SMAD4 Loss Is Associated with Cetuximab Resistance and Induction of MAPK/JNK Activation in Head and Neck Cancer Cells

Head and neck squamous cell carcinoma (HNSCC) is the sixth most common cancer in the world, with an annual incidence of over 50,000 cases in the United States alone (1). HNSCC is a heterogeneous disease that includes squamous cell carcinomas (SCC) of the oral cavity, pharynx, and larynx. The disease is more prevalent in males, and those individuals who smoke or chew tobacco and/or consume alcohol are at much higher risk for HNSCC (2). Human papillomavirus (HPV) is also associated with oropharyngeal primary sites with significantly better prognosis than patients with HPV-negative tumors (3, 4). In particular, HPV-negative HNSCC is notorious for poor prognosis, which reflects the propensity of the tumor to be treatment resistant. Although patients with early-stage disease (stage I or II) have better prognosis, these patients are still at risk for developing locoregional recurrences, distant metastases, and second primary tumors. Despite improvements in molecular diagnosis, cetuximab is the only FDA-approved targeted therapy available for this disease (5). Cetuximab, a mAb targeting EGFR, has been the standard of care in the treatment of HNSCC for many years. However, the response rate to cetuximab monotherapy is low (10%–13%), and the majority of patients develop resistance even after an initial response (6, 7). Thus, HNSCC continues to represent a challenging disease to treat.

SMAD4 is a well-known tumor suppressor and the central signal transduction component of TGFβ, a multifunctional cytokine that regulates cell growth and differentiation and is frequently upregulated in many human cancers including HNSCC (8–11). Although TGFβ/SMAD signaling exerts a suppressive effect on normal epithelial cells, tumor cells frequently become refractory to the growth-inhibitory effect of TGFβ and acquire an ability to increase expression and secretion of TGFβ (11–13). Previous studies have demonstrated that in tumor cells, TGFβ-activated SMAD4–SMAD2/3 complex stimulates the expression of SNAI1 and TWIST1, which cooperate with SMAD proteins to repress the expression of epithelial genes such as CDH1 (encoding E-cadherin; refs. 14, 15). This switch enables tumor cells to leverage the tumor-promoting effects of TGFβ in the tumor microenvironment to facilitate tumor progression, invasion, and metastasis (10–15). The nature of the switch that determines whether TGFβ acts as a tumor promoter or as a tumor suppressor has been the subject of intense research (14).

While mutant TP53 has been suggested to be one potential explanation (16), more recent studies have shown that loss of SMAD4 activity may also result in such a switch (17, 18). Supporting this concept, the formation and subsequent progression of various solid malignancies was strongly associated with loss of SMAD4 activity via inactivating mutations or loss of heterozygosity within the 18q21 locus, containing SMAD4 (19–22). For example, loss of SMAD4 promotes pancreatic and colorectal tumor progression and is strongly associated with increased metastatic potential, lower overall survival, and poor chemotherapeutic responses (23–25). Likewise, loss of SMAD4 activity was reported to cause spontaneous HNSCC formation, to induce epithelial-to-mesenchymal transition (EMT) and cetuximab resistance, and to result in genomic instability through the downregulation of DNA repair–related genes (18, 26, 27). Although the link between SMAD4 loss or inactivation and tumorigenesis is strong and compelling, mechanisms underlying the resistance to cetuximab, an FDA-approved HNSCC-targeted therapy, in tumors with SMAD4 loss has not been reported.

Previously, we demonstrated that cetuximab-resistant clones of an HPV-negative HNSCC cell line had reduced SMAD4 gene expression (27). Interestingly, HPV-negative tumors in patients with HNSCC also had lower SMAD4 expression compared with HPV-positive tumors, suggesting that the worse overall survival of HPV-negative patients with HNSCC may be in part due to low SMAD4 expression. In this study, we demonstrate that loss of SMAD4 in a cohort of 130 newly diagnosed and 43 patients with recurrent HNSCC is associated with poor outcome. Our work also aimed to discover mechanisms of cetuximab resistance in SMAD4-low HNSCC that can be targeted with combination therapy. Using OncoFinder, a bioinformatics software suite for qualitative analysis of intracellular signaling pathway activation using transcriptomic data (28), we uncovered commonly upregulated pathways in the TCGA HNSCC dataset and an independent oral cavity squamous cell carcinoma cohort with low SMAD4 expression. We find that SMAD4 expression is associated with cetuximab resistance in HNSCC cell lines and that SMAD4-low tumors are associated with worse overall survival of tumor-bearing mice in vivo. Furthermore, our data indicate that JNK and MAPK activation may be potential mediators of this process, and demonstrates the potential utility of JNK or MAPK inhibition as a novel strategy for overcoming cetuximab resistance in HNSCC tumors with loss of SMAD4 expression.

