In all, 44 patients with histologically proven pNETs underwent pancreatic resection at the Tokyo Medical and Dental University Hospital between 2001 and 2013. Of them, 15 patients who underwent surgery from 2012 to 2013 were recruited for Aldeflour assay, 31 cases were randomly selected for immunohistochemical analysis. Written informed consent was obtained from the patients, and our institutional review board approved this study (no. 1080).

Human pNET cell line QGP1 was obtained from the Health Science Research Resources Bank (Osaka, Japan). A murine insulinoma cell line MIN6 was provided by Professor J. Miyazaki (Osaka University, Japan) (15). Both cell lines were used in the present experiments within 10 passages after reception. All cell cultures were routinely tested to rule out mycoplasma infection using PlasmoTest kit (InvivoGen, San Diego, CA, USA). QGP1 cells were cultured in RPMI-1640 medium (Sigma-Aldrich, St. Louis, MO, USA), supplemented with 10% fetal bovine serum (Biological Industries, Beit Haemek, Israel) and 1% Pen/Strep (Gibco, Grand Island, NE, USA) as antibiotics. MIN6 cells were cultured in Dulbecco's modified Eagle's medium containing 25 mmol/l glucose (DMEM; Sigma-Aldrich), supplemented with 10% fetal bovine serum, 1% Pen/Strep and β-mercaptoethanol (2ME) (5 ml/l, Wako, Osaka, Japan). All cell lines were cultivated in a humidified incubator at 37°C in 5% CO₂, and were collected with 0.25% trypsin-ethylene diamine tetraacetic acid (EDTA) (Gibco).

Luciferase expression plasmid pGL4.50 [luc2/CMV/Hygro] (Promega, Madison, WI, USA) was transfected into QGP1 cells according to the manufacturer's instructions, and luciferase-expressing QGP1 cells (QGP1-Luc) were generated as described previously (16).

A portion of a resected tumor was harvested under sterile conditions in a pathology suite and placed on ice in DMEM/F12 (Gibco) containing 10% fetal bovine serum, care being taken to avoid contaminating adjacent normal tissue. Each specimen was mechanically dissociated with sterile scalpels, digested for 1 h with 10 mg/ml type IV collagenase (Sigma-Aldrich) in fresh DMEM/F12 medium containing 10% fetal bovine serum, and then filtered through sterile 100-μm membranes to obtain single-cell suspensions. Red blood cells in the surgical samples were lysed using a red blood cell lysis buffer (Miltenyi Biotec, Bergisch Gladbach, Germany). Cultured cells were collected with Accutase (BD Bioscience, San Jose, CA, USA) as single cells. An Aldefluor kit (Stemcell Technologies, Vancouver, Canada) was used to isolate the population of cells with high ALDH enzymatic activity. Cells were suspended in Aldefluor assay buffer containing ALDH substrate (BAAA, 1 μmol/l) and incubated at 37°C for 30 min. As a negative control, an aliquot of each sample was treated with 50 mmol/l diethyl-aminobenzaldehyde (DEAB), a specific ALDH inhibitor. The fluorescently labeled product, BODIPY-aminoacetate, was produced by cells expressing the ALDH enzyme, and ALDH^high cells were quantified and purified using FACS Aria II (BD Biosciences).

The spheroid assay was performed as described previously (16–18). After FACS sorting, ALDH^high or bulk cells were plated separately at a density of 300 cells on low attachment plates (96-well Ultra Low Cluster Plate; Costar, Corning, New York, NY, USA), and then incubated in serum-free DMEM/F12 medium (n=20 in each). To investigate sphere formation with the CD73 inhibitor, after FACS sorting, 5×10⁴ ALDH^high cells were seeded into two 6-cm dishes. After 24 h, PBS or 12 μM α,β-methylene adenosine 5′-diphosphate (APCP) (Sigma-Aldrich) was added to each dish. After 48 h, the medium was changed to drug-free medium, followed by incubation for 24 h. Using trypan blue exclusion, the remaining viable cells were collected and plated separately at 300 cells per low attachment plates, and then incubated in serum-free medium (n=20 in each). Sphere formation was observed using an AxioObserver (Carl Zeiss, Oberkochen, Germany), and images were acquired digitally using AxioVision software (Carl Zeiss). Experiments were independently performed in triplicate.

