Chemoprevention of Head and Neck Cancer by Simultaneous Blocking of Epidermal Growth Factor Receptor and Cyclooxygenase-2 Signaling Pathways: Preclinical and Clinical Studies

This article is featured in Highlights of This Issue, p. 953

There are more than 52,000 new cases of head and neck squamous cell carcinoma (HNSCC) in the United States per year, with approximately 12,000 deaths (1). The survival rate for patients with HNSCC remains poor despite advances in detection and management (2). An effective preventive approach that could be implemented before the development of invasive cancer would be highly desirable to reduce the incidence of HNSCC (3).

One of the main etiologic factors that contributes to HNSCC development is tobacco carcinogen exposure throughout the upper aerodigestive tract. Extensive campaigns to educate the public about the risk of cigarette smoking have resulted in a substantial reduction in the percentage of adults who actively smoke in the United States (4, 5). However, several studies have shown that former smokers in addition to active smokers remain at a high risk of cancer development in the tobacco carcinogen–exposed field (6, 7), likely resulting from the accumulation of cellular and genetic alterations, resulting in activation of several signaling pathways that support cell proliferation (7, 8). Blocking one or several of these signaling pathways may impede or delay the development of cancer (9). The opportunity to prevent cancer formation has been shown by previous studies involving precancerous and cancerous lesions of the head and neck, including oral premalignant lesions (i.e., leukoplakia, erythroplakia, and/or erythroleukoplakia; refs. 10–12). However, a standard approach to chemoprevention of head and neck cancer has not been established (7, 13, 14). Thus, new strategies are needed to prevent the development of premalignant lesions into invasive cancer.

EGFR overexpression has been documented extensively in a wide variety of malignant tumors, including HNSCC (15–17). Overexpression of EGFR and its ligand TGF-α was observed in 80% to 90% of HNSCC specimens (18). Importantly, we previously showed that EGFR was upregulated in normal human oral epithelium adjacent to tumor and remained elevated throughout the phenotypic progression from hyperplasia to dysplasia to carcinoma in situ (15). EGFR expression was dramatically increased when dysplasia progressed to squamous cell carcinoma, suggesting that EGFR may serve as an excellent target for chemoprevention. Erlotinib (Tarceva, OSI-774), is an orally bioavailable EGFR-tyrosine kinase inhibitor (TKI) that is a low molecular weight quinazoline derivative. Erlotinib has been widely studied as a chemotherapeutic agent in a variety of cancer types including HNSCC (19). Erlotinib is well tolerated, although common toxicities include a characteristic acneiform rash and diarrhea (20). The ongoing erlotinib prevention of oral cancer trial (EPOC) with placebo control is expected to be completed by the end of 2014 (NCT00402779, see website: www.clinicaltrials.gov).

Cyclooxygenase (COX) catalyzes the synthesis of prostaglandin from arachidonic acid. COX-2 is overexpressed in many human cancers (for review, see ref. 21). In HNSCC, COX-2 expression is found to be upregulated at both the mRNA and protein levels. This upregulation is also seen in normal epithelium adjacent to HNSCC, and in premalignant lesions such as leukoplakia and dysplasia (22–26). Thus, selective COX-2 inhibitors (COX-2I) are considered as promising agents for chemoprevention. Treatment with COX-2Is, such as celecoxib, in a cancer chemoprevention trial, reduced the risk of developing familial adenomatous polyposis, some cases of which would inevitably progress to full-fledged colon cancer (27). Although the infrequent cardiovascular side effects of these agents raise concerns about their use in the preventive setting (18, 28), celecoxib as a single agent has been used in more than 20 trials for the chemoprevention of colon, oral, bladder, breast, and prostate cancers (29).

