The transcriptional repressor Delta EF1 controls the EMT phenotype and resistance to the EGFR inhibitor erlotinib in human head & neck squamous cell carcinoma lines

Abstract

Purpose

Although the epidermal growth factor receptor (EGFR) is overexpressed in a majority of HNSCCs, only a minority of patients derives substantial clinical benefit from EGFR inhibitors. We initiated the present study to identify the mechanisms underlying erlotinib resistance in a panel of HNSCC cell lines.

Methods

We used ³H-thymidine incorporation to characterize the heterogeneity of responsiveness to erlotinib-mediated growth inhibition in a panel of 27 human HNSCC cells. We characterized the molecular mechanisms involved in resistance using a representative subset of 6 erlotinib-sensitive and -resistant HNSCC lines.

Results

Erlotinib had heterogeneous effects on DNA synthesis in HNSCC cells that correlated closely with molecular markers of epithelial to mesenchymal transition (EMT). Specifically, the drug-sensitive lines expressed high levels of E-cadherin and demonstrated limited invasion and migration capabilities. In contrast, the erlotinib-resistant HNSCC lines expressed high levels of the E-cadherin repressor deltaEF1 (Zeb-1) and other mesenchymal markers and low levels of E-cadherin, and they were highly invasive and migratory. siRNA-mediated knockdown of deltaEF1 in the erlotinib-resistant cell lines (1386LN and UMSCC1) resulted in upregulation of E-cadherin and increased sensitivity to erlotinib in an E-cadherin-dependent manner.

Conclusions

DeltaEF1 controls the mesenchymal phenotype and drives erlotinib resistance in HNSCC cells. E-cadherin and deltaEF1 may prove to be useful markers in predicting EGFR inhibitor responsiveness.

Introduction

EGFR overexpression occurs in the vast majority (up to 90%) of HNSCC cases and has been correlated with poor prognosis, resistance to both chemotherapy and radiotherapy, increased risk of recurrence and metastasis and reduced overall and disease-free survival (1, 2). Aberrant EGFR signaling has been implicated in many HNSCC malignant features, including uncontrolled proliferation and cell cycle progression, resistance to apoptosis, invasion and metastasis (1). Erlotinib and gefitinib are two small molecule tyrosine kinase inhibitors that have been developed to specifically inhibit the EGFR in cancer (3). These drugs bind to and inhibit the EGFR tyrosine kinase thereby preventing initiation of downstream EGFR signal transduction pathways and causing p27-dependent cell cycle arrest (3, 4). The introduction of EGFR tyrosine kinase inhibitors in the clinic was greeted with tremendous enthusiasm, but they produced modest activity as single agents with response rates hovering between 5 − 15% (2).

One potential explanation for these disappointing results is that ineffective patient selection strategies were used to identify those HNSCC tumors that are actually dependent upon EGFR signaling for their growth and/or survival. Studies in NSCLC have identified several features that appear to distinguish EGFR-dependent and -independent tumors that may help to prospectively identify the subset of patients who are most likely to benefit from EGFR-directed therapy (5). These include activating point mutations within the EGFR tyrosine kinase domain and EGFR gene amplification (present in drug-sensitive tumors) or the presence of mutant K-ras or loss of PTEN (associated with drug resistance) (5). However, other studies have concluded that these features do not account for all EGFR sensitivity and resistance nor does the level of EGFR expression, which served as a basis for patient selection in most of the clinical trials performed with EGFR inhibitors to date (6). These observations have prompted new studies aimed at obtaining a better understanding of the biology of EGFR dependency in NSCLC and other solid tumors.

