|Home | About | Journals | Submit | Contact Us | Français|
The Hedgehog (HH) pathway has been identified as an important deregulated signal transduction pathway in pancreatic ductal adenocarcinoma (PDAC), a cancer type characterized by a highly metastatic phenotype. In PDAC, the canonical HH pathway activity is restricted to the stromal compartment while HH signaling in the tumor cells is reduced as a consequence of constitutive KRAS activation. Here we report that in the tumor compartment of PDAC the HH pathway effector transcription factor GLI1 regulates epithelial differentiation. RNAi-mediated knockdown of GLI1 abolished characteristics of epithelial differentiation, increased cell motility and synergized with TGFβ to induce an epithelial-to-mesenchymal transition (EMT). Notably, EMT conversion in PDAC cells occurred in the absence of induction of SNAIL or SLUG, two canonical inducers of EMT in many other settings. Further mechanistic analysis revealed that GLI1 directly regulated the transcription of E-cadherin, a key determinant of epithelial tissue organization. Collectively, our findings identify GLI1 as an important positive regulator of epithelial differentiation, and they offer an explanation for how decreased levels of GLI1 are likely to contribute to the highly metastatic phenotype of PDAC.
With a 5-year survival rate around five percent and about 37.000 deaths per year, pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal human cancers and ranks fourth among cancer-related deaths in the United States (1, 2). The extremely poor clinical prognosis of PDAC is partially attributed to its high tendency to metastasize (3). However, the molecular mechanisms underlying invasion and metastasis in pancreatic cancer remain poorly understood.
Epithelial-to-mesenchymal transition (EMT) is a process in which cells lose their epithelial character and acquire a migratory mesenchymal phenotype (4). While being crucial for normal metazoan development, EMT (and the reverse process mesenchymal-to-epithelial transition MET) is thought to be recapitulated in metastasizing cancer cells (5).
Loss of the homophilic cell adhesion molecule E-Cadherin, which is a main determinant of epithelial tissue organization and cell polarity, is considered a hallmark of EMT (6). A multitude of in vitro and in vivo models show enhanced invasiveness and metastasis upon E-Cadherin (CDH1) repression, while E-Cadherin overexpression conversely leads to a significant decrease in tumor malignancy (7-9). Repression of E-Cadherin occurs primarily on the transcriptional level and is in many instances mediated by direct binding of transcriptional repressors like SNAIL, SLUG or TWIST to E-Box consensus sequences in the CDH1 promoter (10). Early invasion and metastasis as prevalent traits of pancreatic cancer suggest a prominent role for EMT and its upstream activators in the pathogenesis of PDAC (11-13).
Hedgehog (HH) signaling is one of twelve deregulated signal transduction pathways in pancreatic cancer (14, 15). Recent work shows that HH pathway activity in this cancer type is asymmetrically distributed: While the epithelial tumor compartment constitutes the source of HH ligands, high HH pathway activity is predominantly associated with the stroma (16-19). The HH-activated stroma is in turn responsible for the production of tumor growth-promoting factors (17). HH pathway activity is not absent in the tumor cells, but is significantly lower compared to the stroma of human and mouse PDAC (16). Despite its low abundance in tumor cells, GLI1, a member and a transcriptional target of HH signaling, contributes to in vitro cell proliferation, anchorage-independent growth and cancer cell chemoresistance (18, 20, 21). The low HH/GLI activity in PDAC tumor cells are, at least to some extent, the result of mechanisms activated by mutant KRAS, which is a key driver of malignant development in the pancreas: First, KRAS leads to the abrogation of primary cilia on PDAC cells, an organelle crucial for the reception and transmission of signaling induced by HH ligands (22). Second, KRAS actively suppresses signaling events downstream of the primary cilium (19). It is currently unclear if the asymmetrical distribution of HH/GLI activity in pancreatic cancer is of pathophysiological significance or if it represents a mere byproduct of additional cancerous alterations.
