Abstract
We have previously shown that gene expression profiles of oral leukoplakia (OL) may improve the prediction of oral cancer (OC) risk. To identify new targets for prevention, we performed a systematic survey of transcripts associated with an increased risk of oral cancer and overexpressed in OC vs normal mucosa (NM).
We used gene expression profiles of 86 patients with OL and available outcomes from a chemoprevention trial of OC and NM. MET expression was evaluated using immunohistochemistry in 120 OL patients, and its association with OC development was tested in multivariable analysis. Sensitivity to pharmacological Met inhibition was tested in vitro in premalignant and OC cell lines (n = 33) and in vivo using the 4-NQO model of oral chemoprevention (n = 20 mice per group). All statistical tests were two-sided.
The overlap of 693 transcripts associated with an increased risk of OC with 163 transcripts overexpressed in OC compared with NM led to the identification of 23 overlapping transcripts, including MET. MET overexpression in OL was associated with a hazard ratio of 3.84 (95% confidence interval = 1.59 to 9.27, P = .003) of developing OC. Met activation was found in OC and preneoplastic cell lines. Crizotinib activity in preneoplastic and OC cell lines was comparable. ARQ 197 was more active in preneoplastic compared with OC cell lines. In the 4-NQO model, squamous cell carcinoma, dysplasia, and hyperkeratosis were observed in 75.0%, 15.0%, and 10.0% in the control group, and in 25.0%, 70.0%, and 5.0% in the crizotinib group (P < .001).
Together, these data suggest that MET activation may represent an early driver in oral premalignancy and a target for chemoprevention of OC.
Oral cancer (OC) is one of the most frequent smoking- and alcohol-related cancers in the world and the most common cancer of the head and neck globally. The substantial therapeutic morbidity highlights the need to develop prevention strategies (1–5). Furthermore, in oral leukoplakia (OL), the most common oral premalignancy, clinical (eg, homogenous or in-homogenous) and histologic features, are poorly predictive of cancer risk, and surgical resection has not been shown to reduce oral cancer development (4,6,7).
Recent efforts from various groups and The Cancer Genome Atlas (TCGA) have increased our understanding of the disease by providing a comprehensive genomic characterization of HNSCC (8–10). Integrative analyses have identified a limited number of pathways targeted by frequent genome alterations. However, multiple challenges remain, including the fact that tumors are likely not addicted to all those shared or unique genomic features and that tumors may be addicted to nonactionable drivers or tumor suppressor genomic drivers (eg, HRAS and loss of heterozygosity profiles) (11).
Besides constitutive activation of oncogenes by genomic alterations, oncogene addiction has been documented in genetically engineered conditional mouse models that allow expressing or downregulating an oncogene to induce tumor formation. In various models, elevated expression levels of Myc, Ras, c-erb2, and Met are sufficient for oncogenic transformation (12–16).
Based on our previous report showing a strong association between gene expression profiles of oral leukoplakia and the risk of developing OC (17), we hypothesized that transcripts whose expression is statistically significantly associated with OC development and consistently overexpressed in OC compared with normal mucosa (NM) may represent potential targets for prevention. We identify MET as one of those transcripts and show that its inhibition both in vitro and in vivo may prevent oral tumorigenesis.
Methods
Genome-Wide Expression Profiles of Oral Leukoplakias Collected Prospectively
We included 162 patients, previously described in detail (18). All participants provided written informed consent, and the study was approved by the M.D. Anderson Institutional Review Board. All the samples from patients who developed oral cancer (n = 35) were selected for gene expression profiling, as well as 51 samples from patients who did not develop OSCC, randomly selected among 106 patients. The median follow-up of the 51 patients who did not develop oral cancer was 6.08 years (17). Whole-genome expression profiles from 86 OLs collected at baseline were used to identify transcripts associated with an increased risk of OC (GSE26549) (17).
Expression Profiles, Data Processing, and Analysis of Normal Mucosa, Dysplasia, and OC
A total of one mouse and three human data sets were mined (GSE9844, GSE30784, TCGA, and GSE75421) (10,19–21). In GSE9844, genes differentially expressed in OC vs NM were identified after correcting by Bonferroni method. Detailed information is provided in Supplementary Table 1 and the Supplementary Methods (available online).
MET Immunostaining of Oral Leukoplakias Collected Prospectively
Tissue sections (4 μm thick) from formalin-fixed, paraffin-embedded tissue blocks of oral premalignant lesion (OPL) were mounted on positively charged glass slides. Total MET immunostaining was done using the avidin-biotin peroxidase complex technique, as previously described (10). Detailed information on immunostaining is provided in the Supplementary Methods (available online). MET expression was scored by considering membranous and cytoplasmic staining. The expression intensity was scored from 0 to 3. Scores were based on examination of the whole section in each biopsy under a multiheaded microscope by three observers (PS, WL, and AKEN), who were blinded to the clinical information.
Cell Culture of Premalignant and HNSCC Cell Lines
All 33 cell lines were obtained from Dr. Jeffrey Myers (The University of Texas M.D. Anderson Cancer Center, Houston, TX) and maintained using standard cell culture techniques. They included HNSCC cell lines (n = 20) and immortalized and nontumorigenic keratinocyte cell lines (n = 4). Detailed information is available in the Supplementary Methods (available online).
