2.1 Patient samples and next generation sequencing

Patients at Memorial Sloan Kettering Cancer Center with alterations in MET were identified. Patients’ tumours were sequenced using the MSK-IMPACT next generation sequencing platform, which at the time covered up to 468 genes of interest,^{34, 35} as well as with the RNA-based ARCHER fusion panel.^{36, 37} Liquid biopsies were performed using our institution's MSK-ACCESS assay, which covers 169 genes and accounts for paired leukocyte sequencing.³⁸ Additional patients with MET alterations were identified from our clinical sequencing database, which amalgamates data on sequencing of tumours and/or ctDNA.^{39, 40} This study was approved by the Institutional Review Board of Memorial Sloan Kettering Cancer Center.

2.2 Compounds

Crizotinib (HY-50878), capmatinib (HY-13404), cabozantinib (HY-13016) and foretinib (HY-10338) were purchased from MedChemExpress.

2.3 Immunohistochemistry

Immunohistochemical staining for MET was carried out with the SP44 clone (Ventana/Roche), according to previously described methods.^{41, 42}

2.4 In vitro kinase assays

Recombinant MET wild type and G1090A, Y1230S and Y1230H mutant kinase domains were synthesised by Signalchem (Signalchem Biotech Inc.). Kinases were tested against a titration of crizotinib, capmatinib and cabozantinib (with 2 µM being the highest concentration and 1:3 serial dilutions) in duplicate in the presence of 10 µM fixed ATP concentration by Reaction Biology (Reaction Biology Corporation). Results are presented as % activity (mean ± SD), with activity of 100% in the untreated kinases used as controls.

2.5 Plasmid generation, cell culture and Western blots

The CD47-MET, the CD47-MET G1090A and the CD47-MET Y1230S-encoding PUC-GW-Kan pDONR plasmids were generated by Genewiz. Gateway™ LR Clonase™ II Enzyme mix (Invitrogen; #11791020) was used to clone the above from pDONR into the plenti302 destination vector. These vectors were used to transiently transfect HEK-293T cells or H3122 cells (ALK+ NSCLC) using Lipofectamine™ 3000 Transfection Reagent (Thermo Fisher; #L30000008) according to the manufacturer's protocol. Twenty-four hours later transfected cells were treated with the indicated concentrations of the different MET inhibitors for 30 min. After incubation, cells were frozen. The day after incubation, protein lysates were extracted, quantified and used for Western blots. The following antibodies were used: Met (D1C2) XP® Rabbit mAb (CST, #8198S), Phospho-Met (Tyr1234/1235) (D26) XP® Rabbit mAb (CST, #3077S), ERK (CST, #4695S), pERK T202/Y204 (CST, #9101S) and actin (CST, #4970S). Experiments shown are representative of three biological replicates. Western blots quantifications and analyses were performed using GraphPad 9.2.

2.6 Docking and molecular dynamic simulations

Available structures of MET kinase were used to construct a structural model of the kinase domain of MET (MET) in the active (PDB: 3Q6W,) and inactive forms (PDB: 6SD9). Structural models of MET–inhibitor complexes were generated using the co-crystal structures of either the inhibitor (PDB: 2WGJ: Crizotinib; 5EOB: capmatinib, 5HTI: cabozantinib, 6SD9: foretinib, or of an analogue of the inhibitor (compounds closely related to the inhibitor) bound with MET kinase. Models of the mutant MET kinases in its apo form and in complex with inhibitors were generated using the corresponding wild type structures. The resulting models were further refined using MD simulations with the pmemd.cuda module of the program Amber20.⁴³ The partial charges and force field parameters for each inhibitor were generated using the Antechamber module in Amber20. All atom versions of the Amber14SB force field (ff14SB)⁴⁴ and the general Amber force field (GAFF)⁴⁵ were used for the proteins and the inhibitors, respectively. MD simulations were carried out with a standard protocol.⁴⁶ Simulation trajectories were visualised using VMD⁴⁷ and PyMOL.⁴⁸ Analyses were carried out using in-house scripts and the ptraj module of Amber20. Relative binding free energies were calculated using the thermodynamic cycle, by calculating the differences in the free energy for the binding of an inhibitor to the wild type and to the mutant proteins.⁴⁹