Hypoxic treatment was performed as described previously (18). PNET cell lines were seeded into 96-well plates at 3×10³ cells per well. After 24 h, the cells were exposed to hypoxic conditions (1% O₂, 5% CO₂, and 94% N₂) in an anaerobic workstation (Hirasawa Works, Tokyo, Japan). The oxygen concentration inside the workstation was constantly monitored with an oxygen sensor (MC-8G-S, Iijima Electrics, Gamagori, Japan) and maintained at 1% during the experiment. After 2, 4 and 6 days, the number of living cells was measured by the MTS assay using a Cell Titer 96 AQueous One Solution Cell Proliferation Assay kit (Promega), according to the manufacturer's instructions. The absorbance was read at 490 nm, with 630 nm as the reference wavelength, using a multiwall plate reader (Model 550, Bio-Rad, Hercules, CA, USA), with wells containing medium but no cells serving as blank controls (16,19). The experiments were independently performed in triplicate.

After FACS sorting, pNET cells, ALDH^high or bulk, were seeded into 96-well plates at 3×10³ cells per well. The cell layer was scratched, a 200-μl pipette tip being run the full length of each well. The well was washed once with growth medium to remove cell debris. Images were taken at the time of initial wounding as well as 24, 48 and 72 h post-wounding using an INCell Analyzer 2000 (GE Healthcare, Waukesha, WI, USA). The wound area was determined using INCell Developer Toolbox software (GE Healthcare). We compared the ratio of the area at 72 h after scratching to that at the time of initial wounding. Experiments were independently performed in triplicate.

Female NOD.CB17-PRkdc^Scid/J mice, 16–19 g and aged 4–6 weeks, were purchased from Charles River Laboratory Inc. (Kanagawa, Japan). The care and use of animals were performed in accordance with institutional guidelines. Various amounts of cells (ALDH^high and bulk, from 10¹ to 10⁴ cells each) were mixed with Matrigel (BD Biosciences), and then aliquots were injected subcutaneously into both flanks of mice under isoflurane anesthesia. Tumor formation was then monitored once a week. Cancer initiation frequency was calculated using L-Calc software (Stem Cell Technologies) (20) and significance was determined by Chi-square analysis using ELDA (The Walter and Eliza Hall Institute of Medical Research) (20).

The peritoneal metastatic potential of cancer cells was assessed as reported (21). Briefly, 10⁵ ALDH^high pNET cells or unsorted control cells were injected intraperitoneally into mice (n=4 mice per group). The mice were monitored using a luciferase-luciferin-based imaging IVIS system (Xenogen, Alameda, CA, USA) under isoflurane anesthesia. Images of the mice were taken every week for up to 14 weeks, and then the mice were euthanized by cervical dislocation. Any mice with ascites formation or a loss of body weight of >25% were euthanized. Animal survival data were entered to the Kaplan-Meier Life Table format and presented as a cumulative survival plot. Statistical differences were analyzed by means of log-rank test. For the drug sensitivity assay, 10⁴ freshly sorted ALDH^high cells were used. The mice were randomly divided into 2 groups for the experiments. APCP (400 μg/tumor) or PBS was injected into the tumor twice a week after tumor cell injection. Tumor formation was monitored once a week. The tumor volume was estimated using the following equation: volume = (length) × (width)²/2. All in vivo procedures were approved by the Animal Care Committee of Tokyo Medical and Dental University.

Total RNA was extracted from QGP1 freshly sorted ALDH^high cells and unsorted control cells using a RNeasy kit (Qiagen, Hilden, Germany), and the integrity of the obtained RNA was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). All samples had an RNA Integrity Number of >7.0. Anti-sense RNA was prepared from 100 ng of total RNA using a 3′ IVT Express kit (Affymetrix, Santa Clara, CA, USA). Hybridization and signal detection for HG-U133 Plus 2.0 arrays (Affymetrix) were performed according to the manufacturer's instructions. The microarray datasets for ALDH^high and unsorted control QGP1 cells were normalized simultaneously using the robust multiarray average method found in R statistical software (v. 2.12.1) together with the Bioconductor package. The estimated gene-expression levels were obtained as log2-transformed values. To reveal functional relationships among genes differentially expressed in ALDH^high QGP1 cells, a protein interaction network was analyzed. Genes upregulated >1.3-fold between ALDH^high and unsorted control QGP1 cells were included in the network. Protein interaction data obtained from BIND (http://bond.unleashedinformatics.com), Bio-GRID (http://thebiogrid.org), and HPRD (http://www.hprd.org) were downloaded from the ftp site of the National Center for Biotechnology Information (NCBI; ftp://ftp.ncbi.nih.gov/gene/GeneRIF/interactions.gz). The protein interaction network was analyzed using Cytoscape software.

