Inhibition of the Hedgehog Pathway Targets the Tumor-Associated Stroma in Pancreatic Cancer

This article is featured in Highlights of This Issue, p. 1133

The Hedgehog (Hh) pathway is an essential developmental pathway involved in regulating key aspects of embryogenesis, stem cell maintenance, and tumor biology (1). In humans, the Hh ligands Sonic, Indian, and Desert Hedgehog (SHH, IHH, and DHH) bind to the Patched1 protein (PTCH1) on target cells, which results in the release of inhibition of the SMO receptor. SMO is a 7-transmembrane G-protein–coupled receptor-like protein that, when activated, results in activation of GLI transcription factors and expression of downstream targets GLI1, PTCH, BCL2, myc, and IGF2.

Aberrantly activated Hh has recently been identified in several malignancies, including basal cell carcinomas (2–4), medulloblastomas (5, 6), lung cancer (7–9), prostate cancer (10–12), and gastrointestinal malignancies (10, 13). In pancreatic adenocarcinoma (PDAC), the Hh pathway is considered one of the “core” signaling pathways that is altered, and the majority of these cancers show abnormal expression of SHH, PTCH1, and SMO (14). Moreover, these factors are expressed early in preneoplastic pancreatic intraepithelial neoplasia (PanIN) lesions, whereas they are absent in normal pancreatic tissue, suggesting that these factors are important for early tumorigenesis. Blockade of Hh with SMO inhibitors such as cyclopamine inhibited invasion and metastasis and prolonged survival in mouse models of pancreatic cancer (15–17).

Recently, intriguing evidence has emerged suggesting that the Hh pathway is highly active in a paracrine signaling manner in the tumor microenvironment of some pancreatic tumors. When SMO was genetically ablated in the pancreatic epithelium of PDAC-susceptible mice, development of PDAC tumors was not affected, suggesting that Hh signaling in PDAC does not occur in an autocrine manner (18). Using human–tumor xenograft models, Yauch and colleagues (19) used species-specific expression profiling to show that Hh pathway antagonist treatment resulted in downregulation of Hh target genes only in the murine stromal microenvironment but not within the human tumor epithelial compartment. Similarly, expression of SMO in mesenchymal cells, but not epithelial cells, in the pancreas led to Hh pathway activation, further supporting a paracrine model of Hh-mediated tumorigenesis (20). Finally, treatment of a genetically engineered mouse model of PDAC with the Hh inhibitor IPI-926 resulted in depletion of desmoplastic stroma in pancreatic tumors (21). These observations are consistent with a model in which tumor cells produce Hh ligands that trigger signaling in the stromal microenvironment in a paracrine manner.

Despite these initial observations, the precise role of stromal cells in Hh signaling in pancreatic cancer is not well understood. We have previously shown that the cancer-associated fibroblasts in PDAC (human pancreatic stellate cells, HPSCs) produce secreted factors that promote tumor progression and metastasis in vitro and in vivo (22). In this study, we analyzed the role of HPSCs from the tumor-associated stroma in Hh signaling. In addition, we evaluated the efficacy of a novel SMO inhibitor (AZD8542) on pancreatic tumor progression with an emphasis on the role of the HPSCs from the stroma. We present data that strongly suggests the primary mechanism of action of Hh signaling in PDAC occurs in a paracrine manner with ligand expression by the cancer cells and activation of SMO on neighboring HPSCs in the stromal microenvironment.

NIH-3T3, human embryonic palatal mesenchyme (HEPM), C3H10T1/2, HeLa, and human colon cancer Colo205 cells as well as BxPC3, Panc1, SU86.86, MiaPaca2, and Capan2 pancreatic cancer cell lines were obtained from American Type Culture Collection (ATCC, Manassas, VA). Human pancreatic cancer MPanc96 and human pancreatic ductal epithelial (HPDE) cells were obtained from Dr Timothy J. Eberlein (Washington University, St. Louis, MO, USA) and Dr M. Tsao (Ontario Cancer Institute, Toronto, Ontario, Canada), respectively. L3.6pl cells were obtained from Dr I. Fidler (23) and immortalized HPSCs were isolated as previously described (22). Primary HPSCs were established and cultured as previously described (22). Both immortalized (using hTERT and SV40T) and nonimmortalized primary cells were used in these studies. NIH-3T3, HeLa, pancreatic carcinoma cells, and HPSCs were cultured in Dulbecco's minimal essential medium (DMEM) with 10% FBS (Invitrogen) and 1% l-glutamine. HEPM and C3H10T1/2 cells were cultured in Eagle's minimal essential medium (EMEM) with Earle's BSS + 2 mmol/L l-glutamine + 1.0 mmol/L nonessential amino acids + 1.5 g/L sodium bicarbonate + 10% FBS. HPDE cells were cultured in keratinocyte serum-free media containing 50 μg/mL bovine pituitary extract and 0.2 ng/mL recombinant epidermal growth factor (all from Invitrogen). Colo205 was maintained as an adherent culture in DMEM containing 10% FBS at 37°C in a humidified atmosphere of 6% CO₂. All other cells were cultured at 37°C in a humidified atmosphere of 5% CO₂.