Raw RNA-Seq or microarray data had been retrieved from publicly available TCGA or NCBI GEO repository databases. RNA-Seq data preprocessing, and normalization steps were performed in R version 3.1.0 using DEseq package from Bioconductor. The resulting matrix contained mRNA expression information for over 20K genes across all analyzed samples. Normalized gene expression data were loaded into Oncofinder pathway activation scoring platform (28). The software enables calculation of the Pathway Activation Score (PAS) for each of the 271 pathways analyzed, a value which serves as a quantitative measure of differential pathway activation between the two states. The signalome knowledge base developed by SABiosciences (http://www.sabiosciences.com/pathwaycentral.php) was used to determine structures of intracellular pathways, which was used for the computational algorithm as described previously (29, 30). Pathways with positive PAS values are upregulated, while negative PAS values correspond to downregulated pathways. The algorithm used to calculate PAS is as follows:

Here, CNR_n is the ratio of the expression level of a gene n in the tumor sample and in the control; BTIFn is a value of beyond tolerance interval flag, which equals 0 or 1; and ARR_n is an activator/repressor role equal to −1, −0.5, 0, 0.5, or 1, defined by the role of protein n in the pathway. More information can be found in previous publications (29–31). Pathway activation strengths were determined using the default parameters of OncoFinder, a sigma filter of 2 and a CNR value less than 0.67 or higher than 1.5. PAS heatmap generation and hierarchical clustering were performed using R package gplots. Statistical tests and correlation analysis were done with the Microsoft Excel software.

Two tissue microarrays (TMAs) containing tumor cores from 133 newly diagnosed patients with HNSCC were obtained under Institutional review board–approved protocols from Stanford University (Stanford, CA). Detailed clinical description of the TMAs has been published (32, 33). Cores from three patients lacked tumor and were excluded from the study. In addition, 43 unstained formalin-fixed paraffin-embedded tumor slides from retrospectively selected patients for having recurrent HNSCC with available tumors were also obtained under Institutional review board–approved protocols at Vanderbilt University (Nashville, TN).

FaDu, SCC25, and SCC61 cells were harvested by trypsinization when 70% confluent, washed twice in PBS, resuspended in 10% neutral buffered formalin (NBF), and pipetted into a 0.6-mL microfuge tube containing 200 μL of solidified agarose (2% in PBS). The microfuge tube was centrifuged at 200 × g (5 minutes at room temperature) to form a uniform layer on the agarose bed. Supernatant was aspirated, and pellet was carefully overlaid with 200 μL 10% NBF. After repeating centrifugation, the microfuge tube was carefully submerged in 10 mL 10% NBF solution. Cell pellet was allowed to fix for 24–48 hours and submitted for processing and paraffin embedding.

Slides were stained at the Johns Hopkins Oncology Tissue Immunostaining Core Facility, a Clinical Laboratory Improvement Amendments (CLIA)-certified lab, with SMAD4 antibody at 1:500 dilution (Santa Cruz Biotechnology, Inc.) in Universal IHC Blocking Diluent (Leica Biosystems) for 15 minutes using an automated IHC system (Leica BOND-MAX), and a p16 antibody (predilute, mtm-CINtech, E6H4) using iView DAB detection and Ventana BenchMark XT staining system (Roche) as described previously (4, 24, 34). Formalin-fixed paraffin-embedded (FFPE) cell pellet blocks generated from FaDu cells with complete loss of SMAD4 and stained under the same conditions as patient samples served as a negative control, and blocks from SCC25 and SCC61 cells served as a positive control. Scoring of the IHC staining was performed by trained head and neck cancer pathologist. Scoring was based upon the percentage of tumor cells stained in an overall assessment of all tumor tissue available. Cases were scored for SMAD4 and p16 as dichotomous variables based on the presence or absence of protein expression in the tumor. p16 was used as a surrogate marker for high-risk HPV and positive was defined as >70% nuclear or nuclear and cytoplasmic expression. SMAD4 loss was defined as <5% tumor cells to minimize confounding effects introduced by varying degrees of SMAD4 downregulation.

Patient characteristics were analyzed by SMAD4 status using Fisher exact test for categorical factors and Wilcoxon test for continuous variables. To evaluate the prognostic effect of SMAD4, the clinical outcome of overall survival, defined as the elapsed time from diagnosis to death due to any cause, were considered. We used the Cox regression model to correlate pertinent markers with overall survival. For multivariate analysis, backward selection technique was used to select the variables to include in the model, while SMAD4 status and known prognostic factors were restricted within the model. The significance level for removing any variable was 0.2. All P values were based on two-sided tests.

FaDu, SCC61, and SCC25 HNSCC cell lines were used for these studies. FaDu and SCC25 cell lines were obtained from ATCC, and SCC61 was received from Dr. Ralph Weichselbaum (University of Chicago, Chicago, IL). Short tandem repeat analysis (Identifiler, Applied Biosystems) authenticated cell lines before use. The cells were periodically monitored for mycoplasma at Johns Hopkins Genetic Resources Core Facility using the MycoDtect kit (Greiner Bio-One). All experiments were performed within 6 months of the mycoplasma screen. Cell lines were cultured in DMEM/F12 medium supplemented with 10% FBS, Penicillin–streptomycin, and 0.4 μg/mL hydrocortisone. Stable cell lines were maintained in culture medium with select antibiotics puromycin (5 μg/mL FaDu, 4 μg/mL SCC61), hygromycin (400 μg/mL FaDu, 100 μg/mL SCC61). Cetuximab (Bristol-Myers Squibb) was purchased from the Johns Hopkins Pharmacy.

