Research ArticleCancer

Dynamic Reprogramming of Signaling Upon Met Inhibition Reveals a Mechanism of Drug Resistance in Gastric Cancer

See allHide authors and affiliations

Science Signaling  22 Apr 2014:
Vol. 7, Issue 322, pp. ra38
DOI: 10.1126/scisignal.2004839


The Met receptor tyrosine kinase is activated or genetically amplified in some gastric cancers, but resistance to small-molecule inhibitors of Met often emerges in patients. We found that Met abundance correlated with a proliferation marker in patient gastric tumor sections, and gastric cancer cell lines that have MET amplifications depended on Met for proliferation and anchorage-independent growth in culture. Inhibition of Met induced temporal changes in gene expression in the cell lines, initiated by a rapid decrease in the expression of genes encoding transcription factors, followed by those encoding proteins involved in epithelial-mesenchymal transition, and finally those encoding cell cycle–related proteins. In the gastric cancer cell lines, microarray and chromatin immunoprecipitation analysis revealed considerable overlap between genes regulated in response to Met stimulation and those regulated by signal transducer and activator of transcription 3 (STAT3). The activity of STAT3, extracellular signal–regulated kinase (ERK), and the kinase Akt was decreased by Met inhibition, but only inhibitors of STAT3 were as effective as the Met inhibitor in decreasing tumor cell proliferation in culture and in xenografts, suggesting that STAT3 mediates the pro-proliferative program induced by Met. However, the phosphorylation of ERK increased after prolonged Met inhibition in culture, correlating with decreased abundance of the phosphatases DUSP4 and DUSP6, which inhibit ERK. Combined inhibition of Met and the mitogen-activated protein kinase kinase (MEK)–ERK pathway induced greater cell death in cultured gastric cancer cells than did either inhibitor alone. These findings indicate combination therapies that may counteract resistance to Met inhibitors.


Activation and amplification of the Met receptor tyrosine kinase (RTK) occur in gastric cancers, as well as in cancers of the lung, breast, and esophagus, and glioblastomas (19). Cancer cells that have MET amplification often have high abundance of Met protein, which is associated with ligand-independent activation and poor prognosis (14). More recently, Met signaling has been identified as a means through which resistance to other kinase inhibitors can occur. For example, MET amplification occurs in non–small cell lung cancer (NSCLC), and activated Met phosphorylates human epidermal growth factor receptor 3 (HER3), promoting cell survival and resistance to epidermal growth factor receptor (EGFR) inhibitors (10, 11). MET-amplified cells are frequently dependent on Met signaling for survival and proliferation and are sensitive to Met-targeted inhibitors (1, 2, 5).

Many studies have examined Met-dependent signals that mediate biological response upon stimulation by the Met ligand hepatocyte growth factor (HGF). However, less is known about the Met-dependent signals activated after constitutive activation of Met as a consequence of gene amplification, or the mechanisms mediating resistance to small-molecule Met inhibitors. HGF-dependent activation of Met initiates a program of invasive growth that modulates the breakdown of cell-cell contacts, promoting cell migration and invasion. This program facilitates the development of various organs during embryogenesis and contributes to wound healing and liver regeneration in the adult (8, 9, 12). Met primarily signals through the growth factor receptor–bound protein 2 (Grb2) adaptor protein [which couples Met to the Ras-MAPK (mitogen-activated protein kinase) pathway] as well as the Grb2-associated binding protein 1 (Gab1) scaffold protein [which indirectly connects Met to the phosphatidylinositol 3 kinase (PI3K)–Akt pathway and modulators of the actin cytoskeleton; reviewed in (9)]. Met is inhibited by recruitment of the Cbl E3 ubiquitin ligase, which promotes Met ubiquitination and trafficking for lysosomal degradation (1315). Dysregulation of Met signaling is observed in multiple human cancers and occurs through point mutations within the Met kinase domain, autocrine or paracrine activation of Met by HGF, increased abundance of Met, genomic amplification, and loss of Cbl-dependent inhibition through alternative splicing of MET (9, 1520).

Gastric cancer is the second deadliest and fourth most common cancer worldwide (21). MET amplification is observed in up to 23% of all gastric cancers and in 38.5% of the scirrhous subtype (2226). Moreover, in the absence of amplification, increased abundance of Met correlates with advanced-stage tumors and poor prognosis (2224, 2730). Most gastric cancer patients are diagnosed with locally advanced or metastatic disease; patients with nonresectable tumors are treated with chemotherapy alone and have a median survival of only 10 months [reviewed in (30)]. Hence, there is a clear need to identify more effective therapeutics for this disease. Small-molecule inhibitors and antibodies against Met are currently being investigated in all phases of clinical trials (6, 7, 31). However, despite the sensitivity of Met-addicted cells to Met inhibitors in vitro, Met inhibitors have not been as effective in patients with gastroesophageal cancer (32), highlighting an urgent need to better understand Met-dependent signaling pathways and the events that occur upon treatment with Met inhibitors that may promote resistance. Furthermore, although the activation of MAPK and Akt pathways has been extensively studied subsequent to ligand-stimulated activation of RTKs, the pathways required for cell survival and proliferation in oncogene-addicted cells (in which RTK signaling is constitutively activated) are still poorly understood. Here, our objectives were to identify pathways that are consistently dependent on Met kinase activity and to understand the signaling changes that occur upon Met inhibition that may contribute to response or resistance to Met inhibitors.


Met abundance in primary gastric tumors positively correlates with cell proliferation

MET amplification and overexpression have been detected in gastric carcinomas with poor prognosis, and cells with high Met abundance are sensitive to targeted Met inhibitors (2, 32, 33). To determine whether Met may be a potential therapeutic target in lymph node–positive gastric cancer patients, we evaluated Met abundance in a tissue microarray (TMA). We performed immunohistochemistry (IHC) on primary tumor cores from 35 patients with resected gastroesophageal junction (GEJ) adenocarcinoma after neoadjuvant chemotherapy during a local phase 2 trial (34) compared with matched lymph node metastases (fig. S1, A and B). A Met-negative gastric tumor is shown for comparison (fig. S1C). Because Met is present in all cells of epithelial origin, we expected and observed both membranous and weak, diffuse cytoplasmic staining in normal gastric epithelial cells (fig. S1D). Met IHC was independently scored by a pathologist using a method described by Graveel et al. (35): 0, no immunoreactivity; 1, weak immunoreactivity; 2, moderately strong immunoreactivity; and 3, strong immunoreactivity (fig. S1E). We observed strong Met staining in 13% of the primary tumors and 10% of the patient-matched lymph node metastases, and moderate Met staining in 23% of the primary tumors and 38% of the lymph node metastases (Fig. 1A). In addition, nuclear Ki67, an established marker for cell proliferation, significantly and positively correlated with strong Met immunostaining (Fig. 1, B and C, and fig. S1E), supporting a link between Met abundance and a more aggressive gastric cancer phenotype. To determine the correlation between intensity of Met IHC and MET amplification status, we isolated DNA from the formalin-fixed paraffin-embedded (FFPE) blocks of the tumors represented in the TMA. Using quantitative real-time polymerase chain reaction (qRT-PCR), we found that an increase in MET expression significantly and positively correlated with a strong Met IHC score (Fig. 1D and table S1). Fluorescence in situ hybridization (FISH) confirmed amplification of MET in a subset of samples with higher qRT-PCR score (Fig. 1, E and F). Because the clinical prognosis for patients with gastric and GEJ tumors remains poor, targeted therapies are being considered in the hopes of improving survival [reviewed in (36)]. Our data support that patients who have gastric tumors and lymph node metastases with strong Met immunostaining may benefit from screening for elevated MET copy number to recommend treatment with Met inhibitors.

Fig. 1 Met abundance in gastric tumors is positively correlated with proliferation.

