|Home | About | Journals | Submit | Contact Us | Français|
Studies have proposed pro-oncogenic effects of glucagon-like peptide-1 receptor (GLP-1R) agonists in the pancreas by promoting GLP1-R over-activation in pancreatic cells. However, the expression of GLP-1R in normal and neoplastic pancreatic cells remains poorly defined, and reliable methods for detecting GLP1-R in tissue specimens are needed.
We employed RNA in-situ hybridization to quantify glp-1r RNA in surgically resected human pancreatic specimens, including pancreatic ductal adenocarcinoma (PDAC), pre-invasive intraepithelial lesions (PanINs), and non-neoplastic ductal, acinar, and endocrine cells. A mixed-effect linear regression model was used to investigate the relationship between glp-1r signals and all cells, ordered by increasing grade of dysplasia.
All cell types had evidence of glp-1r transcripts, with highest expression in endocrine cells and lowest in ductal cells. The slope of the fitted line was not significantly different from zero (0.07, 95% CI -0.0094, 0.244; p=0.39), suggesting that progression from normal cells to PDAC is not associated with a parallel increase in glp-1r RNA. A series of pairwise comparisons between all cell types with respect to their glp-1r expression showed no significant difference in glp-1r in cancer, PanIN, acinar and ductal cells.
Our study supports the lack of evidence for GLP-1R overexpression in PDAC.
Glucagon-like peptide-1 receptor (GLP-1R) agonists are incretin analogues that potentiate glucose-dependent insulin release following food intake1. These drugs are being increasingly used for patients with type-2 diabetes, as they improve blood glucose control and weight loss, with low risk of hypoglycemia. Despite the favorable effects, a number of publications have questioned the safety profile of incretin analogs, raising concern that these drugs may induce pancreatic inflammation and carcinogenesis2. A retrospective analysis of adverse event reports showed increased incidence of pancreatitis, thyroid cancer and pancreatic ductal adenocarcinoma (PDAC) in patients treated with incretin-based therapies2. A similar study from Germany reported a significant association between the use of GLP-1R agonists and a diagnosis of PDAC 3.
Incretin analogs have also been proposed to stimulate the growth of both the exocrine and endocrine pancreas, promoting subclinical pancreatic neoplasms. A recent autopsy study found an association between incretin use and increased volume of the exocrine and endocrine pancreas 4, 5, suggesting that these drugs may have proliferative effects. However, several limitations of the design of this have been highlighted including the small number of pancreas samples analyzed, and differences between the study group and control group with respect to key clinical characteristics, such as age, gender, and duration of diabetes6-9.
Thus far, preclinical studies have largely failed to demonstrate pathologic alterations in the pancreas induced by incretins10-12. Nyborg et al 13 recently reported studies on mice, rats and primates treated long-term with a GLP-1R agonist, and found no pathologic alterations in the exocrine pancreas. Similar results were reported by Gotfredsen et al. 14 after extensive toxicological studies on the pancreas of primates treated with two different GLP-1R analogs. Another study reported the spontaneous formation of pancreatic ductal cell proliferation, atrophy and inflammation in normal and diabetic rats15, questioning the contribution of incretins to the genesis of pancreatic lesions. The lack of unequivocal recognition of which cell types express GLP-1R currently limits our understanding of pharmacologic actions of incretin analogs. Although several rodent and human studies have documented GLP-1R expression in pancreatic ductal cells, acinar and cancer cells and in pancreatic precursor neoplasms, known as pancreatic intraepithelial neoplasia (PanIN) 16-19, data from multiple studies suggest that the methods used to assess GLP-1R tissue expression lack sufficient sensitivity and specificity based on evidence that the antibodies used in these studies either do not bind to GLP-1R by western blot, or also bind to other proteins isolated from cells that do not express glp-1r RNA 6, 20. For example, glp-1r protein expression in animal models has been reported in the heart, in both atria and ventricles21. However, subsequent investigations revealed that lack of glp-1r expression in the ventricles, questioning the reliability of the antibodies previously used22, 23. Furthermore, Panjwani et al 7 recently studied the action of GLP-1R agonists in atherosclerosis and liver steatosis, in a genetically modified murine model. Contrary to previous data, the authors could neither identify glp-1r mRNA transcripts in isolated macrophages or hepatocytes at RT-PCR, nor could they validate the sensitivity and specificity of three commercially available anti-glp-1r antibodies using Western Blot analysis and immunoprecipitation. Overall, there is insufficient evidence to conclude that there is an association between incretin-based therapies and PDAC, but the existing literature has left open several important questions about the physiology of incretins, particularly with regard to GLP-1R localization in normal and neoplastic pancreas.
In this study we employed an RNA in-situ hybridization (ISH) assay to characterize the presence of glp-1r in pancreatic cell types using surgically resected human pancreas specimens that contained PDAC cells, premalignant intra-epithelial lesions (PanINs), as well as acinar and islet cells.