Epithelial-to-mesenchymal transition (EMT) is a process that plays important roles in normal organ development and in cancer progression (7). EMT is characterized by the combined loss of E-cadherin and the gain of mesenchymal markers such as fibronectin or vimentin and increased invasion and migration (7). Delta-crystalin enhancer-binder factor1 (deltaEF1), also known as TCF-8 or Zeb-1, is one of the transcription factors that has been implicated in EMT in tumor cells (7). It is characterized by the presence of two zinc-finger clusters located within its C and N termini and a central homeobox domain (7). DeltaEF1 regulates E-cadherin transcription by simultaneous binding of the two zinc-finger domains to two high affinity binding sites (known as E-boxes) located in the E-cadherin promoter region (7), and ectopic expression of deltaEF1 has been shown to be sufficient to down regulate E-cadherin expression and induce EMT in normal mammary epithelial cells (8).

Here we report that resistance to the EGFR inhibitor erlotinib is linked to EMT in human HNSCC cells. We show that this EMT is driven by deltaEF1 and can be reversed by deltaEF1 knockdown. The results may have important implications for patient selection into EGFR inhibitor trials and identify deltaEF1 as a new therapeutic target in HNSCC.

Materials and Methods

Reagents, cell lines and culture conditions

Erlotinib was obtained from the M.D. Anderson Pharmacy. Stock solutions were prepared in DMSO and stored in −20°C. Small interfering RNA (siRNA) targeting endogenous DeltaEF1 and an off-target control construct (targeting luciferase) were purchased from Dharmacon (Lafayette, CO). Primary antibodies used for immunoblotting were from the following sources: anti-EGFR,anti-p-EGFR(y1086) were obtained from Invitrogen (Carlsband,CA); anti-Zeb-2/SIP1, anti-vimentin, anti-deltaEF1 and anti-fibronectin were obtained from Santa Cruz Biotechnology (Santa Cruz, CA); anti-actin was from Sigma ( St.Louis, MO); anti-p27 was purchased from BD Biosciences Pharmingen (San Diego, CA); anti-E-cadherin was obtained from Zymed (Invitrogen, Carlsbad, CA); anti-lamin B was from Oncogene/ Calbiochem (San Diego, CA); and anti-SNAIL was obtained from Abcam (Cambridge,MA). Secondary antibodies were obtained from Amersham Biosciences (Piscataway,NJ). Dr.Gary Clayman (Department of Head and Neck Surgery, University of Texas M.D. Anderson Cancer Center, Houston, Texas) provided us with TU138, MDA183, TU358B, UMSCC1, MDA1386LN, TU158LN, TU212LN, MDA686TU, MDA686LN, T404, TU167, TU177, T409, MDA886LN,TU182, MDA1483, TU159, and MDA1986LN. Dr.Reuben Lotan (Department of Head & Neck Medical Oncology, U.T. M.D. Anderson Cancer Center) provided us with14A, 14B, 17A, 17B, 22A, 22B, SQCCY1.

Dr.Jeffrey Myers (Department of Head and Neck Surgery, University of Texas M.D. Anderson Cancer Center, Houston, Texas) provided us with the following cell lines: TU167LN, JMAR. All were maintained in DMEM/F-12 Cellgro Mediatech (Herndon,VA) cell culture medium containing 10% fetal bovine syrum (FBS;Life Technologies), sodium pyruvate (BioWhittaker,Rockland,ME), L-glutamine (Bio Whittaker), nonessential amino acids (Life Technologies), Vitamins (Life Technologies) and antibiotics ( penicillin/streptomycin;Bio Whittaker). Adherent monolayer cultures were incubated at 37°C in a mixture of 5% CO₂ in air.

[³H] -thymidine incorporation assays

Cells were plated in 96-well plates at a density of 1×10⁴ per well. Cells were exposed 24 h later to various concentrations of erlotinib (0.01−10 μM) in serum-free DMEM. After 24 hours, the medium was removed and replaced with fresh DMEM containing 10% FBS and 10μCi/mL [³H] thymidine (Amersham Biosciences, Piscataway, NJ). The cells were pulsed with [³H] thymidine for 1 h, lysed by the addition of 0.1N KOH, and harvested onto fiberglass filters. The incorporated tritium was quantified in a scintillation counter.

