The Proteomic Landscape of Growth Factor Signaling Networks Associated with FAT1 Mutations in Head and Neck Cancers

Head and neck squamous cell carcinoma (HNSCC) is a devastating and often fatal disease, which significantly disturbs patients' vital upper aero-digestive function and is the sixth most common cancer worldwide (1). Despite advances in multimodality treatments, including recent advances in immune-therapy, the overall 5-year survival rate for patients with curable HNSCC is only 61% (http://seer.cancer.gov/). Tumor human papillomavirus (HPV) status remains the strongest predictor of survival, with HPV-unrelated disease carrying a 5-year overall survival (OS) rate of less than 50% (2). Significant gaps in understanding the biology of disease continue to be the main reason behind the paucity of effective therapeutic interventions.

The challenges in effectively treating HNSCC are attributed to its extreme heterogeneity as far as anatomic locations and genetic aberrations (3–6). These genetic alterations, especially gene mutations, accumulate during the growth of HNSCC, complicating our understanding of disease biology and limiting the effectiveness of targeted approaches and ultimately the prognosis of patients. The Cancer Genome Atlas (TCGA) has provided the most comprehensive characterization to date of the genomic and proteomic landscape (5, 6). In TCGA dataset, the mutation rate of FAT1 is approximately 23% in HNSCC, ranking FAT1 as the second most mutated gene after TP53 and supporting its important role in HNSCC biology (5). In HPV-negative (HPV⁻) HNSCC, the FAT1 mutation rate is as high as 28% with many truncations (7), suggesting that wild-type (WT) FAT1 serves as a tumor-suppressor gene in this disease (4). The FAT1 mutation rate in HNSCC is the highest among major solid tumors, making its investigation of primary interest in this disease (www.cbioportal.org). The effect of FAT1 mutation on HNSCC malignant phenotypes has not been extensively investigated, and little is known about its clinical implications, highlighting the significance of our findings.

FAT1 is a cadherin-like protein family member, and a large type 1 transmembrane protein of 4,588 amino acid residues. It has 34 cadherin repeats, a laminin G domain, and five EGF-like repeats in the extracellular region, followed by a transmembrane region and a C-terminal cytoplasmic domain containing a PDZ-binding motif (8). Early studies identified FAT1 as an ortholog of the Drosophila fat gene family, and loss of fat leads to cell-cycle dysregulation and hyperproliferation in larval imaginal discs (9). Recently, FAT1 mutation was identified in human cancers and indicated to contribute to Wnt (Wingless and Int-1) activation, consistent with observations that FAT1 serves as a tumor suppressor in human cancer cells (10). In addition, FAT1 mutant was found to inactivate the Hippo regulatory complex, leading to activation of YAP1 (Yes-associate protein 1) signaling in HNSCC (7, 11). However, comprehensive understanding of the mechanisms underlying FAT1 mutation and its associated growth factor–mediated activation of signaling pathways in cancer progression, particularly in HNSCC, remain limited.

We herein report our comprehensive proteomic analysis identifying altered protein expression and activation status associated with FAT1 mutation in HNSCC using the Reverse Phase Protein Arrays (RPPA) dataset from TCGA. This platform represents a powerful approach for analysis of globally functional proteomics, including more than 200 validated antibodies detecting proteins or phosphorylation sites, indicating activation status (Cancer Proteome Atlas, TCPA, http://tcpaportal.org). The selection of this panel of antibodies was based on antigens representing molecules with critical functions in cancer biology, such as growth factors, signaling molecules, transcription factors, cell cycle, and apoptosis modulators. Using TCGA genomic, expression, and RPPA datasets, we compared differentially expressed proteins and phosphorylated sites in HNSCC tissues associated with FAT1 mutation status using systematic biostatistics and bioinformatics analyses. We have identified a panel of differentially expressed proteins, stratified by FAT1 mutation status, involved in the activation of growth factors and signaling pathways with potential effect on cancer cell proliferation, cell-cycle regulation, apoptosis, and immunomodulation. Furthermore, FAT1 mutation and associated protein expression were associated with a statistically significant difference in patients' OS and progression-free survival (PFS), indicating a potential prognostic value for these biomarkers. Our novel findings from proteomic analysis associated with FAT1 mutation, therefore, provide important information as to potential biomarkers that are linked to FAT1 and that deserve future validation in HNSCC.

