Hepatocyte growth factor (HGF) autocrine activation predicts sensitivity to MET inhibition in glioblastoma

Abstract

Because oncogene MET and EGF receptor (EGFR) inhibitors are in clinical development against several types of cancer, including glioblastoma, it is important to identify predictive markers that indicate patient subgroups suitable for such therapies. We investigated in vivo glioblastoma models characterized by hepatocyte growth factor (HGF) autocrine or paracrine activation, or by MET or EGFR amplification, for their susceptibility to MET inhibitors. HGF autocrine expression correlated with high phospho-MET levels in HGF autocrine cell lines, and these lines showed high sensitivity to MET inhibition in vivo. An HGF paracrine environment may enhance glioblastoma growth in vivo but did not indicate sensitivity to MET inhibition. EGFRvIII amplification predicted sensitivity to EGFR inhibition, but in the same tumor, increased copies of MET from gains of chromosome 7 did not result in increased MET activity and did not predict sensitivity to MET inhibitors. Thus, HGF autocrine glioblastoma bears an activated MET signaling pathway that may predict sensitivity to MET inhibitors. Moreover, serum HGF levels may serve as a biomarker for the presence of autocrine tumors and their responsiveness to MET therapeutics.

The intrinsic ability of glioblastoma multiforme (GBM) tumor cells to invade normal brain tissue impedes complete surgical eradication and predictably results in early local recurrence and mortality. Understanding the molecular mechanisms of GBM invasiveness will lead to novel therapeutic strategies. The major signaling pathways driving GBM tumorigenesis and invasion are Ras-MAPK and AKT. Hepatocyte growth factor (HGF) binding to its receptor MET triggers this series of intracellular signaling, leading to tumor cell proliferation, invasion, survival, and angiogenesis (1–6). Aberrant MET activation can occur through numerous mechanisms, such as autocrine or paracrine stimulation, transcriptional regulation (7), ligand-independent or mutational activation (1, 8); MET amplification can be a major driver after acquired resistance to EGF receptor (EGFR) inhibitors (9), because of cross-talk with other receptor tyrosine kinase (RTK) family members.

Most glioblastomas show MET overexpression, and some display HGF autocrine activation of the MET signaling pathway (10). Approximately 88% of GBM patients have an aberrant RTK/Ras-PI3K pathway activity. MET is located on chromosome 7q, and gains of chromosome 7 occur frequently in GBM. Also, though MET mutations are rare (11, 12), a high level of MET amplification is found in ∼4% of GBM tumors. Amplification of EGFR occurs in 45% of GBM tumors and could be associated with aberrant MET expression (11).

HGF can also transcriptionally activate EGFR signaling in GBM cell lines (13), and EGFR variant III (EGFRvIII) can activate MET signaling (14), suggesting the importance of using a combination of MET and EGFR inhibitors in targeting GBM. EGFRvIII and MET inhibitors synergize against PTEN-null/EGFRvIII+ GBM xenografts (15). Because both MET and EGFR inhibitors are being tested against GBM in clinical trials (16–18), it is increasingly important to identify biomarkers that can predict tumor sensitivity. Knowledge of mechanisms determining susceptibility to MET or EGFR inhibitors will improve identification of patient subgroups suitable for MET and EGFR therapeutics.

We investigated in vivo glioblastoma models for their susceptibility to MET inhibitors sustained by either HGF autocrine or paracrine activation or by MET and EGFR amplification. HGF autocrine expression correlated with p-MET levels in HGF autocrine cell lines, and show high sensitivity to MET inhibition in vivo. An HGF paracrine environment could enhance glioblastoma growth in vivo but did not indicate sensitivity to MET inhibition. EGFRvIII amplification predicted sensitivity to EGFR inhibition, but MET polysomy in the same tumor did not display MET activity and did not predict sensitivity to MET inhibition. Thus, HGF autocrine glioblastoma bears an activated MET signaling pathway that may predict sensitivity to MET inhibitors in glioblastoma patients. Moreover, serum HGF levels may serve as a significant biomarker for the presence of autocrine tumors and their response to MET therapeutics.

Results

HGF Expression and MET Phosphorylation in Glioblastoma Cell Lines.

