Combining Hedgehog Signaling Inhibition with Focal Irradiation on Reduction of Pancreatic Cancer Metastasis

Pancreatic cancer continues to be the most difficult malignancy to treat, with the 5-year survival rate of around 5% (1). Unlike most other malignancies, only 15% to 20% of pancreatic tumors are resectable, and there is an 80% chance of recurrence after surgery. In this patient population, survival averages 20 months with the use of standard gemcitabine chemotherapy (2). The use of radiotherapy alone on pancreatic cancer is disputed due to the high-mortality rate and relatively small improvement with chemoradiotherapy (3). Since pancreatic cancer seems resistant to radiation, one strategy is to combine radiotherapy with another treatment option, such as a targeted drug.

Recent studies indicate that sonic hedgehog signaling can protect cancer cells against ionizing radiation therapy (4). The hedgehog (Hh) pathway, initially discovered in Drosophila, is a major regulator for cell differentiation, tissue polarity, and cell proliferation (5). The 7 transmembrane domain containing protein Smoothened (SMO) serves as the key player for the Hh pathway, whose function is inhibited by another transmembrane protein Patched (PTC) in the absence of Hh ligands (6). Binding of Hh to its receptor PTC releases this inhibition, allowing SMO to signal downstream, eventually to Gli transcription factors, primarily Gli2 transcriptional factor. As transcription factors, Gli molecules can regulate target gene expression by direct association with a specific consensus sequence located in the promoter region of the target genes (7, 8). Activation of the Hh pathway has been reported in pancreatic cancer, but the exact role of this pathway in pancreatic tumorigenesis and progression remains largely unclear (9–15).

There is evidence to indicate that Hh signaling is involved in the maintenance of a variety of tissue stem cells and cancer stem cells (CSC; refs. 16–27). Growing evidence indicates that CSCs are inherently resistant to radiation and other therapeutic treatments (28–32). We hypothesized that inhibition of Hh signaling can reduce the regeneration potential of CSCs or cancer progenitor cells, which should facilitate focal radiotherapy of pancreatic cancer.

Using an orthotopic model of pancreatic cancer metastasis, we first tested the effect of irradiation on tumor growth and metastases. While irradiation was effective in reducing tumor growth, tumor metastasis was not significantly affected. We hypothesized that CSCs within the pancreatic tumor may be responsible for this resistance and tumor metastasis. To test our hypothesis, we examined activity of the Hh signaling pathway as well as CSC markers in these tumors because Hh signaling is known to regulate CSCs (33). We first tested our hypothesis in vitro using CSCs-enriched tumor spheres, and then in an orthotopic mouse model.

Two smoothened signaling inhibitors were used in this study: CycT and BMS833923. CycT is a cyclopamine derivative provided by Logon Natural Products (Plano, Texas). CycT has been described in our previous study, including the structure and biologic activities (34). BMS833923 was provided by Bristol–Myers Squibb. BMS833923 is a potent synthetic small molecule (EC50 = 50 nmol/L) with specific inhibition on smoothened signaling. BMS833923 was originally patented by Exelixis and is now licensed to Bristol–Myers Squibb (35). Phase I clinical trial of BMS833923 has been completed, and further clinical trials are being planned.

AsPC1 & MIA PaCa2 were purchased from American Type Culture Collection (ATCC), authenticated by STR profiling, and cultured as instructed by the vendor. Pan02 was purchased from ATCC. MMC16 cell line was generated from a metastatic tumor of mouse pancreatic cancer model (36), and cultured in Dulbecco's Modified Eagle's Medium (DMEM) with 10% fetal bovine serum (FBS).