(A) Percentage of tumors (n = 31) and lymph node (LN) metastases (n = 21) that exhibited weak (<0.33), medium (0.33 to 0.66), and strong (>0.66) Met staining. (B) Representative images of tumors with weak or strong Met staining and the corresponding Ki67 staining. (C) Percentage of Ki67-positive nuclei in tumors from (A) in each of the Met-positive categories. Error bars depict distribution of all data points, and horizontal bars represent the means. *P < 0.05. n.s., not significant. (D) MET amplification determined by real-time qRT-PCR. Data are mean fold change ± SD compared with normal tissue. ***P < 0.001. (E and F) Representative FISH images for MET in two samples (n = 4 each) showing either no MET amplification (E), scored as 0.9, or MET amplification (F), scored as 4.4. Red, MET probe; green, CEP7 probe.

MET-amplified gastric cancer cell lines are dependent on Met kinase activity for cellular proliferation and anchorage-independent growth

To establish a requirement for Met signaling and sensitivity to Met inhibition in gastric cancer, we used four gastric cancer cell lines (Okajima, MKN45, Snu-5, and KATO II) that have increased abundance of Met (2, 20). MET amplification is observed in MKN45, Snu-5, and KATO II cells (2, 37), although Okamoto et al. failed to observe amplification in Okajima cells (37). Thus, to test the MET amplification status in our cell lines, we extracted genomic DNA from each and performed qRT-PCR. The fold change in MET amplification was compared with the breast cancer cell line T47D, which is diploid for MET (Fig. 2A). In all gastric cancer cell lines, we observed at least a 10-fold amplification of MET (Fig. 2A). Met protein was also abundant in each cell line and was phosphorylated in the absence of exogenous ligand (Fig. 2B). Treatment of each cell line with a Met-specific small-molecule inhibitor PHA-665752 (PHA) (38) efficiently inhibited Met kinase activation, indicated by decreased phosphorylation on Tyr1234 and Tyr1235 (Tyr1234/1235) in the activation loop of the Met kinase domain, rapidly and consistently across all cell lines (Fig. 2, B and C).

Fig. 2 Gastric cancer cell lines with constitutively active Met are dependent on Met signaling for anchorage-independent growth and proliferation.

(A) Fold change in MET mRNA abundance relative to that in T47D cells (black bar) by qRT-PCR. (B) Western blots of total and phosphorylated Met (at Tyr1234/1235; pMet) in gastric cancer cell lines treated with 0.1 μM PHA or dimethyl sulfoxide (DMSO; vehicle control) for the number hours specified. (C) Quantification of phosphorylated Met abundance (normalized to total Met) from three independent Western blots of each of the cell lines in (B). (D) Representative images of soft agar colonies that were either untreated or treated with DMSO or 0.1 μM PHA. Images are at ×40 magnification. (E) Quantification of macroscopic colonies treated as described in (D). ****P < 0.0001. (F) Quantification of soft agar colonies cultured with 0.1 μM crizotinib or an equal volume of H2O. (G) Proliferation of cells cultured in DMSO or 0.1 μM PHA determined by the MTS assay and shown as relative absorbance (normalized to that at day 0). (H) Representative of three Western blots of lysates from cells treated with DMSO or 0.1 μM PHA for up to 7 days. All data are means ± SEM from three independent experiments.

Anchorage-independent growth is a well-established indicator of tumorigenic capacity; thus, we used soft agar assays to assess the impact of Met inhibitor treatment. Seeding of cells at low cell density into soft agar demonstrated that all four cell lines were capable of anchorage-independent growth (Fig. 2D). Inhibition of Met with PHA significantly decreased the number of both macroscopic (observable by eye) and microscopic colonies formed by each cell line (Fig. 2E and fig. S2A), whereas untreated cells and those treated with the vehicle control, DMSO, produced a comparable number of colonies in soft agar versus untreated control (Fig. 2E). To confirm that these effects were reproducible with another inhibitor of Met, we repeated the experiment with the dual Met and anaplastic lymphoma kinase small-molecule inhibitor crizotinib (39). Crizotinib also reduced the number of soft agar colonies formed in all four cell lines (Fig. 2F and fig. S2B). In two-dimensional (2D) cell culture, proliferation of each of the cell lines was also inhibited by PHA, as determined with a 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt (MTS) assay (Fig. 2G).

Upon mitogenic signaling, retinoblastoma protein (Rb), an inhibitor of the cell cycle, is inactivated by phosphorylation, thereby permitting the E2F family of transcription factors to stimulate expression of S-phase cell cycle–associated genes [reviewed in (40)]. In the presence of PHA, phosphorylation of Rb decreased in all four gastric cell lines (Fig. 2H), which was consistent with decreased cell proliferation. Together, these data demonstrate both the dependence of MET-amplified gastric cancer cell lines on Met signaling cell proliferation and anchorage-independent growth and the sensitivity of these cell lines to targeted Met inhibition.

Met inhibition produces temporal changes in gene expression

To examine the events that occur upon treatment with a Met-targeted small-molecule inhibitor and identify the pathways dependent on constitutive Met activation in gastric cancer, we extracted protein and RNA from each of the four cell lines at specific time points after the addition of PHA. Gene expression microarrays revealed that, for each of the four cell lines, many genes were differentially expressed at 24 hours after PHA addition compared with untreated controls at time 0 [Fig. 3A; assessed by Linear Models for Microarray Data (LIMMA), in which false discovery rate (FDR) ≤0.05 and fold change ≥2]. Surprisingly, many of these genes (1217 genes) were differentially expressed in all four cell lines (651 genes had increased expression; 566 had decreased expression), suggesting that expression of these genes depends on Met kinase activity.

Fig. 3 Met inhibition results in dynamic changes in gene expression in all four gastric cancer cell lines.

(A) Venn diagrams depicting the number of genes that increased or decreased more than twofold in each cell line 24 hours after addition of 0.1 μM PHA. P < 2.2 × 10−16, Fisher’s exact test approximation. (B) Schematic of the number of genes that increased or decreased in expression more than twofold. Data are average fold change across all cell lines at each time point after Met inhibition. (C) Heat maps of the 10 most differentially expressed genes at each time point upon 0.1 μM PHA treatment. Significant decrease in expression of DEGs (DUSP4 and DUSP6, P ≤ 1.95 × 10−26). (D) Heat map of the genes that significantly changed in expression in T47D 2A cells 0.5 hour after HGF (0.5 nM) stimulation. The corresponding changes in expression of these genes after 0.1 μM PHA treatment in the gastric cancer cell lines are shown in the right panel. Data in (A) to (D) are representative of 132 arrays [Okajima (24), MKN45 (60), Snu-5 (28), and KATO II (20)].

Met inhibition produced temporal changes in gene expression common to all cell lines. Expression of each gene was averaged across the four cell lines, and genes were then grouped by the time point at which their expression increased or decreased by at least twofold (illustrated in Fig. 3B). After Met inhibition, the first changes in gene expression were observed rapidly (within 2 hours), and an increasing number of differentially expressed genes were observed at each subsequent time point measured up to 24 hours (Fig. 3B).

Upon ligand stimulation of an RTK, activation of downstream signaling pathways promotes the expression of “immediate-early genes” (IEGs) such as JUN, FOS, and MYC, followed by “delayed-early genes” (DEGs) and secondary response genes (41). After Met inhibition, genes with an immediate decrease in expression (within 2 hours after PHA treatment) corresponded to the class of IEGs and included EGR1, EGR2, FOS, IER3, MYC, and FOSL1 (Fig. 3C and Table 1). After 4 hours of PHA-induced Met inhibition, expression of DEGs (such as DUSP4 and DUSP6) significantly decreased (Fig. 3C). DUSP4 and DUSP6 encode dual-specificity phosphatases (DUSPs) that inhibit extracellular signal–regulated kinase (ERK) (42). Activation of the Met RTK by its ligand, HGF, promotes epithelial-to-mesenchymal transition (EMT); consistent with this, expression of many genes associated with EMT, such as SNAI1, SNAI2, and EMP1, also decreased after 8 hours of PHA-induced Met inhibition, showing their continued dependence on Met signaling for expression (Table 1). In cells exposed to PHA for 24 hours, the expression of genes encoding proteins involved in the cell cycle and DNA replication was decreased (Fig. 3C and Table 1), consistent with the reduced cell proliferation observed at later time points (Fig. 2G). Genes with increased expression at 24 hours of Met inhibition included those encoding metallothioneins, transmembrane transporters or channels, and transcription factors (Fig. 3C), as well as genes encoding proteins involved in apoptosis and autophagy (Table 1).

