TMPRSS2 expression is regulated by androgens in vivo

The cellular signaling program regulated by the AR is a primary target of prostate cancer therapies, though the critical AR-regulated effector components responsible for the genesis or progression of prostate cancer have not been defined. In vitro experiments determined that TMPRSS2 expression is promoted through AR and potentially via estrogen receptor regulation (10, 14, 15). In support of these observations, we found a positive correlation between AR and TMPRSS2 in microdissected primary tumor epithelium (r²=0.39; p<0.001). This positive correlation was also evident in metastatic castration-resistant human prostate cancers (CRPC), the majority of which exhibit AR signaling (r²=0.37; p<0.001) (Supplementary Fig. S1A, B).

To further establish the role of AR in TMPRSS2 regulation in vivo, we quantitated TMPRSS2 transcripts in prostate epithelium in the context of pharmacological manipulation of the AR-axis. Men diagnosed with localized prostate adenocarcinoma were treated with the LHRH agonist leuprolide, or doses of estradiol sufficient to induce castrate serum levels of testosterone prior to radical prostatectomy. We isolated epithelial cells by laser-assisted microdissection and quantitated TMPRSS2 mRNAs by qRT-PCR. Compared to epithelium from untreated individuals, TMPRSS2 transcripts were significantly lower in men receiving leuprolide or estradiol (p<0.01) (Fig. 1G). In contrast, TMPRSS2 expression was not altered in tumor cells from men receiving cytotoxic chemotherapy (Supplementary Fig. S1C). We next confirmed the in vivo responsiveness of TMPRSS2 expression in the context of AR activity using the Lu-CaP35 xenograft model that regresses following castration (16). Over time, these xenografts develop a CRPC phenotype accompanied by reactivation of the AR signaling program. Using microdissection to directly compare equivalent cancer cell numbers in the xenografted tumors at different time points following castration, we found that TMPRSS2 is expressed highly in Lu-CaP35 tumor cells grown in eugonadal hosts, decreases at days 5 and 20 post-castration, and increases with the emergence of CRPC (Supplementary Fig. S1D).

TMPRSS2 promotes prostate cancer invasion and metastasis

The extracellular protease function predicted by the TMPRSS2 coding sequence coupled with findings of high TMPRSS2 levels in metastatic prostate cancers suggested a functional role for TMPRSS2 in the pathogenesis of prostate cancer metastasis. To test this hypothesis, we evaluated the contribution of Tmprss2 to the high frequency of metastasis observed in the transgenic autochthonous mouse model for prostate cancer (TRAMP) (17). We crossed TRAMP mice to astrain with a targeted deletion of the Tmprss2 serine protease domain (18) and evaluated cohorts of Tmprss2^+/+; TRAMP and Tmprss2^−/−; TRAMP animals at 20 and 32 weeks of age. The frequency of primary prostate tumor development was not different between the two genotypes: 100% of the Tmprss2^+/+; TRAMP and Tmprss2^−/−; TRAMP animals developed prostate tumors grossly visible at necropsy (Fig. 2A). However, primary tumor sizes and weights in the strains were significantly different; at 32 weeks Tmprss2^+/+; TRAMP tumors averaged 2.65±0.65g while thosearising in Tmprss2^−/−; TRAMP animals were more than twice as large, averaging 6.39±1.24g (p=0.01 for difference) (Fig. 2B). Tumors in both strains were highly proliferative with Ki67 indices of 67% and 72% in Tmprss2^+/+ and Tmprss2^−/− tumors, respectively (Supplementary Fig. S2A). Prostate neoplasms that develop in the TRAMP model are known to exhibit a histological spectrum of growth spanning hyperplasia to poorly differentiated carcinomas (19). At 32 weeks, neoplasms developing in bothTmprss2^−/−; TRAMP (n=17) and Tmprss2^+/+; TRAMP (n=16) strains exhibited a range of histological patterns with a tendency toward more poorly differentiated tumors in the Tmprss2^−/−; TRAMP genotype, though this was not statistically significant (Supplementary Fig. S2B).

