This retrospective study involved 210 patients with thoracic oesophageal cancer who underwent surgical resection in our hospital from 1998 to 2007. All 210 patients were confirmed to have OSCC by histopathological examination of endoscopic biopsy and considered suitable for surgery based on preoperative assessment. They included 23 females and 187 males, aged between 38 and 82 (median, 63.1 years). Table 1 lists patient characteristics. Of the total, 110 patients with cN1 received neo-adjuvant chemotherapy (NACT), which consisted of two courses of 5-fluorouracil (5-FU), cisplatin (CDDP), and adriamycin (ADM) (Akita et al, 2006; Yano et al, 2006; Matsuyama et al, 2007; Makino et al, 2008). Curative resection (R0), that is, oesophagectomy with two- or three-field lymphadenectomy, was performed in 199 patients (94.8%), whereas non-curative resection (R2) was carried out in the remaining 11 patients (5.2%); these patients were excluded from the survival analysis. None of the patients died of postoperative complications. Sixty-five patients with multiple metastatic lymph nodes in the surgical specimen received docetaxel or CDDP plus 5-FU postoperatively (Ando et al, 2003). In our hospital, patients with OSCCs that invade the airway or major blood vessels or accompanied by visceral metastasis are not indicated for surgery and therefore receive chemoradiotherapy or chemotherapy alone.

After surgery, the patients were surveyed every 3 months by physical examination and serum tumour markers, every 6 months by CT scan and abdominal ultrasonography, and every year by endoscopy until tumour recurrence was evident. Patients with tumour recurrence received chemotherapy or chemoradiotherapy, as long as their systemic condition permitted. The mean overall survival (OS) was 34.9 months and mean disease-free survival (DFS) was 22.2 months. The mean follow-up period after surgery was 38.8 months. This study was approved by the Human Ethics Review Committee of Osaka University School of Medicine and a signed consent form was obtained from each subject.

Total RNA was extracted from fresh frozen tissue of resected tumours in 21 out of total 210 OSCCs using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). Complementary DNA (cDNA) was generated from 1 μg RNA in a final volume of 20 μl, containing oligo-(dT)₁₅ primer, avian myeloblastosis virus transcriptase, with reverse transcription (RT) system (Promega, Madison, WI, USA). Polymerase chain reaction (PCR) analysis was performed by using LightCycler, a real-time monitoring thermal cycler. Polymerase chain reaction reaction mixture was prepared containing 2 μl of cDNA template, 3 mmol l⁻¹ MgCl₂, 250 nmol l⁻¹ of primer pairs, using LightCycler FastStart DNA Master SYBR Green I (Roche Diagnostics, Mannheim, Germany). The amount of each transcript was normalised against the expression of the housekeeping gene, glydecaldehyde-3-phosphate-dehydrogenase (GAPDH). Standard curve was constructed with 10-fold serial dilutions of cDNA obtained from non-cancerous oesophageal mucosal cell layers of tissue samples from 10 cases as a standard mixture. The sequences of PCR primers for GAPDH, CK18, and CK8 were as follows: forward primer 5′-CAACTACATGGTTTACATGTTC-3′, reverse primer 5′-GCCAGTGGACTCCACGAC-3′ used for amplification of GAPDH, forward primer 5′-ATCTTGGTGATGCCTTGGAC-3′, reverse primer 5′-CCTGCTTCTGCTGGCTTAAT-3′ for CK18, and forward primer 5′-TAGCACTGGGAACAGGAGA-3′, reverse primer 5′-TTTGACATTGGCAGAGCTA-3′ for CK8. The PCR cycling condition was set as follows: an initial denaturing step at 95°C for 10 min and 40 cycles at 95°C for 15 s, 58°C for 10 s, and 72°C for 25 s. The relative amount of cDNA in each sample was measured by interpolation on the standard curve, and then the relative ratio of CK18/GAPDH mRNA or CK8/GAPDH mRNA expression in log2 scale was calculated for each OSCC sample.

