Introduction

Despite efforts to develop effective treatment strategies, pancreatic adenocarcinoma remains one of the most lethal cancers with a 5-year survival rate of only 5% [1, 2]. Known risk factors for the development of pancreatic cancer include smoking and age and various genetic disorders [1]. Currently available treatment strategies consist of surgery, chemotherapy, and radiation [3].

Toll-like receptors (TLR) are pattern-recognition receptors of the innate immune system [4]. In a normally functioning organism, they recognize unique proteins derived from microbes, called pathogen-associated molecular patterns (PAMPs) [5]. The recognition of PAMPs activates multiple pro-inflammatory responses through downstream signaling, ultimately leading to antigen-specific acquired immunity [5]. TLR2 and TLR4 recognize lipopolysaccharides derived from exogenous microbes and are located mostly on the cell membrane [6], whereas TLR9 recognizes viral and bacterial unmethylated CpG-DNA and is typically located intracellularly [7].

Involvement of TLRs in carcinogenesis is becoming undisputed. TLRs recognize damage-associated molecular patterns (DAMPs), released into the system by tissue damage [8]. Chronic activation of TLRs creates a favorable environment for cancer development through pro-inflammatory responses [9]. This leads to enhancement of anti-apoptotic and proliferative properties of the microenvironment and the emergence of cancer cells [10].

TLR involvement has previously been reported in gastrointestinal malignancies such as colon and esophageal cancers [11, 12], as well as prostate and tongue cancers [13, 14]. There is also evidence of TLR involvement in pancreatic cancer [15]. Single studies have shown that TLR2 is expressed in pancreatic cancer tissue and has been suggested as potential target for immunotherapy [16]. There is also some evidence of TLR4 expression in pancreatic cancer [17–19]. TLR9 has previously been linked to acute pancreatitis, its expression has been reported in pancreatic cancer, and it has been suggested to participate in tumorigenesis [15, 20, 21]. In a recent study, our group showed that germ-free mice had significantly higher levels of TLR9 in the exocrine pancreas compared to conventional mice [22]. There is recent evidence that bacteria may affect pancreatic carcinogenesis, and oral pathogens as well as Helicobacter Pylori have been suggested to participate in pancreatic cancer progression [23, 24]. In this study, we aimed to investigate associations between expression of bacteria-sensing TLR2, TLR4, and TLR9 in pancreatic cancer cells and patient survival and correlation with local inflammatory cell response and clinical parameters.

Materials and methods

Patients

The material consisted of paraffin-embedded archival specimens (n = 69) of patients with surgically resected pancreatic cancer. In 65 cases, representative carcinoma tissue was available in the tissue blocks. In the four remaining patients, either only normal pancreas or premalignant ductal alterations were found (Table 1). The specimens were collected from Oulu University Hospital Pathology Archive between the years 1993 and 2011. Patient clinical and laboratory data were extracted from patient records. The time and cause of death were obtained from Statistics Finland. All had been diagnosed as pancreatic adenocarcinoma, confirmed by an experienced gastrointestinal pathologist (TJK). The patients underwent pancreaticoduodenectomy according to Whipple (n = 54), subtotal pancreatic resection (n = 2), total pancreatic resection (n = 11), or total pancreaticoduodenectomy (n = 2). During the operation, in six of the 69 patients, macroscopic intravenous tumor invasion was found. Median age of the patients at diagnosis was 66 years (range 36–77 years). Median follow-up time of the patients was 21 months (range 1–60 months). Use of the samples and patient data were approved by the Oulu University Hospital Ethics Committee and by the National Authority for Medicolegal Affairs (VALVIRA).

Table 1 Baseline characteristics of patients with resected pancreatic ductal adenocarcinoma. Lymph node status was available for 68 patients
Full size table

Immunohistochemistry

Immunohistochemistry was performed on tissue blocks, which were selected on the basis of hematoxylin and eosin staining as representative for the tumor mass in the resected specimen. Antigen retrieval was performed by exposure to high temperature in Tris-EDTA buffer for 15 min (pH 9.0). Immunostaining was performed in a Dako Autostainer (Dako) with mouse monoclonal antibodies against TLR2 (MAB0066, IgG1, clone 1030A5.138, Abnova, Taipei, Taiwan) at a dilution of 1:50, TLR4 (H00007099-M02, IgG2a, clone 3B6, Abnova, Taipei, Taiwan) at a dilution of 1:1000, and TLR9 (IMG-305A, IgG1, clone 26C593.2, Imgenex, San Diego, CA, USA) at a dilution of 1:150. For detection of the first antibody binding, we used Dako Envision flex kit (Dako, Copenhagen, Denmark). Diaminiobenzidine (Dako basic DAB-kit) was used as a chromogen. As negative control, we used omission of the primary antibody and replacement of the primary antibody with non-specific mouse primary antibody isotype.

