Expression array analysis of the hepatocyte growth factor invasive program

Abstract

Signaling by human hepatocyte growth factor (hHGF) via its cell surface receptor (MET) drives mitogenesis, motogenesis and morphogenesis in a wide spectrum of target cell types and embryologic, developmental and homeostatic contexts. Oncogenic pathway activation also contributes to tumorigenesis and cancer progression, including tumor angiogenesis and metastasis, in several prevalent malignancies. The HGF gene encodes full-length hHGF and two truncated isoforms known as NK1 and NK2. NK1 induces all three HGF activities at modestly reduced potency, whereas NK2 stimulates only motogenesis and enhances HGF-driven tumor metastasis in transgenic mice. Prior studies have shown that mouse HGF (mHGF) also binds with high affinity to human MET. Here we show that, like NK2, mHGF stimulates cell motility, invasion and spontaneous metastasis of PC3M human prostate adenocarcinoma cells in mice through human MET. To identify target genes and signaling pathways associated with motogenic and metastatic HGF signaling, i.e., the HGF invasive program, gene expression profiling was performed using PC3M cells treated with hHGF, NK2 or mHGF. Results obtained using Ingenuity Pathway Analysis software showed significant overlap with networks and pathways involved in cell movement and metastasis. Interrogating The Cancer Genome Atlas project also identified a subset of 23 gene expression changes in PC3M with a strong tendency for co-occurrence in prostate cancer patients that were associated with significantly decreased disease-free survival.

Introduction

Signaling by hepatocyte growth factor (HGF) via the MET receptor kinase is critical for normal development and adult homeostasis [1, 2]. Aberrant pathway activation also contributes to tumor growth, angiogenesis and metastasis in many cancers [3, 4]. HGF activity is pleiotropic: it supports cell survival, drives cell cycle progression and cell proliferation, and stimulates cell motility and morphogenesis in a wide spectrum of cell types [4, 5].

The most abundant form of active HGF is a plasminogen-like protein comprised of an amino-terminal heavy chain with a heparan sulfate proteoglycan (HSP) binding domain (N) followed by four kringle (K) motifs, and a carboxyl-terminal light chain with a serine protease-like domain [5]. The human HGF gene encodes five distinct protein isoforms [5]. Transcript variants 1 (NCBI accession NM_000601) and 3 (NM_001010932) encode full-length forms (NP_000592 and NP_001010932.1) consisting of 728 and 723 amino acids, respectively; the latter lacks the sequence FFPSS which is present at positions 162–166 in K1 of isoform 1. Variants 2, 4 and 5 (NM_001010931, NM_001010933 and NM_001010934) encode truncated isoforms that lack multiple 3′ exons present in the full-length transcripts. Isoforms 2 and 4 (NP_001010931 and NP_001010933), also known as NK2 [5], are identical to isoforms 1 and 3 from N through K2, where they end with a short carboxyl terminal sequence encoded by a unique alternative exon; their total lengths are 290 and 285 amino acids, respectively. Isoform 5, also known as NK1, is 210 amino acids in length with a sequence identical to isoform 1 through K1 but ending thereafter with a short, unique carboxyl terminal sequence [5].

Consistent with functional studies [6-9], structural analyses of NK1 locate the primary high affinity MET-binding site within K1 [10-12]. A second lower-affinity MET-binding site, present only in full-length HGF isoforms, is located in the amino-terminal region of the serine-protease-like domain [13-16]. Although the truncated HGF isoforms bind to MET and HSP, they differ from full-length HGF in biologic activity. Fike full-length HGF, NK1 stimulates mitogenesis, motogenesis and morphogenesis, although at reduced potency and with greater HSP dependence [9, 17-20]. In contrast, NK2 can competitively antagonize mitogenicity stimulated by HGF or NK1 [6, 18, 19], but retains potent motogenic and invasive activity in vitro [18, 19] and strongly promotes HGF-driven tumor metastasis in vivo [21], activating the MET kinase and many of the same intracellular signaling pathways activated by HGF and NK1 [22].

Although human HGF (hHGF) and MET have become highly sought targets in cancer drug and diagnostic development, preclinical studies have been hampered by the fact that murine HGF (mHGF) fails to stimulate the growth of human tumor xenografts, effectively restricting such studies to human tumor cell lines that possess autocrine pathway activation or other more rare MET activation mechanisms. This situation prompted the development of hHGF transgenic [23] and knock-in mice [24] so that preclinical studies in mice could more fully encompass the spectrum of pathway involvement in human cancers. Interestingly, although mHGF does not stimulate the proliferation of human cells, it binds human MET with high affinity [25]. Moreover, we found that the MET-selective competitive antagonist NK1 3S [20], which binds similarly to human MET or murine Met, was equally effective in blocking metastasis in mice by murine B16 melanoma cells or human PC3M prostate adenocarcinoma cells, implying that murine HGF was a critical driver of metastasis in both cell models. Together these findings suggest that mHGF activates human MET with an outcome similar to NK2 in promoting cell motility and tumor metastasis, but not proliferation.

We hypothesized that the functional redundancy of mHGF and hNK2 would strengthen a global approach to characterize the HGF invasive program by expression array profiling and pathway analysis. Such an analysis of gene modulation events common to hHGF, mHGF and hNK2, all of which promote motogenesis, identified canonical pathways associated with enhanced cell motility and tumor cell invasiveness, non-canonical networks linked to cancer and inflammation, and a subset of genes modulated similarly in The Cancer Genome Atlas (TCGA) human prostate tumor sample set that were associated with decreased disease-free survival. Ongoing and planned analyses to define HGF-driven cell survival, cell proliferation and morphogenic programs will be described in future reports.

Materials and methods

Reagents and cell culture

Purified HGF and MET recombinant proteins, and antibodies against MET and HGF, were from R&D Systems; specific anti-phospho-receptor antibodies or 4G10 were from Millipore. The cell lines PC3M, PC3M-luc, B5/589, U87 MG, SK-FMS-1, 786-0, UOK161, UOK109, UOK122, UOK150, J82, RT-4 and SW780 were cultured as described [6, 7, 9, 17, 18, 20, 22]. NK1 3S protein was recombinantly expressed and purified as described [20].

Electrochemiluminescent two-site immunoassays

MET and HGF content in cell lysates, tissue extracts and conditioned media was determined using 2-site electrochemiluminescent immunoassays developed for use with the Meso Scale Discovery (MSD) SectorImager 2400 plate reader as described previously [20]; other assays were from MSD. MET activation in cell lysates or tumor tissue extracts included parallel detection with anti-receptor antibodies and specific anti-phospho-receptor antibodies or 4G10 (Millipore).