Cells (AsPC-1, MIA PaCa2, Panc02, and MMC16) with stable expression of green fluorescent protein (GFP) and luciferase were harvested in single cell suspension at a concentration of 4 × 10⁶ cells/mL. A total of 2 × 10⁵ cells (in 50 μL of growth medium) were injected into the mouse pancreas using a 27-gauge needle according to a protocol developed in Fidler's laboratory (37). For the human cell lines AsPC1 and MIA PaCa2, we used NOD/scid/IL2Rγ^null mice (NSG). For mouse cell lines Panc02 and MMC16, we used inbred C57Bl/6 mice. Twelve mice were used for each group. Bioluminescent imaging was used to monitor tumor growth. GFP-based whole-body imaging was used to visualize metastases after animal sacrifice. Tumor lesions in pancreas, liver, lung, and lymph nodes were harvested and divided into several portions. Some were used for primary culture; some were snap-frozen in liquid nitrogen for mRNA extraction; some were fixed in 10% buffered formalin and embedded in paraffin for H&E staining and immunohistochemistry. Nu/Nu mice were purchased from Charles River, and NSG mice were provided by the In Vivo Therapeutics (IVT) Core in the IU Simon Cancer Center. Mice were treated with Hh signaling inhibitors [CycT at 25 mg/kg body weight via intraperitoneal injection (34), or BMS833932 (labeled as BMS in the figures, provided by Bristol–Myers Squibb, 30 mg/kg body weight via oral gavage)]. All animal experiments were conducted following protocols approved by the Indiana University School of Medicine Institutional Animal Care and Use Committee.

Cells or tumor spheres were irradiated at a 50 cm distance from the X-ray head with a Precision 320 kVp X-ray machine set at 250 kVp and 12.5 mA, with 2 mm Al filtration with a dose rate of 1.6 Gy per minute as previously reported (38). For the in vivo pancreas irradiation studies, the animals were set up at 50 cm from the X-ray head (confirmed by laser alignment) and irradiated with a 2 × 2 cm lighted collimated field at a dose rate of 1.3 Gy per minute. X-ray dosimetry of the Precision X-ray unit is tested every 6 months with both ionization chamber and TLD measurements by the Department of Radiation Oncology at IU School of Medicine.

Fluorescence-based whole body imaging was carried out using an illumatool LT-9900 Bright Light System (LightTools, Optical Research) according to manufacturer's instruction that has been previously described by others (39). Bioluminescent imaging was conducted using a Berthold Nightowl system (Berthold Technologies). Animals were anesthetized with 2% to 4% isoflurane and administered 150 mg/kg d-luciferin (Caliper Life Sciences) intraperitoneally. Mice were immediately transferred to the heated stage (40 ± 1°C) and imaged at 2-minute intervals for 44 minutes with image integration times ranging from 1 to 5 seconds/image. The animals were imaged weekly for 5 weeks beginning on day 7 after cell implant. Areas of interests were drawn at each time-point using a semiautomated Maximum Entropy algorithm, and the peak intensity was also determined. Data were expressed as the fold-increase in the peak intensity (photons/sec mm²) from week 1 for each mouse, and then averaged for each treatment group.

Pancreatic tumor tissues were cut with 21-blade scalpel into less than 1 mm³ size and placed in 1 mg/mL collagenase IV (Worthington) for 2 hours at 37°C with constant agitation. Single cells were pelleted at 800 g after passing the digested tissues through a 70 μm cell strainer filter (BD Falcon). The cell pellets were subsequently washed twice with sphere growing medium (Neural Basal medium with B-27, 10 μg/mL of EGF, 10 μg/mL of bFGF, and 2.5 unit/mL of heparin) and then plated in 6-well nonadherent culture plates with 2 × 10⁶ cells/mL. Spheres generally appear 3 days after culture, and medium was changed every other day. Secondary spheres were generated by obtaining single cells from tumor spheres through gentle incubation with accutase (Innovative Cell Technology) with constant trituration for 5 minutes, and then the cells were re-seeded into sphere growing medium for an additional 3 to 6 days.

Single cells from primary and metastatic tumors were incubated with fluorescence conjugated antibody against human ESA, CD44, and CD24 (all from eBioscience) for 30 minutes at 4°C. After washing, the cells were stained with 1 μg/mL DAPI to detect dead cells.