A flow chart of data extraction and analyses is presented in Supplementary Fig. S1A–S1C. TCGA specimens were obtained with written informed consent under an Institutional Review Board (IRB) approved protocol in accordance with recognized ethical guidelines (6, 12, 13). FAT1 mutation, gene expression, protein expression, and associated clinical features of TCGA HNSCC cohort were downloaded from the Broad Institute Firebrowse website (http://firebrowse.org/). These datasets include level 3 RNA-seq data (presented as log₂-transformed RNA-seq by expectation maximization; RSEM), normalized protein levels of RPPA (12), and curated MAF files. Patient information is summarized in Supplementary Table S1. In brief, these databases contain 489 cases of HNSCC, with 348 WT and 141 cases showing FAT1 alterations, including mutations and other genetic variations (28.8%). There are 329 patient samples with RPPA data, including 96 with altered FAT1 (29.2%) and 233 with WT FAT1 status.

We examined the correlation of protein expression with patients' OS and PFS. OS was calculated as the time from date of diagnosis to death or last patient contact. PFS was calculated as the time from date of diagnosis to date of disease progression, death, or last contact, whichever was earliest. Biomarkers were further dichotomized by the optimal cutoff point [high (≥cutoff point) versus low (<cutoff point)], which corresponded to the most significant relationship with OS or DFS based on log-rank statistic (14).

In the univariate survival analysis, Kaplan–Meier survival estimates were calculated for each group of patients dichotomized by each biomarker's optimal cutoff point and compared between the two groups with the log-rank test among all patients, respectively (15). For the most significant biomarkers (such as HER3, CYCLINE, IRS, RET-pY905, cMYC, P16INK4A, etc.), we further investigated whether FAT1 can modify their effects on survival by subgroup analyses.

Ingenuity Pathway Analysis (IPA) is an all-in-one, annotated, web-based software application that enables analysis, integration, and understanding of data from gene expression and proteomics (analysis.ingenuity.com). IPA allows searches for targeted information on genes, proteins, chemicals, and drugs, which could build interactive models of experimental systems. The software is backed by the Ingenuity Knowledge Base of highly structured, detail-rich biological and chemical findings. Detailed analysis is described in Supplementary Materials and Methods.

Unsupervised clustering analysis was performed using 20 signature protein molecules after extraction and normalization from the HPV(−) patient group, by comparing RPPA protein levels between patient groups with FAT1 WT versus mutant (P < 0.1). The 329 total patient samples with RPPA data were subjected to hierarchical clustering by Manhattan distance in the column, Euclidean distance in row, and the Ward.D linkage method was performed using the Pheatmap package (version 1.0.12) of R software. On the basis of cluster formation in both row and column directions, samples with associated signature protein molecules were pooled and assigned subcluster numbers. For each subcluster, protein scores were calculated using overall samples' protein levels measured by RPPA, compared using one-way ANOVA and reported as P values. Box and whisker plots of median distributions of protein levels for each subcluster were made using GraphPad Prism (Version 6.1). PFS in various patient subgroups were tested for differences using Kaplan–Meier analysis. Survival curves for each patient subgroup were created using GraphPad Prism (Version 6.1).

Supplementary Materials and Methods include additional and detailed information for clinical characteristics of patients with HNSCC from TCGA and International Cancer Genome Consortium (ICGC) datasets, statistical analysis of FAT1 mutation-associated protein expression, OS analysis of patient subgroups classified by heatmap and cluster analyses, FAT1 gene knockout (KO), Western blot analysis, and ELISA.