Open in a separate window

Figure 2

Tmprss2 influences prostate cancer growth and the development of metastasis

(A) Representative anatomical and histological images of prostate glands from wild type (WT; WT) and TRAMP mice with (Tmprss2^+/+; TRAMP) and without (Tmprss2^−/−; TRAMP) Tmprss2 protease activity. Both TRAMP genotypes exhibited a spectrum of well- to poorly-differentiated carcinomas. P, prostate gland; SV, seminal vesicle. Prostate glands are circled. (B) Weights of the genitourinary tracts (GU-SV) excised from TRAMP mice with (Tmprss2^+/+; TRAMP) or without (Tmprss2^−/−; TRAMP) Tmprss2 activity. At 32 weeks of age, prostate tumors in Tmprss2^−/−; TRAMP mice were substantially larger than those tumors excised from Tmprss2^+/+; TRAMP animals (P=0.01). (C) The frequency of lymph node metastasis in Tmprss2^−/−; TRAMP versus Tmprss2^+/+; TRAMP genotypes was similar. The frequency of metastasis to solid organs in TRAMP mice with inactive Tmprss2 was significantly reduced compared to TRAMP mice with wild-type Tmprss2 (P=0.02). (D) Basement membrane invasion assays demonstrating enhanced invasion of primary cells from Tmprss2^WT; TRAMP tumors relative to benign epithelium or primary tumor cells deficient in Tmprss2 activity (Tmprss2^−/−; TRAMP). (E) Soft-agar colony formation assay demonstrating enhanced anchorage-independent growth of DU145 prostate cancer cells expressing wild-type active human TMPRSS2 (TMPRSS2^WT) versus a vector control or protease-dead TMPRSS2 mutant (TMPRSS2^PM). (F) Schematic of the metastasis assay: primary tumors developing in Tmprss2^−/−; TRAMP or Tmprss2^+/+; TRAMP were resected, cells dissociated, and injected into tail veins of recipient hosts. (G) Detection of S40T antigen by PCR in blood (b), liver (l) or lung (lu) of mice at designated time points following tail-vein injections with primary prostate tumor cells from Tmprss2^+/+; TRAMP or Tmprss2^−/−; TRAMP mice. TCR, T-cell receptor.(H) Representative anatomical and histological images of livers from mice receiving vascular injections of cells from Tmprss2^−/−; TRAMP andTmprss2^+/+; TRAMP tumors.

At necropsy, we evaluated each animal for the presence of metastasis both grossly and by tissue histology. At the 32-week time point, tumor cell dissemination to regional lymph nodes was common in both Tmprss2^+/+ (78%) and Tmprss2^−/− (83%) backgrounds (p=0.6) (Fig. 2C). In contrast, the incidence of metastasis to distant solid organs was substantially lower in the TRAMP tumors arising in the Tmprss2^−/− background. Of 18 Tmprss2^+/+; TRAMP animals examined, 11 had gross metastasis to solid organs such as the liver and lung (61%) while only 2 of 30 Tmprss2^−/−; TRAMP animals had macroscopically visible tumors (7%) with an additional 2 animals having microscopic foci in the liver (p=0.016) (Fig. 2C). The AR and Tmprss2 were expressed in primary tumors of both strains and inmetastaticTmprss2^+/+; TRAMP tumor cells in liver and lung, though there were regions of metastasis with low to absent AR staining (Supplementary Fig. S2C–E).

