A Functional Epidermal Growth Factor (EGF) Polymorphism, EGF Serum Levels, and Esophageal Adenocarcinoma Risk and Outcome

Esophageal adenocarcinoma (EAC) is an aggressive tumor with a dismal 5-year overall survival (OS) rate of <15% (1, 2). Over the past 30 years, EAC incidence has surpassed that of esophageal squamous cell carcinoma in the Western world (3). The majority of patients have unresectable tumors or detectable metastases at the time of diagnosis (4). Many clinical risk factors have been established for EAC, including gastroesophageal reflux disease (GERD), Barrett's esophagus (5), tobacco use, obesity, and male gender (2, 4, 6). In the era of targeted therapy, evidence for molecular risk factors for EAC is mounting (7), although the prognostic value of these predictors is largely unknown.

Examples of molecular targets evaluated in EAC therapy include epidermal growth factor receptor (EGFR), matrix metalloproteinases, and cyclooxygenase-2. EGF is an important gene for proliferation and differentiation of epithelial cells. Its overexpression in esophageal cancer has been implicated in tumor invasion and prognosis (8–10). Moreover, a single nucleotide A→G polymorphism at position +61 in the 5′-untranslated region of the EGF gene has been associated with an elevated risk of glioma (11, 12), malignant melanoma (13, 14), gastric cancer (15), and lung cancer (16) and transformation of cirrhosis to hepatocellular carcinoma (17).

In the present study, we evaluated the role of the EGF A61G polymorphism in EAC development and risk using a case-control design. To further study the potential functional effect of EGF A61G in EAC risk, we measured serum EGF levels in a GERD population with endoscopic evidence of Barrett's esophagus. Because several studies suggested malignant progression and worse prognosis in other types of cancer patients carrying the G variant of EGF A61G (12, 15), we also evaluated this polymorphism for EAC survival.

Study populations. All participants provided informed written consent. Approval for this study was granted by the institutional review boards at the Massachusetts General Hospital, Harvard School of Public Health, and Princess Margaret Hospital.

The case-control study enrolled 313 incident EAC patients, all of which had DNA available, where 312 had successful DNA genotyping. Patients were recruited at the Massachusetts General Hospital between November 1999 and December 2005. The 447 controls with successful genotyping (454 screened and 454 with DNA) consisted of healthy friends and non–blood-related family members who were visiting hospital patients; these hospital patients were admitted either with upper aerodigestive cancers or for cardiothoracic surgery. No controls were actual hospital patients, and none reported ever having a diagnosis of cancer. We selected controls with no self-reported history of GERD. Cases and controls were frequency matched for age and gender distributions. Recruitment rate was >85% for cases and controls.

All participants were interviewed by a trained research nurse, who collected important covariate information such as age, gender, height, weight, ethnicity, gastroesophageal reflux symptoms, presence of Barrett's esophagus, and tobacco and alcohol exposure. Healthy adult body mass index (BMI) was estimated using the self-reported average heights and weights of participants during their third decade of life.

Blood samples from a cancer-free population were collected on the day of esophagoscopy for 144 patients with a history of GERD and processed within 2 h. Thus, 82 GERD patients with histologically confirmed Barrett's esophagus and 62 GERD patients with endoscopically normal esophagus were analyzed.

DNA extraction and genotyping of EGF gene. DNA was extracted from whole blood for cases and controls using the Puregene DNA Isolation kit (Gentra Systems). Commercial primers were used to do PCRs using Taqman (Applied Biosystems) to identify a single nucleotide polymorphism at position 61 in the 5′-untranslated region of the EGF gene (rs4444903). The primers and conditions are available on request and were obtained from Assays-by-Demand. Cases and controls were genotyped blinded to all clinical information and case-control status in mixed batches. Two coauthors checked all results independently and a third arbitrated discrepancies. A random 15% of the samples and all equivocal or failed genotyping results were repeated.

EGF ELISA. For serum analysis, 5 to 6 mL of blood were collected in an uncoated plastic blood tube at the time of esophagoscopy and processed within 2 h. EGF levels were measured in serum as opposed to plasma because the major source of EGF in the blood is platelets (18). The serum was allowed to clot for 30 min at 4°C before centrifugation at 2,000 rpm for 10 min at 4°C. Serum was isolated and stored at −80°C before use. EGF protein was subsequently quantified in a blinded fashion using an ELISA (PeproTech). Each ELISA plate well was incubated overnight with 100 μL of capture antibody (1 μg/mL) before blocking with 1% bovine serum albumin in PBS for 1 h. Serum samples (diluted 1:20 in PBS containing 25% fetal bovine serum) were analyzed in triplicate and incubated in the plate for 2 h followed by addition of 100 μL of detection antibody (0.25 μg/mL) for 2 h and 100 μL of avidin peroxidase (1:2,000) for 30 min. Wells were washed four times with PBS containing 0.05% Tween 20 between each step. Color development was monitored after the addition of 100 μL of 3,3′,5,5′-tetramethylbenzidine (R&D Systems) using a spectrophotometric plate reader (Emax, Molecular Devices). An equal number of samples representing each genotype were run on every ELISA plate and the average concentration of each sample was determined based on a plate-specific EGF standard curve.

