Molecular characterization of MET fusions from a large real‐world Chinese population: A multicenter study

Abstract Purpose MET is a notable driver gene in the diversity of aberrations with clinical relevance, including exon 14 skipping, copy number gain, point mutations, and gene fusions. Compared with the former two, MET fusions are severely under‐reported, leaving a series of unanswered questions. In this study, we addressed this gap by characterizing MET fusions in a large, real‐world Chinese cancer population. Methods We retrospectively included patients with solid tumors who had DNA‐based genome profiles acquired through targeted sequencing from August 2015 to May 2021. MET fusion‐positive (MET+) patients were subsequently selected for clinical and molecular characterization. Results We screened 79,803 patients across 27 tumor types and detected 155 putative MET fusions from 122 patients, resulting in an overall prevalence of 0.15%. Lung cancer comprised the majority of MET+ patients (92, 75.4%). Prevalence was markedly higher in liver cancer, biliary tract cancer, and renal cancer (range 0.52%–0.60%). It was lower in ovarian cancer (0.06%). A substantial proportion (48/58, 82.8%) of unique partners were reported for the first time. High heterogeneity was observed for partners, with ST7, HLA‐DRB1, and KIF5B as the three most common partners. Mutational landscape analysis of lung adenocarcinoma (n = 32) revealed a high prevalence of TP53 in MET+ alterations, EGFR L858R, EGFR L861Q, and MET amplification. Conclusion To our knowledge, this is currently the largest study in characterizing MET fusions. Our findings warrant that further clinical validation and mechanistic study may translate into therapeutic avenues for MET+ cancer patients.


| INTRODUCTION
MET as an oncogenic driver in lung cancer (LC) has been reported in a number of other malignancies, including gastric cancer, hepatocellular carcinoma, papillary renal cell carcinoma, thyroid cancer, glioma, and sarcoma. [1][2][3][4] Compared with other driver genes, MET is notable in the diversity of its aberrations with established clinical relevance, including exon 14 skipping, copy number gain (CNG), point mutations, and gene fusions. 1,4,5 The MET gene is located on chromosome 7q21-23 and encodes for a receptor tyrosine kinase, which homodimerizes upon binding to hepatocyte growth factor (HGF) and activates downstream signaling pathways such as RAS/ERK/ MAPK and PI3K/AKT, thereby promoting tumorigenesis and tumor progression. 6 Normally, activated MET can be autoregulated by transphosphorylation-mediated recruitment of CBL E3 ligase, which then mediates MET degradation and thereby establishes a negative feedback loop. However, cells may undergo MET-driven oncogenesis or acquisition of therapeutic resistance in the case of elevated MET signaling through gene amplification, constitutive activation or expression via gene fusions, and/ or loss of the transphosphorylation site through exon 14 skipping or point mutations. 1 MET aberrations are currently most extensively characterized in LC, which shows a prevalence of approximately 5%, 5 including exon 14 skipping 7-9 and CNG. 4,[10][11][12] In comparison, there is a paucity of data on MET fusions. MET fusions are clinically relevant in that it results in constitutive MET activation, thereby promoting oncogenesis or therapeutic resistance. 1 Apart from isolated reports of response to MET tyrosine kinase inhibitors (TKIs) [13][14][15] or mechanism of acquired resistance to EGFR-TKIs, 16 current knowledge regarding MET fusions is mostly derived from three studies, which identified 2 (0.59%, 2/337) MET fusions in lung adenocarcinoma, 17 1 (0.04%, 1/2410) 18 and 15 (0.26%, 15/5695) fusions involving the MET kinase domain (KD) in non-small cell lung cancer (NSCLC), respectively. 19 The low prevalence and heterogeneity in partner genes and breakpoints warrant the screening of large cohorts for characterizing MET fusions.
Herein, we report a retrospective study that screened 79,803 patients and identified 155 putative MET fusions, which involved 58 unique partners, 48 of which reported for the first time. Subsequent characterization revealed the genomic landscape of these fusions and MET fusionpositive lung adenocarcinoma cases. We also highlight the therapeutic relevance of MET fusions for targeted therapy with select LC cases.