Real-time quantitative PCR

For RNA extraction, cells were allowed to reach 70% confluence, medium was removed and cells were immediately lysed in Tri-Reagent (Sigma). Total RNA was prepared following the manufacturer's instructions and further purified using Rneasy Mini Kit (Qiagen, Hilden,Germany). For the real time PCR, theTagMan-system (Applied Biosystems, Foster City, CA) with assays-on-demand (Applied Biosystems) was used following the manufacturer's instructions. Assay ID numbers: DeltaEF1: Hs00232783-m1, E-cadherin: Hs00170423-m1, vimentin: Hs00185584-m1, fibronectin: Hs00415006-m1, cyclophilin: Hs00170423-m1. The relative expression of deltaEF1, E-cadherin, vimentin and fibronectin were normalized to the amount of cyclophilin A in the same cDNA using the comparative CT method.

Immunoblot analyses

Cells were scraped from tissue culture plates with a rubber policeman into Triton lysis buffer [1% Triton X-100, 150 mmol/L NaCl, 25 mmol/L Tris (pH 7.5), 1 mmol/L glycerol phosphate, 1 mmol/L sodium orthovanadate, 1 mmol/L sodium fluoride, and one Complete Mini protease Inhibitor Cocktail tablet (Roche, Indianapolis, IN)] and were incubated for 1 hour on ice. The lysates were then clarified by centrifugation before resolution by 10% SDS-PAGE. Polypeptides were transferred to nitrocellulose membranes. Membranes were blocked in 5% nonfat milk in a TBS containing 0.1% Tween 20 (TBS-T) for 2 hours at 4°C, incubated overnight with relevant antibodies, washed, and probed with species-specific secondary antibodies coupled to horseradish peroxidase (Amersham Biosciences), and detected by enhanced chemiluminescence (Renaissance, New England Nuclear, Boston, MA).

Cell cycle analysis

Cells were grown in six-well plates in the presence of 10% DMEM. After reaching 70% confluence, the cells were exposed to various concentrations of erlotinib for 24 hours. Cells were harvested by trypsinization and pelleted by centrifugation. The pellets were then resuspended in PBS containing 50 μg/mL propidium iodide, 0.1% Triton X-100, and 0.1% sodium citrate. Samples were stored at 4°C for 1 hour before analysis. Propidium iodide fluorescence was measured by fluorescence-activated cell sorting (FACS) analysis (FACScan FL-3 channel; Becton (Dickinson, Mountain View, CA). Percentages of cell populations in different stages of the cell cycle were documented.

Wound healing assay and invasion assay

Cell motility and migration were determined by measuring the movement of cells to close an artificial wound created by 10-μL pipette tip (time 0). The speed of wound closure was monitored by phase contrast microscopy at 0 h, 12 h, and 24 h time points. For invasion assays, invasion chambers containing Matrigel- coated polyethylene terephthalate membranes with 8-μm pores in a 24-well plate format were purchased from BD Biosciences. Cells (5×10⁵) were released from their tissue culture flask by EDTA (1 mmol/L), centrifuged, resuspended in DMEM, and placed in the upper compartment of the invasion chamber. Each cell line was plated in duplicate. NIH-3T3- conditioned medium was placed in the lower compartment as a chemoattractant. After incubation for 24 hours at 37°C, a cotton swap was utilized to scrape and remove cells from the upper surface of the membrane. Cells that had invaded and adhered to the bottom of the Matrigel membrane were stained with crystal violet and visualized utilizing an inverted microscope at 10× magnification. The number of invading cells was counted in 5 separate fields per membrane.

Small interfering RNA- mediated knockdown of deltaEF1

MDA1386LN and UMSCC1 cells were grown in six-well plates until they reached 60% confluence and then they were transfected with small interfering RNA (siRNA) targeting endogenous deltaEF1 or E-cadherin or both deltaEF1 and E-cadherin or a non-targeting negative control using oligofectamine (Invitrogen). All the siRNA constructs were obtained from Dharmacon Research, Inc (Lafayette CO). 60 hours after transfection, total protein lysate was collected and analyzed by Western blotting or total RNA was extracted and analyzed by real-time PCR. In a parallel experiments, 60 hours post transfection, siRNA transfected cells were either trypsinized and plated in 96 well plates for [³H]thymidine incorporation assays, or subjected to EDTA and plated in invasion chambers for invasion assays or allowed to reach 100% confluence for wound healing assays.