We previously showed that GBM cells are invasive and can be highly metastatic (19). Commonly used GBM cell lines (U251, U87, and DBTRG-05MG) have subpopulations with metastatic potential that can be selected. Compared with the parental cells, these metastatic sublines (called U251M2, U87M2, and DBM2) not only induced lung metastases, but also grew more aggressive and showed significantly reduced survival time in orthotopic mouse models. The M2 derivatives all expressed elevated levels of IL-6, IL-8, GM-CSF, and BDNF, factors associated with either cancer metastasis or GBM malignancy (19).

To identify additional markers of invasion in gliomas, we used microarray technology to compare the three GBM-M2 lines with their parental lines both in vitro and in an in vivo orthotopic model. A paired analysis identified 1,008 genes differentially expressed in vitro between the three GBM-M2 lines and their respective parental lines (cutoff P ≤ 0.05 in a paired Student t test; multivariate permutation test = 0.06) (Fig. S1A). The same analysis identified 1,764 genes differentially expressed between in vivo-derived intracranial tumors and the parental lines (multivariate permutation test P = 0.008) (Fig. 1A). We found 206 genes that were common to the in vitro and in vivo analyses.

Fig. 1.

HGF expression level and MET phosphorylation in GBM cell lines. (A) Heat map of HGF expression changes between parental and M2 cell lines in vivo and in vitro. (B) Western blot of GBM cell lines and subclones showed that up-regulation of HGF expression is accompanied by increased p-MET.

Because intracranial tumors likely provide a better representation of GBM tumor biology, we used in vivo microarray data to analyze signaling pathways potentially responsible for glioma invasiveness. The average gene expression values in the three pairs of cell lines (Fig. S1A) were evaluated using ingenuity pathway analysis, and HGF signaling was one of the top eight canonical pathways and the one with the highest percentage of up-regulated genes (Fig. S1B, red). HGF transcription was significantly elevated in U87M2 relative to its parental cell line both in vivo and in vitro (Fig. 1A). By contrast, the MET level remained unchanged (Fig. 1B and Fig. S1). Moreover, increased HGF transcription paralleled increase in up-regulation of the Ras-MAPK and AKT pathways, the leading pathways involved in gliomagenesis (6, 11). MET transcriptional levels were unaffected (Fig. S1C). These observations suggested that HGF expression, rather than MET expression, determined the MET signaling activity.

To validate the microarray results, we compared the three pairs of GBM cell lines together with six other GBM lines for HGF, MET, and p-MET expression by Western blots (Fig. 1B). Elevated HGF expression in U87M2 cells was consistent with the microarray up-regulation result. U87M2 cells also displayed higher levels of p-MET relative to the U87 parental cells (Fig. 1B). Thus, HGF may be the primary regulator of MET activity in autocrine tumors. SF295SQ1, a subline from SF295 characterized by enhanced tumor growth in SCID mice, and like U87M2, showed elevated HGF with little or no change in MET expression (Fig. 1B), indicating it was autocrine for HGF. We observed that all 12 cell lines expressed MET to varying degrees: four expressed detectable HGF, and five displayed p-Met. All HGF-positive lines displayed p-MET, but only one, SF268, was among the eight HGF-negative lines. These findings suggested that HGF expression and autocrine signaling were the dominant regulators of MET activity in GBM. Based on these results, we asked whether HGF autocrine expression determines sensitivity to MET inhibition.

HGF Autocrine Activation and Sensitivity to MET Inhibitors.

Urokinase (uPA) activity has been used as a surrogate for HGF-induced invasion (20, 21). We tested the uPA response of GBM cell lines to HGF with and without the MET inhibitor SGX523. In U251M2, T98G, and DBM2 cells (which do not display autocrine HGF expression), exogenous HGF up-regulated their uPA activity, and SGX523 at 0.1 μM inhibited that HGF-induced increase (Fig. S2 A–C). EGF also up-regulated uPA activity in the cells and was specifically blocked by the EGFR inhibitor erlotinib. There was no evidence of cross-inhibition between SGX523 and erlotinib, indicating the two signaling pathways were distinct, at least at an early (24 h) time point in vitro. By contrast, the HGF autocrine U87M2 cells did not respond to either HGF or EGF, but SGX523 did inhibit the basal level of uPA activity even when the cells were stimulated by EGF (Fig. S2D), indicating that the uPA response in U87M2 cells was highly dependent on endogenous HGF expression.

