Distinctive E-cadherin and epidermal growth factor receptor expression in metastatic and nonmetastatic head and neck squamous cell carcinoma

Predictive and prognostic correlation




The authors investigated whether coexpression and localization of E-cadherin (E-cad) and epidermal growth factor receptor (EGFR) had predictive and/or prognostic correlations with lymph node metastasis and/or survival in patients with squamous cell carcinoma of the head and neck (SCCHN).


Immunohistochemistry (IHC) of archival tissue was performed to measure expression of EGFR and E-cad in surgical specimens of SCCHN (n = 143) that included primary tumors (PTs) with positive lymph nodes (Tu+Met) and their paired lymph node metastases (LnMet), PTs with negative lymph nodes (Tu−Met), and benign tissue biopsies as normal controls. IHC staining was quantified as a weighted index and as the ratio of membrane to cytoplasmic staining. Correlative expression between EGFR and E-cad also was examined in SCCHN cell lines by immunoblotting and immunofluorescence analyses.


Three distinct expression patterns of EGFR and E-cad were observed. Membrane localization of E-cad was significantly lower in the Tu+Met group than in the Tu−Met group (P = .01) and was associated inversely with lymph node status (P = .009). Wilcoxon analysis of the combined markers demonstrated that expression and/or membrane localization of EGFR and E-cad were correlated with disease-free survival and overall survival in patients with SCCHN. The study of SCCHN cell lines demonstrated that cells with positive but low EGFR expression and with negative E-cad expression were relatively resistant to the EGFR tyrosine kinase inhibitor erlotinib.


The current study suggested that examining not only the expression but also the localization of EGFR and E-cad simultaneously may have clinical relevance in predicting lymph node metastasis, patient survival, and response to EGFR-targeted therapy in patients with SCCHN. Cancer 2008. © 2008 American Cancer Society.

Head and neck cancer (HNC) is the sixth most common cancer and is responsible for almost 200,000 deaths worldwide each year.1, 2 Despite advances in understanding the molecular mechanisms of HNC along with improved diagnosis, the 5-year survival rate has remained virtually unchanged in the past 30 years. Both locoregional recurrences and lymph node metastasis (LNM) are associated with a poor prognosis. The 5-year survival rate of HNC is <50% for patients with a single ipsilateral LNM and <25% for patients with bilateral LNM. Therefore, the identification of biomarkers associated with both locoregional failure and LNM could predict tumor behavior and guide treatment of HNC.

The epidermal growth factor receptor (EGFR) is a 170-kDa transmembrane protein with intrinsic tyrosine kinase activity that regulates cell growth in response to binding of its ligands, including epidermal growth factor (EGF) and transforming growth factor-α (TGF-α). EGFR overexpression has been documented extensively in a wide variety of malignant tumors, including squamous cell HNC (SCCHN).3–10 Overexpression of EGFR and its ligand TGF-α reportedly is observed in 80% to 90% of SCCHN specimens.11–14 Several studies have demonstrated that EGFR overexpression correlates with reduced disease-free survival (DFS) and overall survival (OS).5, 7–9, 14 Therefore, many strategies, including the use of specific tyrosine kinase inhibitors (TKI) and monoclonal antibodies to target EGFR, have been developed for the treatment of SCCHN. However, resistance to EGFR-TKI treatment has been observed in lung cancer, SCCHN, and other types of cancer.15

A recent publication suggested that an epithelial-to-mesenchymal transition (EMT) is a determinant of the sensitivity of cancer cells to EGFR inhibition.16 The EMT biomarker E-cadherin (E-cad) was used to predict clinical activity for patients with lung cancer who received combined therapy with erlotinib and chemotherapy.17 Restoring E-cad expression enhanced sensitivity to EGFR-targeted therapy,18 suggesting that E-cad expression may be required for EGFR-dependent tumor cell progression. Currently, the molecular relation between EGFR and E-cad is unclear.