The ALDEFLUOR Kit (StemCell Technologies) was used to isolate cells with high aldehyde dehydrogenase (ALDH) enzymatic activity according to the manufacturer's instruction. Briefly, new isolated single cells were suspended in ALDEFLUOR assay buffer containing ALDH substrate (BAAA, 1 μmol/L per 1 × 10⁶ cells) and then incubated for 45 minutes at 37°C. For each sample, an aliquot of cells was stained under identical conditions with 15 mmol/L diethylaminobenzaldehyde (DEAB), a specific ALDH inhibitor, as a negative control. In all experiments, the ALDEFLUOR-stained cells treated with DEAB served as ALDH-negative controls.

The side population was detected by staining cells with Hoechst 33342 dye (Molecular Probes-Invitrogen) using the methods described by Goodell and colleagues (40). Briefly, cells were resuspended at 1 × 10⁶/mL in DMEM with 2% serum and 10 mmol/L HEPES buffer. Hoechst 33342 dye (final concentration 5 μg/mL) was added, and the mixture was incubated at 37°C for 90 minutes with shaking. The cells were then washed with ice-cold HBSS with 2% fetal calf serum (FCS) and 10 mmol/L HEPES, centrifuged down at 4°C, and resuspended in ice-cold HBSS containing 2% FCS and 10 mmol/L HEPES. An aliquot of cells was used as a negative control by adding ABC transporter inhibitor fumitremorgin C (FTC, 10 μmol/L; to block the uptake of the dye) 30 minutes before addition of Hoechst 33342.

Flow cytometry was carried out on a LSR407 device (Beckton Dickinson) and data were analyzed using FlowJo software.

Total RNAs of cells were extracted using Tri-RNA reagent from Sigma according to the manufacturer, and real-time quantitative PCR analyses were conducted according to a previously published procedure (41–43). Triplicate C_T values were analyzed in Microsoft Excel using the comparative C_T(ΔΔC_T) method as described by the manufacturer (Applied Biosystems). The amount of target (2^−ΔΔC_T) was obtained by normalization to an endogenous reference (glyceraldehyde-3-phosphate dehydrogenase: GAPDH) and relative to a calibrator. All primers and probes were purchased from Applied Biosystems Inc. and the sequences are available upon request.

Comparison of the difference between 2 groups was carried out using the chi-square test, Dunnett's test or Student t test. P value less than 0.05 was regarded as statistically significant. A Bliss independence model was employed to evaluate combination effects. The Bliss expectation was calculated with the equation (A + B)−(A x B), in which A and B are the fractional growth inhibitions induced by agents A and B at a given dose, respectively. The difference between the Bliss expectation and the observed growth inhibition induced by the combination of agent A and B at the same dose is the Bliss excess (44).

To monitor tumor growth and metastasis in mouse models, we generated cells with expression of GFP–luciferase fusion protein, which allowed us to monitor tumor growth by bioluminescence imaging and to detect metastases by fluorescence-based whole-body imaging. As shown in Fig. 1, the level of luciferase activity increased steadily in the pancreas area following pancreatic injection of GFP–luciferase expressing AsPC1 cells in NSG mice, indicating rapid tumor growth in this model. Tumor metastasis is not very easily detected using luciferase activity due to sporadic metastases in organs like lymph nodes, peritoneum, diaphragm, mesentery, liver, and lung. Tumor metastases can be detected by GFP expression after dissection (Fig. 2). Using this mouse model, we examined the effect of radiation on tumor growth and metastases of pancreatic cancer.

