Hypoxia Alters the Response to Anti-EGFR Therapy by Regulating EGFR Expression and Downstream Signaling in a DNA Methylation–Specific and HIF-Dependent Manner

Increased cell proliferation and oxygen consumption result in lower oxygen availability in solid tumors as compared with normal tissue (1, 2). Intratumoral hypoxia has been associated with invasion, metastasis, treatment failure, and patient mortality (3, 4). In murine models of metastasis, cells exposed to hypoxia in the primary tumor were able to metastasize five times more readily than their oxygenated counterpart (5). Cancer cells survive and adapt to hypoxic conditions, in part, through the activation of hypoxia-inducible factor 1 (HIF1) and HIF2, which induce the expression of gene products involved in angiogenesis, glucose utilization, invasion, and metastasis (6). HIF1 is a heterodimeric protein composed of a constitutively expressed HIF1β subunit and an O₂-regulated HIF1α subunit (7). Our recent work suggests that tumors may have a unique transcriptional response to hypoxia with a select number of conserved genes that are induced or repressed across 31 individual cell lines (8). We selected the EGFR, which was induced in seven of 31 cell lines under hypoxic conditions to determine the mechanisms and potential clinical implications of the heterogeneity in the hypoxic response.

The EGFR is a member of the ErbB (avian erythroblastosis oncogene B) family of receptors and activates multiple signaling pathways, including MAPK/ERK and PI3K/V-AKT murine thymoma viral oncogene homolog (AKT) pathways (9, 10). The activation of EGFR has many implications in tumor biology, such as cell proliferation, invasion, metastasis, and apoptosis (11, 12). EGFR is overexpressed in various human cancers, including lung cancer, breast cancer, colon cancer, and glioblastoma, and is associated with tumor malignancy and poor prognosis (13, 14). Approximately half of the cases of triple-negative breast cancer (TNBC) and inflammatory breast cancer present with the overexpression of EGFR (11). Several studies have shown an inverse correlation between EGFR expression and disease-free and overall survival of patients with breast cancer (14, 15). Taken together, these findings have prompted the evaluation of EGFR inhibitors for the treatment of TNBC (16). However, the results of such studies in breast cancer treatment have been disappointing (16–18), partially due to the lack of biomarkers to predict which patients are most likely to respond to treatment with EGFR inhibitors (18).

Under normal circumstances, EGFR expression is primarily regulated by the abundance of its mRNA (19). EGFR gene amplification is a common mechanism of overexpression in high-grade gliomas (20), but it is less common in other solid tumors (21). A recent study of non–small cell lung cancer (NSCLC) found that only 6% of primary NSCLC tumors have gene amplification of EGFR (22). Epigenetic regulation is a biological mechanism by which gene expression is modulated through DNA methylation or histone modifications (23). DNA methylation of cytosine at CpG dinucleotides is an important and well-studied regulatory modification throughout the genome (24). Hypermethylation of the promoter region of EGFR has been described in several types of cancer and alters EGFR expression (25, 26). Whether and how hypomethylation of EGFR can alter gene expression has not been previously considered.

Here, we demonstrate that EGFR is induced under hypoxic conditions. We uncover a functional hypoxia response element (HRE) that is activated upon HIF1 binding to an intron region of the EGFR gene. In normal breast tissue, intron 18 of EGFR is methylated, which prevents EGFR induction. The treatment of cells with a DNA methyltransferase inhibitor causes the demethylation of intron 18 of EGFR, thereby restoring the hypoxic regulation of EGFR. In cancer tissue and cancer cell lines with nonmethylated EGFR, hypoxia leads to AKT, ERK, and Rb phosphorylation, as well as induction of cyclin D1 and repression of p21 in an HIF1α-dependent manner, resulting in cell proliferation and migration. On the other hand, increased levels of EGFR under hypoxia enhance the efficacy of EGFR inhibitors. Taken together, our data suggest that EGFR inhibitors, in combination with methyl transferase inhibitors or in a subset of patients with hypomethylated EGFR, may have a therapeutic benefit for patients with hypoxic tumors.

All cell lines, with the exception of SUMs, were obtained from the ATCC. SUMs were purchased from Asterand Bioscience. Cells were cultured as per the manufacturer-provided protocols. The MCF10A- and MCF10A ER-expressing cells were a kind gift from Ben Ho Park (Vanderbilt University Medical Center, Nashville, TN) and were cultured as described previously (27, 28). CRISPR-edited MCF7 HIF1α, HIF2α, and control knockout cell lines were generated previously in our laboratory (8). All cell lines used in the study were authenticated by short tandem repeat sequencing and confirmed to be Mycoplasma free. Cells were maintained in a humidified environment at 37°C and 5% CO₂ during culture and live cell imaging. Hypoxic cells were maintained at 37°C in an in vivo 200 hypoxia workstation equipped with a digitally controlled oxygen regulator and maintained at 1% O₂, 5% CO₂, and 94% N₂. Live-cell microscopy experiments were conducted in a McCoy Incubator maintained at 1% O₂, 5% CO₂, and 94% N₂ and imaged with a Lionheart Microscope (BioTek).