| Patient information and samples
This study screened the genomic profiles of patients with solid tumors for MET fusions and analyzed the clinical and molecular characteristics of the MET fusion-positive patients. The study was approved by the institutional review board of Tianjin Medical University Cancer Hospital and Institute. Patients with solid tumors who had available genome profiles acquired through targeted sequencing from August 2015 to May 2021 were retrospectively included. There was no preselection based on tumor type, histologic subtype, age, sex, clinical stage, or sample type. A total of 79,803 patients were screened, corresponding to 41,404 tumor tissue samples, 35,770 plasma, 2,242 pleural effusion, and 387 cerebrospinal fluid. In this work, only one sample of each patient was included. After we reviewed the sample records, several patients had two or more types of samples or had more than two specimens with the same sample type (such as plasma, tissue). For those who had two or more types of samples, only the tissue sample was included. For those who had more than two specimens with the same sample type, only the sample detected the first time was included. Genomic profiling was performed with hybrid capture-based next-generation sequencing (NGS) using panels targeting 8, 41, 68, 108, 168, or 520 cancer-related genes (Burning Rock Biotech). These panels were designed and validated for the identification of base substitutions, insertions, deletions, copy number variations (CNV), and gene fusions. 20 Conclusion: To our knowledge, this is currently the largest study in characterizing MET fusions. Our findings warrant that further clinical validation and mechanistic study may translate into therapeutic avenues for MET+ cancer patients.

K E Y W O R D S
KIF5B-MET, lung cancer, MET fusion, novel partners, prevalence 2.2 | DNA extraction, library construction, targeted sequencing, and bioinformatic analysis All wet-lab procedures were performed at a College of American Pathologists (CAP)-and Clinical Laboratory Improvement Amendments (CLIA)-certified clinical diagnostic laboratory (Burning Rock Biotech). Briefly, DNA from tumor tissues was extracted with a QIAamp DNA tissue kit, and cell-free DNA from liquid biopsies was extracted with a QIAamp Circulating Nucleic Acid kit (Qiagen). DNA library construction and targeted sequencing with a commercial panel of 8, 41, 68, 108, 168, or 520 genes (Burning Rock Biotech) were performed as previously described. 21,22 Sequencing was conducted on Nextseq500 (Illumina) at a target depth of 1000× for tumor samples and 10,000× for liquid biopsy samples.

| NGS data analysis
All the reads were mapped to the human genome (hg19) with Burrows-Wheeler Aligner (BWA). 23 Local alignment optimization, mark duplication, and variant calling were performed using Genome Analysis ToolKit (GATK), 24 picards, and VarScan. 25 Gene rearrangements were called with FACTERA, 26 and copy CNVs were analyzed based on sequencing depth. Variants were filtered using the VarScan fpfilter pipeline, and loci with depths <100 were filtered out. At least two and five supporting reads were needed for insertions and deletions (indels), while eight supporting reads were needed to call SNVs, in both plasma and tissue samples. According to the ExAC, 1000 Genomes, dbSNP, ESP6500SI-V2 database, variants with population frequency over 0.1% were grouped as SNP and excluded from further analysis. The remaining variants were annotated with ANNOVAR 27 and SnpEff. 28 Tumor mutation burden (TMB) was calculated as number of nonsynonymous somatic alterations on the coding regions of the targeted genes per million base pairs after excluding alterations with an allelic frequency of <2% from tissue samples or <0.2% from liquid biopsy samples. The concurrent alterations with MET fusions were investigated in 27 MET fusion-positive lung adenocarcinoma (LUAD) patients who underwent targeted sequencing using a 168-or 520-gene panel. In addition, the concurrent lung cancer driver alterations with MET fusions were also investigated in 32 MET+ LUAD patients who underwent targeted sequencing using panels targeting 8, 41, 68, 108, 168, or 520 cancer-related genes. These panels included eight identical genes, also known as classic lung cancer oncogenes, including EGFR, KRAS, BRAF, ALK, ROS1, MET, RET, and ERBB2.

