Capmatinib (INC280) Is Active Against Models of Non–Small Cell Lung Cancer and Other Cancer Types with Defined Mechanisms of MET Activation

A plethora of preclinical and clinical observations spanning several decades has established the receptor tyrosine kinase (RTK) MET (c-Met, cMET, or c-MET) as an oncogene and attractive therapeutic target for cancer therapy (1). Alterations of MET that are thought to be oncogenic include activating mutations, overexpression, gene amplification, and translocations. Furthermore, MET is aberrantly activated in cancer through its only ligand hepatocyte growth factor (HGF). Based on these observations, numerous agents targeting MET or HGF have been discovered and clinically developed to various stages (2). However, the establishment of predictive biomarkers for efficient clinical development of such agents has proven challenging (3). One factor impeding progress in this area is that some clinically studied agents are not MET selective. For example, tivantinib was initially described as a selective MET inhibitor, while later studies revealed that it also acts as a microtubule-disrupting agent, substantially complicating the interpretation of clinical data (4). Likewise, several multikinase inhibitors such as cabozantinib inhibit multiple relevant cancer targets along with MET, such as vascular endothelial growth factor 2 [VEGFR2 (KDR); ref. 5], making it difficult to dissect the contribution of MET inhibition to any observed effects. In addition, multiple mechanisms of MET activation (including mutation, amplification, overexpression, ligand-mediated activation) have been associated with MET dependency in the preclinical literature, some of which are overlapping. Thus, evaluation of multiple biomarkers and definition of appropriate cutoffs is required to predict response to MET inhibitors.

Crizotinib was among the first MET kinase inhibitors that helped gain a clearer understanding of the therapeutic potential of MET inhibition, because its other primary targets such as anaplastic lymphoma kinase (ALK) and ROS1 are only relevant in rare and translocation-defined cancers that generally do not overlap with cancers in which MET is the dominant oncogenic driver (6). Meanwhile, the clinical activity of crizotinib in MET activated lung cancer is well documented, and the acquisition of MET resistance mutations in initially responsive tumors demonstrated conclusively that this activity was indeed due to MET inhibition (7–9).

Capmatinib (INC280, formerly INCB28060) is a highly selective and potent MET inhibitor with in vitro and in vivo activities against preclinical cancer models with MET activation (10). Capmatinib is being tested both as a single agent and in combination in multiple clinical trials that are guided by biomarker-based patient selection criteria. Here, we further elaborate on the preclinical profile of capmatinib and describe data guiding the clinical biomarker strategy.

Capmatinib hydrochloride [2-fluoro-N-methyl-4-(7-(quinolin-6-ylmethyl)imidazo[1,2-b][1,2,4]triazin-2-yl)benzamide dihydrochloride monohydrate, C₂₃H₁₇FN₆O·2ClH·H₂O] was synthesized at Novartis. All other compounds were obtained from commercial sources.

All cell lines were obtained from commercial sources and screened for compound sensitivity in the context of the Novartis/Broad Institute Cancer Cell Line Encyclopedia project (11). The details can be found in the Supplementary Materials and Methods.

Cells were seeded at 2,000 cells per well in 96-well plates in 100 μL per well and incubated for 24 hours at 37°C in 5% CO₂. Capmatinib was then added from a 10 mmol/L DMSO stock solution using a HP D300 Digital Dispenser (Tecan). After 5 days of incubation, Hoechst 33342 and propidium iodide were added to the culture medium at final concentrations of 1 and 2 μg/mL, respectively, and incubated for 45 minutes at 37°C and 5% CO₂. The number of Hoechst 33342-stained nuclei and propidium iodide-stained dead cells was then quantified following image acquisition on a Cellomics VTi automated immunofluorescence microscope (ThermoFisher Scientific) using the appropriate excitation/emission filter sets.

For all studies, animals were housed in a 12-hour light/dark cycle facility and had access to food and water ad libitum. Mice were maintained and handled in accordance with Novartis Institutes for BioMedical Research (NIBR) Institutional Animal Care and Use Committee (IACUC) regulations and guidelines. All studies were approved by the NIBR IACUC.