Total RNA was extracted from cells using a Direct-zol RNA Miniprep kit with DNase I treatment as per the manufacturer's instructions (Zymo Research). One microgram of total RNA was used for first-strand DNA synthesis with the iScript cDNA Synthesis System (Bio-Rad). qPCR was performed using human-specific primers and iTaq SYBR Green Universal Master Mix (Bio-Rad). The expression of each target mRNA relative to 18s rRNA was calculated on the basis of the threshold cycle (C_t) as 2^−Δ(ΔC_t), where ΔC_t = C_t,target – C_t,18s and Δ(ΔC_t) = ΔC_t,test − ΔC_t,control. Primer sequences are presented in Supplementary Table S1.

Aliquots of whole-cell lysates were prepared in NP-40 buffer (150 mmol/L NaCl, 1% NP-40, and 50 mmol/L Tris-HCl, pH 9.0) and fractionated by 10% or 12.5% SDS-PAGE. Proteins were transferred from the gel to a nitrocellulose membrane for 15 minutes using a Trans-blot Turbo (Bio-Rad). The nitrocellulose membrane was blocked in 5% milk (w/v) in TBS and 0.1% Tween-20 (TBS-T) for 30 minutes. Antibodies against EGFR (1:1,000, Proteintech, 18986-1-AP), phosphorylated EGFR Y1068 (1:500, Sigma, SAB4300063), HIF1α (1:500, BD Biosciences, 610958), phosphorylated Erk T202/Y204 (1:1,000, Cell Signaling Technology, 9106S), Erk (1:1,000, Cell Signaling Technology, 4695S), phosphorylated Akt S473 (1:1,000, Cell Signaling Technology, 9271L), Akt (1:1,000, Cell Signaling Technology, 9272S), phosphorylated Rb S780 (1:500, Cell Signaling Technology, 9307S), Rb (1:500, Cell Signaling Technology, 9309S), cyclin D1 (1:1,000, Cell Signaling Technology, 2922S), cyclin A (1:500, Santa Cruz Biotechnology, Sc-271682), cyclin E (1:500, Cell Signaling Technology, 4129S), cyclin B1 (1:500, Santa Cruz Biotechnology, Sc-245), C-myc (1:500, Cell Signaling Technology, 13987S), and p21 Waf1/Clip1 (1:500, Cell Signaling Technology, 2947S) were used with overnight incubation at 4°C with orbital shaking. Blots were washed three times with TBS-T. β-actin-HRP (1:10,000, Proteintech, HRP-60008), secondary anti-mouse-HRP (Azure Biosystems, AC2115), and anti-rabbit-HRP (Azure Biosystems, AC2114) were then utilized with 1.5-hour incubation at room temperature with orbital shaking following by three additional TBS-T washes. Enhanced Chemiluminescent Substrate (PerkinElmer) was utilized as the substrate for horseradish peroxidase (HRP)-catalyzed detection and imaged using a c300 Imager (Azure Biosystems).

The Cancer Genome Atlas (TCGA) breast cancer (BRCA) transcriptional data and clinical data were downloaded from the NIH GDC Data Portal. The transcriptional data were quantile normalized before analyses. The BRCA TCGA methylation (HumanMethylation 450k) datasets were downloaded from the website: firebrowse (http://firebrowse.org). Statistical analysis on the TCGA data was performed with R software (version 3.6.0). All Mann–Whitney P values were calculated with the R function, wilcox.test. The correlation plots and Pearson correlation statistics were calculated with the R function, cor.test. The hypoxia score is the average of the z-score of each of the 42 genes in the hypoxia signature. The hypoxia signature was defined as presented in our prior article (8) by comparing the transcriptional profile of 34 breast cancer cell lines exposed to hypoxia to identify genes with consistent regulation under hypoxic conditions.

Animal research complied with all relevant ethical regulations according to the protocols approved by the Johns Hopkins University (Baltimore, MD) Animal Care and Use Committee. Female NSG mice, 5- to 7-week-old, were anesthetized and 2 × 10⁶ BT-474 cells were injected into the mammary fat pad. Slow-release estradiol pellets (2 mg/pellet) were implanted subcutaneously 3 days prior to BT-474 cell injection. Tumors were excised when they reached 0.5 mm in diameter. Excised tumors were formalin-fixed (Sigma-Aldrich) and paraffin-embedded.