The majority of genetic alterations of the FAT1 gene in cancers are mutations, whereas only a low percentage of copy-number variations, predominantly deep deletions, have been observed (Supplementary Fig. S2A). HNSCC (including both HPV status) has the highest FAT1 mutation rate compared with other major cancer types, revealed by the PanTCGA project. We analyzed FAT1 mutation status in HNSCC using TCGA genomic and proteomic datasets and investigated associations with patients' clinical characteristics (Table 1). TCGA datasets include 489 total cases with complete genomic information, 329 of which have protein data from RPPA. FAT1 mutation status is defined by TCGA, including point mutation, deletion, indel insertion, translocation, and fusion. FAT1 mutation rates and clinical characteristics in the genomic and proteomic datasets showed similar frequencies and distributions (Table 1; Supplementary Table S1). We observed that the FAT1 mutation rate is significantly higher in females than in males in both genomic (38.76% vs. 25.28%, P = 0.004) and proteomic (39.36% vs. 25.11%, P = 0.01) datasets. Patients with FAT1 mutations tend to be older with a mean age of 65 years compared with 59 years among patients with FAT1 WT (P < 0.001 in both datasets). Interestingly, among patients with FAT1 mutations, case numbers were statistically higher in the oral cavity (OC) and larynx and hypopharynx (LH) than in oropharynx (OP), and the FAT1 mutation rate was higher in HPV(−) than HPV(+) patients in both datasets. There was no statistically significant association of FAT1 mutation with race, alcohol usage, smoking status, disease stage, or lymph node status (Table 1).

FAT1 gene mutation status (left) and mRNA expression (right) were compared after stratification by HPV status (Fig. 1A), primary tumor location (Fig. 1B), disease stage (Fig. 1C), and lymph node status (Fig. 1D). More patients with FAT1 mutations and higher FAT1 gene expression levels were observed among HPV(−) than HPV(+) patients. FAT1 gene expression levels were significantly lower in patients with FAT1 mutations than WT for both HPV statuses (Fig. 1A). Comparing tumors from different primary locations, the percentage of patients with FAT1 alterations was higher among patients with OC and LH (mainly HPV-negative) than in OP (mainly HPV-positive) tumors. FAT1 gene expression levels were significantly lower in OC and LH tumors with mutation than WT, but were similar between FAT1 WT and mutant OP tumors. WT FAT1 gene expression in OP tumors was significantly lower than in OC and LH (Fig. 1B). The percentages of patients with FAT1 alterations were similar across different disease stages (including both HPV status), but there were higher case numbers of stage IV tumors. For each disease stage, FAT1 gene expression was significantly lower among patients with mutant FAT1 than WT. However, no statistical significance was observed when comparing among different stages within both FAT1 genetic status (Fig. 1C). The percentages of patients with FAT1 mutation were similar in tumors with different lymph nodes statuses (including both HPV status). FAT1 gene expression was lower in patients with mutant status, except for N3 stage, which could be due to the lower number of cases. There was no difference in FAT1 gene expression among patients with different lymph node stages within each FAT1 genetic status (Fig. 1D).

We analyzed an independent dataset of oral cancer from the ICGC project, which mainly focused on genetic mutations in 178 donors from India (https://dcc.icgc.org; Supplementary Fig. S3A; refs. 16, 17). In this dataset, the frequency of FAT1 mutation was further compared with TCGA at 37%, ranking #3 among these patients. Interestingly, the top two mutated genes were TP53 and CASP8, and the CASP8 mutation rate was unusually high at 42.1% (Supplementary Fig. S3B). In addition, we studied the association of FAT1 genetic alterations with other top mutated genes in HNSCC from TCGA dataset (Supplementary Fig. S3C and S3D). Although CASP8 mutation was only seen in 13% of cases, this alteration was highly enriched in the group with FAT1 genetic alterations. FAT1 genetic alterations co-occurred with CASP8 or TP53, and this association was statistically significant (Supplementary Fig. S3D–S3F).

To understand whether FAT1 mutation is associated with the expression of other proteins, we compared two patient groups with different FAT1 mutation status for their protein expression assayed by RPPA. We analyzed a total of 329 patient samples, including 287 HPV(−) and 35 HPV(+) tumors, which have complete clinical information (Table 1; Supplementary Table S1). Supplementary Table S2.1 presents protein analysis data for total patients with both HPV status, including 233 patients with WT FAT1 and 96 patients with mutated FAT1, using Student's T-test under a relaxed significance level of 0.1 as a screening threshold. In Table 2, we present the protein names used in RPPA with official gene names and gene symbols, chromosome locations, and a brief description of their functions and potential interacting molecules. We ranked these proteins and corresponding gene expression from TCGA RNA-seq data according to statistically significant differences between different FAT1 mutation status. We also included a few molecules with near statistically significant P values ≤0.05 (Table 2; Supplementary Tables S2 and S3).