MET autophosphorylation in cell lysates or tumor tissue extracts were measured similarly but included parallel detection with anti-receptor antibodies and specific anti-phospho-receptor antibodies and/or 4G10 (Millipore). Cell lysates were prepared as for MET content assays. Tumor tissue non-ionic detergent extracts were prepared by physical disruption in a Mini-BeadBeater-8 (Glen Mills Inc.) and cleared prior to immunoassay.

Cell motility, invasion and mitogenesis assays

Cell motility and invasion was measured using TransWell chambers as described [20]; invasion assays were performed with Matrigel coated barriers, whereas migration assays used uncoated barriers. PC3M proliferation in quadruplicate wells of 6-well plates was measured by counting in a Cellometer (Nexcelom Bioscience). Differences between mean values were determined by t test using GraphPad Prism v5.0.

Tumor xenograft and metastasis assays

Experiments with mice were performed in accordance with NIH Guidelines for Care and Use of Laboratory Animals using protocols approved by the Institutional Animal Care and Use Committee of the Center for Cancer Research, National Cancer Institute. PC3M-luc cells were injected subcutaneously into SCID/Beige mice (Taconic, Inc.; 3 × 10⁶ cells per animal, n = 10/group) and tumor volumes were calculated from caliper measurements as described [20]. Mice were treated by intraperitoneal injection of vehicle or NK1 3S protein at 5 or 25 ug/kg after tumors became palpable (~day 5); metastatic burden was determined by luciferase imaging at study termination.

mRNA profiling arrays

PC3M cells were serum-deprived for 16 h, after which they were treated with recombinant hHGF, hNK1 or mNK1 at 37 °C or left untreated. Cells were harvested at the indicated times after treatment (8, 16 and 32 h) by lysis with cold non-ionic detergent containing buffer for 5 min on ice. Cell pellets were collected by centrifugation at 10,000 rpm for 10 min at 4 °C and stored at −80 °C prior to RNA extraction. A total of 36 RNA samples (derived from three MET ligands plus no treatment control at three time points and three replicate cultures for each sample) were analyzed using Affymetrix HGU 133A Plus 2.0 GeneChips™ to assess gene expression. Percent present call was 50–60 % for all samples.

Total RNA was extracted from cell pellets using Qiagen’s All Prep DNA/RNA Kit™ (Qiagen) following the manufacturer’s instructions. The resulting RNA was quantitated by Nanodrop spectrophotometry and stored at −80 °C until use. Total RNA (50 ng) was amplified using the WT-Ovation FFPE System™ (NuGEN) according to the manufacturer’s instructions. Quantity and quality of the resulting amplified cDNA were measured by Nanodrop spectrophotometry. A total of 4 micrograms of the amplified cDNA was fragmented and biotin labeled using FL-Ovation™ cDNA Biotin Module V2 (NuGEN #4200) according to the manufacturer’s instructions. Labeled cDNAs were hybridized to HG U133A Plus 2.0 Affymetrix GeneChip™ arrays, stained with streptavidin phycoerythrin with antibody amplification, and scanned following manufacturer’s instructions.

Microarray data were analyzed using Partek Genomic Suite software v6.6. Thirty-six CEL files were loaded into Partek and normalized by the RAM (robust multichip average) algorithm. Pairwise comparisons were made between each of the MET ligand (hHGF, hNK2 and mHGF) treated and untreated control RNA samples at matching time points (8, 16 and 32 h) by t-test. Probesets with gene expression changes greater than 2-fold (increase or decrease in the comparison of treatment versus control) and significance of change in p value <0.05 with a false discovery rate (FDR, Benjamini and Hochberg) <0.05 were selected. The probesets selected from three time points for each treatment were combined and the redundant probesets, as well as probesets where opposite direction of change occurred over time (3 genes), were filtered out. The resulting probeset list was used for further study by Ingenuity Pathway Analysis software (described below); initial IPA processing produced the gene list in Table S1.

Ingenuity Pathway Analysis (IPA) of expression array data and comparisons to TCGA data

Of 520 probes with ≥2-fold change over untreated control cells (FDR < 0.05) representing the union of hHGF, mHGF and hNK2 treatments combined with those shared by mHGF and hNK2 treatments, 438 were mapped (82 unmapped) during IPA pre-processing, 363 of which were identified as “analysis ready” and processed by IPA Core Analysis (version: IPA Fall Release, 2014) using the following settings: Reference set = Affymetrix Human Genome U133 Plus 2.0 Array; Relationship to include: Direct and Indirect; Does not Include Endogenous Chemicals; Data Sources = All; Species = All; Tissues and Cell Lines = All; Mutation = All; Filter Summary: Consider only relationships where confidence = Experimentally Observed; Cutoff: −2.0 to 2.0 fold change. For the purpose of IPA analysis, all expression changes were nominally reduced to +2- or −2-fold. P values for overlap between PC3M gene expression changes and IPA molecule groups were derived using the right-tailed Fisher’s exact test or B-H multiple test correction as noted.

The PC3M expression dataset (Supplemental Table S1) was compared with expression data available for the TCGA prostate cancer patient cohort described by Taylor et al. [26] using the cBioPortal [27]. cBioPortal settings were as follows: Select Cancer Study = “Prostate Adenocarcinoma (MSKCC, Cancer Cell 2010)”; Select Genomic Profiles = “mRNA Expression data”, subprofile = “mRNA Expression Z-scores versus normals”, Z-score threshold = “2.0”; Select Patient/Case Set = “All Complete Tumors (85)”. In the Enter Gene Set window, the pull-down menu was set to “User-defined List”. Preliminary filtering was performed by entering gene symbols from Table S1 into the Enter Gene Set window entry box in groups of 50 and examining the resulting OncoPrint tab to those eliminate genes for which no instances of expression change in the TCGA set existed, or for which no instances of expression change in the TCGA set were concordant with the direction of change observed for the same gene in PC3M. Genes for which instances of concordant change were observed (150/318, 47 %) were subjected to secondary analysis by entering symbols individually in the Enter Gene Set window entry box window entry box with the search qualifiers “:EXP > 2” or “:EXP < −2” after the gene symbol (e.g. “PTPRM:EXP < −2”; cBioportal Onco Query Language: http://www.cbioportal.org/onco_query_lang_desc.jsp) to limit the results to those cases for which concordant change of 2-fold increase or 2-fold decrease, respectively, was observed. The results were examined under the Survival tab for significant impact on disease-free survival (DFS) defined as having a Logrank Test P-value <0.05. DFS in this TCGA study was defined as time to biochemical recurrence, i.e. PSA ≥ 0.2 ng/ml on two occasions [26].

