Targeting ALDH^bright Human Carcinoma–Initiating Cells with ALDH1A1-Specific CD8⁺ T Cells

This article is featured in Highlights of This Issue, p. 6107

Cancer-initiating cells (CIC) are characterized as a subpopulation of tumor cells in tumors which exhibit “stem cell–like” properties such as self-renewal, chemo- and radioresistance, and high tumorigenicity at low cell numbers in immunodeficient mice (1–5). Therefore, they are considered responsible for tumor recurrence and metastasis. In several types of tumors, cell populations enriched for cancer-initiating activity are being readily identified and isolated by flow cytometric analysis using the ALDEFLUOR reagent on the basis of their high level of aldehyde dehydrogenase (ALDH) activity and can be referred to as ALDH^bright cells (6–12). The ALDH activity detected by this reagent is primarily attributed to members of the ALDH1 and ALDH3 family of ALDH isoforms. The ability to readily identify and isolate ALDH^bright cells by flow cytometry is facilitating the efforts to develop therapeutic approaches that would target CICs and elicit long-term and effective responses in subjects with cancer (13). ALDH1-targeted immunotherapy represents such an approach, as in a previous study we have shown that the ALDH1A1 isoform can mediate the recognition and lysis of ALDH1A1⁺ squamous cell carcinoma of the head and neck (SCCHN) cell lines by cognate CD8⁺ CTLs. Relevant to the potential clinical use of ALDH1A1-specific CTL-based immunotherapy, ALDH1A1-specific CTLs recognize neither normal differentiated cells, such as fibroblasts, unless they are transfected to express high levels of ALDH1A1, nor normal CD34⁺ hematopoietic stem cells. The latter express ALDH1A1 at a level that is higher than that found in most normal differentiated cells and tissues but lower than that in CICs (14).

In the present study, we have investigated the ability of in vitro generated ALDH1A1-specific CTLs to eliminate ALDH^bright cells present in HLA-A2⁺ human carcinoma cell lines, xenografts, and surgically removed lesions in vitro and the antitumor activity of adoptive immunotherapy with ALDH1A1-specific CTLs in vivo. In addition, we have analyzed the expression of ALDH1A1 and HLA class I antigen (Ag) in normal liver, as normal hepatocytes have been reported to express a high level of this ALDH1 isoform (15) and therefore represent a potential concern in implementing ALDH1A1-based immunotherapy.

The human SCCHN cell lines used in these studies have been previously described (14). The MDA-MB-231 breast and MIA PaCa-2 pancreatic carcinoma cell lines were obtained from American Type Culture Collection. The luciferase-transfected MDA-MB-231-Luc cell line was obtained from Xenogen. KT-64 feeder cells were generously supplied by Dr. Bruce Levine (University of Pennsylvania, Philadelphia, PA; refs. 16–18). Cell lines were maintained in RPMI-1640 medium supplemented with 10% (v/v) FBS, 2 mmol/L l-glutamine, 50 μg/mL streptomycin, and 50 IU/mL penicillin (Life Technologies, Inc.). Tumor and blood specimens were obtained from consented subjects with SCCHN and pancreatic cancer under the auspices of the University of Pittsburgh Tissue Bank with Institutional Review Board (IRB) approval #991206 and Massachusetts General Hospital IRB approval #08-265, respectively. Blood was obtained from HLA-A2⁺ normal donors with IRB approval #980633.

ALDH1A1-specific rabbit monoclonal antibody (mAb; catalogue no. ab52492) was purchased from Abcam. The HLA-A, -B, -C, -E, -F, -G Ag-specific mAb W6/32 (IgG_2a), HLA-A2, -A28 Ag-specific mAb KS1 (IgG₁), HLA-DR Ag-specific mAb L243 (IgG_2a), HLA-A heavy chain–specific mAb HC-2A (IgG₁), and HLA-B, -C heavy chain–specific mAb HC-10 (IgG_2a) have been previously described (19–23). APC anti-CD5, FITC anti-CD8, ECD anti-CD45RA, and PC7 anti-CCR7 mAb were purchased from BD Biosciences. Rabbit anti-histone H3 phosphoserine10 mAb was purchased from Cell Signaling Technology, Inc. The ApopTag Plus Peroxidase In Situ Apoptosis Detection Kit was purchased from Millipore Corp.

