BE(2)-C (CRL-2268) cells were obtained from the American Type Culture Collection (Manassas, VA, USA). Human neuroblastoma NBTT1, NBTT2D and NBTT3 cells were previously described (22,23). Tumor tissue samples were obtained from high-risk neuroblastoma patients with written informed consent. The use of human tissues for this study was approved by the Ethics Committee at Kobe University Graduate School of Medicine and conducted in accordance with the Guidelines for the Clinical Research of Kobe University Graduate School of Medicine.

The rabbit anti-PGK1 antibody was purchased from Sigma (St. Louis, MO, USA), rabbit anti-NF-M antibody from Millipore (Billerica, MA, USA), rabbit anti-ALDH1A2 antibody from Atlas antibodies (Stockholm, Sweden) and rabbit anti-NANOG antibody from ReproCELL (Tokyo, Japan).

The N-terminal 3xFLAG-tagged expression plasmid with IRES-driven GFP and puromycin markers (pRS-3FLAG-IRES-GFP) was constructed. Briefly, the U6 promoter expression unit of pRS vector (Origene, Rockville, MD, USA) was first replaced with the CMV-promoter expression unit of pCMV6-AC-IRES-GFP vector (Origene) using In-Fusion HD Cloning kit (Takara, Otsu, Japan). The 3xFLAG-tag was then inserted into the SgfI site within multiple cloning sites of the resulting plasmid. The full-length ALDH1A2 cDNA (NM_003888) was amplified by PCR using PrimeStar GXL DNA polymerase (Takara), cloned into pRS-3FLAG-IRES-GFP and sequenced using an ABI PRISM 3100 genetic analyzer (Applied Biosystems, Foster City, CA, USA). Scramble and specific short hairpin RNA (shRNA) expression plasmids (pGFP-V-RS-scramble shRNA, pGFP-V-RS-ALDH1A2 shRNA, pGFP-V-RS-ALDH1L1 shRNA and pGFP-V-RS-ALDH3B2 shRNA) were obtained from Origene and their sequences are listed in Table I.

Total RNA from neuroblastoma cells and spheres was isolated with a TRIzol Plus RNA purification kit (Invitrogen, Carlsbad, CA, USA) and reverse transcribed using a QuantiTect Reverse Transcription kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s instructions. Real-time RT-PCR was performed as described previously (22). Primer sequences are listed in Table II.

Parental cells and spheres of BE(2)-C, NBTT1, NBTT2D and NBTT3 cells were cultured as described previously (24). BE(2)-C cells were transfected with expression plasmids using Lipofectamine 2000 transfection reagent (Invitrogen) according to the manufacturer’s instructions. Stably transfected GFP-positive cells were selected by 2.0–3.0 μg/ml puromycin (Invivogen, San Diego, CA, USA) and further isolated by using MoFlo XDP (Beckman Coulter, Brea, CA, USA).

ALDH activity was determined using the Aldefluor kit (Stem Cell Technologies, Durham, NC, USA) according to the manufacturer’s instructions. Briefly, neuroblastoma cells and spheres were dissociated with Accumax (Innovative Cell Technologies, San Diego, CA, USA) and suspended at ~1×10⁶ cells/ml in Aldefluor assay buffer containing BODIPY-aminoacetaldehyde (BAAA) in the presence or absence of 15 μM diethylaminobenzaldehyde (DEAB), incubated at 37°C for 40 min and then treated with 1 μg/ml propidium iodide (PI; Sigma). Flow cytometric analysis was performed using MoFlo XDP. Specific ALDH activity was based on the difference between the presence and absence of DEAB.

Sphere formation was analyzed as described previously (24). For colony formation, BE(2)-C cells expressing the indicated shRNA or cDNA were mixed in 0.325% SeaPlaque agarose (Lonza, Rockland, ME, USA) in DMEM/Ham’s F12 (3:1) (Wako Pure Chemical, Osaka, Japan) with 10% FBS and plated at 1,000 cells/well onto a solidified bottom layer of 0.6% SeaPlaque agarose in DMEM/Ham’s F12 with 10% FBS in a 6-well plate. Cells were incubated at 37°C for 21 days, stained with 0.5 mg/ml 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT; Sigma) and photographed. The total number of colonies was counted manually.

