CD133⁺ Tumor Initiating Cells in a Syngenic Murine Model of Pancreatic Cancer Respond to Minnelide

This article is featured in Highlights of This Issue, p. 2239

Pancreatic ductal adenocarcinoma (PDAC) is currently the fourth most frequent cause of cancer-related deaths (1). Aggressive biology, chemoresistance, and recurrence of the tumor make PDAC one of the most devastating cancers.

Since their discovery, tumor-initiating cells (TIC) have referred to a quiescent population of cells that is capable of causing recurrence, both local and distal to the site of tumor. Pancreatic TICs were first isolated as a CD24⁺CD44⁺ESA⁺ population from human xenografts by Li and colleagues (2). Subsequently, CD133 was identified as a TIC for PDAC (3). Both groups demonstrated tumor-initiating properties in the cells enriched for these markers. Pancreatic TICs have been further characterized using other surface markers, including CXCR4 (3, 4), Met (5), aldehyde dehydrogenase activity (6–8), or ABCG2 (9, 10).

CD133, a transmembrane pentaspan protein, was initially described as a surface antigen specific for human hematopoietic stem cells (11, 12). Although the biologic function of CD133 remains unknown, CD133 is well recognized as a stem cell marker for normal and cancerous tissues. It has recently been demonstrated that CD133 is a marker of putative pancreatic progenitor cells in mice (13). In humans, CD133 is expressed on terminal ductal cells in the adult pancreas and carcinoma cells in pancreatic ductal adenocarcinoma (14, 15).

TICs are hypothesized to arise by either accumulation of mutations in stem cells or progenitor cells resulting in dysregulation of several self-renewal pathway genes. Consistent with this, pancreatic TICs have been reported to show increased expression of developmental transcription factors such as Sox2, Nanog, Oct4 (16), and several Hedgehog and Notch pathway genes (2, 17). Apart from the above features, resistance to apoptosis by upregulation of Bcl-2 and Survivin genes has been associated with pancreatic TICs (18). TICs are also associated with increased invasiveness. Pancreatic TIC population CD133⁺/CXCR4⁺ or Met⁺/CD44⁺ has been reported to be more invasive than either population alone.

Most studies on pancreatic TICs have involved isolation of cells from patient tumor tissue and propagation in immunocompromised mouse models such as severe combined immunodeficient (SCID) mice or athymic nude mice (2, 3, 19). However, xenografts in immunocompromised mice lack the syngenic interface between tumor and host microenvironment crucial for understanding tumor dynamics. One of the immunocompetent mouse models for pancreatic carcinogenesis is based on the pancreas-specific expression of endogenous mutant Kras and Trp53 alleles (LSL-Kras^G12D, LSL-Trp53^R172H, Pdx-Cre or KPC mice). These KPC mice spontaneously develop primary pancreatic tumors that recapitulate the clinical and histopathologic features of the human disease (20, 21).

TICs often display resistance to cytotoxic cancer therapies, permitting the repopulation of tumors after radiation or chemotherapy. Several groups have demonstrated that TICs from multiple cancer types exhibit resistance to conventional cancer therapies (22). PDAC is known to be resistant to most chemotherapeutic drugs. However, triptolide, a diterpene triepoxide from the Chinese plant Trypterygium wilfordii, downregulates heat shock genes (23, 24) and induces apoptotic death in pancreatic cancer cells (25–27) has been an exception to this. Triptolide and its water soluble prodrug Minnelide was recently reported to be very effective in tumor regression in a number of murine models (26). Other authors have demonstrated the efficacy of triptolide by inhibition of proliferation within a number of additional malignancies, including cholangiocarcinoma (28, 29), osteosarcoma (30), and neuroblastoma (31). However, efficacy of triptolide has not been tested on CSCs.

In this study, we have identified a population of CD133⁺ cells from the tumors developed from the KPC transgenic mouse model of PDAC. We have shown that this population expresses a number of CSC markers (surface markers, transcriptional markers, and developmental markers); has a significantly higher expression of prosurvival genes like the heat shock proteins, Bcl-2 and Survivin; higher NF-κB activity and has tumor initiating properties in a syngenic, immunocompetent system. We have further shown that these cells, and the tumors derived from these cells, respond to Minnelide, which effectively lowers the pro-proliferative pathways and induces cell death.