Immunoblot Analysis of Protein Lysates
Detailed information is provided in the Supplementary Methods (available online).
Cytotoxicity Assay
In vitro sensitivity to ARQ 197 and crizotinib, two inhibitors of MET activity, was studied in MSKLeuk1, MSKLeuk1s, HaCaT, and HOK-16B, as well as the 20 cancer cell lines derived from the oral cavity. Cytotoxicity was assessed using a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay, as described previously (22). Under each experimental condition, at least four independent wells were treated. Half-maximal inhibitory concentration was calculated using GraphPad Software (GraphPad Software, Inc., San Diego, CA, USA).
Administration of 4-NQO and Treatment With Crizotinib
A total of 40 male CBA mice (n = 20 mice per group), six to eight weeks of age, were purchased from the Jackson Laboratory (Bar Harbor, ME, USA), housed in the Animal Resource Facility under controlled conditions, and fed normal diet and autoclaved water. All animal procedures were carried out in accordance with Institutional Animal Care and Use Committee–approved protocols. Mice were given 4-NQO (100 μg/mL) in their drinking water for a period of eight weeks at the required dose (100 μg/mL), as previously described (see the Supplementary Methods, available online) (23), returned to normal water, and then randomly assigned to observation or treatment by crizotinib at the beginning of week 9. Initiation of crizotinib treatment at this time point was chosen because we have previously shown that following the eight weeks of 4-NQO administration, mice develop histologically identifiable premalignant lesions, mimicking the clinical setting.
Crizotinib was dissolved in DMSO (P8192-5X10ML, Sigma Chemical, Sigma-Aldrich, Saint Louis, MO, USA). Mice receiving crizotinib treatment were given a daily dosage of 25 mg/kg/d for 24 weeks via oral gavage. The rationale for this dose came from preclinical studies evaluating crizotinib activity in xenografts or patient-derived xenografts that found that 25 and 50 mg/kg/d showed similar activity (24–26). Because we were treating premalignancy rather than cancer, we decided to use the lower active dose (25 mg/kg/d), consistent with other recent preclinical studies (27–30). The high dose (100 mg/kg/d) is typically used in studies of crizotinib resistance (31,32).
Mice were killed in accordance with Institutional Animal Care and Use Committee recommendations. Specifically, cervical dislocation was done after anesthesia by i.p. injection of xylazine and ketamine. Immediately following death, the tongues were excised, longitudinally bisected, and processed in 10% buffered formalin and embedded in paraffin. Fifty 5 μm sections from each specimen were then cut, and the 1st, 10th, 20th, 30th, 40th, and 50th slides were stained with hematoxylin and eosin for histopathologic analysis. Histologic diagnoses were rendered using well-established criteria that we use in diagnostic pathology and in our lab when using this model (23,33–35).
Statistical Analysis
Summary statistics, including median and range values, were used to describe the distribution of candidate genes in different data sets. Cancer and normal samples were compared using a log2 fold-change and a paired or nonpaired two-sided Student’s t test. The Kruskal-Wallis test was used when more than two groups were compared. The Kaplan-Meier method was used to construct oral cancer–free survival (OCFS) curves, and the log-rank test was used to test the difference by covariate levels. For time-to-event analysis, Kaplan-Meier curves were plotted. The median time to event with 95% confidence interval (CI) and event-free survival rates at years 5 and 10 were determined. Cox proportional hazards models were used for univariate and multivariable analyses. The weighted residuals score test was used to verify the assumption of proportionality (36). All statistical tests were two-sided, and P values of .05 or less were considered statistically significant.
Results
Identification of Transcripts Associated With an Increased Risk of OC and Overexpressed in OC
From our previous study (17), a comprehensive list of 2182 transcripts associated with OCFS time was defined. Among them, 693 transcripts were associated with an increased risk of OC development, with a false discovery rate of 11.0% using the BUM model (37). We found an overlap of 23 transcripts with the 163 transcripts overexpressed in OC vs NM in GSE9844 (Figure 1). The complete list is provided as Supplementary Table 2 (available online). Of note, no association between MET expression levels with high-risk dysplasia compared with lower-risk dysplasia was observed (data not shown). We focused our attention on targetable genes, especially the MET oncogene, a tyrosine-kinase hepatocyte growth factor (HGF) receptor. When MET expression was dichotomized into high vs low levels based on the median expression values, the median OCFS was 6.22 years in patients with high MET levels, while it was not reached in patients with low MET levels (log-rank P = 0.008). In GSE9844, MET expression was higher in OC vs NM (two-tailed unpaired t test P < 0.001) (Figure 1). Of note, HGF gene expression was not associated with oral cancer–free survival and was not different in patients with oral leukoplakia who developed or did not develop oral cancer during follow-up (data not shown).
Overlapping transcripts associated with an increased risk of developing oral cancer in patients with oral leukoplakia (GSE26549) (A) and overexpressed in a set of oral cancer patients compared with normal oral mucosa patients (GSE9844) (B). Statistical significance was given by the Wald test of a Cox proportional hazard model and by a log-rank test for the Kaplan-Meier curve (A) and by a two-tailed unpaired t test (B). All statistical tests were two-sided. FC = fold-change; HR = hazard ratio; nal = normal; NM = normal mucosa; OC = oral cancer.