Mice with pancreatic tumors were subjected to 6 Gy focal radiation once, 3 days after pancreatic injection of cancer cells, and were examined for tumor size by bioluminescence on a weekly basis, and for tumor metastasis by GFP-based imaging at the end of the study. The value of the bioluminescence (indicating primary tumor growth) decreased in magnitude 1 week after irradiation, and was not significantly increased 5 weeks after irradiation (Fig. 1A and B). In contrast, the bioluminescence value in the control group increased over 50-fold during the same period of time, indicating that radiation was effective in reducing primary tumor growth (Fig. 1A and B). The irradiated mice had no signs of body weight change or reduction of survival in the first 3 weeks, suggesting that focal radiation at 6Gy had limited damages in mice. Starting from week 4, mice with or without radiation started to show signs of weakness, loss of weight, and eventually succumbed to pancreatic cancer. We noticed that the number and the size of metastatic nodules were not significantly decreased in the irradiated mice (Fig. 2). This result indicates that despite effective reduction in primary tumor size, focal radiation did not affect tumor metastases of pancreatic cancer.

It is widely accepted that pancreatic cancer is hierarchically organized with a small population of CSCs that possess a capacity to self-renewal (45, 46). CSCs may be responsible for tumor metastasis and radiation resistance. It has been reported that cells with surface expression of CD44, CD24, and ESA or CD133 exhibit features of CSC in pancreatic cancer (46–48). To assess whether metastatic tumors in our orthotopic model are associated with elevated expression of CSC markers, we conducted 2 types of assays. First, we compared primary pancreatic tumors with metastatic tumors for cell surface expression of CD44, ESA, and CD24 by flow cytometry. By RT-PCR, we detected CD24 expression in AsPC1 cells (Supplementary Fig. S1). In the AsPC1-based orthotopic mouse model of pancreatic cancer metastasis, we first gated ESA-positive cells and then examined the percentage of CD24⁺CD44⁺ cells. We found that the percentage of CD24⁺CD44⁺ cells in metastatic tumors was several fold higher than that of pancreatic tumors (Fig. 3A), indicating an increase of CSC population during metastasis.

Next, we examined the transcripts of Hh target genes GLI, PTCH1, and found an increase of GLI and PTCH1 in lymph node (LN) and liver metastatic tumors (Fig. 3B). We also noticed that expression of Hh target genes in primary pancreatic cancer was higher than that of cultured cells, suggesting an effect of tumor microenvironment on Hh signaling. This phenotype was not limited to AsPC1 cells because models derived from Panc02, MIA PaCa2, and MMC16 cells all had elevated Hh signaling in metastatic tumors (Supplementary Fig. S2). The exact reason for the pathway activation was not entirely clear but we found elevated expression of SHH and IHH, major ligands for Hh signaling activation, in metastatic tumors in comparison with the primary tumor (Supplementary Fig. S3). In addition, we detected stromal Hh target gene expression in models derived from AsPC1 and MIA PaCa2 (Supplementary Fig. S4), which was consistent with previous reports (49, 50). Taken together, these results suggest that metastasis of pancreatic cancer is associated with elevated signaling of the Hh pathway and an increase of the CSC population.

Based on our data, we hypothesized that residual cells of the pancreatic cancer surviving the radiation are enriched in CSC-like cell population that can be regulated by Hh signaling. We propose that in the primary pancreatic cancer, the majority of cancer cells are destroyed by radiation, but CSC-like cells continue to grow and metastasize. Because Hh signaling is highly activated in metastatic tumors, we predicted that radiation in combination with Hh signaling inhibition would be more effective in reducing tumor metastasis. To test our hypothesis, we first evaluated the combined effect of radiation and Hh signaling inhibition in CSC-enriched cells. Although cell sorting using specific cell surface markers can yield a more pure CSC population, it is difficult to yield sufficient cell numbers for further experiments. Alternatively, tumor spheres, which are known to be enriched in CSC, can be used for such in vitro functional studies.