Tumor cell lines were harvested with 1 mmol/L EDTA (Sigma), and xenografts and lesions were disaggregated with Collagenase Type 4 (Worthington Biochemical). Duplicate aliquots of tumor cell samples were incubated with ALDEFLUOR (Stem Cell Technologies), with or without the ALDH inhibitor diethylaminobenzaldehyde (DEAB; control) according to the manufacturer's instructions (14). To identify ALDH⁺ and ALDH^bright cells, the control aliquot of the sample was analyzed by flow cytometry and set for detection of 0.2% or more ALDH⁺ and 0% ALDH^bright cells in the aliquot. Using this cutoff value, the test aliquot was analyzed to identify its ALDH⁺/ALDH^bright cell content. The results for human tumor cell lines or lesion samples can vary depending on in vitro propagation of cell lines, lesion disaggregation conditions, and/or reagent lot. Cells were sorted with a Dako-Cytomation MoFlo sorter (Dako North America) at 1.5 × 10³ events/s.

Cells were surface stained for HLA class I Ag-specific mAb by standard procedures. Flow cytometry was carried out with an FC500 cytometer (Beckman Coulter), which was calibrated daily with fluorescent beads; all samples were run by identical settings to collect a minimum of 10,000 gated events, when possible. Analyses were conducted with EXPO32 ADC software (Beckman Coulter) or Summit V4.3 (Dako).

Expression of ALDH1 isoform mRNA relative to that of β-glucuronidase (an endogenous control or housekeeping gene) mRNA was determined with commercially available and custom-designed ALDH1 isoform primer and probe sets and the Applied Biosystems 7700 Sequence Detection Instrument as previously described (14). The following primers/probe sets were used to measure ALDH1A1 mRNA: forward 5′-cgcaagacaggcttttcag-3′, reverse 5′-tgtataatagtcgccccctctc-3′, probe 5′-FAM-attggatccccgtggcgtactatggat-3′; and ALDH1A2 mRNA, forward 5′-agctttgtgctgtggcaata-3′, reverse 5′-gatgagggctcccatgtaga-3′, probe 5′-FAM-taagccagcagagcaaacaccactcag-3′. Applied Biosystems TaqMan Gene Expression Assay systems Hs00167476_m1 and Mm03003537_s1 were used to measure ALDH1A3 mRNA and β-glucuronidase mRNA, respectively.

ALDH^bright and ALDH^neg cells sorted from tumor cell lines were collected in 2 mL RPMI-1640 medium with 20% FBS and irradiated (300 Gy) bulk parental tumor cells, centrifuged, and the supernatant saved for later use. The pellets were suspended in a predefined volume of the saved supernatant and equal volume of Matrigel (BD Biosciences) so that a 100μL aliquot contained 500 sorted ALDH^bright or ALDH^neg tumor cells and 1 × 10⁴ irradiated carrier/feeder cells. These aliquots were injected subcutaneously in the right and left flanks or intraperitoneally, respectively, in groups of NOD.CB17-Prck^scid/J [NOD/severe combined immunodeficient mice (SCID); The Jackson Laboratories] female (6–8 weeks of age) mice each. The tumorigenicity of MDA-MB-231-Luc cells was monitored by bioluminescence imaging with a Xenogen IVIS 50 instrument (Xenogen) according to the manufacturer's recommended protocol at the University of Pittsburgh Cancer Institute In Vivo Imaging Facility.

HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells were induced/expanded by in vitro stimulation (IVS) of CD8⁺ T cells isolated from peripheral blood obtained from normal HLA-A2⁺ donors with either ALDH1A1_88–96 peptide–pulsed autologous dendritic cells and OKT-3 mAb-activated KT64 feeder cells (the ratio of CD8⁺ T cells:dendritic cells:KT64 cells being 2:1:2) or ALDH1A1_88–96 peptide–pulsed artificial antigen-presenting cells (16–18). The yields of effector cells using artificial antigen-presenting cells as stimulators was 3-fold greater than that using peptide-pulsed dendritic cells and feeder cells and more than 10-fold greater than the use of peptide-pulsed dendritic cells only (data not shown). CD8⁺ T cells obtained from HLA-A2⁺ normal donors and IVS with the HLA-A2–restricted, HIVgag_362–370 peptide were used as controls in adoptive therapy experiments. Peripheral blood of HLA-A2⁺ normal donors and patients with SCCHN as well as IVS cultures was analyzed for HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells by flow cytometry using phycoerythrin-conjugated HLA-A2/ALDH1A1_88–96 peptide tetramer complexes obtained from the NIH Tetramer Facility as previously described (24).

Enzyme-linked immunosorbent spot (ELISPOT) INFγ assays were conducted as previously described (14), using the ELISPOT 4.14.3 analyzer (Zeiss). Values were considered significantly different from control values based on the double permutation test. Assay performance and reproducibility were monitored with aliquots of cryopreserved peripheral blood mononuclear cells (PBMC) obtained from a single donor stimulated with phorbol-12-myristate-13-acetate (10 ng/mL) and ionomycin (250 ng/mL; Sigma). The coefficient of variation for the assay was 15% (n = 50). For mAb blocking experiments, target cells were preincubated with either the blocking mAb or an isotype-matched mAb (10 μg/mL) for 30 minutes at room temperature.

Tumor cell lines, disaggregated xenografts, and lesions (5 × 10⁵ cells) and HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells [2.5:1 effector/target (E/T) cell ratio] were incubated for 4 hours at 37°C, centrifuged, trypsinized, washed, incubated with ALDEFLUOR, and analyzed for ALDH⁺ and ALDH^bright cells by flow cytometry. For mAb blocking experiments, target cells were preincubated with mAb (10 μg/mL) for 30 minutes at room temperature.

Three distinct types of adoptive therapy experiments involving xenograft-bearing immunodeficient C.B-17 scid female mice (Taconic Farms) and HLA-A2–restricted, ALDH1A1-specific CD8⁺ T cells were carried out in this study. In a fixed endpoint experiment, C.B-17 scid mice (n = 15) were challenged with surgically implanted 5-mm pieces of a serial passage PCI-13–derived xenograft. Seven days later, the mice were placed by a stratified randomization based on tumor burden into 3 groups of 5 mice each and adoptive therapy initiated biweekly by intravenous injection with HLA-A2–restricted, ALDH1A1-specific CD8⁺ T cells, irrelevant HIVgag_362–370-specific CD8⁺ T cells (2 × 10⁶/mouse) or left untreated. Tumor volumes (mm³) were calculated by the following formula: (a × b²)/2, where b is the smaller of the 2 diameter measurements (25). The experiment was terminated approximately 21 days later, the mice were euthanized, and xenografts were removed for analyses.

In the second model, experimental lung metastases of MDA-MB-231 cells were established in C.B-17 scid mice (n = 9) following intravenous injection of 1 × 10⁶ cells to each mouse. On day 3, mice were randomly divided into 3 groups of 3 mice each. Group 1 was injected intravenously with HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells (2 × 10⁶/mouse) twice per week for 5 weeks and then once per week for 4 weeks; group 2 was injected intravenously with irrelevant HIVgag_362–370-specific CD8⁺ T cells (2 × 10⁶/mouse) following the same schedule; and group 3 received no CD8⁺ T cells. All 3 groups received PEG-IL2 (equivalent of 6.6 × 10⁴ IU/mouse) by intraperitoneal injection twice on the day CD8⁺ T cells were administered (26). All the mice were euthanized on day 65 and their lungs were harvested and fixed in 10% formalin for further analysis.