Histological analysis

Immunohistochemical staining of TLR2, TLR4, and TLR9 was evaluated by two observers independently (J. L, O. H), blinded to patient clinical data as described earlier [25, 26]. Staining intensity and the percentage of positive epithelial cells were assessed separately in pancreatic ductal adenocarcinoma (n = 65) and adjacent normal pancreatic tissue, when available (for TLR2 in 37 cases, for TLR4 in 38 cases, and for TLR9 in 38 cases. Tissue specimen containing cancer tissue was unavailable for four patients. Membranous, cytoplasmic, and nuclear staining was evaluated independently, and staining intensity was given a score from 0 (absent) to 3 (strong intensity). In addition, the percentage of cells positive for each subcellular localization was estimated (from 0 to 100%). When inter-observer difference was less than one point in intensity or less than 30% in the proportion of positive cells, the mean value of the independent estimates was used for statistical analysis. Cases with higher differences were re-evaluated, and a single score was given by consensus between the two observers. Only one case stained for TLR9 needed re-evaluation. Mean intensity and mean percentage were then multiplied to obtain a histoscore (0–300). For statistical evaluation, histoscore for each stain (TLR2, TLR4, and TLR9) was then dichotomized by median value into two equal sized groups (low and high).

The inflammatory cell response in the invasive front of the cancer and in the bulk of the cancer was evaluated as described earlier [27] by two independent observers (J. L, O. H). The response was given a score ranging from absent (score 0) to strong increase (score 3). The mean value of two independent estimates was then used to divide the cohort into three equal groups (low, moderate, and high).

Statistical analysis

Statistical analysis was done using IBM SPSS statistics for Windows, version 22.0. Armonk, NY: IBM Corp. Wilcoxon paired test was used to assess differences in TLR expression between cancer tissue and adjacent normal pancreatic cancer. Pearson’s chi-square test was used to calculate statistically significant correlations between clinicopathological variables (TNM-staging, tumor grade, tumor size, BMI, gender, age at diagnosis) and TLR histoscore. Kaplan-Meier method was used to calculate life tables, and survival curves were compared with the log-rank test. Cox proportional hazards model was calculated for the following covariates: age at diagnosis (<65 or ≥65), gender (male or female), tumor stage (I, II, and III–IV), and cytoplasmic TLR9 histoscore (low or high). The hazard ratios (HRs) and 95% confidence intervals (CIs) are provided for each covariate in the model after the adjustment. p values less than 0.05 were accepted as statistically significant.

Results

Expression and localization of TLR2, TLR4, and TLR9 in pancreatic cancer and adjacent normal pancreatic tissue

TLR2 staining was found in adjacent normal pancreatic tissue on cell membranes in only 1/37 cases and in the cytoplasm in 34/37 cases (mean histoscore 75, SD 47, range 0–200) (Fig. 1b). In 24/65 cancers, TLR2 expression was found on the cell membrane (mean histoscore 27, median histoscore 0, SD 48, range 0–180) and in all cancers in the cytoplasm (mean histoscore 223, median histoscore 250, SD 70, range 85–300). (Fig. 1a, Fig. 2a, b). Both membranous and cytoplasmic TLR2 staining was more intense in cancer tissue compared to adjacent normal pancreatic tissue (p = 0.001, p < 0.001, respectively).

Fig. 1
figure 1

High magnification images (×20) illustrating typical subcellular localization patterns of the staining in pancreatic cancer tissue (a, c, and e) and examples of typical positive staining of adjacent normal pancreatic tissue (b, d, and f). Moderate intensity cytoplasmic TLR2 expression, without nuclear or membranous staining (a). High membranous and cytoplasmic TLR4 expression without nuclear staining (c). Moderate cytoplasmic TLR9 staining without membranous or nuclear staining (e). Examples of typical staining of adjacent normal pancreatic tissue for TLR2 (b), TLR4 (d), and TLR9 (f)