We and others have proposed blocking the interaction loop between the EGFR and COX-2 signaling pathways as a strategy for chemoprevention (for review, see refs. 30–32). We suggested that a combination of 2 agents for prevention and treatment of HNSCC, namely an EGFR-selective TKI and a COX-2I, act on different biologic targets, phosphorylated EGFR, and COX-2, respectively, but might affect the same signaling pathways. Therefore, we hypothesized that these 2 agents would function synergistically to inhibit tumorigenesis of the head and neck, which was supported by our preclinical studies using EGFR-TKI gefitinib and COX-2I celecoxib (33, 34). Recently, we have enrolled patients with premalignant lesions of the head and neck in a phase I chemoprevention trial using erlotinib and celecoxib. We observed a preliminary overall response rate of 71% (35). Tissue specimens from this clinical trial were obtained and examined for the modulation of biomarkers by the combined treatment. Along with preclinical studies using HNSCC cell line models in vitro and in vivo, we identified several biomarkers for which alterations in expression level correlated with clinical response.

The EGFR-selective TKI erlotinib was provided by Genentech. The COX-2 inhibitor celecoxib was obtained from G.D. Searle & Co. and Pfizer Inc.. Both drugs were dissolved in dimethyl sulfoxide (DMSO; Sigma Chemical) or Tween 80 (Sigma Chemical) in appropriate concentrations for in vitro or in vivo studies, respectively. Human HNSCC cell lines Tu212 and Tu686 were established from primary HNSCCs as described in our previous studies (33). The genotyping of these 2 cell lines confirmed their human origins (IDEXX RADIL Lab Animal and Biological Materials Diagnostic Testing, Columbia, MO). The cell lines were grown in Dulbecco's modified Eagle's medium (DMEM)/F12 (1:1) supplemented with 10% FBS. Human umbilical vein endothelial cells (HUVEC) were purchased from Lonza and cultured in EGM-2 medium (Lonza).

Cell growth inhibition was measured by determining cell density with sulforhodamine B assay (SRB; ref. 36) 72 hours after addition of the drugs as described in our previous studies (33). The drugs were added in a range of concentrations as single agents [erlotinib (0–10 μmol/L) and celecoxib (0–40 μmol/L)] and in combinations. Percentage inhibition was determined by comparison of cell density in the drug-treated cells with that in the untreated cell controls in the same incubation period (percentage inhibition = 1 − cell density of a treated group/cell density of the control group). All experiments were repeated 3 times. Combination index (CI) was determined using CalcuSyn software (Biosoft Inc). A CI value of more than 1 is defined as antagonism, equal to 1 as additive, and less than 1 as synergy.

The effects of each of the single agents and the combination on cell-cycle arrest were analyzed in Tu212 and Tu686 cell lines by propidium iodide/RNase staining (BD PharMingen). Cells were treated with erlotinib (1.0 μmol/L), celecoxib (10 μmol/L), and the combination for 24, 48, and 72 hours. The concentration of erlotinib used is similar to the achievable level in the blood of the patients (400 μg/L, 75 mg QD) observed in the pharmacokinetic study of the phase I clinical trial, whereas the celecoxib concentration used is less than the double dose of the achievable blood level (2,000 μg/L, 400 mg b.i.d.) (35). The experiments were repeated at least 3 times independently.

Annexin V–phycoerythrin (Annexin V–PE) staining was used to examine apoptosis induced by treatment with erlotinib and celecoxib in Tu212 and Tu686 cell lines. The cells were then resuspended in 1× Annexin binding buffer (BD PharMingen), and stained with Annexin V–PE (BD PharMingen) and 7-Aminoactinomycin D (BD PharMingen) for 15 minutes at room temperature. The stained samples were measured using a fluorescence-activated cell sorting caliber bench-top flow cytometer (Beckman Coulter).

Whole-cell lysates were extracted from drug-treated cells using lysis buffer and immunoblotted with specific antibodies as described previously (33, 34), except the antibody against phosphorylated ribosomal protein S6 (pS6) that was purchased from Cell Signaling technology. Mouse anti-β-actin antibody (Sigma Chemical) was used as a sample loading control. Immunostained protein bands were detected with an Enhanced Chemiluminescence Kit (Thermo Scientific).