We observed that the molecules most significantly associated with FAT1 mutation are related to growth factor receptors and their downstream signaling modulators under the RPPA protein selection criteria. In patient samples with FAT1 mutation, we observed increased HER3_pY1289 (P = 0.004), insulin receptor substrate 1 (IRS1, P = 0.017), and cellular membrane protein CAVEOLIN 1 (P = 0.017). VEGFR 2 exhibited a P value barely over the cutoff (VEGFR2, P = 0.054). PDL1 was included due to its established clinical importance as a therapeutic target and CMYC given its significant genetic amplification (5). Among these proteins, HER3_pY1289 and IRS1 were significantly or near significantly upregulated in HPV(−) HNSCC (P = 0.011 and 0.06, respectively), but not in HPV(+) HNSCC (P > 0.1). Upregulation of HER3_pY1289 and VEGFR2 was only observed at the protein level, whereas upregulation of IRS1 and CAVEOLIN 1 was also observed at the mRNA level. We identified 12 proteins that were significantly increased in tumors with WT FAT1 status (Table 2). The immune checkpoint molecule PDCD1 (PD1) was increased in total patient samples with WT FAT1 (P = 0.054), and in HPV(−) cases (P = 0.051).

Because RPPA data did not include the protein expression of immune signatures for CD3⁺, CD4⁺, CD8⁺ T cells, and Treg cells, we analyzed their immune gene signatures and related inflammatory cellular and cytokine markers using TCGA RNA-seq data. We did not find any significant difference in most immune gene expression signatures associated with FAT1 mutation in either total patients' samples or HPV(−) samples (Supplementary Table S4.1 and S4.2). However, we observed decreased gene expression of immune cytokine IL2 and growth factors FGF1 and FGF2, and increased NRG1 (neuregulin, protein ligand for ERBB2/3) and TGFβ1 in total patient samples with FAT1 mutation (Supplementary Table S4.1 and Supplementary Fig. S4). In HPV(−) patient samples with FAT1 mutation, NRG1 exhibited increased expression with statistical significance, whereas the CD274 (PDL1) expression was marginally increased (P = 0.06), consistent with its protein expression (Supplementary Table S4.2).

To further investigate the clinical significance of FAT1 mutation and associated proteins, we analyzed OS and PFS of total and HPV(−) patients using both univariate and multivariable survival analyses, presented as median survival months. From the HNSCC TCGA dataset, we observed a trend toward decreased PFS of patients with FAT1 mutation alone, which approached statistical significance (Supplementary Fig. S5, P = 0.06). To explore the covariates that could influence the effect of mutant FAT1 status on survival, we performed survival analysis, presented as Kaplan–Meier plots, after stratification by protein expression and FAT1 mutation status. This revealed that mutant FAT1 together with HER3_pY1289, IRS1, and CYCLINE2 was significantly associated with worse OS (Fig. 2A). Patients with lower expression of HER3_pY1289 exhibited better OS, whereas FAT1 mutation status alone showed a lesser impact. In patients with WT FAT1, IRS1 expression strongly affected OS. Patients with lower IRS1 expression exhibited the best survival, whereas patients with mutant FAT1 status had intermediate OS. In contrast, patients with mutant FAT1 status and lower expression of CYCLINE2 exhibited better OS than patients with lower CYCLINE2 expression and WT FAT1 status (Fig. 2A). In HPV(−) patients, higher IRS and lower RET_pY905 were dominantly and significantly associated with worse OS (Fig. 2B). Higher P16INK4A and lower CMYC were significantly associated with better PFS in patients with WT FAT1 (Fig. 2C). Pairwise comparison of the survivals for Fig. 2 is presented in Supplementary Table S5.1. Altered expression of several molecules did not significantly affect survival under the analysis with stratified FAT1, but exhibited statistical significance in a pairwise analysis in each group (Supplementary Fig. S6A–S6D, Supplementary Table S5.2). Univariate and multivariate statistical analyses of OS using the COX model are presented in Supplementary Table S6, PFS in Supplementary Table S7 (18).