Paraffin-embedded tissue sections were dewaxed with xylenes and hydrated with decreasing gradients of ethanol. Tissue sections were treated with Tris-EDTA buffer (10 mmol/L Tris-Cl and 1 mmol/L EDTA, pH 9.0) at near-boiling temperature for 20 minutes for antigen retrieval. IHC was conducted with the Vectastain Elite ABC HRP Kit (Vector Laboratories, PK-7200) and DAB Peroxidase (HRP) Substrate Kit (Vector Laboratories, SK-4100) according to the manufacturer's instruction. Primary antibodies against EGFR (1:400, Proteintech, 18986-1-AP) and HIF1α (1:400, BD Biosciences, 610958) were diluted in 1% BSA in PBS and applied to slides for 1 hour at room temperature. Slides were imaged in bright field on a Cytation 5 Cell Imaging Multi-Mode Reader (BioTek).

Cells (1–2 × 10⁵) were seeded in 96-well plates and exposed to 20% or 1% O₂ in the presence of Erlotinib (Selleck Chemicals), gefitinib (Selleck Chemicals), or DMSO at the indicated dose. After 48 hours, cells were washed and harvested in 0.5 mL of trypsin. An additional 0.5 mL of media were added, and cells were counted after 0.4% Trypan Blue (Gibco) staining using a Countess II FL Automated Cell Counter (Thermo Fisher Scientific). Alternatively, cells were treated with the aforementioned drugs for 2 days and then incubated with 10% AlamarBlue Cell Viability Reagent (Thermo Fisher Scientific) for 4 hours. Media (100 μL) from each well were collected and transferred to a black, clear-bottom 96-well plate. The fluorescence intensity was measured at an excitation wavelength of 560 nm and emission wavelength of 590 nm on a Cytation 5 Multi-Mode Reader (BioTek). The fluorescence intensity of AlamarBlue media in a control well with no cells was subtracted from the measurement of all experimental samples prior to analysis. After measuring the AlamarBlue intensity, the cells were fixed with 0.5% Crystal Violet (Sigma-Aldrich) with methanol for 10 minutes. Then, crystal violet solution was removed and followed by 3–5 washes with PBS. The plate was left to dry overnight and imaged with Cytation 5 Cell Imaging Multi-Mode Reader (BioTek).

MCF7 cells were plated in serum-free medium for 24 hours. Cells were then exposed to hypoxia condition for 24 hours and then treated with 100 ng/μL EGF while being maintained under hypoxia condition for an additional 16 hours. Cells were then pelleted, resuspended in water, and fixed by adding 100% ethanol drop-wise to a final concentration of 70%. Fixed cells were maintained on ice for 2 hours. Cells were then washed with PBS, pelleted, and incubated in staining buffer (PBS with 100 μg/mL RNase A and 50 μg/mL propidium iodide) overnight at 4°C in the dark. Flow cytometry for cell-cycle analysis was performed on an LSR II Flow Cytometer (BD Biosciences). Data were analyzed with FlowJo V10 Software (Tree Star Inc.).

Ki67- and DAPI-stained cells were imaged in a 3 × 3 montage per well with a Cytation 5 Cell Imaging Multi-Mode Reader (BioTek). Using the Gen5 3.05 Software (BioTek), the DAPI channel was used to count the number of cells in a field of view and to create an individual mask on the area of each nucleus. Ki67 intensity within each individual mask was then quantified. By visually observing cells, a threshold intensity of 8,000 was selected. Any object with an intensity greater or equal to the threshold was deemed positive, and all others were deemed negative. Percentage was determined for each well by dividing the number of positive cells over the total number of counted cells.

Cells were cross-linked with 1% formaldehyde for 10 minutes and quenched in 0.125 mol/L glycine. Chromatin was sheared by sonication using a Covaris Sonicator [settings: power, 150 W; duty factor, 5%; cycles, 200; and treatment time, 420 seconds (7 minutes)]. Sonicated lysates were precleared with Salmon Sperm DNA/Protein A Agarose Slurry (Millipore). IgG (Santa Cruz Biotechnology) or primary antibodies against HIF1α (Santa Cruz Biotechnology), HIF2α (Novus Biologicals), or HIF1β (Novus Biologicals) were added and incubated overnight with precleared lysates. The following day, salmon sperm DNA/protein A agarose beads were added for 4 hours at 37°C. The agarose beads were collected and washed sequentially with: low- and high-salt immune complex wash buffers, LiCl immune complex wash buffer, and twice with TE buffer. The DNA was eluted from the agarose gel in 1% SDS/0.1 mol/L NaHCO₃ and cross-links were reversed by addition of NaCl to a final concentration of 0.2 mol/L. Proteinase K was added to degrade protein in the lysate. DNA was recovered by phenol–chloroform extraction followed by ethanol precipitation, treated with RNase, and analyzed by qPCR. Fold enrichment was calculated on the basis of the cycle threshold (C_t) as 2^−Δ(ΔC_t), where ΔC_t = C_t,IP − C_t,Input and Δ(ΔC_t) = ΔC_t,antibody − ΔC_t,IgG.