To conduct the capillary tube formation assay, 24-well plates were coated with Matrigel (250 μmol/L/well; BD Bioscience). HUVEC cells suspended in EGM-2 medium (Clonetics Co.) containing DMSO (control), erlotinib (1 μmol/L), celecoxib (10 μmol/L), or the combination of erlotinib (1 μmol/L) and celecoxib (10 μmol/L) were added to each Matrigel-coated well (40,000 cells/well) and incubated overnight. The status of capillary tube formation by HUVECs was recorded at 2× using an Olympus inverted microscope (CKX40; Olympus). Experiments were repeated twice.

The invasion chamber was precoated with 27.2 μg Matrigel (BD Bioscience). HUVEC cells were suspended in serum-free medium containing 0.1% bovine serum albumin (BSA) with or without DMSO (control), erlotinib (1 μmol/L), celecoxib (10 μmol/L), or the combination of erlotinib (1 μmol/L) and celecoxib (10 μmol/L) in triplicate for 8 hours before being placed in the lower chamber as described previously (37). After 36 to 40 hours of incubation, cells at the upper membrane were removed with a cotton swab, and the membrane was stained with the Hema-3 kit (Fisher Scientific). The number of invaded cells was expressed as the sum of 10 random fields under ×200 magnification. Experiments were repeated twice.

The animal experiments were approved by the Animal Care and Use Committee of Emory University (Atlanta, GA). Before inoculation of 5 ×10⁶ Tu212 cells by subcutaneous injection into the right flank, nude mice (athymic nu/nu, Taconic) aged 4 to 6 weeks (about 20 g weight) were randomly assigned to 4 groups. Animals received oral gavaging with vehicle control (0.1% Tween 80 and 0.5% methylcellulose, n = 6), erlotinib (75 mg/kg body weight, n = 6), celecoxib (50 mg/kg body weight, n = 7), or the combination (n = 8) of erlotinb (75 mg/kg body weight) and celecoxib (50 mg/kg body weight) using a blunt-tipped 20G 1.1/2 needle (Popper and Sons). The dose selection was based on our previous publication (34), and is about half to two-thirds of the dose level when the agents were used as single drugs. The tumor size and body weight were measured thrice or twice weekly, respectively. The tumor volume was calculated using the formula: V = π/6 × larger diameter × (smaller diameter)² as reported previously (34).

Staining of Ki67, EGFR, phosphorylated extracellular signal regulated kinase (pERK), pS6, pAKT, COX-2, CD34, and terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) was conducted with formalin-fixed, paraffin-embedded tumor tissues as previously described in detail (34, 38). The antibodies against Ki67, EGFR, pERK, pAKT, and COX-2 were described in our previous studies. Antibody against pS6 was the same as used in immunoblotting. Antibody against CD34 was purchased from AbCam (Dilution 1:25). For Ki67 and TUNEL, the absolute number of labeled nuclei per 1,000 cells was counted in 10 randomly selected areas from each slide. Necrotic areas were excluded for analysis. The intensity of EGFR, pERK, pS6, and COX-2 staining was measured using a numerical scale (0–3 with 3 as the strongest staining) and quantified as weighted index [WI = % positive staining (> 0) in tumor × intensity score]. Positive CD34 signals were counted in 10 random fields at × 100. Microvessels were quantified as described previously (39).

Prostaglandin E2 (PGE2) is rapidly converted in vivo to its metabolites, with more than 90% of circulating PGE2 cleared by a single passage through the lung. To estimate the actual PGE2 production more reliably, the PGE2 metabolite (PGEM) in blood was measured using a PGEM EIA Kit following the standard manufacturer's protocol (Cayman Chemical). Blood from human or mouse was processed within 30 minutes after being drawn and plasma samples were stored at −80°C until use. The PGEM concentration was calculated using a standard curve generated from PGEM standards provided by the manufacturer. The experiment was repeated 3 times.