| Statistical analysis
All statistical analyses were performed using the R programming language. Chi-square or Fisher's exact test was used to compare between two groups in the proportions of values of a nominal variable, such as percentage of MET fusion-positive patients. Wilcoxon signed-rank test was used to compare the central tendency of a continuous variable, such as TMB, between two groups. All tests were two-sided with a significance level set at p < 0.05.

| Patient characteristics
In this study, we retrospectively screened 79,803 patients with solid tumors who had DNA-based genomic profiles acquired with panel-based targeted sequencing. A total of 155 putative MET fusions events were identified in 122 patients (151 samples), resulting in an overall prevalence of 0.15% (122/79803). Figure 1 shows the characteristics of MET fusions (hereafter referred to as MET+) across cancer types. LC represents the largest subset (n = 92, 75.4%), followed by colorectal cancer (CRC, n = 11, 9.0%). These two cancer types showed similar prevalence rates of MET fusions (0.16% and 0.19%), which were comparable with overall prevalence. Although liver cancer (n = 3), biliary tract cancer (BTC, n = 2), and renal cancer (n = 2) patients showed higher prevalence ranging from 0.52% to 0.60%, ovarian cancer (OC, n = 1) showed a markedly lower rate at 0.06%, these results may be artifacts due to the small number of MET+ patients. Pairwise comparison between different tumor types revealed no significant difference in prevalence levels ( Figure S1). Other clinicopathologic characteristics of the patients are summarized in Table 1. Sixty-three (51.6%) of the MET+ patients were men and 53 (43.4%) were women. Gender information was not documented for the remaining six cases (n = 1.2%). Median age was 58 years with a range of 27-87 years. More than half had stage III disease (n = 65, 53.3%), followed by stage IV disease (n = 41, 33.6%).

MET fusions
A by-patient view of clinical and molecular characteristics for each MET+ patient is provided in Table S1. Most (87/122, 71.3%) patients harbored ≥1 fusion in which MET was the 3' partner, among whom 21 (17.2%) carried multiple MET fusions, including 12 (9.8%) cases in which MET was also found as a 5' partner, and the remaining 9 patients had >1 (range 2-5) fusions, all of which with MET as the 3' partners. In terms of fusion events, MET was the 3' partner in 99 (63.9%) events and the 5' partner in the remaining 56 (36.1%; Figure 2A).
The majority (48/58, 82.8%) of the MET partners identified in this study, to our knowledge, has not been previously reported. 1 As summarized in Figure 2B, in addition to LC (n = 47), these novel fusions were also identified in CRC (n = 7), BTC (n = 2), liver cancer (n = 2), gastroesophageal cancer (GEC, n = 1), breast cancer (n = 1), pancreatic cancer (1 event), and other cancers (n = 1). A considerable proportion of MET fusions were intergenic (65/155, 41.9%). Precise 59 partners were found in nonintergenic fusions, with 76.3% of partners (n = 45) identified only once and 86.4% of partners (n = 51) identified ≤2 times. ST7 stood out as the most common partner occurring in 7 events, followed by HLA-DRB1 and KIF5B (5 events each), and CAPZA2 and CFTR (4 events each). Notably, all nonintergenic fusions with HLA-DRB or CAPZA2 had MET at the 3' end and occurred in LC. Fusions with KIF5B were also all detected in LC and had MET as the 3' partner except for one occasion. Besides LC (4 events), ST7-MET was also present in CRC, OC, and other cancer types (1 event for each). Interestingly, apart from LC (2 events), fusions with CFTR were detected in 1 GEC and 1 breast cancer case in the forms of MET-CFTR and CFTR-MET, respectively. Analysis of breakpoint distribution indicated exon 15 as the most common exon ( Figure 2C). Figure 2D shows a by-patient view of clinical stages and fusions-specific features of the 70 nonintergenic fusions with MET as the 3' partner. Fewer than half of them (29/