A detailed description of experimental procedures and calculations can be found in the Supplementary Materials and Methods. In brief, dose matrices were set up in multiwell plates (96 or 384) using a HP D300 Digital Dispenser. Wells were DMSO normalized and randomized to avoid systematic position effects. After incubation with the drugs, effects were quantified either by staining with propidium iodide and Hoechst 33342 or by CellTiter-Glo (CTG; Promega) including a readout for untreated cells (“day 0″). Both methods allowed to quantify the extent of cell killing in the respective experiments.

A model of capmatinib bound to the ATP site was constructed based on the crystal structure of MET in complex with 6-(difluoro(6-(4-fluorophenyl)-[1,2,4]triazolo[4,3-b][1,2,4]triazin-3-yl)methyl)quinoline (PDB code: 5EOB; ref. 12) representative of the binding mode of the class of highly selective MET inhibitors, to which capmatinib belongs. In this binding mode, the imidazotriazine core of capmatinib makes an aromatic stacking interaction with MET residue Y1230 while its quinoline moiety interacts with the hinge region of the kinase. The stacking interaction is made possible by a particular conformation of the kinase activation loop (A-loop) stabilized by a salt bridge between residues D1228 and K1110. An intramolecular hydrogen bond between the amide nitrogen and the fluoro atom of capmatinib is postulated. Additional information can be found in the Supplementary Materials and Methods.

Capmatinib (Fig. 1A) had previously been screened against 57 human kinases and was found to be selective for MET within this panel (10). To extend this kinase selectivity profiling, we measured the affinity of capmatinib in a set of 442 kinases and disease-relevant variants using the KINOMEscan selectivity screening platform. At a screening concentration of 10 μmol/L, which is more than a 1,000-fold above the reported on-target IC₅₀ in biochemical assays (10), nine kinases scored as hits with the predefined cutoff of ≥65% reduction in binding to the capture matrix compared with a vehicle control (Fig. 1B). These hits included MET and two mutant variants thereof. Given that the kinase panel was screened at a concentration of capmatinib that is much higher than its active concentration against MET, we determined the binding constants (K_d) for all nine hits (Fig. 1C). The K_d values for MET and two mutant variants were subnanomolar, and were lower by a factor of approximately 1,000 or more compared with all other hits. Of note, the MET mutations M1250T and Y1235D did not have a notable impact on capmatinib binding. In summary, these data confirm that capmatinib is a highly selective MET inhibitor.

Structural modeling of the MET kinase domain bound with capmatinib revealed that the phenol moiety of Y1230 directly binds to the central aromatic ring of capmatinib in a pi stacking interaction, while D1228 forms a salt bridge with K1110 that stabilizes the MET activation loop in a conformation that is necessary to support the Y1230–capmatinib interaction (Fig. 1D). This binding interaction is similar to crizotinib and other selective MET inhibitors, and although Y1230 and D1228 are conserved in other tyrosine kinases such as IGF1-R and KDR, the required conformation of the activation loop is also stabilized by multiple hydrophobic interactions between residues of the activation loop and residues of helix C that are specific to the MET kinase (13, 14). To validate the structural model experimentally, we made use of a panel of BaF3 cells transformed with TPR-MET constructs bearing MET kinase domain mutations. Some of these mutants had been obtained in an unbiased cellular resistance screen with a selective MET inhibitor that is structurally related to capmatinib (13). As expected, significant resistance was observed when BaF3 cells bearing MET D1228 and Y1230 mutations were treated with capmatinib, while much smaller shifts in the IC₅₀, if any, were seen with other variants (Fig. 1E; Supplementary Table S1). These observations are in line with the proposed structural model of the MET–capmatinib interaction. Importantly, recent clinical case reports documented MET D1228 or Y1230 mutations in lung cancers with acquired resistance to MET inhibitors (7–9, 15).