Western blot analysis showed results consistent with the uPA assay (Fig. S2E). Both HGF and EGF activated the MAPK and AKT pathways in DBM2 cells. There was no evidence of cross-inhibition between the MET and EGFR pathways in vitro. Thus, SGX523 did not inhibit EGF-induced p-EGFR, and erlotinib did not inhibit HGF-induced p-MET. Again, U87M2 was highly dependent on the autocrine HGF-activated signaling pathway (Fig. S2E). DBM2 required HGF stimulation to initiate the signaling pathways, whereas U87M2 displayed constitutive MET activation in the absence of external HGF stimulation. At 1–10 μM, SGX523 inhibited p-MET and the downstream MAPK and AKT pathways in U87M2 regardless of additional HGF or EGF stimulation (Fig. S2E). These results indicate that HGF autocrine status could be the dominant determinant of MET signaling pathway activity.

To test whether an active HGF autocrine loop predicts sensitivity to MET inhibitors in vivo, we inoculated GBM cell lines (either with or without HGF autocrine status) into mice subcutaneously and compared their sensitivity to SGX523. HGF autocrine xenograft tumors from U87M2, U118, and SF295SQ1 (Fig. 2 A–C) were all sensitive to SGX523 in a dose-dependent manner (P < 0.05 at all three doses). SGX523 caused dramatic tumor growth inhibition and regression within 2 wk. These results indicate that HGF autocrine status may be useful as a predictive marker for targeting GBM with MET inhibitors.

Fig. 2.

HGF autocrine GBM tumors are sensitive to SGX523 in vivo. GBM cells (5 × 10⁵) were inoculated subcutaneously into SCID and SCIDhgf mice. When tumors had grown to 100–120 mm³, the mice bearing tumors of similar size were grouped for treatment as indicated. (A–C) HGF autocrine U87M2, SF295SQ1, and U118 tumors were sensitive to SGX523 treatment alone. (D–E) Non-HGF autocrine U251 and DBM2 tumors were not.

HGF paracrine activation was evaluated by comparing xenograft tumor growth with and without expression of human HGF in SCIDhgf-Tg vs. SCID mice (22, 23). The HGF autocrine U87M2 and SF295SQ1 cells displayed similar tumor growth potential in both types of mice (Fig. 2 A and C). U251M2 and DBM2 cells, lacking HGF autocrine activity, showed a partial growth advantage to paracrine HGF stimulation in SCIDhgf mice (unpaired t test, unequal variance DBM2: P < 0.05; U251M2: P > 0.05; Fig. 2 D and E), but showed no response to SGX523 (60 mg/kg). Increasing the dose to 120 mg/kg for U251M2 cells did not improve the response (Fig. 2D). Interestingly, though SF295 wild-type tumors had a significant growth advantage in SCIDhgf-Tg mice vs. SCID mice, the SF295SQ1 subclone, which became HGF autocrine dominant after tumor passage (Fig. 1B), displayed the same growth in both models (Fig. 2C), further supporting the concept that endogenous HGF production influences the MET dependency of HGF autocrine GBM tumor. SF268 is an interesting model that has p-MET but is not HGF positive. However, this line was not tumorigenic when tested in either SCID on SCIDhgf mice, and therefore was not tested further. We conclude that HGF autocrine status predicts HGF-dependent susceptibility to MET inhibitors and hence may be useful as a marker for targeting HGF autocrine GBM.

Serum HGF Level Indicates Therapeutic Efficacy in HGF Autocrine Xenograft Models.

Because HGF is secreted by autocrine tumor cells into the circulation, a reduction of tumor size should result in a decrease of HGF production and could serve as a biomarker of therapeutic response. Using ELISA, we determined the human HGF levels in the serum of the HGF autocrine tumor-bearing mice from the in vivo MET drug study (Fig. 2). SCID mice in the SGX523 alone and combination groups had much smaller tumors (Fig. 2 A–C), accompanied by a significantly lower serum HGF level (Table S1). This result suggested that the HGF autocrine status was sufficient to maintain activation of the MET signaling pathway and was the key determinant of sensitivity to MET inhibition.

Combination of EGFR and MET Inhibitors Enhances Efficacy Against GBM.