The following primary antibodies were from Cell Signaling Technology: phospho-SMAD2^Ser465/467 (138D4), SMAD2/3 (D7G7), phospho-STAT3^Tyr705 (D3A7), STAT3 (79D7), phospho-AKT^Ser473 (D9E), AKT(C67E7), phospho-EGF receptor Tyr1045, Tyr1068 (D7A5), and Tyr1173 (53A5), EGF receptor (D38B1), phospho-p44/42 MAPK^{Thr202/Tyr204}, p44/p42 MAPK (137F5), GAPDH (14C10), phospho-JNK^{Thr183/Tyr185} (81E11), JNK (56G8), BCL-2 (50E3), BCL-XL (54H6), BIM (C34C5), β-actin (13E5). SMAD4 (B-8) was from Santa Cruz Biotechnology.

Protein lysates were prepared in RIPA lysis buffer, and protein concentration was measured by the bicinchoninic acid method (Thermo Scientific). Proteins from each sample were resolved under reducing conditions using NuPAGE gels (Novex; Invitrogen) according to manufacturer's instructions and transferred to PVDF membrane (EMD Millipore). Membranes were blocked in 5% BSA in TBS and incubated with primary antibodies overnight at 4°C followed by incubation with HRP-linked secondary antibodies. Protein bands were visualized by chemiluminescence using the ECL Western blotting Detection System (GE Healthcare) or SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific/Pierce Biotechnology).

SCC61 or SCC25 cells were seeded in 6-well plates 24 hours prior to transduction. As per manufacturer's instructions, an appropriate amount of lentiviral transduction particles of Smad4 (clone number: TRCN0000010321, Sigma) or control transduction particles (#SHC-002V, Sigma) were added into the media in the presence with hexadimethrine bromide (Sigma). After puromycin selection for 2 weeks, surviving cells were pooled and knockdown was confirmed by real-time PCR using primers for Smad4 (Hs00929647-m1) and β-actin (ACTB, Hs00357333-g1), or by immunoblotting.

To establish the SMAD4 expression vector, pLenti-CMV puro lentiviral vector was used as the backbone. pLenti-CMV was a kind gift from Dr. Charles Rudin and Dr. John Poirer (Oncology, Johns Hopkins University, Baltimore, MD). The SMAD4 gene sequence was obtained from the pDPC-WT vector, a kind gift from Dr. Scott Kern (Oncology, Johns Hopkins University, Baltimore, MD), and pLenti-Smad4-CMV was made by Gateway Technology (Life Technologies). All vectors were confirmed to have correct gene sequence. Viral particles were produced by cotransfecting 900 ng of each expression vector, 100 ng of psPAX2 as packaging vector, and 1 μg pMD 2G as envelope vector into HEK-293T cells using 10 μL of Lipofectamine 2000. Virus culture supernatants were obtained 24–48 hours after transfection. FaDuL cells were exposed to virus-containing media for 24 hours, and were selected using puromycin (Life Technologies) 5 μg/mL for at least two weeks. For confirmation of Smad4 expression, total RNA was extracted from transduced cells using Qiagen RNeasy Mini kit (Qiagen) according to the manufacturer's protocol. Smad4 overexpression was confirmed by conventional PCR using forward primer 5′-CCATTTCCAATCATCCTGCT-3′ and reverse primer 5′-ACCTTTGC CTATGTGCAACC-3′, or by immunoblotting.

A total of 1 × 10³ cells with 100-μL media were seeded to each well of a 96-well dish coated with 45 μL of Matrigel (BD Biosciences) and reagents were added next day. The media and reagents were replaced every 3 days. Colonies were scanned and analyzed with GelCountTM (Oxford Optronix Ltd) on day 7 after MTT (4 mg/mL 3-(4,5-Dimethythiazol-2-yl)-2,5-diphenyltetrazolium bromide, Sigma) staining for 2 hours. The total area of colonies was calculated by average colony area multiplied by colony number.

Cells were counted with Trypan blue staining using a TC20 Automated Cell Counter (Bio-Rad) and 3,000 cells in 100 μL were seeded in 96-well culture plates in triplicate. DMSO, SP600125, or U0126 was added 24 hours later. Cell survival rates were calculated upon measuring absorbance following 72 hours using the CellTiter 96 AQ_ueous One Solution Cell Proliferation Assay (Promega) or AlamarBlue (Thermo Fisher) according to the manufacturer's recommendations. Percent viability was determined by comparing DMSO treatment to inhibitor treatment.

Bioluminescent images of mice orthotopically implanted with luciferase-transduced cells were acquired using a Xenogen IVIS Spectrum system (Xenogen Corporation). Mice were anesthetized with 2% isoflurane (Abbott Laboratories), and images were acquired at 10 minutes postinjection of 50 mg/kg i.p. dose of luciferin (Xenogen Corporation).

Female athymic nude mice, 6–10 weeks of age, were kept in a specific pathogen-free animal facility. The animals were fed clean and autoclaved food and water. All of the animal procedures were performed in accordance with a protocol approved by the Institutional Animal Care and Usage Committee. All mice were anesthetized with Ketamine (Sigma) 80–100 mg/kg and xylazine (Sigma) 5–10 mg/kg before injection. Tumor cells (5 × 10⁵) were prepared in 50 μL of Hank's balanced buffered solution (Life Technologies), and injected submucosally directly into the anterior tongue with a 27-gauge needle attached to a 1-mL tuberculin syringe. Mice were then examined 1 or 2 times a week for the development of tongue tumors and weight loss. The mice were euthanized by CO₂ inhalation when they had 25% body weight loss as when compared with their preinjection weight. After euthanasia, mouse tongue, cervical lymph node, salivary gland, lungs, and liver were resected. All tissues were fixed in formalin, embedded in paraffin, and stained with hematoxylin and eosin.