MET gene amplification, leading to overexpression and autophosphorylation of the MET protein, has been linked to MET inhibitor sensitivity in cell lines (16–19). In addition, response to capmatinib has also been reported in two preclinical models that express both MET and its ligand HGF (10). To assess predictors of response to capmatinib in an unbiased and systematic manner, we tested the activity of capmatinib against more than 600 well-characterized cancer cell lines in the Cancer Cell Line Encyclopedia (CCLE) project (11). Cell line screens were conducted twice independently in a high throughput format, where dose–response curves were generated for capmatinib after a 3-day incubation period. After quality control, we obtained interpretable results for a total of 605 cell lines (458 in the first screen and 364 in the second screen, with an overlap of 217 cell lines; Supplementary Table S2). We considered both the maximal effect (A_max) and the EC₅₀ (inflection point) of the fitted sigmoid dose–response curve to determine sensitivity (Supplementary Fig. S1A). With a low stringency (A_max ≤ −25% and inflection point ≤100 nmol/L), we observed a total of 13 responders or partial responders among all tested cell lines (Fig. 2A). The two screens were largely concordant in terms of capmatinib response for the 217 cell lines tested in both occasions, with the exception of two cell lines that scored as modestly sensitive in one screen and completely resistant in the other. Interestingly, all responsive cell lines except these two discordant lines were characterized by one of two genomic profiles: (i) MET gene amplification, leading to pronounced MET mRNA overexpression (Fig. 2B) or (ii) high expression of the MET ligand HGF (Fig. 2C). The expression of HGF by cancer cell lines may be indicative of an autocrine loop that activates MET in these cells. Indeed, we found a good correlation between HGF mRNA expression and the amount of HGF protein in cell culture supernatants (Supplementary Fig. S1B). Four of the seven cell lines in the autocrine category were derived from glioblastoma, presumably related to the observation that glioblastoma shows frequent gain of chromosome 7 regions encompassing both MET and HGF (20).

Only two MET-amplified cell lines with known dependence on MET (17, 19) displayed profound responses to capmatinib (A_max close to −100%) at low concentrations (inflection point <10 nmol/L). All HGF-expressing cell lines and two of the MET-amplified cell lines showed partial responses (A_max > −60%). In some of the cell lines expressing HGF, the dose–response curve was very shallow, suggesting only a moderate reduction in growth upon MET inhibition under the screening conditions.

To investigate whether these observations are generalizable to selective MET inhibitors, we combined the CCLE screening results of capmatinib with results from three other MET inhibitors in the same screening format, each tested twice independently like capmatinib: crizotinib, JNJ-38877605 (2), and PF-4217903 (14). The latter two compounds are highly selective MET inhibitors with chemical structures similar to capmatinib. For crizotinib, cellular activity explainable by ALK translocations was disregarded for this combined analysis. An overall number of 709 cell lines could be analyzed that had been tested in more than one screen. Sensitive cell lines (“hits”) were scored as for capmatinib, but adapting the inflection point cutoff to the relatively lower potency of the other inhibitors. A total of 16 hits were observed that scored with more than one MET inhibitor and included 10 of the hits previously identified with capmatinib alone (overall hit rate 16/709 = 2%; Supplementary Fig. S1C; Supplementary Table S3). All hits were associated with high expression and/or copy number of MET or they coexpressed MET and HGF. When defining thresholds for those biomarkers guided by the hit with the respective lowest value, we noted that the hit rate among cell lines with high MET copy number (amplified) was relatively high (4/6 = 67%), followed by cell lines showing MET overexpression (5/9 = 56%, four of these five also amplified), suggesting that these biomarkers, which are largely overlapping, might be suitable predictive markers for a selective MET inhibitor (Supplementary Fig. S1C). Conversely, among the cell lines with MET/HGF coexpression (putative autocrine), the hit rate was lower (11/32 = 34%), which could be due to at least two factors: (i) maximal growth inhibition in this category was mostly modest, which makes detection in a high-throughput screen less likely. (ii) HGF-mediated MET activation does not lead to MET-dependent growth in a fraction of these cell lines.