Full size image
Fig. 2
figure 2

Examples of variation of TLR staining intensity in pancreatic cancer visualized with low magnification (×10). TLR2 expression in pancreatic cancer in samples with low (a) and high (b) intensity cytoplasmic expression. TLR4 expression in pancreatic cancer samples with low cytoplasmic (c) and high cytoplasmic as well as membranous (d) staining pattern. TLR9 expression in pancreatic cancer tissue with low (e) and high (f) cytoplasmic expression

Full size image

In all the studied samples, TLR4 staining was present in the cytoplasm but not on the membrane of adjacent normal cells (mean histoscore 141, SD 53, range 100–250), (Fig. 1d). In cancer tissue, membranous TLR4 staining was found in 25/65 cases (mean histoscore 53, median histoscore 0, SD 76, range 0–270) and cytoplasmic staining in all cases (mean histoscore 211, median histoscore 200, SD 74, range 90–300). (Fig. 1c, Fig. 2c, d). Cytoplasmic TLR4 staining was significantly higher in cancer tissue compared to adjacent normal pancreatic tissue (p < 0.001).

In all studied samples, in adjacent normal pancreatic tissue, TLR9 staining was present in the cytoplasm (mean histoscore 150, SD 94, range 0–300), but not on the membrane (Fig. 1f). In cancer tissue, TLR9 staining was found on the membrane in 20/65 cases (mean histoscore 31, median histoscore 0, SD 60, range 0–270) and in the cytoplasm of 61/65 cases (mean histoscore 130, median histoscore 113, SD 84, range 0–300). (Fig. 1e, Fig. 2e, f). There was no statistically significant difference in cytoplasmic TLR9 expression between cancer and adjacent normal pancreatic tissue.

Nuclear TLR2 staining in cancer cells was observed in 42/65 cases. Similarly, nuclear TLR4 was seen in 5/65 cases and nuclear TLR9 in 19/65 cases.

TLR expression and correlation with clinicopathological variables

Pearson’s chi-square test was used to investigate associations between TLR expression and clinicopathological variables (TNM-stage, tumor grade, tumor size, BMI, gender, age at diagnosis). High cytoplasmic TLR4 expression is associated with high tumor grade (p = 0.010) and low cell membrane TLR9 expression with low BMI (p = 0.031). There were no other significant correlations between TLR expression and clinicopathological variables (Table 2).

Table 2 TLR2, 4, and 9 histoscores compared to clinicopathological variables, preoperative blood leukocyte count, and local inflammatory response of the cancer. Statistically significant p values are italicized. Lymph node status was unavailable for one patient. Tumor grade was missing from 16 patients
Full size table

Correlation between TLR expression and local inflammatory response

Pearson’s chi-square test was applied to assess associations between TLR expression and the inflammatory response. Low cell membrane TLR2 expression was associated with low local inflammatory response in the tumor bulk (p = 0.019; Table 2). We found no other associations between local inflammatory response and TLR expression.

Local inflammatory response and survival

To test associations between local inflammatory response and 5-year patient survival, the cohort was subdivided according to the density of inflammatory cells in three equal groups (low, moderate, and high). Inflammatory response was assessed separately in the bulk of the cancer and in the invasive front. Only a high density of inflammatory cells in the invasive front showed a trend towards association with longer survival (p = 0.087, log rank test).

Correlation between TLR expression and systemic inflammatory response

We assessed correlations between preoperative peripheral blood leukocyte counts and TLR expression. The only significant correlation found was between higher than median (>7.3 × 109/l) blood leukocyte count and low cytoplasmic TLR9 expression (p = 0.031). We found no correlation between blood leukocyte count and local inflammatory response to the tumor.

TLR expression and patient survival

We assessed correlations between expression of TLR2, 4, and 9 and 5-year survival. Survival time for patients with high cytoplasmic TLR9 expression (mean = 36 months; 95% CI 29–42) was significantly longer compared to patients with low cytoplasmic TLR9 expression (mean = 21 months; 95% CI 15–26); p = 0.001 (log-rank; Fig. 3).

Fig. 3
figure 3

Kaplan-Meier curve visualizing the association between cytoplasmic TLR9 expression and patient survival. Patients with high cytoplasmic TLR9 expression had significantly longer mean survival time (36 months) compared to patients with low cytoplasmic TLR9 (21 months); (p = 0.001)

Full size image

Multivariate analysis identified low cytoplasmic TLR9 expression as independent prognostic factor (HR 3.090, 95% CI 1.673–5.706; p = 0.000) for pancreatic cancer-specific death. Age, gender, or tumor stage were not significantly associated with survival as shown in Table 3. Expression of TLR2 or TLR4 was not associated with survival, neither in uni- nor in multivariate analysis (data not shown).

