Synergistic Activity of Sorafenib and Sulforaphane Abolishes Pancreatic Cancer Stem Cell Characteristics

Pancreatic cancer is an aggressive malignancy characterized by an extensive local invasion, early systemic dissemination, and pronounced resistance to chemotherapy and radiotherapy (1). Increasing evidence indicates that human pancreatic cancer is driven by “tumor-initiating cells” popularly also known as “cancer stem cells” (CSC) that may contribute to tumor metastasis and therapeutic resistance (2–5). We have recently shown that a CD24⁻/CD44⁺ phenotype characterizes the aldehyde dehydrogenase 1 (ALDH1)–positive, therapy-resistant, spheroid, and colony-forming population in MIA-PaCa2, AsPC1, Capan-1, and BxPc-3 pancreatic cancer cell lines (6). Importantly, conventional chemotherapy or radiotherapy does not completely eradicate CSCs in pancreatic cancer (7, 8), suggesting the need for novel therapies based on the elimination of CSCs.

Sorafenib (SO), a novel multikinase inhibitor, also known as BAY 43-9006 or Nexavar, has shown promising antitumor activity in several clinical phase I/II combination studies with oxaliplatin (9), 5-fluorouracil and leucovorin (10), gemcitabine (11), doxorubicin (12), docetaxel (13), and paclitaxel plus carboplatin (14). Multiple phase I, II, and III trials with SO alone or in combination with other chemotherapeutic drugs are ongoing. Mechanistically, SO inhibits vascular endothelial growth factor (VEGF) receptor-2 and VEGF receptor-3 expressed by endothelial cells and CD133⁺ circulating hematopoietic progenitor cells (15) and platelet-derived growth factor receptor-β, FLT3, Ret, and c-Kit expressed by vasculature-associated cells and tumor cells (16, 17). These kinases are involved in tumor angiogenesis, as well as in growth and progression of a variety of tumor types such as leukemias and renal cell and hepatocellular carcinomas, suggesting CSC targeting by SO. However, recent studies in mouse models of pancreatic neuroendocrine tumors, glioblastoma, and breast cancer showed an unwanted side effect of SO and related antiangiogenic substances: After initial antitumor activity, enhanced tumor progression and increased metastasis occurred (18, 19). The underlying mechanism may be due to induction of hypoxia by antiangiogenic treatment and selection of highly resistant CSCs adapted to depletion of oxygen and nutrition. Therefore, combination of SO with a “sensitizer” may be necessary for complete elimination of CSCs. A promising candidate for combination is the dietary food component sulforaphane (SF). This isothiocyanate is enriched in the plant family of Brassicaceae and present in high concentration in broccoli. SF inhibits phase 1 cytochrome P450 enzymes, induces phase 2 metabolic enzymes, and acts as an antioxidant by increasing reduced glutathione levels as well as inducing cell cycle arrest and apoptosis. In addition, SF exhibits anti-inflammatory properties and inhibits angiogenesis (20). SF has recently been shown to sensitize pancreatic CSCs by downregulation of NF-κB along with inhibition of CSC properties (e.g., ALDH1 and self-renewal activity, apoptosis resistance, and growth in immunodeficient mice; ref. 6).

In the present study, we used in vitro assays and in vivo xenograft models to examine the effects of SO on the pancreatic CSC population. We showed that CSCs are diminished by SO, although some clones acquire resistance. This may be due to the observed strong induction of NF-κB signaling by SO. Therefore, we used SF cotreatment, which completely downregulated NF-κB activity and strongly increased the anti-CSC properties of SO.

BxPc-3 (CSC^low) and MIA-PaCa2 (CSC^high) pancreatic cancer cell lines were obtained from the American Type Culture Collection and authenticated throughout the culture by the typical morphology. Mycoplasma-negative cultures were ensured by weekly tests. Primary skin fibroblasts were kindly provided by Dr. H-J. Stark (DKFZ, Heidelberg, Germany). Cells were cultured in DMEM (PAA) supplemented with 10% heat-inactivated FCS (Sigma) and 25 mmol/L HEPES (PAA). Human mesenchymal stem/stromal cells were isolated from bone marrow and cultured as described (21).