Due to the germ-line Tmprss2 loss-of-function mutation in the Tmprss2^−/− strain, the elimination of Tmprss2 protease activity is not exclusive to prostate epithelium. To evaluate the possibility that Tmprss2 influences tumor dissemination through non-prostatic cell types, we excised primary tumors, dispersed the tumor cells, ensured cellular viability, and measured growth in semi-solid medium, and invasion through modified basement membrane by transwell assays. Tmprss2^WT; TRAMP tumor cells were substantially more capable of invasion than either benign epithelium or cells from TRAMP tumors lacking Tmprss2 protease (p<0.01) (Fig. 2D). We next introduced shRNAs targeting TMPRSS2 into LNCaP C4-2B cells and noted a significant reduction in cell proliferation and cell invasion compared to scrambled shRNA controls (p<0.001) (Supplementary Fig. S3A–B). Additionally, DU145 prostate cancer cells, a line that normally expresses no TMPRSS2, engineered to express wild-type TMPRSS2 were more capable of anchorage independent growth than vector control DU145 cells or those expressing a protease-dead mutant form of TMPRSS2, TMPRSS2^PM (Fig. 2E).

We next excised primary tumors from Tmprss2^+/+; TRAMP and Tmprss2^−/−; TRAMP strains, dissociated the tumors, and injected viable tumor cells into the tail vein of wild-type recipient mice (Fig. 2F). Using PCR to detect the presence of SV40T antigen as a surrogate for the presence of tumor cells, we assessed blood, liver and lung over a time course of 28 days. SV40T antigen was readily detectable in the blood and tissues of mice injected with Tmprss2^−/−; TRAMP tumor cells up to 30 minutes, and in blood up to 24 hours, after which the SV40T levels diminished markedly and were undetectable by day 14 (Fig. 2G). Non-viable tumor cells were rapidly cleared and undetectable after 4 hours (Supplementary Fig. S4A). In contrast, SV40T DNA was readily measureable in the lung and liver 3 days postTmprss2^+/+; TRAMP tumor cell injection and remained detectable in these tissues until the time of necropsy at day 28 at which time gross metastasis were evident. Primary tumors resected from 5 different Tmprss2^+/+; TRAMP donor animals aged 32 weeks, produced grossly visible liver or lung metastasis in 21 of 21 recipients, while tumor cells derived from Tmprss2^−/−; TRAMP primary tumors failed to generate metastasis in any of 28 recipients (p<0.01) (Table 1 and Fig. 2H).

We next evaluated the capability of Tmprss2 to directly affect the survival and growth of prostate cancer cells in a distant site. The bone is a frequent location of prostate cancer metastasis in humans, though this predilection is poorly recapitulated in any spontaneous mouse model of prostate cancer for reasons that remain unclear. We injected cells directly harvested from 32 week Tmprss2^+/+; TRAMP or Tmprss2^−/−; TRAMP primary tumors into the tibia of recipient hosts. Mild radiographic abnormalities were noted in bone engrafted with Tmprss2^−/−; TRAMP cells, but tumor growth or bone destruction was not evident (Supplementary Fig. S4B). Radiographs of bone engrafted with cells from Tmprss2^+/+; TRAMP tumors demonstrated a marked expansile mass with mixed lytic and blastic features (Supplementary Fig. S4B). Histological examination determined that 21 of 32 (66%) recipients of tumor cells derived from 6 different 32-week Tmprss2^+/+; TRAMP donors had tumors, while none were identified in 29 recipients of tumors cells from 8 different Tmprss2^−/−; TRAMP donors (p<0.001). Two of seven animals with Tmprss2^+/+; TRAMP tibial grafts where distant organs were examined also had liver metastasis. No distant metastasis were identified in mice grafted with cells from Tmprss2^−/−; TRAMP tumors.