Clinically, response to MET inhibitors has been observed in patients with lung cancer whose tumors contained mutations leading to MET exon 14 skipping (21). In our tested cell line panel, two models contained such mutations: the gastric cancer cell line Hs 746.T and the lung cancer cell line NCI-H596. Hs 746.T responded to capmatinib treatment in vitro, but MET is also highly amplified in this cell line. Thus, it is difficult to assess the contribution of MET exon 14 skipping to capmatinib sensitivity in this model. NCI-H596 cells were resistant to MET inhibition in vitro. However, in this cell line, we observed more persistent MET phosphorylation in response to HGF stimulation (Supplementary Fig. S1D), which is consistent with the reported functional consequence of MET exon 14 deletion (22).

Dependency on MET was evaluated genetically in a large-scale pooled short hairpin RNA (shRNA) screen across 398 cell lines interrogating cell-autonomous dependencies of 7,837 genes each targeted by 20 shRNAs (23). As in the screen with capmatinib, only the two MET-amplified cell lines EBC-1 and MKN-45 showed strong dropout that was clearly distinct from the rest of the screened cell lines (Fig. 2D). Autocrine lines were enriched among the cell lines with MET-dependent growth, but the signal was less pronounced. No clear dependencies were detected upon HGF knockdown (data not shown). This is generally expected for genes encoding secreted factors, because in a pooled shRNA screening format only a tiny fraction of cells will bear shRNAs that target HGF, with negligible impact on the total level of HGF protein in the cell culture medium.

Combining our pooled shRNA screening data with two additional published screens strengthened the link between MET amplification and MET dependency (Supplementary Fig. S2A). Interestingly, a publicly available genome-wide CRISPR screen revealed a marked MET-dependency signal for several cancer cell lines expressing HGF, unlike the RNAi data sets (Supplementary Fig. S2B). This finding recapitulates the previously observed responses to capmatinib and other MET inhibitors seen in autocrine cell lines. Conversely, the apparent MET dependency of MET-amplified cell lines was much less pronounced in the CRISPR screen, which is likely explained by the need to computationally adjust dependency scores for amplified genes (24, 25). The more sensitive detection of dependencies in HGF-expressing cell lines may be related to a superior signal-to-noise ratio of CRISPR versus RNAi, enabling the detection of more subtle effects on growth.

In summary, all genetic MET dependencies and responses of cell lines to capmatinib and other selective MET inhibitors can be explained by either very strong MET overexpression, mostly as a consequence of MET gene amplification, or by coexpression of MET and its ligand HGF.

The MET-amplified lung cancer cell line EBC-1 was found to be exquisitely sensitive to MET inhibition in our cellular screens. This was confirmed by measuring the impact of a diverse series of clinically relevant MET inhibitors on proliferation of this cell line (Supplementary Fig. S3A). Each of the MET inhibitors caused profound inhibition of proliferation though with different potencies.

We then confirmed the capmatinib sensitivity of the EBC-1 cell line in vivo (Fig. 3A). Remarkably, even large EBC-1 xenograft tumors underwent pronounced regression upon treatment. To further characterize the activity of capmatinib in lung cancer in vivo, we first analyzed the Novartis patient-derived xenograft models (PDX) collection (26), but did not identify any lung cancer models with MET amplification or exon 14 skipping mutations (data not shown). Therefore, we turned to an external well-annotated PDX collection (27) of 66 lung cancer PDX models with gene expression data (by Affymetrix HG U133 plus 2.0 array and RNA-seq), gene copy number (GCN; by Affymetrix SNP 6.0 array), and whole exome sequencing data (Supplementary Table S4). The measurements of MET mRNA expression by Affymetrix array and RNA-seq were in excellent agreement, and we chose the three lung adenocarcinoma models with highest MET expression for further studies (Supplementary Fig. S3B). High total and phospho-MET protein levels had also been observed for two of those models (27). Interestingly, MET gene copy numbers were more distinct, with high-level amplification in two models (14 and 11 copies in LXFA 526 and LXFA 1647, respectively, as part of 1–2 Mb amplicons) and only moderate, very broad copy number gain in the third model (LXFA 623; Supplementary Fig. S3C). This constellation enabled us to investigate whether high MET expression in the absence of amplification could be sufficient to predict response to capmatinib. Indeed, all three models underwent profound regression upon MET inhibition with capmatinib (Fig. 3B), including complete responses in a subset of mice for two models (Supplementary Fig. S3D). Treatments were well tolerated as far as determined by body weight monitoring (Supplementary Fig. S3E). However, all tumors grew back after cessation of treatment on day 21, indicating persistent disease.