Outcome data collection. Of 312 cases, 262 had follow-up outcome data available. Notes and results from Massachusetts General Hospital, referring and primary care physicians, and the social security death index were all used to determine OS and failure-free survival (FFS). FFS is defined as the time from the date of first histologic diagnosis to the first date of either (a) cancer progression or recurrence per Response Evaluation Criteria in Solid Tumors (19) or (b) death from any cause. Individuals, who at their last radiologic, physical, or pathologic evaluation had no evidence of recurrence or progression by Response Evaluation Criteria in Solid Tumors, were censored at the date of their last evaluation for FFS. The last date known to be alive was used for censoring in OS analyses.

Statistical methods. Hardy-Weinberg equilibrium (HWE) was assessed in cases and controls separately using χ² tests. Unconditional logistic regression was used to analyze the case-control study. Although cases and controls were frequency matched on age and gender, we adjusted for age, adult BMI, and pack-years of smoking using continuous variables, and gender and smoking status (current, ex-smoker, or never smoker) using indicator variables. Crude odds ratios (OR) and adjusted ORs (AOR) with 95% confidence intervals (95% CI) were calculated for homozygous (G/G) and heterozygous (A/G) genotypes using the wild-type genotype (A/A) as the reference. Survival curves for OS and FFS were constructed using the Kaplan-Meier method and compared using the log-rank test. Cox proportional hazards models for both OS and FFS were used to adjust for clinical covariates identified as being significant in univariate analyses (e.g., stage, performance status, age, gender, and smoking history).

For the serum analysis, we first assessed assay variability by comparing the percent differences in absolute serum levels between each of the triplicate samples and the serum level corresponding with the arithmetic mean of the raw data obtained from the triplicate serum analysis (termed the average serum level). The average of these percent differences (assay variability) was compared with the percent differences in absolute serum levels across different individuals (interindividual variability). The average serum level was used for all subsequent analyses. EGF serum levels were first analyzed using the Wilcoxon rank sum test. Then, we compared G/G with A/G + A/A using t tests. Because serum EGF levels were not normally distributed, we log transformed the serum levels first before parametric testing. The statistical packages Statistical Analysis System 9.1 9 (SAS) and R were used for all analyses.

Demographics. Basic demographic data are presented in Table 1. Cases and controls had similar demographic distributions of age, gender, race, and pack-years. Of the known risk factors for EAC, healthy adult BMIs were significantly higher and there were substantially greater numbers of ever smokers in the cases.

EGF genotype and EAC risk. The concordance rate for the repeated genotyping samples was 99%. The χ² test for HWE was not significant for the 447 controls (P > 0.5) but was significant for the 312 cases (P = 0.04). There were no statistical differences between cases with and without survival outcome data for the following variables: age, gender, smoking status, performance status, and stage (P > 0.10 for each comparison; Table 2).

The G/G genotype was significantly more common in the cases than controls (P = 0.01, χ² test; Table 1). The risk of EAC in patients carrying the EGF A61G homozygous (G/G) genotype was significantly increased (Table 2) in both crude analyses (OR, 1.83; 95% CI, 1.2-2.7; P = 0.003) and after adjusting for age, gender, smoking status, cumulative tobacco exposure, and adult BMI (AOR, 1.81; 95% CI, 1.2-2.7; P = 0.005). No increased risk of EAC was observed with the heterozygote (A/G) when compared with the wild-type A/A genotype (Table 2).

Because Barrett's esophagus is a known independent risk factor for the development of EAC (5, 6), we explored two subsets of EACs, each compared with the entire set of controls (Table 2). In the first subset analysis, we evaluate patients with EAC that had pathologic or endoscopic evidence of Barrett's esophagus. In this subgroup, the G/G genotype was significantly associated with EAC risk (AOR, 2.18; 95% CI, 1.3-3.7; P = 0.004). In the second subset of patients that had been resected and found not to have Barrett's esophagus, the AORs were lower (AOR, 1.37; 95% CI, 0.7-2.6; P = 0.65).