A subset of the proprietary AstraZeneca compound collection (40,000 compounds) with similarity to cyclopamine was screened using a GLI1 luciferase reporter assay to identify inhibitors of the Hh pathway (24).The GLI1 luciferase construct consists of 8 Gli-binding sites upstream of a luciferase reporter gene (25). The construct was transfected into NIH-3T3 cells along with a constitutively active Renilla luciferase construct as a control. Stable cells were selected and stimulated with 50% SHH-containing conditioned medium. Conditioned media was generated by transfecting HEK293 cells with an SHH expression vector and collecting media 48 to 96 hours after transfection. Cells were treated with Hh inhibitor compounds (in dimethyl sulfoxide) at varying concentrations for 24 hours and assayed for luciferase activity using the Dual-Glo Luciferase Assay System (Promega) per the manufacturer's instructions. HEPM cells and C3H10T1/2 cells were treated with inhibitor compounds for 24 hours and assayed for luciferase reporter activity using the Steady-Glo Luciferase Assay System (Promega) per the manufacturer's instructions. Plates were read on the Tecan Ultra microplate reader (Tecan, Mannedorf, Switzerland) at 50 ms integration time per well.

To evaluate the efficacy of SMO inhibitor compounds, an osteoblast differentiation assay was done using C3H10T1/2 cells, which differentiate into osteoblasts after stimulation with SHH or Wnt. In brief, C3H10T1/2 cells were plated into 384-well plates, and media was changed to low serum (2% FBS) or conditioned media containing either SHH or WNT3a. Cells were then treated with inhibitor compounds for 72 hours, and alkaline phosphatase activity was measured. Cells were lysed in 15 μL of 1× RIPA cell lysis buffer, incubated at −80°C for 30 minutes, treated with p-nitrophenyl phosphate at 1 mg/mL in diethanolamine buffer (pH 9.8), incubated at 30°C overnight for color development, and read at absorbance of 405 nm.

HeLa cells were transfected with pcDNA vector expressing human or mouse myc-tagged SMO (GeneCopoeia; FuGENE transfection reagent, Promega).

After 24 hours incubation, media was changed to low serum (0.5% FBS). The cells were then pretreated with various concentrations of SMO inhibitors for 20 minutes, followed by 3 nmol/L BODIPY-labeled cyclopamine (Toronto Research Chemicals) and incubated at 37°C for 4 hours. After incubation, the cells were fixed with 3% paraformaldehyde and 0.5% Triton X + 1× PBS (Invitrogen). The cells were then washed, blocked with 10% goat serum (Fisher Scientific), and incubated with myc-tagged antibody (Cell Signaling Technology) for 2 hours followed by Alexa Fluor 594 anti-rabbit antibody (Invitrogen). Cells were stained with Hoechst dye (Invitrogen), and fluorescence was detected using an ImageXpress system (Molecular Devices).

HPSCs were seeded at 2,000 cells per well in triplicate in 96-well plates and cultured in DMEM containing 10% FBS. After overnight attachment, media was changed to DMEM containing 1% FBS, and varying concentrations (0, 1.0, 1.5, and 2 μg/mL) of recombinant SHH (rSHH; R&D Systems) were added to the wells. Cell proliferation was analyzed at 72 hours using MTS reagent (Promega) added 1 hour before taking a spectrophotometric reading according to the manufacturer's instructions.

HPSCs were seeded in 6-well plates and grown to 70% confluence in DMEM containing 10% FBS. The media was then changed to serum-free DMEM overnight. rSHH or rIHH (R&D Systems) was added to the wells (2 μg/mL) along with varying concentrations (0, 10, 100, and 1000 nmol/L) of the SMO inhibitor AZD8542 (AstraZeneca). Cells grown in 1% DMEM without rSHH, rIHH, or AZD8542 served as controls. RNA was isolated after 24 hours and expression of GLI1 was measured by quantitative reverse transcriptase (RT)-PCR (qRT-PCR) and normalized to HPRT.

Animal protocols for the colon cancer models were approved by the AstraZeneca R&D Boston site's Institutional Animal Care and Use Committee. All animal work was done in accordance with applicable internal standards and external local and national guidelines, regulations, and legislation. Female Ncr nude mice aged 6 to 8 weeks were maintained under specific-pathogen-free conditions in a facility accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International.