Here we provide evidence that the low GLI1 level usually found in the epithelium of pancreatic carcinoma primes the cells towards an EMT. GLI1 is a transcriptional activator of E-Cadherin (CDH1), a main determinant of epithelial tissue organization and cell polarity (6). Lowering the endogenous GLI1 levels in PDAC cells experimentally results in loss of E-Cadherin expression, changes in cell morphology typically associated with a mesenchymal phenotype and an increase in cancer cell motility. We find that GLI1 levels significantly correlate with CDH1 expression in pancreatic cancer cell lines and in primary patient material. Interestingly, the effects of GLI1 on CDH1 expression do not require the upregulation of several well-established EMT inducers including SNAIL and SLUG and are instead mediated by the direct binding of GLI1 to the CDH1 promoter. Moreover, moderately decreased GLI1 expression significantly synergizes with stroma-derived EMT- and migration-inducing factors such as TGFβ and HGF. These data ascribe a functional role for HH pathway suppression in cancer and propose that the reduced expression of an oncogene might be of functional relevance for certain aspects of tumor development.
PDAC cell lines were obtained from ATCC and were cultured at 37 °C and 5 % CO2 in DMEM (high Glucose) plus 10 % heat-inactivated FBS plus 1 mM Na-Pyruvate and Penicillin/Streptomycin. Cell lines were not longer passaged than 6 months. Recombinant TGF-β1 and recombinant HGF (both R&D Systems) were used at final concentrations of 5 ng/ml and 10 ng/ml respectively. SB-431542 (Sigma) was employed at a final concentration of 10 μM; corresponding amounts of DMSO were added to the untreated samples. Cells were usually treated 24 h after siRNA transfection and treatment was maintained for 48 h. The E-Cadherin-Luciferase reporter construct was a kind gift of Lluis Lajas (INSERM, Montpellier, France).
Cells were seeded at 50-70% confluency at transfected with siRNA using Dharmafect1 according to the instructions of the manufacturer. Fresh medium was added 24 h after the beginning of the transfection. If not stated otherwise, cells were harvested 72 h after transfection.
Cultured cells were lysed in SDS-Buffer, proteins were separated on SDS gels and transferred onto PVDF membranes. Detection of blotted proteins was by incubation of the membranes using the following antibodies: anti-βActin (Sigma (A5441));; anti-HA Tag (Cell Signaling (#2367)); anti-E-Cadherin (BD Biosciences (#610181)); anti-β-Catenin (Cell Signaling (#9587)), anti-α-E-Catenin (Cell Signaling (#3236)); anti-Twist1/2 (Santa Cruz (sc-15393)); anti-Zeb1 (Cell Signaling (#3396)); anti-Zeb2 (Sip1) (Santa Cruz (sc-271984)); anti-Snail (Cell Signaling (#3879)); anti-Slug (Cell Signaling (#9585)).
Cells were seeded on cover slips or chamber slides (Nunc) and were fixed for 10 minutes in 4% formaldehyde/PBS at room temperature. Staining of permeabilized cells was performed using the following antibodies: α-E-Cadherin (Cell Signaling (#4065)); α-β-Catenin (Cell Signaling (#9587); anti-α-E-Catenin (Cell Signaling (#3236)); α-Cytokeratin (CAM 5.2) (BD Biosciences (#347653)); α-Keratin 14 (Covance (#PRB 155-P)); Texas Red Phalloidin (Invitrogen). Subsequently, cells were mounted in Vectashield containing DAPI (Vector Labs). Microscopy was performed on a Leica DMR fluorescence microscope, pictures were taken with a QuantiFire XI camera (Intas) and processed with Photoshop (Adobe).
In vitro scratch assay was performed according to Liang et al (23). Pancreatic cancer cell lines suitable for this assay were chosen according to growth characteristics and their ability to form monolayers in the applied time frame. Cells seeded in 24 well cell culture plates were transfected at 60 – 70% confluency in triplicates. Scratching was performed 48 hours post transfection. Growth medium was removed and straight incisions were made with a P200 pipette tip at a 90° angle. Cells were washed several times with PBS to remove detached cells and supplied with new growth medium. Pictures of the scratches were taken at 0 and 24 hours using an Axiovert 35 inverted microscope (Zeiss) equipped with a Scion digital camera. To ensure the recording of the same scratch area, reference points were made on the culture plate with a marker pen prior to photography. Migration was measured as closure of the scratched area as suggested in ref. (23). Images were processed using TScratch (CSElab, ETH Zurich; http://www.cse-lab.ethz.ch/) (24). Open scratch area was determined automatically according to developer's suggestions and – to acquire higher accuracy – adjusted manually. Migration was subsequently defined as ratio of open scratch area after 24 h and initial scratch area. At least three independent experiments were performed in triplicates.