Met Protein Expression in OL and the Time to Oral Cancer Development
In order to confirm this observation, we measured total MET protein expression levels by immunohistochemistry on paraffin-embedded, formalin-fixed sections from the same cohort of patients in 120 individuals with available material. No MET expression was observed in 26 OPL (21.7%). MET was expressed at low (score 1), mild (score 2), and high (score 3) levels in 34 (28.3%), 39 (32.5%), and 21 (17.5%) OPL examined, respectively (Figure 2, A–C). Time to OC development was statistically significantly different in the four groups of patients (log-rank test P = .001) (Supplementary Figure 1, available online). When dichotomized into two groups, patients with a score of 0 or 1 vs patients with a score of 2 or 3, the outcome difference was even more striking (log-rank test P < .001) (Figure 2D). In the univariate analysis, sex, race, treatment arm, smoking, and alcohol history were not statistically significantly associated with OC development. Age and dysplasia were associated with OC development. In particular, patients with dysplasia had an OCFS rate at 10 years of 55.0% (95% CI = 0.40 to 0.76) vs 78.0% (95% CI = 0.68 to 0.89, P = .04). OCFS was statistically significantly decreased in patients with high MET expression levels (P < .001, log-rank test) (Figure 2). At 10 years after the random assignment date, the OCFS rate for the patients with a MET score of 0 or 1 was 86.0% (95% CI = 76.0% to 97.0%) compared with 52.0% for the patients with a score of 2 or 3 MET-positive OPL (95% CI = 39.0% to 69.0%) (Table 1). For multivariable analysis, the Cox proportional hazards model was fitted for MET expression (score 2–3 vs score 0–1) (Table 2). Age, treatment arm, histology, and MET score were entered in the model. MET score was the only independent factor, with a hazard ratio of 3.84 (95% CI = 1.59 to 9.27, P = .003) (Table 2). Histology and age did not reach statistical significance in multivariable analysis.
Univariate analysis for clinical, pathological variables and MET expression*
| Variable | No. | Event | Median time to OC (95% CI), y | OCFS rate at 5 y (95% CI) | OCFS rate at 10 y (95% CI) | P† |
|---|---|---|---|---|---|---|
| All patients | 120 | 31 | NA (NA to NA) | 0.78 (0.71 to 0.86) | 0.69 (0.60 to 0.80) | |
| Sex | ||||||
| Female | 59 | 15 | NA (NA to NA) | 0.76 (0.65 to 0.88) | 0.69 (0.57 to 0.84) | .88 |
| Male | 61 | 16 | NA (10.7 to NA) | 0.8 (0.70 to 0.91) | 0.69 (0.56 to 0.84) | |
| Race | ||||||
| Other | 12 | 3 | NA (5.06 to NA) | 0.77 (0.53 to 1.00) | 0.64 (0.39 to 1.00) | .83 |
| White | 108 | 28 | NA (NA to NA) | 0.78 (0.70 to 0.87) | 0.7 (0.60 to 0.80) | |
| Histology | ||||||
| Dysplasia | 44 | 16 | NA (6.65 to NA) | 0.68 (0.54 to 0.84) | 0.55 (0.4 to 0.76) | .04 |
| Hyperplasia | 76 | 15 | NA (NA to NA) | 0.84 (0.76 to 0.93) | 0.78 (0.68 to 0.89) | |
| Treatment | ||||||
| 13cRA | 63 | 18 | NA (NA to NA) | 0.74 (0.64 to 0.86) | 0.66 (0.54 to 0.81) | .67 |
| BC-RP | 32 | 7 | NA (NA to NA) | 0.83 (0.71 to 0.98) | 0.78 (0.64 to 0.96) | |