2.7 NanoBret target engagement intracellular kinase assay

NanoBret target engagement (TE) intracellular Kinase assay K10 adherent format was purchased from Promega (cat # N2521) as well as the Met Nanoluc construct (cat # NV1751). Met Nanoluc was mutagenised to generate the G1090A, Y1230H and Y1230S mutations using the Phusion site directed mutagenesis kit (Thermofisher cat #F541) following the manufacturer's instructions. HEK293T cells were then counted and resuspended at 2 × 10⁵ cells per mL, transfected with the above constructs with lipofectamine 3000 (Thermofisher cat #L3000015) and plated in polystyrene white plates (Corning cat # 3917) at 2 × 10⁴ cells per well. Past 30 h, a final concentration of  .25 uM of a cell-permeable fluorescent tracer was added to the cells as well as serial dilutions of a Type I crizotinib or Type II cabozantinib (1 µM:2) test compounds. The cells were incubated at 37°C for 2 h. After equilibrating the plates at room temperature for 15 min, 50 uL of a solution containing complete substrate and inhibitor solution was added to the wells and incubated for 3 min. Donor NanoLuc signal (460 nm) and Acceptor fluorescent Tracer emission (618 nm) were read on a GloMax multi detection Plate Reader. MilliBRET ratio was calculated as follow: [Acceptor/Donor × 1000].

3.1 Identification of MET mutations in clinical samples

We identified the emergence of four (three previously uncharacterised and one known) MET mutations (MET^G1090A, MET^D1213H, MET^R1227K and a MET^Y1230S) in samples from patients with multiple solid tumours, including patients who had been previously treated with MET inhibitors. Importantly, two of these mutations, the MET^G1090A and the MET^Y1230S substitutions, were found in more than 1 patient, indicating that they are recurrent events.

In one patient (Pt. 1 in figures) all four mutations were detected in the ctDNA collected at progression to targeted therapy. Specifically, this patient had stage IV PLEKHA6-NTRK1, MET amplified (fold change: 12.3) cholangiocarcinoma which initially responded and then became resistant to the combination therapy with the TRK inhibitor selitrectinib and the type I MET inhibitor crizotinib. As our group previously reported,⁵⁰ ctDNA sequencing at progression (Table S1) uncovered the emergence of 13 different MET mutations, including some known to confer resistance to crizotinib (i.e., MET^D1228H, MET^Y1230S)³³ and three that have only been previously described in this patient and which have yet to undergo extensive clinical and structural characterisation (i.e., MET^G1090A, MET^D1213H, MET^R1227K; Figure 1A).

Patient 2 (Pt. 2) was a 61-year-old female, never smoker, who was initially diagnosed with stage IV NSCLC (Figure 1B). A pleural biopsy at the time showed only a TP53 mutation on DNA sequencing. She was initially treated on an immunotherapy trial, but had rapid progression within 2 months. She was then treated with carboplatin, pemetrexed and bevacizumab followed by pemetrexed/bevacizumab maintenance for just over 7 months, prior to stopping for renal insufficiency and being followed with expectant management. One year and 4 months later, she underwent an ablation for progression of disease in the lung. At the time, next generation sequencing of a right lung specimen showed a CD47-MET rearrangement as well as a new MYH15-MET rearrangement (Figure 1B and Table S2). Additional alterations were seen in TP53 and DOT1L, with APC and DNMT3A alterations in the matched normal blood (Table S2). The RNA-based ARCHER fusion panel showed only the CD47-MET rearrangement (Figure S1A). High MET expression was confirmed by immunohistochemistry (IHC) that showed both diffused cytoplasmic and membranous subcellular localisation of the fusion (Figure S1B). Approximately 8 months later, the patient required further systemic therapy and was started on a MET-directed antibody–drug–conjugate (ADC) trial, but did not respond, likely due to the absence of the MET extracellular ADC binding site in the MET fused protein. She was then treated with crizotinib, to which she responded (Figure 1B). After 5 months on therapy, she developed a significant pleural effusion that required pleurX placement. Sequencing of the DNA collected from the fluid-derived cell pellet showed a MET^G1090A mutation, in addition to the CD47-MET rearrangement (Figures 1B and S1A). Other alterations are shown in Table S3. A plan was made to add chemotherapy to crizotinib given her clinically significant pleural effusion which was felt to be consistent with progression. However, in light of concerns about the risks of immune suppression due to chemotherapy during the Covid pandemic and her relative stability following pleurX placement, she was maintained on crizotinib alone. Nearly a year after the initial MET^G1090A was discovered, she developed near complete radiographic consolidation of her right lung.

In addition to patient 1, the MET^Y1230S mutation was identified via IMPACT sequencing in the tumour of a 53-year-old male patient (Pt. 3) with resected papillary RCC (Figure 1C). Additional relevant alterations identified included a frameshift in the epigenetic regulator ARID1A and MET copy gain (2.5-fold; Figure 1C). The patient was enrolled in a trial testing the activity of the combination of everolimus and bevacizumab for non-clear cell kidney cancers. After an initial mixed response (stable disease by RECIST), tumour resistance was observed about 200 days later and treatment was discontinued.