This study was approved by the Johns Hopkins Medical Institutional Review Board. We utilized two tissue microarrays (TMAs) of normal and neoplastic archival formalin-fixed paraffin-embedded (FFPE) pancreatic tissues from patients who underwent resection of a primary pancreatic adenocarcinoma. The cases were selected only on the basis of having their neoplastic and normal tissues available on existing TMAs. There was no selection of cases based on any clinical criteria (such as diabetes) or pathological criteria other having PDAC and PanIN tissues. We used tissues from cases that had undergone pancreaticoduodenectomy because these represent ~80% of the resections for pancreatic head tumors. The areas of normal pancreas selected for analysis were chosen to avoid as much as possible areas of chronic pancreatitis that can result from ductal obstruction by the tumor. TMA1 included representative cores of PDAC. TMA2 included representative cores of PDAC and PanIN lesions of all grades (PanIN-1-2-3) identified from the patients' resected pancreata. TMA methods are further described in 24. After in situ hybridization, sufficient tissue was available for analysis from 32 patients. Each TMA also contained histologically normal pancreatic tissues from patients' resected pancreata. Matching normal tissue was available for most cases. Clinico-pathological data of the patients included were retrieved from the Johns Hopkins Surgical Pathology database.
In-situ hybridization was performed using RNAScope (v2.0 High-Definition–Red; Advanced Cell Diagnostics, Inc. Hayward, CA), an assay based on branched-chain amplification steps that allows for sensitive detection and quantification of RNA transcripts in tissue sections 25. Four μm-thick tissue sections of TMA1 and TMA2 were deparaffinized, and ISH was performed manually in two consecutive experiments using positive and negative controls, in accordance with the manufacturer's recommendations. Representative tissue cores of non-neoplastic prostate and lymphoid tissues present in the TMAs were also evaluated as internal negative controls, as these tissues have been shown to have low or absent GLP-1R expression26. Following ISH, every tissue core was digitally photographed to capture images of sufficient numbers of cells from each pancreatic cell type using an Olympus BX51 microscope equipped with an Olympus DP26 camera (Olympus America Inc. NY USA).
A pathologist with extensive experience in pancreatic pathology (HK) reviewed tissue cores to confirm the histological diagnosis and quantify hybridization signals. After using a training set of slides to gain experience in scoring, signals and nuclei were manually counted by two authors (MDM and HK) in an independent way using computer image analysis software ImageJ,v1.46 (available at http://rsb.info.nih.gov/ij/index.html). ISH signals were calculated as signal-to-cell ratios: for each digital image, the total number of ISH signals counted for each cell lineage (cancer, PanINs, endocrine, acinar, ductal)was divided by the number of nuclei present in the same cell population. This counting procedure was applied on tissue cores hybridized for GLP-1R as well as for the DNA-directed RNA polymerase II subunit RPB1 (POLR2A), which we employed as positive control in order to assess the level of GLP-1R expression against a common housekeeping gene.
Since most PDAC cells are several-fold larger than normal pancreatic ductal epithelial cells, and RNA abundance is thought to be associated with cell size 27, we reasoned that large cells such as cancer cells might have more abundant mRNA transcript levels than smaller cells, such as non-neoplastic pancreatic ductal cells. We therefore employed a statistical model in which the calculated glp-1r signals in each cell type were corrected for cell size (further explained in the Statistical analysis section). To calculate cell size we measured the average cell area as a surrogate indicator of size. Cell area was calculated by multiplying the two largest diameters in each cell. Each cell type was measured in the same way.
To accurately measure cell area, we immunolabeled additional TMAs slides for E-cadherin. E-cadherin was chosen because of its exclusive membrane localization. The staining allowed us to better visualize cell perimeter and obtain more precise measurements of cell diameters.
Immunohistochemistry (IHC)was performed using primary anti-E-cadherin rabbit monoclonal antibody (EP700Y, Cell Marque, Rocklin, CA), as previously described 24. After IHC, each core was photographed as described above for ISH. Measurements of the two largest cell diameters were taken using ImageJ software. A total of 50 images were evaluated for each cell type.
Values are presented as median and range of the in situ hybridization/cell volume ratios and as mean and standard deviation (SD) of the log-transformed ratios, as well as total and mean number of cells analyzed. Two mixed-effects linear regression models were used to evaluate the association between ISH signal intensity and cell type, with log-ratios as the response and fixed effect terms for cell type, cell size and TMA. We included cell type using two approaches: in the first approach, cell types were ordered according to the pancreatic carcinogenesis model from no dysplasia to progressively higher grades of dysplasia (PanIN-1, PanIN-2, PanIN-3), to PDAC cells (Table 1), and the slope of the linear relationship calculated.