The third model used was a postsurgery and metastases survival model with a survival endpoint. NODscid female mice (n = 27) were challenged in the mammary fat pad with 1 × 10⁶ MDA-MB-231 cells, and 30 days later, when the tumors were, on average, 0.8 cm in diameter, the xenografts were surgically removed. The mice were randomized into 3 groups of 9 mice each, 2 of which received weekly intravenous injections of either HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells or irrelevant HIVgag_362–370 peptide–specific CD8⁺ T cells (2 × 10⁶/mouse). All 3 groups of mice received PEG-IL2 (equivalent of 6.6 × 10⁴ IU/mouse) by intraperitoneal injection twice on the day CD8⁺ T cells were administered.

A liver tissue microarray (catalogue no. BN03011; US Biomax) was stained and analyzed for HLA class I Ag and ALDH1A1 expression in normal liver hepatocytes. The HC-10 and HC-A2 mAbs were used to stain for HLA class I Ag, and ALDH1A1 was detected with the rabbit anti-ALDH1A1 mAb by standard procedures.

Tumor areas in experimentally induced pulmonary metastases were analyzed with 4-μm thick formalin-fixed paraffin-embedded sections of lung tissues stained with 0.5% alcoholic solution of hematoxylin (Sigma-Aldrich, Inc.). Photographs were taken with a Nikon Eclipse E800 microscope, and the areas of tumor nodules in 5 randomly selected fields per section (magnification 200×) were measured and calculated by the SPOT Advanced Imaging software (Diagnostic Instruments, Inc.). For analysis of proliferative and apoptotic cells in untreated and treated xenografts, formalin-fixed paraffin-embedded sections of xenografts were stained for histone H3 phosphoserine10⁺ and terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling⁺ (TUNEL⁺) cells and analyzed by standard procedures. A minimum of 5 sections from each xenograft was stained, and 5 microscopic fields per section were counted manually in a double-blinded fashion by board-certified pathologists.

Two-tailed Student's t test was carried out to interpret the differences between experimental groups. Kaplan–Meier analysis was used to calculate significance of median survival in the adoptive therapy in the postsurgery and metastases xenograft experiment.

Using flow cytometry and the ALDEFLUOR reagent, established human tumor cell lines and digests of tumor xenografts or surgically removed lesions were analyzed for their ALDH⁺ cell content. Sorting by flow cytometry was used to isolate the ALDH^bright cell population from the cell lines. The ALDH^bright cells identified in these samples had an ALDH mean fluorescence intensity (MFI) twice that of the bulk ALDH⁺ cell population. A representative flow cytometric analysis of the human SCCHN cell line PCI-13 was done to identify ALDH⁺ and ALDH^bright cells in the cell line and set the parameters for sorting ALDH^neg, ALDH⁺, and ALDH^bright cells, together with the reanalysis of the sorted populations for ALDH⁺ cells and ALDH MFI, is shown in Figure 1A. Sorted ALDH⁺ PCI-13 cells were found to contain only 67% ALDH⁺ cells, whereas the sorted ALDH^bright cell population had a purity of 94%. The sorted ALDH^neg cells contained less than 1% ALDH⁺ cells and no ALDH^bright cells.