Results and discussion

PC3M cells display robust paracrine HGF/MET signaling

We analyzed cultured cell supernatants and cell lysates from PC3M, the normal mammary epithelial cell line B5/589, the glioblastoma derived cell line U87 MG and the leiomyosarcoma derived cell line SK-LMS1 for HGF production using a two site electrochemiluminescent immunoassay. Both U87 MG and SK-LMS1 cells secreted hHGF into the culture medium at levels that were readily detectable and consistent with prior reports [17, 28], whereas for both PC3M and the negative control B5/589 cells [29] no HGF was detected (assay detection limit: 50 fg/25 ul or 2 pg/ml; Fig. 1a). HGF protein was similarly undetectable in cell lysates prepared from these lines (data not shown). A panel of 10 cell lines, including B5/589, PC3M, renal cell carcinoma derived cell lines 786-0, UOK161, UOK109, UOK122, UOK150 and bladder carcinoma derived cell lines J82, RT-4 and SW780 were analyzed for MET content by two site electrochemiluminescent immunoassay (Fig. 1b). PC3M cells contained ~90 ng MET per mg total cell protein, comparable to that of most of the genitourinary cancer derived cell lines in this panel (Fig. 1b). These results suggest that PC3M cells have the potential for robust paracrine response to HGF.

Fig. 1.

Paracrine HGF signaling in PC3M cells. a Secreted HGF protein content (ng/mg total cell protein) in media conditioned for 24 h by PC3M, B5/589, U87 MG and SK-LMS-1 cells. Values represent mean ± SD of triplicate samples. b MET protein content (ng/mg total cell protein) in a normal mammary epithelial cell line (unfilled bar), PC3M cells (black bar), renal cell carcinoma derived cell lines (light gray bars) and cell lines derived from urothelial carcinoma of the bladder (dark gray bars). Values represent mean ± standard deviation (SD) of triplicate samples. c PC3M migration (fold mean stained area ± SD vs controls from triplicate samples) over 24 h by untreated cells or cell treated with hHGF, mHGF (each 1 nM) or hNK2 (30 nM). d PC3M invasion (fold mean stained area ± SD vs controls from triplicate samples) over 24 h by untreated cells or cell treated with hHGF, mHGF (each 1 nM) or hNK2 (30 nM). e MET tyrosyl autophosphorylation (mean signal intensity ± SD normalized to MET content, ng/mg total protein, from triplicate samples) in PC3M cells treated with hHGF, mHGF (each 1 nM) or hNK2 (1 and 30 nM) for 20 min. f PC3M growth in culture (mean cell number ± SD from triplicate samples) in the absence (circles) or presence of hHGF (1 nM, squares), mHGF (1 nM, triangles) or hNK2 (30 nM, inverted triangles) measured at seeding and days 3 and 6

Both hHGF and mHGF bind comparably to human MET or murine Met ectodomain-Ig fusion proteins [25], and mice in which the murine HGF gene has been replaced with the human HGF gene are phenotypically normal [24], implying that hHGF fully activates the murine Met pathway. However, this relationship is not reciprocal: mHGF fails to adequately replicate hHGF/MET signaling in human tumor xenografts [23]. Consistent with those observations, we found that PC3M cells stimulated with hHGF, mHGF (each 1 nM) or hNK2 (30 nM) responded with significantly increased migration and invasion across semi-permeable membrane barriers (Fig. 1c, d), which was associated with robust MET tyrosyl autophosphorylation as measured by two site electrochemiluminescent immunoassay (Fig. 1e). However, of these three ligands, only hHGF stimulated significantly increased cell proliferation over untreated controls (Fig. 1f). Thus mHGF resembles hNK2 in driving motility, but not proliferation, through hMET.

Spontaneous metastasis by PC3M cell xenografts is MET mediated

Subcutaneous implantation of PC3M cells into immunodeficient mice leads to primary tumor formation and efficient metastasis to major internal organs [30]. We previously reported the development of a potent and selective competitive antagonist of HGF-stimulated MET activation, NK1 3S [20]. NK1 3S has three amino acid substitutions at primary HSP binding residues in the N domain and a normal K1 (receptor binding) domain, and acts by binding MET and disrupting ternary complex formation between ligand, receptor and HSP that is required for receptor kinase activation. Because it was engineered from human NK1, and hHGF fully substitutes for mHGF, NK1 3S can competitively antagonize signaling by human MET or murine Met [20]. In intact PC3M cells, NK1 3S significantly reduced basal MET autophosphorylation (Fig. 2a) and significantly decreased the level of autophosphorylation associated with maximal hHGF stimulation (6-fold higher than basal MET autophosphorylation; Fig. 2b). Consistent with our prior findings, NK1 3S administered by intraperitoneal injection every other day significantly suppressed metastasis in mice implanted subcutaneously with PC3M-luc cells (Fig. 2c), but had no effect on primary tumor growth (Fig. 2d). Because antagonism by NK1 3S is MET pathway selective and because mHGF was the only HGF ortholog present in that system, these results imply that mHGF was a critical driver of metastasis, but not proliferation, by human PC3M cells. In summary, our results indicate that, like human NK2, mHGF binds to and activates hMET, thereby enhancing PC3M cell motility, matrix invasion and tumor metastasis, but not primary tumor growth.

Fig. 2.

A selective HGF antagonist inhibits MET activation and spontaneous metastasis in PC3M cells implanted into SCID/Beige mice. a MET autophosphorylation level (mean signal intensity ± SD normalized to MET content, ng/mg total protein, from triplicate samples) in resting, 24 h serum deprived PC3M cells (“0”) or cells treated NK1 3S (100 nM). b Dose-dependent inhibition of MET autophosphorylation (% maximum) in hHGF treated (1 nM) PC3M cells by NK1 3S. Values are mean ± SD of triplicate measurements. c Mean metastatic burden (total photon flux ± SD on day 28 post-implantation) in mice (n = 10/group) implanted subcutaneously with PC3M-luc cells and treated on day 5 and daily thereafter with NK1 3S protein by intraperitoneal injection at 5 (medium gray bar) or 25 (dark gray bar) mg/kg or treated with vehicle alone (light gray bar). d Time course of primary tumor growth (mean tumor volume, n = 10/group) PC3M-luc cells implanted subcutaneously into SCID/Beige mice treated as described for panel C