| RP only | 25 | 6 | 6.65 (6.65 to NA) | 0.82 (0.67 to 1.00) | ||
| Smoking status | ||||||
| Current | 42 | 6 | NA (NA to NA) | 0.86 (0.76 to 0.98) | 0.81 (0.67 to 0.97) | .13 |
| Former | 45 | 17 | NA (6.21 to NA) | 0.72 (0.6 to 0.87) | 0.58 (0.43 to 0.77) | |
| Never | 33 | 8 | NA (NA to NA) | 0.77 (0.64 to 0.94) | 0.73 (0.59 to 0.91) | |
| Alcohol use | ||||||
| Current | 70 | 17 | NA (NA to NA) | 0.81 (0.71 to 0.91) | 0.71 (0.59 to 0.85) | .47 |
| Former | 10 | 4 | 6.21 (2.27 to NA) | 0.7 (0.47 to 1.00) | 0.35 (0.08 to 1.00) | |
| Never | 40 | 10 | NA (NA to NA) | 0.76 (0.64 to 0.91) | 0.73 (0.60 to 0.89) | |
| MET expression | ||||||
| 0–1 | 60 | 7 | NA (NA to NA) | 0.92 (0.85 to 1.00) | 0.86 (0.76 to 0.97) | <.001 |
| 2–3 | 60 | 24 | NA (5.06 to NA) | 0.64 (0.52 to 0.78) | 0.52 (0.39 to 0.69) |
| Variable | No. | Event | Median time to OC (95% CI), y | OCFS rate at 5 y (95% CI) | OCFS rate at 10 y (95% CI) | P† |
|---|---|---|---|---|---|---|
| All patients | 120 | 31 | NA (NA to NA) | 0.78 (0.71 to 0.86) | 0.69 (0.60 to 0.80) | |
| Sex | ||||||
| Female | 59 | 15 | NA (NA to NA) | 0.76 (0.65 to 0.88) | 0.69 (0.57 to 0.84) | .88 |
| Male | 61 | 16 | NA (10.7 to NA) | 0.8 (0.70 to 0.91) | 0.69 (0.56 to 0.84) | |
| Race | ||||||
| Other | 12 | 3 | NA (5.06 to NA) | 0.77 (0.53 to 1.00) | 0.64 (0.39 to 1.00) | .83 |
| White | 108 | 28 | NA (NA to NA) | 0.78 (0.70 to 0.87) | 0.7 (0.60 to 0.80) | |
| Histology | ||||||
| Dysplasia | 44 | 16 | NA (6.65 to NA) | 0.68 (0.54 to 0.84) | 0.55 (0.4 to 0.76) | .04 |
| Hyperplasia | 76 | 15 | NA (NA to NA) | 0.84 (0.76 to 0.93) | 0.78 (0.68 to 0.89) | |
| Treatment | ||||||
| 13cRA | 63 | 18 | NA (NA to NA) | 0.74 (0.64 to 0.86) | 0.66 (0.54 to 0.81) | .67 |
| BC-RP | 32 | 7 | NA (NA to NA) | 0.83 (0.71 to 0.98) | 0.78 (0.64 to 0.96) | |
| RP only | 25 | 6 | 6.65 (6.65 to NA) | 0.82 (0.67 to 1.00) | ||
| Smoking status | ||||||
| Current | 42 | 6 | NA (NA to NA) | 0.86 (0.76 to 0.98) | 0.81 (0.67 to 0.97) | .13 |
| Former | 45 | 17 | NA (6.21 to NA) | 0.72 (0.6 to 0.87) | 0.58 (0.43 to 0.77) | |
| Never | 33 | 8 | NA (NA to NA) | 0.77 (0.64 to 0.94) | 0.73 (0.59 to 0.91) | |
| Alcohol use | ||||||
| Current | 70 | 17 | NA (NA to NA) | 0.81 (0.71 to 0.91) | 0.71 (0.59 to 0.85) | .47 |
| Former | 10 | 4 | 6.21 (2.27 to NA) | 0.7 (0.47 to 1.00) | 0.35 (0.08 to 1.00) | |
| Never | 40 | 10 | NA (NA to NA) | 0.76 (0.64 to 0.91) | 0.73 (0.60 to 0.89) | |
| MET expression | ||||||
| 0–1 | 60 | 7 | NA (NA to NA) | 0.92 (0.85 to 1.00) | 0.86 (0.76 to 0.97) | <.001 |
| 2–3 | 60 | 24 | NA (5.06 to NA) | 0.64 (0.52 to 0.78) | 0.52 (0.39 to 0.69) |
Median time to oral cancer development (in years) and the oral cancer–free survival rates at years 5 and 10, along with the 95% confidence intervals for all patients and for each group. 13cRA = 13-cis-retinoic acid; BC = beta-carotene; CI = confidence interval; OC = oral cancer; OCFS = oral cancer–free survival; RP = retinyl palmitate.
The log-rank test was used to compare the difference in survival between the prognostic factor groups, and all tests were two-sided.
