Crizotinib in Patients with Tumors with MET Amplification or Exon 14 Deletion: Results from the NCI-MATCH ECOG-ACRIN Trial (EAY131) Subprotocols C1 and C2

Abstract

Background:

The NCI-MATCH trial enrolled patients to subprotocols C1 and C2 to evaluate the METtyrosine kinase inhibitor crizotinib for efficacy in patients with MET amplification (METamp; C1) or MET exon 14 skipping mutation (METex14; C2).

Methods:

Tumors harboring METamp or METex14 were confirmed by Oncomine assay, RNA-sequencing, and Anchored Multiplex PCR. Patients received 250 mg crizotinib orally daily until disease progression or unacceptable toxicity. Primary endpoint was objective response rate (ORR); secondary endpoints included progression-free survival (PFS), 6-month PFS, and overall survival (OS).

Results:

C1 efficacy analysis comprised 28 of 44 enrolled patients (17 gastrointestinal, 7 lung, 4 other tumor types). Four patients had a partial response (PR), 10 had stable disease (SD), 13 had progressive disease (PD); one unevaluable. ORR was 14% (4/28, 90%; CI 5.0–29.8%). Median (m) PFS was 3.4 months (90% CI 1.8–3.7), mOS 7.1 months (90% CI 5.0–11.5). C2 included 14 of 20 patients (5 gastrointestinal, 6 lung, 3 other). Two patients had PR, 4 SD (2 >6 months), and 4 PD were observed. ORR was 14% (2/14, 90% CI 2.6–38.5%). mPFS and mOS were 2.0 months (90% CI 1.4–4.1) and 10.2 months (90% CI 2.3–19.6).METex14 cases with read count ≥50,000 had mPFS 8.8 months (90% CI 2.1–NA) versus 1.7 months (90% CI 1.1–3.7).

Conclusion:

Crizotinib demonstrated clinical activity across tumors with METamp and METex14.Subprotocol C1, but not C2, met its primary endpoint. In METex14 disease, a read count cutoff >50,000 may help distinguish true pathogenic variants from low-level splice transcripts and enable more accurate classification.

Background

MET (MET proto-oncogene or hepatocyte growth factor [HGF] receptor) is a receptor tyrosine kinase, expressed predominantly in epithelial cells, and together with its ligand, HGF, has been an important target of drug development. MET signaling can become dysregulated through several mechanisms, including gene amplification, receptor protein overexpression, mutations (germline or somatic), and chromosomal fusion.(1) Aberrant MET signaling can promote growth, survival, invasion, migration, angiogenesis, and metastasis.(2,3)

Driver mutations in MET are rare, occurring at varying frequencies across tumors (2.6% of all cancers),(4) with the greatest prevalence in lung adenocarcinoma, colon adenocarcinoma, melanoma, and endometrial adenocarcinoma.(4) Across cancers, MET amplification (METamp; 0.69%), mutation (1.90%), MET exon 14 skipping variant (METex14) caused by mutations affecting splicing (0.23%) are the most frequent alterations,(4) whereas germline activating mutations in the tyrosine kinase domain of MET occur in 100% of hereditary papillary renal cell carcinomas.(5) Somatic amplification of MET and/or MET mutations are also observed in sporadic papillary renal cell carcinoma, lung cancer,(6,7) head and neck squamous cell carcinoma,(8) hepatocellular carcinoma,(9) gastric cancer,(10) esophageal cancer,(11) colorectal cancer,(12) gliomas,(13) and clear cell ovarian cancer.(14)

More recently, METex14 has been recognized as a driver event in non-small cell lung cancer (NSCLC) and has been a bona fide target with FDA-approved targeted therapies since 2020.(15) METamp is a frequent driver of targeted therapy resistance in some subsets of NSCLC (e.g., EGFR mutant), and high-level METamp is viewed as a de novo primary oncogene driver in NSCLC.(16) METex14-driven NSCLC has been reported in a significant percentage of non-adenocarcinomas (up to 23%−26%),(17,18) including sarcomatoid (2%−9%)(18–20) and squamous cell carcinoma (6%−9%),(18–20) while high-level amplification of MET occurs at a frequency of 2%−4% overall in NSCLC.(4)