It is still a matter of debate whether PDAC originates from ductal or acinar precursor lesions. The commonly accepted morphological pathway of pancreatic carcinogenesis postulates a stepwise progression from ductal cells to intraepithelial lesions (PanINs) and eventually cancer 28. However, since experimental evidence suggests that PDAC cells may originate from an acinar progenitor cell 29-31, we included the expression levels of both acinar and ductal cells in our model.
In the second approach, indicator variables for each cell type were included in the models, yielding 15 pairwise comparisons between combinations of cell types. Both modeling approaches included a random intercept for each patient, to account for within-patient correlation across cell types. Results are expressed as the difference in log ratios between cell types, and as the ratio of means of the untransformed ratios, calculated by exponentiating the difference. All pairwise differences in signal-to-cell ratios between cell types were estimated, and the type-I error rate was controlled using a Bonferroni correction, resulting in p-values <0.05/15 = 0.003 as a threshold for statistical significance. All analyses were performed using R version 3.0 (R Core Team-2014; R Foundation for Statistical Computing, Vienna, Austria, http://www.R-project.org).
The clinicopathological characteristics of the 32 patients included are presented in Table 1. None of the patients were known to have been taking GLP-1R agonists.
Glp-1r mRNA signals were found in endocrine, acinar, ductal, PanIN and PDAC with variable levels of expression (Figure 1 and Table 2). No signals were observed in prostate or lymphoid tissues (Supplemental Figure 1),in agreement with previous data 26. To identify potential differences between ISH experiments, we compared mean signal-to-cell ratios of acinar cells, which was the most represented cell type, between TMA1 and TMA2 and found no significant difference (p=0.68, two-tailed t-test; data not shown).
Acinar cells were the most prevalent cell type with a median number of 1020 cells analyzed across all tissue cores. The least represented cell type were ductal cells (median number: 85), although we observed a considerable variability in the number of ductal cells across different tissue cores (Table 2). Overall, GLP-1R signal-to-cell ratios were < 0.5 in every cell type observed. Endocrine cells had the highest (0.196) and ductal cells had the lowest (0.01) median signal-to-cell ratio, whereas PanINs, acinar cells and PDAC cells showed intermediate values. We did not find any outliers with respect to GLP-1R signals of individual cells within any of the cell groups, nor did we not observe significant inter-patient variability with respect to signal to cell ratios within the various cell types.
In support of previous data 32, PDAC cells were the largest cell type, with a mean value of 592 μm2, whereas ductal cells had a mean value: 265.8 μm2 and were the smallest cell type observed (Table 3). A significant difference in cell area (hereinafter referred to simply as “size”) was observed across cell types (1-way ANOVA, p<0.001). To examine the relative level of GLP-1R expression, we compared the signal-to-cell ratios calculated for islet, acinar, ductal and PDAC cells with the signal-to-cell ratios obtained from the positive control (POLR2A gene), in the same cell types (Welch two-sample t-test, Figure 2). The mean GLP-1R signal-to-cell ratio was 0.221 in endocrine cells, 0.097 in PDAC cells, 0.012 in ductal cells and 0.043 in acinar cells. The mean POLR2A signal-to-cell ratio was 0.333 in endocrine cells, 0.564 in PDAC cells, 0.273 in ductal cells and 0.338 in acinar cells. Analysis of endocrine cells showed no significant difference in the level of GLP-1R expression compared with POLR2A (95% Confidence Interval (CI): -0.259; 0.036, p=0.122). Expression of POLR2A was significantly higher than GLP-1R in cancer cells (95% CI: -0.650, 0.284; p<0.001), acinar cells (95% CI: -0.429, 0.160; p=0.001) and ductal cells (95% CI: -0.330, 0.192; p <0.001).
We next compared the level of GLP-1R expression in non-neoplastic acinar, endocrine and ductal pancreatic cells (Table 4). The GLP-1R signal-to-cell ratio was significantly higher in endocrine cells than in acinar and ductal cells (p=0.002 and p<0.001, respectively). The signal-to-cell ratios in acinar cells tended to be higher than the ratios in ductal cells, although the difference was not significant (p=0.061).
The first mixed-effect linear regression modeling approach was used to investigate the relationship between glp-1r signal-to-cell log-ratios and cancer cells and their putative precursor lesions. In the regression model, which included acinar cells, ductal cells, all grades of PanIN and cancer cells, the slope of the fitted line was not significantly different from zero (Slope = 0.07; 95% Confidence Interval: 0.91 to 0.244, p=0.39), suggesting that progression from normal pancreas cells (acinar or ductal cells) to PanINs and eventually PDAC is not associated with a significant increase in glp-1r expression.