Fig. 2 shows that different GBM models were preferentially sensitive to either MET or EGFR inhibitor treatment alone (e.g., U87M2, U118, and SF295SQ1 are more sensitive to SGX523, whereas DBM2 was somewhat sensitive to erlotinib). U251M2 tumors, which tolerated either SGX523 or erlotinib alone, were significantly more sensitive to the combination, even in SCID mice (Fig. 2D; P < 0.05), where there was no HGF paracrine activation and the tumor cells had no basal p-MET in vitro. We have observed similar results with other xenograft models in SCIDhgf or SCID mice (24). SGX523 induced U87M2 tumor regression in the first week after treatment (Fig. 2A). However, the tumors started to return after 2 wk, even with continuous dosing (Fig. 3), implying selection of a rescue pathway. Because EGFR activation has often been linked to MET signaling (14, 24, 25), we tested whether erlotinib would be efficacious with MET inhibition. We showed that a combination of SGX523 and erlotinib prolonged U87M2 growth inhibition (Fig. 3A) (P < 0.05), and erlotinib alone failed (Fig. 3B), suggesting that the combination of MET and EGFR inhibitors may have clinical value in GBM.

Fig. 3.

Erlotinib/SGX523 inhibition of U87M2 tumors. (A) SGX523 (90 mg/kg) in combination with erlotinib (150 mg/kg) inhibited U87M2 tumor growth. (B) Erlotinib (150 mg/kg) alone did not inhibit U87M2 tumor growth.

Gain of Chromosome 7^MET in GBM and MET Inhibition in Vivo.

Recent studies have shown that ∼88% of GBM patients bear tumors having an altered RTK/Ras/PI3K pathway activity resulting from EGFR (45%), platelet-derived growth factor receptor (PDGFR; 13%), or MET (4%) amplification (11, 12). To determine whether amplified MET or EGFR predicts sensitivity to their specific inhibitors, we examined the GBM stem cell line X01GB (26). Through chromosome analysis and FISH, we found that X01GB had a ploidy level ranging from 4n to 6n and harbored over 100 copies of EGFR and 7–10 copies of chromosome 7 and MET. The amplification of EGFR (EGFR^amp) was in the form of double minutes (dmin), and that of MET was in the form of gain of chromosome 7 (7gain^MET) (Fig. 4A). However, the activation levels of the two receptors were very different. For example, X01GB cells expressed high levels of EGFR and the mutant derivative form EGFRvIII at both the transcriptional level (Fig. 4B) and protein level; the expression of p-EGFRvIII indicates the activation of EGFR pathway (Fig. 4C). The expression of MET was quite low, and there was no detectable p-MET either in vitro (Fig. 4C) or in vivo (Fig. S3A). No HGF expression was detected in the tumors (Fig. S3B), indicating that MET was inactive. Moreover, the X01GB tumors did not respond to the MET inhibitor SGX523 but were very sensitive to erlotinib at 75 mg/kg (Fig. 4D). These results suggested that though EGFR^amp might be used as a predictive marker for EGFR inhibitor sensitivity, 7gain^MET may not indicate sensitivity to MET inhibition.

Fig. 4.

GBM cells with 7gain^MET are not sensitive to MET inhibitors in vivo. (A) Cytogenetic analysis of X01GB stem cells in interphase (Center) and in metaphase (Left and Right), shown at 100× magnification. FISH signals detecting MET (red) and EGFR (green) showed EGFR amplification as dmin and a gain of MET. (B) RT-PCR of MET, EGFR, and EGFRvIII levels in X01GB and V13 xenografts. (C) Western blot of MET and EGFR expression and activation in X01GB and V13. (D) In vivo treatment efficacy of SGX523 and erlotinib on X01GB tumors. (E) Cytogenetic analysis of V13 cells and tumors shown at 100× magnification. Primary tumor nuclei and xenograft tumor nuclei show the same abnormalities. SKY analysis (Top Left) showed trisomy 7 in the V13 cell line. FISH signals showed MET (red), EGFR (green and aqua), HGF (green), and chromosome X (red). 7gain was detected by SKY and metaphase FISH; a high level of EGFR amplification occurred as dmin (Top Right). (F) In vivo treatment efficacy of SGX523 and erlotinib on V13 tumors.