The pharmacodynamic effect of capmatinib was measured at the end of the study by quantifying total MET and phospho-MET in tumor lysates using a multispot ELISA. LXFA 623 tumors showed markedly lower total and phospho-MET levels than the 2 MET-amplified models (Fig. 3C; Supplementary Fig. S3F). MET inhibition was clearly detectable at 2 hours after the last dosing, with some degree of phospho-MET recovery in two of three models at 12 hours after dosing.

In a third PDX model collection, a lung cancer model named LU5381 with MET exon 14 skipping mutation and moderate MET copy number gain (∼5) was identified, thus dissociating MET exon 14 skipping from high-level MET amplification. When treating mice bearing LU5381 xenografts with capmatinib, we observed tumor regression (Fig. 3D; Supplementary Fig. S3G).

Notably, capmatinib was also active against a liver cancer xenograft model, in which the MET gene is amplified (Fig. 3E; ref. 16).

In the in vitro screens, putative autocrine cell lines generally showed relatively subtle responses to capmatinib treatment (Figs. 2A–C). Yet, the in vivo response of xenografts derived from such models was much more dramatic, as exemplified by the glioblastoma cell line U87-MG (10). Thus, experimental conditions can have a strong impact on the apparent sensitivity of such preclinical models. Regression of additional MET/HGF autocrine glioblastoma xenografts in response to MET inhibitors had been reported previously (28). When we treated xenografts of the gastric cancer cell line IM95, which expresses higher levels of HGF mRNA than U87-MG and produced comparable amounts of HGF as detected in cell culture supernatants (Supplementary Fig. S1C), a significant growth reduction but no regression was observed (Supplementary Fig. S3F). This result confirms that HGF-expressing cancer models can show pronounced responses to capmatinib in vivo, but the level of HGF expression does not appear to be sufficient to make quantitative predictions about response depth.

We analyzed the response of two MET-amplified lung cancer cell lines EBC-1 and NCI-H1993 (17) to capmatinib in more detail, aiming to distinguish growth arrest from cell death. To this end, we quantified total and dead cells by automated microscopy using specific fluorescent dyes. Interestingly, EBC-1 cells displayed a markedly higher rate of cell death upon capmatinib treatment, albeit not reaching 100%, whereas the effect in NCI-H1993 was largely restricted to inhibition of proliferation (Fig. 4A). This observation indicates that the reductions of growth and viability following MET inhibition are not always strictly coupled. Next, we studied the effect of MET inhibition on cellular signaling in these two MET-amplified lung cancer cell lines. As expected, MET phosphorylation as well as phosphorylation of AKT and ERK were suppressed at low single-digit nanomolar concentrations of capmatinib in both cell lines (Fig. 4B). In line with the effects on cellular proliferation, suppression of protein phosphorylation occurred at slightly lower concentrations in EBC-1 than in NCI-H1993, but the maximally achievable effects were comparable. Thus, the cellular phosphorylation events studied here do not provide an obvious explanation for the observed differences in cell death upon capmatinib treatment.

Intrigued by the observation that capmatinib arrests growth of MET-amplified NCI-H1993 cells but failed to induce cell death, we tried to improve this outcome using combination treatments. We reasoned that cotargeting members of the BCL2 family of antiapoptotic proteins might be a good starting point. We used previously described selective inhibitors of BCL2, BCL2L1 (BCL-xL), or MCL1 (29–31) and combined them with capmatinib in a concentration matrix followed by direct quantification of cell death using propidium iodide and Hoechst 33342 staining. Combined inhibition of MET and either MCL1 or BCL2L1 led to synergistic killing of a substantial fraction of cancer cells (Fig. 4C; Supplementary Fig. S4A), whereas combined BCL2 inhibition was inactive (Supplementary Fig. S4A). Yet, under the tested conditions not all cancer cells were killed even with combination treatment. We also examined the effect of the same combinations in EBC-1 cells, although in those cells capmatinib on its own is already inducing pronounced cell death. Interestingly, however, the fraction of dead cells was further increased by concomitant MCL1 or BCL2L1 inhibition (Supplementary Fig. S4B).

