonical WNT signaling (+CID), acini were more numerous and developed, although their overall size was small. Interestingly, WNT-induced stroma was thicker with spindle cell-like morphology and contained vascular (Fig. 4A) components. Likewise, in the absence of CID, JOCK1 LSC and WT stroma had poor engraftment. The induction of FGFR1 signaling in the epithelium was sufficient to improve engraftment, without aberrant pathology, as was seen when combined with embryonic stroma (Fig. 2). Without CID, Ubi-Cat stroma and JOCK1 LSC grafts had similar poor engraftment rates and small acini (Fig. 4A and 4B). However, in the presence of WNT signaling, acini contained proliferative lesions and were significantly larger than CID-treated WT LSC + Ubi-Cat stroma (Supplementary Fig. S6). As shown in Fig. 4A, a reactive host response by histologic parameters is apparent in CID-treated JOCK1 LSC + Ubi-Cat stroma grafts. Meanwhile, control grafts (i.e., minus CID or CID-treated WT LSC + Ubi-Cat stroma) exhibited very little reactive stroma (Fig. 4A). These experiments demonstrate that WNT signaling alone in the stroma is associated with a reactive stroma phenotype in synergy with FGFR1-induced proliferation and premalignant transformation.

To elucidate the mechanism behind the WNT-accelerated FGFR1 adenocarcinoma, we performed laser capture microdissection (LCM) to separate precancerous (hyperplasia and mPIN) epithelia and adjacent reactive stroma cells. We utilized frozen samples from the same timepoints as described earlier, and performed gene expression profiling (Fig. 5A). Comparing the more aggressive Pro-Cat × JOCK1 precancerous tissue to JOCK1, we discovered modulation of the FGF-axis receptors (i.e., FGFR1, 2, 3, and 4) and ligands (i.e., FGFs 3, 7 and 8) in both the epithelium and stroma, demonstrating a feed-forward loop that could lead to accelerated CID-independence, previously described in JOCK1-driven tumors (17, 26). Furthermore, we observed an upregulation of various metastasis-associated genes (Snail2, Twist1, and R-Ras) in the epithelium of both bigenic lines versus JOCK1 (Supplementary Fig. S7). Interestingly, we also observed a significant upregulation of TGF-β ligands in the epithelium of double-transgenic mice compared with JOCK1 mice and a downregulation of TGF-β receptors in both the epithelium and stroma (Fig. 5B). This is consistent with the recent discovery that loss of TGF-βRII in a subset of stromal cells (but not all) promotes a more aggressive disease (34). Although Pro-Cat × JOCK1 and Ubi-Cat × JOCK1 tissues were compared, no notable differences were observed at this particular time point of “accelerated initiation”; however, later time points in disease progression might reveal gene expression changes that could distinguish an “accelerated progression” signature.

To explore which cells were responding to the increased TGF-β ligands, we examined nuclear localization of phosphorylated-Smad2 (p-Smad2) as a read-out for TGF-β receptor activation in JOCK1, Pro-Cat × JOCK1, and Ubi-Cat × JOCK1 tissues, and in a human tissue array, consisting of normal, mPIN and tumor samples from matched donors. Approximately 90% of the normal prostate epithelium darkly stained for p-Smad2. In light of the established antiproliferative role of TGF-β, this supports a tightly controlled signaling mechanism to maintain epithelial homeostasis (Fig. 5C). Interestingly, mPIN lesions showed high expression of nuclear p-Smad2, whereas hyperplasic areas showed p-Smad2 heterogeneity, suggesting that a change in homeostatic signaling permits neoplasia. Furthermore, most of the nuclei in normal stroma faintly to modestly stain for nuclear p-Smad2, whereas reactive stroma stained more intensely and heterogeneously. To investigate the association and potential role of this heterogeneity, we analyzed the percentage of nuclear p-Smad2⁺ stromal cells adjacent to various types of epithelium. Overall, the expression of tumor-associated stromal p-Smad2 correlated with increased disease aggressiveness in the mouse models (Fig. 5D). This suggests that the canonical WNT-driven partial loss of TGF-β receptors and increase in ligands identified by LCM could initiate the reactive stroma, which collaborates with neoplastic epithelium to drive accelerated adenocarcinogenesis in double-transgenic mice. Moreover, the higher overall TGF-β signaling observed in Ubi-Cat × JOCK1 stroma suggests that slight changes in TGF-β signaling in an already primed proneoplastic background elicits more rapid progression.