E-cad is a cell-cell adhesion transmembrane molecule. It plays important roles not only in cell adhesion and morphogenesis but also in cellular signal transduction in collaboration with EGFR/ERK- and c-Src-mediated pathways. E-cad expression in SCCHN tissue specimens has been reported in several studies.19–22 Bosch et al. studied primary tumors (PTs) and LNMs and observed that decreasing E-cad expression was a dominant prognostic factor in SCCHN.20 Kyzas et al. reported that loss of E-cad expression was associated with increased dysadherin that is related to lymphangiogenesis and may be a promising prognostic marker for SCCHN.22 However, those authors did not report that E-cad alone had any significant correlation with LNM. In those studies,E-cad was not examined in association with EGFR.

By using both clinical specimens and cell line models of metastatic and nonmetastatic SCCHN, we examined the coexpression of EGFR and E-cad, the correlation between expression levels and localization of these 2 biomarkers, and factors that were correlated with patient outcome. The identification of significant associations between EGFR and E-cad in SCCHN may provide new insights into the biology of tumor progression and metastasis of this disease. This information may have clinical implications in the management of SCCHN.


Patient Materials

Using an institutional review board-approved consent for tissue acquisition, specimens for this study were obtained from surgical specimens of patients who had SCCHN diagnosed at Emory University Hospital and whose initial treatment was surgery and who had never received prior treatment with radiation and/or chemotherapy. The selection criteria of the available formalin-fixed and paraffin-embedded tissue blocks included 2 groups: primary squamous cell carcinoma (SCC) with positive lymph nodes(N-positive) (n = 49 patients; the Tu+Met group), their paired LNMs (n = 47 patients; the LnMet group), and primary SCC with negative lymph nodes (N-negative) (n = 47 patients; the Tu−Met group). In the Tu−Met group, if any patient developed metastases within 2 years of the initial procedure, then they were excluded from the study. In addition, 10 benign oral soft tissue specimens from noncancer patients were collected as normal controls. The clinical information on the samples was obtained from the surgical pathology files in the Department of Pathology at Emory University according to the regulations of the Health Insurance Portability and Accountability Act. The clinicopathologic parameters for the 2 study groups, including age, sex, tobacco history, tumor location, and disease stage, are listed in Table 1. Each patient's DFS and OS were documented through December 7, 2006.

Table 1. Clinicopathologic Features of the Metastatic and Nonmetastatic Study Groups
CharacteristicPercentage of patients
No metastatic diseaseMetastatic disease
  • WD indicates well differentiated; MD, moderately differentiated; PD, poorly differentiated.

  • *

    Six patients had unknown tobacco history.

  • Eight patients had unknown tobacco history.

Average age (mean), y63 (63)60.1 (58)
 Ratio of men:women1.9:14.4:1
Tobacco history83*90.2
Tumor location  
 Oral cavity5527
Tumor classification  
Lymph node status  
Histologic grade  

SCCHN Cell Lines

The SCCHN cell line 686LN was established from an LNM of a primary base of tongue SCC.23 686LN-M4e is a highly metastatic cell line generated by in vivo selection from 686LN cells that had low metastatic potential in nude mouse lymph node, as described previously.11, 24 The SCCHN cell line 686LN-R30 is an EGFR–TKI-resistant cell line that was established from 686LN cells by single-cell cloning after challenge with gradually increased concentrations of EGFR-TKI gefitinib. Two other SCCHN cell lines, UPCI-37A and UPCI-37B, were established from larynx (epiglottis) at the University of Pittsburgh Cancer Institute (Pittsburgh, Pa): The UPCI-37A cells were from a PT, whereas the UPCI-37B cells were from an LNM. The cell lines were maintained as monolayer cultures in Dulbecco modified Eagle medium/F12 medium (1:1) supplemented with 10% fetal bovine serum (FBS) at 37 °C in a humidified atmosphere with 5% CO2.