To each cell pellet, 100 μL extraction buffer was added and the tubes were placed at 4°C for 30 minutes with occasional shaking. Tubes were spun at 10,000 rpm for 15 minutes at 4°C. Supernatants were transferred to fresh tubes and the protein level estimated using the BCA kit from Promega. Cell lysates were labeled with Cy3 and applied to the microarray slides for 2 hours according to the manufacturer's instructions (Cancer/Apoptosis phosphorylation antibody array, Full Moon Biosystems). Signals obtained from the arrays were normalized to the controls spotted on each array. Levels of phosphorylated proteins were normalized to the total proteins. To correct for differences due to exposure between the 2 slides and protein loading (normalization to housekeeping gene), 2 factors were introduced in the data analysis following the manufacturer's introductions: a baseline correction factor (= average of the signal from each empty well − the median of the averages of the empty well signals from 2 slides) and a scaling factor (= average of β-actin signal − average of signal from empty wells).

We had access to tissue samples, cytobrushed cells, and blood prospectively collected from patients who had premalignant lesions (moderate to severe dysplasia and carcinoma in situ) and participated in a phase I chemoprevention trial conducted at the Emory University Winship Cancer Institute. The clinical trial was approved by the Institutional Review Board and continues to accrue patients. Tissue samples from participating patients were obtained and evaluated pathologically at 3, 6, and 12 months of therapy. All patients received celecoxib 400 mg orally twice daily and erlotinib was dose escalated at 3 dose levels of 50, 75, and 100 mg orally once a day continuously for 6 months. Pathologic complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD) were defined as CR: complete disappearance of dysplasia from the epithelium, PR: improvement of dysplasia at least 2 degrees (i.e., severe dysplasia becomes mild dysplasia) of dysplasia after treatment, SD: minor focal improvement without change of degree of dysplasia (i.e., focal improvement from moderate dysplasia to mild dysplasia with overall still moderate dysplasia) or no pathologicchanges after treatment, and PD: worsening of at least 2 degrees of dysplasia (i.e., mild dysplasia to severe dysplasia) or development of invasive cancer after treatment, accordingly. This clinical trial was first reported at the American Society of Clinical Oncology annual meeting 2010 (35).

A linear mixed model with random intercept was fitted to the log-transformed data to test for the effects of treatment on tumor growth. The Kruskal–Wallis and Wilcoxon tests were adopted to test the pairwise percentage of positive Ki67 expression, weighted index of pS6, and microvessel density in 4 treatment groups, respectively. Pairwise comparisons of PGEM concentration were conducted using a Wilcoxon test.

Spearman's correlation coefficient was then estimated to measure the relationship between the change in the value of each gene/protein with the clinical response of patient, assuming that the pathologic response in increasingly worse order was CR<PR<SD<PD. t test was further used to test whether the Spearman's correlation coefficient for each gene/protein was significant. All 122 genes/proteins were sorted by their Spearman's correlation coefficient in a descending order. The top 27 genes/proteins were plotted in a second heat map and analyzed with cluster analysis. Genes/proteins at the significance level of 0.1 were further plotted in a third heat map and analyzed with cluster analysis again. The SAS statistical package V9.2 (SAS Institute, Inc.) was used for data management and Spearman's correlation coefficient analyses. The heat map and cluster analysis were conducted with R language.

To understand the effect of erlotinib and celecoxib on the growth of HNSCC, we examined two HNSCC cell lines (Tu212 and Tu686), which express comparable amounts of activated and total EGFR and COX-2 proteins at basal level (data not shown). Cells were treated with erlotinib (0 to 40 μmol/L) and/or celecoxib (0 to 20 μmol/L) for 72 hours. A cell growth inhibition assay showed that both erlotinib and celecoxib as single agents inhibited growth of both HNSCC cell lines in a dose-dependent manner (Fig. 1A and B).

To evaluate the effect of the combined treatment with erlotinib and celecoxib on growth of HNSCC cells, a combination index was determined through a series of 2-fold reductions of erlotinib and celecoxib concentrations. The same concentration range used for the single drug was applied for the combination. The combination indices ranged from 0.29 to 0.48, indicating that the 2-drug combination of erlotinib and celecoxib synergistically inhibited cell growth in the tested HNSCC cell lines, T212 and Tu686.