We also tested a GBM patient's specimen, and, for comparison, V13 was subjected to similar cytogenetic analyses as X01GB. We performed spectral karyotyping (SKY) and FISH on the V13 cell line and FISH on the primary and xenograft tumors. The V13 primary tumor had three copies of HGF and MET (chromosome 7) and an amplification of EGFR (50–100 copies; Fig. 4E). The SKY and FISH analysis of the V13 cell line clearly showed a 7gain and 10–100 dmin of EGFR (Fig. 4E), similar to the V13 xenograft tumor. The FISH analyses of the V13 cell line, the primary tumor, and the xenograft tumor are summarized in Table S2. V13 had a cytogenetic profile similar to that of X01GB.

We examined MET and EGFR expression levels in the V13 cell line and xenograft tumors. The V13 xenograft tumors were very similar to X01GB with low MET expression, lack of phosphorylation (Fig. 4C and Fig. S3A), and no HGF expression (Fig. 4 B and C, and Fig. S3B). We did not detect MET or EGFR signaling in the V13 cell line, probably because the majority of cells isolated from the primary tumor (78%) appeared normal (Table S2). The V13 xenograft tumor was transplanted into SCIDhgf mice at an early passage number (p2) for an in vivo efficacy study in response to SGX523 or erlotinib. The tumor showed identical results to X01GB cells (Fig. 4F) in that it was extremely sensitive to erlotinib at 75 mg/kg but showed no response to SGX523. The combination of both inhibitors did not enhance the efficacy, possibly due to the strong effect of erlotinib alone. Our data show that the lack of association between 7gain^MET and MET activity also held for an ex vivo GBM, showing that it was not unique to a long-term cell line such as X01GB.

Aberrant Expression of HGF, MET, and EGFR in Human GBM.

To estimate the frequency of HGF autocrine or paracrine status, and of 7gain^MET with EGFR^amp in GBM patients, we analyzed the Cancer Genome Atlas (TCGA) Network datasets from 202 patients via in silico assays for the expression of EGFR, HGF, and MET. A self-organizing heat map displaying the HGF, EGFR, and MET genes segregated the patients into four groups (Fig. 5A). A majority of samples (group A, n = 79) displayed high expression of EGFR and low expression of MET and HGF. A significant number of samples (group C, n = 52) showed low expression of EGFR and high expression of HGF and MET. The remaining samples showed either low (group D) or high (group B) expression of all three genes. Though 90 cases (45%) overexpressed EGFR, 63 cases (31.2%) showed overexpression of both HGF and MET, suggesting the possibility of autocrine HGF signaling.

Fig. 5.

In silico analysis of EGFR, HGF, and MET genetic aberrancy in GBM patients. (A) Self-organizing heat map based on transcriptional profiling of 202 GBM samples assayed by TCGA Network (11) on an Agilent 244K platform array. The map displays HGF, MET, and EGFR transcripts (rows) across the GBM samples (columns). The dendrogram indicates the degree of similarity among GBM samples using Pearson's correlation coefficient. Genes were projected using log2 intensity, and gene ratios were average corrected across experimental samples and displayed according to the uncentered correlation algorithm. Red indicates overexpression; green, underexpression; black, unchanged expression; and gray, no detection of expression (intensity of both Cy3 and Cy5 was below the cutoff value). (B) Matrix similarity based on Pearson's correlation for the HGF, MET, and EGFR transcripts in GBM samples. (C) CGH analysis of the frequency of amplifications occurring in the A–D groups. P value refers to the significance of correlation between EGFR, HGF, and MET gene copy number alteration and their transcriptional levels as shown in A. (D) Scatter plot between the average HGF intensity from the four groups described in Table S5 and the percentage of samples determined to have HGF autocrine activation. (E) Scatter plot between the average MET intensity from the four groups described in Table S5 and the percentage of samples determined to have HGF autocrine activation.

Given that the levels of EGFR and MET expression were opposite in a majority of patients (groups A and C), negative regulation may occur between the two pathways; the matrix in Fig. 5B shows an overall negative correlation between MET and EGFR (n = 202, Pearson's P = –0.4), suggesting that different GBM tumors could be preferentially sensitive to one inhibitor depending on the activation of the MET or EGFR pathways. We do not see this negative MET/EGFR relationship in melanoma patient samples (Fig. S4), suggesting this effect might be organ specific.