Based on the high propensity of pancreatic cancer cells for early metastasis it is generally assumed that these cells harbor intrinsic alterations which prime them towards an EMT, an important pro-migratory and pro-metastatic driving force in tumors (6). Since HH/GLI activity has been linked to the induction of an EMT in other settings (25-27) we were interested to elucidate the role of GLI transcription factors in pancreatic cancer. We transiently transfected PDAC cell lines with small interfering RNA (siRNA) targeting either of the two activating GLI isoforms, GLI1 or GLI2 (Fig. 1A). Knocking down endogenous GLI1 or GLI2 in Panc1, Su86.86, AsPC1 or Hs766T cells induced a striking reduction of CDH1 expression. As a positive control, we applied recombinant transforming growth factor β (TGFβ), a potent EMT inducer in PDAC cells (11, 13). Interestingly, CDH1 reduction was more pronounced in siGLI-transfected cells than in TGFβ-treated samples, suggesting a powerful contribution of GLI factors to the regulation of CDH1 expression. Given that GLI2 and GLI1 can induce each other's expression (28, 29), it was possible that the effects of siGLI2 were, at least in part, mediated by a reduction in GLI1 expression. However, this GLI1/GLI2 regulatory circuit was not fully functional in AsPC1 and Hs766T cells, allowing for the selective knock-down of GLI1 without changes in GLI2 expression (Fig. 1A). These experiments showed that knocking down GLI1 alone was sufficient for the repressive effects on CDH1 expression. The significant reduction of CDH1 expression upon loss of the GLI transcription factors could also be verified on the protein level using immunoblotting (Fig. 1B) and immunofluorescent staining experiments (Fig. 1C). Supporting these findings and excluding off-target effects, similar results were obtained using additional siRNA constructs to knock down GLI1 (Suppl. Fig. 1). Conversely, overexpression of GLI1 strongly induced the transcription of CDH1 compared to mock-transfected cells. The increased CDH1 expression in these cells could not be documented on the protein level, suggesting that MiaPaCa2 cells harbor additional post-transcriptional mechanisms to decrease CDH1 protein amounts. Similar findings were recently reported by Song et al (30). Interestingly, we were unable to detect a further CDH1 induction upon transfection of GLI1 in Panc1 and Su86.86 cells (Suppl. Fig. 2), suggesting that the GLI1-CDH1 axis is saturated in these cells. Hypothesizing that limiting amounts of transcriptional GLI-cofactors could be the reason for the lack of CDH1 induction after GLI1 transfection, we assumed that titrating out these cofactors should interfere with the capability of endogenous GLI1 to induce E-Cadherin expression. To address this question we cloned a GLI1 construct lacking the five zinc-finger domains which are required for DNA binding (GLI1ΔZF; Suppl. Fig. 3a). As expected and in contrast to full-length GLI1, this construct was unable to activate a luciferase reporter containing GLI-binding sites (Suppl. Fig. 3b). Despite the inability to bind to DNA however, Panc1 cells stably expressing GLI1ΔZF showed a prominent scattering and mesenchymal phenotype (Suppl. Fig. 3c). In line with the morphological appearance, GLI1ΔZF-expressing cells had significantly reduced levels of E-Cadherin mRNA and protein (Suppl. Fig. 3d and and3e).3e). Taken together, these results strongly suggest that transcriptional GLI cofactors are limiting and set the upper limit for GLI-induced CDH1 induction.