Formalin-fixed, paraffin-embedded tissue sections were used for immunohistochemistry (IHC), as described previously.25 The slides were incubated with primary antibodies (EGFR, clone 31G7: 1:50 dilution; Zymed Labs, South San Francisco, Calif; E-cad, clone 36: 1:2000 dilution; BD Biosciences, Franklin Lakes, NJ). The staining of the antibodies was observed by 3,3′diaminobenzidine tetrahydrochloride peroxidase substrate solution (Vector Laboratories, Burlingame, Calif). Cell nuclei were counterstained by using Hematoxylin QS (Vector Laboratories). Mouse immunoglobulin G (IgG) was used as a negative control, and normal epithelial tissue with known positive immunoreactions to both anti-EGFR and anti-E-cad antibodies were used as positive controls.

The intensity of immunohistochemical staining was measured by using a numerical scale (0, no expression; 1+, weak expression; 2+, moderate expression; and 3+, strong expression) and was quantified as the weighted index (WI = %positive stain [>0] in tumor × intensity score), as described previously by our laboratory.25 Membrane and cytoplasm distribution of each biomarkers were recorded as the ratio of membrane:cytoplasm distribution (RMC = membrane distribution/[membrane distribution + cytoplasm distribution]). Both the WI and the RMC were determined by 2 individuals, and the final values were the average of the 2 readings.

Immunoblotting and Immunofluorescence

Immunoblotting was performed as described in our previous reports.6, 24 Primary antibodies for immunoblotting were monoclonal antibodies against E-cad (clone G-10; 1:1000 dilution), vimentin (V9; 1:500 dilution), and polyclonal antibodies against EGFR (clone 1005; 1:500 dilution) and phosphorylated EGFR (p-EGFR) (Tyr 1173; 1:500 dilution; Santa Cruz Biotechnology, Santa Cruz, Calif); glyceraldehyde-3-phosphate dehydrogenase (1:3000 dilution; Trevigen, Inc., Gaithersburg, Md) was used as an internal control.

Localization of EGFR was observed by immunofluorescent staining according to the protocol described in our previous study.24 In brief, cells were incubated with rabbit polyclonal anti-EGFR antibody (1:400 dilution; Cell Signaling Technologies, Beverly, Mass). For negative controls, rabbit IgG was used at the same concentration. After washing, the cells were incubated with Alexa Fluor488 goat antirabbit IgG (Molecular Probe, Carlsbad, Calif). Nuclei were counterstained with 4′,6-diamidino-2-phenylindole. The results were observed under Zeiss LSM 510 confocal microscopy.

Cell Growth Inhibition Assay

To test the effect of erlotinib on cell growth of SCCHN, the sulforhodamine B (SRB) cytotoxicity assays were adapted from the method described by Skehan et al.26 Cells that were maintained in medium with 5% FBS were seeded in 96-well plates at a density of 4000 cells per well overnight before drug treatment. Afterward, drugs were added as single agents at various concentrations (range, 0-40 μM), followed by incubation at 37 °C and 5% CO2 for 72 hours. Cells were fixed for 1 hour with 10% cold trichloroacetic acid. Plates were washed 5 times in water, air dried, then stained with 0.4% SRB for 10 minutes. After the cells were washed 4 times in 1% acetic acid and air dried, bound SRB was dissolved in 10 mM unbuffered Tris base, pH 10.5. Plates were read in a microplate reader by measuring A564. Then, the percent of cells that survived was calculated based on the absorbance values relative to untreated samples. The experiment was repeated at least twice.

Statistical Analysis

Because metastatic and nonmetastatic tumors may have different etiology/biology, separate summary statistics were calculated within each of the following groups: LnMet, Tu+Met, and Tu−Met. Because the sample sizes were relatively large in each group, all significance levels were calculated by using the 2-sample t test. Significance levels for comparisons of matched samples (ie, LnMet vs Tu+Met) were calculated by using the t test for paired data. Distributions of EGFR and E-cad expression (WI) and location (RMC) by tumor groups were described by using box plots.