Crizotinib is a multitargeted small-molecule ATP-competitive multikinase inhibitor, with potent activity against ALK, ROS1, and MET.(19,21) Crizotinib is a type Ia MET tyrosine kinase inhibitor, FDA-approved for treatment of ALK- and ROS1-rearranged non–small cell lung cancers (NSCLC) and it has demonstrated activity in tumors harboring MET alterations, particularly MET exon 14 skipping and high-level MET amplification.(1) Drilon et al(19) demonstrated activity in patients with chemotherapy-refractory METex14-aberrant NSCLC in the PROFILE 1001 study (NCT00585195), the first prospective study of a MET inhibitor in METex14-altered NSCLC, with an objective response rate (ORR) of 32% among 65 treatment-naive patients, a median duration of response of 9.1 months, and median progression-free survival (PFS) of 7.3 months.(22) More recently, more potent and selective MET-targeting agents, such as capmatinib, tepotinib, and savolitinib, have yielded superior clinical results in MET-driven lung cancer.(17,20,23) For example, in the VISION trial which investigated tepotinib in patients with METex14 NSCLC, meaningful activity across subgroups was achieved, with ORR 46%, and disease control rate (DCR) 65.7%.²³ In the METex14 cohort of the GEOMETRY-mono-1 trial, capmatinib demonstrated similar activity, ORR 40.6% in previously treated patients, and ORR 67.9% in treatment naïve patients.(17) These data have informed guideline recommendations and provide a strong rationale for evaluating crizotinib in biomarker-defined cohorts in NSCLC and beyond.(24)

The National Cancer Institute (NCI)–Molecular Analysis for Therapy Choice (MATCH) platform trial, a collaboration between the ECOG-ACRIN Cancer Research Group and the NCI, was initiated to discover efficacy signals by matching patients with refractory malignancies to treatments targeted to molecular drivers across tumors.(25) Subprotocols C1 and C2 investigated crizotinib in MET activation: C1 used crizotinib in patients with confirmed MET amplification (METamp), and C2 used crizotinib in patients with MET exon 14 mutation/deletion (METex14). While METamp and METex14 are rare driver events across tumors, we hypothesized that the identification of true driver events and true pathogenic variants could be crucial in identifying patients who may benefit from targeted therapy using a MET kinase inhibitor. Here, we report the clinical efficacy and safety of crizotinib in both substudies.

Methods

Study Design and Population

An overview of the National Cancer Institute Molecular Analysis for Therapy Choice (NCI-MATCH) trial (ClinicalTrials.gov identifier: NCT02465060; ECOG-ACRIN identifier: EAY131), developed by the NCI and the ECOG-ACRIN Cancer Research Group, has been published previously.(26) Patients enrolled to the NCI-MATCH trial were assigned to biomarker-matched treatment subprotocols based on their tumor’s genomic profile, using a prospectively defined, NCI-designed computational central algorithm (MATCHBOX) that prioritized alterations according to predefined tiers of clinical actionability. Patients whose tumors had an actionable alteration were assigned by MATCHBOX to one of the available treatment subprotocols (Figure 1). If multiple actionable variants were present, the patient was assigned by the variant with the highest level of evidence, followed by the variant with the highest allele frequency. If still equivalent, assignment was to the subprotocol with the fewest patients or random if accrual was equal. Assignment to arms C1 and C2 followed these criteria. This precision medicine study comprised 38 target-specific subprotocols and accrued patients at over 1000 centers in the United States (National Clinical Trials Network and NCI Community Oncology Research Program sites). The study was performed in accordance with provisions of the Declaration of Helsinki and Good Clinical Practice guidelines and approved by the NCI Central Institutional Review Board. Written informed consent was obtained for all participants.

Figure 1. Responses for patients in the subprotocol C1 cohort.

A) Best lesion size change from baseline. Waterfall plot (color coded by histologic category) shows best confirmed response of target lesion(s) according to RECIST 1.1 (n=23). Five patients were excluded from the plot, including 1 patient with unevaluable response and 4 patients who had best response of PD determined due to new lesions without successful assessment of the target lesions. Asterisks indicate new lesions in patients with the best response of PD. B) Duration of treatment. Swimmer plot (colored by histologic category) for patients with best response of SD or PR (n=14). The bar lengths indicate duration of treatment. One patient is still receiving treatment.

