For the present study, the cases were retrospectively selected from the University of São Paulo Gastric Cancer Database: 152 patients with gastric adenocarcinoma operated at the Stomach and Small Bowel Unit Hospital das Clinicas, University of São Paulo, School of Medicine, São Paulo, Brazil, between February 1993 and December 2002. In total 416 different gastric tissues were analysed, including 163 normal mucosa, 59 metaplastic mucosa, 152 primary tumors and 42 lymph node metastasis samples. The samples were retrieved from the files of the Department of Pathology, and organized in tissue microarrays (TMAs). To achieve representative sampling and minimize sample loss, each case was included in duplicate in the TMAs. Additionally, formalin-fixed and paraffin-embedded prostate carcinoma and GISTs samples were used as positive and negative controls for immunohistochemistry, as previously described (27).

Relevant patient clinical data available included patients age, gender, tumor size and location, WHO classification, Lauren’s classification, TNM staging, depth of invasion, lymph node metastases and lymphatic, vascular and neural invasions, desmoplasia, inflammatory infiltrated and follow-up, as previously described (34). The inclusion criteria were: patients with primary gastric adenocarcinoma submitted to subtotal or total gastrectomy with D2 lymphadenectomy, submucosa layer invasion or deeper (T1b or higher), at least 25 lymph nodes retrieved per case, absence of distant metastasis (M0), and available follow-up data. All patients were treated according to a well-established surgical protocol following the Japanese Gastric Cancer Association rules.

The histological sections, obtained from original paraffin blocks and, after hematoxylin and eosin staining, were submitted to histological review by senior pathologist. Primary tumors were histologically classified according to World Health Organization (35) in tubular, intestinal, signet-cell and mucinous carcinomas. Lauren and Ming (36,37) classifications were also applied to primary tumors, stratifying lesions into diffuse or intestinal or infiltrating or expansive types, respectively, the latter derived from evaluation of the deepest tumor edge. Lymphatic, vascular or perineural invasion was assessed as non-detected or present. Pathological nodal status (pN) was determined histologically by counting the affected lymph nodes and classified as pN0, pN1, pN2 or pN3 according to AJCC TNM Staging system (38). Information on the tumor size as well its main location, classified as proximal or distal tumors, were obtained from original surgical pathology reports. Peri/intratumoral inflammatory infiltrate was semi-quantified as absent, mild/moderate and intense whereas desmoplasmatic stromal response as absent to mild or moderate to intense. Final pathological TNM stage was also assessed.

Tissue microarray (TMA) slides with 3 μm-thick sections were subjected to immunohistochemical analysis according to the streptavidin-biotin peroxidase complex system (UltraVision Large Volume Detection System Anti-Polyvalent, HRP; LabVision Corporation, CA, USA), as previously described (27,30,39). Briefly, deparaffinised and rehydrated slides were submitted to heat-induced antigen retrieval for 20 min at 98°C with 10 mM citrate buffer (pH 6.0). After incubation with the primary antibody raised against RKIP (dilution 1:800; incubation 2 h at RT; Upstate Biotechnology, Lake Placid, NY, USA), the secondary biotinylated goat anti-polyvalent antibody was applied for 10 min followed by incubation with the streptavidin-peroxidase complex. The immune reaction was visualized by 3,3′-diamonobenzidine (DAB) as a chromogen. All sections were counterstained with Gill-2 haematoxylin. For negative controls, primary antibodies were omitted and also replaced by a universal negative control antibody (CEA, rabbit anti-human, Dako Corporation, Carpinteria, CA, USA). Prostate carcinoma and GISTs cases, previously analysed (27,30,39), were used as positive and negative controls. Sections were independently scored by two of the authors (K.S. and A.L.F.), following a semi-quantitative criterion: (−), 0% of immunoreactive cells; (+), <5% of immunoreactive cells; (++), 5–50% of immunoreactive cells; and (+++), >50% of immunoreactive cells. Samples with scores (−) and (+) were considered negative, and those with scores (++) and (+++) were considered positive.

Either the χ² test or Fisher’s exact test (when required) was done to determine the correlation between RKIP expression status and clinicopathologic features. Survival curves were estimated using the Kaplan-Meier product-limit method, and the significance of the differences between survival curves was determined using a log-rank test. Multivariate survival analyses were done using the Cox proportional hazards model. Results were considered to be statistically significant for p<0.05. All statistical analyses were conducted using SPSS 16.0 statistical software program.

We have available tissue in 152 patients, 114 with advanced and 38 early gastric cancer. The mean age was 60.89 (range: 26–87 years). The majority of patients had tumors located at the distal part of the stomach. The mean tumor size was 3.92 (range: 0.8–20 cm). There was a predominance of the Lauren intestinal type. All patients were submitted to D2 lymphadenectomy with a mean number of retrieved lymph nodes of 43.7 (range: 25–113). The mean follow-up was 62.3 months (range: 6–76 months), and 13.4% was lost to follow-up.

Four hundred and sixteen gastric tissue samples were studied for RKIP expression by immunohistochemistry. RKIP staining was always present in the cytoplasm of the cells and, according to the immunohistochemistry score, we found RKIP-positive expression in 84.7% (138/163) of normal gastric tissue, in 79.7% (47/59) of gastric mucosa metaplasia, in 44.1% (67/152) of gastric primary tumors, and in 9.5% (4/42) of gastric lymph node metastasis (Table I and Fig. 1). There are no statistically significant differences between RKIP expression in normal and metaplasias (p=0.377). However, a statistically significant (p<0.001) decrease of RKIP expression was found in primary tumors, compared to normal and metaplasia tissues, and in lymph-node metastasis, when compared to normal, metaplasia and primary tumors (Table I). Additionally, in 35 patients we had normal, tumor and metastatic tissues available. We found that RKIP was positive in non-neoplastic tissues in ~83% of the patients, but was absent ~69% of the primary tumors and completely absent in metastatic tissues (data not shown).

The correlations between RKIP expression and clinicopathologic features are summarized in Table II. We found that RKIP is differently expressed between the different WHO histological types (p=0.03), being highly expressed in intestinal type, and lost in tubular, mucinous and signet-ring cell carcinomas. At variance, no statistically differences were observed among the Lauren subtypes. RKIP is significantly lost in advanced gastric cancer when compared with early tumors (p<0.001). Additionally, absence of RKIP expression was statistically associated with tumors with higher tumor size, with higher TNM stage, with the presence of vascular, lymphatic and neuronal invasion and with the presence of lymph node metastasis (Table II).

We found that additionally to the above mentioned clinical factors, the absence of RKIP expression is also significantly (p<0.001) associated with poor overall survival in gastric carcinoma patients (Table III and Fig. 2). To evaluate whether RKIP expression is an independent prognostic factor, we carried out a multivariate Cox regression analysis and found that absence of RKIP expression is independently associated with patients poor survival with a 4.53 hazard ratio (Table III). Additionally, male gender, higher tumor size and presence of lymph node metastasis were also found to be independent prognostic factors in our series of gastric carcinomas (Table III).