CHIP assay

Cells were allowed to reach 70−80% confluence then they were processed for CHIP analyses using the CHIP-IT express kit (Active Motif, Carlsbad, CA) following manufacturer's protocol. PCR reactions were performed with the following primer: Human E-cadherin promoter: forward 5′-AACTCCA GGCTAGAGGGTCA-3′, reverse 5′-GGGCTGGAGTCTGAACTGA-3′.

Statistical analysis

All assays were performed at least 3 independent times. Numerical data are presented as mean ± SD. Comparisons between test and control groups were evaluated using Student's t test. Statistical significance was set at P< 0.05.

RESULTS

Heterogeneous effects of erlotinib on DNA synthesis in human HNSCC cells

To determine their relative sensitivities to EGFR inhibition, we exposed a panel of 27 human HNSCC cells to increasing concentrations of erlotinib and measured DNA synthesis (as a marker of cell cycle progression) by ³H-thymidine incorporation. The lines displayed marked heterogeneity in responsiveness (Fig. 1), consistent with data we and others have obtained in HNSCC and other solid tumors (9-13). Based on their marked differences in erlotinib responsiveness we selected three representative drug-sensitive cell lines (TU138, MDA183, and TU358B) and three representative drug-resistant cell lines (MDA1386LN, UMSCC1, and TU167LN) for further analysis.

Figure 1.

Heterogeneous growth inhibitory effects of erlotinib in 27 HNSCC cell lines. Cells that displayed more than 50% growth inhibition at 1 μM of erlotinib were arbitrarily classified as “sensitive”, and cells with less than 50% growth inhibition at 1 μM of erlotinib were identified as resistant.

In order to obtain an independent confirmation of the ³H-thymidine incorporation results, we used propidium iodide staining and fluorescence activated cell sorting analyses (PIFACS) to measure the percentages of cell populations in different stages of the cell cycle after exposure to increasing concentrations of erlotinib. Because cyclin-dependent kinase inhibitors have been implicated in growth arrest (4, 11), we also used immunoblotting to determine the effects of erlotinib on the expression of the cyclin-dependent kinase inhibitor p27 in the subset of six HNSCC cell lines. The three sensitive cell lines exhibited significantly-induced increases in the percentages of cells that accumulated within the G1 phase of the cell cycle, with corresponding significant decreases in the percentages of cells found in S phase (Table I). On the other hand, the three resistant cell lines showed minimal response with respect to cell cycle progression following exposure to increasing erlotinib concentrations (Table 1). Consistent with these observations, 1 μM erlotinib increased expression of p27 in the sensitive cell lines after 24 h exposure, but erlotinib induced no change in p27 expression in the 3 resistant cell lines (Figure 2A).

Table 1.

Effects of Erlotinib on Cell Cycle Distribution in HNSCC cell lines^*

SENSITIVE	%G1±SD	%S±SD	RESISTANT	%G1±SD	%S±SD
TU138			1386LN
Control	54±9.7	30±6.8	Control	60±4.2	26±6.7
Erl, 10 Nm	57±8.3	27±10.5	Erl, 10nM	64±2.05	24±1.2
Erl, 100 nM	66±8.6	19±5.9	Erl, 100nM	63±7.3	24±7.1
Erl, 1 uM	73±11.5	14±4.8	Erl, 1uM	61±5.6	25±6.8
MDA183			UMSCC1
Control	57±5.8	27±2.7	Control	58±5.7	27±3.1
Erl, 10nM	64±6.9	20±4.1	Erl, 10nM	58±2.2	26±1.9
Erl, 100nM	72±8.2	13±6.2	Erl, 100nM	61±6.8	26±2.0
Erl, 1 uM	76±3.1	8±3.3	Erl, 1uM	62±7.1	26±2.6
358B			167LN
Control	58±5.4	24±3.6	Control	70±12.3	16±2.2
Erl, 10nM	68±10.2	23±9.8	Erl, 10nM	70±2.4	14±3.6
Erl, 100nM	71±2.9	12±5.4	Erl, 100nM	73±7.1	10±1.8
Erl, 1 uM	76±3.1	8 ±4.0	Erl, 1uM	72±6.8	14±4.5

