PROX1 and β-catenin are prognostic markers in pancreatic ductal adenocarcinoma

This study is based on a series of 189 consecutive PDAC patients surgically treated in 2000–2011 at the Department of Surgery, Helsinki University Hospital. Only patients with verified PDAC were included in this study. Median age at operation was 64 (range 39–84) years. Twenty-one patients, who received neoadjuvant chemotherapy, were excluded from the study. Eight patients were eventually diagnosed with stage IV disease with distant metastases according to the American Joint Committee on Cancer Pancreatic Cancer Staging System [30], and four patients lacked data on stage. They were excluded from the study. Altogether, 156 patients were included in the study. Patients’ records, the Finnish Population Registry and Statistics Finland were used to obtain survival data and cause of death of the patients. A description of the study cohort is in Table 5.

Formalin-fixed and paraffin-embedded surgical tissue samples were collected from the archives of the Department of Pathology, Helsinki University Hospital. Experienced pathologists (J.H. and S.N.) re-evaluated all samples for confirmation of the histopathological diagnosis of PDAC. Representative regions of tumor specimens were defined and tumor areas were marked on hematoxylin- and eosin-stained tumor slides for preparation of tissue microarray blocks (TMA). Two 1.0-mm cores were taken from each tumor block with a semiautomatic tissue microarrayer (Tissue Arrayer 1, Beecher Instruments Inc., Silver Spring, MD, USA). In order to evaluate TMA representativeness compared to whole tissue blocks, we examined altogether six spots per patient taken from different areas/parts of the tumor.

TMA blocks were freshly cut into 4-μm sections. After deparaffinization in xylene and rehydration through a gradually decreasing concentration of ethanol to distilled water, slides were treated in a PreTreatment module (Lab Vision Corp., Fremont, CA, USA) in Tris–HCl (pH 8.5) and Tris-EDTA (pH 9) buffer for 20 min at 98 °C for antigen retrieval. Staining of sections was performed in an Autostainer 480 (Lab Vision Corp., Fremont, CA, USA) by the Dako REAL EnVision Detection system, Peroxidase/DAB+, Rabbit/Mouse (Dako, Glostrup, Denmark) for β-catenin, and by ImmPRESS HRP Polymer Detection Kit, Peroxidase, Anti-Goat IgG (Vector Laboratories, Burlingame, CA, USA) for PROX1. Tissues were incubated with beta-Catenin Antibody (Invitrogen, Thermo Fisher Scientific, Inc., Waltham, MA, USA; diluted to 1:500 = 5 μg/ml) for one hour at room temperature, and with Anti-human Prox1 Antibody (R&D Systems, Inc., Minneapolis, MN, USA; diluted to 1:1500 = 15 μg/ml) for overnight at room temperature. Samples of colon tissue and normal lymph node served as positive controls in each staining series (see Additional files 1 and 2). We also chose 13 whole tumor tissue blocks and corresponding lymph node metastases from the patient cohort to compare PROX1 expression in the tumor and its lymph node metastases.

Cytoplasmic stainings of PROX1 and β-catenin were scored as negative (0), weakly positive (1), moderately positive (2), or strongly positive (3) according to staining intensity. Also, β-catenin membranous staining was evaluated. In the samples, where no membranous staining was seen, there was no cytoplasmic staining either. The highest score of each sample was considered representative for analysis. Scoring was performed by two independent investigators (K.S. and J.H.) without knowledge of clinical data and outcome. In case of differing scores, consensus score was discussed and determined.

Categories of β-catenin and PROX1 were dichotomized for statistical purposes into low (scores 0–1) and high (scores 2–3). A three-class categorization was created to study these two tumor markers together: low (PROX1, and β-catenin low), moderate (either PROX1, or β-catenin high), and high (PROX1, and β-catenin high).

Associations between tumor marker expression and clinicopathological parameters were assessed by the Fischer’s exact-test or the linear-by-linear association test. The Kaplan-Meier method and log-rank test were used for survival analysis. The Bonferroni correction was used for multiple comparisons by dividing the probability level by the number of comparisons. The Spearman correlation coefficient with bootstrapped (1000 resamples, bias corrected) confidence intervals was calculated to find out correlations between PROX1 and β-catenin expression. Uni- and multivariate survival analyses were carried out by the Cox regression proportional hazard model adjusted for age, gender, stage, metastasized lymph node ratio (LNR) ≥/<20 % (cut-off ≥/<20 %), perivascular invasion, and postoperative adjuvant therapy. Since stage and LNR are internally correlated to each other, a combination variable was formed for multivariate analyses (see Table 6). Interaction terms were considered. The Cox model assumption of constant hazard ratios over time was tested. For each testable variable at a time, a time-dependent covariate was included separately. All variables fulfilled the assumption. Stage and lymph node ratio were combined into a single variable to simplify the model. A p-value <0.05 was considered significant and all tests were two-sided. Statistical analyses were computed with SPSS version 22.0 (IBM SPSS Statistics, version 22.0 for Windows/MAC; SPSS, Inc., Chicago, IL, USA, an IBM Company).