Colo205 cells were implanted subcutaneously in the right flank (4 × 10⁶ cells per mouse) in 0.1 mL of serum-free media. Tumors were allowed to grow until they reached an average volume of 200 mm³, and mice were randomized (N = 5 per treatment group.) AZD8542 was suspended in 0.5% (v/v) hydroxypropyl methyl cellulose (HPMC) in sterile water and administered orally once daily (20 or 40 mg/kg). Tumor and blood samples were collected in RNAlater (Ambion) and EDTA, respectively, at 1, 4, 6, 8, 12, and 16 hours after dosing. In a coimplantation xenograft model of colon cancer, mice were injected with both HT29 (0.3 × 10⁶) and MEF (1.5 × 10⁶) cells subcutaneously in the right flank (100 μL per injection) for a final tumor:stroma (T/S) ratio of 1:5. When tumors reached 70 to 100 mm³, animals were randomized to receive either SMO inhibitor (20–80 mg/kg) or vehicle (0.5% HPMC/0.1% Tween 80; N = 10 per group). Tumor measurements and body weights were recorded twice weekly. Data are expressed as percentage tumor growth inhibition at day 29. For pharmacodynamic studies, tumors were collected 8 hours after each dose in RNAlater (Qiagen) or formalin (Newcomer Supply). Blood was collected for pharmacokinetic analysis.

Tumors were lysed using a Lysing Matrix D tube (MP Biomedicals) with RLT buffer containing 1% b-mercarptoethanol and homogenized with the Fast-Prep-24 Instrument (MP Biomedicals). RNA was isolated using the RNeasy Midi Kit Column (Qiagen) and converted to cDNA using the High-Capacity cDNA Archive Kit (Applied Biosystems). Quantitative RT-PCR was done using primers for mouse GLI1 (Applied Biosystems) on the 7900HT TaqMan real-time PCR instrument (Applied Biosystems). Data were normalized to housekeeping gene HPRT.

All pancreatic cancer animal experiments were reviewed and approved by the M.D. Anderson Institutional Animal Care and Use Committee. An orthotopic nude mouse model of pancreatic cancer using BxPC3 pancreatic tumor cells labeled with firefly luciferase (BxPC3-FL) has previously been described by this lab (22). All mice were divided into groups with varying tumor:stroma (T/S) ratios with intrapancreatic injections of: (a) 1 × 10⁶ BxPC3-FL cells alone (T/S 1:0), (b) 1 × 10⁶ BxPC3-FL cells and 1 × 10⁶ HPSCs (T/S 1:1), or (c) 1 × 10⁶ BxPC3-FL cells and 3 × 10⁶ HPSCs (T/S 1:3) suspended in 50 μL of HBSS (Sigma-Aldrich). Groups were further subdivided into those treated with AZD8542 and those receiving vehicle only. Treatment with 80 mg/kg AZD8542 suspended in 200 μL of vehicle—0.5% (hydroxypropyl) methyl cellulose (Sigma-Aldrich) with 0.1% Tween 80 (Fisher Scientific)—or with vehicle alone was administered daily by oral gavage starting on day 4 after tumor implantation. Bioluminescence imaging was done twice weekly to assess the luciferase signal from BxPC3 cells using the IVIS imaging system (Caliper Life Sciences; ref. 22). Mice were euthanized, and tumors were harvested, measured, and either snap-frozen or fixed in formalin for further processing. The same experiment was done using MPanc96-FLcancer cells.

Briefly, for Ki67 staining, formalin-fixed paraffin embedded slides were deparaffinized in xylene, rehydrated, and blocked with 3% H₂O₂. Antigen retrieval was done by steaming in antigen unmasking solution (Vector Lab). Slides were blocked with 4% fish gelatin and incubated overnight with rabbit anti-Ki67 antibody (1:300; Fisher) at 4°C and washed with PBS. After incubation with biotinylated goat anti-rabbit IgG (Vector Lab) for 1 hour at room temperature, slides were washed and incubated for 30 minutes with VECTASTAIN® ABC Reagent (Vector Lab). A positive reaction was detected by exposure to stable 3,3′-diaminobenzidine (Trevigen) for 10 to 20 minutes. Slides were counterstained with Gill's hematoxylin. The Aperio ScanScope CS Slide Scanner (Aperio Technologies) was used to digitally scan the slides with a 20× objective and images were analyzed using Aperio Imagescope software for nuclear staining with Ki67.

For CD31 staining, frozen slides were fixed in cold acetone and endogenous peroxidase was blocked with 3% H₂O₂. Slides were blocked with 5% normal horse serum + 1% normal goat serum in PBS, incubated with rat antimouse CD31 antibody (1:50; BD PharMingen) at room temperature for 2 hours and washed with PBS. After incubation with peroxidase-AffiniPure goat anti-rat IgG (1:200; Jackson) for 1 hour at room temperature, a positive reaction was detected by exposure to stable 3,3′-diaminobenzidine (Trevigen) for 5 minutes. Nuclei were counterstained with Gill's hematoxylin.