Given the fact that a knock-down of GLI1 was sufficient to regulate CDH1 expression without alterations in GLI2 levels, we focused on GLI1 in our further studies. In order to expand our findings on the GLI1-CDH1 interplay we expanded our analysis and measured GLI1 and CDH1 expression in 15 PDAC cell lines. Figure 2A (and suppl. fig. 4) shows a statistically significant positive correlation between GLI1 and CDH1 expression in this panel. Interestingly, cell lines which harbor wild-type KRAS did not fully fit into the correlation (Fig. 2A). However, the correlation data suggested a more direct interaction between GLI1 and CDH1. Bioinformatics studies identified 13 candidate binding sites for GLI transcription factors located in 5 different clusters within 3500 bp upstream of the transcriptional start site in the CDH1 gene (Fig. 2B). In agreement with our previous results, overexpression of GLI1 in Panc1 cells showed that this transcription factor is able to increase the activity of a luciferase construct driven by a 3 kb fragment of the CDH1 promoter (Fig. 2C). Finally, chromatin immunoprecipitation experiments demonstrated binding of endogenous GLI1 to the CDH1 promoter (cluster #4) in Panc1 cells (Fig. 2D). In order to investigate if this GLI1-CDH1 relationship can also be found in vivo, we analyzed 144 primary tumor samples from pancreatic cancer patients and found a highly significant positive correlation between the GLI1 and CDH1 expression levels (Fig. 2E). Taken together, these results establish CDH1 as a novel transcriptional target of GLI1 in PDAC cells.
Since the repression of CDH1 expression during an EMT is normally executed by transcriptional repressors such as SNAIL, which directly bind to the CDH1 promoter, we wondered if the well-established EMT repressors are induced in pancreatic cancer cells after GLI1 knockdown. As can be seen in figure 3A, neither SNAIL (SNAI1) nor SLUG (SNAI2) are induced by siGLI1. A lack of SNAIL induction was also verified on the protein level using siGLI1-transfected Panc1 cells (Fig. 3B). In contrast, TGFβ was capable of significantly inducing SNAIL expression on the mRNA and protein level in these cells (Fig. 3A and 3B). In line with these results, other established CDH1 repressors (TWIST1, TWIST2, ZEB1, ZEB2, TCF3, TCF4, GSC) were not induced after siGLI1 transfection, strongly suggesting a mechanism which does not rely on the induction of these CDH1 repressors (Suppl. Fig. 5a). Since TGFβ is a very potent EMT inducer and many PDAC cells express endogenous TGFβ (31) we wanted to investigate if a reduction in GLI1 levels would sensitize the cells to TGFβ. However, using a selective inhibitor of TGFβ signaling (SB-431542) we found no effect on the ability of siGLI1 to repress CDH1 expression regardless if only endogenous or additional recombinant TGFβ was present (Fig. 3C). The functionality of the TGFβ antagonist was verified by measuring the expression of the TGFβ target gene PAI1 (Fig. 3C). Taken together, the effects of siGLI1 on CDH1 do not require the transcriptional induction of classical EMT inducers. Given that some EMT molecules are regulated on the protein level, we further analyzed the protein levels of SLUG, TWIST1/2, ZEB1 and ZEB2 in Panc1 cells transfected with siGLI1 (Suppl. Fig. 5b). In particular ZEB1 was stabilized in cells with GLI1 knock-down. However, these alterations were also observable in cells with CDH1 knock-down (Suppl. Fig. 5b), arguing that the changes in ZEB1 protein amount are most likely associated with the induction of an EMT phenotype per se and are not GLI-mediated.
The cell-cell adhesion molecule E-Cadherin is a major determinant of the organization of epithelial tissues and its loss is a hallmark of an EMT. E-Cadherin loss is associated with characteristic changes in cellular morphology resulting from reduced cell-cell contact and the gain of mesenchymal traits (7, 8). As shown in figure 4A, similar to TGFβ, the GLI1 knockdown led to specific alterations in PDAC cell morphology. These cells adopted a more mesenchymal, spindle-like shape and left the epithelial cell clusters which are normally found in control samples. Staining the actin cytoskeleton with phalloidin furthermore revealed a rearrangement of F-actin filaments from the cortical periphery into internal stress fibers, similar to what can usually be found in mesenchymal cells. The changes in morphology could also be observed by transient cotransfection of cells with a short hairpin RNA (shRNA) targeting GLI1 (shGLI1) (Fig. 4B).