Identification of TMPRSS2 substrate specificity by Positional Scanning of Synthetic Combinatorial Peptide Libraries (PS-SCL)

To identify substrates of TMPRSS2 that could contribute to the metastatic phenotype we took an approach designed to establish a hierarchy of likely target proteins based on biochemically-defined cleavage specificity. To assess the primary (P1) and extended (P2–P4) TMPRSS2 cleavage sequence, we screened pools of synthetic peptides each with a structure of Ac-P4-P3-P2-P1-ACC (7-amino-4-carbamoylmethylcoumarin) (20, 21). We tested TMPRSS2 against 160,000 distinct peptides for proteolytic activity in vitro. At the P1 position, TMPRSS2 strongly preferred substrates with an arginine rather than a lysine (Fig. 3A). The P2–P4 specificity of TMPRSS2 is distinct from the optimal cleavage specificities defined for other extracellular serine proteases including hepsin, MT-SP1/matriptase, thrombin, plasmin, tryptases βI and βII, prostasin and KLK4 (Supplementary Fig. S5), but TMPRSS2 also exhibits a degree of overlap with hepsin and MTSP1 substrates (22–25). We next created a database of human sequences that contain a trypsin-like fold as extracted from Pfamand MEROPS databases. Based on the information obtained from the TMPRSS2 PS-SCL peptide profiling, we identified serine protease-fold like proteins that had (i) activation domains comprising a known or predicted P4–P1 sequence, (ii) P1 position comprised of arginine, and (iii) a P1′ position comprising a hydrophobic amino acid (isoleucine or valine) to bind to the large, hydrophobic S1 pocket of TMPRSS2 (Supplementary Fig. S6A–C).

Open in a separate window

Figure 3

Identification of TMPRSS2 protease substrates

(A) Results of positional scanning of synthetic combinatorial peptide libraries (PS-SCL). The amino acid cleavage preferences are: P1 (R); P2 (T, F, W, A, V); P3 (E, M, Q); P4 (G, I, M). (B) pro-HGF is a substrate for TMPRSS2. The arrow denotes the single-chain (pro-) HGF and is not visualized following incubation with active TMPRSS2. The inactive TMPRSS2 protease-dead mutant (TMPRSS2^PM) is incapable of catalyzing HGF proteolysis. TMPRSS2 does not degrade the BSA carrier protein (dominant bands). Western blot analysis using a polyclonal anti-HGF antibody that recognizes multiple epitopes in the α and β chains of HGF demonstrates the loss of single chain (pro) HGF following incubation with active TMPRSS2 (lane 2), but not protease-dead TMPRSS2 (lane 3). (C) pro-HGF proteolyzed by TMPRSS2 activates c-Met as determined by immunodetection of phosphorylated c-Met (p-cMet) in DU145 prostate cancer cells. Control reactions with matriptase also demonstrate c-Met activation. Pro-HGF alone and the addition of anti-HGF neutralizing antibody abolishes c-Met phosphorylation.

Results of TMPRSS2 substrate screening identified the activation sequences PQFR in the zymogen precursors of tissue plasminogen activator (PLAT), and IQSR in human glandular kallikrein 2 (hK2). Of interest, both PLAT and hK2 participate in protease cascades culminating in the anti-coagulation of blood or seminal fluid, respectively (26, 27). We confirmed that TMPRSS2 proteolyzed pro-hK2 and exhibited no activity toward MMP2 and MMP9, proteins without predicted TMPRSS2 cleavage sites (Supplementary Fig. S7A–D). hK2 is produced by benign prostate epithelial cells and is secreted into the glandular lumen where it is known to activate prostate specific antigen (PSA), a protease that in turn lyses the procoagulant seminogelin seminal fluid proteins. These results indicate that TMPRSS2 participates in proteolytic cascades of relevance for the normal physiological function of the prostate.