The EGFR, mutated EGFR, or LDHA luciferase reporter vectors were generated by annealing 10 μmol/L of the forward strand and 10 μmol/L of reverse strand of 60-bp oligonucleotide sequence of EGFR or LDHA containing the HIF1α binding site. The 60-bp fragment was cloned into a pENTR TOPO vector using a pENTR Directional TOPO Cloning Kit according to the manufacturer's protocol (Life Technologies). The pENTR plasmids were then recombined into the pGL4.23-GW plasmid upstream of firefly luciferase using Gateway LR Clonase Enzyme (Thermo Fisher Scientific). Oligonucleotide sequences are listed in Supplementary Table S1. Plasmid constructs were then confirmed by Sanger sequencing.

MCF7 and 293T cells were seeded overnight in 24-well plates and cotransfected with 0.4 μg of the indicated vectors, 0.05 μg of psVmRL Renilla luciferase vector, and 0.05 μg of pcDNA3-EGFP per well using PolyJet In Vitro DNA Transfection Reagent (SignaGen Laboratories). psVmRL Renilla luciferase vector was used as an internal control. Approximately 16 hours later, media were refreshed, and the transfected cells were exposed to 20% or 1% O₂ for 24 hours. Cells were analyzed for luciferase activity using a Dual-Luciferase Reporter Assay System per the manufacturer's instructions (Promega).

DNA (200 ng) from each of the indicated samples was treated with sodium bisulfite using the EpiTect Bisulfite Kit (Qiagen, 59824) according to the manufacturer's protocol. The bisulfite-treated DNA (20 ng) was used for methylation-specific PCR (MSP) or Sanger sequencing. PCR was used to amplify bisulfite-treated DNA prior to Sanger sequencing. Primers used for PCR are provided in Supplementary Table S1. Sanger sequencing was performed by Johns Hopkins Genetic Resources Core Facility.

EGFR forward and reverse primers specific for the detection of only nonmethylated bisulfite-treated DNA are provided in Supplementary Table S1. The specificity of the primer pair was assessed using fully methylated or nonmethylated Synthetic DNA (Ultramer DNA Oligonucleotides, Integrated DNA Technologies). The ultramers were mixed to achieve the following percentage of methylated DNA: 0%, 6.25%, 12.5%, 25%, 50%, 75%, and 100%. PCR was performed in a final volume of 20 μL, containing 10 μl of Precision Melt Supermix (Bio-Rad, 172-5110), 2 μmol/L of each primer, 20 ng of bisulfite-modified DNA template, and remaining volume of DNase-free water. Each reaction was performed in triplicate. All analyses were run according to the following conditions: one cycle of 95°C for 2 minutes, 43 cycles of 95°C for 10 seconds, Ta for 30 seconds, and 60°C for 30 seconds; followed by a high-resolution melt (HRM) step at 95°C for 30 seconds and at 60°C for 1 minute, 65°C for 15 seconds, and continuous acquisition to 95°C at one acquisition per 0.2°C.

The percent methylation was calculated after normalizing to DAPIKI expression on the basis of the threshold cycle (C_t) as 2^−Δ(ΔC_t), where ΔC_t = C_t,sample – C_t,DAPIKI and Δ(ΔC_t) = ΔC_t,test − ΔC_t,control. DAPIKI primers were designed to detect both nonmethylated and methylated DAPIKI. For samples with no signal detection, the C_t value was set to 43. Raw data from MSP were analyzed utilizing web-based high-resolution DNA melting analysis software (uAnalyze 2.0), with normalized curves for comparison among samples (29).