A panel of established human breast, pancreatic, and SCCHN cell lines and digests of disaggregated surgically removed pancreatic and SCCHN lesions were then analyzed by flow cytometry to identify their ALDH⁺ and ALDH^bright cell contents. Representative analyses are shown in Figure 1B, and the results are listed in Supplementary Table S1. In summary, the data indicate that the percentages of ALDH⁺ and ALDH^bright cells varied with each sample regardless of its tumor type or source. High percentages of ALDH⁺ cells in a sample did not automatically correlate with high percentages of ALDH^bright cells, but, on average, the ALDH^bright cell content was about 10% of the ALDH⁺ cell content. The ALDH^bright cell content ranged from a low of 0.02% in the SCCHN PCI-30 cell line to a high of 4.6% in the MIA PaCa-2 pancreatic carcinoma cell line–derived xenograft. The percentages of ALDH^bright cells in the MDA-MB-231 and MIA PaCa-2–derived xenografts were higher than those in the parental cell lines, whereas the ALDH^bright cell content of the PCI-13–derived xenograft was lower than that of the parental cell line. Additional passages of the xenografts did not result in significant changes in their ALDH^bright cell content (data not shown).

To confirm that the ALDH^bright cell population was highly tumorigenic, a critical characteristic of CICs, ALDH^bright cells sorted from the PCI-13, MIA PaCa-2, and MDA-MB-231-Luc cell lines were tested for their tumorigenicity by challenging groups of 3 or 5 immunodeficient mice each at a dosage of 500 cells. Xenografts were established in 3 of 3 mice challenged with ALDH^bright PCI-13 cells, 2 of 3 mice challenged with ALDH^bright MIA PaCa-2 cells, and 4 of 5 mice challenged with ALDH^bright MDA-MB-231-Luc cells within 6 months of challenge (Supplementary Fig. S1). The tumorigenicity of the ALDH^bright MDA-MB-231-Luc cells was monitored by bioluminescence imaging. None of the ALDH^bright cell–derived xenografts can be attributed to the irradiated tumor feeder cells used in the inoculums, as ALDH^neg challenges, which also included the same number of irradiated tumor feeder cells, failed to yield xenografts in the same mice.

According to its manufacturer, ALDH activity detected by the ALDEFLUOR reagent can be attributed to ALDH1 and ALDH 3 isoforms, with the emphasis on ALDH1 isoforms. Four ALDH1/3 isoforms have been identified, ALDH1A1, -1A2, -1A3, and ALDH3A1. A real-time reverse transcriptase PCR (qRT-PCR) analysis of the levels of expression of these 4 isoform mRNAs in bulk PCI-13 cells indicated predominant expression of ALDH1A1 mRNA compared with the other 3 isoform mRNAs. Little to no ALDH1A2 mRNA was expressed, and the level of ALDH1A1 mRNA was approximately 50 times greater than that of ALDH1A3 and ALDH3A1 mRNA, a finding consistent with ALDH1A1 contributing to the ALDH activity detected by ALDEFLUOR. Furthermore, the analysis indicated that ALDH^bright cells expressed approximately 8-fold higher level of ALDH1A1 RNA than bulk population of tumor cells (48.2 ± 5.6 vs. 6.2 ± 0.3). This result correlates well with the nearly 10 times higher ALDH MFI of ALDH^bright PCI-13 cells than that of the ALDH^neg PCI-13 cells. In addition, qRT-PCR analysis of the sorted ALDH^bright populations indicated that ALDH^bright PCI-13, MDA-MB-231, and MIA PaCa-2 cells uniformly express higher levels of ALDH1A1 mRNA than ALDH1A3 mRNA (Supplementary Table S2). Essentially no ALDH1A2 mRNA was detected in these cells (data not shown).

We previously identified ALDH1A1 as a tumor-associated antigen defined by HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells based on the ability of the ALDH1A1_88–96 peptide to stimulate the in vitro induction/generation of these effector cells from PBMCs obtained from most normal donors, as well as a subject with SCCHN (14). In vivo, the immunogenicity of ALDH1A1_88–96 epitope was confirmed in a limited HLA-A2/ALDH1A1_88–96 peptide tetramer–based flow cytometric analysis of PBMCs obtained from HLA-A2⁺ subjects with SCCHN and normal donors. Representative results of this analysis are shown in Supplementary Figure S2. On the basis of a cutoff frequency of 1/8,000 determined with PBMCs obtained from HLA-A2^neg normal donors, the frequency of tetramer⁺ cells detected in PBMCs of HLA-A2⁺ normal donors was comparable with that of negative controls. In contrast, relatively high frequencies of tetramer⁺ cells in the range of 1/500 to 1/2,000 were detected in the peripheral circulation of subjects with SCCHN. The CCR7/CD45RA phenotypes of the tetramer⁺ CD8⁺ T cells varied with each subject; they had a predominately memory and terminally differentiated phenotype in subject 1, a predominately naive phenotype in subject 2, and a mixture of naive and terminally differentiated phenotypes in subject 3.

