CRCs were collected from 567 patients who underwent surgical resection at Seoul National University Hospital (SNUH) between 2004 and 2006. Clinicopathological data of patient age, sex, tumor location, size, histological subtype, differentiation, presence of lympho-vascular invasion, American Joint Committee on Cancer/International Union against Cancer (AJCC/UICC) cancer stage (seventh edition), time of death, tumor recurrence, and follow-up time were obtained by reviewing the clinical and pathological reports. Fifty-six CRC samples were obtained from patients at Jeju National University Hospital (JNUH); 22 cases were CRCs arising from preexisting adenomas, and 34 cases were conventional CRCs with lymph node metastasis. The histopathologic features were evaluated by two gastrointestinal pathologists (J.M.B and G.H.K) for CRCs from SNUH and by one pathologist (B.G.J) for CRCs from JNUH. In addition, 32 paired, fresh-frozen CRC tissues and matched normal tissues were obtained from the Jeju National University Hospital Biobank, a member of the National Biobank of Korea, for which informed consent was obtained from all subjects. All procedures were in accordance with the Helsinki Declaration of 1964 and later versions, and this study was approved by the Institutional Review Boards of SNUH (IRB No. 1502-029-647) and JNUH (IRB No. 2017-04-001).

Thirteen tissue microarrays (TMAs) containing 642 CRCs from SNUH were generated as previously described [16]. In brief, through histologic examination, the representative cancer areas containing more than 70% of cells were marked in each case. For each CRC, core tissue biopsies (2 mm in diameter) were obtained from individual formalin-fixed paraffin-embedded samples (donor blocks) and arranged in a new recipient paraffin block (tissue array block) using a trephine apparatus (SuperBioChips Laboratories, Seoul, Korea). With CRCs from JNUH, two TMAs including 22 pairs of adenoma and carcinoma areas and five TMAs including 34 ulcerofungating CRCs with lymph node metastasis were constructed with 4 mm cores. For ulcerofungating CRCs, superficial and invasive portions, and metastatic cancers in the lymph node were included.

Immunohistochemistry was performed on 4-μm TMA sections using a Ventana BenchMark XT Staining systems (Leica Microsystems, Wetzlar, Germany) according to the manufacturer’s instructions. The primary antibodies used were anti-EPHB2 (1:700, R&D Systems, Minneapolis, MN, USA), anti-CD44 (1:100, Novocastra Laboratories Ltd., Newcastle upon Tyne, UK), and anti-β-catenin (1:800, 17C2, Novocastra Laboratories). The expression of EPHB2 and CD44 was determined by examining the tumor cell membrane. For each tumor, the intensity and percentage of tumor cells expressing EPHB2 (n = 567) or CD44 (n = 87) were evaluated. Histoscores (H-scores) were calculated by multiplying the intensity score (0, negative; 1, weak; 2, moderate; and 3, strong) and percentage of positive tumor cells (range, 0 to 100), ranging from 0 to 300. For statistical analyses for EPHB2, we used a cutoff of 40 on the basis of the distribution of the H-scores (median value, 40). CRCs with H-score of 40 or lower were classified as negative, while cases with H-score higher than 40 were classified as positive. β-Catenin staining was considered positive when more than 10% of tumor cell nuclei were strongly stained.

Total RNA was extracted from 32 paired fresh-frozen CRCs and matched non-cancerous colon tissues using TRIZOL reagent (Invitrogen, Carlsbad, CA, USA). RNA (1 μg) was subjected to reverse-transcription with oligo-dT primers and the GoScript reverse transcription system (Promega, Madison, WI, USA). Complementary DNA was used to conduct real-time polymerase chain reaction (PCR) with Premix EX Taq (Takara Bio, Shiga, Japan) following the manufacturer’s instructions. The cycling conditions were as follows: initial denaturation for 30 seconds at 95˚C, followed by 40 cycles of 95˚C for 1 second and 60˚C for 5 seconds in a StepOne Plus real-time PCR system (Applied Biosystems, Foster City, CA, USA). The TaqMan expression assays were performed for the following genes: Hs00362096_m1 (EPHB2), Hs00610344_m1 (AXIN2), Hs02379687_s1 (CD24), Hs01075684_m1 (CD44), Hs01009250_m1 (PROM1/CD133), Hs00233455_m1 (ALCAM/CD166), and Hs02786624_g1 (GAPDH). GAPDH served as the endogenous control.