Sorafenib tosylate [N-(3-trifluoromethyl-4-chlorophenyl)-N′-(4(2-methylcarbamoyl pyridin-4-yl)oxyphenyl) urea] was kindly provided by Bayer Corporation (West Haven, CT) and dissolved in DMSO (Sigma-Aldrich) to a 100 mmol/L stock. This stock was further diluted for in vivo studies in a Cremophor EL/100% ethanol mixture (1:1 ratio) to a 4× stock. Sulforaphane [1-isothiocyanato-4-(methylsulfinyl)-butane] (Sigma Aldrich) was dissolved in 100% ethanol to a stock solution of 100 mmol/L. For in vivo studies, this stock was further diluted with sterile PBS. A 40 mmol/L Z-VAD-FMK stock (Promega) was prepared in DMSO. The concentration of DMSO in the working solutions did not exceed 0.1% (v/v).

MIA-PaCa2 tumor cells (2 × 10⁷ in 200 μL of PBS) were injected s.c. into the left and right flanks of 4-week-old NMRI (nu/nu) female mice. Two weeks later, xenografts were minced and subtransplanted into 6-week-old mice. After the tumors had reached a mean diameter of 8 to 10 mm, mice were randomized to six groups of six animals per group. The mice were treated i.p. on 3 consecutive days with SF at a dose of 3 mg/kg or with SO at doses of 30, 40, and 60 mg/kg or with a combination of SF (3 mg/kg) and SO (60 mg/kg). The control group received a Cremophor EL/95% ethanol/PBS mixture (1:1:6 ratio). Tumor growth was monitored as described recently (6). Animal experiments have been carried out at the animal facility of the University of Heidelberg after approval by the Baden Württemberg animal oversight committee (Regierungspräsidium Karlsruhe, Germany).

Cells were stained with Nicoletti buffer as described (22) and DNA fragmentation was identified by flow cytometry (FACScan, BD Biosciences).

Viability was measured using MTT as described previously.

Tumor cells were cultured in human NeuroCult NS-A basal serum-free medium with supplements (StemCell Technologies, Inc.) and spheroid formation was evaluated as described (6).

Tumor cells were seeded at a density of 3 × 10⁵ per well in six-well tissue culture plates (BD Falcon). Twenty-four hours later, the cultures were pretreated with SF for 24 hours and subsequently treated with SO for 48 hours. Afterward, cells were trypsinized, washed with PBS, and resuspended in cell culture medium at a density of 200 per well in six-well tissue culture plates and colony formation was evaluated as described (23).

MIA-PaCa2 cells (6 × 10⁵) were seeded in six-well plates and grown to confluence. Twenty-four hours after treatment, a line was scraped within confluent cells using the fine end of a 10-μL pipette tip (time 0).

ALDEFLUOR substrate (5 μL; Aldagen, Inc.) was added to 1 × 10⁶ treated tumor cells in 500 μL of assay buffer and incubated for 60 minutes at 37°C. Pretreatment with the ALDH1 inhibitor diethylaminobenzaldehyde was used as a negative control.

Kits providing fluorochrome inhibitors of caspases (FLICA) were used according to the manufacturer's recommendations (Immunochemistry Technologies) and caspase activity was analyzed by flow cytometry or immunofluorescence microscopy as described (6).

Cell proteins were isolated and Western blot analysis was done as described recently (6).

Whole-cell extract was harvested as described recently (24) and electrophoretic mobility shift assay (EMSA) was done as described (6).

Immunohistochemistry on 6-μm frozen or paraffin-embedded tissue sections was done as described previously (6). Antibodies used were rat anti-mouse CD31 monoclonal antibody (mAb; PharMingen) and rabbit polyclonal antibodies against cleaved fragment of activated human caspase-3 (R&D Systems), human Ki-67 (Thermo Scientific), and human ZEB1 (Santa Cruz). Mouse mAbs were anti-human Twist2, anti-vimentin (Abcam), and anti–HIF-1α (BD Biosciences).

For MTT and fluorescence-activated cell sorting (FACS) measurements, data are presented as the mean ± SD. Data were analyzed using Student's t test for statistical significance. P < 0.05 was considered statistically significant. For xenografts of nude mice, a distribution free test for tumor growth curve analyses for therapy experiments was used as described by Koziol and colleagues and Kallifatidis and colleagues (25, 26).