To understand the underlying mechanisms of cell growth inhibition induced by the combination of erlotinib and celecoxib, we further investigated the effects of the combination on cell-cycle progression in Tu212 and Tu686 cells. As shown in Fig. 2A and B, cells were treated for 24, 48, and 72 hours with 1.0 μmol/L erlotinib and 10 μmol/L celecoxib. Compared with the control group, erlotinib alone induced significant cell-cycle arrest at the G₀–G₁ phase in both Tu212 and Tu686 cells as early as 24 hours after treatment (P = 0.0002 and 0.0017, respectively). Celecoxib alone exerted limited or no effect on cell-cycle arrest. The combined treatment induced a higher level of cell-cycle arrest at 72 hours in both Tu212 and Tu686 cells than celecoxib (P = 0.01 and P < 0.0001, respectively) or erlotinib (P = 0.04 and P = 0.002, respectively) alone. The 2 drug combination did not affect cell-cycle changes at either S or G₂–M phase (data not shown). Consistent with the G₀–G₁ arrest, we observed upregulation of the cell-cycle regulators p21 and p27 at 48 hours after treatment in these cells (Fig. 2C).

To examine whether growth inhibition of HNSCC cells by erlotinib, celecoxib, and their combination is also attributable to programmed cell death, we evaluated the apoptotic effect of the combination on Tu212 and Tu686 cells. Both cell lines were treated with erlotinib and celecoxib as either single agents or the combination for 72 and 96 hours. Approximately, 25% to 30% apoptosis was observed 96 hours after treatment with erlotinib alone and the combination. Treatment with celecoxib did not induce apoptosis under the current conditions (data not shown).

To elucidate the molecular functions by which the combination of erlotinib and celecoxib acts in HNSCC, we examined whether these 2 drugs could modulate activation of EGFR and its downstream AKT and ERK proteins, and S6 as a readout for mTOR signaling. HNSCC cell lines Tu212 and Tu686 cells maintained in medium with 5% serum were treated with celecoxib (10 μmol/L), erlotinib (1.0 μmol/L), or their combination (10 + 1.0 μmol/L) for 24, 48, and 72 hours. Immunoblotting analyses showed that at the concentration used, erlotinib as a single agent inhibited the phosphorylation of EGFR, AKT, ERK, and S6 as early as 24 hours after treatment, whereas celecoxib alone did not affect these pathways (Figure 3). Compared with the single drugs in both cell lines, the combination had an effect similar to erlotinib alone.

To assess potential antiangiogenesis activity, the effect of the combination of erlotinib and celecoxib on vascularization was assessed in HUVEC cells. Compared with control group, treatment with celecoxib alone or the combination significantly suppressed the level of tubular formation (P < 0.05), however, this process was not affected by erlotinib treatment (Fig. 4A). In addition, the ability of HUVEC cells to invade into tumor cells was significantly inhibited by the combination treatment but not by either single agent alone (P < 0.05; Fig. 4B), supporting the antiangiogenesis function of the combined treatment.

In this study, mice were given the chemopreventive agents erlotinib, celecoxib, and the combination approximately one week before HNSCC cell inoculation. All 3 treatments suppressed tumor growth significantly compared with the vehicle control (P < 0.0001, Fig. 5A). Importantly, the combination treatment showed a marked advantage over either erlotinib or celecoxib alone (P < 0.0001). The slightly greater potency of erlotinib over celecoxib in suppressing tumor growth was not significant.

All 3 treatments significantly reduced the expression of pS6 and microvessel density when compared with the staining in tumor tissues treated with the vehicle control (P < 0.05). Cell proliferation, reflected by Ki67 staining, was significantly inhibited by all treatment groups (P < 0.0001), and while the combination treatment reduced Ki67 expression more effectively than either single agent alone (P < 0.01), there was no significant difference between treatment with erlotinib and celecoxib (Fig. 5B and C).

The combined treatment tended to reduce the expression of EGFR, however, this effect was not statistically significant (data not shown). In addition, the treatments only slightly reduced the level of plasma PGEM collected at the time of termination (Fig. 5C). Consistent with the in vitro observation, apoptosis detected by TUNEL assay was not significantly higher in the combined treatment group than in the vehicle control (data not shown).