The specificity of the HLA-A2–restricted, ALDH1A1-specific CD8⁺ T cells used in this study for the ALDH1A1_88–96 peptide in ELISPOT IFNγ assays is shown in Supplementary Figure S3A. These effectors recognize ALDH^bright target cells but not the bulk cell population or ALDH^neg target cells sorted from the HLA-A2⁺ PCI-13 SCCHN cell line (Fig. 2A). Recognition of the ALDH^bright cells by the ALDH1A1_88–96 peptide–specific CD8⁺ T cells was blocked by the HLA-A2–, HLA-28–specific mAb KS1 but was not affected by the HLA-DR–specific mAb L243. In an HLA-A2–restricted manner, these effector T cells also recognize ALDH^bright target cells, but neither bulk cell population nor ALDH^neg target cells sorted from the HLA-A2⁺ basal breast carcinoma MDA-MB-231 and pancreatic carcinoma MIA PaCa-2 cell lines (Supplementary Fig. S3B and C).

In vitro recognition by HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells of ALDH⁺/ALDH^bright cells present in the established human carcinoma cell lines was also measured in a cell-mediated cytotoxicity (CMC) assay using flow cytometry. As indicated in Figure 2B, 83% and 70% decreases in the percentages of ALDH^bright and ALDH⁺ PCI-13 cells were observed following incubation of the tumor cells with the effectors, which can be attributed to the differential levels of ALDH1A1 expression in these cells. Recognition of ALDH^bright/ALDH⁺ cells was blocked by the HLA-A2, -28 Ag-specific mAb KS1. Importantly, comparable results were also obtained with cells derived from in vivo propagated tumor cells, namely, a PCI-13–derived xenograft and an HLA-A2⁺ surgically removed SCCHN lesion (Table 1). These results further confirm that HLA-A2⁺ ALDH^bright tumor cells are recognized by HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells.

The efficacy of adoptive therapy with ALDH1A1_88–96 peptide–specific CD8⁺ T cells was evaluated in immunodeficient mice bearing either subcutaneous xenografts derived from SCCHN PCI-13 cells or experimental or spontaneous pulmonary metastases derived from basal breast carcinoma MDA-MB-231 cells. In a fixed time point experiment involving immunodeficient mice bearing subcutaneous PCI-13–derived xenografts, adoptive therapy with intravenous ALDH1A1-specific CTLs was administered. The experiment was terminated on day 21 postimplantation to obtain sufficient residual xenograft specimens for subsequent analyses of their ALDH^bright cell content, as well as proliferation and apoptotic indices, measured by staining for histone H3 phosphoserine10⁺ cells (27, 28) and TUNEL⁺ cells, respectively. The results indicate that treatment of the xenografts with ALDH1A1_88–96 peptide–specific CD8⁺ T cells, but not with irrelevant CD8⁺ T cells, significantly inhibited their growth (Fig. 3A). This inhibition was concordant with a significant decrease in both their ALDH^bright cell content and proliferative index but a significant increase in the apoptotic index compared with xenografts obtained from the control groups of mice (Table 2 and Supplementary Fig. S4).