We also did exploratory stratified risk analyses (Table 3). For continuous variables such as age and pack-years, we stratified based on median pack-years and quartiles for age in the controls. We stratified BMI using the standard obesity definition cutoff of 25. Deviation from HWE was not found in any control subset (P > 0.20 for each subgroup) or in most of the case subsets, except for female cases (P < 0.01). The risk conferred by EGF A61G seems to be limited to younger patients, earlier stages of disease, ever smokers particularly light smokers, and individuals who were not obese at a younger age. Given these findings, we also reanalyzed the data adjusting for healthy adult BMI 6 months before diagnosis (BMI-6M) instead of in the third decade of life (BMI). We found similar results for both groups of patients using this new definition: the AOR for the G/G genotype was 2.23 (95% CI, 1.1-3.6; P = 0.001) for BMI-6M < 25 and 1.74 (95% CI, 1.0-3.0; P = 0.94) for BMI-6M > 25. Although the G/G-associated risk of EAC was conferred to both genders, a substantially greater risk was found among females.

EGF genotype and EAC prognosis. Median OS was 27.1 months and median FFS was 15.1 months. There were 161 deaths of 262 evaluable patients and 184 recurrences or disease progression of 258 evaluable individuals. Median follow-up for OS was 29.4 months and 31.4 months for FFS (Figs. 1 and 2).

Individuals with EAC and a G/G genotype had similar OS (P = 0.69, log-rank test; Fig. 1) and FFS (P = 0.84, log-rank test; Fig. 2) compared with either the A/A or A/G genotype. When adjusted for disease stage, performance status, age, and pack-years (each of which was statistically significant in univariate analyses), the adjusted hazard ratios for the G/G and A/G genotypes (compared with reference group A/A) were 1.02 (95% CI, 0.7-1.6) and 1.03 (95% CI, 0.7-1.5), respectively, for OS. For FFS, the models were adjusted for stage and performance status only (the only statistically significant variables in univariate analyses), with adjusted hazard ratios of 1.05 (95% CI, 0.7-1.5) for the G/G genotype and 1.12 (95% CI, 0.8-1.7) for the A/G genotype. The G/G genotype did not confer a worse prognosis in any subgroup analyses, including stratified analyses by stage, smoking status, performance status, treatment, or presence or absence of concurrent Barrett's esophagus.

EGF serum levels and GERD. We assessed the serum EGF levels of an endoscopically evaluated population of GERD patients known to be free of esophageal cancer. This population had a median of four episodes of either acid reflux, regurgitation, or heartburn per week (range, 2-20 episodes). Eighty-two patients are known to have histologic evidence of Barrett's esophagus, whereas the remaining 62 each have an endoscopically normal esophagus. Of the 82 patients with Barrett's esophagus, 40 had moderate to severe dysplasia, whereas 42 had no or mild dysplasia (Table 4).

Patients with a G/G genotype (n = 34) had a statistically significant higher median serum EGF level compared with the individuals carrying at least one A allele (242 versus 216 pg/mL; P = 0.03, Wilcoxon rank sum test). In a subanalysis, serum EGF levels were also significantly elevated in the subset of GERD patients with Barrett's esophagus carrying G/G versus other genotypes (median, 254 versus 216 pg/mL; P = 0.05) but not in endoscopically normal patients [median, 214 pg/mL (G/G) versus 216 pg/mL (A/G + A/A); P > 0.20]. There was no correlation with the degree of Barrett's dysplasia and EGF level.

The mean variation in the average EGF concentrations across different individuals (interindividual variation) was 6.0-fold greater than the mean variation in the repeated triplicate samples (test variability) based on assessment of the first 100 samples.

This study provides the first report of a relationship between the EGF A61G homozygous variant genotype and risk of EAC. More specifically, patients with the G/G genotype were found to have almost a 2-fold greater risk of EAC compared with the wild-type genotype. In contrast, the risk associated with the G/A heterozygote seems to be similar to the wild-type genotype. We also detected elevated EGF levels in tumor-free individuals with GERD and G/G genotype when compared with individuals with all other genotypes, suggesting that EGF and its signaling pathways may potentiate esophageal tumorigenesis in patients with GERD, a well-known clinical risk factor of EAC. Because EGF/EGFR has been implicated in the progression of metaplasia to dysplasia to adenocarcinoma in patients with Barrett's epithelium (20), EGF genotype may serve as a novel biomarker to assess the risk of malignant transformation in endoscopic surveillance programs for patients with Barrett's esophagus.