We then tested these molecules showing significant association with FAT1 mutation using univariate analysis of Kaplan–Meier plots, but without stratification by FAT1 mutation status. We observed that higher expression levels of HER3_pY1289, CYCLINE2, and IRS1 were significantly associated with worse OS among all patients (Supplementary Fig. S7A). HPV(−) patients exhibited worse OS with lower expression of RET_pY905 and mTOR_pS2448, and higher IRS (Supplementary Fig. S7B). Among the total patients, lower expression of P16INK4A and higher expression of CAVEOLIN1, IRS1, and CMYC were significantly associated with worse PFS (Supplementary Fig. S7C). Among HPV(−) patients, only lower expression of mTOR_pS2448 was significantly associated with worse PFS (Supplementary Fig. S7D).

To understand potential signaling pathways and networks mediated by FAT1 alterations in HNSCC, we performed IPA using 31 molecules identified and combined from total and HPV(−) patient samples. As expected, the top pathways identified by IPA included Molecular Mechanisms of Cancer, ERBB Signaling, Neuregulin Signaling, and EGF Signaling, ranked by high statistical significance and positive z-scores (Supplementary Fig. S8A). The molecules involved in these pathways are presented in Supplementary Table S8. Upstream analysis predicted TP53 and MYC as essential transcriptional regulators, and ERBB2 as the potentially critical receptor that regulates downstream pathways (Fig. 3A; Supplementary Fig. S8A). Supporting these predictions, we observed the co-occurrence of TP53 and FAT1 genetic alterations in the HNSCC TCGA dataset (Supplementary Fig. S3F). Altered CMYC protein expression and ERBB2 phosphorylation were associated with FAT1 mutation (Table 2; Supplementary Table S2), of which, increased CMYC protein expression also contributed to worse patient survival (Supplementary Fig. S7C, Fig. 2C). Growth factors and cytokines, such as TGFB1, IL6, and FGF2, were identified as potential upstream regulators (Fig. 3B). These predictions are supported by RNA-seq data in TCGA analysis (Supplementary Fig. S4). We analyzed the downstream cellular functional categories regulated by the proteins identified, and observed a decrease in apoptosis activity and increase in cell-cycle progression and proliferation activities (Fig. 3C). Furthermore, we uncovered potential therapeutic drugs that could target these molecules, including sirolimus/rapamycin (mTOR inhibitor; ref. 19), curcumin (anti-inflammatory dietary supplement; ref. 20), and tanespimycin (17-AAG, HSP90 inhibitor; Supplementary Figs. S8B and S9; ref. 21).

To validate whether there is a functional relationship between FAT1 inactivation and observed major signaling pathways, we knocked out the FAT1 gene in two HNSCC cell lines, MDA686TU and UMSCC-1P, using CRISPR-Cas9 technology, and established both parental (WT) and FAT1-KO HNSCC cell line pairs (Fig. 3D). Western blot analysis illustrated decreased pEGFR, pHER2, and pERK proteins, indicating inactivation of the EGFR signaling axis. pHER3 and HER3 were consistently elevated but to a lower degree after FAT1 protein KO. These experimental results support the causal relationship of FAT1 mutation and altered signaling pathways identified in this study.

As several upstream signaling cytokines and growth factors were predicted by IPA (Fig. 3B) and confirmed by gene expression analysis (Supplementary Table S4, Supplementary Fig. S4), we further analyzed the protein expression of IL6 and TGFβ in supernatants from FAT1 WT and KO MDA686TU and UMSCC-1P cells by ELISA (Supplementary Fig. S10). We confirmed elevations of IL6 and TGFβ proteins in FAT1 KO HNSCC cell lines, inferred above by IPA and gene expression data.

