Immunohistochemical analysis of changes in signaling pathway activation downstream of growth factor receptors in pancreatic duct cell carcinogenesis

This study has been approved by the Research Ethics Board of the University Health Network (Toronto, ON, Canada) in compliance with applicable national Tri-Council ethics principles. The formalin-fixed and paraffin embedded (FFPE) samples used in this study were tumor and adjacent normal pancreatic tissue obtained within 30 minutes after Whipple's resection for pancreatic ductal adenocarcinoma (PDAC) or non-PDAC conditions. The FFPE blocks were made from thin cross sections of tissues collected for snap-frozen tumor banking and were initially intended for a histological quality check of the banked tissue. These samples were particularly suitable for this study due to their limited delay between sampling and fixation.

Histology was reviewed to assure the correct diagnosis of the 26 PDAC specimens and 12 specimens that were considered "normal" pancreas from patients with non-pancreatic conditions (e.g. Ampulla of Vater, bile duct and stomach cancers) [see Additional file 1]. The clinicopathological parameters of individual cases are listed in an additional file [see Additional file 2]. H&E slides were used to guide the selection of representative 1.5 mm diameter cores from paraffin blocks containing viable tumor areas and morphologically normal pancreatic parenchyma. The cores were constructed into a tissue microarray (TMA) using a manual tissue arrayer (Beecher Instruments, Sun Prairie, WI). The final TMA contained 67 of 71 (94.3%) cores, (four cores were lost during processing) and each PDAC case was represented by at least one core [see Additional file 3]. Results are reported on 26 tumor cores (T), 13 cores containing adjacent normal parenchyma and non-neoplastic ducts (Dt), and 12 cores containing parenchyma with non-neoplastic tissue from non-PDAC conditions (D).

Antibodies and staining methods are summarized in an additional file [see Additional file 2]. Microwave antigen retrieval was performed for all immunostains except CK7, EGFR and p-S6 which were treated with pepsin digestion. Secondary antibodies (anti-mouse and anti-rabbit) were used as provided by the IDetect Ultra HRP system (ID Labs, London, ON). Goat polyclonal antibody (biotin-conjugated anti-goat IgG, 1:300 dilution), NovaRed peroxidase substrate and hematoxylin counterstain were purchased from Vector Laboratories (Burlingame, CA). A negative control slide was processed with a mix of pooled secondary antibodies, omitting primary antibody incubation [see Additional file 1].

The immunostained slides were scanned using an Aperio CS Scanner (Vista, CA) at 20× magnification. Either slides or digital images were used for scoring by three independent evaluators (NAP, JS, VI) without prior knowledge of the core source or stained target antigen. Final scores were based on a consensus of the three evaluators. A single intensity score was obtained since the intensity of staining within each core was mostly homogeneous. Intensity was scored as 0 for absence of staining, 1 for weak, 2 for moderate, and 3 for strong staining. Scores were recorded for the different cell types as well as their nuclear and cytoplasmic compartments. EGFR and MET receptor were scored for their plasma membrane staining only.

Unsupervised hierarchical clustering using the agglomeration rule average linkage placed cases and antibodies next to each other if they were most similar in their IHC profiles (free software Genesis [23]). The relationship of two categorical variables was calculated using Fisher's exact test. Data of categorical variables were arbitrarily split into low and high groups to maximize statistical power obtained by equalizing the number of samples in each groups. The strength of the relationship between two variables was calculated using Spearman's rank correlation coefficient, Rho (ρ). No adjustment for multiple testing was done. Statistical analyses were performed in SAS Version 9.1 (SAS Institute, Cary, NC).

A heat map in Figure 1 shows relative expression levels of eight signaling proteins. The plasma membrane expression of EGFR (p = 0.02) and MET receptor (p < 0.0001) and the cytoplasmic proteins ADAM9 (p < 0.001), SRC (p < 0.0001), PKBβ (p = 0.0005) and βCAT (p = 0.005) were significantly higher in PDAC cells than in histologically normal ducts. In contrast, levels of the two known tumor suppressor genes cytoplasmic PTEN (p = 0.01) and nuclear SMAD4 (p = 0.0005) were significantly lower in PDAC cells than in normal ducts. The p-values of Fisher's exact tests were calculated using proportions of PDAC cells and ducts that expressed high levels of proteins [see Additional file 3]. Unsupervised hierarchical clustering segregated tumor specimens (group a) and normal ducts adjacent to PDAC and non-PDAC pancreas (group b). As expected, cytokeratin 7 was expressed in all PDAC cells and normal ducts (results not shown).