Moribund animals of the LSL-Kras^G12D, LSL-Trp53^R172H, and Pdx-Cre genetic background was sacrificed and single-cell suspension was isolated by digestion with collagenase B and dispase II. Cells were plated in medium containing growth factors and 2% serum. Medium was replaced with serum-free medium after 48 hours and cells were maintained in it for 2 to 3 weeks in the absence of serum until all fibroblasts were removed. Cells were then grown in Dulbecco's modified Eagle medium (DMEM) with 10% serum for all experiments. The 2 cell lines derived were named KPC001 and KPC023. All experiments were carried out in 2 derived cell lines and 2 to 3 primary tumors from KPC animals.

The CD133⁺ population was separated from the mouse progenitor cells and other CD133⁻ cells using MACS separation (Miltenyi Biotech) using manufacturers protocol. Single-cell suspension was generated from tumors in KPC mice according to the Li and colleagues (32). Non-epithelial progenitor cells were removed using anti-CD31-Biotin (BD Biosciences) and anti-CD45 Biotin (BD Bioscience) using MACS technique. The flow through free from the mouse progenitor cells was bound to anti-mouse CD133-Microbeads for 10 minutes on ice and positively purified for CD133⁺ cells by MACS. The purity of separation was tested for each batch by performing a FACS analysis using anti-CD133-PE antibody AC141 (Miltenyi Biotech). The separated populations were used for RNA, protein, and FACS analysis. Cells growing in culture were scraped gently into centrifuge tube and washed once in wash buffer (PBS, 0.5% BSA, 2 mmol/L EDTA) before binding to anti-mouse CD133 microbeads and proceeded as described earlier.

For flow cytometric analysis of tumor tissues, single-cell suspension was prepared according to the protocol of Li and colleagues (32). Cells isolated from tumor, as well as those growing in culture, were stained with directly conjugated monoclonal antibodies in the presence of FcR blocking reagent (Miltenyi Biotech) with anti-CD133-PE (Miltenyi Biotech), anti-CD44-FITC (BD Biosciences), ESA-APC (BD Bioscience), and CD24-PE (BD Biosciences). The Aldefluor assay was performed per manufacturer's instructions (Stem Cell Technologies) complete with DEAB controls. Immunoglobulin G (IgG) isotype controls corresponding to each directly conjugated fluorophore were utilized to identify, quantify, and positively select desired cell populations. All FACS analyses were performed on a BD FACSCanto II (BD Biosciences) using FACSDiva (BD Biosciences) and FlowJo (Tree Star) software. Debris and cell clusters were excluded during side- and forward-scatter analyses.

Migration assay for different cell lines was conducted by Electric Cell–Substrate Impedance Sensing (ECIS; Applied Biophysics). In ECIS, the cells are grown on the surface of small and planar gold-film electrodes and the AC impedance of the cell-covered electrode is measured continuously at a frequency of 64 kHz. A total of 4 × 104 pancreatic cancer cells (MiaPACA-2, S2VP10, S2013, AsPC-1, and KPC001) were seeded into an 8E1W ECIS array (Applied Biophysics containing a 250 μmol/L electrode. Upon confluence, a high field current with frequency 48 kHz and amplitude 5 was applied for 10 seconds, killing cells overgrowing the electrode, creating “wounds” in the wells devoid of cells.

The migration was compared between each cell line at the end of each experiment, which was presented as a measure of changing impedance over time after normalizing to the time for wounding.

For immunohistochemistry, paraffin tissue sections were deparaffinized in xylene and hydrated through graded ethanol. Slides were steamed with a Reveal Decloaker (Biocare Medical) to minimize background staining. Sniper Universal Blocking Sera (Biocare Medical) were used throughout the protocol. The slides were stained using anti-Ki67, anti-HSP 70 antibody (Abcam), and anti-GRP78 antibody (Cell Signaling Technology). Staining was detected using anti-rabbit secondary antibody conjugated to horseradish peroxidase followed by Diaminobenzidine Peroxidase Substrate Kit (Vector Laboratories). The tissue sections were counterstained with Gill hematoxylin (Vector Laboratories). The primary antibody was omitted for the negative controls. TUNEL was performed using the In Situ Cell Death Determination Kit (Roche Diagnostics) according to manufacturer's protocol. For immunofluorescence, fluorescent antibody conjugates were used after primary antibody staining. Slides were counterstained with 4′,6-diamidino-2-phenylindole and visualized in a Nikon fluorescent microscope. Tissue samples were incubated with mouse IgG1 isotype controls (BD Biosciences) and did not demonstrate any specific staining.