A total of 11 patients received treatment with erlotinib and celecoxib. Seven had both baseline and follow-up biopsies and cytobrushed cells from oral mucosa and sera and were evaluable for response using the last documented histologic response at 3, 6, or 12 months. Among them, based on the pathologic evaluation of biopsy samples by pathologists, 3 of 7 achieved a complete pathologic response (CR 42%), 2 of 7 a partial response (PR 29%), and 2 of 7 progressive disease (PD 29%) with an overall response rate (RR) of 71%. The mean duration of response for the 7 evaluable patients defined as the interval from the time of enrollment to the last pathologic evaluation, or last follow-up visit without evidence of progression, was 20 months (range, 6–36 months).

We explored whether changes in the levels of several biomarkers, which were affected by erlotinib and celecoxib in vitro, including EGFR, COX-2, pERK, pS6, and pAKT, before and after the treatment correlated with clinical response (Fig. 6A). IHC staining was quantified as described in the Materials and Methods section. Assuming a pathologic response in an increasingly worse order of CR< PR<SD<PD, we noted that a decrease in the expression level of EGFR and pERK at the last clinical response of the patient was significantly and positively associated with clinical response (P = 0.019 and P = 0.006 for EGFR and pERK, respectively; Fig. 6B). In addition, decreasing pS6 expression was associated with clinical response though this was not statistically significant (P = 0.16), perhaps due to the small sample size. Using the same analysis, there was no significant relationship between change in the expression level of COX-2 or pAKT at the last clinical response of the patient from baseline and actual clinical response (data not shown).

In addition, proteins isolated from the cytobrushed cells of the oral mucosa before and 3 months after treatment were used to screen potential biomarkers using an antibody array. A total of 7 patients were included in this analysis. Figure 6C shows a heat map of the 6 phosphoproteins that were found to have a significance level of 0.1 among all 7 patients (Supplementary Table S1). On the basis of the overall expression of these 6 proteins, patients with PD were completely separated from patients with other types of clinical response by the cluster analysis.

Numerous preclinical studies have flourished ever since the concept of chemoprevention was proposed by Dr. Sporn in 1976 (40). However, limited studies have been translated into therapeutic options as standard care for patients with premalignant lesions. Furthermore, to date, there have been 1,741 studies related to “prognostic markers” in head and neck cancer since 1983 in the PubMed database, but no biomarkers are recommended or approved by the U.S Food and Drug Administration (FDA) for routinely diagnosing premalignant lesions. Limiting factors for the development of premalignant biomarkers include: (i) paucity of preventive clinical trials; (ii) biostatistical challenges in biomarker validation (41); (iii) the need for repeated biopsies for pharmacodynamic biomarker study; (iv) biomarker development is often not in sync with targeted drug development, making it difficult to integrate and translate into clinical practice; and (v) the demand for large resources for validation trials, which have often been conducted far later than the drug intervention study has been completed.

Hence, to promote the integration of biomarker development into individualized targeted therapy (42), we propose adopting an incremental approach by which we seek every drug development clinical trial as an opportunity for biomarker identification and validation, bypassing this need for large and time consuming separate validation studies. In light of the success with the Dx/Rx model in breast cancer (pairing the HercepTest with Trastuzumab) and recent guidelines issued by the FDA promoting companion diagnostics (43), we propose that biomarker identification and incremental validation should be carried out in parallel in all clinical trial phases. Taking advantage of our phase I chemoprevention trial, we used 3 different patient tissue sources to examine biomarkers and identify those that correlate with clinical response to the combination treatment, that is, exfoliated cells of the oral mucosa collected by cytobrush before and 3 months after treatment, patient plasma samples at different time points, and patient biopsies at baseline and after treatment. To make comparisons across clinical trials in the future, we stratified the changes seen in biomarker expression levels based on pathologic response. In addition, we explored the involvement of these biomarkers in the mechanism of action of the combined treatment via in vitro and in vivo studies.