Next, we analyzed changes in epithelial marker expression upon knock-down of GLI1. In line with the loss of E-Cadherin in siGLI1-transfected cells, additional important epithelial marker genes were also significantly decreased in numerous PDAC cell lines, such as Keratin 19 (KRT19) or the adherens junctions components EVA1 and PTPRM (Fig. 4C) (32). In contrast, Integrin β1 (ITGB1), which mediates the interaction with the extracellular matrix (ECM) and which has previously been implicated in cancer cell metastasis (33), was clearly upregulated (Fig. 4C). Moreover, global analysis of gene expression by means of cDNA microarray revealed a widespread reduction of cytokeratins in Panc1 cells transfected with GLI1 siRNA compared to cells transfected with control siRNA (Fig. 4D). Reduction of cytokeratin expression was also verified by immunostaining in Su86.86 cells (Fig. 4E). Thus, together these findings define GLI1 as a gatekeeper of the epithelial phenotype in PDAC cells.
In epithelial cells E-Cadherin is physically linked to catenins, which mediate the connection to the actin cytoskeleton. EMT induction in pancreatic cancer cells by exposure to TGFβ results in the disassembly of these adherens junctions concomitant with a destabilization of α- and β-Catenin (34). In line with these findings, α- and β-Catenin were destabilized in siGLI1-transfected PDAC cells and were grossly reduced in immunofluorescence and western blotting experiments (Fig. 5A and 5B). Importantly, the remaining β-Catenin in siGLI1-transfected cells was mainly localized to the cytoplasm and was absent from the nucleus (Fig. 5A). In line with these observations, knockdown of GLI1 did not result in an induction of WNT signaling as measured by means of WNT luciferase reporter assays (Fig. 5C). In support of these findings, the WNT target gene AXIN2 was not induced upon transfection of cells with siGLI1 (data not shown). Similar results were recently described by Herzig et al in a model of β-cell carcinogenesis showing lack of WNT pathway activation after loss of E-Cadherin (35). These data support a function for GLI1 in the maintenance of adherens junctions and in WNT pathway responsiveness, thus, further emphasizing its role as an epithelial differentiation factor.
The loss of CDH1 has been shown to promote cancer cell motility (7, 8). In order to elucidate if low GLI1 levels enhance cellular migration we performed in vitro scratch assays with confluent PDAC cells which had been transfected with control siRNA, siCDH1 (as a positive control) or siGLI1. In agreement with our previous finding that knock-down of GLI1 results in decreased CDH1 expression, PDAC cells transfected with siGLI1 closed the scratch area much more efficiently than control cells (Fig. 6A and 6B). With respect to their migratory potential, the siGLI1-transfected cancer cells were equipotent to the positive control, siCDH1-transfected cells (Fig. 6A and 6B). Similar results were obtained using an independent set of GLI1-specific siRNA (Suppl. Fig. 6). Importantly, the increased closure of the scratch was not due to changes in cell number, as shown in figure 6C.
Pancreatic tumor and stroma cells express a variety of EMT-inducing factors, such as TGFβ and hepatocyte growth factor (HGF) (12). In order to investigate the functional interplay of lowered GLI1 levels with these factors, we used siRNA concentrations which minimally affected GLI1 expression. In addition we applied recombinant TGFβ in a concentration range between 0.1 and 1 ng/ml, which had no discernable effect on CDH1 or GLI1 expression when applied alone (Fig. 7A). However, the combination of low siGLI1 and low TGFβ concentration had a synergistic effect on both, CDH1 expression (Fig. 7A) and cellular motility (Fig. 7B and 7C). A similar synergistic effect on E-Cadherin was also obtained using Su86.86 cells (not shown). To expand our analysis on synergism we tested HGF as another motility-promoting factor present in the PDAC microenvironment. As can be seen in figure 7B and 7C, 2.5 ng/ml HGF (which on its own had no effect on cell migration) potently synergized with low siGLI1 in the in vitro scratch assay. Taken together we could observe a synergistic behavior of only slightly reduced GLI1 levels with TGFβ or HGF on pancreatic cancer cell motility.