NF-κB activity assay was performed on CD133⁺ sorted cells and minnelide-treated CD133⁺ generated tumors. DNA binding and downstream activation of NF-κB pathway was assessed using a transcription factor binding ELISA (Thermo Scientific) according to manufacturer's instruction.

Microarray studies were performed at the University of Minnesota Biomedical Genomics Center. RNA samples from 3 independent cell sorts were submitted to the facility for Illumina microarray analysis. Results were analyzed using GeneData Expressionist software at the Minnesota Supercomputing Institute.

All animal experiments were performed according to the University of Minnesota Animal Care Committee guidelines.

Sorted cells were washed with serum-free HBSS and suspended in serum-free DMEM/Matrigel mixture (1:1 volume). A total of 10, 50, 100, and 500 cells (CD133⁺/CD133⁻/unsorted) were injected subcutaneously into the right and left flank of age and gender matched C57BL/6 mice (Jackson Laboratories). A total of 8 to 10 mice were used in each group. Tumors were measured weekly and tumor volume was documented. Tumors were allowed to grow until they reached a volume of 1 cm³, at which the mice were sacrificed and the tumor tissue was harvested and processed for other experiments.

For serial transplantation, tumor measuring ∼700 to 800 mm³ was chopped up into approximately 1 to 2 mm. Tumor pieces were implanted subcutaneously into right and left flank of age matched 10 C57BL/6 mice. Tumors were measured weekly to monitor their progress.

Minnelide treatment was started 5 days after tumor implantation following randomization of animals. 0.42 mg Minnelide/kg body weight was administered intraperitoneally every day for 25 days. Tumors were measured as stated above. At the end of study, animals were sacrificed according to the University of Minnesota Animal Care guidelines.

For orthotopic model, 500 CD133⁺ cells or 5,000 unsorted KPC cells were injected into the pancreas of 24 C57BL/6 mice. A total of 5 to 6 mice were sacrificed every 10 days to monitor development of tumor in each group.

Values are expressed as the mean ± SEM. All in vitro experiments were performed at least 3 times. Statistical significance of results was calculated using the Student t test. Columns represent mean; bars represent SE (n = 4; *P < 0.05).

We analyzed 3 primary KPC tumors and 2 cell lines derived from KPC mouse tumors for PDAC for the different stem cell markers. A flow cytometric analysis showed these cells had 6% to 9% of CD133⁺ population (Fig. 1A and Supplementary Table S1). A population of CD24⁺/CD44⁺/ESA⁺ cells were also present but to a much lesser percent (3% to 4%) compared with the CD133⁺ population (Fig. 1B). A minor population of these cells (∼2%) also showed higher Aldh1 activity (Fig. 1C). To put our study in perspective, we studied TIC markers from a classical TIC model: human tumors transplanted in mice as well as one that was freshly isolated tumor. These tumors showed 3% to 4% CD133⁺ cells whereas 1% to 2% CD24⁺CD44⁺ESA⁺ cells.

To see if the percent of CD133⁺ cells differed in metastatic tumors or in accumulated ascitic fluid in KPC tumor bearing animals, we analyzed cells from accumulated ascitic fluid from these animals. Our analysis showed ∼2% CD133⁺ cells were present in ascitic fluid from 3 animals tested. Similarly, cells from a liver metastasis nodule from a pancreatic tumor bearing KPC animal also showed ∼2% CD133⁺ cells. The relative abundance of CD133⁺ cells from primary tumor, malignant ascites, or metastatic nodules is summarized in Table 1.