BT-474 cells (1 × 10⁴) were plated in 6‐well plates coated with soluble rat tail Type I Collagen (Corning). Cells were incubated overnight, and phase contrast images were taken every 5 minutes for 23 hours using a Lionheart (BioTek). MetaMorph software was used to determine x and y coordinates at each time interval and to construct cell trajectory maps. The cell trajectories were fit using an anisotropic persistent random walk (APRW) model of cell motility to calculate distance traveled from origin and total cell diffusivities (Dtot). APRW model analysis was performed as described in detail using MATLAB (30). Three‐dimensional cell trajectory data were used to statistically profile cell migration using the MSD, which can be obtained from (x[t], y[t]) coordinates of cells with time (t). MSD (τ) = (x[t + τ] − x[t]) + (y[t + τ] – y[t]), where τ = 5 min * frame number. Values of persistence and speed were obtained from APRW model fitting and expressed as speed (S) and persistence (P) of cells, which can be used to calculate total cell diffusivity (D_tot). D_tot = (S_p²P_p + S_np²P_np)/4, where both speed (S) and persistence (P) were calculated along both the primary and nonprimary axes.

All the values in text and figures are presented as mean ± SEM, unless otherwise stated. Statistical significance was determined when appropriate by using Student t test or one/two-way ANOVA with Bonferroni post-test. P < 0.05 was considered significant.

In our previous work (8), we performed an RNA sequencing analysis of 31 breast cancer cell lines exposed to 20% or 1% O₂ conditions for 24 hours. Our results showed that more than 1,000 genes were induced or repressed in each cell line in response to hypoxia; however, only 42 genes shared a conserved response to hypoxia. Intriguingly, EGFR was among the genes that showed induction under hypoxic conditions in some, but not all, breast cancer cell lines. To confirm this finding, we performed qPCR and verified that only seven of the 31 breast cancer cell lines had a 2-fold or greater increase in EGFR expression upon exposure to hypoxia (Fig. 1A). Subsequently, we determined the baseline and hypoxia-induced mRNA and protein expression of EGFR in luminal and basal cell lines (Fig. 1B–D; Supplementary Fig. S1A and S1B). One basal cell line, SUM149, showed increased EGFR expression under hypoxic conditions, whereas all of the luminal cell lines, except MDA-MB-175, displayed increased EGFR expression (Fig. 1B–D; Supplementary Fig. S1A and S1B). EGFR was localized in hypoxic regions in orthotopic tumors derived from luminal, BT-474 cells (Supplementary Fig. S1C).

Given that the luminal cell lines express the estrogen receptor (ER), whereas basal cell lines do not, we reasoned that ER expression may play a role in EGFR induction upon exposure to hypoxia. To test this hypothesis, we inhibited ER activity by treating cells with 4-hydroxytamoxifen or fulvestrant. Both treatments inhibited ER-regulated TFF1 induction, but did not affect EGFR expression under hypoxic conditions in both BT474 and MCF7 cells (Supplementary Fig. S1D–S1F). We also utilized an MCF10A human breast epithelial cell line, engineered to overexpress an ER cDNA (28), to determine whether ER expression would promote EGFR expression under hypoxic conditions. ER expression did not promote EGFR expression under hypoxic conditions (Supplementary Fig. S1G and S1H). The results verify that hypoxia selectively induces EGFR expression in some breast cancer cell lines, but manipulating ER expression did not alter this response. Using RNA expression data from TCGA, we found that EGFR expression correlates with expression of our hypoxia score (8) in samples from patients with either luminal (ER⁺) or basal (ER⁻) breast cancer (Fig. 1E). However, it is important to note that both the hypoxia score (8) and EGFR expression (31) have been shown to be enriched in basal breast cancer and may play a role in the aforementioned result.

To determine whether HIFs are required for EGFR induction under hypoxic conditions, we assessed the expression of EGFR in CRISPR-depleted HIF1α- or HIF2α-knockout MCF7 subclones (Supplementary Fig. S2A). The knockout of HIF1α abrogated EGFR induction upon exposure to hypoxia at both the mRNA and protein levels, whereas the knockout of HIF2α did not, demonstrating that HIF1α (but not HIF2α) is required for EGFR induction under hypoxic conditions (Fig. 2A and B). To determine whether the increase in EGFR levels under hypoxia is sufficient to activate the EGFR pathway, we stimulated cells with EGF. MCF7, BT474, and HCC1428 cells showed increased phosphorylation of AKT and ERK under both hypoxic and normal O₂ conditions following 30 minutes of stimulation with EGF, albeit the levels of pAKT and pERK induction under hypoxic conditions varied between the cell lines, with HCC1428 cells showing the most striking induction (Fig. 2C; Supplementary Fig. S2B and S2C). The knockout of HIF1α abrogated this effect (Fig. 2D). Next, we stimulated MCF7 and BT474 cells with EGF and exposed the cells to hypoxia in the presence of the EGFR inhibitor, gefitinib. The robust increase in AKT and ERK phosphorylation in response to hypoxia was abrogated in a dose-dependent manner in response to gefitinib treatment (Fig. 2E; Supplementary Fig. S2D). Taken together, the results demonstrate that HIF1α increases EGFR expression under hypoxic conditions, leading to robust activation of the EGFR pathway in response to ligand (Fig. 2F).