Four-week-old male athymic BALB/cAJcl nu/nu (nude) mice were obtained from CLEA (Shizuoka, Japan). All procedures involving animals were approved by the Animal Care and Use Committee of Kobe University Graduate School of Medicine and carried out in strict accordance with the Guidelines for the Care and Use of Laboratory Animals of Kobe University Graduate School of Medicine. BE(2)-C cells expressing the indicated shRNA or cDNA in 50% Matrigel (BD Biosciences, Franklin Lakes, NJ, USA) were injected subcutaneously into the flank of nude mice at a density of 1×10⁵ cells per injection site. Tumor growth was monitored 3 times per week by external caliper and tumor volume (V) was calculated by the formula: V = 1/2 (L × W²), where L and W were the greatest longitudinal and transverse diameters (25). Mice were dissected when the greatest diameter of tumor reached 20 mm. Xenograft tumors were fixed in 20% buffered neutral formalin solution (Muto Pure Chemicals, Tokyo, Japan) and embedded in paraffin. Tumor sample was sectioned 4-μm thick, deparaffinized in xylene, rehydrated in alcohol and stained with hematoxylin and eosin (H&E).

Immunohistochemical staining of neuroblastoma xenografts was performed on the 4-μm thick sections of tumor samples. After antigen retrieval using a conventional steamer with citrate buffer (pH 6.0), the section was immunostained with a primary antibody in REAL antibody diluent (Dako, Glostrup, Denmark). The section was then blocked with peroxidase blocking reagent (Dako) and incubated with EnVision labeled polymer peroxidase (Dako). The immune complex was visualized using 3, 3′-diaminoben-zidine (DAB; Dako) as a chromogen and hematoxylin as a counterstain. To quantify the average percentage of immunostained cells, three fields containing ≥300 neuroblastoma cells were randomly selected from each sample. Positive cells were identified microscopically as brown cytoplasmic staining and manually counted.

Phase-contrast images were acquired using a BZ-9000E fluorescence microscope (Keyence, Osaka, Japan). Western blotting was performed as described previously (26).

To begin to characterize ALDH in neuroblastoma, NBTT2D, NBTT1 and NBTT3 cells established from distinct high-risk neuroblastoma patients were grown as spheres in a serum-free non-adherent condition as described previously (24). The ALDH activities in spheres and parental cells were then determined by Aldefluor assay. Higher ALDH activity was constantly detected in spheres compared to parental cells (Fig. 1A and B). Because the ALDH isoform responsible for high ALDH activity measured by Aldefluor assay is specific for each cancer type (18), we next analyzed the fold-change of ALDH isoform expression in spheres compared to parental cells. Among all 19 ALDH isoforms expressed in human cells, ALDH1A2, ALDH1L1 and ALDH3B2 expression was consistently induced in spheres compared to parental cells (Fig. 1C). While the induction of ALDH1A2 expression was not so high in NBTT1 and NBTT3 cells, ALDH1A2 showed the best correlation between mRNA induction and enzymatic activity induction among ALDH1A2, ALDH1L1 and ALDH3B2 isoforms.

To gain insight into the function of ALDH isoforms induced in spheres, we first examined ALDH1A2, ALDH1L1 and ALDH3B2 mRNA expression in spheres and parental cells of neuroblastoma BE(2)-C cells. Like NBTT2D, NBTT1 and NBTT3 cells, BE(2)-C cells also showed the induction of expression of these ALDH isoforms in spheres (Fig. 2A). We then generated BE(2)-C cells stably expressing scramble, ALDH1A2, ALDH1L1 and ALDH3B2 shRNA. ALDH1A2 shRNA #2, ALDH1L1 shRNA #2 and ALDH3B2 shRNA #1 achieved effective knockdown and were used in the present study (Fig. 2B). Next, we performed the sphere and colony formation assays that are widely used to examine the CSC properties in vitro. In the sphere formation assay, cells are cultured in a serum-free non-adherent condition so that only cells with ability to self-renew will form spheres. Knockdown of ALDH1A2, ALDH1L1 and ALDH3B2 in BE(2)-C cells significantly impaired the sphere formation (Fig. 2C). In the colony formation assay, cells with a capacity of anchorage-independent growth can grow in soft agar and form colonies. The colony formation was also significantly impaired upon ALDH1A2, ALDH1L1 and ALDH3B2 knockdown (Fig. 2D). Among these ALDH isoforms, ALDH1A2 knockdown showed the most profound effect on both the sphere and colony formation.