RNA was isolated from homogenized snap-frozen tumor tissue and cultured cell lines using TRIzol (Invitrogen) according to the manufacturer's instructions. cDNA was made from RNA using the QuantiTect Reverse Transcription System kit (Qiagen) according to the manufacturer's instructions. qPCR for SMO, SHH, PTCH1, GLI1, and HPRT (Applied Biosystems) was done using TaqMan gene expression assays (Applied Biosystems) according to the manufacturer's specifications. Quantification was accomplished using the standard curve method. qPCR was also done for GLI1 in HPSCs labeled with green fluorescent protein (HPSC-GFP) with or without coculture with BxPC3 cells for 96 hours.

Cell lysates were prepared from HPSC and pancreatic cancer cells and protein concentrations were measured by BioRad reagent. Protein (40 μg) was loaded onto 10% SDS-PAGE gels and Western blotting was done using a rabbit primary antibody against SHH or IHH (Cell Signaling) at a 1:1000 dilution and goat anti-rabbit secondary antibody (LiCor) at a 1:10,000 dilution. Blots were reprobed for β-Actin (1:1000 dilution; Abcam), which served as loading control.

Experiments were done in triplicate and representative data are shown. Statistical analysis was done using GraphPad Prism software (GraphPad). Comparisons were made using the 2-tailed Student t test, and significant difference was defined as P < 0.05. Data are shown as mean ± SEM.

Screening of the AstraZeneca compound library using a GLI1 luciferase reporter assay identified several compounds with an IC₅₀ of <10 nmol/L (Supplementary Fig. S1A) and no activity against the control Renilla vector. To show specificity of the compounds for the Hh pathway, we utilized an osteoblast differentiation assay stimulated by 2 different ligands known to act at different points in the differentiation process. After stimulation with conditioned media containing either SHH or Wnt3A, mouse C3H10T1/2 cells were analyzed for secreted alkaline phosphatase as an indicator of osteoblast differentiation. Several compounds were identified to have an IC₅₀ of <20 nmol/L with SHH stimulation (Supplementary Fig. S1), but none were shown to inhibit osteoblast differentiation when stimulated with Wnt3a (data not shown), indicating the specificity of the compounds for the Hh pathway. None of the compounds identified were able to inhibit proliferation in a panel of 60 tumor cell lines, consistent with previous reports (19, 26). The most promising clinical candidate was AZD8542 (Fig. 1A).

To determine whether the inhibitors acted specifically on SMO, we used a cyclopamine displacement assay. HeLa cells engineered to express either human or mouse SMO were pretreated with SMO inhibitor compounds, followed by cyclopamine labeled with BODIPY. To validate the assay, unlabeled KAAD-cyclopamine was used as a competitor for BODIPY-labeled cyclopamine; it showed a similar affinity for both mouse and human SMO (2 nM vs. 0.8 nM, data not shown). Fluorescent images of cells treated with AZD8542 are shown in Fig. 1B. In this assay, AZD8542 shows a K_d of 5 nmol/L for mouse SMO and a K_d of 20 nmol/L for human SMO, indicating that the compound has a similar range of affinity for both mouse and human SMO (Fig. 1C).

Furthermore, we tested the ability of AZD8542 to not only bind SMO but also block downstream GLI1 signaling in both mouse and human cells. C3H10T1/2 (mouse) and HEPM (human) cells expressing GLI1-responsive luciferase reporter genes were stimulated with SHH conditioned media and treated with AZD8542. AZD8542 was able to effectively inhibit GLI1 activity with an IC₅₀ of 2.9 nmol/L in the mouse cell line and an IC₅₀ of 0.02 nmol/L in the human cell line, indicating increased potency against the human reporter line (Supplementary Fig. S1B). These data establish that AZD8542 is an ideal candidate to evaluate effects on the Hh pathway in models containing both human and mouse components.

A panel of cells including pancreatic cancer, normal pancreatic duct, and HPSCs from PDAC were analyzed for expression of genes involved in the Hh pathway (Fig. 2). SMO was expressed at high levels in HPSCs and at lower levels in all other cancer cells tested. The expression level in a few cancer cells (SU86.86, MiaPaca2, and Capan2) was moderate (Fig. 2A). In contrast, the SHH ligand was absent in HPSCs but was expressed in several cancer cell lines (Panc1, MPanc96, Capan2; Fig. 2B and Supplementary Fig. S2A). Similarly, IHH was expressed in cancer cells BxPC3 and MPanc96 but not HPSCs (Supplementary Fig. S2A). The downstream targets PTCH1 and GLI1 were expressed in both cancer and stellate cells to varying degrees (Fig. 2C and D). To confirm the generalizability of these results to other HPSC cell lines, we analyzed the expression of SMO, SHH, PTCH1, and GLI1 in primary HPSCs (nonimmortalized) obtained from 3 unique patients and found that expression levels of these genes were similar to those of the HPSC cell line used in these studies (data not shown).