Adult patients with any solid tumor, lymphoma, or myeloma who had disease progression during standard treatment or for whom no standard treatment was available were eligible. Adequate hematopoietic, liver, and kidney function; an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1, and availability of fresh biopsy tissue were required. Tumor tissue was obtained from either the primary tumor or a metastatic site, depending on sample availability and quality. Tissue adequacy was assessed locally, and molecular profiling was conducted as outlined below. There were two mechanisms of recruitment. Between August 2015 and May 2017, prescreening for enrollment on the NCI-MATCH trial was accomplished using biopsies collected at time of enrollment and sent for testing at one of four central laboratories. Central testing was performed using an Oncomine AmpliSeq assay at Clinical Laboratory Improvement Amendments (CLIA)–certified laboratories running the harmonized central assay as described by Lih et al.(27) Patients found to have eligible molecular alterations were assigned to subprotocols using a prospectively defined NCI-designed informatics algorithm (MATCHBOX).(28) After May 2017, mutational analysis was completed by designated academic and industry CLIA-certified laboratories using next generation sequencing platforms (Designated Lab Network); for patients identified through the Designated Lab Network, samples were requested to attempt central confirmation using the Oncomine AmpliSeq platform. Confirmatory central testing was attempted for all patients whose tumors were deemed eligible based on results from designated outside laboratories.

For subprotocol C1, patients were required to have MET amplification defined as ≥7 copies/cell as identified by the Oncomine® Assay, or the Oncomine® Assay equivalent of 7 or greater as identified by a Designated Laboratory. For subprotocol C2, METex14 was identified using the Oncomine® Assay RNA-sequencing approach, and confirmation was performed retrospectively at the DNA level using Anchored Multiplex PCR, utilizing a custom targeted DNA sequencing panel and analysis pipeline that permitted analysis of the MET exon 14 coding region and surrounding intronic sequences.(29) Contextualisation of the study cohort relative to external reference populations is provided in Supplementary Table S1.

Study Therapy and Assessments

Patients assigned to the C1 and C2 subprotocols received 250 mg of crizotinib orally twice daily in 28-day cycles until disease progression or unacceptable toxicity. Tumor assessment using RECIST version 1.1 criteria for solid tumors occurred every two cycles.(30)

Toxicity was evaluated using Common Terminology Criteria for Adverse Events v4.0. Patients were monitored closely for toxicity, and the dose of crizotinib was adjusted per protocol. Intrapatient dose reduction by one and, if needed, two dose levels was permitted depending on the type and severity of toxicity encountered. Patients requiring more than two dose reductions due to treatment-related toxicity and patients requiring treatment to be held for >4 weeks were removed from the trial.

Statistical Analysis

The overall NCI-MATCH statistical design and rationale have been previously described.(25,28) The primary objective for each subprotocol was to evaluate the objective response rate (ORR). The original accrual goal per subprotocol was 35 patients, aiming to enroll 31 eligible patients, providing 92% power to distinguish an ORR of 25% from a null of 5% with a one-sided type 1 error of 1.8%. Based on the goal of 31 eligible patients included in the primary analysis, an ORR of at least 5/31 patients (16%) was to be considered a signal of promising activity and the endpoint met. Expansion up to a total of 70 enrolled patients was permitted based on protocol criteria, allowing up to 6 months of additional accrual and a maximum of 35 additional patients, in circumstances where clear signals had not emerged, and it was desired to gain more experience with certain tumor types and specific variants.

After 35 patients were accrued, the number of patients with each tumor type that could be enrolled was capped at 10. Patients who were enrolled based on the NCI-MATCH central screening assay or based on an outside assay and whose tumor molecular abnormalities were confirmed by the NCI-MATCH assay were included in the primary efficacy analysis. All treated patients were evaluable for toxicity assessment. When fewer than 31 patients were included in the primary efficacy analysis for a subprotocol (the situation for C1 and C2), the primary endpoint was assessed using a 5% one-sided exact binomial test of the null hypothesis that the ORR is ≤ 5%. Secondary objectives included evaluation of PFS, PFS rate at 6 months, overall survival (OS), toxicity, and predictive biomarkers (co-occurring mutations or other biomarkers predictive of response).

METex14 Quantification

At the time of the development of NCI-MATCH assay, there was a lack of data defining the optimal cutoff to differentiate low and high level of METex14 skipping RNA counts that would define biological meaningful subsets. During the early phase of NCI-MATCH assay development for the clinical study, it was determined that a 1,000-count RNA read depth for METex14 skipping would be used as minimal threshold to enroll patients in subprotocol C2. The 1,000-count RNA read depth was established, following discussion with the assay manufacturer and input from pharmaceutical industry colleagues.