To further characterize the function of ALDH isoforms induced in spheres, we next analyzed the correlation of ALDH1A2, ALDH1L1 and ALDH3B2 expression with overall survival probabilities of neuroblastoma patients. For this purpose, we used the bioinformatics program R2 (http://r2.amc.nl) and the NB88 dataset (Tumor Neuroblastoma-Versteeg-88-MAS5.0-u133p2) that consisted of 88 primary neuroblastoma tumors of all stages. High expression of both ALDH1A2 and ALDH3B2 was associated with low overall survival probabilities (Fig. 3A and C). In contrast, low ALDH1L1 expression tended to have low overall survival probabilities (Fig. 3B). Among ALDH1A2, ALDH1L1 and ALDH3B2 isoforms, only ALDH1A2 expression was significantly correlated with overall survival probabilities.

Based on the above results, we focused on ALDH1A2 isoform in the subsequent study. If ALDH1A2 were involved in the sphere and colony formation in neuroblastoma cells, ALDH1A2 overexpression would promote the sphere and colony formation. To test this possibility, we generated BE(2)-C cells stably expressing ALDH1A2 cDNA. ALDH1A2 overexpression was detected by western blotting (Fig. 4A). The sphere and colony formation were significantly promoted by ALDH1A2 cDNA expression (Fig. 4B and C). These results suggested that ALDH1A2 was involved in the sphere and colony formation in neuroblastoma cells.

As CSC is defined as a subpopulation of cancer cells that recapitulates their heterogeneous populations in xenograft tumors, we next examined the function of ALDH1A2 in xenograft tumors. To this end, we injected BE(2)-C cells stably expressing scramble shRNA, ALDH1A2 shRNA, control cDNA and ALDH1A2 cDNA into the flank of nude mice. All mice developed palpable tumors within 10–15 days and tumor volume was determined in 3–4 weeks. Tumor volume tended to decrease upon ALDH1A2 knockdown and to increase upon ALDH1A2 overexpression, albeit it was not statistically significant (Fig. 5A and B). We then performed the pathological examination with H&E staining. Scramble shRNA and control cDNA tumors showed typical characteristics of a small round cell tumor (Fig. 5C and D). Compared to scramble shRNA tumors, ALDH1A2 shRNA tumors contained more stromal structures and neuronal fibers (Fig. 5C, H&E). In contrast, stromal structures were scarcer in ALDH1A2 cDNA tumors than in control cDNA tumors (Fig. 5D, H&E). The tumors were further examined by immunostaining with antibodies against ALDH1A2, NANOG and NF-M. ALDH1A2-positive cells were reduced in ALDH1A2 shRNA tumors compared to scramble shRNA tumors, whereas they were increased in ALDH1A2 cDNA tumors compared to control cDNA tumors (Fig. 5C and D, ALDH1A2). ALDH1A2 shRNA tumors showed more differentiated phenotypes with NF-M-positive and NANOG-negative cells than scramble shRNA tumors, while ALDH1A2 cDNA tumors had more undifferentiated phenotypes with NANOG-positive and NF-M-negative cells than control cDNA tumors (Fig. 5C and D, NANOG and NF-M). These data suggested that ALDH1A2 was involved in the growth and undifferentiation of neuroblastoma xenografts.

Because CSC contributed to the chemoresistance in addition to the sphere, colony and tumor formation, we finally examined the function of ALDH1A2 in the resistance of neuroblastoma cells to 13-cis-RA, which is currently used in the maintenance therapy for high-risk neuroblastoma patients. As RA generally induces the growth arrest and differentiation of neuroblastoma cells, we first analyzed the growth arrest induced by 13-cis-RA treatment for 72 h in BE(2)-C cells expressing scramble shRNA, ALDH1A2 shRNA, control cDNA and ALDH1A2 cDNA. The 13-cis-RA-induced growth arrest was significantly promoted by ALDH1A2 knockdown and inhibited by ALDH1A2 overexpression (Fig. 6A and C). We then investigated their differentiation by phase-contrast microscopy. In scramble shRNA and control cDNA cells, the elongation of neurites started at ~48 h and became evident at ~72 h after 13-cis-RA treatment. Compared to scramble shRNA cells, ALDH1A2 shRNA cells showed more elongated neurites at 48 h after 13-cis-RA treatment (Fig. 6B). In contrast, ALDH1A2 cDNA cells did not elongate the neurites compared to control cDNA cells at 72 h after 13-cis-RA treatment (Fig. 6D). These results suggested that ALDH1A2 was involved in the resistance of neuroblastoma cells to 13-cis-RA.