Analysis of HGF motogenic signaling by gene expression profiling and Ingenuity Pathway Analysis

To distinguish and characterize the HGF invasive program by gene expression profiling, a set of 36 gene expression arrays (triplicate cell culture samples of 3 time points for 4 groups) were analyzed for untreated PC3M and cells treated with hHGF (1 nM), mHGF (1 nM) or hNK2 (30 nM) for 8, 16 and 32 h. Microarray data has been deposited at GEO (accession GSE71404). Gene expression changes (≥2-fold, FDR < 0.05, union of three time points) associated with treatment by each ligand displayed a balance of unique and overlapping expression changes that was critical to the feasibility of our approach (Supplement Fig. S1). All three ligands induce motility and invasion, therefore expression changes shared by all three (393 probe sets, Fig. S1C) were deemed most likely to include events related to PC3M cell motility, invasiveness and metastasis. In light of evidence that both NK2 and mHGF are independently capable of driving tumor metastasis in mouse models [20, 21], this set was expanded to include events shared by these ligands (127 probe sets), yielding a total of 520 probe sets representing 345 unique known genes (183 upregulated and 162 downregulated; Supplement Table S1).

Of the 520 probes defined above, preprocessing for Ingenuity Pathway Analysis yielded 438 mapped IDs and 82 unmapped IDs (see Methods for parameter settings) and identified the “Associated Network Functions” and “Top Diseases and Bio Functions” summarized in Table 1. The former included those associated with cell movement and cancer, and cancer was also the top disease identified among the latter. “Cellular Movement” was the top “Molecular and Cellular Function” identified (P-values: 7.35E–6 to 1.95E–02, right-tailed Fisher’s exact test; 4.24E–02 to 1.41E–01, B-H correction), which contained functionally annotated subsets with highest significant overlap (P-values: 7.35E–6 to 8.92E–04, right-tailed Fisher’s exact test; 6.43E–02 to 8.88E–02, B-H correction) related to migration and invasion by cancer cells. The 72 genes comprising all of these cell movement networks are listed in Supplemental Table S2. “Physiological System Development and Function” subcategories with significant overlap included “Embryonic Development”, “Tissue Morphology”, and “Skeletal and Muscular System Development and Function” (P-values: 8.56E–05 to 1.95E–02, right-tailed Fisher’s exact test; 5.24E–02 to 1.41E–01, B-H correction), each of which encompasses well established aspects of HGF biology where motogenesis is a critical component [31]. Restricting the analysis to the 393 probe sets shared by all three ligands provided the same top cell motility related networks with modestly reduced p-values (2.86E–06 to 1.43E–02, right-tailed Fisher’s exact test). The top 4 cell movement networks specifically related to cancer cells (a total of 32 genes, P-values: 7.35E–06 to 8.92E–04, right-tailed Fisher’s exact test; 4.24E–02 to 8.88E–02, B-H correction; Table 2) were further interrogated by literature search. For the majority of these genes (69 %), the direction of expression change we observed was concordant with a predicted or functional role in increasing motility and/or metastasis. To relate these expression changes to cell motility, prostate cancer and metastasis, they were organized into the following 6 functional categories, which are depicted schematically in Supplement Figure S2.

Table 1.

Ingenuity Pathway Analysis summary: 520 motility-associated probe IDs up- or downregulated ≥2-fold relative to median by HGF family members

Associated network functions	IPA scorea	# Molecules
Cell death and survival, cellular compromise, neurological disease	29	20
Lipid metabolism, molecular transport, small molecule biochemistry	29	20
Connective tissue development and function, tissue morphology, organismal injury and abnormalities	27	19
Cell-to-cell signaling and interaction, cancer, cellular movement	27	19
Cellular assembly and organization, cellular compromise, hereditary disorder	25	18

Top diseases and bio functions	p value rangeb	p value rangec	# Molecules
Diseases and disorders
Cancer	2.97E–05 to 1.95E–02	4.85E–02 to 1.41E–01	212
Endocrine system disorders	2.97E–05 to 9.75E–03	4.85E–02 to 1.41E–01	23
Gastrointestinal disease	2.97E–05 to 1.68E–02	4.85E–02 to 1.41E–01	141
Molecular and cellular functions
Cellular movement	7.35E–06 to 1.95E–02	4.24E–02 to 1.41E–01	72
Component functional annotationsd
Migration of tumor cells	3.50E–04	6.43E–02	14
Invasion of cells	3.90E–04	6.43E–02	34
Migration of tumor cell lines	7.08E–04	8.01E–02	28
Cell movement of cancer cells	8.92E–04	8.88E–02	10
Cell-to-cell signaling and interaction	2.86E–04 to 1.95E–02	6.43E–02 to 1.41E–01	49
Cell death and survival	3.79E–04 to 1.95E–02	6.43E–02 to 1.41E–01	116
Physiological system development and function
Embryonic development	8.56E–05 to 1.95E–02	4.85E–02 to 1.41E–01	51
Tissue morphology	9.06E–05 to 1.95E–02	4.24E–02 to 1.41E–01	70
Skeletal and muscular system development and function	1.41E–04 to 1.95E–02	5.24E–02 to 1.41E–01	49

^aIPA Score is generated using a right-tailed Fisher’s exact test and represents the absolute value of the exponent to which base 10 is raised in the resulting p value, e.g. an IPA Score of 29 indicates a p value of 10E-29

^bRight-tailed Fisher’s exact test

^cBenjamini-Hochberg multiple test correction

^dFunctional annotations containing 2-4 molecules have been omitted for brevity; their constituents were also present in the annotations shown

Table 2.