When the patient data was grouped according to pathologic scores, we discovered that Gleason 7 relative to Gleason 5 to 6 tumors and stage T2/T3a relative to T3b or N1 tumors revealed a strong trend (P = 0.073 and P = 0.112, respectively) toward higher TGF-β signaling in the stroma immediately adjacent to the tumors; however, the relatively limited cohort size (n = 49) may have tempered statistical significance. Interestingly, tumor-associated stroma containing greater than 60% nuclear pSmad2⁺ cells is significantly associated with more aggressive tumors than stroma that was less than 60% positive (P = 0.003; Fig. 5D). Finally, four patients had recorded biochemical recurrence with a trend in higher tumor-associated TGF-β; however, the small number of patients did not lend statistical significance. Together, these data demonstrate that both bigenic mouse lines model several key attributes of human prostate cancer progression: (i) pathologic progression, (ii) expression of many important gene signatures, and (iii) alterations in stromal TGF-β signaling. This last point is particularly relevant, as patients with higher stromal TGF-β signaling present with more aggressive disease, which is paralleled in the more aggressive Ubi-Cat × JOCK1 compared with Pro-Cat × JOCK1 or monogenic JOCK1 tumors.

Although the role of FGFR1 in human prostate carcinogenesis is well established, a regulatory role of canonical WNT signaling during therapy resistance in response to treatment-induced wound healing is only beginning to be unraveled, and WNT signaling in the pretreated stroma has not been investigated. To this end, we used the ontogenic analysis tool in DAVID Bioinformatics Resources 6.7 to interrogate the microarray data collected by Dakhova and colleagues, who compared normal human stroma to reactive stroma grade 3 (RSG3, characterized by at least a 1:1 ratio between reactive stroma and cancer epithelium; ref. 35). We discovered a distinct WNT signature, consisting of the upregulation (≥1.2-fold and P < 0.001) of 15 genes implicated in a WNT signaling cascade (Supplementary Fig. S8A; refs 36, 37). Although the WNT signature observed in the CAT × JOCK transgenic animals contains different misregulated genes (Supplementary Fig. S8B), it, nonetheless, validates the importance of stromal WNT signaling in progression of human carcinogenesis. Furthermore, this WNT signature reveals the modulation of several WNT ligands (Wnt7a, Wnt3, and Wnt6) in Pro-Cat × JOCK1 and a dramatic upregulation of Wnt3 in Ubi-Cat × JOCK1, indicating that induced FGF and canonical WNT signaling could induce the production of more WNT ligands, as seen in development, that could act on the stroma needed for cancer initiation (Supplementary Fig. S6; ref. 38).

Tumor-associated epithelial cells are surrounded by supporting cells and matrices, together termed “stroma”. The stroma contains vasculature, smooth muscle, nerves, fibroblasts, and immune cells (33). Following wounding, this diversity further expands in response to various stimuli (e.g., TGF-β, FGF and WNT signaling) to facilitate fibroblast migration, myofibroblast differentiation, epithelial cell proliferation, and homing of inflammatory cells to close the breach and eliminate pathogens (33). These same innate functions are capable of responding to basement membrane breaches caused by neoplastic growth. However, growth stimuli further exacerbate tumorigenicity, eliciting a classic “chicken-or-egg” dilemma, as to which tissue-type evolves first and promotes/recruits the other. This report supports tumor-promoting mutations in the epithelial cells occur first, which subsequently induce reactive stroma growth. In turn, reactive stroma is then required to coevolve with a dysplastic epithelium to drive malignant transformation to adenocarcinoma.

We and others have previously shown the transforming ability of chronic FGFR1 stimulation when expressed in prostate epithelia cells (17, 39). Using the Pro-Cat × JOCK1 and Ubi-Cat × JOCK1 mice, we show that canonical WNT signaling accelerates FGFR1-driven adenocarcinoma, due largely to an accelerated conversion to a fibroblastic reactive stroma. Although the mechanisms whereby WNT accelerates this wound-like reactive stroma are unknown, migratory and metastatic roles for the WNT pathway have been reported. For example, FGF and canonical WNT signaling are highly expressed and result in the osteoblastic phenotype of AR-negative bone metastasis (40). Furthermore, anticancer therapeutics promote canonical WNT signaling in prostate cancer stroma via WNT16B, leading to carcinoma progression and therapeutic resistance (24). As a correlate with the role of WNT signaling, we have identified the decrease of TGF-βRII and an increase in TGF-β ligands in these bigenic tumors as the focus of future study.