Table 3 Multivariate analysis of prognostic factors in pancreatic cancer
Full size table

Discussion

We report here an association between TLR9 expression and patient survival in pancreatic cancer. High cytoplasmic TLR9 expression was associated with a mean survival advantage of 15 months (p = 0.001). Multivariate analysis identified low cytoplasmic TLR9 expression as an independent factor predicting cancer-specific death. Abundant inflammatory response was not related with TLR expression. In agreement with previous studies, expression of TLR2 and TLR4 was higher in pancreatic carcinoma cells as compared to adjacent normal pancreatic tissue [18, 28]. TLR2 expression was not associated with patient survival. TLR4 expression has been previously associated with poor survival, which was not observed in our material [18]. This difference might be due to different geographical locations and populations.

It is noteworthy that we studied histologically normal pancreatic tissue adjacent to cancer tissue, which might not be functionally identical to more distant pancreatic tissue or tissue from a normal pancreas. However, we minimized this bias by carefully assessing normal histology. Additionally, a previous study suggested that adjacent carcinoma has only limited or no impact on TLR expression in normal epithelium, suggesting that there is no field effect in pancreatic cancer in terms of TLR expression [22].

Associations between TLR9 expression and patient survival appear to be related to tumor type. Previously, high TLR9 expression has been associated with adverse prognosis in esophageal adenocarcinoma, squamous cell carcinoma of the tongue, and prostate cancer [29–31]. In contrast and in agreement with our observations, in triple-negative breast cancer, renal cell carcinoma and mucoepidermoid salivary gland carcinoma high TLR9 expression are associated with better prognosis [32–34]. The authors hypothesized that TLR9 expression in triple-negative breast cancer and renal cell carcinoma might be explained by hypoxia, but this purported biological link needs further evidence [34]. Of note, high expression of TLR is linked to better survival in organs without physiological exposure to bacteria. In contrast, in organs where bacteria are present in abundance such as gastric and colorectal cancers and squamous cell carcinoma of the mobile tongue, TLR9 ligation induces cell invasion and increased viability and migration of esophageal cancer cells [30, 35, 36]. In pancreatic cancer, however, TLR9 ligation has been shown to have mostly inhibitory effects on cancer cells [21]. Studies in mice suggest that TLR9 ligation with synthetic CPG ODNs, in combination with vaccines based on immune stimulatory complexes (ISCOM) or gemcitabine, may inhibit pancreatic cancer cell growth and invasive properties, enhance immunological response to cancer, and lead to tumor regression and prolonged survival [21, 37–39]. On the other hand, TLR9 ligation induced epithelial cell proliferation in pancreatic stellate cells, suggesting that TLR9 may also have tumor-promoting effects on pancreatic cancer [20]. Taken together, the role of TLR9 in pancreatic cancer remains controversial.

Pancreatic cancer risk is elevated in carriers of oral pathogens such as Fusobacterium species and Porphyromonas gingivalis [23, 40, 41]. Bacteria may affect pancreatic cancer progression, reaching the pancreas through the bloodstream or biliary tract [23, 24, 42]. Local presence of oral Fusobacterium species might associate with poor outcome in pancreatic cancer, as well as in colorectal and esophageal cancers [43–45]. This has been explained by reactive oxygen species (ROS) and inflammatory cytokines of TLR signaling pathways such as IL-6 and TNF [45]. There is evidence that bacteria might have a downregulating effect on TLR levels [46]. TLR9 expression in the normal pancreas might be also downregulated by exposure to bacteria, as suggested by a study comparing pancreatic TLR expression in germ-free and conventional mice [22]. We also found patients with low TLR9 expression to have higher blood leukocyte count, suggesting the possibility of activation of a systemic inflammatory response. There might be differences in the regulation of innate immunity receptors in the presence of bacteria. High TLR9 expression leading to improved survival in pancreatic cancer might be related to lack of bacterial exposure in this group of patients. This explanation, however, remains hypothetical [20, 47]. Blood leukocyte count is a highly unspecific clinical marker elevated in various inflammatory states and, therefore, does not unequivocally equate to bacterial exposure.

In conclusion, we found high TLR9 expression to be associated with improved survival of patients with pancreatic ductal adenocarcinoma. Expression of TLR2 and TLR4 is not associated with pancreatic cancer survival. The prognostic effect of TLR9 might be associated with bacterial exposure, but this needs further evidence.