Constituents of ingenuity pathway analysis top 32 cell movement networks

Symbol	Entrez gene name	Subcellular location	Molecular function
IGFBP3	Insulin-like growth factor binding protein 3	Secreted	Protein binding
CXCL8	Chemokine (C-X-C motif) ligand 8	Secreted	Cytokine
CXCL11	Chemokine (C-X-C motif) ligand 11	Secreted	Cytokine
CXCL1	Chemokine (C-X-C motif) ligand 1 (melanoma growth stimulating activity, alpha)	Secreted	Cytokine
CXCR4	Chemokine (C-X-C motif) receptor 4	Plasma membrane	G-Protein coupled receptor
IFNAR1	Interferon (alpha, beta and omega) receptor 1	Plasma membrane	Transmembrane receptor
SOCS3	Suppressor of cytokine signaling 3	Cytosol	Protein binding
DDX58	DEAD (Asp-Glu-Ala-Asp) box polypeptide 58	Cytoplasm	Enzyme: RNA helicase
PTP4A3	Protein tyrosine phosphatase type IVA, member 3	Plasma membrane	Phosphatase
DNAJB6	DnaJ (Hsp40) homolog, subfamily B, member 6	Cytoplasm, Nucleus	Protein binding, DNA binding
RNF41	Ring finger protein 41, E3 ubiquitin protein ligase	Cytosol	Enzyme: ubiquitin ligase
ASNS	Asparagine synthetase (glutamine-hydrolyzing)	Cytosol	Enzyme: ligase
SCNN1G	Sodium channel, non-voltage-gated 1, gamma subunit	Plasma membrane	Ion channel
NOV	Nephroblastoma overexpressed	Secreted	Matrix component, growth factor
FN1	Fibronectin 1	Secreted	Matrix component, protein binding
MMP1	Matrix metallopeptidase 1 (interstitial collagenase)	Secreted	Peptidase
MMP13	Matrix metallopeptidase 13 (collagenase 3)	Secreted	Peptidase
SERPINE1	Serpin peptidase inhibitor, clade E (nexin, plasminogen activator inhibitor type 1), member 1	Secreted	Serine protease inhibitor
TFPI	Tissue factor pathway inhibitor (lipoprotein-associated coagulation inhibitor)	Secreted or microsome membrane	Serine protease inhibitor
CDH3	Cadherin 3, type 1, P-cadherin (placental)	Plasma membrane	Cell adhesion
AKAP12	A kinase (PRKA) anchor protein 12	Cytoplasm	Protein binding
PTPRM	Protein tyrosine phosphatase, receptor type, M	Plasma membrane	Phosphatase
NEDD9	Neural precursor cell expressed, developmentally down-regulated 9	Cytoplasm, nucleus	Protein binding
CTTN	Cortactin	Cytoplasm, cytoskeleton	Protein binding
S100P	S100 calcium binding protein P	Cytoplasm	Protein binding
RAP1A	RAP1A, member of RAS oncogene family	Cytoplasm, lipid anchor	Enzyme: GTPase
TPM3	Tropomyosin 3	Cytoplasm, cytoskeleton	Protein binding: actin
ARRB1	Arrestin, beta 1	Cytoplasm, nucleus, cell membrane	Protein binding
RUNX2	Runt-related transcription factor 2	Nucleus	Transcription regulator
SNAI2	Snail family zinc finger 2	Nucleus	Transcription regulator
ATF3	Activating transcription factor 3	Nucleus	Transcription regulator
DIDO1	Death inducer-obliterator 1	Cytoplasm, Nucleus	Poly(A) RNA binding

Upregulated genes are shown in italics, downregulated in bold; order in table is as presented in “Results and discussion“ section

Cytokine, stress, and growth factor signaling (IGFBP3, CXCL8, CXCL11, CXCL1, CXCR4, IFNAR1, SOCS3, DDX58, PTP4A3, and DNAJB6)

This group can be further subdivided into secreted (IGF1BP3, CXCL8, CXCL11 and CXCL1), membrane (CXCR4 and IFNAR1), and intracellular effector (SOCS3, DDX58, PTP4A3 and DNAJB6) molecules. The downregulation of IGFBP3 we observed is consistent with reported roles as a suppressor of cell migration, tumor progression and metastasis for prostate, colorectal, gastric and endometrial cancers [32-36]. The type 1 insulin-like growth factor receptor (IGF1R) is overexpressed in prostate cancer (PCa; [37]) and its function in the PC3M parental cell line, PC3, is well documented, suggesting that HGF-induced IGF1BP3 downregulation in PC3M would most likely increase IGF1R responsiveness to available IGF1, and in turn enhance cell survival, proliferation and motility. Chemokine (C-X-C motif) ligand (CXCF) 8 is generally pro-angiogenic and pro-inflammatory; the roles of several cytokines in PCa and cancer progression generally is complex (reviewed in [38]), thus the impact of HGF-induced CXCL8 downregulation on cell invasion or metastasis by PC3M is unclear. CXCL11 can be angiostatic [39, 40], thus HGF-induced CXCL11 suppression could be pro-metastatic, though that would be at odds with reports that plasma levels of its product are increased in advanced PCa patients relative to controls [41]. In contrast, HGF-induction of CXCL1 is completely consistent with reports that its expression is increased in high grade PCa [42], and that its product increases PCa cell migration and invasion [43], as well as PCa metastasis and chemoresistance [44, 45].

HGF-induction of CXCR4 is consistent with reports that its product promotes metastasis in PCa [46, 47] and other cancers where exogenous CXCL12 is available [48]. Similar to the multifaceted role of many inflammatory cytokines in cancer. Interferon alpha receptor 1 (IFNAR1) can mediate inflammation associated with PCa progression [49], but can also enable increased infiltration of T- and NK cells so as to prevent tumor de-differentiation and progression [50], suggesting that HGF-induction of IFNAR1 will have context dependent effects on cell migration and/or tumor metastasis.

Suppressor of cytokine signaling 3 (SOCS3) inhibits PCa cell growth and migration mediated by the JAK/STAT and MAPK pathways [51, 52], and becomes an even more critical negative regulator when PTEN and p53 are inactivated [53], as they are in PC3M. SOCS3 downregulation in concert with CXCL1 and CXCR4 induction and IGFBP3 suppression could produce additive or synergistic motogenic signaling. Decreased SOCS3 expression also identifies a subset of PCa patients with more aggressive disease [54, 55] and has a pro-metastatic role in other cancers [56, 57]. DDX58 encodes an interferon-inducible mediator of inflammatory and antiviral responses associated with the NFkB and Akt pathways, and can activate or inhibit tumor cell proliferation in a context dependent manner [58, 59]; the impact of HGF induction of DDX58 on PC3M invasiveness is unclear. Although PTP4A3 was downregulated in our dataset, mRNA abundance and copy number alterations in PCa patients suggest that enhanced expression is related to PCa aggressiveness [60]. PTP4A3 also enhanced cell migration and tumor metastasis in (non-prostate) model systems [61]. A mass-spectrometry based interrogation of the PTP4A3 phosphoproteome revealed PTP4A3-dependent activation of an extensive phosphotyrosyl signaling network normally governed by ligand-activated transmembrane growth factor, cytokine and integrin receptors, with Src in a pivotal role immediately downstream of PTP4A3 [62]. HGF-induced changes in DDX58 and PTP4A3 in concert with changes in other pro-inflammatory and innate immune system mediators in this data set, while not cohesively predictive of invasive activity, identify pathways likely to be important in regulating HGF/MET driven metastasis. Consistent with a potential pro-invasive impact of DNAJB6 upregulation in our data set, silencing of the encoded chaperone/stress response protein inhibited tumor cell migration and invasion [63] and enhanced DNAJB6 expression correlated with metastasis and poor prognosis in breast cancer [64].