Others have shown that the decrease in TGF-βRII and concomitant increase in overall signaling is found in more than 60% of human prostate cancers (41), and is sufficient to induce malignant transformation in prostate reconstitutions and mouse models (42–45). Using known prostate-associated oncogenes, we serendipitously observed a similar increase in TGF-β signaling, which correlates strongly with the coevolving neoplasia (mPIN) and reactive stroma, as a likely mechanism for progression from mPIN to adenocarcinoma. Likewise, in a human tissue array, we observed the same pattern of Smad2 activation. Furthermore, tumors with greater than 60% stromal TGF-β signaling significantly correlated with higher Gleason grades, resembling the signaling pattern of the more aggressive Ubi-Cat × JOCK1 mice. These findings agree with the “two-step computational model”, whereby TGF-β–nonresponsive stromal cells are required for the first (epithelial) transformational step, and TGF-β–responsive cells are needed for the second (invasion) step of prostate cancer progression (46). Interestingly, this model demonstrated that canonical WNT signaling was a vital paracrine factor secreted by the TGF-βRII–deficient stromal cells as a mechanism for epithelial transformation. Although this model does not rule out other explanations, this bipartite TGF-β model could further explain the higher levels of nuclear p-Smad2 staining seen in Ubi-Cat × JOCK1 animals, as enforced transgenic WNT signaling in the stromal compartment could circumvent the initial requirement of TGF-β–nonresponsive cells to secrete Wnt3a for the proliferation step, allowing TGF-β–responsive (p-Smad2⁺) cells to efficiently promote epithelial invasion and accelerated progression.

The observed epithelial phenotype of our double-transgenic mice correlates with progression hallmarks identified in patients, even to the extent that sarcomatoid carcinomas are now accepted as histologic manifestations of EMT (17, 33, 47). The tumor-associated stroma, however, reveals both similarities and distinct differences between mouse and human diseased tissue (Table 1). The reactive stroma seen in these mice “closely resembles the stroma of [the] aggressive human prostate cancer subgroup (RSG3)” (33). Furthermore, these models display a fibroblastic reactive stroma (i.e., high-level collagen and vimentin staining, a decrease in SMA, and focal areas of tenascin c), all of which indicate a fibroblastic reactive stroma. In humans, these fibroblasts, more commonly referred to as cancer-associated fibroblasts (CAF), make up approximately 50% of reactive stroma in Gleason 3 foci and are progressively replaced by myofibroblasts in Gleason 4 to 5 foci (30). In more advanced foci, myofibroblasts are the majority and share both fibroblastic (i.e., tenascin c and vimentin) and smooth muscle markers (i.e., SMA). This difference correlates with the dissimilarities seen in the normal stroma of mice and humans. Normal murine stroma is sparse, with only a thin layer of smooth muscle surrounding the acini, whereas human stroma consists of dense areas of smooth muscle filling the space surrounding the glands. Therefore, it is easy to conceptualize that the characterization of the tumor-associated fibroblasts found in each species will correlate with the characteristics of the normal stroma. In contrast, the predominant myofibroblasts may be derived from CAFs or from another stem/progenitor cell types. Therefore, the lack of myofibroblasts in murine prostate cancer may not be due to the starting material, but a difference in progression and/or development. Despite this difference, these murine models clearly model CAFs, infiltration of immune cells, and upregulation of key signaling pathways, like those initiated by TGF-β, FGFs, and WNTs, which are all present in human cancers.

In conclusion, we have demonstrated the ability of canonical WNT signaling to accelerate FGFR1-driven carcinogenesis, by promoting a fibroblastic reactive stroma, characterized by increased TGF-β heterogeneity. As reactive stroma has now been shown to be predictive of prostate cancer-specific death (32), this work provides a powerful new model to test and validate the importance of reactive stroma in human prostate cancer progression.

No potential conflicts of interest were disclosed.

Conception and design: J.L. Carstens, P. Shahi, S. Tsang, J.M. Levitt, D.R. Rowley, D.M. Spencer

Development of methodology: J.L. Carstens, P. Shahi, B. Smith, A. Seamans, D.R. Rowley, D.M. Spencer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.L. Carstens, P. Shahi, A. Seamans, M. Seethammagari, I. Vedula, M.M. Ittmann, D.R. Rowley

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.L. Carstens, P. Shahi, S. Tsang, C.J. Creighton, Y. Zhang, J.M. Levitt, M.M. Ittmann, D.M. Spencer

Writing, review, and/or revision of the manuscript: J.L. Carstens, C.J. Creighton, J.M. Levitt, M.M. Ittmann, D.R. Rowley, D.M. Spencer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.L. Carstens, P. Shahi, S. Tsang, B. Smith, A. Seamans, J.M. Levitt, M.M. Ittmann, D.R. Rowley, D.M. Spencer

Study supervision: J.L. Carstens, J.M. Levitt, D.R. Rowley, D.M. Spencer

The authors thank W. Decker, K. Chan, J. Rosen, L. Xin, L Donehower, and their laboratory staff for their technical expertise, generous sharing of reagents, and invaluable discussions. AP20187 was generously provided by ARIAD Pharmaceuticals (Cambridge, MA).

This work was financially supported by the mouse models for human cancer consortium (MMHCC) grant U01-CA84296, NIH R01 CA58093; the Cytometry and Cell Sorting Core at Baylor College of Medicine with funding from the NIH (AI036211, CA125123, and RR024574); and the National Cancer Institute P30 Cancer Center support grant (P30 CA125123) for support of the Human Tissue Acquisition and Pathology and Genetically Engineered Mouse Cores.

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.