Our results showed that CD133⁺ population was higher (6%–9%) in the primary tumor whereas it was ∼2% to 3% in malignant ascitic cells or metastasizing cells. We next looked for expression of genes involved in development of the pancreas and known to be upregulated in pancreatic TICs. KPC cells showed several hundred fold overexpression of CD133, Sox2, Nanog, Aldh1, CXCR4, SHH, Gli, and Notch genes (Fig. 1D) compared with the normal pancreatic ductal cells. These data indicated that KPC cells harbor a population of TICs characterized by the same markers as was previously reported from the human tumor xenografts and pancreatic cancer cell lines (2, 3).

Because ∼6% to 9% of cells derived from the KPC tumor were found to be CD133⁺, we used magnetic cell sorting (MACS) with anti-mouse CD133 (AC133) coupled microbeads to isolate this population. The enrichment for CD133 was verified by flow-cytometry using CD133 antibody (AC141) against a different epitope after purification. Approximately, 97% purity was achieved consistently (Fig. 2A) by this cell separation technique. Sorted cells were labeled with CD24, CD44, and ESA antibody and assayed for Aldh1 activity. It was observed that CD133⁺ population from KPC001, KPC023 as well as cells isolated from KPC primary tumors, included the CD24⁺CD44⁺ESA⁺ population, whereas the CD133⁻ did not (Fig. 2B and Supplementary Fig. S1A).It was also seen that CD133⁺ population harbored cells with maximum Aldh1 activity whereas the CD133⁻ population lacked any Aldh1 activity (Fig. 2C and Supplementary Fig. S1A). This indicated that both CD24⁺CD44⁺ESA⁺ population and Aldh1⁺ population were subpopulations of CD133⁺ cells. This was further confirmed with immunofluorescence, which showed that only CD133⁺ cells costained with Aldh1 (Fig. 2D).

CD133 expression has been correlated with poor prognosis in hepatocellular carcinoma (33). To evaluate if higher levels of CD133⁺ cells correlated with increased aggressiveness in pancreatic cancer cells, CD133 expression in normal pancreatic ductal cells was compared with cancer cell lines (MIA-PaCa2, BxPC3, Capan1, S2013, and S2VP10, AsPC1), human patient tumor xenografts, and KPC tumor–derived cells. KPC001, KPC023 derived from the tumor had increased (6%–9%) expression of CD133 compared with 1%–4% in all the other cells (Supplementary Fig. S1B). No significant expression of CD133 was seen in normal pancreatic ductal cells (0.01%) and 0.5% to 1% expression of CD133 was observed in the less aggressive pancreatic cancer cells such as MIA-PaCa2. The KPC001 cells (with 8% CD133) showed increased rate of migration (assayed as a measure of wound healing ability) compared with the other pancreatic cancer cell lines (Supplementary Fig. S1C). Migration was measured by ECIS and the data were represented as changing impedence over time. Because KPC cells showed increased invasiveness and because they had higher CD133 expressing cells, we studied the expression of genes involved in EMT in CD133⁺ and CD133⁻ cells isolated from the KPC cell lines (both KPC001 and KPC023). Consistent with the functional assay, expression of EMT-associated transcription factors such as SNAI1, SLUG, and Twist were 10- to 80-fold higher in CD133⁺ cells (Supplementary Fig. S1D).

Tumorigenicity of CD133⁺ population was next tested by injecting 100, 50, and 10 MACS sorted cells subcutaneously in both flanks of C57BL/6 mice. Tumor volume was measured weekly to monitor tumor growth (Fig. 3A). Tumors formed from 10 CD133⁺ cells in these mice within 3 weeks of cell injection. Neither CD133⁻ cells nor unsorted cell (Fig. 3B) population formed tumors up to 2 months post-inoculation at these cell numbers. Unsorted KPC cells did not show tumor formation until 20,000 cells were injected. Tumorigenicity of CD133⁺ cells is summarized in Supplementary Table S1. To put our study in perspective, we isolated CD133⁺ cells from 3 human tumor xenografts propagated in mice and implanted these at low numbers in immunocompromised SCID mice. CD133⁺ cells showed tumor initiation at as low as 100 cells whereas CD133⁻ cells did not even at 1,000 cells (Supplementary Fig. S2A–S2D).