The cells were sorted using the FACSAria II and were incubated for 24 h on glass slides. After fixation with 10% trichloroacetic acid for 15 min at 4°C, the cells were incubated in permeabilization buffer (0.2% Triton-PBS) for 5 min at room temperature. After incubation in a blocking buffer, 3% bovine serum albumin (BSA)-PBS, for 1 h, the slides were incubated with the primary antibodies, CD73 (IE9; 1:50; Santa Cruz, Dallas, TX, USA) for QGP1 or CD73 (C-20; 1:50; Santa Cruz) for MIN6 for another hour. The secondary antibody, Alexa Fluor 647 tetramethyl rhodamine isothiocyanate-conjugated goat anti-rabbit IgG (1:200; Sigma-Aldrich), was diluted in 3% BSA-PBS, and then incubated with the cells for 30 min. The Hoechst 33342 solution was added for nuclear staining. After mounting, the slides were observed under a fluorescent microscope, AxioObserver (Carl Zeiss).

Immunohistochemical analysis was performed on pNET tissue samples. For the tissue analysis, 31 samples were available, which were stained using the anti-CD73 primary antibody (IE9; Santa Cruz) at 1:50 dilution with PBS followed by reactions in an automated immunostainer (Ventana XT System; Ventana, Roche, Basel, Switzerland) using a standard DAB detection kit (Ventana, Roche). The immunostaining was evaluated quantitatively by counting ≥500 cells in three different random fields under a light microscope. Tissue samples with >5% of strong staining in the pNET tumor cells were diagnosed as positive, and the others were diagnosed as negative immunohistochemically. The immunostaining was evaluated under a light microscope by two independent investigators.

Statistical comparisons of clinicopathological characteristics for significance were performed using a Chi-square test or Fisher's exact test with a single degree of freedom, and Student's t-test was used to determine the differences between continuous values. p-values of <0.05 were considered to be statistically significant. All statistical analyses were performed using SPSS version 17.0 (SPSS, Chicago, IL, USA).

A total of 15 pNET specimens (7 primary lesions, 7 liver metastatic lesions and 1 lymph node metastatic lesion) were analyzed by Aldefluor assay. Flow cytometry was used to isolate viable single ALDH^high cells from pNET specimens, with ALDH expression greater than that of the top 0.1% of DEAB-treated negative control cells (Fig. 1A). The ALDH^high cells were observed in pNET surgical sections, the mean proportion of the subpopulation being 2.1±1.7% (range, 0.2–4.6%).

We also found a small population of cells expressing ALDH^high in pNET cell lines, QGP1 (human somatostatinoma) and MIN6 (murine insulinoma). Flow cytometric analysis revealed that ALDH^high cells represented from 1 to 4% of the population in both pNET cell lines, i.e., QGP1 and MIN6 (Fig. 1B).

In order to investigate whether or not ALDH^high cells were characterized as CSCs, sphere-forming ability was evaluated by sphere formation assay. The results showed that ALDH^high cells formed spheres in contrast to unsorted cells for both QGP1 (Fig. 2A and C; p<0.001) and MIN6 (Fig. 2B and D; p<0.001).

Since the pluripotent potential in embryonic stem cells is efficiently maintained under low oxygen levels (22), and hypoxia contributes to CSCs maintenance (23), the effects of hypoxic treatment on ALDH^high and unsorted pNET cells were evaluated utilizing QGP1. Cell proliferation was 39% reduced under hypoxic conditions (1% O₂) compared to normoxic conditions for control cells, in contrast, no significant difference in proliferation was observed between under hypoxic and normoxic conditions for ALDH^high cells (Fig. 2E). These results are consistent with the reports showing that hypoxic conditions serve as a stimulus for reprograming cells towards normal stem cells and CSCs (22,23).