^*Cells were allowed to reach 70% confluence, and then they were exposed to various concentrations of erlotinib (1μM, 10nM, 100nM) for 24hours followed by propidium iodide staining for 1hour at 4°C. Propidium iodide fluorescence was measured utilizing fluorescence activated cell sorting (FACS) analysis. Percentages of cell populations in different stages of the cell cycle were documented. Mean values from three independent experiments are shown (± Standard Deviation).

Figure 2.

Effects of erlotinib on (p27, EGFR and p-EGFR) expression in HNSCC cell lines. A.Western blot analysis showed upregulation of p27 in the three sensitive cell lines (TU138, MDA183 and TU358B) after 24h exposure to 1 μM of erlotinb. While the three resistant cell lines (MDA1386LN, UMSCC1 and TU167LN) exhibited no change in p27 expression levels after 24h exposure to 1 μM of erlotinib. The expression of actin is shown as a loading control. B. Status of baseline (EGFR and p-EGFR) in erlotinib-sensitive and erlotinib-resistant cell lines. Note that erlotinib inhibited baseline phosphorylation of EGFR in all of the erlotinib-sensitive cell lines but had no effect on the baseline phosphorylation of EGFR in the erlotinib-resistant cell lines.

Baseline EGFR and p-EGFR expression does not correlate with response to erlotinib

To identify molecular markers predicting erlotinib sensitivity, we measured the expression of the erlotinib target protein EGFR as well as its active (phosphorylated) form (p-EGFR) in erlotinib sensitive and erlotinib resistant cells at baseline and after erlotinib exposure by immunoblotting. Although baseline levels of total and phosphorylated EGFR were not related to erlotinib responsiveness, erlotinib induced a concentration-dependent reduction in EGFR phosphorylation in all of the sensitive lines but had no effects in any of the resistant lines (Figure 2B). These results demonstrate that EGFR phosphorylation is uncoupled from EGFR tyrosine kinase activity in the erlotinib-resistant cells, consistent with our previous experience in human bladder and pancreatic cancer cell lines (Shrader et al, MCT 2007; Pino et al, Cancer Res. 2006).

Erlotinib-resistant cells display features of EMT

Recent studies have demonstrated that EMT correlates with EGFR inhibitor resistance in diverse solid tumors, including HNSCC (9-11, 13). Therefore, we used RT-PCR and immunoblottting to characterize the expression of E-cadherin, vimentin, fibronectin, and deltaEF1 in the panel of six HNSCC cell lines. Consistent with the previous studies (9), E-cadherin levels were high and the levels of vimentin, fibronectin, and deltaEF1 were low in the erlotinib-sensitive cells (Figure 3). Conversely, E-cadherin levels were low, while fibronectin and deltaEF1 levels were high in the erlotinib-resistant cells (Figure 3).

Figure 3.

Expression of EMT markers in HNSCC cells. mRNA expression levels of E-cadherin, vimentin, fibronectin and DeltaEF1 were measured by quantitative real-time RT-PCR, and protein expression was measured by immunoblotting. The expression of actin is shown as a loading control. A. E-cadherin expression. B. Vimentin expression. C. Fibronectin expression. D. DeltaEF1 expression. The expression of lamin B is shown as loading control of the nuclear fraction in the panel of HNSCC cell lines.