Because one of the hallmarks of cancer stem cells is their ability to generate serially transplantable tumors that recapitulate the cellular and histologic structure of the parent tumor (3, 34), we injected 100 CD133⁺ cells subcutaneously into the flanks of the C57BL/6 mice. Tumors formed in an average of 21 days. 1 mm³ tumor pieces were implanted into a new set of C57BL/6 mice and observed for a month once the parental tumor reached a volume of 700 to 800 mm³. Similar transplantation was done with the second-generation tumors, once they reached a size of 700 to 800 mm³ (Supplementary Fig. S3A and S3B). The tumors from all the 3 generations were histologically identical as seen by H&E staining (Supplementary Fig. S3C), and showed identical proliferation pattern as seen by Ki67 staining (Supplementary Fig. S3D). This indicated that tumor-initiating CD133⁺ cells had tumorigenic properties that were sustained with serial transplantation in animals.

Five hundred CD133⁺ cells were implanted in the tail of the pancreas in 20 mice and tumorigenesis was compared with 5,000 unsorted KPC cells. Groups of 5 to 7 mice were sacrificed at days 10, 20, and 40 after implantation and tumor development was assessed upon necropsy. Mice with CD133⁺ cells developed visible and palpable tumors by day 20 whereas the mice with 5,000 unsorted KPC cells showed a delayed tumor growth (Fig. 3C). By day 40 of implantation, tumors derived from CD133⁺ cells had an average volume of ∼900 mm³, whereas tumors derived from unsorted KPC cells had an average volume of ∼300 mm³. The experiment was terminated at day 40 owing to morbidity in the CD133⁺ group. CD133⁺ mice also showed increased ascites volume (0.978 ± 0.234 mL) compared with unsorted KPC cells (0.098 ± 0.05 mL), indicating increased aggressiveness of the CD133⁺ derived tumor (Supplementary Fig. S3E). Histology of the pancreas in the 2 groups at day 20 and day 40 after tumor implantation showed early onset of tumor in CD133⁺ inoculated animals compared to the unsorted KPC cells (Fig. 3D). These data indicated that CD133⁺ cells are responsible for early tumor formation resulting in increased accumulation of ascitic fluid.

We next analyzed the expression of proinvasion and survival genes in the CD133⁺ population of KPC cells. Antiapoptotic and prosurvival genes Bcl-2 and Survivin were 4- to 5-fold higher in the CD133⁺ population compared with the CD133⁻ population (Fig. 4A). Similarly 3- to 4-fold higher expression of TIC-specific genes such as Nanog (4-fold) and Sox2 (2-fold) were also observed in CD133⁺ population versus the CD133⁻ population (Fig. 4B). CD133⁺ cells also showed ∼4-fold increased expression of CXCR4 compared with CD133⁻ cells. Heat shock proteins are overexpressed in pancreatic cancer allowing increased survival. HSP70, GRP78, HSP27, HSF1 were found to be at least 3- to 4-fold higher in CD133⁺ cells compared with CD133⁻ cells (Fig. 4C). This increase in mRNA expression corresponded to an increase in protein expression of GRP78, HSP70, and Bcl-2 in CD133⁺ cells versus CD133⁻, unsorted KPC or normal pancreatic ductal cells (Fig. 4E). Both GRP78 and HSP70 (Supplementary Fig. S4A–S4C) showed increased staining in immunohistochemistry CD133⁺ cell–derived tumor.

A gene expression profile of CD133⁺ cells as compared with CD133⁻ cells was performed using an Illumina microarray on RNA isolated from these cells. One of the genes overexpressed in CD133⁺ subset was IDH1, a metabolic enzyme that has been as a potential oncogene in a number of cancers (35, 36). Increased expression of IDH1 in CD133+ was confirmed by quantitative real-time (qRT)-PCR (Supplementary Fig. S5).

Because NF-κB signaling pathway regulates survival, invasion, and apoptosis in tumor cells, we analyzed DNA binding activity of CD133⁺ cells and compared it to the CD133⁻ cells. CD133⁺ cells had increased (∼2.5-fold) NF-κB DNA binding activity compared with CD133⁻ cells (Fig. 4D).