Table 1 Categorized list of average change in expression of genes upon Met inhibition.

Genes were obtained from 132 arrays of gastric cancer cell lines treated with PHA for the indicated time points. Time is shown in hours, and colored boxes indicate an increase (red) or decrease (blue) in fold change.

View this table:

Positive effectors of Met signaling appeared to be differentially regulated upon Met inhibition. Gab1 and Gab2 are both scaffold proteins that bind to Met and are critical to Met-induced biological responses (4345). Expression of both of these genes increased twofold upon Met inhibition (Table 2). In contrast, PTPN11, which encodes an Shp2 tyrosine phosphatase required to sustain ERK signaling downstream from Met (4548), decreases in expression upon Met inhibition (Table 2).

Table 2 Average fold change in expression of genes in the Met signaling pathway after Met inhibition.

Expression of genes involved in Met signaling was obtained from 132 arrays of gastric cancer cell lines treated with PHA for the indicated time points. Time is shown in hours, and colored boxes indicate an increase (red) or decrease (blue) in fold change.

View this table:

To determine whether the genes that decreased in response to Met inhibition were induced in response to HGF, we stimulated epithelial T47D cells stably expressing wild-type Met (T47D 2A) (14) with HGF for 0.5, 3, and 24 hours. All of the genes that significantly increased in expression after 0.5 hour of HGF corresponded to IEGs, and most of these genes rapidly decreased in expression upon PHA-induced Met inhibition in the gastric cell lines (Fig. 3D). Amit et al. (41) also demonstrated rapid and transient expression of the IEGs EGR1, FOS, FOSL1, and IER3 upon EGF stimulation in HeLa and MCF10A cells, confirming that induction of these IEGs occurs after HGF or EGF stimulation (fig. S3). A sustained increase in expression of the DEGs DUSP4 and DUSP6 was also observed after HGF or EGF stimulation (fig. S4A) (41) and decreased in expression after Met inhibition. At the protein level, DUSP4 and DUSP6 initially decreased in abundance after HGF stimulation before recovering at later time points, suggesting that DUSP4 and DUSP6 are differentially regulated at the gene and protein levels (fig. S4, B and C).

Genes differentially expressed at 3 hours after HGF stimulation showed opposite expression patterns after Met inhibition (fig. S5A). It is not unexpected that many genes induced by Met stimulation (at 0.5 and 3 hours) decreased in expression upon Met inhibition. However, for many genes assessed for expression at 24 hours, the change in expression after Met activation positively correlated with that after PHA addition (fig. S5B). Pathway analysis of the genes that change in expression in the same direction indicated enrichment for those encoding proteins involved in DNA replication, cell cycle, cell division, and apoptosis (fig. S5, C and D). Together, these data suggest that genes induced alike at later time points (24 hours) by either Met stimulation or inhibition represent genes that are responsible for promoting Met signaling and Met-dependent biological responses, such as cell proliferation (fig. S5, C and D).

PHA promotes rapid and sustained inhibition of the Akt and STAT3 pathways

To further characterize the downstream signaling pathways that were dependent on constitutive Met kinase activity, we performed reverse-phase protein array (RPPA) on whole-cell lysates from MKN45 and KATO II cells, either untreated or treated for 24 hours with DMSO or PHA. According to the abundance of total and phosphorylated proteins, PHA-treated samples clustered together and separately from the untreated or DMSO control cells regardless of cell line (Fig. 4A). Both the Akt and STAT3 (signal transducer and activator of transcription 3) pathways appeared to depend on Met kinase activity because the amounts of phosphorylated STAT3 (at Tyr705) and phosphorylated Akt (at Thr308 and Ser473) decreased in PHA-treated cells (Fig. 4A). The decrease in Ser473-Akt phosphorylation was validated by Western blot analysis, and rapid and sustained loss of phosphorylated Akt was observed within 30 min of PHA-induced Met inhibition (Fig. 4, B and C). The decrease in Akt phosphorylation after PHA treatment was not unexpected because HGF-mediated Met activation promotes the phosphorylation of Akt, a prosurvival signal, as early as 5 min after stimulation (14). Activated Akt promotes cell survival at least in part through the inhibition of FOXO transcription factors [reviewed in (49)]. Consistent with this, upon Met inhibition, the expression of FOXO3A and gene targets of FOXO transcription factors (such as the antiproliferative and proapoptotic genes CDKN1B and BCL2L11) increased, supporting the induction of cell death pathways (fig. S6B).

Fig. 4 Met inhibition results in sustained inhibition of Akt and STAT3 signaling pathways.

(A) Heat map of total or phosphorylated protein abundance from each of the different samples (MKN45 or KATO II cells that were either untreated or treated with DMSO or 0.1 μM PHA for 24 hours) determined by RPPA. Insets depict total or phosphorylated proteins that decreased in both PHA-treated cell lines (purple box) or increased in PHA-treated KATO II cells but decreased in MKN45 cells (blue box). (B) Western blots of phosphorylated (Ser473) and total Akt amounts in gastric cancer cell lines treated with DMSO or PHA (0.1 μM) for the specified time points. (C) Quantitation (means ± SEM) of samples in (B) from three independent experiments. (D) Representative Western blots of phosphorylated (Tyr705) and total STAT3 amounts in cells treated with DMSO or PHA (0.1 μM). (E) Quantitation (means ± SEM) of samples in (C) from three independent experiments. (F) Heat map of STAT3-regulated genes from (52) that significantly changed in expression after treatment with 0.1 μM PHA. P = 3.1 × 10−5, Fisher’s exact test. The corresponding changes in expression of these genes after expression of activated STAT3 [as shown in (52)] are depicted in the column to the left of the heat map. P = 2 × 10−16, Fisher’s exact test. (G) Number of STAT3-targeted genes that are differentially regulated in gastric cells after addition of 0.1 μM PHA (green circles). STAT3 signatures (purple circles) from GM12878 and HeLa cell lines derived from (54) were applied to identify Met-dependent STAT3-targeted genes.

HGF-induced stimulation of Met promotes the tyrosine phosphorylation of STAT3 (50), which promotes its translocation to the nucleus and induction of target genes (51). In the gastric cancer cell lines, Met inhibition led to rapid (within 30 min) and sustained loss of phosphorylated STAT3 (Fig. 4, D and E). Consistent with a decrease in STAT3 activation, we observed a decrease in expression of STAT3 gene targets, including those that encode inhibitors of cytokine signaling, such as CISH, SOCS2, and SOCS3 (Table 3). Many of the genes differentially expressed upon Met inhibition in our data were also present in STAT3-regulated gene signatures [Fig. 4F (52, 53)], and genes that increased in expression with activated STAT3 decreased in expression upon Met inhibition (Fig. 4F). This inverse correlation is also observed with genes that decrease in the presence of activated STAT3 (52). Because a STAT3 signature is not limited to direct targets of this transcription factor, we used published chromatin immunoprecipitation–sequencing (ChIP-seq) data for STAT3 from GM12878 lymphoblastoid and HeLa cell lines (54). Notably, 76 and 64 STAT3 targets in GM12878 and HeLa cells, respectively, were also differentially expressed in gastric cancer cell lines after Met inhibition (Fig. 4G and fig. S7). Thus, we have identified a Met-dependent, STAT3-mediated gene expression signature and established that the activation of the STAT3 and Akt signaling pathways is dependent on constitutive activation of Met in MET-amplified gastric cancer cell lines.

Table 3 Average gene expression changes of negative regulatory proteins upon Met inhibition.