During the NCI-MATCH screening, it was anecdotally noticed that some METex14-positive cases also possessed other MAPK pathway alterations (e.g., KRAS), that would not be expected in true METex14 cases, and that these cases had borderline positive read counts. Analysis of all cases revealed a dichotomy of the existing data: 1) high METex14 skipping RNA reads, which were typically greater than 50,000 and 2) low METex14 skipping RNA reads, typically less than 5,000. In our data set, cases between 5,000–100,000 RNA reads for METex14 skipping were absent. Based on these data, the plotting of the distribution of all read counts, and observations regarding mutual exclusivity, the METex14 minimum read count was adjusted to 50,000 for subsequent screened cases through an amendment to the protocol dated February 28, 2020.

As part of an ancillary correlative science study, METex14 mutations at the DNA level were explored retrospectively using available cases with sufficient tissue for analysis of DNA mutations surrounding MET exon 14 and control samples with a variable amount of MET exon 14 skipped counts. MET counts were determined using Anchored Multiplex PCR using the MGH CLIA laboratory genotyping platform.(29) 100ng of input DNA was used for library construction, and a minimum of 200X target coverage was obtained. Variant calling was made by manual review of METex14 and surrounding intron 13 and 15 variants by a molecular pathologist (AJI).

We hypothesized that low-level RNA read counts (<5,500 reads) for METex14 would not predict response to therapy, and thus should likely be considered separately from high read count cases. We hypothesized that the correlative study, if successful, could more definitively define a threshold for METex14 skipping RNA reads that would correlate with that of DNA variants predictive of splicing changes. In our NCI-MATCH clinical study, 16 patients with low level METex14 skipping RNA reads were identified, i.e. were noted to have METex14 <5,500 read counts; twelve of these patients were enrolled in EAY131-C2 per protocol, and along with patients having high level METex14 RNA reads, received treatment with crizotinib. As an exploratory analysis, we assessed whether patient response was associated with METex14 read counts and MET DNA mutations, to better guide future studies where METex14 skipping is used as a treatment selection biomarker.

Data Availability Statement:

The data underlying this article are available from the authors upon request and will be made available for request from the NCTN/NCORP Data Archive (https://nctn-data-archive.nci.nih.gov/) upon completing a Data Request Form for data from NCT06357975 (Subprotocol C1) and NCT06360575 (Subprotocol C2).

Results

Accrual for subprotocol C1 occurred from July 29, 2016, to November 12, 2021, and for subprotocol C2, from July 20, 2016, to March 16, 2020 (Supplementary Figure 1 Supplementary Figure 6).

Subprotocol C1 – MET amplification Cohort

Forty-four patients were enrolled in subprotocol C1. Thirteen patients were enrolled from the screening cohort (by central NCI-MATCH assay). Thirty-one patients were enrolled from the Designated Lab Network (DL), of whom 16 had their tumor molecular eligibility confirmed by the NCI-MATCH Oncomine assay (Supplementary Figure 1). Per protocol, all patients confirmed centrally using the NCI-MATCH Oncomine assay were considered analyzable (n=28); this included the thirteen patients enrolled from the screening cohort, and fifteen patients from the DL network (one patient never started treatment). Thus, twenty-eight were eligible and received treatment with crizotinib and were considered analyzable and included in the primary efficacy analysis as prespecified in the protocol. All treated patients were included for toxicity monitoring. The 28 patients included in the primary analysis represented a variety of tumor histologies, with gastrointestinal (GI; n=17) and lung (n=7) being the most frequent; 32% were female (n=9), and the median age was 67 years (range 28–82) (Table 1).

Table 1. Characteristic of Patients in the Outcome Analysis for Subprotocol C1 Cohort.

Characteristic	No. of Pts. (%) n=28
Gender:  Female	9 (32.1%)
Age (years): Range Median	28–82 67
Race:  White	23 (82.1%)
Black	3 (10.7%)
Not reported	2 (7.1%)
Ethnicity:  Non-Hispanic	26 (92.9%)
Not reported	2 (7.1%)
ECOG performance status:  0	9 (32.1%)
1	19 (67.9%)
No. of prior therapies:  0*	1 (3.6%)
1	6 (21.4%)
2	11 (39.3%)
3	4 (14.3%)
>3	6 (21.4%)
Tumor type:
Gastrointestinal carcinoma Adenocarcinoma of esophagus (n=7)  Lower third/distal (n=2)  Esophagogastric junction (n=3)  Not otherwise specified (n=2) Adenocarcinoma of rectum (n=3) Adenocarcinoma of sigmoid colon (n=1) Adenocarcinoma of transverse colon (n=1) Intrahepatic cholangiocarcinoma (n=2) Metastatic colorectal adenocarcinoma, not otherwise specified (n=1) Metastatic gastric adenocarcinoma with signet ring cells (n=1) Mixed adenoneuroendocrine carcinoma/mucinous adenocarcinoma/signet ring-cell carcinoma of cardia (n=1)	17 (60.7%)
Lung carcinoma Adenocarcinoma (n=5) Well-differentiated adenocarcinoma with lepidic pattern (n=1) Atypical carcinoid (intermediate-grade neuroendocrine) tumor of lung (n=1)	7 (25%)
Genitourinary tumors High-grade transitional cell carcinoma of urinary bladder neck (n=1)	1 (3.6%)
Melanoma Malignant melanoma of hard palate mucosa (n=1)	1 (3.6%)
Thyroid carcinoma Anaplastic carcinoma of thyroid (n=1)	1 (3.6%)
Adenocarcinoma of unknown primary	1 (3.6%)