Loss of intercellular adhesion and downregulation of E-cadherin have been associated with enhanced motility and increased invasive and migratory potential (7). We therefore performed wound healing assays in order to compare the migratory capabilities of the 6 HNSCC lines in the presence and absence of erlotinib. When cells reached 100% confluence, we created a scratch with a pipette tip and used light microscopy to monitor wound closure at 0, 12, and 24-hour time points. Resistant cell lines migrated faster to close the wound than did the sensitive cell lines (Figure 4A). The most resistant cell line (MDA1386LN) exhibited the greatest migratory capability. It is important to note that MDA1386LN has the longest doubling time among the 6 HNSCC cell lines, indicating that MDA1386LN is able to close the wound quickly due to its advanced motile and migratory potential, and not due to its proliferation rate. Erlotinib had no effect on rate of wound closure in the “mesenchymal” cell lines (Figure 4), indicating that cells that are resistant to the growth-inhibitory effects of the drug are also resistant to its anti-migratory effects. Conversely, the lines that were sensitive to the growth inhibitory effects of erlotinib also displayed delayed wound closure when they were exposed to the drug (Figure 4), confirming that drug sensitivity was not limited to cell cycle inhibition. We then used Boyden chamber invasion assays to confirm and extend our findings with the wound healing assays and to determine the relative invasive capabilities of both sensitive and resistant cell lines. Consistent with their mesenchymal phenotypes, the resistant cell lines exhibited greater invasive potential than the sensitive cell lines did. Furthermore, erlotinib did not affect the invasive capabilities of resistant cell lines, but it strongly inhibited invasion in the drug-sensitive lines (Figure 4B). Together, these results strongly suggest that erlotinib's anti-metastatic effects are most prominent in HNSCC cells that are sensitive to erlotinib-induced growth arrest and display an “epithelial” molecular phenotype.

Figure 4.

Wound healing and invasion assays in resistant versus sensitive HNSCC cell lines. A. Wound healing assays. Assays were performed as described in Materials and Methods. Note that 24h exposure to 1 μM of erlotinib had no influence on the migration rates of resistant cell lines but strongly decreased the migration rates of the sensitive cell lines. B. Invasion assays. Modified Boyden chamber assays were performed as described in Materials and Methods. Note that erlotinib had no effect on invasion in the resistant lines but caused significant reductions in invasion in the sensitive lines.

DeltaEF1 induces EMT and drives resistance to Erlotinib

The correlation between deltaEF1 expression and erlotinib resistance suggested that a cause-effect relationship might exist between the two. To test this possibility, we knocked down deltaEF1 expression in two erlotinib-resistant lines (MDA1386LN and UMSCC1) and examined the effects on E-cadherin expression and erlotinib sensitivity. Immunoblotting and real time PCR confirmed that silencing resulted in >50% knockdown of deltaEF1 and increased expression of E-cadherin (Fig 5A) and increased cellular sensitivity to erlotinib as measured by inhibition of ³H-thymidine incorporation (Fig 5B). Combined knockdown of deltaEF1 plus E-cadherin reversed the erlotinib sensitization observed with knockdown of deltaEF1 alone (Fig 5B), demonstrating that erlotinib sensitization required E-cadherin reexpression. Furthermore, deltaEF1 knockdown dramatically decreased migration and invasion in 1386LN cell line (Fig 5C, 5D). Together, these data establish that deltaEF1 is a critical component of the molecular mechanisms that drive EMT and EGFR inhibitor insensitivity in HNSCC cells.

Figure 5.

DeltaEF1 maintains the mesenchymal phenotype in HNSCC cells. A. Effects of silencing deltaEF1 on deltaEF1 and E-cadherin mRNA levels in (1386LN, UMSCC1) cell lines. Silencing and real-time PCR were performed as described in Materials and Methods. In addition, Immunoblotting was performed in 1386LN cell line. Expression of actin served as a loading control. B. Effects of silencing deltaEF1 with or without E-cadherin silencing on erlotinib sensitivity. ³H-thymidine incorporation assays were used to measure the effects of erlotinib on DNA synthesis in (1386LN, UMSCC1) cells. Note that deltaEF1 knockdown restored sensitivity to the drug, and these effects were reversed by combined silencing of deltaEF1 and E-cadherin. C. Effects of deltaEF1 knockdown in on cell migration. 1386LN cells were transfected with siRNAs targeting deltaEF1 or an off-target control for 48 h, and wound healing assays were performed as described in Materials and Methods. Note that silencing decreased migration and that migration was further reduced by exposure to 1μM erlotinib. D. Effects of knockdown on invasion. 1386LN cells were transfected with siRNAs targeting deltaEF1 or an off-target control and then analyzed in modified Boyden chambers as described in Materials and Methods. Note that similar effects of silencing and erlotinib were observed in the invasion and migration assays.