3.2 MET^G1090A and MET^Y1230S mutations confer resistance to type I but not type II MET inhibitors

Given that each patient had a potentially targetable MET alteration, we carried out pre-clinical studies to identify whether MET-directed therapy would be predicted to offer benefit and whether specific inhibitors could be used for these patients. Anecdotal examples have demonstrated that some MET mutations that confer resistance to type I MET inhibitors may still respond to type II agents (e.g., MET^D1228H, MET^Y1230C).^{29, 31} Hence, we performed in silico molecular modelling combined with MD simulations to predict the binding of all reported MET resistance mutations (including the four MET mutations we identified; Figure 2A) to a representative type I (crizotinib) and a representative type II (cabozantinib) MET inhibitor.

Notably, we identified additional opportunities for therapeutic intervention for patients with certain MET alterations. We found that some MET mutations, including MET^G1090A and MET^Y1230S, are predicted to impair binding of type I but not type II MET inhibitors (Figure 2A). Our findings suggest that tumours with specific MET mutations may benefit from MET-directed therapy that uses a type II agent.

Specifically, in the case of the MET^G1090A substitution, our simulations indicated that the 2,6-dichloro-3-fluorophenyl moiety present on crizotinib and other type I MET inhibitors directly bind through a pi stacking interaction to the side chain of the Y1230 residue in the activation loop of the MET wild type kinase. In the MET^G1090A mutant kinase, the position of the Y1230 residue is predicted to be flipped by roughly 180°. As a result, the interaction between type I MET TKIs and the Y1230 is compromised; this, in turn, is predicted to reduce crizotinib's binding affinity for the mutant kinase (Figure 2B).

Conversely, the type II agents cabozantinib and foretinib do not interact with the Y1230 and therefore their binding affinity to the mutant MET kinase is predicted to be retained (Figure 2B). Similarly, in the case of the MET^Y1230S, and as also previously shown for the MET^Y1230H mutation,⁵¹ the substitution of the Y1230 with a serine is predicted to disrupt the pi stacking interaction and thus weaken the binding of type I but not of type II MET inhibitors to the MET^Y1230S mutant kinase (Figure 2A). These findings suggested that the MET^G1090A and the MET^Y1230S mutant kinases may be resistant to type I inhibition, but not to type II MET inhibitors.

To further test our hypothesis, we performed in vitro kinase assays using a titration of crizotinib and capmatinib as representative type I agents and of cabozantinib and foretinib as representative type II drugs against MET WT and mutant MET^G1090A, MET^Y1230S and MET^Y1230H (used as control) recombinant kinases (Figure 2C). IC₅₀ values of 7.58 nM and 4.59 nM were obtained for the type I MET inhibitors crizotinib and capmatinib against MET WT. Significantly higher IC₅₀ values were obtained when the same agents were tested against the MET^G1090A kinase (IC₅₀ = 33.99 nM and 23.05 nM for crizotinib and capmatinib, respectively) or the MET^Y1230S kinase (IC₅₀ = 112.6 nM and 3554 nM for crizotinib and capmatinib, respectively). Interestingly, while both type II MET inhibitors cabozantinib and foretinib were slightly less potent than type I agents against MET WT (IC₅₀ = 133 nM and 57.05 nM for cabozantinib and foretinib, respectively), their activity was not significantly inhibited when they were tested against the MET^G1090A kinase (IC₅₀ = 137 nM and 67.44 nM for cabozantinib and foretinib, respectively) or the MET^Y1230S kinase (IC₅₀ = 268.1 nM and 86.13 nM for cabozantinib and foretinib, respectively; Figures 2C and 2SA).

We next transfected HEK-293T cells with plasmids encoding the CD47-MET fusion, the CD47-MET fusion harbouring the G1090A mutation, or the CD47-MET fusion harbouring the Y1230S mutation and evaluated changes in MET phosphorylation following treatment with type I and type II agents. Our results showed that, while crizotinib (type I) and cabozantinib (type II) were equally effective in inhibiting MET phosphorylation in cells transfected with the fusion alone, only type II inhibition with cabozantinib impeded MET activity at low nanomolar concentrations in cells transfected with the mutant MET fusions (Figures 2D and S2B,C). Similar confirmatory results were obtained when the type I capmatinib and the type II foretinib were tested against the same constructs (Figure S2D,E). These in-cell data further validated that type II MET inhibitors retain activity against the MET^G1090A and the MET^Y1230S mutant kinases.