^*No prior therapy due to nonexistence of a standard treatment shown to prolong overall survival.

Of these 28 patients, four patients (14%) had partial response (PR) as their best response; and 10 patients (35.7%) had stable disease (SD). Thirteen patients (46%) had progression of disease (PD) as best response. One patient was unevaluable. Progression of disease was the most common reason for discontinuing study participation (n=19, 68%). The ORR was 14% (4/28, 90% CI: 5.0%−29.8%) (Figure 1); the null hypothesis of response rate not exceeding 5% was rejected at the 1-sided 0.05 significance level (P=0.049), thus meeting the primary endpoint. Median PFS was 3.4 months (90% CI: 1.8–3.7); median overall survival (OS) was 7.1 months (90% CI: 5.0–11.5) (Supplementary Figure 2). The estimated 6-month PFS rate was 14.3% (90% CI: 5.6%−26.8%). The four patients with confirmed PR had lung adenocarcinoma (n=2), GI adenocarcinoma (n=1), and melanoma (n=1). One patient with NSCLC had SD >6 months (Supplementary Table 2). Most patients in subprotocol C1 had ≥1 co-occurring gene alteration (n=27; 96%), with TP53 mutations being the most frequent (n=23, 82%); 21 patients had ≥3 co-occurring mutations (Supplementary Figure 3). No new toxicity or safety signals were identified (Supplementary Table 3).

Subprotocol C2 – METex14 Cohort

Twenty patients were enrolled on subprotocol C2; sixteen patients were enrolled from the screening cohort (by central NCI-MATCH assay) and four patients were enrolled from the Designated Lab Network (Supplementary Figure 4). Fourteen patients were deemed eligible, centrally molecularly confirmed, received treatment with crizotinib, and were included in the primary efficacy analysis (Supplementary Figure 4). Four patients discontinued treatment prior to imaging; all 14 patients were included in the denominator for the intent-to-treat (ITT) analysis, with early discontinuation patients counted as non-responders. Their median age was 68 years (range 57–78), six patients were women (43%), nine patients (64%) had ≥3 lines of prior therapy, and the most frequent tumor histologic types were GI (n=5) and lung (n=6) adenocarcinomas (Table 2). Ten patients were evaluable; two patients (both with NSCLC) achieved PR as best response (14%), four patients had SD (29%), and four had PD as best response (29%) (Figure 2). Two patients (one with NSCLC, one with bladder carcinoma) achieved SD >6 months (Figure 2, Supplementary Table 4). The ORR was 14% (2/14, 90% CI: 2.6%−38.5%). Since the null hypothesis of response rate not exceeding 5% could not be rejected at the one-sided 0.05 significance level (P=0.15), subprotocol C2 did not meet its primary endpoint. Median PFS was 2.0 months (90% CI: 1.4–4.1), and median OS was 10.2 months (90% CI: 2.3–19.6) (Supplementary Figure 5). The estimated 6-month PFS rate was 29% (90% CI: 11.3%−48.7%). Most patients had ≥1 co-occurring gene alterations (n=13, 93%; range 1–4) with TP53 mutations being the most frequent (n=7, 50%) (Supplementary Figure 6). Crizotinib was well tolerated, and no new toxicity or safety signals were identified (Supplementary Table 5).

Table 2. Characteristics of Patients in the Outcome Analysis for Subprotocol C2 Cohort.