Others have shown that activation of the Hh pathway by rSHH does not affect endogenous GLI1 messenger levels in tumor epithelial cells (Bxpc3 and CFPAC-1), regardless of SMO expression in these cells (19). However, it has been suggested that Hh signaling acts in a paracrine fashion because SMO inhibition in the mouse stroma of xenograft tumor models is required for growth inhibition (19). We determined the effects of Hh pathway activation in HPSCs by evaluating cell proliferation, migration, and GLI1 expression. Treatment of HPSCs with rSHH (1.5 and 2.0 μg/mL) resulted in increased cell proliferation with a peak effect of 173.1% of control (P < 0.005; Fig. 3A). Accordingly, downstream GLI1 expression also increased with rSHH stimulation (5.7-fold vs. control, P < 0.05; Fig. 3B). Inhibition of SMO by AZD8542 effectively abrogated the rSHH-mediated induction of GLI1 at 100 to 1000 nmol/L (0.25- to 0.5-fold vs. control, P < 0.05). Similar results were obtained in a human prostate stromal cell line (Supplementary Fig. S1B). Although rSHH stimulated HPSC activity, it had no effect on pancreatic cancer cell (Bxpc3 and Panc1) proliferation or migration (data not shown). HPSCs also responded to IHH treatment with increased cell proliferation and increased levels of GLI1 mRNA (Supplementary Fig. S2).

Although AZD8542 showed potent inhibition of Hh in vitro, we wanted to test the ability of the compound to affect the Hh pathway in vivo. Initially, we used a Colo205 xenograft model, which expresses SHH ligand in the tumor cells but no IHH ligand (Supplementary Fig. S4A). Using species-specific primers, we observed strong inhibition of GLI1 only in the mouse stromal compartment but not the human epithelial compartment (Supplementary Fig. S4B). GLI1 inhibition was time dependent, with maximum inhibition at 8 hours after dosing and recovery by 16 hours. Plasma levels of AZD8542 were significant (60 μg/mL) at 1 hour after dosing, with depletion by 8 hours (Supplementary Fig. S4C). Interestingly, this time frame corresponds to the time required for maximum GLI1 inhibition, indicating a 6-hour time delay between SMO inhibition and the effect on GLI1 transcription. These data show that AZD8542 is orally available and able to inhibit the Hh pathway in the stromal component of the Colo205 xenograft model.

Although AZD8542 was able to inhibit Hh signaling in the Colo205 model, we observed no effect on tumor growth. Thus, we used a coimplant model with a different colon cancer cell line, HT29, known to express SHH combined with mouse embryonic fibroblasts (MEFs)—cells known to be highly responsive to the Hh pathway (19). Animals treated with 80 mg/kg AZD8542 showed significant tumor growth inhibition of 79% compared to vehicle controls (P < 0.001; Supplementary Fig. S5A). The effect on tumor growth was dose dependent, with 52% and 46% tumor inhibition at 40 and 20 mg/kg, respectively (P < 0.001 and 0.005; Supplementary Fig. S5A). The compound was well tolerated, with no significant changes in body weight (not shown). Accordingly, AZD8542 treatment resulted in a dose-dependent decrease in mouse GLI1 and PTCH1 expression in tumors, with nearly complete abrogation of expression at the highest dose (Supplementary Figs. 4B and 5B). Plasma levels of AZD8542 increased with each higher dose (Supplementary Fig. 5C), peaking at 30 μg/mL ± 12.6 with the most effective dose for tumor inhibition (80 mg/kg). To further investigate the effect of AZD8542 on the stromal compartment, we analyzed tumors for alpha smooth muscle actin (αSMA) expression. Tumors treated with AZD8542 expressed markedly less αSMA than did the control group (Supplementary Fig. 5D), suggesting that AZD8542 acts primarily on the tumor-associated stroma rather than the epithelial compartment in this model.