We also analyzed the expression data using a series of pairwise comparisons involving PDAC cells, all grades of PanIN, and non-neoplastic cells (ductal, acinar cells) to test for significant differences in their respective signal-to-cell ratios (second regression, Table 5). Acinar cells and PDAC cells showed higher glp-1r ratios than ductal cells (p=0.006 and p=0.021, respectively). However, these differences were not statistically significant after adjusting for multiple comparisons. There were no other statistically significant differences in any of the other comparisons. Specifically, we did not observe differences in ratios between different PanIN grades (PanIN-1 vs. PanIN-2, p=0.406; PanIN-1 vs. PanIN-3, p=0.154; PanIN-2 vs. PanIN-3, p=0.684). Similarly, there was no significant difference in glp-1r expression between PDAC cells and any grade of PanIN, between PDAC cells and acinar cells, or between PDAC cells and ductal cells. Further more, we did not find any outliers with respect to glp-1r expression in the PDAC cases that might indicate GLP1-R overexpression in one or two of the cancers relative to the group overall or within the subgroup of patients that had been diagnosed with diabetes at the time of surgery.
The most reliable evidence that a gene has oncogenic functions is finding activating mutations, gene amplification or fusions of the gene. This is further substantiated by the observation that selective inhibition of the gene or gene product can suppress its oncogenic phenotype. In the literature genes that are overexpressed in cancer cells are identified as oncogenes, particularly if the gene is thought to have growth-promoting functions.
GLP-1R agonists have been recently suggested to promote PDAC, due to the putative action of GLP-1R on cell growth and proliferation5. Thus far, there is no definitive evidence supporting this hypothesis. Moreover, reliable data regarding the expression of GLP-1R in normal and neoplastic pancreas remains elusive.
In this study we sought to determine if glp-1r expression is a feature of PDACs, PanINs and non-neoplastic pancreatic cells. We chose to employ RNA in situ hybridization to measure gene expression because in contrast to the uncertain antigenic specificity of most commercially available antibodies 6, 7, 20, RNA probes for ISH have been reported to have excellent sensitivity and specificity, even in FFPE tissues33, 34.
Given the relatively low level of GLP-1R expression in pancreatic islets reported by microarray and RNA sequencing studies35, 36, we chose to assess gene expression in a quantitative manner. Although low levels of glp-1r transcripts were present in all cell types, endocrine cells expressed glp-1r at a significantly higher level than PDAC cells or any other cell type. Importantly, the level of expression in PDAC cells was not significantly different from that of normal pancreatic acinar or ductal cells or PanIN cells, which would be expected if GLP-1R were an important contributor to PDAC development or progression.
We would have expected that if GLP-1R were functioning as an oncogene promoting pancreatic carcinogenesis, at least some of the PDACs we tested would have had much higher levels of glp-1r expression than normal pancreas cells. However, we never observed PDAC cells expressing significantly higher levels of glp-1r than normal pancreas cells. Pancreatic expression of GLP1-R using immunohistochemistry and autoradiography was recently reported and found similar results to ours37. These authors did not identify GLP-1R expression in pancreatic adenocarcinomas, although they did find low levels of GLP-1R expression in PanIN-1 and PanIN-2 lesions. Together, these findings and our results indicate that of the level of expression of glp-1r transcripts in normal and neoplastic pancreatic ducts may be below the level required for protein expression to be detected using techniques such as immunohistochemistry.
Interestingly, we also detected glp-1r expression in normal pancreatic acinar cells, confirming the findings recently reported by Waser et al37. This observation may explain the occurrence of pancreatic gland hypertrophy arising from glp-1r agonist therapy described in animal models. Koehler et al, recently reported that glp-1r agonists increase protein, but not DNA expression or cell proliferation in pancreatic acinar cells38.
Although the lack of glp-1r overexpression in PDAC cells suggests that direct effects of incretins on pancreatic cells may not be important in driving carcinogenesis, we cannot exclude effects related to systemic metabolic effects of these agents on the pancreas. Similarly, our findings do not preclude the possibility that PDAC cells could respond to GLP-1R agonists differently to normal pancreas cells without requiring gene overexpression.
Ultimately, the long-term outcome results of ongoing randomized trials will reveal if there is any significant effect of these drugs on the incidence of PDAC or pancreatitis. So far the initial results of randomized trials do not find evidence for an increase in the incidence of pancreatitis or PDAC associated with incretin use 39, 40.
In conclusion, we find that normal and neoplastic pancreatic cells express low levels of glp-1r transcripts. Our results suggest that PDAC cells or their precursor lesions do not overexpress glp-1r compared to non-neoplastic pancreatic cells.
Grant Support: This work was supported by NIH grants (CA62924, R01CA176828), Susan Wojcicki and Dennis Troper and the Michael Rolfe Foundation.
Disclosures: Dr. Goggins has served as a consultant for Takeda, Merck, Novo Nordisk and AstraZeneca/BMS.