DeltaEF1 associates with the endogenous E-cadherin promoter

To determine whether deltaEF1 interacts directly with the endogenous E-cadherin promoter at the chromatin level, we performed chromatin immunoprecipitation (CHIP) experiments. A deltaEF1 antibody (directed against its amino-terminus) efficiently pulled down deltaEF1 together with chromatin fragments comprising the E-cadherin proximal regulatory promoter region in the resistant cell lines, whereas the antibody did not precipitate E-cadherin promoter fragments in the sensitive cell lines (Fig 6). These results establish that deltaEF1 interacts directly with the E-cadherin proximal promoter region in erlotinib-resistant cells. Together with the silencing data they establish that deltaEF1 is responsible for suppressing E-cadherin expression and maintaining erlotinib resistance in the mesenchymal HNSCC cells.

Figure 6.

DeltaEF1 interacts with the E-cadherin promoter in erlotinib-resistant cells. Chromatin immunoprecipitation assays were performed using an anti-deltaEF1 antibody and primers designed to amplify the region containing the E-boxes present in the E-cadherin promoter. Note that deltaEF1 interacts with the E-cadherin promoter in all of the erlotinib-resistant lines tested but does not interact with the promoter in the erlotinib-sensitive cells. Also note that other candidate regulators of E-cadherin promoter activity and EMT (Snail, Zeb-2) were not detected on chromatin in these assays.

Discussion

The results of clinical trials employing EGFR inhibitors in diverse solid tumors indicate that tumors display remarkable heterogeneity in drug responsiveness that does not correlate well with target (EGFR) expression (2, 5, 6, 14). Here we employed a panel of 27 human cell lines to obtain a sense of the heterogeneity in EGFR dependency that exists in HNSCC and to attempt to identify biological markers of drug sensitivity and resistance that might prove useful in distinguishing drug-sensitive from drug-resistant tumors. Our data demonstrate that cells that are sensitive to the EGFR inhibitor erlotinib express high levels of E-cadherin, low levels of deltaEF1, and display low invasive and migratory potential, whereas drug-resistant cells express high levels of deltaEF1 and low levels of E-cadherin and are highly invasive and migratory. Knockdown of deltaEF1 reversed the EMT phenotype associated with erlotinib resistance and restored erlotinib sensitivity in “mesenchymal” cell lines (MDA1386LN and UMSCC1), demonstrating that deltaEF1 plays causative roles in both processes. The effects of deltaEF1 knockdown on erlotinib sensitivity required E-cadherin, because combined knockdown of deltaEF1 plus E-cadherin reversed the erlotinib sensitization observed when deltaEF1 was silenced alone. Our data confirm and extend other recent findings in HNSCC (9) and other solid tumors (15), where EMT and more specifically deltaEF1 have been shown to correlate with resistance to EGFR inhibitors. All of these studies strongly suggest that quantification of E-cadherin and deltaEF1 levels in pretreatment biopsies might allow for the prospective identification of solid tumors that are most likely to be sensitive to EGFR-directed therapy. This hypothesis must now be tested in well-designed clinical trials employing pharmacodynamic markers that are capable of measuring drug-induced cell cycle arrest (p27 accumulation, decreases in Ki-67 or PCNA) and possible inter- and intra-tumoral heterogeneity in the molecular EMT phenotype.