PROX1 expression was cytoplasmic and evenly distributed with no distinctive membranous staining. Cytoplasmic staining was scored as described above. In normal pancreatic tissue apparent nuclear staining is present although all the nuclei are not stained. In two cancer tissue samples we saw staining of the nuclei, and the cytoplasmic staining scores in these samples were 1, and 3. In all the other cancer specimens nuclei were negative. In the whole tumor specimens, there was no nuclear staining in the metastases; only negative or weak cytoplasmic staining was present (Fig. 1).

β-catenin expression was distributed in the cell membrane and within the cytoplasm. Only in a few exceptions the staining was not uniform throughout the cell. With more intense membranous staining, also cytoplasmic staining was stronger. The cytoplasmic expression pattern showed two different types of staining; homogenous and granular. There was no distinct nuclear staining. Only three samples lacked membranous staining (Fig. 2). The membranous and cytoplasmic staining were very difficult to score separately. Because of this, cytoplasmic expression was used in statistical analyses.

PROX1 staining could be evaluated in 154 (99 %) specimens: 20 (13 %) showing negative, 60 (39 %) weak, 66 moderate, (43 %) and 8 (5 %) strong staining (Fig. 3). β-catenin cytoplasmic staining could be evaluated in 153 (98 %) specimens: 1 (1 %) showing negative, 52 (34 %) weak, 63 (41 %) moderate, and 37 (24 %) strong staining (Fig. 4). Combined PROX1 and β-catenin expression was evaluated in 152 (97 %) tumors: 38 (25 %) low, 56 (37 %) moderate, and 58 (38 %) high expression pattern.

There was a statistically significant association between PROX1 expression and age; patients in the low PROX1 expression group were younger than in the high PROX1 expression group (p = 0.038). PROX1 expression did not correlate with gender, stage, LNR, histological grade, perineural, or perivascular invasion (Table 1).

Patients with low β-catenin expression showed a significant association with higher tumor histological grade compared to patients with high expression (p = 0.025). No significant association was found between β-catenin and age, gender, stage, LNR, perineural, or perivascular invasion (Table 2).

There was no correlation between combined PROX1 and β-catenin expression and age, gender, stage, histological grade, LNR, perineural, or perivascular invasion (Table 3). PROX1 and β-catenin expression correlated with each other (Spearman correlation coefficient = 0,371; 95 % CI 0.24–0.50; p < 0.001).

Five-year cancer-specific survival (CSS) was not significantly different for PDAC patients with low PROX1 expression compared to those with high expression (log-rank, p = 0.174, Fig. 5). Five-year CSS was 15.5 % (95 % CI 6.7–24.3 %) for patients with low PROX1 expression, and 20.0 % (95 % CI 9.2–30.8 %) when PROX1 expression was high (Table 4). PDAC patients with low β-catenin expression showed significantly poorer CSS than those patients with high expression (log-rank, p = 0.007, Fig. 6). Five-year CSS for PDAC patients with low β-catenin expression was 11.3 % (95 % CI 2.1–20.5 %), and 22.4 % (95 % CI 13.0–31.8 %) for those with high expression (Table 4).

Combined expression of PROX1 and β-catenin showed significantly poorer CSS for PDAC patients with low compared to high expression (p = 0.013). Between patients with moderate and low expression (p = 0.092), or with moderate and high expression (p = 0.435) no significant difference in CSS was seen (Fig. 7). Five-year CSS for patients with low combined expression was 10.3 % (95 % CI −0.7–21.3 %), with moderate combined expression 18.7 % (95 % CI 9.9–29.5 %), and with high combined expression 21.3 % (95 % CI 8.1–34.5 %) (Table 4).

In univariate analyses high β-catenin expression associated significantly with lower risk of death from PDAC (HR = 0.61, 95 % CI 0.42–0.88; p = 0.008). High PROX1 expression seemed to reduce the risk of death from PDAC, but this result just failed to be statistically significant in univariate analysis (HR = 0.71, 95 % CI 0.49–1.01; p = 0.053). With PROX1 and β-catenin, combined high expression showed lower risk of death from PDAC (HR = 0.52, 95 % CI 0.33–0.83; p = 0.006). With moderate combined expression the risk of death from PDAC was not statistically significant (HR = 0.69, 95 % CI 0.44–1.08; p = 0.103). Other prognostic variables in univariate analyses were stage, lymph node positivity, perivascular invasion, and postoperative adjuvant therapy (Table 5).

In multivariate analyses adjusted for age, gender, stage, LNR, perivascular invasion, and adjuvant therapy high β-catenin expression remained statistically significant for better prognosis (HR = 0.54, 95 % CI 0.35–0.82; p = 0.004), and high PROX1 expression was also statistically significant (HR 0.63, 95 % CI 0.42–0.95; p = 0.026). The combined high expression of β-catenin and PROX1 remained statistically significant (HR 0.46, 95 % CI 0.28–0.76; p = 0.002) (Table 6).