Receptor trafficking (RNF41)

HGF-induction of the E3 ubiquitin ligase Ring finger protein 41 (RNF41) has intriguing potential effects on receptor trafficking and signaling. RNF41 reportedly negatively regulates ErbB3 expression by the androgen receptor in androgen-dependent, but not androgen-independent, PCa cells [65]. RNF41 also negatively regulates the stability of the ubiquitin isopeptidase USP8, a regulator of JAK2-associated cytokine receptor sorting and processing: elevated RNF41 destabilizes the endosomal-sorting-complexes-required-for-transport (ESCRT)-0 complex (Hrs and signal-transducing-and adapter-molecule (STAM)-1 or STAM2), which results in cargo arriving at sorting endosomes being routed to recycling endosomes rather than to lysosomes [66]. By the same mechanism, USP8 also directs the trafficking and lysosomal degradation of CXCR4 [67], MET and epidermal growth factor receptor [68, 69], implying that its loss consequent to increased RNF41 abundance would prolong and potentially amplify invasion signaling by these receptors.

Metabolism and ion transport (ASNS and SCNN1G)

Asparagine synthetase (ASNS) is a key metabolic regulator of cell stress responses and its frequent suppression in acute lymphocytic leukemia cells renders them sensitive to asparaginase treatment by further starving them of asparagine (reviewed in [70]). Potential roles in the progression of solid tumors remain unclear: ASNS suppression in PCa cells inhibited growth [71], however, consistent with its suppression by HGF ligands, ASNS deficiency in rectal cancers [72] and hepatocellular carcinoma [HCC; 73] was associated with shorter metastasis-free survival and worse outcome; ASNS inhibited also the migration and tumorigenicity of HCC cells [73]. SCNN1G encodes a subunit of sodium transporter; inconsistent with its suppression by HGF ligands, engineered SCNN1G suppression in glioma cells impaired their migration through an undetermined mechanism [74].

Extracellular matrix and cell adhesion (NOV, FN1, MMP1, MMP13, SERPINE1, TFPI, CDH3, AKAP12, PTPRM, and NEDD9)

This group also can be subdivided as secreted (NOV, FN1, MMP1, MMP13, SERPINE1 and TFPI), membrane (CDH3), and intracellular effector (AKAP12, PTPRM, and NEDD9) molecules. Increased NOV expression as seen in our data has been associated with enhanced PCa cell motility, PCa bone metastasis and angiogenesis [75-77], as well as the migration of cells derived from pancreatic cancer and melanoma [78, 79]; direct effects on integrin expression, adhesion and actin cytoskeletal reorganization have been reported [80]. Upregulation of FN1 in response to HGF ligands is consistent with invasive, metastatic and therapy resistance roles for this matrix protein in breast cancer (reviewed in [81]); FN1 suppression in PC3 cells was associated with reduced tumorigenesis [82]. HGF-induced suppression of MMP1 and MMP13 is not consistent with their pro-invasive roles in many cancers, including PCa [83-85]. We note, however, that MMP13 expression is positively regulated by RUNX2, which is downregulated in our dataset, and MMP13 induction downstream of parathyroid hormone required both AP-1 and RUNX2 transcription factors [86, 87]. The impact of SERPINE1 (also known as plasminogen activator inhibitor-1, or PAI-1) expression in the tumor microenvironment is context dependent: higher expression is associated with lower malignancy of some cancers (presumably due to plasminogen inhibition), but with poor prognosis in others, perhaps due to plasminogen independent effects on cell survival and growth (reviewed in [88]). Increased SERPINE1 expression reduced migration by normal prostate cells [89]. Consistent with HGF suppression of TFPI, decreased TFPI expression has been correlated with increased cell motility, invasion and tumor metastasis in breast and pancreatic cancers, glioblastoma, melanoma and nasopharyngeal carcinoma [90-94]. Tissue factor pathway inhibitor (TFPI) also interferes with endothelial cell migration by inhibiting the Erk pathway and focal adhesion proteins [95].

Like E-cadherin, P-cadherin (CDH3) expression promoted cell–cell adhesion and countered invasion in melanoma [96], oral squamous cell carcinoma [97] and breast cancer model systems [98]; HGF ligands suppressed CDH3 in PC3M, suggesting release from this negative regulator consistent with enhanced invasion. Similarly, the adhesion adapter AKAP12 repressed prostate metastasis and cell motility by sequestering Src away from FAK-associated adhesion and signaling complexes [99-101], suggesting that AKAP12 suppression by HGF ligands is likely to promote PC3M invasion and metastasis. HGF-suppression of the protein tyrosine phosphatase encoded by PTPRM, a negative regulator of invasion by glioma [102] and breast cancer [103] cells, and upregulation of the integrin adapter encoded by NEDD9, which promotes epithelial-to-mesenchymal transition and invasion by PCa cells [104, 105], also predict enhancement of PC3M invasiveness.

Cytoskeleton and scaffolds (CTTN, S100P, RAP1A, TPM3 and ARRB1)

Cortactin (CTTN) suppression decreased invasion by a PCa cell line [106], but more basic studies indicate that its role in invasion is complex: it is required for E-cadherin-mediated cell contact formation [107], yet mediates the destabilization of invadopodia structures downstream of the Rac1-Pak1 pathway necessary for tumor cell invasion [108]. The impact of CTTN suppression by HGF ligands is not easily predicted and warrants further scrutiny. Suppression of S100P expression in our dataset is not consistent with most scenarios reported in the literature including prostate cancer, where increased expression is associated with increased cell motility and cancer progression [109, 110]. Although the molecular basis for these effects is not well understood and there is evidence of context dependence, interactions with ezrin, tubulin and the actin cytoskeleton affecting focal adhesions have been described [111, 112]. Overexpression of RAS-related RAP1A (upregulated by HGF ligands in PC3M) is associated with poor prognosis for oral cavity squamous cell carcinoma and promotes tumor cell invasion via Aurora-A modulation [113]. Activation of the Rap-1A GTPase also promotes PCa metastasis [114]. HGF-induction of tropomyosin 3 (TPM3), which regulates the recruitment of myosin motors to actin filaments, is likely to be pro-invasive, and while TPM3 upregulation has been associated with increased tumor metastasis, its impact has context-dependent and isoform-specific dimensions (reviewed in [115]). HGF induction of the cytosolic scaffold beta-arrestin (encoded by ARRB1) in PC3M is consistent with its role in promoting PCa through metabolic reprogramming [116], and in signaling downstream of MAPK cascade activation by CXCR4 (also upregulated in our data set) to drive breast cancer cell motility and invasiveness (reviewed in [117]).