| Mutational landscape of MET fusions in lung adenocarcinoma
LUAD was the predominant histologic subtype with MET fusions, we then focused on the 32 MET+ LUAD cases for further characterization ( Figure 3A). These patients showed a median age of 58 years (range 40-79) and equal numbers of men and women (15/32, 46.9% for each; Table 1). Similar to all MET+ cases, most (29/32, 90.6%) had stage III or IV disease.  L iv e r c a n c e r R e n a l c a n c e r  Figure 3B). Figure 3C illustrates the 20 nonintergenic fusions with MET as the 3' partner, of which the majority (17/20, 85.0%) harbored MET KD or lacked exon 14 (11/20, 65.0%; Figure 3D).
In the third highlighted case, a 62-year-old woman (P107) had a KIF5B-MET (K24:M15)-positive LUAD resistant to second-line crizotinib, as indicated by rapid progression within 2 months, which contrasted with previous report of durable response (10 months) to crizotinib in a patient with stage IV LUAD harboring KIF5B-MET (K24:M14). Together, these cases demonstrated heterogeneity in the response to multi-kinase and selective MET-TKIs among carriers of different MET fusions.

| DISCUSSION
Although MET was first identified as part of an oncogenic gene arrangement with TRP, 29 the prevalence of its fusions is far lower than that of exon 14 skipping mutations or amplification. As a result, MET fusions and their therapeutic implications have been largely ignored. In this study, we addressed this challenge with a large-scale screening of 79,803 patients across 27 tumor types, which led to the detection of 155 putative MET fusions from DNA-based genomic profiles of 122 patients. Subsequent analyses answered fundamental questions such as prevalence rates in different tumor types, whether the fusions were enriched in any of these tumors, percentage of MET as the 3' partner and carriers of multiple fusions, location of the MET sequence, and breakpoints and partners. Notably, this study also identified 48 novel partners, which provided candidates for preclinical research into their biology and potential as drug targets. In terms of cancer types, LC patients represented the largest MET+ population (92/122, 75.4%) and LUAD as the largest histologic subtype (32/122, 26.2%). Analysis of the mutational landscape of MET+ LUAD revealed a prevalence of altered TP53 or EGFR much higher than those of other genes and found MET amplification, EGFR L858R, and EGFR L861Q as frequently concurrent aberrations.
MET amplification and exon 14 skipping alterations are well-known oncogenic drivers in multiple cancer types. A number of previous studies on the prevalence and clinicopathological characteristics of MET amplification/exon 14 skipping alterations in solid cancers across countries have been documented. 9,[30][31][32][33] However, MET fusions are poorly defined. To our knowledge, studies on MET fusions in large cohorts of tumors are scarce. Pekova et al. have reported EML4-MET occuring in a pediatric papillary thyroid carcinoma patient (PTC, 1.08%, 1/93) from a large cohort. 34 Although pediatric PTC patients were not included in this work, EML4-MET was detected in a stage IV CRC patient (0.02%, 1/5725). A recent study by Yang et al. has revealed MET fusions in solid tumors from a multicenter study in China, indicating its incidence with 0.34% (37/10882) across cancer types and 0.07% (4/5835) in LC, 35  This study provided a comprehensive account of the frequencies of MET fusions in various cancer types. Prevalence was 0.15% overall and ranged from 0.06% to 1.38% across tumor types. LC, CRC, breast cancer, and GEC each had more than 2,000 screened patients and showed similar, relatively low rates. In contrast, BTC, liver cancer, and renal cancer, all with <500 screened patients, showed markedly higher rates, which may or may not be an artifact due to the small sample size and warrants further investigation. Pairwise Fisher's exact test showed no significant difference among the tumor-specific prevalence rates ( Figure S1). Interestingly, although there is evidence of MET fusions enriched in gliomas, including one report of ~10% prevalence in pediatric glioblastoma 36 and one of 214 fusions detected from 272 glioma samples with RNA-seq, 37 none of the 66 screened gliomas patients in this study was MET+. This discrepancy was likely attributable to the different patient demographics and method of genome profiling, which suggested the significance of combining DNA-and RNA-based detection in future studies of MET rearrangements.