Expression of genes that inhibited the Raf, ERK, JAK-STAT, JNK, p38MAPK, and mTOR (mammalian target of rapamycin) pathways. Data were obtained from 132 arrays of gastric cancer cell lines treated with PHA for the indicated time points. Time is shown in hours, and colored boxes indicate an increase (red) or decrease (blue) in fold change.

View this table:

Loss of ERK inhibitors DUSP4 and DUSP6 corresponds with reactivation of ERK signaling

Upon HGF-induced stimulation of Met, phosphorylation of MAPK kinase (MEK) and ERK occurs rapidly (within 5 min) and peaks before it subsequently returns to baseline within 1 to 4 hours (14). After inhibition of Met kinase by PHA, both MEK and ERK were dephosphorylated in the four gastric cancer cell lines (Fig. 5, A and B, and fig. S8A), whereas phosphorylation of other MAPKs, such as c-Jun N-terminal kinase (JNK) and p38-MAPK, did not decrease (fig. S8, B to D), suggesting that these pathways are Met-independent in gastric cancer cells. Activation of the MEK-ERK pathway increases cyclin D abundance, complex formation between cyclin D and cyclin-dependent kinase 4 (CDK4), subsequent phosphorylation of Rb, release of E2F transcription factors, and expression of E2F gene targets [reviewed in (55)]. Correspondingly, Met inhibition decreased CCND1 and E2F1 transcript abundance and the phosphorylation of Rb in gastric cancer cell lines (Figs. 2H and 5, C and D). Furthermore, 224 Met-dependent E2F1 target genes previously identified in MCF7 cells (54) were differentially expressed upon Met inhibition in the gastric cancer cell lines, with the majority exhibiting a decrease in expression (Fig. 5E and fig. S9A). Of the E2F1 targets previously identified in HeLa cells (54), 214 genes were also differentially regulated in the gastric cancer cell lines (Fig. 5E and fig. S9B). Hence, inhibition of Met decreased signaling through the MEK-ERK pathway and revealed a subset of E2F1 targets that are regulated by Met.

Fig. 5 Met inhibition results in loss of ERK-negative regulators and transient ERK inhibition in Snu-5 and KATO II cell lines.

(A) Western blots of phosphorylated ERK (pERK; Thr202/Tyr204) and total ERK abundance in cells treated with DMSO or 0.1 μM PHA for up to 24 hours. (B) Quantification (means ± SEM) of average phosphorylated ERK/ERK abundance in all four gastric cell lines from (A) at 4 hours in DMSO or PHA (0.1 μM) from three independent experiments. (C and D) Changes in the abundance of (C) CCND1 and (D) E2F1 transcripts after addition of DMSO (dotted lines) or 0.1 μM PHA (solid lines). Expression values are in log2 scale; hypergeometric, P ≤ 0.05. Data are representative of 132 arrays. (E) Number of E2F1-targeted genes that were differentially expressed after addition of 0.1 μM PHA (green circles). E2F1 signature (blue circles) from MCF7 and HeLa cell lines (54) was applied to the genes differentially expressed in gastric cancer cell lines treated with PHA to identify Met-dependent, E2F1-targeted genes. (F and G) Quantification (means ± SEM) of phosphorylated ERK abundance in (F) MKN45 or (G) Snu-5 cells from three independent experiments. (H and I) Changes in amounts of (H) DUSP4 and (I) DUSP6 transcripts in the presence of DMSO (dotted lines) or 0.1 μM PHA (solid lines). (J) Western blots of DUSP4 and DUSP6 in Snu-5 and KATO II cells. (K) Western blots of components of the Met-ERK pathway in cells treated with DMSO, U0126 (20 μM), or PHA (0.1 μM) for 24 hours.

By RPPA analysis, the decrease in phosphorylated ERK was sustained in MKN45 cells at 24 hours but actually increased in abundance in KATO II cells (Fig. 4A). Validation by Western blot analysis confirmed this observation and identified that inhibition of ERK phosphorylation was also transient in Snu-5 cells (Fig. 5, A, F, and G) and was restored at later time points (16 and 24 hours) (Fig. 5A). The mechanism for ERK reactivation appeared to be Met-independent because inhibition of Met phosphorylation was sustained at these time points (Fig. 2B). Notably, many genes that encode proteins that inhibit the ERK pathway decreased significantly in the presence of Met inhibitor (Table 3). Of these, the most differentially expressed included DUSP4 and DUSP6, which decreased 10- and 39-fold, respectively, 24 hours after PHA addition (Fig. 5, H and I). DUSP4 and DUSP6 dephosphorylate ERK in the nucleus and cytoplasm, respectively (56). The loss of DUSP4 and DUSP6 transcripts corresponded with a decrease at the protein level within 4 hours of Met inhibition (Fig. 5J). This precedes the recurrence of ERK phosphorylation, supporting that, in the absence of Met activation, loss of negative regulators enables the reactivation of ERK.

Expression of genes encoding DUSPs occurs downstream from growth factor stimulation and is thought to occur through the MAPK pathway as part of a negative feedback loop (56). Consistent with this, decreased abundance of DUSP4 and DUSP6 after MEK inhibition with U0126 for 24 hours was similar to that seen after Met inhibition, with the exception of DUSP6 in Snu-5 cells (Fig. 5K). This indicates that DUSP4 and DUSP6 abundance downstream from Met is primarily dependent on MEK-ERK signaling. Intriguingly, although ERK phosphorylation decreased in KATO II and Snu-5 cells 2 hours after U0126 treatment (fig. S10A), ERK phosphorylation returned to basal abundance within 24 hours, supporting that the loss of negative feedback enables ERK reactivation at later time points. Furthermore, whereas loss of DUSP4 and DUSP6 occurred in all four cell lines upon Met inhibition, reactivation of ERK occurred only in Snu-5 and KATO II cells, suggesting that although loss of ERK inhibitors may facilitate ERK reactivation, an upstream activating event is also required to promote ERK signaling under these conditions.

ERK reactivation is MEK-dependent in Met-inhibited Snu-5 and KATO II cells

Multiple reports have identified a compensatory increase in activation of alternate RTKs after kinase inhibition, including inhibition of Met (10, 5759). Thus, to examine whether activation of another RTK was responsible for ERK reactivation when Met is inhibited, whole-cell lysates from Snu-5 and KATO II cells were subjected to a phosphorylated RTK array. Under these conditions, no significant increase in the phosphorylation of alternate RTKs was observed (fig. S10, B and C). Consistent with this, the decrease in phosphorylation of the Gab1 scaffold protein, which coordinates ERK activation downstream from Met, and other RTKs (6063) is sustained, suggesting that an RTK-independent mechanism is responsible for ERK reactivation (Fig. 6A).

Fig. 6 Met-independent ERK reactivation occurs through MEK kinase activity.

(A) Western blots of Snu-5 and KATO II cells treated with DMSO or 0.1 μM PHA for 24 hours, followed by a 2-hour treatment with the MEK inhibitor U0126 (20 μM MEK) where indicated. (B to D) Changes in (B) phosphorylated ERK (pERK), (C) phosphorylated MEK (pMEK) (Thr286), or (D) DUSP4 and DUSP6 amounts after treatment with inhibitors detailed in (A). Data are means ± SEM from three independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. (E) Western blots of cells treated initially with DMSO or 0.1 μM PHA for 24 hours, followed by inhibitors to MEK (20 μM U0126), Raf (10 μM sorafenib), Tpl2 (10 μM Tpl2i), or a combination of inhibitors for Raf and Tpl2 (S+T). (F) Model depicting the MEK-ERK pathway in gastric cancer cell lines in conditions in which Met is activated (left panel) or inhibited (right panel) in Snu-5 and KATO II cells. a.u., arbitrary units.