Gene transcription (RUNX2, SNAI2, ATF3 and DIDO1)

Unlike RUNX1 and 3, which are considered tumor suppressors, RUNX2 is a master osteoblast-specific gene regulator, influences androgen receptor (AR) DNA binding, and is implicated in bone metastasis in breast cancer [118] and PCa, where RANKL signaling in PCa cells upregulated both RANKL and MET to promote a metastasis-associated osteomimicry program [119]. RUNX2-AR interactions in PCa cells have intriguing target-specific impact: SNAI2 induction (known to be pro-invasive) required co-activation by RUNX2 and AR, and simultaneous strong immunohistochemical staining for SNAI2, RUNX2 and AR, but not any pair alone, in primary prostate tumor samples was associated with disease recurrence [120, 121]. Since AR is constitutively active in PC3M cells [122], RUNX2 and SNAI2 suppression by MET ligands suggests that a coincident third signal such as RANKL may be required for their induction.

The stress response mediator encoded by ATF3 (suppressed in our data set) displays pro-and anti-invasive activities, depending on cancer type. In PCa, ATF3 binds AR and represses AR signaling [123]; AR signaling has been shown to potentiate Src mediated invasiveness or PCa cells [124]. Since Src is frequently activated downstream of MET [4], the combination of MET-driven Src activation, ATF3 suppression, and constitutive AR activation is likely to be pro-invasive in PC3M. Death inducer-obliterator 1 (DIDO1), upregulated by MET activation, is a histone binding protein that regulates embryonic stem cell self-renewal [125]. Although a role in PCa has not been described, it is also a BMP target gene that promotes BMP-driven melanoma progression [126].

Concordant gene expression changes among PC3M and TCGA datasets associated with significantly altered DFS predict poor outcome

The set of 345 PC3M motility-associated genes (Supplement Table S1) were interrogated for potential relationships to metastasis using PubMed, which directly implicated 130 (38 %) in that process by functional or correlative studies. PC3M motility-associated gene expression changes were also compared for concordance with expression changes reported for the PCa cases compiled by Taylor et al. [26], for which both mRNA expression array data (z-scores vs. normals, ≥2-fold) and disease-free survival (DFS) data (defined as time to biochemical recurrence, i.e. PSA ≥ 0.2 ng/ml on two occasions) were available for 85 cases (all complete tumors). Significant change in expression level was observed in subsets of PCa cases for nearly all (318/345 or 92 %) of the genes implicated in MET-driven motility (Fig. 3a). Of these 318 genes, approximately half (150 or 47 %) changed concordantly with those in TCGA cases, 24 of which were also associated with significantly altered DFS (Table 3 and Fig. 3a). Of the latter, only 1 (MAP2K6) was associated with improved DFS (Fig. 3b), while the other 23 were associated with decreased DFS (Fig. 3c, d). Collectively, these 24 genes were altered in the majority of PCa cases and identified a patient group with decreased DFS relative to those without expression change, though this difference was not statistically significant (Fig. 3e). Omitting MAP2K6 yielded a set of 23 genes whose expression was also altered in a majority of cases, comprising a patient group with significantly decreased DFS (logrank P = 0.007542, Fig. 3f) that was also clinically distinct: increased tumor MAP2K6 expression seldom occurred in the same patients as those with changes in any of the other motility-associated genes: odds ratios were <0.5 for 17/23 genes and 6/23 showed no association (odds ratio between 0.5 and 2; Fig. 3g). In contrast, expression changes in the other 23 genes frequently co-occurred in PCa tumors: 215 of 253 changes (85 %) co-occurred with odds ratios between 2 and 10, 101 of which (40 %) co-occurred with odds ratios >10; most of these tendencies were statistically significant (P < 0.05; Fig. 3g).

Fig. 3.

Concordant gene expression changes among PC3M and TCGA datasets associated with significantly altered DFS predict poor outcome, a Scheme for unsupervised interrogation of concordant gene expression changes associated with enhanced PC3M migration/metastasis and TCGA PCa data sets. Among all (345) migration associated genes with expression changes in PC3M, 92 % were also modulated in PCa patient tumor samples; of these 47 % changed in both sets concordantly, 24 of which were correlated with significantly different disease-free survival (DFS) time (logrank P ≤ 0.05). b through f: Kaplan–Meier curves of PCa patient DFS (months) for b MAP2K6 upregulation; c 11 other upregulated genes; d 12 downregulated genes; e all 24 genes combined; and f combined 23 genes excluding MAP2K6. Percent of patients affected and logrank P values for each group are as listed. g Co-occurrence chart for the 24 gene expression changes significantly affecting PCa patient DFS. Although gene symbols are not listed, the chart represents a 2 × 2 matrix where all 24 events are arranged along the top and right axes, and probabilities of co-occurrence for each possible pair of events (expressed as odds ratios) estimated using the 85 PCa patient cohort are indicated by color as follows: Strong tendency towards mutual exclusivity (odds ratio between 0 and 0.1) is indicated by dark blue; some tendency towards mutual exclusivity (odds ratio between 0.1 and 0.5) is indicated by light blue; no association (odds ratio between 0.5 and 2) is indicated in white; some tendency toward co-occurrence (odds ratio between 2 and 10) is indicated by yellow; and strong tendency towards co-occurrence (odds ratio greater than 10) is indicated by orange. P values < 0.05, as derived via right-tailed Fisher’s exact test (not adjusted for FDR) are outlined in red. Co-occurrence/exclusivity between every other gene and MAP2K6 is indicated in the top row

Table 3.