Having established that AZD8542 is able to inhibit tumor growth in the stromal-dependent HT29/MEF model, we wanted to evaluate the effects of Hh pathway inhibition in a more clinically relevant model containing a human stromal component. We have previously shown that coinjection of HPSCs with cancer cells results in significantly increased pancreatic tumor growth and metastasis in a mouse model (22). After observing that AZD8542 abrogated rSHH-mediated stimulation of GLI1 expression in HPSCs, we sought to determine the in vivo effects of AZD8542 on pancreatic cancer growth and metastasis in our coinjection model of PDAC. Mice were injected in the pancreas with Bxpc3 cells labeled with firefly luciferase (Bxpc3-FL) and HPSCs in T/S ratios of 1:0, 1:1, or 1:3. Four days after cells were injected, mice began treatment with AZD8542 (80 mg/kg oral gavage daily) or vehicle alone. Tumor growth was followed in real time using IVIS imaging. Mice were euthanized on day 32 of treatment.

In the group of mice with Bxpc3 alone without HPSCs (T/S ratio of 1:0), treatment with AZD8542 had no significant effect on tumor growth (data not shown). However, AZD8542 treatment of mice with a T/S ratio of 1:3 resulted in reduced growth of the pancreatic tumors beginning after day 15 of treatment and reached statistical significance at day 30 (Fig. 4A). The final volume of pancreatic tumors after completion of AZD8542 treatment was significantly lower in the group of mice with a T/S ratio of 1:3 compared with the volume in mice with vehicle treatment alone (1054 mm³ ± 224 vs. 4629 mm³ ± 1450, P = 0.04; Fig. 4B). Although tumors in the T/S 1:1 group were smaller with AZD8542 treatment than those in the vehicle control group, this difference did not reach statistical significance (Fig. 4B and Supplementary Fig. S6). Furthermore, expression of downstream GLI1 and PTCH1 was significantly reduced specifically in the stromal component of tumors treated with AZD8542 in the groups with a T/S ratio of 1:3 (Fig. 4C and D), indicating that the Hh pathway was inhibited. However, these markers were not significantly affected in the groups with T/S ratios of 1:0 or 1:1 (data not shown). To confirm whether BxPC3 directly activates Hedgehog signaling in HPSCs, HPSCs labeled with GFP were cultured with BxPC3 for 96 hours and sorted by flow cytometry. Gli1 expression was significantly increased in HPSCs after coculture with BxPC3 cells compared with controls (Fig. 5A), indicating that BxPC3 cells do indeed activate the Hh pathway in HPSCs.

Additional analysis was conducted to elucidate whether SMO inhibition affected cancer or stromal cell proliferation and tumor vascularity. Expression of Ki67 was reduced significantly in the tumors of mice with T/S ratio 1:3 treated with AZD8542 compared to controls (71.9 ± 1.9 vs. 42.2 ± 4.4, P < 0.005; Fig. 5B). The Ki-67 positive cells were primarily in areas of cancer, not stroma, indicating that SMO inhibition affected proliferation of primarily cancer cells but not stellate cells. There was no difference in Ki67 expression in tumors derived from lower amounts of stroma. Tumor vascularity in the T/S 1:3 group was increased with AZD8542 treatment by CD31 immunohistochemistry (24.2 ± 3.1 vs. 82.6 ± 7.1, P < 0.0001; Fig. 5C).

Mice in the T/S 1:3 group also developed fewer liver metastases with AZD8542 treatment compared with control mice (Fig. 6A) with lower luciferase signal (1.86 × 10⁵ vs. 95.9 × 10⁵ photons, P = 0.06), although this difference did not reach statistical significance. Metastatic lesions were confirmed histologically (Fig. 6B). Careful examination by IVIS imaging and gross examination did not reveal a significant volume of other metastatic lesions (e.g., lung, retroperitoneum) other than the liver. Similar to our observations with primary pancreatic tumors, AZD8542 treatment had no effect on liver metastases in the groups with T/S ratios of 1:0 or 1:1 (data not shown). Serial IVIS imaging suggests that the effect of AZD8542 was to inhibit tumor seeding at sites of distant metastasis, rather than to inhibit the expansion of metastatic lesions (Supplementary Fig. S7). Unlike mice in the control group, mice in the drug treatment group did not develop a signal outside the pancreas.

In this study, our results suggest that Hh signaling occurs in a paracrine fashion that is dependent on activation of SMO on neighboring pancreatic stellate cells in the tumor microenvironment of pancreatic cancer. Furthermore, we describe the development of a novel Hh antagonist that potently binds SMO and affects the tumor-associated stroma in mouse models of cancer, resulting in inhibition of tumor growth and metastasis.

Expression of Hh pathway components revealed that SMO and GLI1 were primarily observed in stromal-derived HPSCs, whereas the ligands SHH and IHH were limited to pancreatic tumor cells, suggesting a paracrine signaling mechanism. In support of this, treatment of HPSCs with SHH and IHH activated the Hh pathway, as measured by upregulation of GLI1 mRNA, and stimulated the proliferation of HPSCs. This effect was blocked by the SMO inhibitor AZD8542. No effect of SHH stimulation or SMO inhibition was observed on pancreatic cancer cell activity.