To further determine the clinical significance of SMAD4 loss, 130 tumors obtained from patients with newly diagnosed HNSCCs were analyzed for SMAD4 protein expression by IHC (Supplementary Table S1; Supplementary Fig. S1). The average age of patients was 59.3 years with 73.8% of the population being male. The primary tumor site was the oropharynx for 48.5% of the patients, and 50.8% of those were p16 positive. SMAD4 expression was detected in 86.9% of patients, but was undetectable in 6 (4.6%) tumors (Supplementary Table S1). This is consistent with 7% of HNSCC having SMAD4 alteration via homozygous deletion, and nonsense, missense, and silent mutations resulting in complete loss, while relative downregulation of SMAD4 expression is more commonly seen in 44.8% of HNSCC based on the gene expression analysis of TCGA dataset as previously published by our group (27). Of the 6 SMAD4-negative patients, 5 patients had a tumor site other than the oropharynx and were HPV negative.

We next evaluated SMAD4 status in a cohort of 43 patients with recurrent HNSCC (Supplementary Table S2). Similar to the newly diagnosed patients, the average age was 55.1 years with 72.1% of the population being male. Unlike the newly diagnosed patients, a larger subset of recurrent HNSCCs was SMAD4 negative. Approximately 26% of the recurrent HNSCC samples lacked SMAD4 protein expression and were likely HPV-negative given the primary tumor site and p16 staining status. Univariate analysis of SMAD4 expression in these patients showed that SMAD4 loss is associated with a trend toward worse survival. A strong association between SMAD4 loss and distant metastasis development was also observed (Table 1). Multivariate analysis showed that patients with SMAD4-negative tumors had more than a fivefold increased risk of developing distant metastases [HR, 5.85; 95% confidence interval (CI) 1.83–18.67, P = 0.003; Table 2].

Of the 43 patients with recurrent and/or metastatic disease, 15 were treated with cetuximab monotherapy. Treatment was given off-protocol limiting the ability to assess response or survival. However, cetuximab treatment duration can be used as a surrogate for potential clinical benefit as treatment was halted at time of disease progression. Eight patients with SMAD4-positive tumors received cetuximab treatment significantly longer than the 7 patients with SMAD4-negative lesions (mean length of cetuximab treatment, 140.5 days vs. 40.4 days; median length of cetuximab treatment, 63 days vs. 35 days; P = 0.02). The clinical data suggest that SMAD4 loss is associated with recurrence and cetuximab resistance in HNSCC.

To assess differentially expressed pathways in HNSCC tumors based on the SMAD4 expression levels, we used RNA-sequencing data from The Cancer Genome Atlas (TCGA) HNSCC dataset (n = 528) and categorized SMAD4 mRNA expression as high level (one SD above the mean of the TCGA tumor group) or low level (one SD below the mean of the TCGA tumor group; Fig. 1A). We then used OncoFinder, a new bioinformatic software suite for qualitative analysis of intracellular signaling pathway activation (SPA) based on transcriptomic data (28–30), to analyze pathway activation in TCGA tumors with low (n = 80) or high (n = 75) SMAD4 mRNA levels. Analysis revealed that pathways involved in cancer initiation, progression and maintenance, such as those associated with AKT, JNK, JAK/STAT, ILK, RAS, MAPK/ERK, p38, and WNT signaling, were significantly upregulated in most of the patients in the SMAD4-low expression group (Fig. 1A), whereas pathways associated with apoptosis were significantly downregulated in this cohort (Fig. 1A).

Although most of the HNSCC TCGA samples for whom HPV status is available are HPV negative (35, 36), HPV status has not yet been evaluated for many patients in this dataset. Our SMAD4 IHC dataset suggests SMAD4 loss predominantly occurs in HPV-negative tumors. Therefore, to further validate the role of SMAD4 in regulating signaling networks in HPV-negative patients, we applied a similar analysis to a separate set of transcriptomic data from 430 oral squamous cell carcinoma (OSCC) tumors, the most common subtype of HNSCC (5), derived from the well-annotated publicly available Gene Expression Omnibus (GEO) datasets (GSE9844, GSE41613, GSE30784, GSE42743, GSE31056, and GSE6791). All the datasets were obtained using the microarray platform Affymetrix Human Genome U133 Plus 2.0 Array. HPV infection does not appear to be a significant risk factor for OSCC and the prevalence of HPV-positive oral cavity cancers is relatively low, at less than 6% (37). After categorizing samples based on their SMAD4 mRNA level, we used OncoFinder suite to analyze pathway activation in tumors with low (n = 17) or high (n = 47) SMAD4 expression. Consistent with the results observed in TCGA cohort, well-known cancer-driving signaling axes such as RAS, PAK/p38, MAPK, EGFR, HGFR, JNK and TGFβ pathways were shown to be upregulated in virtually all SMAD4-low tumors, whereas apoptotic pathways had been significantly downregulated (Fig. 1B). Notably, 29 pathways were simultaneously upregulated (Fig. 1C), and 6 pathways were commonly downregulated (Fig. 1D) among the SMAD4-low tumors in both datasets (Fig. 1E). Commonly upregulated pathways reflected activation of ERBB family members and receptors important in EMT phenotype, including WNT, hGH, and TGFβ as well as downstream pathway modulators such as RAS, MAPK, JNK, and mTOR (Fig. 1E). Conversely, commonly downregulated pathways consisted of predominantly decreased apoptosis and cytokine regulation (Fig. 1E). Taken together, these data suggest that SMAD4 may regulate genes associated with key cancer pathways and further support the role of SMAD4 as a potent tumor suppressor in head and neck cancers.