To understand the tumor heterogeneity associated with FAT1 mutation and related protein expression or phosphorylation, we performed hierarchical cluster analysis using a panel of 20 differentially expressed proteins listed in Supplementary Table S2.2. We observed two subgroups of patients with HNSCC within each group of patients with FAT1 mutation (Fig. 4A) or WT status (Fig. 4B). The two observed patient subgroups were represented by hierarchical clusters horizontally on the top of the heatmaps, without identified significant association with HPV status or primary tumor locations. Two major protein expression patterns were revealed by the hierarchical links on the left side within each heatmap. In patients with FAT1 mutation (A), the protein group presented at the top of the cluster included all phosphorylated proteins, such as TUBERING_pT1462, GSK_pS9, AKT_pS473, SHC_pY317, EGFR_pY1068, HER2_pY1248, RET_PY905, CJUN_pS73, whereas the protein cluster at the bottom included most total proteins with only three phosphorylated proteins, HER3_pY1289, SRC_pY527, and MTOR_pS2448. Similar protein clusters were observed in patients with FAT1 WT (B), with the only exception of two phosphorylated proteins, MTOR_pS2448 and SRC_pY527. Next, we divided the heatmaps into four quadrants (labeled I–IV at the tops and bottoms), and calculated the total expression scores of each protein clustering group (Fig. 4C and D). The expression scores were statistically different among the four clusters from both heatmaps (P < 0.001). When we compared the scores between protein clusters, all pairwise comparisons exhibited statistical significance, with one exception of cluster III versus IV from the FAT1 mutant group (Fig. 4C). Furthermore, we did not observe an overall benefit in PFS when comparing patient group I versus II separated by the protein clusters (Fig. 4E and F). However, when particular PFS time points were analyzed, we observed a statistically significant improvement in survival for patient subgroup I regardless of FAT1 status, that subgroup I patients overexpressed EGFR_pY1068, HER2_pY1248, and other phosphorylated proteins. In patients with FAT1 mutant status, the only survival benefit was observed at day 1,200 (3.3 years, Fig. 4E), whereas for patients with FAT1 WT status, the survival benefit was observed relatively later at days 2,500 (6.8 years), 3,000 (8.2 years), and 4,000 (11 years; Fig. 4F; Supplementary Table S9).

Our previous publication revealed that squamous cell cancers (SCC) share many common genetic alterations (6). We explored the broader relevance of FAT1 alterations in HNSCC and other squamous cancer types, such as lung and cervical SCCs, and identified a strong association between FAT1 and EGFR gene expressions (Fig. 5A,–C). This association ranked #1 in lung SCC and #4 in cervical cancer but #24 in HNSCC. The association between gene expression of FAT1 and these genes in lung and cervical SCC is consistent with the data in HNSCC, where FAT1 gene mutations and deletions are associated with decreased FAT1 gene expression (Fig. 5D,–F). In addition, we observed that FAT1 genetic alterations in HNSCC and lung SCC are associated with a high mutation count, whereas in lung SCC, FAT1 genetic alterations are also associated with aneuploidy score (Fig. 5G–I). Furthermore, we displayed FAT1 and EGFR gene expressions by individual HNSCC, lung SCC, and cervical cancer patient samples by Oncoprint (Supplementary Fig. S11A–S11C) and observed that EGFR gene expression was also associated with copy-number alterations (Supplementary Fig. S11D–S11F).

In this study, we characterized the global proteomic landscape associated with FAT1 mutation in more than 300 HNSCC tissue samples from TCGA with available RPPA data (12). The FAT1 mutation rate in HNSCC is 23%–28% (5, 7), and is the highest among other major cancer types, including lung and cervical SCC (11). The frequency of FAT1 mutations is further enriched for oral SCC in the independent ICGC dataset. Together, these observations support a critical role and unique contribution of FAT1 to oncogenesis of a subset of head and neck cancers. Following TP53, FAT1 is the second most commonly mutated gene in HNSCC, especially HPV(−) tumors, underlining its pathological significance (5). TP53 and FAT1 mutations co-occur in HNSCC. Given the challenges in targeting p53 in HNSCC, pathway(s) modulated by mutant FAT1 could have impact on disease outcome and serve as additional targets for prevention or therapy for a significant proportion of patients with TP53 alterations (22). The unique alteration pattern of FAT1 implicates its functional importance in SCC, but with less understood mechanisms than the well-studied TP53 and CDKN2A (p16; ref. 6). Furthermore, FAT1 mutations include a high percentage of truncation and nonsense mutation types, which could eliminate its expression and abolish the entire function of the FAT1 molecule. Accordingly, tumors with FAT1 mutation exhibited significantly lower FAT1 gene expression, which supports its functional role as a tumor suppressor.