Univariate analysis for clinical, pathological variables and MET expression*
| Variable | No. | Event | Median time to OC (95% CI), y | OCFS rate at 5 y (95% CI) | OCFS rate at 10 y (95% CI) | P† |
|---|---|---|---|---|---|---|
| All patients | 120 | 31 | NA (NA to NA) | 0.78 (0.71 to 0.86) | 0.69 (0.60 to 0.80) | |
| Sex | ||||||
| Female | 59 | 15 | NA (NA to NA) | 0.76 (0.65 to 0.88) | 0.69 (0.57 to 0.84) | .88 |
| Male | 61 | 16 | NA (10.7 to NA) | 0.8 (0.70 to 0.91) | 0.69 (0.56 to 0.84) | |
| Race | ||||||
| Other | 12 | 3 | NA (5.06 to NA) | 0.77 (0.53 to 1.00) | 0.64 (0.39 to 1.00) | .83 |
| White | 108 | 28 | NA (NA to NA) | 0.78 (0.70 to 0.87) | 0.7 (0.60 to 0.80) | |
| Histology | ||||||
| Dysplasia | 44 | 16 | NA (6.65 to NA) | 0.68 (0.54 to 0.84) | 0.55 (0.4 to 0.76) | .04 |
| Hyperplasia | 76 | 15 | NA (NA to NA) | 0.84 (0.76 to 0.93) | 0.78 (0.68 to 0.89) | |
| Treatment | ||||||
| 13cRA | 63 | 18 | NA (NA to NA) | 0.74 (0.64 to 0.86) | 0.66 (0.54 to 0.81) | .67 |
| BC-RP | 32 | 7 | NA (NA to NA) | 0.83 (0.71 to 0.98) | 0.78 (0.64 to 0.96) | |
| RP only | 25 | 6 | 6.65 (6.65 to NA) | 0.82 (0.67 to 1.00) | ||
| Smoking status | ||||||
| Current | 42 | 6 | NA (NA to NA) | 0.86 (0.76 to 0.98) | 0.81 (0.67 to 0.97) | .13 |
| Former | 45 | 17 | NA (6.21 to NA) | 0.72 (0.6 to 0.87) | 0.58 (0.43 to 0.77) | |
| Never | 33 | 8 | NA (NA to NA) | 0.77 (0.64 to 0.94) | 0.73 (0.59 to 0.91) | |
| Alcohol use | ||||||
| Current | 70 | 17 | NA (NA to NA) | 0.81 (0.71 to 0.91) | 0.71 (0.59 to 0.85) | .47 |
| Former | 10 | 4 | 6.21 (2.27 to NA) | 0.7 (0.47 to 1.00) | 0.35 (0.08 to 1.00) | |
| Never | 40 | 10 | NA (NA to NA) | 0.76 (0.64 to 0.91) | 0.73 (0.60 to 0.89) | |
| MET expression | ||||||
| 0–1 | 60 | 7 | NA (NA to NA) | 0.92 (0.85 to 1.00) | 0.86 (0.76 to 0.97) | <.001 |
| 2–3 | 60 | 24 | NA (5.06 to NA) | 0.64 (0.52 to 0.78) | 0.52 (0.39 to 0.69) |
| Variable | No. | Event | Median time to OC (95% CI), y | OCFS rate at 5 y (95% CI) | OCFS rate at 10 y (95% CI) | P† |
|---|---|---|---|---|---|---|
| All patients | 120 | 31 | NA (NA to NA) | 0.78 (0.71 to 0.86) | 0.69 (0.60 to 0.80) | |
| Sex | ||||||
| Female | 59 | 15 | NA (NA to NA) | 0.76 (0.65 to 0.88) | 0.69 (0.57 to 0.84) | .88 |
| Male | 61 | 16 | NA (10.7 to NA) | 0.8 (0.70 to 0.91) | 0.69 (0.56 to 0.84) | |
| Race | ||||||
| Other | 12 | 3 | NA (5.06 to NA) | 0.77 (0.53 to 1.00) | 0.64 (0.39 to 1.00) | .83 |
| White | 108 | 28 | NA (NA to NA) | 0.78 (0.70 to 0.87) | 0.7 (0.60 to 0.80) | |
| Histology | ||||||
| Dysplasia | 44 | 16 | NA (6.65 to NA) | 0.68 (0.54 to 0.84) | 0.55 (0.4 to 0.76) | .04 |
| Hyperplasia | 76 | 15 | NA (NA to NA) | 0.84 (0.76 to 0.93) | 0.78 (0.68 to 0.89) | |
| Treatment | ||||||
| 13cRA | 63 | 18 | NA (NA to NA) | 0.74 (0.64 to 0.86) | 0.66 (0.54 to 0.81) | .67 |
| BC-RP | 32 | 7 | NA (NA to NA) | 0.83 (0.71 to 0.98) | 0.78 (0.64 to 0.96) | |
| RP only | 25 | 6 | 6.65 (6.65 to NA) | 0.82 (0.67 to 1.00) | ||
| Smoking status | ||||||
| Current | 42 | 6 | NA (NA to NA) | 0.86 (0.76 to 0.98) | 0.81 (0.67 to 0.97) | .13 |
| Former | 45 | 17 | NA (6.21 to NA) | 0.72 (0.6 to 0.87) | 0.58 (0.43 to 0.77) | |
| Never | 33 | 8 | NA (NA to NA) | 0.77 (0.64 to 0.94) | 0.73 (0.59 to 0.91) | |
| Alcohol use | ||||||
| Current | 70 | 17 | NA (NA to NA) | 0.81 (0.71 to 0.91) | 0.71 (0.59 to 0.85) | .47 |
| Former | 10 | 4 | 6.21 (2.27 to NA) | 0.7 (0.47 to 1.00) | 0.35 (0.08 to 1.00) | |
| Never | 40 | 10 | NA (NA to NA) | 0.76 (0.64 to 0.91) | 0.73 (0.60 to 0.89) | |
| MET expression | ||||||
| 0–1 | 60 | 7 | NA (NA to NA) | 0.92 (0.85 to 1.00) | 0.86 (0.76 to 0.97) | <.001 |
| 2–3 | 60 | 24 | NA (5.06 to NA) | 0.64 (0.52 to 0.78) | 0.52 (0.39 to 0.69) |
Median time to oral cancer development (in years) and the oral cancer–free survival rates at years 5 and 10, along with the 95% confidence intervals for all patients and for each group. 13cRA = 13-cis-retinoic acid; BC = beta-carotene; CI = confidence interval; OC = oral cancer; OCFS = oral cancer–free survival; RP = retinyl palmitate.