As an in vitro model system to validate our computational findings and to further elucidate the functional consequences of SMAD4 loss, we used SMAD4-specific shRNA under a lentiviral promoter to selectively knock down its expression in the HPV-negative SCC61 HNSCC cell line, known to express high endogenous SMAD4 level. Cells infected with the scrambled shRNA were used as a control. The shRNA-mediated knockdown was confirmed by Western blot and real-time PCR analyses and resulted in 80% reduction of SMAD4 protein and mRNA levels (Fig. 2A). SMAD4 depletion in SCC61 cells did not affect SMAD2/3 levels or SMAD2 phosphorylation (Supplementary Fig. S2). To assess the effect of SMAD4 depletion on the long-term cell-growth, Matrigel colony formation assays were performed with either SMAD4-depleted (SCC61-SMAD4KD) or control (SCC61-SC) cell lines. SCC61-SMAD4KD cells displayed significantly higher clonogenic survival over the SCC61-SC cell line (Fig. 2B and C). Moreover, SMAD4 knockdown cells acquired a potent resistance to cetuximab, as demonstrated by markedly increased number and size of the colonies under cetuximab treatment, compared with the SCC61-SC cell line (Fig. 2D and E). To validate that this effect is SMAD4 loss-dependent, we have knocked down SMAD4 expression in additional HPV-negative cell line, SCC25 (Supplementary Fig. S3A). Consistently with our results in SCC61 cells, depletion of SMAD4 in SCC25 cell line rendered them less sensitive to cetuximab (Supplementary Fig. S3B), confirming that modulation of SMAD4 expression levels may mediate cetuximab resistance in HNSCC.

To test whether SMAD4 depletion affects tumorigenicity in vivo, SCC61L-SC and SCC61L-SMAD4KD cells that stably express firefly luciferase were injected into the oral tongue of athymic nude mice (10 mice per group) to generate orthotopic xenograft tumors. Mice were then examined for the development of oral cavity tumors and weight loss. Mice engrafted with the bioluminescent SCC61L-SMAD4KD cells exhibited enhanced tumor growth compared with the animals inoculated with the SCC61L-SC cell line (Fig. 2G). The mice were euthanized when they reached 25% body weight loss compared with their preinjection weight. Kaplan–Meier survival curve indicated that SMAD4 knockdown significantly reduces the overall survival (P = 0.0017) in our orthotopic HNSCC model (average days until euthanasia was 26.8 and 18.5 for SCC61L-SC and SCC61L-SMAD4KD, respectively; Fig. 2F). Notably, metastasis to cervical lymph nodes were detected in 2 (20%) mice injected with SCC61L-SMAD4KD cells, whereas lymph nodes harvested from SCC61L-SC xenografts showed no signs of metastatic lesion (Fig. 2H). Collectively, these findings further support our hypothesis that low SMAD4 expression may contribute to poor clinical outcomes seen in HPV-negative patients with HNSCC.

Multiple pathways predicted to be upregulated in SMAD4-low HNSCC tumors are principally related to proliferative and pro-survival EGFR signaling (Fig. 1E). As EGFR regulates activation of several tyrosine kinase pathways, such as MAPK, AKT and JNK, we next evaluated the activation status of EGFR and several key downstream signaling components (AKT, STAT3, MAPK, and JNK), following the depletion of SMAD4 in SCC61 cells. We first tested the phosphorylation of three major tyrosine autophosphorylation sites within the extreme carboxyl-terminal region of EGFR (Y1045, Y1068, and Y1173). SMAD4 depletion did not affect autophosphorylation in the position Y1048 and Y1068, whereas phosphorylation at residue Y1173 was significantly increased in SCC61-SMAD4KD cells compared with SCC61-SC cells (Fig. 3A). While there were no significant changes in AKT phosphorylation in SCC61-SMAD4KD cells, the levels of pSTAT3, p42/44 MAPK and pJNK were significantly upregulated following SMAD4 depletion (Fig. 3B). Consistently, depletion of SMAD4 in the SCC25 cell line has also resulted in a substantial upregulation of pMAPK and pJNK levels (Supplementary Fig. S3C).

As computational analysis predicted the downregulation of proapoptotic signaling in SMAD4-low HNSCC tumors, we next tested whether SMAD4 depletion in SCC61 alters expression of pro- and antiapoptotic markers. In accordance with our computational analysis, SMAD4 knockdown resulted in an increase of antiapoptotic proteins BCL-2 and BCL-XL (Fig. 3C), whereas protein expression level of BIM, a promoter of apoptosis, was significantly decreased in SCC61-SMAD4KD cells compared with control SCC61-SC cell line (Fig. 3C).