After obtaining tumor spheres (see Supplementary Fig. S5 for morphology), we used 2 approaches to test whether the CSC population was indeed enriched: expression of cell surface markers ESA, CD24, and CD44, and the percentage of ALDH⁺ cell population. By comparing tumor sphere cells with the parental cells, we found about 1% CD24⁺CD44⁺ESA⁺ cells in parental cells but over 8% triple-positive cells in tumor spheres (Fig. 4A). At the transcript level, CD24 was detectable in parental AsPC1 cells and was increased in the tumor spheres (Supplementary Fig. S1).

In addition, CSC population often has an increased level of ALDH activity (33). By comparing AsPC1 cells growing in monolayer with those in tumor spheres, we noticed that ALDH-positive cells were higher in tumor sphere (1.2%) compared with the parental cells (0.159%; Fig. 4B). These data indicate that tumor spheres indeed have elevated expression of CSC markers.

At the same time, we found increased expression of GLI/PTCH1 and other reported Hh target genes [reviewed in (51)] in tumor spheres in comparison with the parental cells (Fig. 4C), suggesting that Hh signaling may be important for tumor sphere maintenance and regeneration. Most notably, we found elevated expression of proliferation-related genes MYCN, CCND2, stem cell signaling molecules JAG2, FST, PDGFRα, WNT2, as well as genes involved in survival (BCL2) and in epithelial-to-mesenchymal transition (SNAI1).

Next, we tested whether combination of Hh signaling inhibition with radiation would be more effective in reducing tumor sphere formation efficiency. For the parental AsPC1 cells, irradiation at 2 Gy was sufficient to reduce cell number by half (Fig. 5A), suggesting that AsPC1 cells are quite sensitive to radiation. Formation of secondary tumor spheres is commonly used to test cell stemness in CSC functional studies (47). Because tumor spheres have more CSCs and express a high level of Hh target genes, we examined whether regeneration of tumor spheres is affected by Hh signaling inhibition and radiation. We first treated single cells from tumor spheres with radiation at different dosages (2, 4, 6, 8 Gy) or left untreated, and examined the efficiency of secondary tumor sphere formation. Effects of irradiation on tumor spheres were assessed by the number of spheres formed from 1,000 cells. As shown in Fig. 5B, we found that 2 Gy only reduced the number of secondary tumor spheres by 30%. We estimated the EC50 dose of radiation for inhibiting tumor sphere formation was over 5 Gy.

The effect of Hh signaling inhibition on tumor sphere formation was tested by addition of Hh signaling inhibitor CycT in the tumor sphere formation medium. We found that Hh signaling inhibition by CycT at 5 μmol/L reduced tumor sphere formation by 50% (Supplementary Fig. S6). We therefore determined whether combination of the 2 will be more effective. After irradiation of tumor sphere cells in the presence or absence of Hh inhibitor CycT, we examined the tumor sphere-forming efficiency. As shown in Fig. 5C, tumor sphere formation was reduced by 30% to 50% after 4 Gy, by 50% after Hh inhibition, but was completely eliminated when radiation was combined with Hh inhibition. By Bliss independence analysis (see Methods), we found that combining focal irradiation with Hh signaling inhibition had more than additive effects (the response fraction for irradiation was 0.08; the response fraction for Hh signaling inhibition was 0.42, whereas the combined treatment had a response fraction of 0.62; 0.62 > (0.42 + 0.08) − 0.42 × 0.080; Fig. 5C). We also found that radiation of tumor spheres further increased the level of Hh target genes (GLI and PTCH1), which were inhibited by BMS833923 (Fig. 5D). Consistent with Hh signaling changes, we discovered that the level of CSC cell surface markers was induced by radiation, but reduced to the basal level after radiation plus Hh inhibition (Fig. 5E). Taken together, our results indicate that the combination of radiation with Hh signaling inhibition will be more effective in treatment of pancreatic cancer metastasis in our mouse model.