The log-rank test was used to compare the difference in survival between the prognostic factor groups, and all tests were two-sided.
Multivariable analysis of MET expression for oral cancer–free survival*
| Covariates | Analysis of maximum likelihood estimates | |
|---|---|---|
| HR (95% CI) | P† | |
| Age | 1.03 (1.00 to 1.06) | .07 |
| Treatment | ||
| BC+RP vs 13cRA | 0.96 (0.39 to 2.36) | .92 |
| RP only vs 13cRA | 0.93 (0.36 to 2.42) | .88 |
| Histology | ||
| Dysplasia vs hyperplasia | 1.53 (0.74 to 3.16) | .26 |
| MET expression | ||
| 2–3 vs 0–1 | 3.84 (1.59 to 9.27) | .003 |
| Covariates | Analysis of maximum likelihood estimates | |
|---|---|---|
| HR (95% CI) | P† | |
| Age | 1.03 (1.00 to 1.06) | .07 |
| Treatment | ||
| BC+RP vs 13cRA | 0.96 (0.39 to 2.36) | .92 |
| RP only vs 13cRA | 0.93 (0.36 to 2.42) | .88 |
| Histology | ||
| Dysplasia vs hyperplasia | 1.53 (0.74 to 3.16) | .26 |
| MET expression | ||
| 2–3 vs 0–1 | 3.84 (1.59 to 9.27) | .003 |
Only MET protein expression is associated with oral cancer development (E/N = 31/120). 13cRA = 13-cis-retinoic acid; BC = beta-carotene; CI = confidence interval; HR = hazard ratio; OC = oral cancer; OCFS = oral cancer–free survival; RET = retinyl palmitate.
The P values were calculated using the Wald test, and all tests were two-sided.
Multivariable analysis of MET expression for oral cancer–free survival*
| Covariates | Analysis of maximum likelihood estimates | |
|---|---|---|
| HR (95% CI) | P† | |
| Age | 1.03 (1.00 to 1.06) | .07 |
| Treatment | ||
| BC+RP vs 13cRA | 0.96 (0.39 to 2.36) | .92 |
| RP only vs 13cRA | 0.93 (0.36 to 2.42) | .88 |
| Histology | ||
| Dysplasia vs hyperplasia | 1.53 (0.74 to 3.16) | .26 |
| MET expression | ||
| 2–3 vs 0–1 | 3.84 (1.59 to 9.27) | .003 |
| Covariates | Analysis of maximum likelihood estimates | |
|---|---|---|
| HR (95% CI) | P† | |
| Age | 1.03 (1.00 to 1.06) | .07 |
| Treatment | ||
| BC+RP vs 13cRA | 0.96 (0.39 to 2.36) | .92 |
| RP only vs 13cRA | 0.93 (0.36 to 2.42) | .88 |
| Histology | ||
| Dysplasia vs hyperplasia | 1.53 (0.74 to 3.16) | .26 |
| MET expression | ||
| 2–3 vs 0–1 | 3.84 (1.59 to 9.27) | .003 |
Only MET protein expression is associated with oral cancer development (E/N = 31/120). 13cRA = 13-cis-retinoic acid; BC = beta-carotene; CI = confidence interval; HR = hazard ratio; OC = oral cancer; OCFS = oral cancer–free survival; RET = retinyl palmitate.
The P values were calculated using the Wald test, and all tests were two-sided.
Met protein expression by immunohistochemistry in oral leukoplakia and time to oral cancer in patients with oral leukoplakia. Sections were immunostained with a total Met-specific antibody. Representative sections with no (A), low (B), or high (C) expression levels are shown. Black scale bar = 200 µm. D) The association of Met protein expression levels with time to oral cancer development was evaluated using a Kaplan-Meier curve. Statistical significance was given by the two-sided log-rank test. E/N = number of events and number of patients.
In order to strengthen these observations, we analyzed MET gene expression in independent series of samples using array-based and RNAseq technologies. Array-based MET gene expression was found to be increasingly expressed in 45 normal mucosa, 17 dysplastic lesions, and 167 OC patients, respectively (P < 0.001) (Figure 3A). In the TCGA, paired normal and SCC samples from the oral cavity were compared, showing an increased MET expression in 27 of 30 patients (paired t test P < .001) (Figure 3B).
MET gene expression levels in two independent data sets from public data repositories. A) Array-based MET expression levels w in normal oral mucosa (n = 45), oral dysplasia (n = 17), and oral cancer (OC; n = 167) patients (GSE30784; P < 0.001); data from The Cancer Genome Atlas showed MET expression levels by RNAseq in 30 paired OC and normal oral mucosa (P < 0.001). B) The horizontal bar represents the log2 median values (A, C). Statistical significance was given by the Kruskal-Wallis test (A, C) and a paired t test (B). All statistical tests were two-sided. NM = normal; OC = oral cancer; OroPh = oropharyngeal; SCC = squamous cell carcinoma.