To further evaluate the role of SMAD4 in HNSCC, we investigated the effect of stable SMAD4 overexpression in HNSCC FaDuL cell-line, which has a complete loss of SMAD4 due to a de novo truncation mutation (Fig. 3D). A control cell line containing vector only was also created. SMAD4 overexpression in FaDu cells conferred potent long-term antiproliferative effect, as seen by a marked reduction in the colony-forming ability of SMAD4-overexpressing FaDu-SMAD4 cells, compared with control FaDu-mock cell line (Fig. 3E–G). Consistent with our OncoFinder result that SMAD4 depletion induces MAPK and JNK activation, reconstitution of the SMAD4 expression in FaDu cell line resulted in significantly lower p42/44 MAPK, pJNK, and pSTAT3 phosphorylation (Fig. 3H). Furthermore, FaDu-SMAD4 cells displayed lower pEGFR (Y1173), BCL-2, and BCL-XL levels compared with the control cell line (Fig. 3H).

To test the effect of SMAD4 overexpression in vivo, FaDu-mock and FaDu-SMAD4 cells were injected into the oral tongue of athymic nude mice (5 mice per group) to generate orthotopic xenograft tumors. Engraftment of control FaDu-mock cell line resulted in successful implantation in 100% of the animals, with metastasis to lymph node detected in 2 (40%) mice. Conversely, injection of FaDu-SMAD4 cells demonstrated a much lower take rate, with only 2 of 5 animals successfully developing tumors and showing no signs of metastatic lesion (Supplementary Table S3). Although the ability to conduct survival analysis was limited by low tumorigenicity of SMAD4-overexpressing cells, our data further support the role of SMAD4 as a potent tumor suppressor gene in HNSCC. Notably, SMAD4 overexpression resulted in a significant increase in cetuximab sensitivity, as demonstrated by Matrigel colony formation assays performed with either FaDu-SMAD4 or control cell lines (Fig. 3I). Taken together with the data revealed by pathway activation analysis, our observations suggest that SMAD4 may regulate cellular survival and proliferation dynamics by modulating activation of key positive regulators of HNSCC.

On the basis of the increased JNK and MAPK activation in SMAD4-depleted SCC61 cells, we targeted these kinases with small-molecule inhibitors in SMAD4 knockdown and control cell lines. Cells were serum starved and then exposed to JNK inhibitor (SP600125) for 24 hours with serum stimulation. Western blot analysis demonstrated that SMAD4-depleted cells express higher total JNK protein compared with the SCC61-SC cells (Fig. 4A). Notably, while JNK phosphorylation was efficiently inhibited by 5 and 10 μmol/L SP600125 in SC cells, SMAD4KD cells displayed only a modest decrease in phospho-JNK levels even when treated with 10 μmol/L SP600125 (Fig. 4A). To examine the effect of JNK inhibition on cell viability, SCC61-SC and SCC61-SMAD4KD cells were cultured in the presence of JNK inhibitor or DMSO for 48 hours and MTS cell viability assay was performed. Although JNK inhibition resulted in a short-term antiproliferative effect in both cell lines (Fig. 4B), a 7-day treatment with SP600125 did not significantly affect colony formation in a Matrigel colony growth assay (Fig. 4C).

Similarly, to evaluate the effect of inhibiting the MAPK pathway, SCC61-SC and SMAD4KD cells were synchronized by serum starvation and treated with a selective MAPK/MEK inhibitor U0126 or DMSO for 24 hours. Western blot analysis confirmed the increased phospho-MAPK level in SCC61-SMAD4KD cells compared with control (Fig. 4D). U0126 treatment significantly lowered MAPK phosphorylation in control cells at 1 μmol/L, whereas in SMAD4KD cells, the best response was observed at 10 μmol/L of U0126 (Fig. 4D). While 48-hour exposure to U0126 selectively reduced survival of SMAD4KD cells over control at all inhibitor concentrations (Fig. 4E), the long-term clonogenic survival of U0126-treated SCC61-SMAD4KD cells was no different or even higher than SCC61-SC (Fig. 4F). Although SMAD4 reconstitution in FaDu cell line confirmed the decrease in pMAPK and pJNK levels and restored dose-dependent inhibition of JNK and MAPK activity in response to SP600125 or U0126 respectively (Supplementary Fig. S4A and S4D), it failed to induce short-term antiproliferative benefit (Supplementary Fig. S4B and S4E) and its impact on long-term clonogenic survival was significant, but not egregious (Supplementary Fig. S4C and S4F). Multifaceted EGFR signaling orchestrates numerous prosurvival cellular processes (38). Therefore, activation of alternative EGFR downstream pathway targets that compensate for the loss of MAPK or JNK activity (i.e., adaptive rewiring) may provide one possible explanation for noneffectiveness of JNK and MAPK inhibitors as a monotherapy.

To examine whether JNK or MAPK inhibitors may enhance the antiproliferative effect of cetuximab, we treated SCC61-SC and SCC61-SMAD4KD cells with cetuximab in combination with either SP600125 or U0126. As expected, SCC61-SMAD4KD cells displayed resistance to cetuximab as a single agent compared with control cells (Fig. 5). Notably, while both JNK and MAPK inhibitors successfully sensitized SMAD4KD cells to cetuximab, JNK inhibitor was considerably more effective in SCC61-KD cells compared with control SCC61-SC cell line (Fig. 5). In alignment with these observations, inhibition of JNK and MAPK signaling effectively enhanced cetuximab sensitivity in SMAD4-null cell line (FaDu-mock), and further augmented its antiproliferative activity in SMAD4-expressing FaDu-SMAD4 cells (Supplementary Fig. S5). Taken together, our data indicate JNK and MAPK activation as potential mediators of cetuximab resistance and suggest the utility of concomitant EGFR/JNK or EGFR/MAPK inhibition as a novel strategy for overcoming cetuximab resistance in HNSCC tumors with low levels of SMAD4 expression.