TMPRSS2 activates hepatocyte growth factor, promotes signaling through c-Met, and engages a gene expression program contributing to metastasis

Results of TMPRSS2 substrate screening identified the peptide KQLR, a sequence identical to the activation sequence of the single chain precursor form of hepatocyte growth factor (HGF). HGF has structural similarity to serine-proteases, and functions as the ligand of the c-Met receptor tyrosine kinase (28). Several studies have shown that HGF and c-Met comprise a signaling pathway of relevance for prostate cancer invasion and metastasis (29, 30). Cleavage of the single chain HGF precursor by known activators such as HGFAC and matriptase, produce a functional α/β-heterodimeric HGF capable of engaging the c-Met receptor (31). To test whether TMPRSS2 could activate HGF we evaluated its proteolytic activity toward a preparation of precursor sc-HGF. Exposure to active TMPRSS2 resulted in the loss of detectable pro-scHGF (Fig. 3B), an observation not recapitulated following incubation with aprotease-inactive TMPRSS2 mutant or vehicle control. We next confirmed that TMPRSS2-processed HGF, HGF_TMPRSS2, activates the c-Met receptor. We exposed DU145 cells, known to express c-Met, with pro-HGF pre-incubated with matriptase, a known activator of HGF. Met activation was determined by demonstrating receptor phosphorylation (Fig. 3C). Exposure of these cells to pro-HGF alone did not result in detectable Met phosphorylation. However, pro-HGF activated by incubation with TMPRSS2 resulted in Met phosphorylation. The inclusion of anti-HGF neutralizing antibody eliminated Met phosphorylation, demonstrating that activation resulted from HGF (Fig. 3C). DU145 cells exposed to HGF proteolyzed by TMPRSS2 were substantially more invasive when compared to cells treated with mock-activated pro-HGF (p<0.01) (Fig. 4A). Invasion was blocked by an HGF neutralizing antibody. The addition of TMPRSS2 to conditioned medium from prostate fibroblasts also significantly increased the invasion of DU145 cells, an effect that was not observed with a TMPRSS2 protease mutant, or the addition of HGF neutralizing antibody or the c-Met inhibitor SU11274 (Fig. 4B). LNCaP and C4-2B cells that natively express TMPRSS2, and PC3 and DU145 cells engineered to express TMPRSS2 exhibited enhanced invasion when exposed to pro-HGF (p<0.01) (Fig. 4C–E and Supplementary Fig. S7E).

Open in a separate window

Figure 4

TMPRSS2-activated HGF consistently promotes invasion but differentially enhances or suppresses proliferation

(A) Exposure of DU145 prostate cancer cells to pro-HGF activated by either TMPRSS2 or matriptase increases cellular invasion. (B) Exposure of DU145 prostate cancer cells to prostate fibroblast conditioned medium and TMPRSS2 increases cellular invasion, which is not observed with an inactive TMPRSS2 protease mutant (TMPRSS2^PM) and abolished with the addition of anti-HGF neutralizing antibody or the cMet inhibitor SU11274. The pro-invasive phenotype associated with TMPRSS2 expression in LNCaP (C) PC3 (D) and DU145 (E) cells is promoted by exogenous paracrine-acting HGF. SF is serum-free medium. (F, G) TMPRSS2-activated HGF enhances the proliferation of DU145 cells and suppresses the proliferation of BPH1 and TRAMP C2 cells (TC2). Cell numbers were determined at 96 hours and represent a mean of 3 replicates (asterix = p<0.01). (H) Invasion of TRAMP C2 cells is promoted by TMPRSS2-activated HGF and attenuated with anti-HGF neutralizing antibody (asterix = p<0.01). Differential effects of activated HGF on cell cycle regulators p21 and p27 in TC2 (I) and DU145 (J) cells.

The unanticipated finding that primary tumors developing in Tmprss2^−/−; TRAMP mice were substantially larger than those arising in Tmprss2^+/+; TRAMP mice, yet possessed attenuated metastatic capabilities, prompted further studies to evaluate the role of HGF in this process. When exposed to HGF, DU145 cells increased rates of proliferation (p<0.01) (Fig. 4F). However, HGF suppressed the proliferation of TC2 cells, a line derived from a TRAMP tumor (p<0.03) (Fig. 4G), while still promoting invasion (Fig. 4H). Similarly, the proliferation of non-tumorigenic BPH1 cells was suppressed by HGF exposure (Fig. 4G). Consistent with these observations, TC2 cells exposed to HGF upregulated the expression of cell cycle inhibitors p21 and p27 with concomitant decreases in proliferating nuclear antigen (PCNA), whereas PCNA expression increased in DU145 cells (Fig. 4I, J). These studies demonstrate that HGF signaling can simultaneously suppress cell proliferation while enhancing invasive phenotypes.