Characteristic	No. of Pts. (%) n=14
Gender:  Female	6 (42.9%)
Age (years):  Range  Median	57–78 68
Race:  White	11 (78.6%)
Black	1 (7.1%)
Asian	2 (14.3%)
Ethnicity:  Hispanic	0
ECOG performance status:  0	4 (28.6%)
No. of prior therapies:  1	3 (21.4%)
2	2 (14.3%)
3	4 (28.6%)
>3	5 (35.7%)
Tumor type:
Gastrointestinal carcinoma Adenocarcinoma of colorectum (n=2)  Mucinous in cecum  Rectum Adenocarcinoma of pancreas (n=2), 1 each tail and head Mixed adenoneuroendocrine carcinoma of esophagogastric junction (n=1)	5 (35.7%)
Lung carcinoma Adenocarcinoma (n=3) Squamous cell carcinoma (n=1) Adenosquamous carcinoma (n=1) Sarcomatoid (pleomorphic) carcinoma (adenocarcinoma component in primary) (n=1)	6 (42.9%)
Genitourinary tumor Transitional cell carcinoma of bladder (n=1)	1 (7.1%)
Gynecologic tumor Clear cell adenocarcinoma of endometrium (n=1)	1 (7.1%)
Melanoma Uveal (n=1)	1 (7.1%)

Figure 2. Responses for patients in the subprotocol C2 cohort.

A) Best lesion size change from baseline. Waterfall plot (color coded by histologic category) shows best confirmed response of target lesion(s) according to RECIST 1.1 (n=10). Four patients were excluded from the plot due to unevaluable response. Asterisks indicate new lesions in patients with the best response of PD. B) Duration of treatment. Swimmer plot (colored by histologic category) for patients with best response of SD or PR (n=6). The bar lengths indicate duration of treatment.

Quantification by MET Count

We hypothesized that the Oncomine assay RNA-level copy threshold/cutoff may have been inappropriately set for calling pathogenic METex14 cases, and that response rates may have been lower than expected due to the assay permitting the inclusion of patients with tumors with low- level exon 14 skipping read counts (<5,500). Therefore, we theorized that DNA mutations could help differentiate low-level splice variant transcripts, which are known to occur in normal tissues,(31) from true pathogenic variants. METex14 mutations at the DNA level were explored retrospectively, using cases with sufficient tissue for analysis of DNA mutations surrounding MET exon 14, and control tumors with a variable amount of MET exon 14 counts using Anchored Multiplex PCR; MET counts were determined using Anchored Multiplex PCR using the MGH CLIA laboratory genotyping platform(29) (Figure 3; Table 3).

Figure 3.

METex14 quantification in subprotocol C2 and controls. Correlation of METex14 supporting RNA reads from Oncomine assay with retrospective MET DNA sequencing results using DNA from cases with sufficient remaining tissue (16 patients from C2 are marked with asterisks; the remaining are comparator samples from the NCI-MATCH screening cohort that were negative for METex14 by Oncomine). Each column represents a single patient; red columns signify the presence of a conclusive pathogenic MET DNA mutation confirming the Oncomine result; blue indicates a lack of confirmatory DNA mutation. Five of five cases with Oncomine read counts >50,000 showed confirmatory DNA mutations. Of 16 cases enrolled on arm C2, three had confirmed Oncomine METex14 counts ≥ 50,000 (a fifth patient had assignment based on a novel indel detected by an outside laboratory, so no partner read count was available). Y-axis cut-off was 150,000 reads.

Table 3. METex14 Quantitation for All Subprotocol C2 Patient with Available Specimens*.

Each row represents a patient enrolled on study.

Cancer Diagnosis	MET Pathogenic Mutation	METex14 Count	Co-occurring MAPK Alteration
Lung adenocarcinoma		326**	KRAS CNV
Adenocarcinoma of the rectum		1012	KRAS
Mucinous adenocarcinoma of cecum		1108
Transitional cell carcinoma of bladder		1509
Adenocarcinoma of head of pancreas		1532	KRAS
Adenocarcinoma of cecum		2280**	KRAS
Lung adenocarcinoma	MET c.3082+3A>C 0.205	2895
Uveal melanoma		3018
Papillary and clear cell carcinoma of kidney		3158**
Adenocarcinoma of sigmoid colon		3212**	BRAF
Papillary serous adenocarcinoma of endometrium		4035
Adenocarcinoma tail of pancreas		5340	KRAS
Lung adenocarcinoma	MET c.2942–44_2942–9 delTGATAGCCGTCTTTAAC AAGCTCTTTCTTTCTCTCT 0.086	24,950**
Adenosquamous carcinoma of lung	MET c.3082+2T>C 0.238	144,833
Lung adenocarcinoma	MET c.2942–29_2945 delACAAGCTCTTTCT TTCTCTCTGTTTTAAGATCT 0.151	282,287
Sarcomatoid (pleomorphic) carcinoma of lung	METp.Asp1028His c.3082G>C 0.875	325,755
Lung squamous cell carcinoma	MET c.3082+2T>G 0.35	N/A***

^*Results for 16 of the 17 cases tested are shown. One additional patient with METex14 RNA read count >50,000 is not displayed due to incomplete tabular data but is included in summary statistics.