In the past few years, we have provided evidence to support our hypothesis that EGFR-TKI and COX-2I can act synergistically to inhibit tumorigenesis of the head and neck by showing gefitinib and celecoxib either additively or synergistically inhibited the growth of 5 HNSCC cell lines (33, 34). We have also confirmed these findings by combining a similar EGFR-TKI, erlotinib, with celecoxib in the current study in which we examined the alterations of some proteins which are directly and indirectly affected by the 2 targeted agents in both HNSCC cell lines and xenograft models. The results confirmed the synergetic/additive inhibitory effects of combined erlotinib and celecoxib treatment on HNSCC cell growth. The phase I chemoprevention trial exploring this combination in patients with premalignant lesions of the head and neck is ongoing and is near completion (thus, the full clinical study will be reported separately). The same proteins examined in the preclinical study were explored as the biomarkers in the phase I trial. The correlations between pathologic responses and levels of these biomarkers were evaluated in serial samples obtained from patients, including tissue biopsy and cytobrushed oral mucosa cells.

Our results from the biopsy tissues by IHC identified that downregulation of EGFR and pERK (Figure 6), but not pAKT, was significantly correlated with the patients' responses to treatment with erlotinib and celecoxib, suggesting that EGFR/ERK pathway is more sensitive to the combined treatment than the AKT pathway. The in vitro data consistently support the targeting effect of erlotinib on EGFR and its downstream pathways, although the findings from cell lines did not match the clinical observations for every biomarker. In the HNSCC cell line study, erlotinib showed strong reduction of pAKT (Fig. 3), but we did not observe a significant correlation of the change in expression of this protein with either clinical outcome or xenograft tumor inhibition. One potential explanation for this is that in addition to EGFR, AKT is also regulated by a complicated upstream network involving G-protein–coupled receptors and insulin growth factor 1 receptors (IGF-1R), the activation of which depends on autocrine and paracrine signals (44, 45). It is also possible that a larger sample size is required to observe whether pAKT modulation can serve as a biomarker for the combined treatment due to its complicated regulatory nature. From another perspective, we should be aware of the limitations of the use of cell lines and xenograft models as guides for the selection of biomarkers in a clinical trial, as there will inevitably be some inconsistent observations between cultured cells/xenograft tumors and human tissues due to different biologic environments. However, human cell lines and xenograft tumor models are still the most commonly used systems for preclinical evaluation of therapeutic agents and their relevant biomarkers. In this study, the use of cell lines and xenograft mouse models serves as a simplified system to provide evidence and a starting point to support the clinical investigation. In light of an excellent review describing why the translation of cancer biomarkers into clinical practice is a highly failure-prone process (46), our consistent findings across 3 systems should be recognized.

As an exploratory study, we conducted phosphorylation antibody array analysis using proteins isolated from the cytobrushed mucosa cells before and 3 months after treatment from 7 patients. Changes in the expression levels of 6 proteins were identified to discriminate progressive disease from the responsive cases (Supplementary Table S1). Among these 6 phospho-proteins, 3 are in the downstream part of the EGFR signaling pathway and 1 is in the downstream section of the mTOR pathway, both major signaling pathways that dominated our biomarker identification in the phase I chemoprevention trial. The other 2 phospho-proteins are a breast cancer suppressor and an NF-κB regulator, which both deserve further investigation. The 6 significant phospho-proteins found in the antibody assay study have potential to be signature proteins for clinical response. However, we should acknowledge that because there were only 6 samples, no conclusion can be made from this study. With a larger sample size, we expect to find significant proteins that can predict clinical response.

Our observation of downregulation of pS6 (Figure 6), a commonly used readout for mTOR activation (47, 48) in premalignant tissues after the combined treatment brought the mTOR pathway to our attention. The modulation of pS6 by the combined treatment was supported by our preclinical studies in both HNSCC cell lines (Fig. 3) and the HNSCC xenograft model (Figure 5C). The contribution of the mTOR pathway to the early stage of head and neck carcinogenesis was reported by Gutkind and colleagues in an oral-specific chemical carcinogenesis model using the mTOR inhibitor rapamycin for prevention of HNSCC (49, 50). On the basis of the previous studies, inhibition of the mTOR pathway may result from the inhibition of its upstream suppressor TSC1/2 by pERK and pAKT (48). As the roles of the mTOR pathway in the early development of HNSCC have not been well elucidated, our finding deserves further investigation.