The strong propensity of pancreatic cancer cells for invasion and metastasis implies that these cells are primed to undergo an EMT due to intrinsic alterations and/or extrinsic signals provided by the abundant microenvironment called desmoplasia. The developmental Hedgehog signaling pathway, which is normally inactive in the adult pancreas, has been shown to be reactivated in PDAC where it promotes stromal hyperplasia, myofibroblast differentiation and production of ECM (36, 37). Contrasting an earlier perception, canonical HH signaling in tumor cells is dispensable for cancer development as was recently shown by the epithelium-specific deletion of Smo, a critical bottleneck of HH signaling, in a PDAC mouse model (18). Interestingly, mice with a homozygous Smo deletion in the pancreatic epithelium had a significantly reduced survival compared to heterozygotes (18), arguing that a reduction in basal HH signaling caused a higher mortality in these mice. In this respect, pancreatic cancer contrasts other malignancies such as basal cell carcinoma or medulloblastoma, in which high GLI levels in the tumor cells are necessary and sufficient for disease progression (38, 39). In PDAC, stromal GLI1 levels are up to 120-150 fold higher than in the neighboring epithelium (16), possibly being the result of mutant KRAS acting on primary cilia and on downstream HH pathway components (19, 22). We could demonstrate for the first time that the low GLI1 levels associated with the epithelial compartment prime tumor cells towards the induction of an EMT. In PDAC cells, GLI1 is a positive regulator of E-Cadherin, one of the major epithelial determinants. Reduced expression of GLI1 results in a widespread loss of epithelial markers and leads to the acquisition of a mesenchymal morphology. Similar effects might be expected from a manipulation of endogenous GLI2, of which the full-length activator form could be detected in PDAC cells (Suppl. Fig. 7). In contrast, transfected GLI3 behaves as a transcriptional repressor on a HH/GLI reporter and an E-Cadherin reporter construct (Suppl. Fig. 8). Thus, the possibility exists that, following GLI1 removal, endogenous GLI3 occupies the empty GLI binding sites in the CDH1 promoter contributing to E-Cadherin repression. Alternatively, GLI3 might repress E-Cadherin via lowering of GLI1 expression.
Surprisingly, we could not detect an upregulation of mesenchymal marker genes such as Vimentin, Fibronectin or N-Cadherin in siGLI1-transfected cells. To rule out the possibility that the time frame of our transient siRNA transfection experiments precluded the detection of secondary alterations in mesenchymal gene expression, we performed two successive rounds of siRNA transfection and thus expanded our analysis to 6 d posttransfection. However, no upregulation of mesenchymal genes was observed (not shown), arguing for a selective control of epithelial marker genes by GLI1. However, despite the lack of altered mesenchymal gene expression, significantly increased cell motility was observed in cells with low GLI1.
Importantly, our data do not dispute a functional role for epithelial GLI1 in tumor growth and in fact, in vitro investigations document the importance for GLI1 in cell proliferation, anchorage-independent growth and chemoresistance (18, 20, 21). Regarding the latter, we could find that Gemcitabine exposure to PDAC cells resulted in upregulation of both, GLI1 and CDH1. More work will have to unravel the exact interplay between GLI1, EMT and chemoresistance in pancreatic cancer. However, given the finding that even very modest reductions in GLI1 mRNA strongly synergized with TGFβ or HGF suggests that the cellular GLI1 expression level acts as a potent sensitizer regulating the epithelial phenotype. These data imply that by adjusting the GLI1 expression levels, a cell could switch between a pro-proliferative (higher GLI1) and a pro-migratory (lower GLI1) state. In fact, it is known that cells undergoing an EMT proliferate less than control cells (40, 41). Therefore, it will be interesting to learn if GLI1 expression levels are undulating in vivo in phases of primary tumor growth, invasive growth and metastasis. In this respect it is interesting to see that MET (mesenchymal-to-epithelial transition), the reversal of EMT, is crucial for successful metastasis (42). In line with this concept and our findings, HH/GLI2 activation has been shown to induce CDH1 expression, maintain an epithelial phenotype in non-PDAC cell types and block keratinocyte invasiveness in 3D culture (43-45).