While our in-cell data using the HEK-293T cells suggest that the differential response to type I and type II MET inhibitors depends on the specific MET substitution, we proceeded to assess the possibility that tissue context in which the mutants are identified or the presence of co-occurring drivers could influence drug sensitivity. Thus, we transfected the ALK fusion-positive NSCLC cell line H3122 with the MET^G1090A and the MET^Y1230S mutations and confirmed these ALK fusion-positive, MET mutant cell lines to be resistant to type I but not to type II MET, mimicking results we obtained in the HEK-293T cell line (Figure S3).

We then performed NanoBret experiments to evaluate the binding affinity of type I and type II inhibitors for WT and mutant MET kinases. This technique allows derivation of binding affinities of drugs through in-cell experiments following transfection with WT and mutant MET constructs and treatment with increasing concentrations of drugs. Data obtained in HEK-293T cells show that cabozantinib binds MET G1090A, MET Y1230S and MET Y1230H kinases with higher affinity than crizotinib (Figure 2E). Similar results were obtained when the same NanoBret assay was conducted on the same cell model using a titration of the type I MET TKI capmatinib and the type II MET TKI cabozantinib (Figure S4).

3.3 MET^D1213H and MET^R1227K mutations do not impact the activity of type I or type II MET inhibitors

A different scenario was observed when we modelled the previously uncharacterised MET^R1227K and MET^D1213H mutations. Our predictions showed that these mutations would just slightly impact type I and type II TKIs response (Figure 3A). Specifically, for the MET^R1227K we found that Arg1227 is involved in a network of interactions with residues Asp1222, Phe1223, Gly1224, His1202 that partially stabilises the bound conformation of both crizotinib and cabozantinib. Replacement of Arg1227 with Lys may partially, but just slightly perturb this network of interactions. Similarly, the MET^D1213H mutation does not seem to destabilise the bound conformations of either crizotinib or cabozantinib; however, the interaction of crizotinib with Met1160 is not retained in the case of MET^D1213H, suggesting this mutation to slightly impair crizotinib binding (Figure 3A). To evaluate the effect that these mutations have on the response to type I and type II MET inhibitors in cells, we transfected HEK-293T cells with plasmids encoding the CD47-MET fusion, the CD47-MET fusion harbouring the R1227K mutation, or the CD47-MET fusion harbouring the D1213H mutation and evaluated changes in MET phosphorylation following treatment with type I and type II agents. Consistent with our predictions, the presence of each of these mutations did not affect the ability of either type I (i.e., crizotinib, capmatinib) or type II (i.e., cabozantinib, foretinib) MET inhibitors to inhibit MET phosphorylation (Figure 3B–E).

Together, these data indicate that MET^R1227K and MET^D1213H mutations do not confer significant resistance to any of the type I or the type II MET TKIs tested and suggest that the presence of these MET mutations in the ctDNA of patient 1 (see Figure 1A) was not the main aetiology of resistance to crizotinib.

3.4 Cabozantinib is active against selected type I agents-resistant MET mutants in patients

Given our pre-clinical findings suggesting that some patients may continue to benefit from MET-directed therapy even after developing alterations in MET known to confer resistance to certain TKIs, we aimed to identify patients who might benefit from treatment with a type II MET inhibitor.

We surveyed our clinical database at MSKCC for patients with MET mutations predicted by our MD simulations and/or cell-line data to respond to type II MET inhibitors. In addition to our three index patients (Pts. 1, 2 and 3), we found nine additional patients: five with RCC and four with NSCLC (Figure 4A). All patients with NSCLC with these alterations in their tumours had progressed on at least one line of therapy, suggesting that the MET mutation was the likely cause of the acquired resistance to prior therapy. By contrast, most of the RCC patients were treatment-naïve (Figure 4A).

Two of the patients (Pts. 2 and 3) were considered candidates for additional therapy. Pt. 2, who had the CD47-MET fusion-positive, MET G1090A mutant NSCLC, and Pt. 3 with the MET Y1230S RCC, both clinically responded to the type II inhibitor cabozantinib (Figure 4B,C).

Unfortunately, despite the robust anti-neoplastic response experienced by Pt. 2, she was unable to tolerate the medication at prescribed doses due to hand–foot syndrome, which eventually led to every-other-day dosing and then treatment discontinuation. Pt. 3 also developed skin toxicity and underwent dose reduction. While tolerability issues limited the time on treatment with cabozantinib, together these data provide clinical proof of principle suggesting that treating patients with type II MET TKIs can be beneficial in molecularly selected cases and emphasise the importance of developing tolerable type II inhibitors.