MEK is a known activator of ERK proteins (64); hence, we examined whether MEK was responsible for ERK phosphorylation after Met inhibition by PHA. Addition of the MEK inhibitor U0126 abrogated ERK reactivation in the presence of PHA, supporting this hypothesis (Fig. 6, A and B). However, ERK rephosphorylation occurred even in conditions where the phosphorylation of MEK and the abundance of DUSP4 and DUSP6 protein significantly decreased compared with basal conditions (Fig. 6, A to D). Phosphorylation of downstream effectors of ERK, such as p90 ribosomal S6 kinase (p90RSK) and S6 protein, also decreased in response to U0126, further demonstrating that Met-independent ERK signaling is MEK-dependent (Fig. 6A). To confirm that these observations were not inhibitor-specific, we repeated this with other inhibitors of Met (crizotinib) and MEK (selumetinib) that are currently in clinical use. Indeed, we observed that, 24 hours after the addition of crizotinib, ERK was reactivated and that addition of selumetinib abrogated the phosphorylation of ERK (fig. S10, D and E). In addition, crizotinib decreased DUSP4 and DUSP6 protein abundance (fig. S10, D and F), together confirming that MEK-mediated ERK reactivation is independent of Met phosphorylation, whereas DUSP4 and DUSP6 abundance is promoted by Met signaling.

Raf kinases are the most well-characterized upstream activators of MEK proteins (65). However, another MAPK kinase kinase, the tumor progression locus 2 (Tpl2), has also been shown to promote MEK phosphorylation in the absence of Raf activation (66). Thus, to determine the role of either of these kinases in ERK reactivation when Met is inhibited, we applied targeted inhibitors to Raf (sorafenib) and Tpl2 (Tpl2i) singly or in combination to Snu-5 and KATO II cells with or without PHA. In the presence of PHA, Tpl2i ablated ERK phosphorylation in Snu-5 cells, whereas sorafenib alone was sufficient to achieve the same effect in KATO II cells (Fig. 6E). Tpl2i or sorafenib also resulted in corresponding inhibition of downstream signaling proteins, p90RSK and S6 protein (fig. S10G), indicating that Raf or Tpl2 may be activated.

Met-dependent gastric cancer cell lines are dependent on STAT3 activity for proliferation

We have shown thus far that multiple signaling pathways in the MET-amplified gastric cell lines were inhibited upon treatment with the Met kinase inhibitor PHA. A similar decrease in phosphorylation of Akt and STAT3 and reactivation of ERK were observed in Snu-5 and KATO II cells in response to one of two orally bioavailable Met inhibitors, crizotinib or EMD-1214063 (fig. S11, A to J), supporting that the Akt, STAT3, and ERK pathways are regulated by Met. As previously shown, inhibition of Met reduced the ability of the gastric cancer cells to proliferate in both 2D and 3D culture (Fig. 2, D to G). Therefore, we sought to determine the contributions of each of these pathways to Met-dependent cell proliferation. Treatment of the cells with single inhibitors to PI3K (LY294002), Akt (Akti), mTOR (rapamycin), or MEK (U0126) partially reduced cell proliferation (Fig. 7, A and B). Combined treatment with U0126 and either LY294002 or Akti further decreased cell proliferation (Fig. 7A), supporting that the MEK-ERK and PI3K-Akt pathways each partially contribute to Met-dependent cell proliferation. Inhibition of STAT3 with one of multiple inhibitors (Stattic, S3I-201, BP-1-102, and SH454) or by stable knockdown with short hairpin RNA (shRNA) reduced cell proliferation in 2D culture, as well as in soft agar, in which colony growth in response to STAT3 inhibition was comparable to that in response to Met inhibition (Fig. 7, A to E, and fig. S12, A to J). The addition of Stattic to Akti and U0126 cotreatment further decreased cell proliferation (Fig. 7A). To confirm that this was not inhibitor-specific, each of the other targeted STAT3 inhibitors was used and similarly inhibited cell proliferation in the gastric cancer cell lines (fig. S12A). Hence, although Met-dependent gastric cancer cell lines use the Akt and MEK-ERK pathways for proliferation, only STAT3 is an essential mediator for cell proliferation and anchorage-independent growth.

Fig. 7 STAT3 and ERK pathways promote Met-dependent proliferation and Met-independent cell survival, respectively.

(A) Proliferation measured by the MTS assay in cells cultured in different inhibitors (20 μM LY294002, 10 μM Akti, 0.1 μM rapamycin, 20 μM U0126, 10 μM Stattic, or 0.1 μM PHA) for 5 days. Data are means ± SEM from three biological replicates. (B) Model illustrating signaling pathways and specific inhibitors downstream from Met. (C) Average number of soft agar colonies formed by cells cultured with 30 μM BP-1-102 or an equal volume of DMSO. Data are means ± SEM from three independent experiments. (D) Cell proliferation of KATO II cells stably expressing one of five shRNAs to STAT3 or a control vector (pLKO). Data are means ± SEM from three independent experiments. ****P < 0.0001. (E) Representative images of KATO II cells from (D). Cells are highlighted by a yellow mask generated by the IncuCyte ZOOM software. (F) Mean tumor volume ± SEM from 10 mice treated with crizotinib (30 mg/kg), BP-1-102 (3 mg/kg), or control. **P < 0.005, ****P < 0.0001. (G) Average growth rate of established tumors (n = 20) after treatment with crizotinib (30 mg/kg) or control for up to 14 days. ****P < 0.0001. (H and I) Average cell viability of (H) Snu-5 or (I) KATO II cells after treatment with DMSO, 20 μM U0126, 0.1 μM PHA, or a combination (PHA/U0126) determined by the trypan blue assay. Data are means ± SEM from three independent experiments. **P < 0.01, ****P < 0.0001.

Phosphorylation and activation of STAT3 can occur downstream from the Janus kinases (JAKs) or non-RTKs such as Src or Abl (51). To test whether Met-dependent phosphorylation of STAT3 was indirect through one of these kinases, we treated cell lines with inhibitors to JAK1 and JAK2 (ruxolitinib), the Src family kinases (PP2), Abl (imatinib), or both Src and Abl (dasatinib) (fig. S13, A and B). Under these conditions, the phosphorylation of STAT3 was not detectably different from the DMSO control, whereas Met inhibition decreased the abundance of phosphorylated STAT3 (fig. S13, A and B). Thus, the phosphorylation of STAT3 is highly dependent on Met activity and does not occur through other known STAT3-activating kinases, such as JAK, Src, or Abl. Also previously suggested by Boccaccio et al. (67), these data indicate that Met may directly phosphorylate STAT3 in these cells.

To determine the impact of STAT3 inhibition on tumor growth in vivo, we subcutaneously injected KATO II cells into nude mice on either side of the abdomen, and mice were treated with BP-1-102 or crizotinib by daily oral gavage. After 12 days, tumor growth was significantly inhibited in BP-1-102– or crizotinib-treated mice (Fig. 7F). However, whereas BP-1-102 treatment impaired tumor growth, its effect was not as marked as that seen in response to crizotinib (Fig. 7F). Collectively, these data further support that STAT3 signaling is critical for cell proliferation in Met-dependent gastric cancer cell lines.

ERK reactivation promotes cell survival in the absence of Met signaling

Although Met signaling was required for cell proliferation in all four MET-amplified gastric cancer cell lines tested (Fig. 2, D to G), Okajima and Snu-5 cells exhibited the greatest sensitivity to Met inhibition; only a third of these cells remained viable after PHA addition compared with 50% of KATO II and 67% of MKN45 cells (fig. S14A). Thus, PHA was most cytotoxic in Okajima cells but exhibited more of a cytostatic effect in MKN45 cells. This is further confirmed by Western blot analysis, which showed that the abundance of cleaved poly(adenosine diphosphate–ribose) polymerase (PARP), a marker of apoptotic cell death, increased from day 0 to day 7 in PHA-treated Okajima cells but was undetectable in PHA-treated MKN45 cells (fig. S14B).

This may have implications in vivo because the treatment of established tumors with a Met-targeted inhibitor may prevent further proliferation but may not promote tumor regression. To test this, nude mice were injected subcutaneously with KATO II cells and treated with crizotinib by daily oral gavage for 14 days. The growth rate of crizotinib-treated tumors was significantly decreased compared with control mice (Fig. 7G). Furthermore, whereas tumors in the control mice continued to grow, the tumor size of crizotinib-treated mice remained static, on average, and was not significantly different from that at the start of treatment (fig. S14C).