Correlations between the WIs and the RMCs for EGFR and for E-cad were calculated by using the Spearman rank correlation. Logistic regression, Kruskal-Wallis tests, and Spearman correlations were used to correlate WIs and RMCs with clinical characteristics. Logistic regression and Cox proportional hazards models were used for the categorical data and time-to-event data, respectively. First, we evaluated both DFS and OS using a Cox proportional model or the log-rank statistic to correlate either WI or RMC for EGFR and for E-cad as single markers. Then, the WIs and the RMCs were categorized as high (greater than the median value) and low (less than or equal to the median value) and were classified into 4 groups, high-high (HH), high-low (HL), low-high (LH), and low-low (LL). Thus, the appropriate double markers were used to perform the correlative analyses for DFS and OS based on the defined cutoff values by using the Kaplan-Meier product limit method.


Expression and Localization of EGFR and E-cad in PTs and LNMs of SCCHN

Expression of EGFR and E-cad proteins in PTs and LNMs of SCCHN was studied by IHC analysis (Fig. 1). The tissue samples included N-positive PTs (Tu+Met), their LNMs (the LnMet group), and N-negative PTs (Tu−Met). Normal oral epithelial tissues were used as positive controls, and tissues stained with the relevant IgGs were used only as negative controls. Both EGFR and E-cad are located in the cell membrane of normal epithelial cells where the expression of both proteins is observed mainly in the basal and spinous layers (data not shown). The average WIs of the 2 proteins present in normal oral epithelium were compared with those in the tumor tissues. Three expression patterns were observed in the tumor tissues. Compared with normal tissues, 40% of all samples had overexpression of both EGFR and E-cad (Fig. 1A), 48% had overexpression of EGFR but reduced expression of E-cad (Fig. 1B,C), and 12% had low or no expression of both EGFR and E-cad (Fig. 1D). When both EGFR and E-cad were overexpressed, the 2 proteins were located mainly in the cell membrane of well differentiated (WD) and in some moderately differentiated (MD) tumors (Fig. 1A). In tissues with EGFR overexpression and reduced E-cad expression, EGFR had either membrane staining (Fig. 1B) or cytoplasmic staining (Fig. 1C), but E-cad had very weak or undetectable cytoplasmic staining in poorly differentiated (PD) squamous cancers and in some MD cancers. In the third pattern, both E-cad and EGFR expression had weak or undetectable cytoplasmic staining, predominately in PD tumor tissues (Fig. 1D). This pattern was observed in 8% of Tu+Met tissues. However, in the corresponding LNMs, it increased to 20%.

Figure 1.

Immunohistochemistry staining patterns of epidermal growth factor receptor (EGFR) and E-cadherin (E-cad) in tissues from squamous cell carcinoma of the head and neck. In tumor cells, EGFR and E-cad are expressed in both the membrane and in the cytoplasm. There are 3 major patterns of E-cad and EGFR expression as labeled in the figure. (A) Forty percent of the total tumor samples had overexpression of both EGFR and E-cad compared with normal control samples. (B,C) EGFR overexpression in membrane and cytoplasm, respectively, and reduced E-cad expression were observed in 48% of the total tumor samples. (D) Reduced or no expression of both EGFR and E-cad was noted in 12% of the total samples.

The 2-sample t test indicated that both the mean WI and the mean RMC of E-cad were lower in metastatic PTs (Tu+Met) than in nonmetastatic PTs (Tu−Met; P = .059 and P = .01, respectively), indicating that E-cad expression was decreased and internalized in the cytoplasm of metastatic tumors compared with that in nonmetastatic tumors (Fig. 2A). The mean WI of EGFR was significantly higher in the Tu+Met group than that in the Tu−Met group (P = .02), but more cytoplasmic staining was observed in the Tu+Met group (Fig. 2B). The results from t tests for paired data comparison between the PTs (Tu+Met) and the LNMs (LnMet) illustrated that the mean WIs for both EGFR and E-cad and the mean RMC for EGFR were significantly lower in the LnMet group than in the Tu+Met group (P = .009, P = .013, and P = .038, respectively), indicating that EGFR was internalized further and resulting in reduced intensity in the LNMs.

Figure 2.