To determine whether or not ALDH^high cells promote cell motility, scar migration assay was performed for QGP1. Cell motility was quantified by measuring the scratch wound closure as a percentage. There was a 20% greater reduction in wound size for ALDH^high cells than for control unsorted cells (Fig. 2F and G; p=0.006).

Different numbers of QGP1 ALDH^high or unsorted cells were injected subcutaneously into non-obese diabetic/severe combined immunodeficient (NOD/SCID) mice in numbers ranging from 10¹ to 10⁴ cells per injection. The results showed ALDH^high pNET cells had higher tumorigenicity than unsorted cells. Surprisingly, for a very small number of ALDH^high cells, as few as 10¹ cells, subcutaneous tumor formation was observed (Fig. 3A and Table I). The cancer initiation frequency was 1 in 83 (95% CI, 33–211) for ALDH^high pNET cells and 1 in 720 (95% CI, 292–1,777) for unsorted cells (p=0.002). These results suggested that CSCs are enriched in the ALDH^high subpopulation.

Furthermore, to investigate whether or not the pNET cells establish in vivo metastasis, ALDH^high or unsorted pNET cells were injected intraperitoneally into a NOD/SCID mouse utilizing QGP1. Peritoneal metastasis was assessed with IVIS imaging system. ALDH^high cells (10⁵) were sufficient to establish metastasis in all transplanted mice, in contrast, no established metastasis was observed in mice transplanted with 10⁵ unsorted cells (Fig. 3B and C).

Comprehensive gene expression analysis of ALDH^high and unsorted control QGP1 cells was performed to acquire a CSC gene profile. The microarray data have been deposited in the Gene Expression Omnibus (GEO) database under accession number GSE62079. The gene expression changes in 185 probe sets were detected by microarray analysis (fold-change value >2 between ALDH^high pNET and control cells). Genes associated with mesenchymal stem cells, including CD73 (5′-nucleotidase, ecto) and CD109, and epithelial-mesenchymal transition (EMT), including caveolin-1 (CAV1) and SLUG (SNAI2), were overexpressed in ALDH^high cells.

A protein interaction network was then constructed using 4,438 probe sets with ≥1.3-fold overexpression in ALDH^high pNET cells. To more closely investigate molecular networks associated with CSCs, a sub-network of 3-hop neighbors from the cell adhesion molecule fibronectin 1 (FN1) genes was generated, including CD73 (Fig. 4), which is 12.332-fold overexperessed in ALDH^high cells compared to the unsorted control cells. On immunocytochemical analysis of QGP1 and MIN6, CD73 overexpression was observed in ALDH^high cells (Fig. 5).

We evaluated the efficacy of CD73 inhibition in ALDH^high cells of pNETs using APCP, which is a known small molecule inhibitor of CD73 in vitro and in vivo (2425). ALDH^high cells treated with or without APCP showed no significant difference in proliferation (data was not shown). To determine whether or not APCP inhibits the CSC properties of ALDH^high cells, sphere-forming and scar migration assay were performed for QGP1 cells. The results showed that the number of spheres was 62.2% decreased (Fig. 6A and B; p=0.002) and cell motility was 39% reduced (Fig. 6C and D; p=0.027) on exposure to APCP in ALDH^high cells.

In order to evaluate the in vivo efficacy of APCP, peritumoral injection of APCP was performed twice a week into the QGP1 ALDH^high subcutaneous murine xenograft model. As shown in Fig. 6E and F, ≤43% tumor growth inhibition was observed in the group of mice treated with APCP compared with in the control group on day 49 (p=0.003). None of the APCP-treated mice showed signs of wasting or other toxicity compared to control mice. APCP was tolerated at the dose at which antitumor efficacy was observed.

Immunohistochemical analysis of tissue samples from 31 patients with pNETs was performed to explore the clinical significance of CD73 expression. As shown in Fig. 7, the CD73 protein was mainly distributed on the cytoplasm of pNET cells. In cases of liver metastasis and peritoneal dissemination, CD73 overexpression was observed (Fig. 7C and D). Consequently, specific overexpression of CD73 was observed in 17 out of 31 cases of primary lesions (54.8%) (Fig. 7A and B). The clinicopathological significance of CD73 expression was evaluated in the CD73-high expression group (n=17), and compared to the low expression group (n=14) of patients with pNETs (Table II). CD73 expression was significantly correlated with invasion into adjacent organs (35% vs 0%; p=0.024).