Because ERK reactivation was observed in the Snu-5 and KATO II cells, this may contribute to the sustained cell viability after PHA treatment, especially in KATO II cells. Thus, to test whether dual Met and MEK inhibition might prove to be more cytotoxic, we cultured Snu-5 and KATO II cells with DMSO, U0126, PHA, or U0126 and PHA in combination. After 7 days of treatment, Snu-5 and KATO II cells cotreated with Met and MEK inhibitors exhibited greater cell death than those treated with either inhibitor alone (Fig. 7, H and I). Hence, although Met inhibition has a strong antiproliferative effect in the MET-amplified gastric cancer cell lines in vitro and in vivo, combinatorial treatment with Met and MEK inhibitors could be more effective in treating gastric cancer.


Gastric and gastroesophageal cancers affect 1 million people per year worldwide and are the second most common cause of cancer-related deaths (21, 68). Although there has been some progress in systemic cytotoxic therapy for gastric adenocarcinoma over the past decade (6973), there is still significant room for improvement, particularly given the persistent poor prognosis for patients with metastatic disease. Whereas targeted therapies were developed and incorporated into standard treatment for other cancers, such as lung or breast, these (including Met inhibitors) are only now being examined in the context of gastric and gastroesophageal cancer (36). Our demonstration of increased expression of MET in patients with poor prognosis lymph node metastases in a North American cohort corroborates recent data demonstrating MET amplification in 5% of high-grade, high-stage adenocarcinomas (32). The observation that Met inhibition of multiple gastric cancer cell lines abrogates cell proliferation, and the positive correlation between increased Met protein abundance and Ki67-positive nuclei in primary gastric tumors, demonstrates that Met signaling exerts a strong proliferative role in MET-amplified gastric cancers. However, although MET amplification has been a good indicator of sensitivity to Met inhibitors in gastric cancer cell lines (2), Met inhibitors in clinical trials have been less successful because patients who respond eventually exhibit further disease progression (32). Thus, there is a need to understand the events that occur in response to Met inhibition and how these might better predict resistance.

Met-dependent signaling pathways and biological responses have primarily been studied after transient activation of Met by its ligand, HGF. The major signaling pathways activated by Met in response to ligand include the Ras-MEK-ERK, PI3K-Akt, as well as JNK and p38 MAPKs, STATs, and RHO guanosine triphosphatases (6). Less well characterized are signaling pathways required for the Met-dependent tumorigenic growth of cancer cells, where Met is constitutively hyperactive in the absence of ligand (2, 10, 74, 75). Met inhibition using several small-molecule inhibitors revealed that activation of the Akt, MEK/ERK, and STAT3, but not JNK and p38 MAPKs, is dependent on Met kinase in all gastric cell lines tested harboring MET amplification. Specificity for Met rather than the related kinase Ron, which is sensitive to PHA at the concentrations used here, was confirmed using the highly specific Met inhibitor EMD-1204063 (76). Using inhibitors for each of these downstream pathways, we identified a critical role for STAT3 alone, but not MEK or Akt alone, for Met-dependent cell proliferation, anchorage-independent growth, and tumor formation, identifying STAT3 as a key Met-dependent signal for gastric cancers. Met-dependent tyrosine phosphorylation of STAT3 has been proposed previously (37, 67, 77), and in support of this, tyrosine phosphorylation of STAT3 in the gastric cancer cell lines tested was Met-dependent and Met-independent of other known regulators of STAT3, including JAK, Src, and Abl. Dual inhibition of MEK and Akt suppressed proliferation greater than inhibition of either kinase alone, indicating convergence of signals for proliferation. This is in agreement with Bertotti et al., who identified a role for Akt- and Ras-dependent pathways for Met-dependent proliferation in MET-amplified tumor cells (75). Consistent with tyrosine phosphorylation on STAT3 and the dependence upon STAT3 for tumor formation, activation of STAT3-dependent gene signatures was observed in all MET-amplified gastric cancer cell lines examined. Although a role for STAT3 in survival of Met-dependent gastric cell lines was known (37), the demonstration of a requirement for STAT3 for tumor growth of Met-addicted gastric cancers now identifies STAT3 as a suitable target for gastric cancers with activation of Met.

Cancer models where negative feedback loops are up-regulated to mitigate oncogenic signaling may be more susceptible to targeted therapeutic resistance through loss of these feedback loops (78). Although the Met inhibitor crizotinib effectively inhibited tumor growth of gastric cell lines, it was unable to promote appreciable regression of established tumors, suggesting that other events promote cell survival in the absence of Met kinase activity. Loss of negative feedback loops upon Met inhibition may facilitate signaling by cell survival pathways. Indeed, we observed loss of negative feedback loops at multiple levels of Met-dependent signaling pathways. These include negative regulators of STAT3 signaling, cytokine-inducible SH2-containing protein (CISH), and suppressor of cytokine signaling (SOCS) 2 and 3 (79), as well as negative regulators of the MEK-ERK pathway (the dual-specificity phosphatases DUSP4 and DUSP6), which target ERK. DUSP expression is Met- and MEK-dependent, indicating that the constitutive activity of the MEK-ERK pathway in these gastric cell lines induces stable expression of multiple DUSPs as part of inhibitory feedback loops. Reactivation of MEK-ERK phosphorylation, but not STAT3 phosphorylation, was observed in two of four gastric cell lines (Snu-5 and KATO II) after short-term inhibition of Met (24 hours), providing a potential survival signal. DUSP4 and DUSP6 proteins decrease in abundance in all gastric cell lines tested after Met inhibition, indicating that loss of these negative regulators alone may be insufficient to promote ERK phosphorylation. Reactivation of ERK in lung and gastric cancer cell lines has been identified as a mechanism of resistance to long-term treatment with Met inhibitors as a consequence of RTK switching to an activated EGFR (80, 81). However, we fail to observe enhanced activity of other RTKs, including EGFR, after short-term Met inhibition, implicating other potential mechanisms.

An increasing body of work supports the robustness of the ERK pathway in various cancer cell models and its role in therapeutic resistance. Reactivation of the ERK signaling pathway may be a common mechanism through which resistance to Met and kinase inhibitors can occur. Although we demonstrate that the loss of negative DUSPs contributes to ERK reactivation, other mechanisms promoting ERK activity and conferring resistance to Met inhibitors exist. Resistance to Met inhibition in GTL-16 gastric cancer cells can occur through the formation of a BRAF fusion protein or KRAS amplification that promotes ERK signaling (8284). The ERK pathway can promote resistance through diverse mechanisms, including Raf-independent MEK activation (to Raf inhibition) (66), KRAS, NRAS, BRAF mutations (to RTK inhibitors) (8588), upstream RTK signaling (to Raf or MEK inhibition) (58, 59, 89), KRAS, BRAF, NRAF amplification (to MEK inhibition), and allosteric activation of Raf (to Raf inhibition) (90, 91). Reactivation of ERK occurred by 24 hours after treatment with three different small-molecule inhibitors of Met, including EMD-1214063 and crizotinib, which are both orally bioavailable, and the latter is currently being tested clinically (32, 76). Hence, because MEK-ERK activation can promote resistance to RTK inhibitors and RTK activation can likewise circumvent Raf or MEK inhibition (58, 59, 8589), dual RTK-MEK inhibition may be an attractive therapeutic combination. In support of this, combinatorial Met-MEK inhibition increased cell death when compared to treatment with each inhibitor alone. Enhanced apoptosis upon dual EGFR-MEK therapy has been observed (86, 92). Thus, combinatorial inhibition of Met and MEK may be a potential strategy to prevent resistance to Met. As Met inhibitors are progressing through all stages of clinical trials (6, 7), it will become increasingly vital to identify mechanisms of both innate and acquired resistance to these inhibitors so that we can better target patients and tailor treatment plans.