Although previous reports have suggested a possible autocrine role for Hh signaling, the concentration of Hh antagonist required to downregulate the Hh pathway in cancer cells was significantly higher (1.9–6 μmol/L; refs. 15 and 19) than the effective dose in our studies (10–100 nmol/L). Our data correlate with observations by Yauch and colleagues (19), who found that the IC₅₀ of Hh antagonist to inhibit growth of a mesenchymal cell line was 400 times lower than that required for the most sensitive cancer cell line (5 nmol/L vs. 1.9 μmol/L). High concentrations of Hh antagonist resulted in repression of unrelated transcriptional reporters, suggesting that previously observed effects in cancer cells may be because of off-target effect (19).

Using in vivo models of colon cancer and pancreatic cancer, we show that AZD8542 inhibits the Hh pathway specifically in the stromal compartments of each model. In the Colo205 colon xenograft model, species-specific RT-PCR showed strong inhibition of GLI1 in the mouse stromal compartment but no downregulation in the human epithelial compartment. No effect on tumor growth inhibition was observed. However, in both the HT29/MEF and BxPC3/HPSC orthotopic pancreatic cancer models, AZD8542 significantly reduced tumor growth. Interestingly, growth inhibition was observed only when tumors contained both stromal cells and cancer cells but not cancer cells alone. Moreover, downstream Hh signaling in both tumor models was inhibited by AZD8542. Although the BxPC3 cells used in the orthotopic pancreatic cancer model shown in Fig. 4 do not express appreciable levels of SHH, they do express IHH that has similar effects on HPSCs as does SHH (Supplementary Fig. S2). Thus, IHH is most likely the ligand participating in Hh signaling in the BxPC3 orthotopic model. In addition, we have conducted the same experiment using MPanc96, which expresses high levels of SHH, with similar effects of AZD8542 on tumor growth inhibition (Supplementary Fig. S3). Taken together, these results indicate that SMO inhibition blocks Hh signaling in the stromal compartment, resulting in an overall reduction in tumor growth and metastasis.

Our data support several recent reports showing that Hh ligands are expressed by epithelial cells, which in turn activate Hh signaling in the adjacent stroma with the overall effect of promoting tumor growth (18–20). Activation of the Hh pathway by expression of mutant SMO (SMOM2) specifically in pancreatic ductal epithelial cells driven by PDX promoter did not initiate tumorigenesis. In these genetically engineered mice, Hh signaling was active in mesenchymal cells but not neoplastic pancreatic epithelial cells (20). Furthermore, species-specific qRT-PCR analysis of primary human tumor xenografts showed correlation between SHH and IHH expression in the tumor and activation of GLI1 in the stromal compartment (19). Finally, analysis of microdissected samples of human PDAC showed that the levels of Hh ligand (SHH and IHH) were significantly higher in the tumor epithelial cells than in stromal cells, whereas GLI1 was much higher in tumor stroma (20).

To our knowledge, this study is the first to investigate Hh signaling using primary HPSCs derived from PDAC. The precise mechanism by which Hh signaling in HPSCs promotes pancreatic tumor growth and metastasis is unknown. We have shown that conditioned media from activated HPSCs stimulates pancreatic tumorigenesis in vitro and in vivo. It is likely that Hh ligand secreted from neighboring cancer cells stimulates HPSCs to produce soluble factors that then feedback on cancer cells to promote their activity. Expression profiling of tumor xenografts treated with Hh antagonist showed repression of the Wnt pathway and insulin-like growth factor receptor in the mouse stroma (19).

Inhibition of tumor growth in the PDAC/HPSC coinjection model with AZD8542 was associated with decreased cancer cell proliferation and increased vascularity. We did not see an obvious difference in stellate cell proliferation or stroma formation in these tumors and yet our data indicate that SMO inhibition with AZD8542 promotes the activity of stellate cells, not pancreatic cancer cells. It is plausible that SMO inhibition may have act directly on HPSCs to alter the nature of their secreted factors, which in turn indirectly acts on cancer cells in a paracrine manner to result in overall tumor inhibition. Although our observation of increased angiogenesis with SMO inhibition correlates with studies by Olive and colleagues using a genetically engineered mouse model of PDAC (21), a recent report by Chen and colleagues demonstrated a proangiogenic role of Hh signaling (27). Treatment of a colorectal cancer xenograft model with another SMO inhibitor resulted in decreased vascularity (27).These observations seem to be contradictory and thus, additional studies to elucidate the effects of Hh signaling on pancreatic stellate cells and angiogenesis are currently in progress.