The combination of a selective MET inhibitor with the microtubule-stabilizing chemotherapeutic docetaxel was found to be active against MET-amplified gastric cancer models (32). Independently, we observed during a systematic combination screen that docetaxel and chemotherapeutics with related mode of action were active in combination with the EGFR tyrosine kinase inhibitor nazartinib in EGFR-mutant lung cancer models (manuscript in preparation). Therefore, we tested the combination of capmatinib and docetaxel in the two available MET-amplified lung cancer cell lines, EBC-1 and NCI-H1993 (Fig. 4D; Supplementary Fig. S4C). In both cell lines, a synergistic boost of cell killing was observed. The EGFR inhibitor erlotinib had previously been reported to prevent outgrowth of resistant EBC-1 cells upon prolonged MET inhibition (33). In line with this report, the combined treatment of EBC-1 cells with erlotinib and capmatinib further increased cell killing similar to the docetaxel combination (Supplementary Fig. S4D), whereas the added benefit of erlotinib against NCI-H1993 was modest (data not shown). In summary, the activity of capmatinib against MET-amplified tumors can be further enhanced by several combination partners with distinct mode of action.

Although cancer models that depend primarily on MET alone are relatively infrequent (Fig. 2A and Supplementary Fig. S1C), MET has also been reported to cause acquired or adaptive resistance to other targeted therapies, which is the basis for an important additional clinical application of MET inhibitors. For example, in lung cancer with EGFR activating mutations, the activation of MET can bypass EGFR dependency, causing resistance to EGFR inhibitors. This was first discovered in an EGFR-mutant lung cancer cell line named HCC827, which contains a minute fraction of MET-amplified subclones that grow out under treatment with EGFR inhibitors (34, 35). The clinical relevance of this resistance mechanism has hence been confirmed in numerous clinical studies.

Using parental HCC827 cells and gefitinib-resistant derivatives (GR) bearing MET amplification, we confirmed that capmatinib can revert gefitinib resistance in the GR variant in vitro, while not adding to the effect of gefitinib in parental cells (Supplementary Fig. S5A). Capmatinib also had a subtle but measurable effect on the growth of HCC827 GR cells as a single agent. Interestingly, when testing the same combination in vivo using HCC827 GR xenografts, we observed a relatively strong antitumor effect of capmatinib even as a single agent, leading to stasis for more than 3 weeks until tumors started to progress again (Fig. 5A). However, combination treatment led to profound and sustained tumor regression. Similar results were obtained when treating a MET-activated HCC827 xenograft derivative with a combination of capmatinib and the third-generation EGFR inhibitor nazartinib (EGF816; ref. 36). Besides MET amplification, activation of MET via its ligand HGF has been proposed as another potential mechanism of resistance to EGFR inhibitors in lung cancer (37), and indeed we confirmed that addition of exogenous HGF to two EGFR-mutant lung cancer cell lines could substantially reduce growth inhibition by gefitinib (Supplementary Fig. S5B).

We hypothesized that—analogous to EGFR-mutant lung cancer—MET may also drive resistance to ALK inhibition in ALK-translocated lung cancer. Although this potential resistance mechanism is not expected in patients treated with the dual MET/ALK inhibitor crizotinib, it may be relevant in patients treated with second-generation selective ALK inhibitors. In support of this hypothesis, we noted in our PDX collection a lung cancer model with EML4-ALK translocation that expressed very high MET mRNA levels without MET amplification, and high phospho-MET protein levels (Supplementary Fig. S5C). Although this model was responsive to crizotinib (Supplementary Fig. S5D), it did not respond to the second-generation ALK inhibitor ceritinib, but regressed when ceritinib was combined with capmatinib (Fig. 5B).