The weighted index (WI) and the ratio of membrane:cytoplasm distribution (RMC) of E-cadherin (E-cad) and epidermal growth factor receptor (EGFR) expression in tissues from squamous cell carcinoma of the head and neck. The expression (WI) and location (RMC) of E-cad (A) and EGFR (B) by tumor groups are described by using box plots, which are graphic representations of the median, 25th percentile, and 75th percentile, and range of the data. The median is represented by the solid horizontal line inside each box, and the dashed line represents the mean. The horizontal bracket and lines that extend outside the box represent the entire range of the data. Tu−Met indicates primary tumors with negative lymph nodes; Tu+Met, primary tumors with positive lymph nodes; LnMet, paired lymph node metastases.

Correlation of Expression and Localizationof EGFR and E-cad

Spearman correlation demonstrated positive correlations between the RMCs for EGFR and E-cad in both the Tu+Met group and the Tu−Met group (P < .0001 and P = .005, respectively) (Table 2). The RMC for EGFR and the WI for E-cad were correlated in theN-positive group (Tu+Met; P < .0001).

Table 2. Spearman Correlation Coefficients Between the Weighted Index and the Ratio of Membrane:Cytoplasmic Distribution of Epidermal Growth Factor Receptor and E-cadherin
Patient groupSpearman correlation coefficient (P values)
  • EGFR indicates epidermal growth factor receptor; WI, weighted index; RMC, ratio of membrane:cytoplasmic distribution; E-cad, E-cadherin; Tu−Met, primary tumors with negative lymph nodes; Tu+Met, primary tumors with positive lymph nodes; LnMet, paired lymph node metastases.

  • *

    Statistically significant P values.

 EGFR WI1.0000.5141 (.0002)*.1616 (.2778).0594 (.6916)
 EGFR RMC1.0000.0848 (.5711).4028 (.0050)*
 E-cad WI1.0000.6524 (<.0001)*
 E-cad RMC1.0000
 EGFR WI1.0000.1016 (.4878).1710 (.2401).1057 (.4874)
 EGFR RMC1.0000.7761 (<.0001)*1.0000 (<.0001)*
 E-cad WI1.0000.7761 (<.0001)*
 E-cad RMC1.0000
 EGFR WI1.0000.65927 (<.0001)*.4354 (.0022)*.4671 (.0009)*
 EGFR RMC1.0000.2594 (.0779).4475 (.0016)*
 E-cad WI1.0000.7553 (<.0001)*
 E-cad RMC1.0000

Association of Expression and Localization of EGFR and E-cad With Clinical Characteristics and Survival

Statistical analyses of the IHC data were performed to correlate WIs and RMCs with clinical characteristics. These factors had no correlations with sex, age, smoking history, or T classification either in patients with metastatic disease or in patients with nonmetastatic disease. The location of E-cad was correlated significantly with PT sites in the Tu−Met group, with predominately membrane staining observed in the oral cavity, both membrane and cytoplasmic staining observed in the larynx, and almost exclusively cytoplasmic staining observed in the oropharynx (median RMC: 0.8, 0.35, and 0.2, respectively; P = .003). The level of E-cad was correlated significantly with histologic grade in the Tu−Met group (median WI: WD, 240; MD, 130; PD, 12.5; P = .002). The localization of E-cad expression was associated significantly with lymph node status, with increasing cytoplasmic staining observed in N-positive PTs (median RMC: N-positive tumors, 0.3; N-negative tumors, 0.5; P = .009). Furthermore, in the Tu+Met group, the RMC for E-cad was associated inversely with N status (N-negative vs N-positive; P = .009) in which N1 tumors had a higher E-cad RMC than N2 and N3 tumors(P = .035).