We used MIA-PaCa2 and BxPc-3 pancreatic cancer cell lines, which we have recently characterized as CSC^high and CSC^low, respectively (Supplementary Table S1). Morphologically, CSC^low cells consist of only one adherent cell phenotype, whereas CSC^high cells contain adherent cells and spheroidal floating cells (Fig. 1A). To investigate whether SO targets pancreatic CSCs, both cell lines were treated with 2, 5, 10, 20, or 50 μmol/L SO, and 48 hours later, cell viability was analyzed by MTT assay. SO treatment resulted in a dose-dependent inhibition of viability in both cell lines (Supplementary Fig. S1A). However, SO at the highest concentration of 50 μmol/L was toxic to nonmalignant primary human skin fibroblasts and bone marrow–derived mesenchymal stem/stromal cells (Supplementary Fig. S1B and C). Therefore, SO was applied at a subtoxic concentration of 20 μmol/L in all further experiments. This concentration significantly reduced colony and spheroid formation (Fig. 1B, compare Fig. 3C). However, some small colonies survived, resulting in regrowth of spheroids 30 days later (Fig. 1C). These data indicate that SO alone does not completely diminish the clonogenicity and self-renewal capacity of CSCs in vitro.

To investigate the effects of SO in vivo, MIA-PaCa2 CSC^high cells were s.c. transplanted into nude mice. Treatment with SO at days 11, 12, and 13 after tumor inoculation strongly inhibited tumor growth in a dose-dependent manner (Fig. 2A). No toxic side effects of SO occurred as concluded from the unchanged general condition of mice and constant body weight during treatment (Fig. 2B and data not shown). In xenograft tissue, obtained 17 days after treatment, SO reduced the percentage of Ki-67–positive tumor cells in a dose-dependent manner, suggesting inhibition of proliferation (Fig. 2C and D). Likewise, blood vessel density was decreased by SO as examined by CD31 staining of endothelial cells. Accordingly, apoptosis was induced as concluded from enhanced activity of caspase-3, which was, however, not significant. These data show that SO retards the growth of pancreatic CSC^high xenografts by diminishing angiogenesis and proliferation and, most likely, by induction of apoptosis. However, SO did not completely eradicate the tumor mass, suggesting the need for combination therapy to increase the effect.

For sensitization, we pretreated BxPc-3 CSC^high and MIA-PaCa2 CSC^low cells with SF for 24 hours followed by incubation with SO for an additional 48 hours. This resulted in a more pronounced induction of cell death compared with each substance alone as indicated by cell morphology (Fig. 3A), colony and spheroid formation (Fig. 3B and C), and ALDH1 positivity (Fig. 3D). Fluorogenic substrate assays indicated that SF or SO single treatment activated caspase-3/7, caspase-8, and caspase-9 only marginally, whereas the combination strongly activated caspase-3/7, caspase-8, and caspase-9 in both cell lines with an additive effect (Fig. 4A). To investigate whether SF- and SO-induced apoptosis could be abrogated by caspase inhibition, we preincubated cells with pan-caspase inhibitor Z-VAD-FMK for 4 hours before treatment (Fig. 4B). Z-VAD-FMK significantly decreased apoptosis in both cell lines as evaluated by Nicoletti staining and detection of DNA fragmentation by FACS analysis. To determine whether the expression of antiapoptotic proteins was altered, we examined cIAP-1, XIAP, and cFLIP expression by Western blot. Generally, there was a difference in basal expression of antiapoptotic proteins between the two cell lines: Whereas cIAP, XIAP, and cFLIP were highly expressed in CSC^high cells, only cIAP was highly expressed in CSC^low cells, which reflects the higher apoptosis sensitivity of this latter cell line (Fig. 4C). Correspondingly, the effects of SO and SF on expression of antiapoptotic proteins were more pronounced in CSC^high cells because both SF and SO reduced the expression of cIAP, XIAP, and cFLIP, whereas the effect was highest with the combination treatment. In combined therapy, SF increases targeting of CSC characteristics by SO, involving colony and spheroid formation, ALDH1 activity, and apoptosis resistance.