To study if CD133⁺ cells were resistant to other cytotoxic chemotherapeutic agents such as paclitaxel, 5FU or current standard of care gemcitabine, KPC cells were sorted for CD133 and both positive and negative cells were treated with different concentration of 5FU, paclitaxel, triptolide, and gemcitabine. 5FU and paclitaxel reduced viability of CD133⁻ cells but did not have any effect on CD133⁺ cells. Gemcitabine did not affect either positive or negative population at a lower dose and reduced the viability of CD133⁻ cells at a higher concentration. Interestingly, triptolide was seen to reduce viability of CD133⁺ (0.4+/0.02) similar to CD133⁻ cells (0.32 ± 0.04; Fig. 5A).

To test the effect of Minnelide on CD133⁺-derived tumors, CD133⁺ cells were implanted in flanks of C57BL/6 mice. Treatment with therapeutic dose of Minnelide (0.42 mg/kg body weight) was started once implanted tumors reached a volume of 300 mm³. Tumor progression as well as endpoint measurements of tumor volumes of treated tumors showed ∼60% decrease in tumor volume in minnelide-treated tumors compared with untreated ones (Fig. 5B). Minnelide-treated tumors also showed decreased number of Ki-67 positive (∼30/field; Supplementary Fig. S6A–S6C) indicating decreased proliferation and increased TUNEL staining (∼60/field; Supplementary Fig. S6D–S6F) indicating apoptosis.

To test the efficacy of Minnelide in human patient tumor–derived xenograft, we sorted CD133⁺ cells from a human tumor and generated subcutaneous tumors on the flanks of NOD/SCID mice (n = 12 tumors). Treatment with Minnelide (0.42 mg/kg body weight) was started when the tumors reached ∼1,000 mm³ in volume. Tumor regression (871 ± 102 mm²) was apparent after 2 weeks of Minnelide treatment (Supplementary Fig. S6G).

Minnelide reduced the expression of the TIC markers in the CD133⁺ cell–generated tumors as studied by flow cytometry. Minnelide-treated tumors were seen to express less CD133 (∼4%) compared with the untreated tumors (∼8%; Fig. 5C). A similar decrease (1.2% ± 0.32 compared with 3.5 ± 1.2) was observed in the expression of CD24⁺/CD44⁺/ESA⁺ and in Aldh1 activity (0.3 ± 0.02 in Minnelide compared with 1.43 ± 0.78 in untreated) in tumors treated with Minnelide as compared with control (data not shown).

To see if similar effects of this drug were obtained in primary KPC tumor–bearing mice, we analyzed the primary tumors treated with minnelide. Our analysis showed that minnelide treated animals has 3% to 4% CD133⁺ cells compared with the untreated animals that had 6% to 9% CD133⁺ cells (Fig. 5C).

Minnelide also decreased the expression of genes involved in Hedgehog signaling (3%), Notch signaling, stemness such as Nanog (19% of untreated), ALDH1 (2%), CXCR4 (23%) prosurvival genes such as HSP70 (26%) and GRP78 (16%) and antiapoptotic genes such as Bcl2 (8%) and Survivin (21%; Supplementary Fig. S6H and S6I) compared with saline-treated animals. In additional, NF-κB activity was decreased in minnelide-treated tumors (∼500 RLU/mg protein) compared with untreated (2,670 RLU/mg protein; Fig. 5D). Thus, minnelide seemed to downregulate pro-proliferative pathways and decrease the CD133⁺ tumor-initiating population in pancreatic cancer cells.

Similarly, Minnelide also decreased expression of the genes involved in the Hedgehog signaling (9% of untreated), Notch signaling (18% of untreated), and stemness genes (21% to 24% of untreated) in KPC animals bearing primary tumors. Consistent with our results obtained from implanted cells in C57BL/6 mice, minnelide also decreased expression of HSP70 (40% of untreated), GRP78 (32% of untreated), and other antiapoptotic genes such as Survivin (41% of untreated) and Bcl-2 (26% of untreated). Along with this, minnelide also downregulated NF-κB activity in the primary tumors (876 RLU/mg protein) compared with untreated tumor (3712RLU/mg protein; Fig. 5D).

TICs are a population of quiescent cells within a tumor that evade apoptosis and cell death induced by standard chemotherapy, often resulting in tumor recurrence. As a result, there is a need to develop therapy that can reduce this population and contribute to tumor regression. In pancreatic cancer, TICs have been isolated using a number of markers, including surface expression of CD133 (2, 3, 5, 10, 32).