Statistical analysis with a Cox proportional model demonstrated that the RMC for EGFR was the only single marker that was correlated with both DFS (P = .011; hazard ratio, 1.85) and OS (P = .026; hazard ratio, 1.68) in the combined patient groups (Tu−Met and Tu+Met), illustrating that a high RMC for EGFR was associated with poor DFS and OS. However, neither of the single markers was correlated significantly with either DFS or OS when the same analyses were performed separately for each of the 2 groups (data not shown). Wilcoxon analysis demonstrated that combining the WI for EGFR with the RMC for E-cad was correlated with DFS in the Tu−Met group (P = .018) (Fig. 3A), whereas combining the RMCs for EGFR and for E-cad was correlated with both DFS and OS in the Tu+Met group (P = .062 and P = .036, respectively) (Fig. 3B).

Figure 3.

Kaplan-Meier plot for disease-free survival (DFS) and overall survival (OS) according to the expression and/or localization of epidermal growth factor receptor (EGFR) and E-cadherin (E-cad). (A) In the patients who had primary tumors with negative lymph nodes (Non-Met Group), the EGFR weighted index (WI) and the E-cad ratio of membrane:cytoplasm distribution (RMC) were categorized as high (greater than the median value) and low (less than or equal to the median value) and were classified into 4 groups, high EGFR WI and high E-cad RMC (HH), high EGFR WI and low E-cad RMC (HL), low EGFR WI and high E-cad RMC (LH), and low EGFR WI and low E-cad RMC (LL). (B) In the patients who had primary tumors with positive lymph nodes (Met Group), both EGFR and E-cad RMCs were categorized as high (greater than the median value) and low (less than or equal to the median value) and were classified into 4 groups, HH, HL, LH, and LL. Differences between these 4 groups and their correlation with DFS and OS were evaluated by Wilcoxon test.

Expression EGFR and E-cad in Metastatic SCCHN Cell Lines and Their Sensitivity to EGFR-TKI

We identified cell line models that represented 2 of the 3 patterns: expression of both EGFR and E-cad and expression of EGFR but no E-cad. Immunoblotting analyses indicated that PCI-37A cells and 686LN cells expressed both EGFR and E-cad but did not express the mesenchymal cell marker vimentin, whereas the cell lines PCI-37B and 686LN-M4e expressed vimentin but did not express E-cad and expressed less EGFR (Fig. 4A). In the E-cad-negative cells, EGFR was internalized, as illustrated by immunofluorescence staining (Fig. 4B). It is noteworthy that both PCI-37B cells and 686LN-M4e cells were less sensitive to EGFR-TKI erlotinib than their counterparts PCI-37A cells and 686LN cells (Fig. 4C), although p-EGFR was reduced in all the cell lines that were treated with erlotinib (Fig. 4A). An EGFR–TKI-resistant derivative of 686LN cells, 686LN-R30 cells, had similar sensitivity to erlotinib. Two-sample t tests indicated a significant difference in growth inhibition (P < .0001) between paired erlotinib-sensitive and erlotinib-insensitive cell lines (686LN cells vs 686LN-M4e cells or 686LN-R30 cells; PCI-37A cells vs PCI-37B cells). Furthermore, expression patterns of reduced E-cad expression and enhanced vimentin expression similar to 686LN-M4e cells suggested that these cells experienced EMT.

Figure 4.

Expression of E-cadherin (E-cad) and epidermal growth factor receptor (EGFR) in squamous cell carcinoma of the head and neck (SCCHN) cell lines and their responses to EGFR-tyrosine kinase inhibitor (TKI). (A) SCCHN cell lines PCI37A, PCI-37B, 686LN, 686LN-M4e, and 686LN-R30 were treated with erlotinib at concentrations of 0.5 μM and 10 μM for 72 hours and were lysed for protein isolation and immunoblotting analysis by using specific antibodies against the indicated proteins. Glyceraldehyde-3-phosphate dehydrogenase (G3PDH) served as a loading control (p-EGFR indicates phosphorylated EGFR). (B) Localization of EGFR was observed by immunofluorescence staining according to the protocol described in the text (see Materials and Methods). Solid arrows indicate membrane staining, and open arrows indicate cytoplasmic staining in EGFR–TKI-sensitive cells and EGFR–TKI-resistant cells, respectively. (C) SCCHN cell lines PCI-37A, PCI37B, 686LN, 686LN-R30, and 686LN-M4 were treated with erlotinib under a range of concentrations as indicated for 72 hours. Then, the cell density was analyzed by sulforhodamine B assay. The percent of surviving cells was calculated as the ratio of live cells treated at each concentration to the untreated cells and is presented as an average of 3 data points with standard deviation. All experiments were repeated at least twice.