To investigate whether SO may sensitize pancreatic CSCs by affecting NF-κB survival signaling, we performed gel retardation assays. Unexpectedly, we found a strong increase of DNA binding induced by SO in both cell lines (Fig. 5A). Although SF itself did not enhance promoter binding of NF-κB complexes, it mediated a marked reduction of SO-induced NF-κB activity in BxPc-3 CSC^low cells and totally abolished it in MIA-PaCa2 CSC^high cells. To elucidate the transactivation activity, we analyzed the composition of the SO-induced NF-κB complexes by incubation of cell extracts with the NF-κB consensus sequence in the presence of specific antibodies recognizing the various NF-κB subunits (Fig. 5B). In both cell lines, a selective supershift of p65 (RelA) was observed, whereas p50 and cRel could not be detected. In addition, RelB- and p52-specific shifts did not occur (data not shown). To evaluate p65 (RelA) activity by its phosphorylation pattern, we performed Western blot analysis with antibodies recognizing total p65 and phosphorylated p65. In line with the gel retardation assays, treatment with SO, but not with SF, led to a stronger phosphorylation of p65 (Fig. 5C). Combination of SO and SF slightly diminished phosphorylation of p65 in CSC^low cells, but not in CSC^high cells. These results indicate binding of a high amount of transactivation-potent p65 (RelA)-containing NF-κB dimers on treatment with SO. SF inhibits DNA binding but not the phosphorylation pattern of p65. Because NF-κB activity is associated with increased cell migration, we investigated whether SO affects the cell motility of MIA-PaCa2 CSC^high cells. Short-term scratch assays were performed under hypoxic conditions to enhance migratory activity. Compared with cells cultured under normoxic conditions, hypoxia led to faster migration of CSC^high cells into the wounded region (Fig. 5D). SF, but not SO, strongly inhibited hypoxia-induced migration. However, SO completely prevented the migratory potential on coincubation with SF.

To evaluate the in vivo effects of combined SF and SO treatment, we s.c. xenografted MIA-PaCa2 CSC^high cells into nude mice. After treatment, tumor size and body weight of mice were monitored daily. Treatment with SO or SF alone retarded tumor growth, but combined SF and SO treatment was strongest and significantly reduced tumor growth (P < 0.05; Fig. 6A). Importantly, there was neither a change in body weight of mice (Fig. 6A, right) nor a significant elevation of plasma levels of lactate dehydrogenase, aspartate aminotransferase, or alanine aminotransferase (Supplementary Fig. S2A), suggesting that the treatment was not toxic. To investigate the potential mechanisms of tumor growth retardation, tissue samples were analyzed by immunohistochemistry. Double staining of CD31 and active caspase-3–positive cells in xenograft samples indicated that tumor vascularity was reduced and apoptosis was induced on combined treatment with SO and SF (Fig. 6B). Furthermore, we observed a strong reduction of tumor cell proliferation after combined SF and SO treatment as analyzed by Ki-67 immunostainings. Finally, we investigated protein expression involved in epithelial-mesenchymal transition (EMT), which has been associated with early signs of metastasis following tumor hypoxia and NF-κB activity (27, 28). First, SO, but not SF, strongly increased the expression of HIF-1α, suggesting induction of hypoxia by SO. Combination with SF decreased the induction of hypoxia by SO (Fig. 6C). Correspondingly, the EMT proteins Zeb-1 and Twist2 were upregulated after SO treatment, whereas SF treatment alone had no effect; however, their expression decreased after cotreatment with SO and SF. Expression of vimentin, a mesenchymal cell marker, was already upregulated in untreated CSC^high tumors and not further affected by SO. In contrast, SF alone or in combination with SO downregulated vimentin expression. Taken together, these data suggest that growth retardation of CSC^high tumor xenografts by combined SF and SO treatment involves antiangiogenic, antiproliferative, and proapoptotic mechanisms. Moreover, SO-induced inhibition of angiogenesis may have been the reason for hypoxia induction, which may be, together with SO-induced NF-κB activity, the reason for the observed enhanced expression of proteins involved in EMT.