However, all of these studies have been performed in either immunocompromised, athymic nude mice, or in SCID mice. Although transplantation into these mice is the best way of assessing tumorigenicity, such studies often cannot address the extent to which these cells might be positively or negatively regulated by the actual physiology of animal. As development of solid tumors is a complex phenomenon involving intricate interaction with the host system, use of a syngenic model is a closer representation to actual tumor development of PDAC from tumor initiating cells. In this study, we report isolation and characterization of CD133-expressing TIC from a cell line derived from the KPC genetic mouse model for pancreatic cancer in a syngenic host.

As reported by other groups, genetically engineered KPC mice are the closest clinical representation of pancreatic adenocarcinoma (20). Our study revealed that 6% to 9% of the cells in primary pancreatic tumors from KPC mice as well as in the cell lines isolated from the KPC tumors are CD133⁺. The CD133⁺ population was also responsible for initiation of a fresh tumor in both subcutaneous and orthotopic models (Fig. 3). The CD133⁺ cells showed early tumorigenesis and increased invasiveness in both these models. Particularly, in the orthotopic mouse model, CD133-derived tumors showed increased accumulation of abdominal ascites compared with unsorted KPC cell–derived tumor-bearing animals (Supplementary Fig. S3E). The CD133⁺ cells derived from the KPC model formed tumors from as low as 10 cells, which reflected their increased tumorigenicity when compared with that of CD133⁺ cells isolated from human tumors (that formed tumors from a minimum of 100 cells). In all these studies, the CD133⁻ population did not form tumors in almost twice the amount of time.

The CD133⁺ population in KPC cells also included a smaller subset of Aldh⁺ population, previously described as a TIC marker in SCID mice (Fig. 2C and D; ref. 8). Furthermore, the CD133⁺ population also contained an independent subset of CD24⁺/CD44⁺/ESA⁺ cells, shown previously in human pancreatic cancer xenografts as a population of TICs (Fig. 2B; ref. 2).

Although pancreatic TICs have been well characterized, there has been limited study on the dysregulation of molecular pathways in these cell populations. Biological function of CD133 is currently unclear. Our study revealed that in addition to an expected increase in expression of Bcl2 and Survivin genes, the CD133⁺ population had increased expression of heat shock proteins; HSP27, HSP70, and GRP78. Previous studies from our group have shown that HSP70 is overexpressed in pancreatic cancer cells and downregulation of HSP70 results in loss of viability in PDAC cells (24, 25).

HSP70 thus functions as an inhibitor of apoptosis in pancreatic cancer. Similarly, HSP27 has been shown to have an increased expression in breast cancer stem cells (37). GRP78, a heat shock protein family that is a gatekeeper of ER stress in eukaryotic cells, has been found to be overexpressed in PDAC and involved in its cell survival. GRP78 was found to be crucial for embryonic stem cell precursor survival and also highly expressed in hematopoietic stem cells (38).

In addition, GRP78 has been shown to be upregulated in TICs in other cancers (39, 40). Our studies show an increased expression of HSP27, HSP70, and GRP78 in CD133⁺ cells isolated from the tumor, as well as in immunohistochemical studies on tumors derived from these cells (Supplementary Fig. S4A–S4C). This may be of significant clinical relevance as therapy targeted toward these proteins can be used for complete eradication of pancreatic TICs.

Along with overexpression of heat shock proteins, the prosurvival NF-κB pathway was also overexpressed in CD133⁺ KPC cells (Fig. 4D). The NF-κB pathway has been reported to also regulate multiple phenomena in TICs from other cancers (41). In the KPC tumor–derived pancreatic cancer cell line, increased expression of NF-κB in the CD133⁺ TICs indicate the role of this pathway in promoting tumorigenesis. Minnelide also reduced the expression of genes that define tumor initiating cells like Hedgehog and Notch signaling pathway genes in CD133⁺-derived tumors (Supplementary Fig. S6H and S6I).

As expression of CD133 in different cell lines correlated with increased invasion, we studied the expression level of EMT genes such as SNAI1, SLUG, Twist, and Vimentin. CD133⁺ cells showed ∼5- to 10-fold increased expression of SLUG and Twist genes whereas almost 50- to 80-fold increased expression of SNAI1 and Vimentin genes compared with CD133⁻ cells (Supplementary Fig. S1D).