^**Indicates patient not included in primary efficacy analysis, ie. patient not included in Table 2 above.

^***One patient had assignment based on a novel indel, i.e., no partner read count, as the case was referred from the designated lab network and not analyzed using Oncomine.

Of the available cases for testing, four patients had a confirmed MET count of ≥50,000 and were identified as “true positive”, i.e., harboring splice site DNA mutations; all were patients with NSCLC. Of these, two patients had PR, and one had prolonged SD as best response. None of the patients with METex14 counts ≥50,000 possessed other MAPK pathway activating mutations (e.g., KRAS), suggesting that MET was the major driver mutation in those tumors (Figure 3; Table 3).

Next, we analyzed PFS and OS, stratifying by MET count among the subprotocol C2 patients for whom MET counts were available (n=13); all these patients were included in the primary analysis cohort of 14 patients. For PFS, for patients with a MET count <50,000, the median PFS was 1.7 months (90% CI: 1.1–3.7). For those with a MET count ≥50,000, the median PFS was 8.8 months (90% CI: 2.1–NA) (Figure 4). All patients with lung cancer with confirmed PR or with SD >6 months had a documented MET count of >50,000. For OS, for patients with a MET count of <50,000, the median OS was 4.8 months (90% CI: 1.4–19.6). For those with a MET count of ≥50,000, the median OS was 20.2 months (90% CI: 2.3-NA).

Figure 4. PFS (A) and OS (B) in the subprotocol C2 cohort by MET count.

For one case, no partner read was present; this case is excluded from the curve.

Discussion

In keeping with previously published studies, we confirmed the clinical activity of crizotinib across tumor types in patients with MET-driven tumors, and reaffirmed the safety and tolerability of crizotinib, with no new safety signals identified. Subprotocol C1 (METamp) met its primary endpoint; however, the primary endpoint was not reached in subprotocol C2 (METex14), with an overall low ORR of 14% in both subprotocols. Our correlative data suggest a bimodal distribution of results with regards to METex14 skipping RNA read counts in subprotocol C2, with METex14 skipping RNA reads documented as either very high (>100,000) or very low reads (< 6,000). To interrogate this, we performed a retrospective analysis using the cases available to confirm MET exon 14 skipping at the DNA level in subprotocol C2.

Our data (Table 3) suggest a bimodal distribution of METex14 counts in the data set, i.e. those with very low reads, thought to represent true biology in normal cells, and those with very high reads, representing likely true pathogenic events, and we confirmed cases with MET read count >50,000 in four patients (all with lung adenocarcinoma). However, this may reflect the small number of non-NSCLC tumors enrolled in Subprotocol C2, and we caution against overinterpretation of tumor-type specificity based on this limited sample. Future pan-cancer studies will be needed to determine whether very high METex14 transcript abundance is uniquely enriched in NSCLC or occurs more broadly across tumor types. Assessment of survival based on MET count in this cohort also suggested a difference in both PFS and OS. Of course, these observations were noted in a limited sample size, statistical significance could not be established, and with an absence of corresponding protein or phospho-protein data, these findings must be as preliminary. These findings will therefore need to be validated prospectively in larger, independent datasets, with integrated transcriptomic and proteomic profiling. Still, these data suggest that identifying a cutoff MET count could be useful in identifying true pathogenic drivers of disease and suggest that, in carefully selected patients, identifying patients with true pathogenic variants from low-level splice variant transcripts may be clinically meaningful. Our data builds on previous reporting that suggests that MET RNA expression may correlate with response to MET TKIs in NSCLC.(32) To the best of our knowledge, we are the first to suggest that utliIizing a rational patient selection strategy utilizing MET read count cutoff of >50,000 for MET exons 13–15 may have clinical implications and may contribute to ongoing efforts to refine METex14 as a predictive biomarker. Prospective studies are needed to define clinically meaningful thresholds and determine whether RNA expression, in combination with other biomarkers, can refine patient selection beyond MET exon 14 skipping alone.