To determine whether HIF1α is a direct transcriptional regulator of EGFR, we searched for putative HIF1α binding sites within the EGFR gene. We also leveraged the results of a previous study that used high-resolution genome-wide mapping of HIF-binding sites in MCF7 cells exposed to 0.5% O₂ or 2 mmol/L of DMOG (32). A high-stringency HIF1α-binding region was identified in intron 18 of the EGFR loci (Fig. 3A). Using a chromatin immunoprecipitation (ChIP) assay, we confirmed that HIF1α and HIF1β, but not HIF2α, were bound to this region, which contained three ACGTG sites with enrichment levels similar to LDHA binding (Fig. 3B; Supplementary Fig. S3A). On the other hand, HIF1α did not bind to a nearby two different regions of intron 17 in EGFR. Likewise, HIFs were not enriched in an intronic region of EGFR, which contained five ACGTG binding sites (Supplementary Fig. S3B and S3C). This demonstrates the specific recruitment of HIF1α to intron 18 of the EGFR gene under hypoxic conditions in BT474 cells.

Because only seven of the 31 breast cancer cell lines displayed a significant induction of EGFR upon exposure to hypoxia (Fig. 1A), we hypothesized that EGFR may contain one or more single-nucleotide variants (SNV) in the HIF1 binding region. To address this consideration, we isolated DNA from 10 cell lines and Sanger sequenced and amplified a 400-bp region of EGFR containing the HIF–DNA binding site (Fig. 3C). The DNA isolated from ZR-75-1, MCF7, and BT-474 displayed a unique point mutation (T > C) that generated an additional ACGTG site (Supplementary Table S2). To determine whether the SNV altered EGFR expression under hypoxia and to verify that we had identified a functional HRE, we utilized a luciferase reporter assay. We inserted a 60-bp sequence spanning the HIF-binding sites in EGFR into the reporter plasmid, pGL4.23-GW-luciferase, in which a basal SV40 promoter drives firefly luciferase expression. Two additional constructs were also generated, one containing the SNV (T > C) of EGFR and one in which all three HIF sites were mutated (EGFR-MUT). In MCF7 and 293T cells, transfected with pGL4.23-EGFR and pGL4.23-EGFR (T > C), luciferase activity was increased 2.5-fold on exposure to hypoxia, whereas in cells transfected with pGL4.23-EGFR-MUT, the hypoxic induction of luciferase expression was abrogated (Fig. 3D and E). We also constructed an LDHA luciferase reporter construct as a positive control. Thus, the ChIP and luciferase reporter assays demonstrate that EGFR is a direct HIF1 target gene. However, the single-nucleotide variation that we identified neither altered luciferase expression nor did it provide evidence for the difference in EGFR regulation by hypoxia between cell lines.

Given that nucleotide variations did not predict for hypoxia-induced EGFR expression among breast cancer cell lines, we next questioned whether the methylation status of the ACGTG binding site might play a role. Methylation of promoter regions is a well-established mechanism for gene silencing (33). To assess the methylation status of the HIF1-binding region, we isolated, bisulfite treated, and PCR amplified a 400-bp region of EGFR containing the HIF–DNA binding site followed by Sanger sequencing of 10 breast cancer cell lines (Fig. 4A and B; Supplementary Fig. S4A and S4B; Supplementary Table S3). The chromatogram analysis demonstrated that all cytosine residues in the amplified region of EGFR in MCF7, CAMA1, HCC1428, and BT-474 cells were unmethylated. On the other hand, every cytosine residue in the HIF-binding region of EGFR in MDA-MB-231, hTERT-HME, MCF10A, HCC1806, BT-20, and SUM159 cells was methylated.

To confirm the results of Sanger sequencing, we developed a MSP assay paired with an HRM curve analysis. To test the ability of our assay to discriminate between nonmethylated and methylated bisulfite-treated DNA, we designed synthetic oligos that represent a PCR-amplified, bisulfite-treated DNA sequence of a fully methylated or nonmethylated HIF1-binding region in the EGFR intron 18. The oligos were mixed at ratios from 0:1 to 1:0 (methylated:nonmethylated) to show the specificity of the primers to detect only nonmethylated DNA (Supplementary Fig. S5A). Next, we used uAnalyze (28), a web-based high-resolution DNA melting analysis tool to determine the unique melt curve signature of the methylated and nonmethylated DNA sequences (Supplementary Fig. S5B and S5C). Oligonucleotide DNA had distinct melting curves of 70°C and 72°C for nonmethylated and methylated DNA, respectively. Mixed ratios of nonmethylated to methylated DNA had bimodal melting curves that reflected the input quantities of each oligo. Given that temperature plays a role in the helical twist of DNA, we also considered the percent helicity of DNA, which ranged between −40% and 20% for methylated and nonmethylated amplicons, respectively.