We estimated the percentage of patients (n = 202) who had both EGFR^amp and 7gain^MET (as found with X01GB and V13) with only EGFR activated. We used CGH data to analyze EGFR, HGF, and MET amplification (Fig. 5C) and correlated the results to the transcriptional data (Fig. 5A). We found that EGFR (94%), MET (78%), and HGF (71%) were frequently amplified in group A, consistent with a high percentage of 7gain (Table S3), as has been reported (27). However, when we characterized group A for levels of transcription, the MET and HGF genes were not highly expressed (Fig. 5A) though were highly amplified (Fig. 5C). By contrast, EGFR amplification in group A (94%) was associated with high EGFR expression, a pattern similar to that in X01GB and V13. These data suggest that the amplification observed in the EGFR gene was significantly correlated with that gene's expression in the four groups at the transcriptional level (Fig. 5A) (P = 5.9E–5). However, no significant correlation was found for the MET gene (P = 0.1), supporting the data (Fig. 4) that amplification of the EGFR gene may predict activation of the EGFR pathway, but amplified MET may not predict MET pathway activation. Interestingly, we show here that the copy number of HGF had a significantly higher correlation to HGF overexpression (P = 0.01), whereas that of MET showed no or low correlation to MET overexpression (P > 0.05), suggesting that HGF might be a better marker than MET for HGF/MET axis therapy.

Discussion

Preclinical studies have shown that targeting MET signaling can have potent antitumor effects, including GBM (28–31). MET inhibitors are currently in clinical trials against several cancers (16, 17). AMG102 (a neutralizing antibody against HGF) was not successful in clinical trials against recurrent GBM (32), but XL184 (a small molecule targeting both MET and VEGF) showed promising efficacy in phase II clinical trials against GBM (33). Therefore, it is important to understand the mechanisms leading to HGF/MET sensitivity and to identify the patient subgroups most likely to benefit from MET-targeted therapeutics. Although MET expression is high in GBM patients (10), and is used as a mesenchymal marker to indicate a more invasive GBM phenotype (12, 27), it may not be a good biomarker for predicting sensitivity to MET inhibitors.

Here, we suggest that HGF autocrine GBM tumors with constitutively activated MET serves as a biomarker for GBM patients who might benefit from MET inhibitor therapy. HGF measured by ELISA in serum or cerebrospinal fluid have been shown to correlate with GBM prognosis (34), and HGF in the cerebrospinal fluid of GBM patients (847 ± 155 pg/mL) was found to be significantly higher than in patients with meningioma (430 ± 28 pg/mL) or from healthy individuals (204 ± 28 pg/mL). High HGF in patients with GBM has also been associated with higher mortality and recurrence rates (35). These data further suggest that HGF expression may be a useful prognostic marker and an indicator for responsiveness to MET inhibitors.

Because high MET expression is found with high-grade tumors (Van Andel Institute, www.vai.org/met/) and is under consideration as a biomarker, we compared the use of HGF and MET as potential biomarkers to predict HGF autocrine activation by using the TCGA database. (Fig. 5 D and E and Tables S4 and S5). Samples with both high HGF and MET expression were defined as having HGF autocrine activation. Based on log2 signal intensity, 202 GBM samples were separately ranked from the highest to the lowest values of HGF and MET, and the average signal intensity for each individual gene was calculated. Because the average signal intensity of HGF or MET significantly dropped after the top 60 cases (for the last 70% of samples, average HGF intensity = –0.46; average MET intensity = –0.51), the top 60 were defined as high HGF or MET.

We showed that among the 20 samples with the highest HGF expression (average HGF intensity = 1.79), 13 also had MET overexpression. As the level of HGF decreased (second 20 samples, average HGF intensity = 1.03), the number of samples with high MET also decreased (n = 11) (Table S5), showing a good correlation (R² = 0.75) (Fig. 5D). Using the same method to evaluate MET overexpression as a biomarker, however, the identification of HGF overexpression was less informative. Among the 20 samples with the highest MET expression (average MET intensity = 2.37), 10 showed high HGF, but in the second 20 samples (average MET intensity = 0.84), 15 had high HGF (Table S5), an overall correlation much lower (R² = 0.23) (Fig. 5E). The data suggest that HGF expression level might be a better biomarker than MET in identifying samples with an active MET pathway.