Overexpression of the IDH1 gene was seen in the gene expression microarray for CD133⁺ cells, as compared with CD133⁻ (Supplementary Fig. S5). However, there is no report of IDH1 overexpression in this disease. IDH1 is a NADP⁺-dependent isocitrate dehydrogenase, which catalyzes the oxidative decarboxylation of isocitrate to α ketoglutarate. Overexpression of IDH1 in CD133⁺ cells of pancreatic cancer may indicate increased metabolic activities in TICs.

It is now well-established standard chemotherapy targets rapidly proliferating cells as a result of which the quiescent TICs evade the cytotoxic effect of the drug. Our studies showed that CD133⁺ cells were indeed very slow proliferating compared with the CD133⁻ cells (Supplementary Fig. S1E). Our lab has previously demonstrated that minnelide treatment prevents tumor recurrence and causes tumor regression in a number of mouse models (26). In this study, minnelide was effective in decreasing CD133⁺ population as well. Our in vitro study showed that the CD133⁺ population responded equally to triptolide (the active compound) as the CD133⁻ population, although both populations remained relatively resistant to gemcitabine, the current first line of therapy for pancreatic cancer. Our study also showed that cytotoxic agents 5FU and paclitaxel decreased the viability of CD133⁻ cells whereas CD133⁺ cells were unaffected (Fig. 5A).

Consistent with our in vitro studies, a 50% to 60% decrease tumor volume within the Minnelide-treated group was observed in our animal studies along with almost 50% decrease in the expression of stem cell marker CD133⁺ (Fig. 5C). Similar regression was also observed on tumors derived from CD133⁺ population of the human tumors (Supplementary Fig. S6G). Analysis of the CD133⁺-derived tumors showed a decrease in tumor volume and less proliferation in minnelide-treated animals and increased cell death by the drug (Fig. 5A and Supplementary Fig. S6). Treatment with Minnelide also showed a similar decrease in expression of prosurvival genes (HSP70, GRP78, Bcl-2, Survivin) in tumors (Supplementary Fig. S6H and S6I). NF-κB activity, which was upregulated in CD133⁺ population, was decreased in minnelide-treated tumors (Fig. 5D). Effect of other compounds targeting TICs have been studied in pancreatic cancer (Supplementary Table S2). However, all of these previous studies have been done in immunocompromised models. This study is the first report showing the effectiveness of minnelide in an immunocompetent syngenic system.

Our study on TIC in a syngenic animal system show that cells derived from KPC tumors have a high percentage of CD133⁺ cells forming tumors from very low cell numbers. This study shows that minnelide regresses CD133⁺-derived tumor volume and reduces the number of TICs in tumor. It also downregulates all the prosurvival proteins overexpressed in these cells. The observation that minnelide can reduce the TIC population strengthens our earlier observation that it can prevent recurrence of pancreatic tumor. This study is the first report of pancreatic TIC characterization in an immunocompetent syngenic system as minnelide is being evaluated in a phase I clinical trial.

R. Chugh has ownership interest in a patent. Selwyn Vickers has ownership interest (including patents) in Minneamrita. A.K. Saluja has ownership interest (including patents) in Minneamrita therapeutics and is a consultant/advisory board member for Minneamrita Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Banerjee, V. Dudeja, S.M. Vickers, A.K. Saluja

Development of methodology: S. Banerjee, V. Sangwan, V. Dudeja, S.M. Vickers

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Banerjee, A. Nomura, V. Sangwan, R. Chugh, A.K. Saluja

Analysis and interpretation of data (e.g., statistical analysis, biostatist\ics, computational analysis): S. Banerjee, A. Nomura, V. Sangwan, R. Chugh, V. Dudeja, S.M. Vickers, A.K. Saluja

Writing, review, and/or revision of the manuscript: S. Banerjee, A. Nomura, V. Sangwan, R. Chugh, V. Dudeja, A.K. Saluja

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Dudeja, S.M. Vickers, A.K. Saluja

Study supervision: A.K. Saluja

This study was funded by NIH grants R01-CA170946 and CA124723 (to A.K. Saluja).

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.