The present study was approved by the Medical Ethics Committee of Sun Yat-Sen University Cancer Center (Guangzhou, China), and all enrolled patients provided written informed consent. A total of 253 patient specimens from 186 males and 67 females; the median patient age was 58 years (age range, 32–80 years) were collected from post-operative patients with ESCC, the validity of which were confirmed by pathological review following IHC staining. All 253 patients were observed at Sun Yat-Sen University Cancer Center (Guangzhou, China) between January 2000 and December 2007, during which they were clinically and histologically diagnosed with ESCC. The histological grade and clinical stage of the tumors were recorded according to the 8th edition of the Tumor-Node-Metastasis (TNM) classification of the International Union Against Cancer (2018) (28). A patient was selected as a qualifiable subject if the following five requirements were met: i) The patient was diagnosed with histologically confirmed primary ESCC, but had not received any previous treatment elsewhere; ii) the patient had no family history of cancer; iii) the patient had undergone radical surgery with lymphadenectomy (limited or extended); iv) the patient had not received neoadjuvant or adjuvant treatments; and v) the clinical information and follow-up data of the patient were documented. As a result, 253 patients constituted the complete set of specimens, and their clinical data were obtained from hospital records. The patients were followed up in 2019 to ascertain the latest disease status.

A tissue microarray was constructed according to previously specified methods (29). Samples from the 253 patient specimens were fixed in formalin for 24 hours at room temperature and then embedded in paraffin. The paraffin blocks were sliced into 3-µm sections. The sections were then reviewed by a senior pathologist and representative tumor regions were defined by hematoxylin and eosin staining. Next, two targeted core samples from each tissue specimen were obtained using a tissue array instrument (MiniCore^® instruments; ALPHELYS). Tissue cylinders with a diameter of 10 mm were punched and arrayed in a recipient paraffin block. Sections of the tissue array were then cut and arranged on glass slides.

Following paraffin embedding, the tissue sections were dried in incubator at 60°C for 2 h, then deparaffinized in xylene and rehydrated in a descending graded alcohol series (100, 95, 90, 85 and 75%), and then incubated in 0.3% hydrogen peroxide for 15 min to block endogenous peroxidase activity. Antigen retrieval was conducted by heating in EDTA buffer (pH 8.0) for 3 min in a pressure cooker. The sections were blocked for 10 min with 10% normal goat serum (cat. no. ab7481; Abcam) at room temperature, and then incubated with a polyclonal antibody against eIF4A2 (dilution, 1:1,000; cat. no. ab31218; Abcam) for 12 h in a moist chamber (4°C). Control samples were incubated in blocking solutions without the primary antibody. The slides were then incubated with horseradish peroxidase for 30 min and visualized using 3,30-diaminobenzidine solution. Mayer's hematoxylin was applied as a counterstain at room temperature for 1 min. Negative control staining was performed by replacing the primary antibody with normal murine immunoglobulin G (1:200; cat. no. A7028; Beyotime) for 12 h in a moist chamber at 4°C. eIF4A2-positive slides were used as positive controls, and the internal negative controls consisted of normal squamous esophageal mucosal samples from cancer-free patients.

Next, two independent pathologists (Lyu and Yan) blinded to the clinicopathological data evaluated eIF4A2 expression in the tissue specimens with a light microscope (magnifications, ×100 and ×200). The evaluation approach followed two scoring criteria: i) The positive cell proportion score was assigned according to the percentage of positive tumor cells: 0 (0%), 1 (1–10%), 2 (11–50%), 3 (51–70%) and 4 (71–100%); and ii) the staining intensity score was graded in accordance with the signal intensity: 0 (no signal), 1 (weak), 2 (medium) and 3 (strong). The immunoreactivity score (IRS) for eIF4A2 expression was calculated by multiplying the positive cell proportion and the staining intensity scores (range, 0–12). The specimens were re-evaluated if the scores submitted by the two pathologists displayed discrepancy. In addition, eIF4A2 expression data were downloaded from UALCAN (http://ualcan.path.uab.edu/index.html), an interactive web portal for the in-depth analysis of The Cancer Genome Atlas (TCGA) gene expression data.

The survival receiver operating characteristic (ROC) package in R (R software version 3.0.1 and the survival ROC package; R Foundation for Statistical Computing; www.R-project.org) was used to determine the IRS cut-off values. The sensitivity and specificity for each outcome were plotted and ROC curves were generated. The score was selected as the cut-off value which was closest to the point with both maximum sensitivity and specificity. For eIF4A2 expression, the cut-off value for the staining intensity score (0–3), proportion score (0–4) and IRS (0–12) were assessed separately. Clinicopathological features, including age (68 years old or >50 years old), sex (male or female), histological grade (1–2 or 3), tumor status (T1, T2, T3 or T4), nodal status (N0 or N1) and TNM stage (1–2 or 3) were also evaluated.