Our data also showed that the serum HGF level correlated with the therapeutic efficacy of SGX523 in HGF autocrine GBM xenograft models (Table S1), suggesting that circulating HGF may be a useful prognostic and therapeutic marker for GBM patients. In the SCIDhgf-Tg mouse preclinical model, we observed that HGF paracrine activation promoted GBM tumor growth (DBM2) (Fig. 2E). However, the activation appeared weaker than that observed in other cancer models we have tested in this system (23, 36), and showed little response to MET inhibitors after 3 wk of dosing. MET amplification has also been found in non–small cell lung cancer patients resistant to EGFR inhibitors (24). Notably, all of these cases had increased expression of HGF in tumor sections, raising the possibility of HGF serving as a therapeutic marker for MET sensitivity. We suggest that autocrine HGF status may be detected by immunohistochemistry of tumor tissue accompanied by ELISA analysis of HGF expression in serum and cerebrospinal fluid.

MET amplification predicts sensitivity to MET inhibitors in gastric cancers (25), and MET amplification occurs in 4% of GBM patients (11). We tested whether the X01GB stem cell and V13 xenograft GBM lines, which have MET amplification, were also sensitive. Both of these lines have a 7gain^MET (3–10 copies) and EGFR^amp with 30–100 dmin. Surprisingly, neither V13 nor X01GB showed strong MET expression or detectable p-MET, but both had strong EGFRvIII expression and p-EGFR, unlike the SNU-5 and H820 lines, in which MET amplification was always accompanied by constitutively activated MET (24, 25). Consistent with the in vivo studies, these tumors did not respond to SGX523, but did respond to erlotinib. Thus, a high level of EGFR^amp did predict sensitivity to EGFR inhibition in GBM, but unless the tumor is HGF autocrine, 7gain^MET may not to be a good predictor for MET sensitivity.

We questioned whether a different profile of MET amplification would result in different MET activity. For both SNU-5 and H820, MET signals are located on chromosome 7 and in areas distinct from the endogenous gene locus (24, 25), whereas for V13 and X01GB, MET signals come only from chromosome 7. Thus, the constitutive activation of MET may be more associated with the aberrant MET locus. In fact, the EGFR amplification in V13 and X01GB occurred as extrachromosomal dmin and was highly associated with EGFR activation and sensitivity to EGFR inhibition. Fish analysis of GBM cell lines in vivo (Table S4) revealed that a gain of MET occurred frequently. This was consistent with our in silico analysis using the TCGA database (Fig. 5). SF295SQ1, U118, and DBM2 all had 7gain, and they were more or less sensitive to SGX523 (Fig. 2). U87M2 showed sensitivity to SGX523 but did not have 7gain, suggesting that even when not amplified, HGF autocrine tumors may be sensitive to MET inhibitors. Thus, it is worthwhile to accurately evaluate GBM patient samples. In particular, we suggest distinguishing 7gain^MET from MET^amp, because they may indicate different cellular activities. Interphase FISH analysis of frozen tumor tissues provides only the number of gene copies. A more informative technique is needed, such as metaphase FISH and chromosome analysis, which uses isolated tumor cells to identify the location of MET amplification or gain.

In this study we investigated HGF autocrine and paracrine status, as well as MET and EGFR amplification, as molecular determinants of HGF/MET dependency. We found that HGF autocrine status could indicate MET activity and predict sensitivity to MET inhibitors. Because MET inhibitors are in clinical trials against GBM, it would be valuable to further evaluate the use of HGF autocrine status as a marker for identifying patient subgroups for MET treatment. We also found that a combination of MET and EGFR inhibitors was effective in the HGF paracrine GBM models we tested, supporting the value of such a combination as a strategy for treating GBM. Because brain tumors are extremely heterogeneous, more information is needed on how the genetic background and overlapping signaling networks influence tumor growth to uncover the full potential of using novel combinations of reagents to treat glioblastoma.

Materials and Methods

See SI Materials and Methods for additional information related to the main text on the following topics: cell culture; HGF-induced uPA activity; HGF-induced downstream signaling pathway; GBM patient-derived xenograft tumor model; RNA preparation, amplification, and labeling for microarray analysis; data processing and statistical analyses; FISH; RT-PCR; in vivo SGX523 and erlotinib therapeutic efficacy study; ELISA analysis; and in silico analysis of TCGA datasets. SGX523 was provided by Eli Lily Pharmaceuticals, and erlotinib was provided by OSI Pharmaceuticals Inc. All animal studies were approved by the Institutional Animal Care and Use Committee at Van Andel Research Institute.