Although both EGFR and E-cad are considered to contribute to metastasis of HNC, their coexpression has not been examined carefully, particularly in paired primary and metastatic SCCHN. In the current study, we identified a correlation between EGFR and E-cad and investigated whether this correlation has any biologic significance. To our knowledge, this is the first report to investigate colocalization of these 2 proteins in SCCHN cells using both human tissue specimens and cell lines. We were able to illustrate the progressive loss of E-cad expression from nonmetastatic PTs, to metastatic PTs, and finally to LNMs (Fig. 2A). In addition, we demonstrated that cytoplasmic distribution of both EGFR and E-cad, as indicated by reduced RMC values, were associated with LNM (Fig. 2B), which may serve as a predictor for metastasis of SCCHN.

Examining correlative expressions between EGFR and E-cad has attracted much attention in recent years, since 3 major publications reported that sensitivity to EGFR-TKI is correlated to E-cad expression in lung cancer.16–18 Although those publications did not provide mechanisms to explain their observations, they promoted the study of a correlation between EGFR and E-cad and their correlative biologic significance. In an in vitro study, we identified several EGFR–TKI-resistant cell lines and their EGFR–TKI-sensitive counterparts. One pair of cell lines, PCI-37A and PCI-37B, was established from a primary SCCHN and its LNM. Another pair of cell lines, 686LN and 686LN-R30, was established under constant selection pressure with gefitinib. It is noteworthy that the highly metastatic cell line 686LN-M4s, including 686LN-M4e, which were selected through in vivo selection of 686LN cells by using an SCCHN metastatic xenograft mouse model,24 had reduced sensitivity to EGFR-TKI similar to that observed in 686LN-R30 cells (Fig. 4C). Consistent with the observation in lung cancer, we demonstrated loss of E-cad in EGFR–TKI-resistant SCCHN cell lines regardless of their origins (Fig. 4A). To support our observation, similar findings were reported recently in other types of cancer, including colorectal and pancreatic cancers.27 Furthermore, although both EGFR–TKI-sensitive and EGFR–TKI-resistant SCCHN cell lines express EGFR, the expression level of this protein is reduced, and it is localized mainly in the cytoplasm in the EGFR–TKI-resistant cells compared with their EGFR–TKI-sensitive counterparts (Fig. 4B), suggesting that EGFR may no longer respond to extracellular signals, such as growth factors, and may not be a major support for cell proliferation in these cells. Several mechanisms have been considered to contribute to cancer cells resistant to EGFR-targeting therapy in HNC, including overexpression of EGFR ligands,12 increasing EGFR gene copy numbers,28, 29 overexpression of other members of the EGFR family,30 and an existing EGFRvIII mutation.31 Our study illustrates an additional possibility: that cytoplasmic localization of EGFR and not EGFR expression, along with reduction of E-cad, may explain the clinical findings of a poor response to EGFR-TKI and, particularly, to therapeutic antibody against the extracellular portion of EGFR. The surgical specimens that we used in this study were collected retrospectively in patients without prior chemotherapy. Further proof of this hypothesis will require tissue samples that are collected before and after treatment with EGFR-TKI, such as erlotinib.