Motility-associated genes with significant impact on disease-free survival (DFS) among 85 prostate cancer patients

Symbol	Entrez gene name	TCGA cases w/change (%)	DFS Logrank P
Expression increased ≥2-fold relative to Median
ATP5E	ATP synthase, H+ transporting, mitochondrial F1 complex, epsilon subunit	20	0.03397
CBX2	Chromobox homolog 2	7	0.04198
DIEXF	Digestive organ expansion factor homolog (zebrafish)	22	0.00341
GSPT1	G1 to S phase transition 1	7	0.00197
GUSBP3	Glucuronidase, beta pseudogene 3	5	0.00026
HELLS	Helicase, lymphoid-specific	8	0.00987
LIPG	Lipase, endothelial	4	0.00011
MAP2K6	Mitogen-activated protein kinase kinase 6	56	0.04918a
PAG1	Phosphoprotein associated with glycosphingolipid microdomains 1	6	0.03854
PARPBP	PARP1 binding protein	13	0.00404
TUBAL3	Tubulin, alpha-like 3	4	0.00174
WDR76	WD repeat domain 76	14	0.00880
Expression decreased ≥2-fold relative to median
ALDH1A3	Aldehyde dehydrogenase 1 family, member A3	9	0.02312
BOD1L1	Biorientation of chromosomes in cell division 1-like 1	2	0.00108
COG3	Component of oligomeric golgi complex 3	7	0.00753
CTBS	Chitobiase, di-N-acetyl-	6	0.00356
DYNC2LI1	Dynein, cytoplasmic 2, light intermediate chain 1	5	0.00174
FAM122C	Family with sequence similarity 122C	6	0.03444
HLF	Hepatic leukemia factor	36	0.00487
LOC100131564	Uncharacterized LOC100131564	14	0.00109
NT5E	5′-nucleotidase, ecto (CD73)	27	0.02611
PTPRM	Protein tyrosine phosphatase, receptor type, M	13	0.00190
RBM24	RNA binding motif protein 24	18	0.01039
SUV420H1	Suppressor of variegation 4–20 homolog 1 (Drosophila)	9	0.00184

^aImproved DFS

We compared gene expression changes observed in an experimental model related to metastasis with those observed in the TCGA to help prioritize genes for further analysis of potential pro-invasive function and impact, not to develop a gene signature that was predictive of metastasis in patients. We note that studies have shown that even random gene sets can provide statistically significant association with prognosis in cancer patients [127]. While provocative, the potential predictive value of the 23 genes identified here is a subject for future study. We note here that of these 23 genes, two (PTPRM, discussed above and NT5E, below) have been directly linked to cell migration and invasion, but several others have been implicated in the tumorigenesis, progression and metastasis of several malignancies, including PCa. CBX2 encodes a component of a Polycomb group (PcG) multiprotein PRC1-like complex; complexes of this class maintain the transcriptional repression of many genes, including HOX genes, throughout development. Increased CBX2 expression and immunohistochemical protein staining is correlated with metastasis and overall survival (OS) in breast cancer [128]. In zebrafish, DIEXF regulates the p53 pathway to control the expansion growth of digestive organs [129]; increased DIEXF expression is also associated with significantly decreased OS in breast invasive carcinoma, affecting 4.6 % of 482 cases [130]. HELLS regulates gene silencing through DNA methylation; elevated HELLS expression correlated significantly with melanoma progression and metastasis [131]. High expression of LIPG, encoding endothelial lipoprotein lipase, occurs in testicular germ cell tumors [132]. PARPBP (PARP-1 binding protein or PARI) overexpression promotes genomic instability and pancreatic tumorigenesis [133]. PARPBP overexpression in 11 % of the PCa cases analyzed by Taylor et al. [26] and amplification in 3 % of PCa cases in a later TCGA cohort [134] was associated with significantly decreased DFS and OS, respectively.

Decreased expression of ALDH1A3 in non-small cell lung cancer patients was associated with poor overall survival [135]. The transcription factor HLF (hepatic leukemia factor) is linked to acute lymphoblastic leukemia primarily though chromosomal rearrangements with another transcription factor, TFPT (also known as E2A) resulting in E2A-HLF fusion proteins that presumably transform through aberrant gene activation (reviewed in [136]). However, no such rearrangements were observed in the Taylor et al. TCGA cohort [26], where 36 % of the patient tumor samples showed HLF suppression associated with significantly decreased DFS. Ecto-5′-nucleotidase (NT5E/CD73) is an ecto-nucleotidase that generates extracellular adenosine, a potent immunosuppressor, from AMP. It is reportedly overexpressed in certain malignancies and has been linked to migration and tumor angiogenesis (reviewed in [137]), although there is evidence of context dependent effects. Significant NT5E suppression was seen in 27 % of the PCa patient tumor samples interrogated by Taylor et al. [26], which was associated with significantly decreased DFS; no NT5E overexpression was observed in that cohort.

Conclusions

Ingenuity Pathway Analysis of significant gene expression changes induced by MET ligands that were associated with increased motility and invasion by PC3M cells, and spontaneous metastasis by PC3M xenografts in mice, identified 72 out of 345 (21 %) genes with significant involvement in IPA cell motility and/or tumor metastasis networks. The majority (69 %) of these expression changes occurred in a direction (increase or decrease) that was concordant with predicted or documented promotion of motility and/or metastasis. For 32 genes in the IPA functional annotations of tumor cell migration that most significantly overlapped the PC3M dataset, published evidence indicated activity through (1) regulatory and/or structural roles in motility/invasion machinery, (2) regulation of stress and inflammatory pathways that independently enable motility and potentially amplify and perpetuate HGF-induced motility and invasion, or (3) regulation of broader, generally pro-invasive transcriptional programs implicated in PCa progression and metastasis. These findings provide an overview of the highly integrated HGF invasive program and a foundation for further analysis of individual pathways engaged by activated MET in PCa. Finally, 24 of the 345 gene modulation events associated with MET-mediated PC3M cell motility and metastasis (7 %) were similarly modulated in a collective majority of 85 PCa cases, 23 of which were associated with significantly decreased DFS. The strikingly high rate of co-occurrence of these changes in this PCa patient cohort supports the likelihood that “master switches” such as the HGF/MET pathway may be important drivers of PCa metastasis that warrant further scrutiny for the development of molecular diagnostics and therapeutics.

Abbreviations

HGF

Hepatocyte growth factor

HSP

Heparan sulfate proteoglycan

TCGA

The Cancer Genome Atlas

h

Human

m

Murine

p

Phospho-

IPA

Ingenuity Pathway Analysis

JAK

Janus kinase

STAT

Signal transducer and activator of transcription

MAPK

Mitogen activated protein kinase

PTEN

Phosphatase and tensin homolog

NFkB

Nuclear factor kappa-light-chain-enhancer of activated B cells

PTP

Protein tyrosine phosphatase

FAK

Focal adhesion kinase

PSA

Prostate specific antigen

DFS

Disease-free survival

FDR

False discovery rate

PCa

Prostate cancer

SD

Standard deviation