An unsupervised clustering of cytoplasmic activated protein levels shows that tumor specimens (Figure 2A, group a) are more similar to each other than to histologically normal ducts (group b). Only five of twenty five specimens of normal ducts were misclassified as tumor, while none of the tumors were misclassified as normal ducts. The cytoplasmic levels of activated p-ERK (p < 0.001), p-SRC (p = 0.0005), ^T308p-PKB (p < 0.0001), p-p38 (p = 0.02), and their putative downstream substrates p-JNK (p < 0.0001), ^S727p-STAT3 (p = 0.005), ^Y705p-STAT3 (p = 0.0005), p-GSK3β (p = 0.006), p-RAF (p = 0.01) and p-mTOR (p = 0.05) were significantly higher in PDAC cells than in ducts. Degradation-targeted p-βCAT levels significantly decreased (p < 0.0001) in PDAC cells compared with ducts, consistent with the observation of an accumulation of βCAT in PDAC compared with ducts (Figure 1). The four cytoplasmic proteins, ^S473p-PKB, p-S6K, p-S6 and p-NFκB were similarly expressed in PDAC cells and ducts (Table 1). Although the cytoplasm is the known cellular compartment of activity for the mTOR/S6K/S6 pathway (Figure 2B, i and 2B, ii), a subset of activated signaling proteins are also known to translocate into the nucleus. For example, p-ERK was detected at high levels in the cytoplasm as well as the nuclei of cancer cells (Figure 2B, iii). Low levels of nuclear p-ERK were detected in normal pancreatic ductal cells and acinar (Figure 2B, iv).

Since many signaling proteins translocate into the nucleus to serve as transcription factors after being activated by phosphorylation, we explored their profiles in PDAC cells compared with histologically normal ductal epithelia. An unsupervised clustering of the nuclear activated protein levels showed an absence of clustering between tumor and normal ducts (Figure 3). Levels of nuclear activated proteins, including p-NFκB, p-STAT3, p-βCAT, p-GSK3β and ^T308p-PKB, were similarly expressed in PDAC cells and normal ducts. However, the levels of four proteins, p-ERK (p = 0.001), p-p38 (p = 0.004), p-JNK (p = 0.01), and ^S473p-PKB (p = 0.001) were significantly higher in nuclei of PDAC cells than in normal ducts (Table 1). Cancer cells also displayed significant moderate to strong relationships between cytoplasmic and nuclear levels for several proteins: p-ERK (ρ = 0.72, p < 0.001), p-p38 (ρ = 0.56, p = 0.004), ^T308p-PKB (ρ = 0.52, p = 0.007), SMAD4 (ρ = 0.63, p < 0.001) and PTEN (ρ = 0.87, p < 0.001). These associations were absent in ductal epithelia from non-PDAC specimens except for an association of cytoplasmic and nuclear levels for p-p38 (ρ = 0.61, p < 0.001) in the ductal epithelia adjacent to PDAC [see Additional file 4].

Table 2 shows that proportions of PDAC specimens characterized by the ERK pathway activation were associated with all specimens of high p-mTOR levels (100%, p = 0.05), and its downstream substrates p-S6K (100%, p = 0.02), and a proportion of specimens with high p-S6 (86%, p = 0.005). The activation of p-S6K also correlated with high levels of ^S727p-STAT3 (86%, p = 0.005) and shows a positive trend with high levels of ^T308p-PKB (100%, 0.06). The activation of p-STAT3 correlated with high levels of PKBβ (59%, p = 0.004) and the presence of SMAD4 (70%, 0.02). An association was observed between high levels of ^T308p-PKB and p-JNK (80%, p = 0.05). High levels of inactive p-RAF was correlated with PKBβ (82%, p = 0.009), PTEN (91%, p = 0.01) and p-βCAT (91%, p = 0.01). Protein levels were compared with clinicopathological characteristics of tumors. There was no significant association between tumor grade and individual protein levels (p-value > 0.06, Fisher's exact test, results not shown). However, advanced stage (III and IV) showed a trend in association with higher levels of βCAT (p = 0.04), p-NFκB (p = 0.05) and p-GSK3β (p = 0.05). In contrast, high levels of p-STAT3 (p = 0.01) and loss of PTEN (p = 0.01) appeared not to be associated with cancer progression. Table 3 lists the distribution of PDAC stages that showed trends in associations with proteins levels.

A subset of the signaling proteins examined in this study differentially characterized histologically normal ducts adjacent to PDAC compared with ducts in non-PDAC pancreas (Figure 4A). Mean levels of five proteins, cytoplasmic PKBβ (p = 0.03) and p-S6 (0.03), and nuclear p-GSK3β (p = 0.008), p-ERK (p = 0.04) and p-p38 (p = 0.03), were significantly higher in ducts adjacent to PDAC than in non-PDAC pancreas [see Additional file 5]. EGFR levels showed a tendency (p = 0.055) to be higher in ducts adjacent to PDAC compared with non-PDAC pancreas.

Centroacinar cells, which are the terminal ducts lining the centre of the acini, as expected showed a protein profile that was similar compared with larger ducts (Figure 4B). The seven proteins, CK7, ADAM9, EGFR, p-βCAT, p-GSK3β, p-mTOR and p-p38 were present in both centroacinar and ductal components (Figure 4C). The high expression of p-βCAT, EGFR and ADAM9 characterized islet cells, and relatively lower levels of all seven proteins characterized acinar cells.