To further support our data derived from AsPC1 cells, we used another cell line, MMC16, for tumor sphere studies. MMC16 cells were derived from a mouse model of pancreatic cancer (36), with a high efficiency in tumor sphere formation. For this mouse cell line, we used side population as a marker for CSC population because the mouse cell surface markers are not widely used. The side population is the cells with high expression of ABC transporter protein family members, rendering these cells less effective in uptake of fluorescent dyes (52). Like normal tissue stem cells, CSCs are also reported to possess this property (24). It has been reported that the side population of pancreatic cancer cells has some features of CSCs, such as high invasive and metastatic potential, and susceptibility to EMT (24). After staining with the fluorescent dye Hoechst 33342, we can detect side population (SP) by flow cytometry. For MMC16 cells, the side population was less than 2%, whereas radiation increased the percentage of the population to greater than 6% (Supplementary Fig. S7A). This increase was specific because addition of the ABC transporter inhibitor fumitremorgin C reduced the side population (data not shown). Consistent with an increase of the side population, we observed elevated Hh target gene expression in tumor spheres after radiation (Supplementary Fig. S7B). Further analysis indicate that while radiation and Hh signaling inhibition both reduced tumor sphere formation efficiency, combination of Hh signaling inhibition with radiation at 8 Gy resulted in no visible tumor spheres (Supplementary Fig. S7C). Similar to AsPC1-derived tumor spheres, Bliss independence analysis showed that the combination had greater than additive effects when irradiation was over 2Gy (Supplementary Fig. S7C).

Once placed in regular culture medium, however, tumor sphere cells began to lose expression of the cell surface markers as well as Hh target genes (not shown), suggesting a unique microenvironment is required to keep the stemness of CSCs. Consistent with the high expression of CSC markers, single cells from tumor spheres were also more efficient in tumor sphere formation (20–30 spheres per 1,000 cells from tumor sphere cells vs. 2–3 spheres per 1,000 cells from parental AsPC1 cells). These results suggest that the CSC-enriched tumor spheres require a tumor environment for CSC to be functional.

Our results from tumor sphere studies indicate that Hh target gene expression is associated with CSC markers in 2 independent cell lines, and radiation plus a Hh signaling inhibitor have more than additive effects in tumor sphere formation. Because tumor metastasis is associated with high Hh target gene expression and increased expression of CSC markers, our data support that combining Hh signaling inhibition with radiation can be more effective in suppressing tumor metastasis in pancreatic cancer.

To test our hypothesis derived from tumor sphere studies, mice with orthotopic pancreatic tumors were treated with radiation (6 Gy) with or without Hh inhibitor BMS833932 (as Hh-I in the figure), or left untreated. The mice were imaged weekly, and were sacrificed 5 weeks after pancreatic injection. Hh signaling inhibition by BMS833932 was shown to reduce expression of Hh target genes (Supplementary Fig. S8). Typical images from each group were shown in Fig. 6A. We found that the combined treatment group gave the best response. Tumor growth was monitored by bioluminescence imaging. Hh signaling inhibition had no significant effect on primary tumor growth, whereas tumors from mice with radiation did not grow (Fig. 6B). Despite lack of any significant effects of Hh inhibitor BMS833932 for primary tumor growth, the number of metastatic tumor nodules in lymph nodes was significantly reduced (Fig. 6C). We observed that combining Hh inhibition with focal irradiation gave a greater than additive effect in reducing lymph node metastases based on Bliss independence analysis. Combination of IR and Hh inhibition was also more effective than either single treatment alone on suppressing lung metastases (Supplementary Fig. S9). Thus, our data have shown more therapeutic efficacy when irradiation is combined with Hh inhibition.