Here, we show that the multikinase inhibitor SO alone diminishes pancreatic CSC-like tumor cells, but depletes them significantly only in combination with the phytochemical SF. The major CSC characteristics of the CSC^high MIA-PaCa2 and CSC^low BxPc-3 cell lines have been described in our recent study (6). In the present work, we describe the presence of two cell populations, which can be found in CSC^high but not in CSC^low cells, namely, spheroidal cells and fibroblast-like adherent cells. These two populations may correspond to primitive CSCs and differentiated daughter cells, suggesting differentiation potential of in vitro cultured pancreatic CSCs. SO targeted both cell subpopulations present in the CSC^high cell line in vitro. This was due to the induction of apoptosis, DNA fragmentation, and downregulation of antiapoptotic proteins. In addition, SO strongly diminished clonogenicity, spheroid formation, and ALDH1 activity. Most importantly, the concentration of 20 μmol/L SO was selectively toxic to CSCs but not to nonmalignant primary skin fibroblasts and bone marrow–derived mesenchymal stem/stromal cells. This finding corresponds to our in vivo results showing that treatment of mice harboring CSC^high xenografts with SO reduced tumor growth in a dose-dependent manner, without obvious side effects to mice, as the general state of health, body weight, and liver parameters were constant during and after treatment. Reduction of pancreatic tumor xenografts was due to inhibition of proliferation and angiogenesis and induction of apoptosis. Our results confirm data obtained in thyroid carcinoma and hepatocellular carcinoma xenograft models (29, 30), in which the authors describe inhibition of angiogenesis and tumor growth along with induction of apoptosis following treatment of the mice with SO.

We are the first to report increased NF-κB activity on SO treatment. We found induction of p65 (RelA)-containing transactivation-competent dimers by SO as shown by supershift assays and detection of phosphorylated p65 proteins by Western blot analysis. This result was surprising because NF-κB activity may render tumor cells resistant to conventional cancer therapy. However, many chemotherapeutic agents and the apoptosis-inducing ligand tumor necrosis factor–related apoptosis-inducing ligand share this unwanted feature of SO and can activate NF-κB (6). Basal NF-κB overexpression is observed in various malignant tumors including pancreatic adenocarcinoma and has been made responsible for high therapy resistance of pancreatic tumor cells (31, 32). We assume that enhanced NF-κB binding by SO could be the reason for the observed selection of resistant CSC^high clones on long-term treatment with SO. NF-κB was recently identified as a central mediator of EMT and metastasis in a model of breast cancer (28). Our xenograft studies suggest that NF-κB activity may be involved in induction of EMT following SO treatment, resulting in the selection of a more invasive and metastatic phenotype. Such a scenario has recently been described, as transcription factors involved in EMT are possible downstream targets of NF-κB (33, 34).

To overcome SO-induced NF-κB activity and associated resistance, we chose SF for cotreatment because we recently showed that this broccoli-derived isothiocyanate is able to sensitize pancreatic CSCs toward conventional therapy by strong downregulation of NF-κB activity (6). Indeed, SF totally abolished SO-induced NF-κB binding in CSC^high cells, whereas no effect in CSC^low cells was observed. SF may directly interfere with promoter binding of p65 (RelA), rather than downregulating p65 phosphorylation, as we conclude from our supershift assays and Western blot analysis of phosphorylated p65. Furthermore, combination of SO with SF considerably reduced the clonogenicity and self-renewal potential of CSCs and enhanced apoptosis in vitro. In addition, suppression of NF-κB sensitized CSCs to SO-induced apoptosis, as measured by caspase activity and DNA fragmentation. Although we were not able to detect any NF-κB changes in xenografts, this transcription factor may be involved in the observed effects in tumors because several in vivo studies show that blocking of NF-κB in pancreatic cancer affects angiogenesis, apoptosis, and proliferation (35).

A notable recent observation by other authors is that SO and related antiangiogenic substances induce pronounced initial tumor growth retardation. However, this is followed by adaptation of the tumor to treatment and subsequent regrowth (18, 19). Our results confirm this observation because a few clones of our long-term treated CSC^high cells adapted to SO treatment and survived. These results match our in vivo findings, where inhibition of angiogenesis by SO may have induced hypoxia and EMT. Another potential mechanism explaining the tumor regrowth in the presence of SO could be that after transient vessel reduction induced by the antiangiogenic treatment, the delivery of therapeutic substances is impaired.

In conclusion, our data show that although SO is initially effective, it does not lead to total elimination of CSCs, which start to regrow after an interim. Combination with an NF-κB inhibitor such as SF may be a therapeutic option to improve the therapeutic effect of SO.

No potential conflicts of interest were disclosed.

We thank Bayer Corporation for sorafenib and H-J. Stark for primary skin fibroblasts.

Grant Support: Bundesministerium für Bildung und Forschung (01GU0611), Tumorzentrum Heidelberg/Mannheim (D10027(6)350), Stiftung Chirurgie Heidelberg, Dietmar-Hopp Stiftung, Deutsche Krebshilfe (107254), and Deutsche Forschungsgemeinschaft (SFB 518/A17).

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.