The observation of PGEM downregulation in mouse blood is mainly attributed to the COX-2I celecoxib, although it is not significant. It seems that celecoxib does not play a role in most of the pathways examined in our study. However, consistent with previous reports from our group and others (33, 51), we observed that celecoxib inhibited microtubular and mouse vasculature formations both in vitro (Fig. 4) and in vivo (Fig. 5B and C), respectively, indicating the antiangiogenesis activity of this agent. Celecoxib also contributed to regulation of proliferation as evidenced by the induction of p21 and p27 in vitro (Figs. 1 and 2C) and reduction of Ki67 in vivo (Fig. 5C). In addition, blockage of EGFR and its downstream mTOR pathways by erlotinib also showed an antiangiogenesis effect in vitro and in vivo (Figs. 4A and 5C, respectively). Therefore, it was expected that the combination of celecoxib and erlotinib would achieve greater growth inhibition than either single agent both in vitro and in vivo (Figs 1A and B and Fig. 5A, respectively).

The observations of our biomarker studies are supported by recent publications from Kao and colleagues who reported both a preclinical study and a phase I trial of concurrent erlotinib, celecoxib, and reirradiation therapy for recurrent head and neck cancer (52, 53). Their studies suggest that combination therapy with erlotinib and celecoxib can also be used as a radiation-sensitizing approach in the treatment of recurrent head and neck cancer. For treatment of recurrent and metastatic HNSCC, Wirth and colleagues also conducted a phase I clinical trial using a combination of gefitinib and celecoxib, which achieved a PR rate of 22% (54). In addition, Ferris and colleagues recently reported a pilot randomized trial in a neoadjuvant setting using erlotinib plus sulindac, a nonspecific inhibitor targeting COX1/2. The major goal of this study was to study modulation of serum proteins upon the treatment (55).

In summary, we observed correlations between changes in the activated protein levels of the EGFR signaling pathway with the combined therapy using erlotinib and celecoxib in both cell lines and a preventive xenograft model. As expected, the changes in some protein markers were significantly correlated with the clinical outcome of patients on a phase I chemoprevention trial, although the sample size is small. Our preclinical studies not only confirm the clinical findings but also provide insight into the mechanisms of action of combined treatment with erlotinib and celecoxib in head and neck cancer prevention. Together, our investigation combining both preclinical studies and the biomarker-driven phase I clinical trial serves as a proof-of-principle to justify future chemoprevention trials that target multiple signal transduction pathways involved in carcinogenesis of head and neck cancer.

J. Beitler is a consultant/advisory board member of ACR and NIH. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D.M. Shin, Z.G. Chen, H. Zhang, A.Y. Chen, J.C. Brandes, A.R.M.R. Amin

Development of methodology: Z.G. Chen, H. Zhang, S.M. Thomas, A.R.M.R. Amin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.M. Shin, Z.G. Chen, G. Sica, N.F. Saba, A.Y. Chen, S. Nannapaneni, S. Muller, S.A. Kono, J. Beitler, F.R. Khuri, M.M. Lewis, W.J. Grist, S.M. Thomas, R. Moreno-Williams

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.M. Shin, Z.G. Chen, H. Zhang, N.F. Saba, S. Nannapaneni, S. Muller, Z. Chen, H.J.C. Shin, A.R.M.R. Amin

Writing, review, and/or revision of the manuscript: D.M. Shin, Z.G. Chen, H. Zhang, N.F. Saba, A.Y. Chen, S. Nannapaneni, J.C. Brandes, Z. Chen, F.R. Khuri, W.J. Grist, J. Grandis

Study supervision: D.M. Shin, Z.G. Chen, N.F. Saba, F.R. Khuri

We thank Genentech for providing erlotinib for this study. We also thank Dr. Anthea Hammond for her critical reading of the article.

This study was supported by grants from the NIH (U01 CA101244 and Specialized Programs of Research Excellence in Head and Neck Cancer, P50 CA128613; R01 CA112643 to D.M. Shin) and R01CA0983722, P50CA097190, and the American Cancer Society (to J.R. Grandis).

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.