Our analysis revealed substantial heterogeneity in MET fusion partners, which was distinct from other oncogenic driver fusions such as those involving ALK 38 or RET. 20 Among the 58 unique partners, 44 were identified only once and 50 no more than twice. The top 3 most common partners ST7, HLA-DRB1, and KIF5B were detected in 11.0% (17/155) events. In stark contrast, in a study of NSCLC patients confirmed to be ALK-positive via immunohistochemistry and fluorescence in situ hybridization, those carrying fusions with one of the three most common partners (EML4, KIF5B, and HIP1) constituted 86.1% of (136/158) cases. 38 A study screening DNAbased genomic profiles for RET-positive cancer patients showed that 74% of fusions involved KIF5B, CCDC6, or NCOA4. 20 In comparison, MET was similar to NTRK in that both fusions were rare (NTRK: 899/295675, 0.30%) and demonstrated no marked partner preference. In a similar study of NTRK fusions in cancer patients, those harboring fusions with any of the 3 most frequent partners ETV6, TPM3, and LMNA comprised 22.3% (198/889) of all NTRK-positive cases, a percentage closer to our observation with MET (13/122 patients, 10.7%).
One interesting observation was the high frequency of TP53 alterations in MET+ LUAD, which amounted to 74.1% (20/27). In comparison, TP53 has altered in 52.3% of (296/566) LUAD cases in The Cancer Genome Atlas database 8 and 61.1% (151/247) LUAD cases in the MSK-IMPACT database, respectively. 39 The difference in incidence rates was not statistically significant for either comparison, which could be due to the small number of MET+ LUAD cases. This finding warrants further validation and may provide alternative therapeutic targets for MET+ LUAD patients. MET fusions are prohibitively rare for clinical trials of a MET-altered cohort and as a result, no such trial has been reported to date. 1 Our findings also provided candidates for a mechanistic study that may translate into therapeutic inventions. Of the 58 unique fusion partners identified in our study, 48 were novel. Importantly, among the corresponding fusions, 13 were in-frame with KD-intact MET as the 3' partner, involving 5' partners actionable targets such as protein kinase A (PRKAR1A), 40 targets with ongoing drug development such as Tousled-like kinase 2 (TLK2), 41 proposed prognostic biomarkers such as LRIG 42 and DOCK4, 43 and genes with preclinical evidence of cancer-associated activity such as DIXDC1 44 and YEATS2 . 45 Functional insights into the updated list may open avenues to new therapies for MET+ cancers. Also noteworthy were the fusions with MET as the 5' partner. The putative protein products may not undergo dimerization and subsequent MET transactivation and yet still be clinically relevant, as suggested by an interesting case report of MET-UBE2H as a mechanism of acquired EGFR resistance in LC. 16 In addition to partners, breakpoints that result in the inclusion of MET exon 14 may result in elevated oncogenic activity of the chimeric protein. For instance, our third highlighted case and a previously reported one 13 shared tumor histology (LUAD), 5' partner (exons 1-24 of KIF5B), and regimen, although the former lacked MET exon 14. Nonetheless, the two cases showed vastly different responses to crizotinib as the former had a PFS of 2 months with best response of stable disease, while the response lasted for 10 months for the patient whose KIF5B-MET included exon 14. This contrast was consistent with the observation that the prognosis for patients with MET exon 14 skipping was generally worse compared with METamplified NSCLC. 46 This study was limited by several factors. First, the low prevalence of MET fusions and heterogeneity of partners and breakpoint require large cohort sizes for highconfidence characterization. Second, although this study is to date the largest effort in this field, MET+ patients were still scarce except in LC, which limited the depth of insights gained from the remaining tumor types. Third, MET fusions were identified from DNA-based genomic profiles and were not validated on the mRNA or protein level. Therefore, the putative fused genes reported in our study will need further validation on the RNA and/or protein level. Fourth, the therapeutic data of most enrolled patients were not available. Therefore, it is challenging to determine whether the MET fusion is a primary or acquired event. The study could benefit from more detailed clinical information, especially those regarding treatment and response/ progression, which could have led to the characterization of drug efficacy for different MET fusions and potential resistance mechanisms of MET+ patients.

| CONCLUSION
To our knowledge, this study is currently the largest in characterizing MET fusions. Our findings provide insights into the structure, partner distribution, and concurrent genomic alterations of MET fusions. This study warrants further clinical validation and mechanistic investigations that may translate into therapeutic avenues for MET+ cancer patients.

FUNDING INFORMATION
This study was supported by the Translational research project of the Medical Oncology Key Foundation of