In summary, our data indicate that short-term Met inhibition under conditions of long-term receptor hyperactivity leads to specific silencing of restricted Met signaling pathways, including the MEK-ERK, Akt, and STAT3 pathways. Gastric cancer cells addicted to Met are highly sensitive to STAT3 inhibitors, and the emergence of new orally available STAT3 inhibitors (93) may provide new strategies for treatment of gastric cancers with amplified Met. The observation that multiple kinases are amplified in gastric cancer, including HER2 and EGFR (up to 37% of gastric cancers) (94, 95), which have the capacity to promote tyrosine phosphorylation of STAT3, raises the possibility that STAT3 may be a common target in RTK-dependent gastric cancer.


Antibodies and reagents

Antibody 148 was raised in rabbit against a C-terminal peptide of human Met (96). Antibodies against ribosomal S6 protein and Src were acquired from Santa Cruz Biotechnology. Antibody against actin was obtained from Sigma-Aldrich. Antibodies against phosphorylated Met at Tyr1234/1235, ERK1/2 at Thr202/Tyr204, MEK at Ser217/221, Akt at Ser473, HER2 at Tyr1221/1222, STAT3 at Tyr705, JNK at Thr183/Tyr185, p38 MAPK at Thr180//yr182, p90RSK at Ser380, S6 at Ser235/236, Rb at Ser807/811, and total ERK1/2, MEK, Akt, STAT3, DUSP4, DUSP6, RSK1/2/3, mTOR, JNK, and p38 MAPK were purchased from Cell Signaling Technology. Antibody against Rb was purchased from BD Biosciences. Antibody against phosphorylated Src (Tyr418) was obtained from Covance.

HGF was a gift from Genentech and used at 0.5 nM final concentration. LY294002 was purchased from BIOMOL Research Labs. EMD-1214063, rapamycin, Stattic (STAT3 Inhibitor V), and Tpl2 inhibitors were obtained from Calbiochem/EMD Millipore. U0126 was purchased from Promega, and Met inhibitor (PHA) was a gift from Pfizer (final concentration, 0.1 μM). Dasatinib was obtained from LC Laboratories. Sorafenib, ruxolitinib, imatinib, and S3I-201 were purchased from Selleckchem. Crizotinib was purchased from Active Biochem, and SH454 and BP-1-102 were gifts from P. Gunning.

Cell cultures and viral transduction

T47D 2A, Okajima, and MKN45 cells were cultured as previously described (14, 20). Snu-5 and KATO II cell lines were cultured in RPMI supplemented with 20 and 10% fetal bovine serum (FBS), respectively. Snu-5 cells were also cultured in the presence of 4 mM l-glutamine. Human embryonic kidney (HEK) 293T cells were cultured in Dulbecco’s modified Eagle’s medium with 8% heat-inactivated FBS.

HEK293T cells were used to produce lentiviral supernatants as described at The calcium phosphate method was used for the transfection of HEK293T cells. Infected cells were selected for successful retroviral integration using puromycin (1 μg/ml).

Lentiviral shRNA vectors were retrieved from The RNAi Consortium (TRC) arrayed human genome-wide shRNA collection (TRC-Hs1.0). Additional information about the shRNA vectors can be found at using the TRC number. The following lentiviral shSTAT3 vectors were used: shRNA1, TRCN0000329888; shRNA2, TRCN0000353630; shRNA3, TRCN0000329811; shRNA4, TRCN0000329887; and shRNA5, TRCN0000329886.

Soft agar, cell proliferation, and viability assays

Soft agar assays were performed in six-well plates. Bottom agar (3.2 mg/ml) was used in each well, and 10,000 cells were plated in duplicate into top agar (6 mg/ml) in either untreated, PHA-treated (0.1 μM), or DMSO-treated (equal concentration) conditions. Snu-5 and KATO II cells were cultured in RPMI, and Okajima and MKN45 cells were grown in Temin’s modified Eagle’s medium (Invitrogen). Fresh top agar and medium with or without PHA were replaced twice a week, and colonies were grown for at least 3 weeks. Images of colonies were captured at ×40 magnification, 10 images for each condition, and colony size was measured with INFINITY ANALYZE software (Lumenera). Whole-well images were taken at ×3.5 magnification, and colony number was measured with OpenCFU software ( Soft agar assays were performed in triplicate.

For the MTS assay (in Figs. 2G and 7A and fig. S12A), 5000 cells were plated in triplicate in 96-well plates on day 0 in either inhibitor- or DMSO (control)–treated conditions. Cell proliferation was assayed over 7 days with the CellTiter 96 AQueous One Solution Cell Proliferation Assay kit (Promega) according to the manufacturer’s instructions.

For cell proliferation assays, each cell line was plated in triplicate in 96-well plates at a density of 2000 cells per well. Cells were then cultured in the IncuCyte ZOOM Live Kinetic Imaging System (Essen BioScience) to capture phase-contrast images at ×10 magnification every hour and to measure confluence. All proliferation assays were performed in triplicate.

Cell viability was assessed using trypan blue dye (Invitrogen). Cells (3 × 105 per well) were plated in a six-well dish and treated with the specified inhibitors for the various time points. Cells were then pelleted and resuspended in 250 μl of 1× phosphate-buffered saline (PBS) per well, mixed 1:1 with trypan blue dye, and incubated for 3 min at room temperature. Cells were then counted with a hemocytometer, and dead cells were identified by an inability to exclude the dye.

In vivo xenograft assays

For the tumorigenesis assays, KATO II cells were injected subcutaneously (1 × 106 cells in 100 μl of PBS) into each side of 4- to 5-week-old female nude mice (CD1 nu/nu; Charles River Breeding Laboratories). Mice were administered crizotinib (45 mg/kg, dissolved in water), BP-1-102 [3 mg/kg, dissolved in 1:1 water/poly(ethylene glycol), molecular weight 300 (PEG300) (Sigma-Aldrich)], or control (1:1 water/PEG300) daily by oral gavage. Ten mice were treated per condition.

For xenograft assays, KATO II cells were injected subcutaneously (5 × 105 cells in 100 μl of PBS) as described above, and after tumors were established at day 10, mice were treated with crizotinib or control daily for 2 weeks. Tumors were measured periodically, and mice were sacrificed before tumors reached 1 cm3 or ulcerated.

Western blotting

All cells were harvested in a Triton-glycerol-Hepes (TGH) lysis buffer (20). Proteins were resolved by SDS–polyacrylamide gel electrophoresis, transferred to nitrocellulose membranes, and visualized as previously described (20). For blots that required quantitation, proteins were transferred to polyvinylidene difluoride membranes, blocked with Li-COR blocking buffer (Li-COR Biosciences), incubated with primary antibodies and then infrared (IR)–conjugated secondary antibodies, and detected and analyzed on the Odyssey IR Imaging System (Li-COR Biosciences).

IHC and copy number assays

Human patient tumor and lymph node metastasis samples were selected from a tissue bank of patients who received surgical treatment at the Montréal General Hospital (MGH) and were provided courtesy of V.A.M. and L.E.F. The TMA consisted of 35 patients diagnosed with gastric carcinoma at MGH. IHC was performed with the antibody against total c-Met (SP44; rabbit monoclonal) on a Ventana automated stainer (Ventana Medical Systems Inc.). Met positivity and the percentage of Ki67-positive nuclei were determined with Aperio ImageScope software.

Total genomic DNA was isolated from cell lines with the Qiagen AllPrep Mini Kit. MET amplification was determined by qRT-PCR of the genomic DNA with the qBiomarker Copy Number PCR Assay (assay VPH107-0581564A, SABiosciences) according to the manufacturer’s instructions. qBiomarker Multicopy Reference Assay (assay VPH000-0000000A, SABiosciences) was used as the internal control, and MET amplification in the gastric cell lines was compared to that in the breast cancer cell line T47D (which is diploid for MET) with the ΔΔCT method as outlined by SABiosciences.