Another potential mechanism by which Hh signaling in the stroma might regulate both tumor growth and chemosensitivity is through the promotion of cancer stem cell-like properties. Several investigators have showed the role of Hh signaling in regulating cancer stem cell renewal in breast cancer, gliomas, and leukemia (28–31). Cells that undergo epithelial–mesenchymal transition (EMT) have many properties of cancer stem cells, such as self-renewal and resistance to toxic injuries, including chemoradiation. Although pancreatic stellate cells have not clearly been shown to participate in cancer stem cell renewal, they do promote EMT changes in pancreatic cancer cells (32). Thus, Hh signaling in HPSCs might also affect stem cell properties in PDAC with an overall effect of increased tumorigenesis as well as resistance to chemotherapy.

In PDAC and other solid malignancies, the tumor-associated stroma has been implicated as a physical barrier to the delivery of chemotherapy (33). The effects of Hh blockade on chemoresistance were not studied in this article but have been described previously by others. Using a genetically engineered mouse model of PDAC that develops extensive stroma, Olive and colleagues (21) showed improved efficacy of gemcitabine when combined with an Hh antagonist. Tumors in this model are poorly perfused, which hampers the delivery and efficacy of gemcitabine treatment. However, treatment with an Hh antagonist depleted tumor-associated stroma and improved tumor vascularity, increased intratumoral concentration of gemcitabine, and stabilized disease (21). In support of those findings, we have shown in our HT29/MEF model a reduction of αSMA staining upon treatment with AZD8542. Not only does stroma act as a physical obstacle to delivery of chemotherapy, but it also inhibits its efficacy on a molecular level. We have shown that conditioned media from HPSCs protects cancer cells from apoptosis induced by chemotherapy or radiation (22), although the precise mechanisms are unclear. Hh signaling has been shown to induce chemoresistance as well as resistance to ionizing radiation (34–36). Therefore, it is possible that activation of Hh signaling in HPSCs may also inhibit neighboring cancer cell responsiveness to chemotherapeutic agents or radiation.

A range of clinical trials targeting the Hh pathway has emerged recently. A Phase II study of the SMO inhibitor vismodegib (Genentech) showed efficacy in objective response rate in patients with basal cell carcinoma, resulting in FDA approval in early 2012. However, a recent Phase II trial by Infinity Pharmaceuticals to evaluate the SMO inhibitor saridegib (IPI-926-03) in combination with gemcitabine in patients with metastatic pancreatic cancer was halted after preliminary analysis showed a difference in survival favoring the placebo plus gemcitabine arm. The reasons for failure are unknown since the full data analysis from this trial has not been published. As showed by our data and others, the efficacy of Hh antagonists in PDAC seems to be dependent on the stromal compartment; whether stroma is a significant component and driving force in metastatic pancreatic cancer is unknown. Perhaps the key to success with Hh inhibitors may be the ability to better select patient populations that have tumors dependent on the stroma and the Hh pathway for growth and maintenance.

In summary, our results are direct evidence of Hh signaling in pancreatic stellate cells in the tumor-associated stroma of pancreatic cancer. Activation of the Hh pathway in PDAC occurs in a paracrine fashion, and disruption of the pathway with a novel SMO antagonist, AZD8542, reduced pancreatic tumor growth and metastasis specifically by inhibiting the stromal compartment. Thus, preclinical cancer models that lack stroma are not adequate tools for testing Hh-targeted therapies.

M. M. Hattersley, M. Scarpitti, B. Yang, E. Devereaux, J. L. Brown, and R. Godin are employees of AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R. F. Hwang, T. T. Moore, V. Ramachandran, B. Ji, C. D. Logsdon, J. L. Brown, R. Godin

Development of methodology: R. F. Hwang, T. T. Moore, M. M. Hattersley, B. Yang, E. Devereaux, R. Godin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. F. Hwang, M. M. Hattersley, M. Scarpitti, B. Yang, E. Devereaux, V. Ramachandran, B. Ji, R. Godin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. F. Hwang, T. T. Moore, M. M. Hattersley, M. Scarpitti, E. Devereaux, C. D. Logsdon, J. L. Brown, R. Godin

Writing, review, and/or revision of the manuscript: R. F. Hwang, M. M. Hattersley, C. D. Logsdon, J. L. Brown, R. Godin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. F. Hwang, T. T. Moore, V. Ramachandran, R. Godin

Study supervision: R. F. Hwang, M. M. Hattersley, J. L. Brown, R. Godin

Teaching techniques to authors: T. Arumugam

The authors thank James Janetka, Mukta Bagul, Troy Patterson, Chris Pien, Alex Hird, Les Dakin, Qibin Su, Alex Cao, Patrick Brassil, Dan Russell, and Yan Liu for their contributions and support to the project team.

This work is supported by the National Institutes of Health through grant 1K08CA138912-01A1 to R. F. Hwang and through MD Anderson's Cancer Center Support Grant, 5P30CA016672.

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.