However, our findings in pancreatic cancer cells are to some degree in contrast to certain earlier reports in non-transformed kidney cells, prostate carcinoma and nonalcoholic fatty liver disease (25-27). One potential reason being that the alimentary tract including the pancreas and the stomach are endoderm-derived organs, suggesting that certain lineage- or tissue-specific effects might be of importance for the GLI1-CDH1 regulation. In fact, we could find a CDH1 decrease upon siGLI1 transfection also in carcinoma cells of the lung, another endodermally-derived organ (not shown).
One recent report in PDAC cells claims a GLI1/MUC5AC-dependent E-Cadherin destruction process (46). However, this study describes a posttranslational mechanism which could only be detected under certain culture conditions and which utilized non-physiological GLI1 levels (46), raising doubts about the generalization of these events. Since our data show that overexpression of a mutant GLI1 lacking the zinc-finger domains is capable of inducing an EMT, similar indirect effects might be obtained by strong overexpression of full-length GLI1, giving a potential explanation for some of the differences also to another study in which ectopic GLI1 expression in normal pancreatic duct cells reduced CDH1 expression (47). In some instances, the in vitro use of pharmacological inhibitors such as Cyclopamine might have been misleading given their HH-unspecific nature (17). In addition to potential off-target effects in the tumor cells, the possible influence of Smo inhibitors on the Gli-positive stroma in vivo might complicate the interpretation of the results (47). In an effort to address the reason for some of the discrepancies, we determined if the disease stage of the tumor cell (primary tumor versus metastasis) might play a role in the GLI1-responsiveness but could not find an altered CDH1 regulation in a primary tumor/metastasis cell line pair (Suppl. Fig. 9).
Pancreatic ductal adenocarcinoma is characterized by the vast presence of stromal cells and the high content of extracellular matrix (ECM). In this context it is very interesting that PDAC cells express Integrin β1 upon a reduction of GLI1. Integrins are key molecules implicated in the binding to and the communication with the ECM and the subsequent signaling by focal adhesion kinase (FAK) (48). As a result, the cellular GLI1 levels might not only determine the cell-specific epithelial phenotype but also define its interactive potential with the environment. In fact, Integrin β1-FAK signaling has been shown to be essential for the proliferation of mammary carcinoma cells at metastatic niches (33).
Interestingly, TGFβ has recently been shown to act as an inducer of GLI2 and, as a result, also GLI1 (18, 49, 50). This raises the question of whether the EMT-promoting capability of TGFβ is simply stronger than the concomitant (MET-promoting) GLI induction or if the induction of GLI factors constitutes part of a negative feedback mechanism.
In summary, we identified GLI1 as a potent positive regulator of E-Cadherin (CDH1), the major gatekeeper of the epithelial phenotype in pancreatic cancer cells. Since this regulation involves direct GLI1 binding to the CDH1 promoter and does not depend on the upregulation of well-established CDH1 repressors like SNAIL, SLUG or TWIST, it adds a novel mechanism to the regulatory circuits defining the differentiation state of tumor cells. Our findings place GLI1 in a central position within the tumor signaling network and associate key cellular decisions with the HH pathway.
We are grateful to Wolfgang Meissner for RNA quality control, the LOEWE-financed Genomics Core Facility at the IMT and Florian Finkernagel for bioinformatics analysis. We are furthermore indebted to Bernhard Wilke for superb technical assistance and to Drs. R. Toftgard, F. Aberger and C. Bruns for the kind gift of materials.
This work was financed by grants from LOEWE (Tumor and Inflammation) to ML, SJ, VR, PSH and MK. MEFZ was supported by the Mayo Clinic Cancer Center, NCI CA136526, Mayo Clinic Pancreatic SPORE P50 CA102701, by the Mayo Clinic Center for Cell Signaling in Gastroenterology (P30 DK084567) and by the NIH grants CA136526, CA102701, DK084567. RRM obtained funding from the NIH grant K07 116303 and RRW from K07 116303.
Grant support: ML, SJ, VR, PSH, MK: LOEWE Program ‘Tumor and Inflammation’; MEFZ: NCI CA136526, Mayo Clinic Pancreatic SPORE P50 CA102701, Mayo Clinic Cancer Center, and Mayo Clinic Center for Cell Signaling in Gastroenterology (P30 DK084567); NIH grants CA136526, CA102701, DK084567; RRM: K07 116303; RRW: K07 116303.
Conflict of interest: None