The amounts of CK18 and CK8 proteins in the tissues were examined by immunohistochemical staining of formalin-fixed and paraffin-embedded serial sections. One representative slide with the deepest tumour invasion was selected from each patient and subjected to immunohistochemistry using the streptavidin–peroxidase method. Briefly, after deparaffinisation in xylene and dehydration in graded ethanol, endogenous peroxidase activity was blocked by incubation with 30 ml l⁻¹ hydrogen peroxide for 20 min. Then tissue sections were heated for 40 min at 95°C in citrate buffer (0.05 mol l⁻¹, pH 6.0) for antigen retrieval. After incubation with mouse monoclonal primary antibody DO10 (Bartek et al, 1991) (Novocastra, Newcastle, UK, dilution 1 : 40) for CK18 and TS1 (Johansson et al, 1999) (Novocastra, dilution 1 : 200) for CK8 for 14 h, the sections were stained by the labelled streptavidin biotin method. Negative controls of immunohistochemical reactions were performed by omitting the primary antibody. Positive staining of normal oesophageal gland at the non-cancerous area in the same section was used as an internal positive control. Furthermore, CK18 immunoreactivity was examined using the same antibody and methods in pretreated fiberscopic biopsy samples obtained from 83 of the 210 patients. The presence of CK18 or CK8 protein was judged ‘positive' when the proportion of immunohistochemically stained cells was more than 50% of all observed cancer cells (Lam et al, 1995), or otherwise ‘negative'. All slides were assessed by two observers independently and then in conference in a blinded manner without any prior knowledge of the clinicopathological parameters. Characteristically, immunostaining of CK family of proteins yields clearly recognised tumour cells with sufficient intensity and frequency, with little or no background or non-specific staining. Therefore, the judgment of immunostaining was always consistent between the two observers and the results were the same on different cut-off lines around 10–50% of the proportion of immunohistochemically stained cells.

Western blot analysis was performed to confirm the specificity of CK18 and CK8 antibodies (Miyata et al, 2001). Briefly, the protein extracted from tissue samples (20 μg), MCF7 whole cell lysates (0.5 μg) and recombinant protein of each CK18 and CK8 (ProSpec-Tany Technogene, Rehovot, Israel) (0.1 μg) were separated using 7.5% polyacrylamide gel electrophoresis, followed by electroblotting onto a polyvinylidene difluoride membrane. The membrane was incubated with the primary antibodies at appropriate concentrations (anti-CK18 1 : 200 dilution, anti-CK8 1 : 500 for 12 h at 4°C, and anti-actin 1 : 2000 for 1 h at room temperature). Protein bands were detected using the Amersham-enhanced chemiluminescence detection system (Amersham Biosciences Corp., Piscataway, NJ, USA).

Data are expressed as mean±s.d. Differences in continuous parameters between two groups classified by CK18 or CK8 protein expression were evaluated by the Mann–Whitney's U test. Correlations between CK18 expression on immunohistochemistry and various clinicopathological parameters were evaluated by the χ² test and Fisher's exact probability test. Regression analysis was used to determine the correlation between CK18 and CK8 gene expressions using Pearson's correlation coefficient (R value). Prognostic variables were assessed by the log-lank test, and DFS and OS were analysed by the Kaplan–Meier test. Cox's proportional hazard regression model with stepwise comparison was used to analyse the independent prognostic factor(s). All statistical analyses were carried out using SPSS for Windows release 10 (SPSS, Inc, Chicago, IL, USA). A P-value of <0.05 was accepted as statistically significant.

Non-cancerous squamous epithelium showed no immunohistochemical staining for both CK18 and CK8, but adjacent proper oesophageal glands always showed strong immunostaining for both, which served as an internal positive control (Figure 1A and B). All 210 samples, including cancer and non-cancerous lesions in the same section, were evaluated for CK18 and CK8 protein expression by immunohistochemical analysis (IHC), respectively. Positive CK18 expression was identified in 90 (42.9%) cases, and the staining was mainly observed in the cytoplasm of tumour cells. The remaining 120 (57.1%) cases were negative (Figure 1C and D). Furthermore, 85 (40.5%) cases showed positive immunostaining for CK8 in the cytoplasm of tumour cells, whereas 125 (59.5%) were negative (Figure 1E and F). The positive staining for CK18/8 was almost homogeneous at single cancer nest and among different areas (surface, central, and deepest areas) of the cancer lesion. There was a significant correlation between CK18 and CK8 immunostaining (P<0.001); 69 (32.9%) patients showed positive immunostaining for both CK18 and CK8, with largely matching distribution, whereas 104 (49.5%) were negative for both. On the other hand, 37 (17.6%) patients showed discordant immunostaining, including CK18 positive and CK8 negative in 21 cases, and CK18 negative and CK8 positive in 16 (Table 2). Intra-epithelial neoplasia (dysplasia) was observed in 55 cases. Among them, CK18 and CK8 expressions were detected in 13 (23.6%) and 14 (25.5%) cases, respectively (Figure 1G and H).

Western blot analysis demonstrated strong expression of CK18 and CK8, with molecular weights of 45 and 52.5 kDa, respectively, in whole cell lysates of MCF7 and recombinant proteins of CK18 and CK8. In cancer tissue obtained from one representative patient positive for both CK18 and CK8 expression by immunohistochemistry, CK18 and CK8 were strongly expressed, but not in normal squamous epithelial cells (Figure 2).