After verifying that our MSP-paired HRM assay can successfully distinguish between methylated and nonmethylated EGFR amplicons, we used the assay to confirm the methylation status of the cell lines previously tested using Sanger sequencing of bisulfite-treated DNA (Fig. 4C and D). To determine whether the HIF1-binding site is methylated in normal epithelial cells, we tested cells from four different individuals and determined that they were 100% methylated (Fig. 4E). Intrigued by the result, we assessed the 450K methylation array data from TCGA, which contains probe cg20062492, to detect the methylation status of the EGFR region of interest (Supplementary Fig. S6A). The results show that normal breast tissue displayed a higher level of methylation in HIF-binding region compared with breast cancer tissue (Fig. 4F). The methylation levels are significantly lower for breast cancer compared with normal breast tissue. Together, these results suggest that the methylation status of the HIF1-binding site prevents EGFR induction under hypoxic conditions.

Our results demonstrate that the methylation status of the HIF-binding region correlates with increased expression of EGFR under hypoxic conditions. To determine whether demethylation could restore EGFR induction under hypoxic conditions, we treated MDA-MB-231 and SUM-159 cells with 500 nmol/L azacytidine or 100 nmol/L of decitabine for 3 days followed by drug withdrawal for an additional 6 or 10 days. The MSP-paired HRM (Fig. 5A and B; Supplementary Fig. S7A and S7B) showed that both azacytidine and decitabine decreased methylation at this site by 80%. Using a ChIP assay, we also confirmed that decitabine treatment enhances HIF1α (but not HIF2α) to the HRE of the EGFR gene under hypoxia (Fig. 5C and D; Supplementary Fig. S7C). Treatment with either azacytidine or decitabine also restored EGFR mRNA and protein induction by hypoxia in MDA-MB-231 or SUM-159 cells (Fig. 5E–H; Supplementary Fig. S7D). The results demonstrate that cytosine methylation within the HIF-binding region prevents the induction of EGFR under hypoxia, which can be restored by treatment with demethylating agents.

Mitogen activation stimulates the Ras/MEK/ERK and Ras/PI3K/AKT pathway, leading to the induction of cyclins, c-myc, and Rb phosphorylation (34–37). We found that hypoxia, EGF, or the combination of EGF and hypoxia led to an increase in the level of almost every cyclin tested, promoted Rb phosphorylation, and enhanced E2F levels, while concomitantly decreasing p21 levels at 20% O₂ conditions (Fig. 6A and B; Supplementary Fig. S8A). The expression of cyclin D1, c-myc, and phosphorylated RB was reduced in HIF1α-knockout cell lines (Supplementary Fig. S8B).

To determine whether EGFR induction in hypoxic breast cancer cells promotes cell-cycle progression, we stimulated MCF7 cells with EGF for 0, 24, or 48 hours and assessed cell proliferation (Fig. 6C and D; Supplementary Fig. S8C). EGF treatment led to a robust increase in Ki67 expression under hypoxia (Fig. 6C and D) accompanied by increased proliferation (Fig. 6E). EGF stimulation led to a 21% increase in the number of cells in S-phase when cultured under hypoxia as compared with a 12% increase in the number of cells in S-phase under normal O₂ conditions (Supplementary Fig. S8C).

Given the enhanced expression and activation of EGFR under hypoxic conditions, we reasoned that similar to cells with an EGFR amplification (38, 39), cells exposed to hypoxia may be more sensitive to EGFR inhibitors, gefitinib and erlotinib. Under hypoxic conditions, MCF7 and BT474 cells were more sensitive to both erlotinib and gefitinib (Fig. 6F and G; Supplementary Fig. S8D and S8E), whereas HIF1α-knockout subclones were resistant to gefitinib treatment under hypoxia (Fig. 6H). Collectively, the results suggest that HIF1α induction enhances EGFR expression to promote activation by ligand or suppression via EGFR-targeted therapies.

In addition to stimulating cell proliferation, EGF stimulation (40, 41) and hypoxia (42) independently promote two-dimensional cell motility. To determine whether EGF and hypoxia would synergistically enhance cell motility, EGF-stimulated BT474 cells were exposed to hypoxia, and migration was monitored using time-lapse phase microscopy. The data demonstrate that cells exposed to hypoxia in the presence of EGF showed the highest increase in motility (Fig. 6I–K).