To investigate the molecular mechanisms by which Hh signaling regulates CSC and tumor metastases, we examined putative Hh target genes in our mouse models. As shown in Supplementary Fig. S10, we found elevated expression of PDGFRα, WNT2, OPG, and IGFBP4 (Supplementary Fig. S10A), and BMS833923 inhibited expression of PDGFRα,WNT2, and OPG (Supplementary Fig. S10B). As described earlier, PDGFRα is a known Hh target gene (53–56), and may be responsible for Hh signaling-mediated cell proliferation and tumor metastasis. Similarly, WNT2 is reported to be a specific marker for circulating pancreatic cancer cells in the patients with pancreatic cancer and metastatic disease (57), and was shown to be regulated by Hh signaling in another study (50). OPG is known to be involved in cancer metastasis, and is regulated by Indian hedgehog. We believe that PDGFRα and WNT2 are good candidates for the underlying molecular mechanisms by which Hh signaling mediates cell proliferation and maintenance of CSC population.

During tumor development of pancreatic cancer, increasing evidence suggests 2 major changes: a cell-autonomous noncanonical Gli regulation and Hh ligand-dependent paracrine Hh signaling (12, 49, 50, 58–61). It has been shown that Gli1 or Gli2 activation cooperates with KRAS to promote pancreatic cancer formation (12, 61). However, Hh signaling during pancreatic cancer metastasis is less clear, although it is generally believed that similar signaling events also occur. Variable rates of metastasis in different GEM models and the dependence of the phenotypes on the mouse genetic background make the study of metastasis more challenging. Using AsPC1 and MIA PaCa2 cells in the orthotopic mouse models of pancreatic cancer metastasis, we were able to specifically examine Hh signaling in the tumor compartment (human Hh target genes) and the stroma (mouse Hh target genes). Our data revealed that Hh signaling is activated in the stroma and in the tumor compartment. Our data are consistent with previous studies in E3LZ10.7 cell-derived orthotopic model (33). We further indicated that SHH and IHH were highly expressed in metastatic tumors (Supplementary Fig. S3). We believe that ligand-mediated signaling is at least partially responsible for Hh signaling activation because SMO signaling inhibitors (CycT and BMS833923) did reduce the levels of human and mouse Hh target genes. Similar results were also observed in isolated tumor spheres where only human Hh target genes are detectable (Fig. 5D). The high association between human Hh target gene expression and CSC markers in tumor spheres and in mouse models indicates that tumor-specific Hh signaling may be important for maintenance of CSC population during pancreatic cancer metastasis.

The CSC theory has explained several phenotypes of human cancer, including high metastatic potential, tumor recurrence, and therapeutic resistance. Accumulating evidence indicates the existence of CSCs in pancreatic cancer (45). Using cell surface markers, Li and colleagues have shown that a subset of pancreatic cancer cells with CD24⁺CD44⁺ESA⁺ or CD44⁺cMET^high is more potent in generating tumors (29), and inhibition of cMET reduces tumor metastasis (47). Furthermore, pancreatic cancer cells with high ALDH activity more readily form tumors in immune-deficient mice (33). In addition, the side population of pancreatic cancer cells is more metastatic when implanted in immune-deficient mice (62). Our studies showed that tumor spheres are enriched in CD24⁺CD44⁺ESA⁺ cells and ALDH+ cells (Fig. 4). Functional analyses using tumor sphere regeneration indicate that inhibition of Hh signaling can sensitize tumor spheres to radiation (Fig. 5). The data were further confirmed in a mouse orthotopic model of pancreatic cancer metastasis, in which Hh signaling inhibition significantly reduced tumor metastasis (Fig. 6). Similar studies from other groups also indicate a role of Hh signaling in pancreatic CSCs (15, 27, 33). Understanding the molecular mechanisms by which Hh signaling regulates CSCs require additional studies. For example, we do not know whether CSCs with genetic depletion of Hh signaling can still form tumor metastases. We know very little of the interaction between CSC and the tumor microenvironment. It is known that Hh signaling activation is a major factor for the development of desmoplastic stroma (63), which provides a physical barrier for drug delivery to the tumor.