Quantitative RT–PCR analysis was performed in 21 representative cases to elucidate the mechanisms of gene transcription and the relationship between protein and mRNA expression levels of these molecules. It showed the expression level of CK18 mRNA in CK18-positive tumours (n=5) was significantly higher than that in CK18-negative cases (n=16) (1.78±0.92 vs −0.001±0.87, P=0.005). Similarly, the expression level of CK8 mRNA in CK8-positive tumours (n=6) was significantly higher than that in CK8-negative cases (n=15) (3.08±1.46 vs 0.84±1.75, P=0.016) (Figure 3). Regression analysis showed a significant correlation between CK18 and CK8 mRNA expression (R=0.822, R2=0.675, P<0.001).

Table 3 lists the correlations between CK18 and CK8 expression and various clinicopathological parameters. In comparison with CK18-negative OSCCs, the proportion of CK18-positive tumours was significantly higher among moderately poorly differentiated OSCCs (65.8% vs 92.2%, respectively, P<0.001), patients treated with NACT (45.0% vs 62.2%, respectively, P=0.018), advanced pathological T stage (pT3,4) (58.3% vs 73.3%, respectively, P=0.029), large number (⩾4) of pathologically positive lymph nodes (21.7% vs 44.4%, respectively, P=0.001), and advanced pathological stage (pStage III/IV) (45.0% vs 70.0%, respectively, P=0.0045). On the other hand, there were no significant correlations among CK18 expression and other parameters listed in Table 3, such as age, gender, tumour location, and clinical response to NACT.

Similarly, compared with CK8-negative tumours, the proportion of CK8-positive cases was significantly higher among moderately poorly differentiated OSCCs (65.6% vs 94.1%, respectively, P<0.001), patients treated with NACT (46.4% vs 61.2%, respectively, P=0.0485), and large number (⩾4) of pathologically positive lymph nodes (25.6% vs 40.0%, respectively, P=0.0340). On the other hand, there were no significant correlations among CK8 expression and other parameters listed in Table 3, such as age, gender, tumour location, pT, pStage, and clinical response to NACT.

Disease recurrence after curative resection was diagnosed in 90 (45.2%) of 199 patients with curative resection (R0) and the mean time to recurrence was 8.8 months. Deaths because of primary cancer occurred in 83 (39.5%) of the 210 patients and the mean time between surgical resection and death was 15.3 months. The 5-year DFS and OS rates for all patients were 51.6% and 54.0%, respectively. Patients with CK18-positive tumours had significantly poorer DFS and OS than those with CK18-negative tumours (5-year DFS: 30.6% vs 65.8%, respectively, P<0.001, 5-year OS: 32.8% vs 68.4%, respectively, P<0.001) (Figure 4A). CK8-positive tumours were also significantly associated with poorer DFS and OS than CK8-negative ones (5-year DFS: 40.0% vs 59.0%, respectively, P=0.043, 5-year OS: 39.6% vs 63.9%, respectively, P=0.017) (Figure 4B). There was significant prognostic difference in survival between CK18-positive and -negative groups according to pStage II (5-year DFS: 33.6% vs 70.6%, respectively, P=0.033) and pStage III (5-year DFS: 27.8% vs 67.9%, respectively, P=0.006), but not at pStage I (5-year DFS: 100% vs 92.3%, respectively, NS) or pStage IV (5-year DFS: 18.2% vs 35.0%, respectively, P=0.579) (Figure 4C).

Univariate analysis using Cox's proportional hazard model showed that the following parameters correlated significantly with DFS: pT stage, number of pathologically positive lymph nodes (pN number), lymphatic invasion (ly), venous invasion (v), CK18 expression, and CK8 expression (Table 4). Finally, the above parameters were entered into multivariate analysis. The results showed that pT, pN number, and CK18 expression significantly and independently influenced prognosis (i.e., DFS) (HR=1.909, P=0.020; HR=2.095, P=0.001; and HR=1.879, P=0.004, respectively, Table 4).

In 83 patients, fiberscopic biopsy samples obtained before treatment, which included 2–5 tissue pieces in each patient, were also investigated for CK18 expression. Immunostaining for CK18 was basically similar among the obtained tissue pieces. Immunostaining showed CK18 expression in 47 (56.6%) cases (Figure 1I). The classification of CK18 immunostaining matched that of surgical specimens in 67 (80.7%) cases. Survival analysis also showed that pretreated CK18-positive tumours were also significantly associated with poorer OS than biopsy CK18-negative ones (5-year OS: 30.4% vs 62.2%, respectively, P=0.045) (Figure 5).