Our most significant and novel finding is the identification of a broader proteomic landscape of predominant oncogenes and tumor suppressors and potentially druggable targets associated with mutant or WT FAT1. We observed increased expression of a network of proteins that includes cell surface receptors, such as HER3_pY1289, VEGFR2, and PDL1, plus IGFR signaling mediator IRS1, and cell-cycle modulator CMYC, in more than 90 HNSCC patient samples with FAT1 mutation (Supplementary Figs. S12 and S13). Although the highly mutated FAT1 molecule itself has not been identified as a therapeutic target in HNSCC, due to its nature as a tumor suppressor, these surface receptors and signaling molecules associated with FAT1 mutation are well-investigated therapeutic targets with available targeted agents and possible links to immune-related pathways. Specifically, our data showed that in both total and HPV(−) patients, HER3_pY1289 was upregulated in FAT1-mutated HNSCC (Table 2; Supplementary Table S2). HER3 is one of the EGFR family members and HER3_pY1289 is the activated form that transduces signaling after partnering with other EGFR family members (23, 24). We found significant correlation between FAT1 and EGFR expression in lung, cervical, and head and neck SCC, among which, this correlation was ranked #1 in lung SCC. We also observed that FAT1 KO could reduce EGFR expression and signaling in HNSCC cell lines, indicating the potential impact of FAT1 in EGFR signaling. In addition, IRS1 (25), a key regulator of IGF-1R, is also upregulated in FAT1-mutated HNSCC. Cross-talk between HER3 and IGF-R1 (26) signals could synergistically activate ERK/MAPK, PI3K/AKT, and RAS/RAF pathways and promote cell proliferation/survival, protein synthesis, and cell cycle through CMYC (Supplementary Figs. S12 and S13; ref. 27). Among other associated proteins, CAVEOLIN (28, 29) can alter cell shapes, and together with VEGFR2 (30), enhance cell angiogenesis, stemness, migration, and invasion potential (Supplementary Fig. S13). These findings are consistent with a recent publication suggesting that FAT1 contributes to epithelial-to-mesenchymal transition, tumor cell stemness, and metastasis in both skin and lung cancer models (31). Furthermore, expression of a different checkpoint protein, PD1, was higher in the tumor microenvironment with WT FAT1, which could induce immunotolerance and suppress antitumor immunity (32).

IPA analysis revealed dominant upstream regulators, including TP53, CMYC, and ERBB2, as potentially activated transcription factors and cell surface molecules, plus upregulated growth factors and proinflammatory cytokines, such as TGFβ1 (33), IL6 (34, 35), and FGF2 (36). These predictions were confirmed by gene expression data from TCGA RNA-seq dataset (Supplementary Fig. S4 and Supplementary Table S4), and/or the protein data tested by ELISA in HNSCC FAT1 KO cell lines (Supplementary Fig. S10). Increased TGF1β expression in the FAT1-altered group was predicted by IPA, and confirmed by gene expression analysis from TCGA datasets and protein expression in cell lines by ELISA. Although increased IL6 was not observed in the current gene expression analysis, moderate upregulation of IL6 was observed in the supernatant of FAT1 KO cells by ELISA. Though we did not observe FAT1 alteration affecting immune CD3⁺, CD4⁺, CD8⁺ T cells, and Treg cells, we observed altered growth factors and pro-inflammatory cytokines, which could modulate the immune response in the tumor microenvironment. These predicted signaling pathways and regulatory networks and in vitro experimental results warrant further investigation regarding their mechanistic association with FAT1 mutation and clinical implication.

HNSCC is remarkably heterogeneous, partially due to the complex anatomical structures and different primary tumor locations. Two significant etiologies of tobacco usage and HPV infection represent two major types of diseases, which activate distinct biological pathways in carcinogenesis. We found that the FAT1 mutation rate is significantly higher in HPV(−) than HPV(+) patients in this cohort, consistent with a previous publication (7). In that study, the FAT1 mutation rate was reported as up to 28% in HPV(−) versus 2.8% in HPV(+) patients, suggesting that FAT1 mutation may play more critical roles in HPV(−) than HPV(+) HNSCC. It is not surprising that the FAT1 mutation rate is higher in OC and LH than in OP, because most OP tumors are HPV(+). Significantly, patients with FAT1 mutation were older and female, which could be compounded by HPV(−) status. Also, we identified a group of proteins differentially expressed in FAT1 mutated HPV(+) cases, which were distinct from those proteins identified in HPV(−) cases (Supplementary Table S2.3). However, we cannot conclude the possible role of FAT1 mutation in HPV(+) HNSCC due to the small number of cases, and this should be validated in a larger cohort.