Matched hematoxylin and eosin–stained slides for gastric tumor tissue blocks were assessed by a clinical pathologist (V.A.M.), and areas of primarily tumor or normal tissue were marked. The corresponding areas were manually excised from the appropriate FFPE blocks, and DNA was extracted with the QIAamp DNA FFPE Tissue Kit (Qiagen) according to the manufacturer’s instructions. DNA was quantitated with a NanoDrop spectrophotometer (Thermo Scientific); samples with sufficient yield were subjected to qRT-PCR for MET with the qBiomarker Copy Number PCR assay VPH107-0581564A (SABiosciences). All samples were assayed in triplicate.

For selected samples, FISH assays were conducted to confirm qRT-PCR results. Briefly, 4-μm FFPE sections were deparaffinized, treated with 0.2 N HCl and then with 1 M NaSCN for 30 min at 80°C, and digested with a pepsin solution (1500 U/ml, Sigma-Aldrich) for 3 min at 37°C. Sections were then hybridized to a locus-specific identifier MET SpectrumRed probe (Vysis) mixed with CEP 7 SpectrumGreen probe (Abbott). Co-denaturation and hybridization were performed on a ThermoBrite system for 5 min at 73°C and 36 hours at 37°C, respectively. Post-hybridization washes were conducted in 2× SSC/0.3% Igepal for 8 min at 72°C and then in 2× SSC for 1 min at room temperature. After application of 4′,6-diamidino-2-phenylindole counterstain, coverslips were applied to slides, and slides were analyzed with a Metafer system (MetaCyte V3.8.6).

Microarrays, normalization, quality control, qRT-PCR, and copy number PCR assay

Total RNA was isolated with QIAzol and the RNeasy Mini Kit (Qiagen) as per the manufacturer’s instructions. RNA amplification was performed with Ambion AM 1753 Amino Allyl MessageAmp II aRNA kit (Invitrogen). RNA quality was assessed with a Bioanalyzer microfluidics workstation (Agilent Technologies) and quantitated with a NanoDrop spectrophotometer. Total RNA (500 ng) was subjected to a single round of amplification with the Amino Allyl MessageAmp II aRNA kit (Ambion/Life Technologies) according to the manufacturer’s instructions, and the amplified product was labeled with a CyDye Post-Labeling Reactive Dye Pack (GE Healthcare Bio-Sciences Inc.). Cy3-labeled antisense RNA (aRNA) samples (825 ng) were cohybridized with 825 ng of Cy5-labeled reference aRNA (Universal Human Reference RNA) to 4x44K Whole Human Genome Arrays (Agilent Technologies). Slides were washed according to the manufacturer’s instructions and scanned with an Agilent dual-laser scanner (P/N: G2505B). Feature extraction was performed with Agilent Feature Extraction software (FE

In total, 132 arrays were hybridized. The arrays were normalized with normexp (robust multiarray analysis) background correction and quantile between array normalization. All arrays passed quality control checks. Raw and normalized microarray data are available in the Gene Expression Omnibus (GEO) database (accession number GSE54532). In each cell line, for each gene, the expression was normalized to the average expression of the gene at time 0 to account for basal differences in gene expression between the cell lines. To assess differential expression in individual cell lines, in each cell line, for each time point (2, 4, 8, 16, and 24 hours), the differential expression was evaluated between that time point and the initial time point. In some cell lines, there were relatively few arrays done at the 0-hour time point. Thus, the 0.5-hour arrays were pooled with the 0-hour arrays as the initial time point. There were no differentially expressed genes between 0 and 0.5 hour. For combined differential expression, differential expression analysis was also performed at each time point on all four cell lines pooled together. All differential expression analysis was performed with LIMMA in R/Bioconductor. At each time point, a gene was considered as differentially expressed if it had an FDR-corrected P value less than 0.05 and a relative fold change greater than 2. If multiple probes for a gene were differentially expressed, the probe with the earliest differential expression was considered. To obtain average expression profiles for each of the different time points, the expression profiles of genes with a fold change greater than 2 at that time point were averaged and are depicted in Fig. 3B.

Statistical analysis

Statistical analyses were performed with a two-tailed Student’s t test (Fig. 7G) or analysis of variance (ANOVA) and Tukey’s (Figs. 1, C and D, 2E, and 7, H and I, and fig. S2), Dunnett’s (Figs. 6, C and D, 7, D and F, and figs. S12, D, G, and J, and S14C), or Sidak’s (fig. S14A) multiple comparisons test. Analyses were performed with GraphPad Prism software.


Fig. S1. Met IHC staining ranges in intensity in gastroesophageal primary tumors and lymph node metastases.

Fig. S2. Met inhibitor decreases the number of soft agar colonies.

Fig. S3. HGF-induced IEGs decrease upon Met inhibition.

Fig. S4. DUSP4 and DUSP6 are induced upon growth factor stimulation.

Fig. S5. Late expression of genes after HGF stimulation and PHA inhibition indicates proteins involved in the abrogation of cell proliferation.

Fig. S6. FOXO3A expression and downstream genes are differentially regulated upon Met inhibition.

Fig. S7. STAT3-targeted genes are differentially expressed in gastric cancer cell lines upon Met inhibition.

Fig. S8. Not all MAPK pathways are dependent on Met kinase activity.

Fig. S9. E2F1-targeted genes are differentially expressed in gastric cancer cell lines upon Met inhibition.

Fig. S10. DUSP loss upon Met inhibition is MEK-dependent, and subsequent ERK reactivation promotes phosphorylation of downstream effectors.

Fig. S11. Met inhibitors crizotinib and EMD-1214063 recapitulate effects on signaling observed with PHA.

Fig. S12. Inhibition of STAT3 hinders cell proliferation.

Fig. S13. STAT3 phosphorylation is dependent on Met kinase activity.

Fig. S14. Met inhibition promotes variable cytotoxicity in gastric cancer cell lines.

Table S1. MET amplification in primary gastric tumors.


Acknowledgments: We thank members of the Park laboratory for helpful comments on the manuscript. The Met inhibitor PHA was a gift from Pfizer. HGF was donated by Genentech. BP-1-102 and SH454 were gifts from P. Gunning. Patient samples were provided by V.A.M. and L.E.F. Snu-5 and KATO II cells were gifts from D. Haber. We also thank J. Lavoie for assistance with FISH analysis. Funding: This research was supported with funds from the Canadian Institutes for Health Research (CIHR) to M.P., a CIHR Master’s Award to R.M.P., a CIHR Doctoral Research Award, a Rolande and Marcel Gosselin Graduate Studentship, and a CIHR/Fonds de la recherche en santé du Québec training grant in cancer research (FRN53888) of the McGill Integrated Cancer Research Training Program to A.Z.L. and C.T. S.H. was funded by a CIHR grant (MOP-130540). We acknowledge infrastructure support and technical assistance from the Breast Cancer Functional Genomics Group, which is partially supported by funds from the Terry Fox New Frontiers Program. The RPPA core facility was funded by a National Cancer Institute grant (CA16672). Author contributions: S.C. and A.T. performed the bioinformatics analysis. H.Z. carried out the microarray experiments. B.S. performed the HGF stimulation experiment. C.T. executed the qRT-PCR of genomic DNA. R.M.P. built the TMA. M.-C.G. performed the Met IHC and FISH. V.A.M. scored the Met and Ki67 IHC. E.S.B. and M.N. contributed cell culture expertise. M.G. contributed Western blot expertise. S.H. performed lentiviral transduction and infection of MKN45 cells. A.M. executed the xenograft assays. B.D.G.P. and P.T.G. synthesized the STAT3 inhibitors (BP-1-102 and SH454). A.Z.L. performed lentiviral infection of KATO II and Snu-5 cells, cell culture, cell proliferation and viability assays, biochemistry experiments, and data analysis. A.Z.L. and M.P. conceived, designed, and directed the research. A.Z.L. wrote the manuscript, and C.T., S.C., N.B., L.E.F., M.H., E.S.B., M.G., and M.P. edited the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The accession number for the gene expression microarray data is deposited in GEO (, accession number GSE54532. P.T.G. and the University of Toronto have a patent pending on the STAT3 inhibitor (SH-4-54) used in this work.
View Abstract

Navigate This Article