The ability of HGF to diminish the effect of kinase inhibition through MET activation has also been described in several other contexts beyond EGFR-mutant lung cancer (38–40). For example, HGF can reduce the effect of ERBB2 inhibition in ERBB2-amplified cancers. In keeping with a previous report (40), we observed no or partial rescue by exogenous HGF in four ERBB2-amplified breast cancer cell lines (data not shown). In an ERBB2-amplified lung and gastric cancer cell line, however, which were both sensitive to lapatinib, the effect of HGF was more pronounced, in particular by enhancing overall growth but also reducing the maximal inhibitory effect of lapatinib (Fig. 5C). Interestingly, the esophageal cancer cell line OE33, which is MET-amplified and partially sensitive to capmatinib (Fig. 2A), also displays ERBB2 amplification and high ERBB2 mRNA expression, suggesting that both RTKs could be activated (41). In support of this hypothesis, combined treatment with capmatinib and lapatinib resulted in more pronounced growth inhibition than either single agent alone (Fig. 5D).

Another example where HGF was reported to drive resistance is BRAF-mutant melanoma treated with BRAF inhibitors (39, 40). Although HGF may be produced by noncancer cells in the tumor microenvironment, such as cancer-associated fibroblasts, we also identified a BRAF-mutant colorectal cancer cell line (RKO) where autocrine MET activation may play a role: the modest growth inhibitory effect upon targeting mutant BRAF signaling with dabrafenib plus trametinib in these cells could be enhanced by capmatinib treatment, albeit not to an extent that resulted in cell killing (Supplementary Fig. S5E).

In summary, activation of MET, either by direct alterations of the MET gene itself or through HGF, can cause resistance to various kinase inhibitors, which may substantially expand the clinical utility of a MET inhibitor like capmatinib in combination therapies.

Systematic screening across broad cancer cell line panels revealed that sensitivity to the selective and highly potent MET inhibitor capmatinib and/or genetic MET dependency can be explained by distinct mechanisms of MET activation that could serve as predictive biomarkers. Among those, MET amplification and pronounced MET overexpression were associated with robust sensitivity to capmatinib in vitro and in vivo. The percentage of models with these two MET-activating features is low across cancer types, indicating that a very stringent patient selection approach might be needed in contrast to the approach taken in several previous negative clinical trials with MET-targeting agents. Furthermore, MET-amplified models generally also displayed overexpression, whereas the reverse was not always true, raising the question whether MET expression or MET GCN is the more efficient predictive biomarker.

These observations formed the basis for clinical exploration of capmatinib with an initial focus on patient selection markers and cutoffs. A phase I study examined the predictive value of MET expression (IHC) versus MET GCN (FISH) in a lung cancer expansion cohort and reached the conclusion that GCN-based selection will likely result in a higher response rate (manuscript in preparation; ref. 42). GCN-based selection is now further refined in a phase II study with cohorts covering several GCN ranges. This study is also recruiting lung cancer patients whose tumors bear MET exon 14 skipping mutations (METex14), which partially overlaps with MET amplification (43). The predictive value of METex14 has likely been underestimated preclinically due to the lack of models and overlap with amplification, and only emerged as a potential stratifier based on clinical evidence and exome sequencing data from very large cancer sample sets (21). This case illustrates that even the most extensive cancer model collections (e.g., CCLE) do not cover every possible cancer dependency. The incidence of this genetic alteration in lung cancer is nearly 3% (21, 44), whereas the incidence of “MET amplification” is a function of the determined copy number cut-off, and will need to be defined in ongoing trials. Additional candidate biomarkers that require clinical exploration for lack of preclinical models are MET activating kinase domain mutations (45) and MET chromosomal rearrangements (46, 47).

Capmatinib was also investigated clinically as single agent in liver cancer, revealing that both MET amplification and MET overexpression can contribute to the pre-selection of responding tumors (manuscript submitted; ref. 48). No clinical trials with capmatinib have yet been performed that utilized HGF as selection marker, in part due to the finding that the majority of presumable autocrine models displayed only minor growth reductions under treatment in vitro (45).