Consistent with models in SCCHN cell lines, our investigation on 143 tissue specimens from 96 patients identified 3 major populations with distinct expression patterns of EGFR and E-cad. Among them, the population with EGFR overexpression and reduced E-cad expression accounted for 48% of the total samples. A similar percentage was reported in lung cancer by several researchers.17, 32 Like what has been reported by other studies in SCCHN, a significantly greater level of EGFR expression was observed in PTs with metastasis than in nonmetastatic tumors (Fig. 2B). Unlike previous studies, we examined the localization of EGFR and E-cad and observed that membrane localization of EGFR as a single marker was correlated significantly with poor DFS and OS for patients with or without metastatic SCCHN. Cell membrane distribution of EGFR is essential for response to circulating EGFR ligands. Therefore, it is expected that, under this condition, the EGFR signaling pathway may be activated fully to support tumor progression, rendering cancer cells more sensitive to EGFR-targeted therapy. This would not be surprising if the correlates obtained from specimens were at variance at least in part with the correlates obtained from cell lines in which potential responsiveness to EGFR-TKI was being considered. Because each of the cell lines was established from a population of tumor cells and had been cultured in vitro for a period of time, some genetic alterations may have caused further genotypic and phenotypic alterations, such as activation of another pathway, resulting in more variations in response to EGFR-TKI.

It is noteworthy that, in the current study, no other single marker that we examined had any significant prognostic value in our tissue sets. However, analyzing EGFR and E-cad coexpression did demonstrate significance with regard to patient survival. In patients with nonmetastatic tumors (Tu−Met), high expression of EGFR and low membrane localization of E-cad was associated significantly with shorter DFS than was observed in other patients (Fig. 3A). This coexpression pattern was reported previously in infiltrating ductal carcinomas of breast.33 Supporting the existence of this cancer cell population was an in vitro study from Lu et al, who demonstrated that chronic EGF treatment disrupted cell-to-cell adhesion, reduced the expression of E-cad, and caused EMT in human tumor cells that overexpressed EGFR.34 Because only a few patients died of disease in our nonmetastatic group, we did not identify any statistically significant correlation with OS by using combined WIs and RMCs.

Conversely, the metastatic tumors (Tu+Met) with high membrane localization of both EGFR and E-cad were associated with poor DFS and OS (Fig. 3B). We noted that the median value of the E-cad RMC in Tu+Met group already was significantly lower than that in the Tu−Met group (0.3 vs 0.5; P = .008) (Fig. 2A), so that high or low E-cad RMC in the Tu−Met and Tu+Met groups were not comparable. However, it is clear that tumors in the Tu−Met and Tu+Met groups had distinguishable biology in terms of EGFR andE-cad activities. This notion is supported by the observation that, in the Tu−Met group, the lowest rate of 5-year DFS was 50% and observed s in those patients with high EGFR WI and low E-cad RMC. In contrast, in the Tu+Met group, the lowest rates of 5-year DFS and OS were only 0% and 20%, respectively. This poor survival was observed in 31% of the patient population with metastatic tumors exhibiting relatively high RMCs for both EGFR and E-cad. Conversely, for patients with low RMCs for both EGFR and E-cad (35% of the population), the rates of 5-year DFS and OS also were low (40%), suggesting that these 2 populations, which had the majority of metastatic tumors, were distinguishable in terms of the mechanism of progression from the population with nonmetastatic tumors. This may account for our inability to identify a significant correlation when we combined the Tu−Met and Tu+Met groups for the same correlative survival analysis. Currently, the biologic mechanisms underlying these observations are under investigation.

A hypothesis of interaction between adhesion molecules like E-cad and EGFR was proposed several years ago.35 In our previous study, we illustrated that proteins that may regulate EMT, such as TGF-β receptors, integrin β1, and activated c-Src and p38α, were up-regulated in highly metastatic 686LN-M4s cells.24 Because of the loss of E-cad, β-catenin was internalized in 686LN-M4s cells, although it was colocalized on the tumor cell surface with E-cad in the parental cell line 686LN. Whether alterations of these proteins are involved in the regulation of EGFR currently is under investigation.

Our current results suggest that coexpression of E-cad and EGFR is present in SCCHN in unique patterns and that specific staining patterns are associated with a poorer outcome. Examining not only expression but also localization of these biomarkers simultaneously may have clinical relevance in predicting tumor behavior, including LNM, patient survival, and response to EGFR-targeted therapy.


We thank Dr. Anthea Hammond for her critical reading of the article.