The heterogeneity of HNSCC is also influenced by the large variety of genomic, expression, and proteomic alterations that drive carcinogenesis. This study observed two major subgroups of samples exhibiting different protein signatures within each group classified by different FAT1 mutation status. Hierarchical clustering analysis of protein expression revealed two distinct clusters of protein molecules that were mutually exclusive in HNSCC subgroups with either FAT1 mutant or WT status (Fig. 4). We observed mutually exclusive protein expression and activation, such as phosphorylation of EGFR/HER2 in one subgroup of HNSCC, versus activated HER3 and IGFR signaling IRS protein in the different patient subgroups regardless of FAT1 mutation status. The underlying mechanisms and clinical impacts of these protein clusters shown by hierarchical heatmaps remain unclear. These observations are consistent with our and other previous studies showing that HER3 activation is one of the primary mechanisms that bypasses EGFR blockade (23, 37). Literature suggests that HER3 might interact with IGF-1R to activate common downstream signaling (38), contributing to cetuximab resistance and impacting clinical outcomes (39).

One of the clinically significant observations from this study comes from the survival analysis of the protein biomarkers and their relationship with the significant impact of FAT1 mutations through statistical analysis (Fig. 2; Supplementary Fig. S6). FAT1 mutation status exhibited a stronger impact on PFS than OS. Our observations of the impact of mutant FAT1 on PFS are supported by a previous study on patients with HNSCC in Taiwan, which reported a 29% mutation rate of FAT1 and significant correlations with lymph node status and worse disease-free survival, but not OS (40). Although we did not find that FAT1 mutations alone were associated with either PFS or OS in univariate analysis of this cohort, statistical significance could be diminished by the long survival time of follow-up for >200 months (>16 years) in this cohort, where other causes affect survival. There was a survival benefit in PFS among all patients with FAT1 WT status that approached significance (P = 0.06; Supplementary Fig. S2B), which could reach statistical significance if shorter-term survival of 5 years was analyzed. It is worth noting that the PFS analysis at earlier time points revealed statistical significance and that patients with activated HER3_pY1289 and other protein overexpression exhibited worse PFS. Furthermore, an earlier effect on PFS was seen among patients with mutated FAT1 status than those with FAT1 WT (Fig. 4E and F). These observations of survival benefits may help classify patients with HNSCC with differential treatment plans and prognostic outcomes, which warrant future clinical investigations. Better stratification of patients with HNSCC based on these alterations could be instrumental in addressing the therapeutic dilemma that has surrounded patients with HPV-unrelated disease, who exhibit TP53 alteration and resistance to EGFR antagonists. Exploring immune-related correlates to these pathways may unlock critical information instrumental in the design of future combinatorial approaches in patients with HPV-unrelated disease.

The content is solely the authors' responsibility and does not necessarily represent the official views of the National Institutes of Health.

N.F. Saba reports advisory role to Merck, Pfizer, GlaxoSmithKline, Blueprint, Kura, CUE, Vaccinex, and Biomtech. No disclosures were reported by the other authors.

Zhengjia Chen: Conceptualization, software, formal analysis, supervision, validation, investigation, methodology, writing–original draft, project administration. C. Zhang: Data curation, software, formal analysis, validation, investigation, visualization, methodology. J. Chen: Resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft. D. Wang: Investigation. J. Tu: Data curation. C. Van Waes: Writing–review and editing. N.F. Saba: Conceptualization, investigation, analysis of data, writing–review and editing. Z.G. Chen: Conceptualization, supervision, investigation, project administration, writing–review and editing. Zhong Chen: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This study was supported by grants from a pilot Winship Invests grant from Emory University Winship Cancer Institute (to C. Van Waes, Z.G. Chen, and N.F. Saba) and NIDCD intramural projects Z01-DC-00016, 73, 74 (to Zhong Chen). The research reported in this publication was partially supported by the University of Illinois Cancer Center Biostatistics Shared Resource Core (BSRC). The authors thank Li Jia, Bioinformatician from National Institutes of Health Library for analysis of proteomic datasets of head and neck cancers. The authors thank Jianwei Chen who participated in the study of TCGA datasets of head and neck cancer and analyzing the association of the FAT1 mutation with pathogenesis of head and neck cancer. They also thank Drs. Anthea Hammond, Gregory B. Lesinski, and Shi-Yong Sun (Emory University) for their suggestions and critical reading of the article.

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