Although the HRE consensus sequence 5′-ACGTG-3′ contains a methylation-prone CpG dinucleotide (43, 44), the extent and role of methylation on hypoxic gene regulation have been limited. Our results uncovered a functional HRE in an intron region of EGFR that is activated under hypoxic conditions in an HIF-dependent manner. The methylation status of the CpG dinucleotide within the HIF-binding region in intron 18 correlates to the ability for EGFR to be induced under hypoxic conditions in breast cancer. In cell lines with methylation of this region, treatment with a demethyltransferases restored EGFR regulation under hypoxia. Similar to our findings with EGFR, erythropoietin (EPO) is a also well-known hypoxia-regulated gene whose expression correlates inversely with methylation (45, 46). Direct methylation of the HRE sequence of the EPO gene has been shown to abrogate both HIF1 DNA binding and hypoxic reporter gene activation (45). Likewise, a study of the MUC17 gene in pancreatic cancer demonstrated a robust hypoxic induction only in cell lines without methylated HRE regions (47, 48). Further investigation is warranted to determine how globally the methylation status of DNA affects the expression of hypoxia-regulated genes and the mechanisms that are involved.

DNA derived from the normal human mammary epithelial cells of four donor patients indicated that the EGFR region was methylated in normal mammary epithelium. This finding is also supported by the high methylation beta values in this region of EGFR reported for more than 60 patients that donated normal tissue for TCGA project. Therefore, the data from patient tissue and breast cancer cell lines suggest that hypomethylation of the region occurs in cancer and correlates with increased expression under hypoxia. Although DNA hypomethylation was the initial epigenetic abnormality recognized in human tumors (49), hypermethylation of promoters of genes that are silenced in cancers (e.g., tumor suppressor genes) have been the most well studied to date. Recent high-resolution genome-wide studies confirm that DNA hypomethylation is as prevalent as DNA hypermethylation (50). One study found that in addition to global hypomethylation of repeat sequences, hypomethylation of certain genes in cancer, especially in genes linked with signaling pathways (e.g., BCR, LYN RAB8A, and NFKBIB), chromatin modifications (e.g., CHD2, CHD3, and SMARCB1), cell growth and development (e.g., EBF1, EGR1, EGFR, ERBB2, and MYC), apoptosis inhibition (e.g., BCL2 and TRAF1), and cell proliferation (e.g., CCND1, LYN, and BCL3) can occur. It will be important to determine how hypomethylation of HREs occurs and whether hypomethylation correlates strictly within HREs found in intronic regions on the genome.

There are several functional implications for the activation of EGFR in response to hypoxia. First, cells exposed to hypoxia will be more sensitive to EGF-induced signaling with enhanced AKT and ERK phosphorylation that occurs in an HIF-dependent manner. We found that hypoxia promotes G₁–S-phase progression, progression from G₂-phase to M-phase, and also prevents apoptosis in the presence of EGF by reducing the levels of p21. Some tumor types produce EGF in excess, which would amplify the activation of the EGFR under hypoxic conditions. On the other hand, hypoxia sensitizes cells to EGFR inhibition. To date, six phase II clinical trials to investigate the efficacy and safety of anti-EGFR mAbs in patients with TNBC have been reported (18). In breast cancer, the clinical trials of EGFR inhibitors have shown low response rates, however, some patients have shown a meaningful response. Therefore, it may be necessary to stratify patients to allow those patients that may benefit from EGFR-targeting agents to have access to agents currently used in the clinic. Our results suggest that patients with hypoxic tumors and hypomethylated EGFR gene may be candidates for the addition of EGFR inhibitors to their current treatment regimens.

No potential conflicts of interest were disclosed.

M. Mamo: Formal analysis, validation, investigation, methodology, writing-original draft, writing-review and editing. I.C. Ye: Data curation, formal analysis, validation, investigation. J.W. DiGiacomo: Investigation, writing-review and editing. J.Y. Park: Data curation, validation. B. Downs: Data curation, formal analysis. D.M. Gilkes: Conceptualization, resources, formal analysis, supervision, funding acquisition, methodology, writing-original draft, project administration, writing-review and editing.

We thank Julia Ju for helping with sample preparation for this work and Sara Sukumar and Mary Jo Fackler for helpful advice on methylation-specific PCR assays and design. Work in the Gilkes laboratory was supported by U54-CA210173 (NCI), R00-CA181352 (NCI), Susan G. Komen Foundation (CCR17483484), The Jayne Koskinas Ted Giovanis Foundation for Health and Policy, The Emerson Collective, The Allegany Health Network, the NIH Predoctoral and Postdoctoral Training Program (T32 CA 153952), and the SKCCC Core Grant (P50CA006973, NCI).

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.