In consistent with the role of Hh signaling for CSC, we detected elevated expression of several stem cell signaling molecules in tumor spheres, including JAG2, FST, WNT2, and PDGFRα. The CSC-enriched tumor spheres also have elevated expression of proliferation-related genes such as MYCN and CCND2. In addition, survival gene BCL2 and EMT regulatory gene SNAI1 were also upregulated. Further understanding of the roles of Hh signaling for maintenance of CSCs and for the tumor microenvironment may help generating new therapeutic options for pancreatic cancer.

Surgery, radiation, and chemotherapy have been the mainstays of treatment for human malignancies for several decades. Technical advances in irradiation have made the treatment very focused on cancer tissues with limited radiation exposure to normal adjacent tissues. The use of a combination of radiation and chemotherapy, chemoradiation, has been proven to be effective in many tumor types. In treatment of pancreatic cancer, combination of gemcitabine with irradiation is effective only in a subset of patients (64). Thus, there exists the need to develop better combinations of radiation with chemotherapy or targeted therapy.

Our data provide evidence to support that the combination of radiation with Hh signaling inhibitor results in greater than additive inhibition of tumor metastasis in mouse orthotopic model of pancreatic cancer. We found elevated expression of CSC markers following irradiation as well as increased Hh signaling. We showed that Hh signaling regulates expression of WNT2 and PDGFRα (Supplementary Fig. S10), molecules known to regulate cell proliferation and cell stemness (53, 54, 57). This CSC-enriched tumor spheres are also more resistant to radiation, and contain activated Hh signaling. They became sensitive to radiation after Hh signaling inhibition (Fig. 5C and Supplementary Fig. S7C). The relevance of our data in vitro to pancreatic cancer treatment was also shown in our animal model (Fig. 6). Since Hh signaling activation in tumor metastases has been observed in the 2 models tested, we believe that this mechanism is common for most pancreatic cancer.

Several Hh signaling inhibitors are in clinical trial, and all are targeted at smoothened. Off-target effects have been reported when the inhibitors are overdosed (10 mmol/L for cyclopamine; ref. 65). To avoid off-target effects of CycT and BMS833923, we first tested their biologic effects in Hh-dependent tumor models (34) to obtain the EC50. In tumor sphere assays, BMS833923 (EC50 = 50 nmol/L) is more potent than CycT (EC50 = 5 μmol/L; Supplementary Fig. S6). Furthermore, we correlate the change of metastatic tumor nodules with Hh target gene expression.

Taken together, our studies suggest that combining focal irradiation with Hh signaling inhibition may be more effective in future pancreatic cancer treatment. Focal irradiation reduces the size of pancreatic cancer but has little effects on tumor metastases, which are associated with elevated expression of CSC cell surface markers and Hh target gene expression. Combination of irradiation with Hh signaling inhibition gives more additive effects than single treatment in the tumor sphere assay and in an orthotopic mouse model for pancreatic cancer metastasis.

No potential conflicts of interest were disclosed.

Conception and design: J. Xie

Development of methodology: D. Gu, H. Liu, H. Hanenberg, M.S. Mendonca

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Gu, H. Liu, X. Zhang, H. Chin-Sinex, M.S. Mendonca, J. Xie

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Gu, H. Liu, M.S. Mendonca, H.E. Shannon, E. Gabriela Chiorean, J. Xie

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Gu, G.H. Su, E. Gabriela Chiorean

Writing, review, and/or revision of the manuscript: G.H. Su, M.S. Mendonca, H.E. Shannon, E. Gabriela Chiorean, J. Xie

Study supervision: J. Xie

The authors would like to thank Anthony L. Sinn, Brian P. McCarthy, Drs. Karen Pollock, and Paul R. Territo for technical support as well as the core facilities (In Vivo Therapeutics and In Vivo Biomedical Imaging) for services.

This work was supported by grants from National Cancer Institute (R01CA94160, R01CA155086) to J. Xie, from The Wells Center for Pediatric Research to J. Xie, from IU Simon Cancer Center and The Signature Center for Pancreatic Cancer in Indiana University to J. Xie.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.