As HGF signaling is known to promote a pro-metastatic epithelial to mesenchymal transition (EMT) (32, 33), we examined the expression of genes contributing to such a phenotype. We first confirmed that Tmprss2 and c-Met were expressed in primary and metastatic Tmprss2^+/+; TRAMP tumors (Fig. 5A). We next microdissected benign prostate epithelium from wild type animals, and neoplastic prostate epithelium from age-matched Tmprss2^+/+; TRAMP and Tmprss2^−/−; TRAMP strains. Microarray-based transcript profiling determined that genes associated with EMT are expressed at substantially higher levels in Tmprss2^+/+; TRAMP tumor cells relative to Tmprss2^−/−; TRAMP cells or benign epithelium (Fig. 5B, C). We confirmed increases in the mesenchymal marker vimentin (2.1-fold), and EMT-associated transcripts encoding Twist1 (5.7-fold) and Twist2 (1.7-fold) by qRT-PCR (Fig. 5D). HGF signaling has been shown to promote invasive phenotypes through transcriptional regulation of CXCL12/CXCR4 chemokine components (34). Tumors excised from the highly metastatic TRAMP strain expressed 15-fold higher levels of Cxcl12α (Fig. 5D), whereas Tmprss2^−/−; TRAMP cells expressed Cxcl12α at levels equivalent to benign epithelium. Immunohistochemical staining confirmed that Tmprss2^WT; TRAMP tumor cells express high levels of the EMT-associated proteins vimentin and N-cadherin, and reductions of both were apparent in tumors arising in the absence of Tmprss2 (Fig. 5E). These findings indicate that Tmprss2 influences an EMT phenotype associated with enhanced cellular motility, invasion, and metastasis (Fig. 5F).

Open in a separate window

Figure 5

TMPRSS2 associates with a gene expression program involved in epithelial to mesenchymal transition (EMT)

(A) AR and c-Met transcript levels in primary and metastatic TRAMP tumors determined by qRT-PCR.(B) Transcript profiling of mRNAs from microdissected prostate epithelium. Genes associated with EMT are elevated in Tmprss2^WT; TRAMP tumors (yellow) and decreased in Tmprss2^−/−; TRAMP tumors (blue). (C) Gene set enrichment analysis (GSEA) confirms a significant enrichment in EMT-associated genes in Tmprss2^WT; TRAMP tumors (p<0.001). (D) qRT-PCR quantitation of gene expression from benign and neoplastic epithelium microdissected from TRAMP mouse strains with and without Tmprss2 activity. (E) Immunohistochemical staining of TRAMP tumors for EMT-associated proteins vimentin and N-cadherin. (F) Schematic view of the proteolytic cascades influenced by TMPRSS2 in the context of normal physiology in benign epithelium and in neoplasia where substrates normally confined to the stroma are available for interactions with the TMPRSS2 protease. SC, secretory epithelial cell; BC, basal epithelial cell; BM, basement membrane; CC, cancerous epithelial cell, EMT, epithelial to mesenchymal transition.

Gene expression assays

Laser assisted microdissection was used to capture prostate epithelia from the ventral and anterior prostatic lobes as previously described (49). Microarray hybridization and qRT-PCR were used to quantitate mRNA abundance as detailed in Supplemental Methods. Microarray data are accessible in GEO using accession GSE58822. Primer sequences and PCR conditions for the various genes analyzed are provided in Supplemental Methods.