TGFβ is a multifunctional cytokine that is overexpressed in a majority of cancers (8–11). The high-affinity binding of TGFβ to TGFβ receptor II recruits TGFβ receptor I into a heterotetrameric complex that initiates SMAD-mediated transcriptional activation or repression of a network of genes that control cell growth, differentiation, and migration (14). While dual effects of TGFβ signaling on tumor initiation and progression are cell specific and have yet to be determined under particular contexts, a number of studies have indicated that SMAD4 (a key downstream mediator of TGFβ signaling) plays crucial roles in maintaining tissue homeostasis and suppressing tumorigenesis (39–41). Although a causal link between somatic inactivation of SMAD4 and tumor progression was observed in various solid malignancies, including HNSCC (23, 26, 42, 43), the molecular mechanisms underlying SMAD4 loss–associated HNSCC tumorigenesis are not yet fully elucidated.

We have previously shown that in HNSCC obtained from TCGA dataset, HPV-negative tumors are more likely to have lower SMAD4 expression than HPV-positive HNSCC (27). Furthermore, we showed that decreased SMAD4 expression induces HPV-negative tumor cells to undergo EMT and renders them resistant to the EGFR inhibitor cetuximab (27). Consistent with these observations, our analyses of HNSCC tumors obtained from newly diagnosed or recurrent patients revealed that lack of SMAD4 expression is highly associated with HPV-negative tumor status. Notably, the frequency of SMAD-negative tumors was significantly higher among patients with recurrent HNSCC and strongly associated with distant metastasis development, further supporting the observation that in some cases loss of SMAD4 may facilitate a rapidly progressing and aggressive tumor phenotype (44–46). Moreover, recurrent and/or metastatic patients with SMAD4-negative tumors underwent significantly shorter cetuximab treatment than patients whose tumors expressed markedly higher SMAD4 levels, suggesting that loss of SMAD4 expression is a molecular signature that characterizes aggressive, cetuximab-resistant tumors. Further studies in a larger cohort of clinical samples are necessary to address the clinical significance of varying degrees of SMAD4 downregulation and functional SMAD4 insufficiency at a less stringent staining criterion.

The response to EGFR-targeted agents is inversely correlated with EMT in multiple types of tumors without known EGFR mutations, including NSCLC, head and neck, bladder, colorectal, pancreas, and breast carcinomas (47–50). These data suggest that EMT is a common denominator of tumors that are resistant to EGFR inhibitors. Although the significance of SMAD4 downregulation in the induction of EMT was recently described in our previous study (27), a recent work in pancreatic cells reported a paradigm in which TGFβ-induced Smad4-dependent EMT perturbs a protumorigenic transcriptional network resulting in apoptosis, which is prevented by Smad4 loss (51). While this study further confirms a pivotal role of SMAD4 in mediating EMT-associated signaling, the molecular mechanisms underlying this association in HNSCC have not yet been defined, and remain the subject of future research.

In the current investigation, we used the new bioinformatics software suite for the analysis of intracellular signaling pathway activation using transcriptomic data (29–31), for quantitative and qualitative comparison of the signaling pathway activation between SMAD4-low and SMAD4-high HNSCC tumors. Our comprehensive computational analysis has predicted that several oncogenic signaling pathways important for induction of the EMT phenotype in various solid malignancies, such as WNT (52), hGH (53, 54), and TGFβ (14, 52), as well as downstream pathway modulators such as RAS (52), MAPK (52), JNK (52), PAK/p38 (52), and mTOR (52), were significantly upregulated among the SMAD4-low tumors in both TCGA-HNSCC dataset and a subset of HPV-negative OSCC tumors obtained from GEO. These prosurvival signaling axes play a crucial role in cancer initiation, progression, and maintenance, and may contribute to survival and promotion of cetuximab resistance via inhibition of apoptosis and induction of cell proliferation. In line with these observations, proapoptotic pathways were predicted to be significantly downregulated in most of the SMAD4-low tumors.

In accordance with the in silico analysis, SMAD4 depletion induced cetuximab resistance in HPV-negative HNSCC tumor–derived human cells and was able to promote tumorigenicity in both in vitro and orthotopic in vivo models, further supporting the role of SMAD4 as a potent tumor suppressor in HNSCC, especially in HPV-negative tumors (27). Interestingly, the intracellular p53-regulatory network was predicted to be upregulated in tumors with decreased SMAD4 level. As the viral oncoprotein E6 inactivates p53 in HPV-positive HNSCC tumors (55) and low SMAD4 expression was predominantly seen in HPV-negative cases (27), one possible explanation for the increased p53 signaling in SMAD4-low samples is that these patients may be TP53 WT, HPV-negative HNSCC, and SMAD4 loss is an alternative tumor suppressor functional loss to TP53. It was reported that mutation of TP53 occurs in approximately 50% of HNSCC (21, 56), resulting in altered p53 expression and function. However, the differing nature and effects of these alterations on p53 expression and activation are frequently discordant. Therefore, additional studies are warranted to further define the role of SMAD4 functional loss in context of p53 signaling in tumorigenesis and therapeutic resistance of HNSCCs.