Thorough microscopic examination, representative cancer areas in each case were marked, microdissected, and then incubated at 55˚C with lysis buffer and proteinase K for 2 days. KRAS codons 12 and 13 and BRAF codon 600 mutations were determined by PCR-restriction fragment length polymorphism and direct sequencing techniques. Among the 567 CRCs, 29 and five samples were excluded from the KRAS and BRAF mutation analyses, respectively, due to an insufficient DNA amount. The microsatellite instability (MSI) status of each cancer tissue was determined using five National Cancer Institute recommended markers (BAT25, BAT26, D2S123, D5S346, and D17S250). The MSI status of each case was classified into three groups: MSI-high (two or more unstable markers among the five markers), MSI-low (one unstable marker), or microsatellite stable (no unstable marker).

CpG island methylator phenotype (CIMP) status was determined by MethyLight assay. Sodium bisulphite modification of genomic DNA samples was conducted for all 567 CRC tissues, as previously described [17]. The methylation statuses of eight CIMP-specific CpG islands (MLH1, NEUROG1, CRABP1, CACNA1G, CDKN2A [p16], IGF2, SOCS1, and RUNX) were quantified. CIMP-high was defined as having five or more hypermethylated markers, and CIMP-low was defined as having four or fewer hypermethylated markers out of eight. CIMP-negative was defined as having no hypermethylated marker.

All data were analyzed using SPPSS statistical software ver. 18.0 (SPSS Inc., Chicago, IL, USA) and Prism ver. 5.0 (Graph-Pad Software, San Diego, CA, USA). The correlations between EPHB2 and cancer stem cell markers or CD44 were evaluated by the Spearman correlation test. Between-group comparisons of the real-time PCR data were performed using Student’s t-test. The correlations between EPHB2 positivity and clinicopathological parameters were tested using Pearson’s chi-square test or Fisher exact test. Survival curves were estimated using the Kaplan-Meier method, and log-rank test was used to compare groups. Using Cox proportional hazards model, multivariate analyses were performed to identify independent prognostic factors. A p-value of < .05 was considered statistically significant.

To measure the expression level of EPHB2 in CRCs, we performed real-time PCR with 32 pairs of fresh-frozen human CRC samples and adjacent normal colon tissues. Comparing individual cases, EPHB2 mRNA expression was higher in CRC samples than in non-cancerous tissues in most samples (77%, 25 out of 32 cases) (Fig. 1A), and the mean EPHB2 level was also significantly higher in CRCs than in matched colon tissues (p < .01) (Fig. 1B). EPHB2, a Wnt signaling target gene, showed a strong positive correlation with AXIN2, a direct target of the Wnt pathway (p < .001), confirming enhanced Wnt signaling activity in CRCs (Fig. 1C). As EPHB2-high cancer cells have been suggested as a cancer stem cell population in CRCs [18], we investigated the correlations between EPHB2 expression and expression of other candidate cancer stem cell (CSC) markers CD24, CD44, CD133, and CD166. Notably, we found that only CD44 showed a positive association with EPHB2 (p = .030) (Fig. 1D–G), which was confirmed at the protein level by a positive correlation between EPHB2 and CD44 H-scores in CRCs (n = 87) (Fig. 2).

To explore the expression profile of EPHB2 expression during CRC progression, we collected 22 cases of CRCs arising from pre-existing adenomas and measured the H-scores of EPHB2 in adenoma and carcinoma portions (Fig. 3A). However, no difference in H-scores was observed between the adenoma and carcinoma areas (Fig. 3B). Next, we collected 34 ulcerofungating CRCs with lymph node metastasis and examined the H-scores in three spots for each cancer: the superficial fungating area, invasive fronts, and metastatic cancer cells (Fig. 3C, D). EPHB2 expression was slightly lower in invasive fronts compared to surface area (p < .05), but did not differ between superficial and metastatic areas (Fig. 3E).

To investigate the prognostic value of EPHB2 in CRCs, immunohistochemical staining was performed on 12 TMAs, and 567 cases were included for the analysis. The clinicopathological relevance of EPHB2 positivity is summarized in Table 1. EPHB2 expression was significantly higher in CRCs from old patients (p = .002). EPHB2 positivity exhibited negative correlations with lymphatic (p < .001) and venous (p = .001) invasion, T category (p < .001), N category (p < .001), M category (p = .013), and tumor stage (p < .001). Nuclear β-catenin staining was observed more frequently in EPHB2-positive CRCs (Table 1, Fig. 4). EPHB2 positivity was higher in CRCs with MSI-high than in those with MSI-negative/low, whereas EPHB2 expression had no correlation with mutation status of BRAF, KRAS, or CIMP (Table 2). Kaplan-Meier analysis demonstrated that CRC patients with EPHB2 positivity showed better clinical outcomes in both overall (p = .048) (Fig. 5A) and recurrence-free survival (p = .015) (Fig. 5B). However, multivariate analysis failed to show EPHB2 as an independent prognostic marker in CRCs (hazard ratio, 0.692; p = .692) (Table 3).