Met Expression and Activation in Head and Neck Cell Lines
We then evaluated MET protein expression levels and their activation in a large panel of cell lines, including head and neck cancer cell lines and premalignant cell lines (Supplementary Figure 2, available online). Total MET was found to be expressed in almost all the cell lines except one, which was derived from the larynx (UMSCC11B). MET activation was evaluated by the expression levels of phospho-MET. In the vast majority of the cell lines, phospho-MET expression at Tyr1234/Tyr1235 and Tyr1349 was consistent. Some level of MET activation at Tyr1234/Tyr1235 was found in 22 of 33 (66.7%) cell lines, including three of four (75.0%) premalignant and 19 of 29 (65.5%) head and neck cancer cell lines. MET activation at Tyr1349 was found in 24 of 33 (72.7 %) cell lines, including all premalignant cell lines. Supplementary Figure 2 (available online) shows that epidermal growth factor receptor (EGFR) was activated in a number of cell lines comparable with the one harboring Met activation. In general, the level of phospho-EGFR in preneoplastic cell lines was lower compared with HNSCC cancer cells. Total and phospho-Akt as well as total and phospho-Erk 1/2 were expressed homogenously across all cell lines with few exceptions.
Pharmacological Inhibition of Met in Vitro
We next evaluated the effect of MET inhibition on cell growth and tumor formation using a pharmacologic approach (Figure 4). In vitro, both crizotinib and ARQ197 were tested in 19 cell lines. With crizotinib, median half-maximal inhibitory concentrations (IC50) of premalignant (median = 1.4 µM, range = 0.3–2.0 µM) vs OC (median = 1.7 µM, range = 0.6–2.9 µM) cell lines were not statistically significantly different (Figure 4A). MSKleuk1s was the more sensitive cell line (0.3 µM) and had, together with five of 15 OC cell lines (MDA686TU, UMSCC1, 1483, Tu138, and UMSCC47) an IC50s of less than 1 µM. Consistently, phospho-MET inhibition was observed at increasing doses of crizotinib (Figure 4C). An extra-band with 3.2 µM crizotinib in pMt1349 was observed that may correspond to the expression of a MET isoform induced by a bypass mechanism. With ARQ197, premalignant cell lines had lower IC50s (median = 0.3 µM, range = 0.0–0.04 µM) compared with OC cell lines (median = µM, range = 0.2–1.3; P = .005, Student’s t test) (Figure 4B). However, the range of cell line IC50s with ARQ197 was much lower compared with the dose required to inhibit MET activation, suggesting alternate mechanisms of action of the drug (Figure 4D).
Efficacy of Met inhibition in vitro and in vivo. A total of four premalignant and 20 oral cancer cell lines were treated with crizotinib (A, C) or ARQ 197 (B, D) to determine their sensitivity to Met inhibition. The effect of crizotinib (C) and ARQ 197 (D) on Met phosphorylation at increasing concentrations is shown. The effect of Met inhibition with crizotinib was evaluated in vivo in the 4-NQO model of oral tumorigenesis (a total of 40 mice were analyzed, 20 in the control group and 20 in the crizotinib-treated group; chi-square test, P < .001) (E). MET gene expression in normal mucosa, hyperkeratosis, dysplasia, and squamous cell carcinoma in the 4-NQO model of oral carcinogenesis (Kruskal-Wallis, P = .01) (F). The horizontal bar represents the log2 median values (A, B). All statistical tests were two-sided. Ctrl. = control group; IC50 = half maximal inhibitory concentration; NM = normal mucosa; OC = oral cancer; Premalig. = premalignant; SCC = squamous cell carcinoma.
Met Inhibition in the In Vivo Model of 4-NQO-Induced Oral Tumorigenesis
Finally, we tested whether crizotinib would prevent oral cancer development in the 4-NQO model. During the 24-week chemoprevention regimen, no statistically significant differences in food and fluid consumption, activity, or weight were observed when comparing vehicle to crizotinib (data not shown). At the completion of the 32-week study, there was a statistically significant difference in the incidence of dysplasia and OSCC in the crizotinib treatment group compared with the control group (n = 20 mice per group). Overall squamous cell carcinoma (SCC), dysplasia and hyperkeratosis were observed in 75.0%, 15.0%, and 10.0% in the control group, and in 25.0%, 70.0%, and 5.0% in the crizotinib group (P < .001, chi-square) (Figure 4E). This represented a 67.0% decrease in the rate of SCC. Interestingly, while crizotinib blocked the transformation from dysplasia to invasive SCC, it seemed to have little effect to prevent the appearance of dysplasia. Consistently, MET transcript was found to be increasingly expressed during oral tumorigenesis, especially in SCC as compared with normal hyperkeratosis and dysplasia (Kruskal-Wallis P = .01), while HGF expression was low in general and was slightly increased in dysplastic lesions compared with other stages, although not reaching statistical significance (Supplementary Figure 3, available online).
Discussion
In this report, we identify MET among transcripts that are both associated with an increased risk of OC and overexpressed in OC vs normal mucosa. Patients with OL expressing high MET protein levels had a statistically significantly decreased OCFS time. MET overexpression during oral tumorigenesis was observed in two other data sets of patients and the 4-NQO murine model of oral carcinogenesis. Finally, we show that inhibition of MET activation with crizotinib was effective in a panel of tumorigenic and premalignant head and neck cell lines and the 4-NQO model. We conclude that evaluation of MET expression by immunohistochemistry may 1) help identify patients with OL at high risk of OC and 2) represent a new target for chemoprevention.