Cell culture and phenotypic assays

Cell lines used in these studies were used within 6 passages of original receipt or authenticated by comparing transcript profiles with those generated from original stocks. Additional details are provided in Supplemental Methods. The TMPRSS2 coding sequence was cloned into the pCR2.1-TOPO vector following the manufacturers protocol (Invitrogen, Carlsbad, CA), transfected into prostate cancer cells, and stable cell lines generated under antibiotic selection as detailed in Supplemental Procedures. The effect of TMPRSS2 on prostate cancer cell proliferation was determined by measuring the growth rates of individually selected TMPRSS2 expressing cell line clones in triplicate in 6-well tissue culture plates. Cells were trypsinized, stained with trypan blue and counted every 2 days for 8 days. The number of metabolically active cells was measured every day for 8 days using the Celltiter 96 Cell Proliferation Assay from Promega (Madison, WI). Cell invasion assays were performed using the QCM 96-Well Invasion Assay from Chemicon (Temecula, CA). Anchorage-independent growth differences were assayed following the procedure of Arteaga, et al. (1988).

Metastasis and Xenograft Assays

Primary prostate tumors were removed from Tmprss2^−/−; TRAMP or Tmprss2^WT; TRAMP mice at 28–32 weeks of age and transferred into phenol red free RPMI with 10% fetal bovine serum and antibiotics. Part of each tumor was fixed or frozen for histological analysis. Tumor tissue was minced with scalpels and passed through a 100 μm cell strainer (BD Falcon). Live cell number was estimated by trypan blue exclusion using a hemocytometer. Approximately 5 × 10⁵ live cells were injected into the tail vein of 6–8 week old male ICR SCID mice (Taconic). SCID recipients were euthanized 8 weeks after injection. Intratibial injection of primary tumor cells was performed essentially as previously described (50). Mice were euthanized 8 weeks after injection. Radiographic images were obtained using a Faxitron with Kodak minR2000 film.

TMPRSS2 Substrate Identification

The methods for Positional Scanning of Synthetic Combinatorial peptide Libraries (PS-SCLs) were performed as described previously (20) and specific details are provided in Supplemental Methods. To identify preferred protein substrates for TMPRSS2, we used the information obtained from the PS-SCL peptide profiling and the testing of individual peptide substrates to create a database of human sequences that contain a trypsin-like fold as extracted from Pfamand MEROPS databases. Candidate proteins were clustered and ranked based on matches with the preferred amino acids at the P1-P4 positions. We then tested the ability of TMPRSS2 to cleave selected candidate substrates from each cluster such as the HGF precursor, PLAT and KLK2 using standard in vitro digests, PAGE, and western blotting as described in the Supplemental Methods.

Financial Support: This work was supported by the Prostate Cancer Foundation (EM, PSN) and grants from the NIH (P01CA85859 (JML, CH, TK, SAH, MSM, LDT, CM, EC, NC, IC, PSN), R01CA165573 (IC, PSN), U01CA164188 (PSN), and the Pacific Northwest Prostate Cancer SPORE P50CA97186 (LDT, CM, BM, EM, PSN).

We thank Alex Moreno for administrative assistance, Ruth Etzioni and Roman Gulati for statistical input, Valera Vasioukhin for helpful discussions, Funda Vakar-Lopez and Sue Knoblagh for pathology support, and the Tissue Acquisition Necropsy Team (Robert Vessella, Celestia Higano, Evan Yu, William Ellis, Martine Roudier, Beatrice Knudsen, Jennifer Noteboom, Paul Lange). We thank Norm Greenberg for providing the TRAMP mouse strain. We are also very grateful to the prostate cancer patients participating in these studies. This work was supported by the Prostate Cancer Foundation and grants from the NIH (P01CA85859, R01CA165573, U01CA164188, and the Pacific Northwest Prostate Cancer SPORE P50CA97186).