Not all models bearing predictive MET alterations respond to capmatinib to the same extent. This is illustrated by the MET-amplified NCI-H1993 cell line that fails to undergo cell death upon MET inhibition. Of note, NCI-H1993 was derived from a metastasis, whereas another cell line (NCI-H2073) was derived from the primary tumor of the same patient and lacks MET amplification (49), highlighting that MET amplification is not always a truncal event, and that it may be important to determine whether it is present as a clonal rather than subclonal event in enrolling patients. In support of this notion, a recent clinical report on the activity of the MET inhibitor AMG337 in esophagogastric cancer described that MET amplification was detected solely in a metastasis but not the primary tumor in two of six cases, where it appeared to be associated with less clinical benefit (51).

Several capmatinib combinations are being tested in clinical trials. The concept of combining capmatinib and EGFR inhibitors in EGFR-mutant lung cancer with MET dysregulation is clinically validated (52) and has been explored in further trials (NCT02468661, NCT02335944). However, our preclinical data suggest that capmatinib combinations can be effective beyond EGFR-targeting agents, both in tumors where MET is the dominant oncogenic driver, and in tumors with other co-occurring drivers. Exemplifying the former category, we observed that combinations with BH3 mimetics and docetaxel enhance the anticancer activity of capmatinib in MET-amplified lung cancer models. In addition to the role of MET as a cancer cell-autonomous driver, MET activation in immune cells has been linked to immune suppression via various mechanisms (54), and a recent study showed that capmatinib can enhance the activity of various cancer immune therapies (manuscript in preparation; ref. 55). The combination of capmatinib with anti-PD1 antibodies is currently being evaluated in two clinical trials (NCT02323126, NCT02795429).

S. Baltschukat, B.S. Engstler, H.-E.C. Bhang, F. Hofmann, and R. Tiedt hold ownership interest (including patents) in Novartis. W.R. Sellers holds ownership interest (including patents) in Novartis and is a consultant/advisory board member for Servier Pharmaceuticals, Sanofi Pharmaceuticals, and Array Biopharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Huang, J. Liang, H.-E.C. Bhang, Y. Wang, R. Tiedt

Development of methodology: H.Q. Wang, J. Liang, R. Tiedt

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Baltschukat, A. Tam, H.Q. Wang, J. Liang, M.T. DiMare, H.-E.C. Bhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Baltschukat, A. Huang, H.-X. Hao, H.Q. Wang, J. Liang, H.-E.C. Bhang, Y. Wang, P. Furet, R. Tiedt

Writing, review, and/or revision of the manuscript: S. Baltschukat, A. Huang, H.-X. Hao, H.Q. Wang, J. Liang, M.T. DiMare, H.-E.C. Bhang, Y. Wang, P. Furet, W.R. Sellers, F. Hofmann, J. Schoepfer, R. Tiedt

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.S. Engstler, J. Schoepfer

Study supervision: A. Huang, J. Liang, H.-E.C. Bhang, W.R. Sellers, F. Hofmann, R. Tiedt

The authors thank Christopher J. Wilson and team for conducting the large-scale cancer cell line screens with capmatinib and the Novartis DRIVE team for conducting the pooled shRNA screen. They also thank Chen Liu for technical assistance in the RKO experiments, Markus Wartmann and Andreas Hueber for help with live/dead cell imaging, and Sabine Zumstein-Mecker for help with EBC-1 combination experiments. PDX studies with the models LXFA 526, LXFA 623, and LXFA 1647 were conducted at Charles River Laboratories (former Oncotest), Freiburg, Germany. The LU5381 PDX study was conducted at Crown Biosciences, San Diego, California. The HCC827GR derivatives used in this study were kindly provided by Jeffery Engelman and Pasi Jänne. The authors thank the capmatinib global project team as well as Peter Hammerman for review and helpful comments on this manuscript and Pushkar Narvilkar, Novartis Healthcare Pvt. Ltd., for providing medical editorial assistance. These studies were sponsored by Novartis.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.