A challenge in the development of MET-driven studies to date has been selecting the cutoff for MET activity, i.e., the level at and above which a MET-targeted compound might be clinically active.(1) The relationship between MET copy number and protein expression is overall weak. Previous studies have highlighted that MET protein detection using IHC is not an effective screening method for MET activity, as MET IHC can essentially fail to identify patients with METamp or copy number gain.(1) PhosphoMET assays may add value to determining clinically actionable pathogenic MET status,(33) although data suggest that the assay requires site-specific phosphorylation interrogation and rapid snap freezing of tissue in order to permit accurate results—not easily achieved in a large multi-institution trial. Our data suggest that utilizing a MET read count cutoff is key to identifying patients with “true” pathogenic variants and suggest that having METex14 DNA coverage may be essential for accurate calling of METex14 disease. It is important, however, to highlight that in our dataset, counts were not normalized to sample input or any reference gene, a caveat therefore, in trying to set a threshold that could broadly apply to other assays, which may use a different input level or normalize their data. Based on these data, in consideration of METex14 cases, it may be prudent to rely on splice site mutations in DNA, until this is resolved for the many available RNA assays.

METex14 alterations are also a challenge diagnostically, as a high diversity exists in the composition of their sequence, number of novel variants, and indel mutations that can be difficult to detect.(34,35) Typically, in the clinic these variants are detected by DNA sequencing of the tumor by identifying a variant that alters or ablates a splicing site, or by using RNA sequencing or RT-PCR.(36–38) Across clinical trials, there has been inconsistency in diagnostic testing of METex14. For example, the registrational trial for capmatinib identified METex14 using RT-PCR with Foundation Medicine’s targeted capture next-generation sequencing (NGS) panel on tumor tissue,(17) while the tepotinib registrational trial utilized the Guardant360 NGS panel (73-gene) to analyze plasma-circulating tumor DNA and/or the Oncomine Focus Assay (52-gene) to analyze RNA collected from tumor tissue biopsy.(20) Our data highlight the limitations of calling METex14 a true variant using RNA only, and suggest that confirming RNA variant calls with DNA analysis for mutations can impact METex14 splicing.

While high METex14 RNA expression may reflect oncogene dependency, the absence of protein-level data in our study is a limitation. Notably, recent trials of telisotuzumab vedotin (Teliso-V), a MET-targeting antibody drug conjugate,(39) have incorporated MET IHC positivity as an enrollment criterion (defined as ≥50% tumor cells with 2+/3+ staining), reinforcing the potential value of protein-level biomarkers predicting MET pathway addiction. Future studies should explore whether high MET RNA read counts correlate with MET protein overexpression or phospho-MET activity, and whether such features better predict sensitivity to MET-directed therapies.

Data on co-occurring mutations were available for patients in both the C1 and C2 subprotocols; we confirmed the frequent presence of co-mutations in the context of both METamp and METex14. TP53 was the most altered gene in both subprotocols, which has been demonstrated previously as a frequent finding in prospectively sequenced metastatic tumors.(40) True METex14 mutations were exclusive of other MAPK pathway mutations. Our molecular analysis of subprotocol C2 has some limitations: the analysis was performed retrospectively in a small cohort and will require confirmation in prospective studies. If confirmed, our data could inform clinical practice in what is a rare molecularly driven cohort.

In summary, we demonstrate clinical activity using crizotinib in patients with METamp and METex14, with no new safety signals identified; subprotocol C1 met its primary endpoint, however, subprotocol C2 did not. In patients with METex14 tumors, we demonstrate that identifying patients with true pathogenic variants from low-level RNA splice variant transcripts may be impactful. In the context of METex14-driven clinical trials to improve outcomes for patients, studies should aim to identify and enroll patients with true pathogenic variants for treatments with selective next-generation MET-targeted therapies.

Translational Relevance.

MET (MET proto-oncogene) is a receptor tyrosine kinase increasingly recognized as a target across tumors. Recently, MET exon 14 skipping mutation (METex14) has been recognized as a driver event in NSCLC. We conducted a phase II trial within the NCI-MATCH trial of TKI crizotinib in tumors with MET amplification (METamp) or METex14. Crizotinib demonstrated clinical activity across tumors with METamp and METex14, though only subprotocol C1 met its primary end point.

During NCI-MATCH assay development, optimal cutoffs for distinguishing low vs. high METex14 skipping RNA counts were undefined. We hypothesized that RNA read thresholds could correlate with DNA variants driving METex14 splicing. Our analysis demonstrated increased survival associated with METex14 read counts ≥50,000. In METex14 tumors, identifying patients with pathogenic variants from low-level splice variant transcripts may provide clinical benefit in carefully selected patients. Our data may guide future studies where METex14 skipping is used as a treatment selection biomarker.