Although dysregulation of signaling pathways downstream of EGFR is a common feature of EMT progression (57–59), many studies focusing on elucidating the role of EGFR signaling in EMT regulation often do not distinguish the specific phosphorylation site differences. Our data reveal that while Y1048 and Y1068 positions of EGFR are constitutively phosphorylated independent of SMAD4 status, phosphorylation at residue Y1173 was significantly upregulated in SMAD4-depleted cetuximab-resistant SCC61 cells, compared with SMAD4-expressing cetuximab-sensitive control. Subsequently, reconstitution of SMAD4 expression in FaDu cells resulted in a substantial decrease in Y1173 level. Interestingly, it was recently reported that phosphorylation at Y1173, which is a major target of EGFR kinase inhibitors (60), leads to the activation of MAPK/ERK signaling cascade and correlates with poor prognosis in patients with OSCC tumors (61). In accordance with our in silico pathway activation analysis, phosphorylation of STAT3, MAPK, and JNK was significantly upregulated following SMAD4 depletion. Although these results are consistent with EGFR activation at Y1173, which has been implicated in downstream activation of the MAPK/ERK cascade (62), additional studies are warranted to better understand how SMAD4 loss might influence site-specific phosphorylation changes of activated EGFR. Furthermore, a recent study reporting that SMAD4 may execute its tumor-suppressive function via attenuation of JNK/MAPK activation (63), parallels our observation that pJNK level was significantly upregulated in SMAD4-depleted cells. Moreover, reconstitution of the SMAD4 expression in a SMAD4-negative cell line resulted in marked inhibition of cell growth and concomitant reduction in MAPK and JNK phosphorylation, suggesting that dysregulation of MAPK and JNK signaling may contribute to cetuximab resistance in subset of HPV-negative HNSCC tumors expressing low SMAD4 level.

Chemoresistance often ensues as a result of the concomitant activation of multiple, often overlapping signaling pathways. Therefore, for some cancers, inhibition of multiple, cross-talking pathways involved in cell growth and survival control with combination therapy is usually more effective in decreasing the likelihood that cancer cells will develop therapeutic resistance than with single-agent therapy. Interestingly, while JNK or MEK inhibitors as standalone agents had limited effects on growth and survival of SMAD4-depleted cetuximab-resistant cells, dual targeted inhibition of the MEK or JNK signaling combined with cetuximab, significantly reduced colony formation ability of SMAD4 knockdown cells, highlighting the potential therapeutic benefit of concomitant EGFR and JNK or MAPK inhibition in patients with HNSCC with low SMAD4 expression. Recent preclinical studies in patients with colorectal cancer and pancreatic ductal adenocarcinoma (64), have demonstrated that treatment with MEK inhibitors combined with EGFR-targeting agents are well tolerated and display modest antitumor activity. As low SMAD4 expression is more prevalent in HPV-negative HNSCCs (27, 65), it is tempting to assume that the level of SMAD4 expression may act as a determinant of sensitivity/resistance to EGFR/MAPK or EGFR/JNK inhibition in HPV-negative tumors. Nevertheless, further exploration in a wider range of cell lines and tumor samples of HPV subgroups is warranted to determine the ability of SMAD4 to stratify the subset of patients whose tumors could respond and who would clinically benefit from the combination therapy.

In summary, our data implicate JNK and MAPK activation as mediators of cetuximab resistance and provide the foundation for the concomitant EGFR/JNK or EGFR/MAPK inhibition as a novel strategy for overcoming cetuximab resistance in a subtype of HNSCC tumors with loss of SMAD4 expression.

C.H. Chung was a consultant in ad hoc scientific advisory board meetings held by Bristol Myers Squibb and Lilly Oncology. The terms of this arrangement are being managed by Johns Hopkins University in accordance with its conflict of interest policies. No potential conflicts of interest were disclosed by the other authors

Conception and design: E. Izumchenko, E. Makarev, A. Markovic, A. Bedi, C.H. Chung, H. Ozawa, R.S. Ranaweera, A. Zhavoronkov

Development of methodology: E. Izumchenko, E. Makarev, A. Markovic, J. Perez, C.H. Chung, H. Ozawa, R.S. Ranaweera, A. Zhavoronkov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Ranaweera, E. Izumchenko, J.D. Howard, A. Markovic, Q.-T. Le, C.S. Kong, R.C.K. Jordan, C.H. Chung, H. Kang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Ranaweera, E. Izumchenko, E. Makarev, E.J. Fertig, J.D. Howard, A. Markovic, A. Bedi, C.S. Kong, R.C.K. Jordan, H. Wang, H. Quon, D. Sidransky, C.H. Chung, A. Zhavoronkov

Writing, review, and/or revision of the manuscript: R. Ranaweera, E. Izumchenko, E.J. Fertig, J.D. Howard, A. Markovic, A. Bedi, R. Ravi, Q.-T. Le, C.S. Kong, R.C.K. Jordan, H. Wang, H. Quon, D. Sidransky, C.H. Chung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Ranaweera, E. Izumchenko, Q.-T. Le, C.H. Chung

Study supervision: E. Izumchenko, C.H. Chung

This project was supported by NIH grants 5R21DE023430 and RO1 DE017982 (to C.H. Chung), K25 CA141053 (to E.J. Fertig), and SPORE P50 DE019032 (to D. Sidransky).

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.