In several malignant tumors, MET pathway activation was shown to increase tumor growth, angiogenesis, and metastasis. Knowles et al. have shown that 80% of HNSCCs express MET and its ligand HGF or both (38). Seiwert et al. have shown that 84% of HNSCCs analyzed in their study were overexpressing MET (39). MET gene copy number was increased in 13% to 65% of HNSCC tumors, depending on the cutoff used to define increased gene copy number. MET inhibition was associated with decreased proliferation, migration, and motility, as well as angiogenesis (39). MET tyrosine kinase mutations have been detected in 10% to 14% of HNSCC patients, but their functional and prognostic role and statistically significance remain to be determined (40–42). In the TCGA, numbers are even lower, with two of 279 (0.7%) tumors harboring MET amplification and one of 279 (0.4%) tumors having a missense mutation (cBioportal). MET amplification was never observed in a recent study using fluorescence in situ hybridization (43). Finally, Sun et al. identified the role of the c-MET/FZD8 axis in the self-renewal of cancer stem-like cells (CSCs) in HNSCC and suggested targeting this axis to inhibit the CSC population and thus prevent cancer development (44).
One way to interpret our 4-NQO data is that crizotinib had no impact on evolution from normal mucosa to hyperplasia but delayed the progression of dysplasia to carcinoma. In our experimental design, all mice were killed at the same time, making it unfeasible to address this subject. Crizotinib has an excellent toxicity profile, and newer inhibitors in this class are being developed with even lower toxicity (45,46). From data of recent clinical trials, the most frequently reported adverse events were visual changes, nausea, and diarrhea. We also found activation of AKT in a large panel of head and neck cancer and premalignant cell lines (Figure 4), supporting dual targeting of MET and AKT/mTOR signaling (a convergent signaling pathway/driver in oral tumorigenesis) to increase activity and limit potential drug resistance (47). Metformin is a very safe repurposed agent shown to target this pathway and prevent oral cancer in the 4NQO model (48,49).
Crizotinib was initially designed and developed as a potent MET inhibitor. It was then found to inhibit ALK and is currently approved in patients with non–small cell lung cancer who are positive for the ALK fusion gene (50). ARQ197 was initially reported to be a MET-specific inhibitor (51). More recently, several groups questioned ARQ197 specificity and suggested that its cytotoxic activity may actually be related to inhibition of tubulin polymerization (52–56). Our results in head and neck cell lines show that the cytotoxic effect of ARQ197 may not be related to MET inhibition and may involve alternate mechanisms. A recent study reported that ARQ197 antitumor effects were independent of MET inhibition in OSCC cell lines but related to downregulation of FAK (57).
MET transcript was included in a set of 36 transcripts, all associated with an increased risk of OC development in patients with OLs and overexpressed in OC vs normal mucosa (Figure 1), including secreted phosphoprotein 1 (SPP1), TP63, serine/threonine kinase 3 (STK3), and transferrin receptor (TFRC). They may also be of interest as biomarkers of risk and/or drug targets. SPP1, a recognized target of MET pathway activation and member of the small integrin binding ligands N-linked glycoprotein (SIBLINGs) family, and certain members of this family were found to be associated with an increased OC risk (58); STK3 is targetable, and TFRC has been recently reported to be overexpressed in OC compared with dysplasia and to have some activity on tumor growth in a murine xenograft model of OC, suggesting its potential for chemoprevention (59). Although we identified 23 transcripts from the clinical OL studies, we focused the preclinical studies in this report on the MET finding because of its translational potential to repurpose crizotinib, a US Food and Drug Administration–approved drug, for clinical trial in high-risk MET overexpressing OL.
Our study has several limitations. The association between Met expression and time to oral OC development was found in a single cohort of patients included in a randomized clinical trial evaluating retinoids for chemoprevention with no placebo arm. Further validation in independent cohorts of patients, especially in patients not receiving any experimental therapy, is required. Also, while the 4-NQO model of oral tumorigenesis is popular in the head and neck chemoprevention field, it may not reflect the biology of all oral premalignant lesions (20). A better understanding of the molecular heterogeneity of oral premalignancies will be required to tailor chemoprevention strategies.
In summary, this work identifies c-met as a potential prognostic marker of cancer risk in patients with OL and a possible and predictive marker/driver to prevent malignant transformation in this high-risk population.
Funding
This work was supported by the University of Texas SPORE in Head and Neck P50 CA97007 (PS, SML), the University of Texas SPORE in Head and Neck Career Development Award 2011-2012 (5 P50 CA097007 09 to PS), the Fondation pour la Recherche Médicale (JPF), the Cancéropôle Lyon Auvergne Rhône-Alpes (CLARA) 2014–2016 Programme structurant (No. CVPPRCAN000153-International Head and Neck Prevention Act-IHNPACT to PS), and the LYric Grant INCa-DGOS-4664. SML is supported in part by a National Institutes of Health–National Cancer Institute grant (P30CA023100-30).
Notes
The funders had no role in design of the study; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.
The authors have no conflicts of interest to declare.