Survival analysis was performed using SPSS version 25.0 (IBM Corp.). Pearson's χ² test was used to determine the association between eIF4A2 expression and patient clinicopathological features. Disease-free survival (DFS) was defined as the time between surgery and regional relapse or the development of distant metastasis. Overall survival (OS) was defined as the time between surgery and death. The Kaplan-Meier estimator was used to assess DFS and OS, the results of which were further compared using the log-rank test. Multivariate survival analysis was performed using the Cox regression model for all variables that were identified as significant during univariate analysis. For all analyses, P<0.05 (two-sided) was considered to indicate a statistically significant difference.

Of the study population, 59 patients were diagnosed with well-differentiated squamous cell carcinoma (SCC), 163 patients presented with moderately-differentiated SCC and 31 with poorly-differentiated SCC. The number of patients confirmed to have T1, T2, T3 and T4 disease was 6, 61, 183 and 3, respectively, and 118 of the 253 patients exhibited nodal metastases. With regards to TNM staging, 156 patients were diagnosed with stage 1 and 2 disease, and 97 patients presented with stage 3 disease. The mortality rate across all patients was 52.6% (133 out of 253), and the median OS and DFS times were 40 (4–115) and 38 (1–115) months, respectively.

The results of the present study demonstrated that eIF4A2 expression is localized to the cytoplasm of ESCC cells. eIF4A2 was also revealed to be more highly expressed in esophageal carcinoma tissues than in the normal control samples, displaying a similar expression trend to that determined by TCGA database analysis, though not statistically significant (Figs. 1 and 2). In the present study, samples with a signal intensity score of 3 (strong) were considered to exhibit high eIF4A2 expression, while those with a score of 1 (weak) or 2 (medium) were considered to exhibit low expression intensity. High eIF4A2 expression was detected in 53 patients, while the remaining 200 samples presented with low eIF4A2 expression. None of the samples scored 0 (Fig. 2). The cut-off value for the proportion score (0–4) was 1, indicating that samples with a score ≤1 were considered to exhibit a low proportion of eIF4A2 expression, and those with a score of >1 were regarded to express a high proportion of eIF4A2. There were 10 patients with a low, and 243 patients with a high proportion of eIF4A2 expression. For the IRS (0–12), ESCC samples with a score >9 were defined as highly eIF4A2-immunoreactive. Therefore, in the present study, 214 patients possessed high eIF4A2 immunoreactivity and the remaining 39 patients presented with low immunoreactivity. However, Kaplan-Meier analysis revealed that only the intensity score was significantly associated with DFS and OS, hence a high eIF4A2 intensity was defined as high eIF4A2 expression, and low intensity samples were defined by low eIF4A2 expression (Table I).

There were no statistically significant associations between the general clinicopathological characteristics of the 253 patients and eIF4A2 expression (Table II).

According to the results of Kaplan-Meier analysis, specific clinicopathological characteristics, including age, histological grade, tumor, nodal and TNM staging were revealed to be significantly associated with DFS and OS time(P=0.002; Fig. 3C and D); the cut-off value for age was set at 68 years. eIF4A2 expression was also revealed to be associated with the prognosis of ESCC. The median DFS time was 40 months for patients with low eIF4A2 expression and 16 months for those with high eIF4A2 expression (P<0.001; Fig. 3A). In addition, the median OS time was 48 months for patient with low eIF4A2 expression, but 25 months for those with high eIF4A2 expression (P<0.001; Fig. 3B). When patient specimens were stratified into N stage subgroups, the association between eIF4A2 expression and DFS or OS became more significant. In the N0 subgroup, the median DFS and OS times of patients with low eIF4A2 expression were 62 and 65 months, respectively, while the median DFS and OS times of those with high eIF4A2 expression were both 40 months (both P=0.002). In the N1 subgroup, the median DFS and OS times of patients with low eIF4A2 expression were 19 and 27 months, respectively, while the median DFS and OS times of those with high eIF4A2 expression were 10 and 15 months, respectively (both P=0.002). These data indicate that high expression levels of eIF4A2 are associated with a poor prognosis in patients with ESCC (Fig. 3).

Univariate analysis was performed using the Cox proportional hazards model to assess the importance of multiple factors on the survival times of patients with ESCC. The results indicated that age, histological grade, tumor and nodal status, TNM stage and eIF4A2 expression are significantly associated with DFS and OS time (Table I). Multivariate analysis was then applied to investigate these identified parameters. The results verified that only eIF4A2 expression and age were independent prognostic factors for DFS and OS time in ESCC (Table III).