Other recent studies have demonstrated that, like deltaEF1, E-cadherin also plays a causative role in determining EGFR inhibitor sensitivity (16). E-cadherin is known to form complexes with erbB family members that modulate downstream signal transduction (17-20), and it is also possible that the homotypic adhesion mediated by E-cadherin facilitates interactions between the EGFR and surface-tethered ligands (HB-EGF, TGF-α) expressed on neighboring cells. Preclinical studies have demonstrated that knockdown of E-cadherin decreases EGFR inhibitor sensitivity (10); conversely, forced expression of E-cadherin in EGFR-independent cells can restore EGFR inhibitor sensitivity (16). These observations are somewhat inconsistent with previous work showing that the EGFR can drive EMT, invasion, migration, and metastasis (21, 22). It is possible that these observations can be reconciled if the EGFR can interact with other (mesenchymal) growth factor receptors to drive tumor progression in a manner that is insensitive to the currently available EGFR antagonists. This possibility is consistent with our observation that the constitutive EGFR phosphorylation that is observed in mesenchymal HNSCC cells is refractory to erlotinib (Fig.2B). Another important question that remains unresolved concerns how deltaEF1 expression is regulated in HNSCC cells. One attractive possibility is that deltaEF1 expression is increased by cytokines and other signals, including the TGF-β, Shh, Notch, COX-2, and NFκB pathways, that have been implicated in driving EMT in other models (7, 23). However, recent work has demonstrated that members of the miR-200 family of micro RNAs (miRNAs), control deltaEF1 expression and the EMT phenotype in the “NCI-60” and other model systems (24-27). Specifically, the transcripts encoding deltaEF1 (Zeb-1) and its homolog Zeb-2 (Sip-1) contain multiple binding sites for these miRNAs, and forced expression of members of the miR-200 family directly bind to the deltaEF1 and Zeb-2 transcripts and decrease translation of their protein targets (24-27). Furthermore, profiling studies have demonstrated that miR-200 family miRNAs are lost in cell lines that display high levels of deltaEF1 and Zeb-2, low levels of E-cadherin, and a “mesenchymal” phenotype (24). We have found that expression of miR-200c inversely correlates with deltaEF1 expression in bladder and pancreatic cancer cells, and reintroduction of the miRNA restores an epithelial phenotype and EGFR inhibitor sensitivity in “mesenchymal” bladder cancer cells (L, Adam et al, T. Arumugam et al, manuscripts submitted). We measured miR-200c expression in the subset of 6 HNSCC lines characterized here and found that it is expressed at low levels in the most mesenchymal line (MDA1386LN) but not in the UMSCC1 or TU167LN cells, even though the latter are also erlotinib-resistant, highly migratory, and express high levels of fibronectin (data not shown). We are currently expanding these studies to include our whole panel of HNSCC lines and additional members of the miR-200 family, but these preliminary results suggest that other mechanisms may be involved as well.

DeltaEF1 represses gene expression by serving as a scaffold for recruitment of HDACs to chromatin (7). This observation provides an immediate means of translating our findings into clinical application. HDAC inhibitors are being developed aggressively for cancer therapy in large part because early studies demonstrated that they could promote differentiation in erythroid leukemia cells (28), a process that is reminiscent of EMT reversal. Vorinostat (SAHA) is the first of this class of agents to receive FDA approval for cancer treatment (for the therapy of cutaneous T-cell lymphoma) (29-31). Although it displays little (if any) single agent activity in solid tumors, preclinical studies have demonstrated that MS-275 is capable of reversing EMT and restoring sensitivity to EGFR inhibitors in NSCLC cells (16), and our own results in pancreatic cancer cells confirm these observations (K. Fournier, W. Choi, manuscript in preparation). Importantly, the EMT reversal induced by vorinostat in vitro or in vivo is relatively long lasting (K. Fournier, manuscript in preparation), suggesting that proper scheduling might maximize the beneficial effects of the drug in solid tumors while avoiding some of the toxic effects that might be caused by combination therapy. We are currently characterizing the effects of HDAC inhibitors on sensitivity to erlotinib and other drugs in preclinical in vitro and in vivo models of HNSCC to determine whether they can be used to restore sensitivity to EGFR inhibitors and other agents.