Caution must be made about overinterpreting interesting results from exploratory analyses. For example, EGF A61G seems to confer a substantially greater risk in women, but the confidence interval is wide and women formed a small proportion of study subjects. Nonetheless, these exploratory results do suggest that EGF A61G may confer risk to important subsets of patients that previously had no known risk factor (i.e., females, early-onset EAC, and nonobese individuals). The greater association in light smokers may represent confounding by gender and age of onset of EAC because both factors are associated with lower cumulative smoking exposures. Alternatively, heavier smokers may result in the activation of alternative carcinogenic pathways that bypass the role of the EGF A61G polymorphism. The association between EGF A61G G/G and EAC risk in earlier-stage patients might suggest that the EGF A61G is associated with a less aggressive phenotype, paralleling lung cancer where individuals with EGF pathway-driven tumors seem to have fewer molecular alterations (21, 22); such a conclusion is premature. A greater association of this polymorphism and EAC risk in nonobese individuals is only hypothesis generating because using a different definition of BMI strata (i.e., BMI based on patient 6 months before diagnosis in cases and 6 months ago for controls) found similar results across strata. We did not have established non-GERD healthy controls to assess EGF levels and, thus, could not exclude the possibility that serum EGF levels are higher in individuals independent of GERD.

Because none of our controls had GERD symptoms, we were not able to analyze GERD-polymorphism relationships. Because we shared a common pool of controls with other aerodigestive cancers (from which the present age and gender frequency-matched control sample is derived), we had only limited information on GERD in the controls. Thus, we opted to use a homogeneous set of controls all lacking a history of GERD symptoms, as we have previously done (23).

Although our EAC cases were not in HWE, we have also genotyped an additional 2,500 cancer cases and 1,500 controls for other studies using the identical technique; each group of these other samples was in HWE (P > 0.10). This suggests that our cases were properly genotyped and that the cases out of HWE are reflective of a strong risk association for the G/G genotype.

We selected a set of endoscopically evaluated GERD patients for comparing constitutive serum EGF levels and EGF genotype. A little over half of this set of patients had Barrett's esophagus with GERD, and the remainder had GERD without Barrett's esophagus. GERD is the most clinically assessable risk factor for EAC risk, and we wanted to determine if EGF A61G G/G is linked with serum levels of EGF in this population. Previous studies (13) have already shown that the G allele was linked to elevated EGF production in lymphocytes in a general population; we are able to confirm an association with serum EGF levels in vivo in a more specific and clinically relevant population at risk of developing EAC. Although we would have wanted to evaluate serum EGF levels in EAC patients before the onset of cancer, this was clearly not feasible in a case-control setting. Finally, serum EGF levels may not entirely reflect the endoluminal milieu of the esophagus because EGF is abundant in saliva and esophageal mucosa. Our overall results, nonetheless, are consistent with a genetically recessive genotype-phenotype correlation in at-risk GERD patients.

In contrast to investigations in gastric cancer (15) and glioma (11), the presence of an A61G single nucleotide polymorphism in the EGF gene did not impart a more aggressive phenotype or affect prognosis in patients with EAC. There are several potential explanations. First, esophageal cancer is among the most lethal of cancers. Thus, even in the earliest clinical stages, multiple pathways are already deregulated and the cancer is unlikely to be driven by the EGF pathway alone. Second, EGF A61G may not be prognostic but predictive in the setting of clinical inhibition of the EGF pathway. However, <1% of our population received an EGFR pathway inhibitor such as an EGFR tyrosine kinase inhibitor or targeted antibody. Third, the entire EGF pathway may be affected by racial differences. In lung cancer patients, being of Asian descent (where the prevalence of a GG allele is ∼40%) significantly increases the chance of responding to an inhibitor of this pathway. Of note, the distribution of EGF genotype in our control population (predominantly Caucasian) was strikingly similar to previously reported studies in Caucasians (12, 24).

There are studies examining populations with melanoma (24, 25), gastric cancer (26), glioma (27), and cervical cancer (28) that have not shown a consistent correlation between EGF A61G polymorphism and risk of disease. This discordance can be partially attributed to ethnic variation and population stratification.

In summary, the potential future role of EGF A61G may lie in its ability to identify patients at risk for developing EAC. This novel correlation requires verification in other patient populations with EAC, particularly in other races, because our population was overwhelmingly Caucasian. EGF genotyping may thus eventually form one of a panel of molecular markers that includes p53 expression status (29) and cyclin D1 genotyping (30, 31) that is clinically useful in predicting which GERD patients should have increased surveillance endoscopy. Although these molecular markers are not yet applicable in clinical practice, the EGF A61G polymorphism may be among the first markers that may identify EAC risk, with particular interest in women, nonobese individuals, and early-onset EAC, where few risk factors (either clinical or molecular) have been identified.

No potential conflicts of interest were disclosed.

We thank Peggy Suen, William Puricelli, Richard Rivera Massa, Andrea Shafer, and Matthew Kulke.