MDA-MB-231 cells readily form pulmonary metastases following intravenous injection or spontaneously following surgical removal of primary orthotopic xenografts. Adoptive therapy of mice bearing experimentally induced MDA-MB-231 pulmonary metastases by systemic administration of ALDH1A1-specific CD8⁺ T cells resulted in fewer and smaller tumor nodules with a significantly reduced total tumor area in the lungs of mice compared with that of the control groups of mice [irrelevant CTLs + interleukin (IL-2) and IL-2; Fig. 3B]. The total tumor area in the lungs of mice was quantified to determine the efficacy of the treatment in this experiment, because the metastatic lesions in the lungs of the control groups of mice had grown so extensively, had fused to form large tumor masses, and were not individually discernible (Fig. 3C; ref. 29).

In the clinically relevant, postsurgery and metastasis survival model, mice succumb primarily to lung metastases following surgical removal of the primary MDA-MB-231–derived orthotopic xenograft. Groups of immunodeficient mice were treated with ALDH1A1-specific CD8⁺ T cells + IL-2, irrelevant CD8⁺ T cells + IL-2, or IL-2 only following their surgery. Only adoptive therapy with ALDH1A1-specific CTLs significantly prolonged their survival (P < 0.001) compared with the control groups of mice, as shown in Figure 4. All mice in the 2 control groups died from lung metastases by day 87 postsurgery, whereas 80% of the ALDH1A1-specific CD8⁺ T cells–treated mice exhibited no signs of disease at day 210 postsurgery. The results of these in vivo human tumor xenograft experiments show the efficacy of adoptive therapy with ALDH1A1_88–96 peptide–specific CD8⁺ T cells to effectively target ALDH^bright cells and control tumor growth and metastases.

In view of the potential clinical application of these results, we sought to address the question of whether ALDH1A1-based immunotherapy could target normal liver hepatocytes, which are reported to express a high level of ALDH1A1, and cause deleterious side effects (15). Therefore, we tested for the expression of HLA class I Ag and ALDH1A1 at the protein level in normal liver tissue by immunohistochemical staining with mAb. A liver tissue microarray composed of 63 cores derived from 3 nondiseased and 16 diseased livers (e.g., cirrhosis and fatty degeneration), 3 hepatocellular carcinomas, and an abnormal spleen was analyzed: 23 cores were considered to be normal liver tissue, 9 of which came from 3 nondiseased livers. Only 1 of these 9 cores stained for HLA class I Ag, and it showed weakly patchy staining. The remaining 14 “normal liver tissue” cores came from nondiseased regions of diseased livers; 2 showed strong cytoplasmic HLA class I Ag but weak ALDH1A1 expression, whereas 2 others showed HLA class I Ag membrane staining but no ALDH1A1 expression (see Supplementary Table S3 and Fig. S5). In contrast, HLA class I Ag expression was prevalent in multiple cores of diseased tissue taken from diseased livers.

The results of this translational preclinical study show that a subset of tumor cells in human carcinomas identified as ALDH^bright cells are recognized and eliminated in vitro and in vivo by HLA-A2–restricted, ALDH1A1_88–96 peptide–specific CD8⁺ T cells. In human tumor xenograft models, we have shown that adoptive transfer of ALDH1A1-specific CD8⁺ T cells inhibited growth of subcutaneously growing xenografts and experiment-induced lung metastases. In addition, following surgery to remove a primary tumor, this therapy inhibited spontaneous metastases and prolonged survival of mice.

It was recently reported that the ALDH activity in breast cancer stem cells detected by ALDEFLUOR is primarily due to ALDH1A3, rather than ALDH1A1, expression (30). In our study, however, we have shown that sorted ALDH^bright cells express higher levels of ALDH1A1 mRNA than ALDH1A3 mRNA. Nonetheless, as previously reported (14), the ALDH1A-specific CD8⁺ T cells used in this study recognize the ALDH1A1_88–96 peptide (LLYKLADLI) but not the highly related peptides derived from the ALDH1A2 (LLDKLADLV) and ALDH1A3 (LLHQLADLV) isoforms. Therefore, regardless of which ALDH1 isoform is prevalently expressed, the recognition of ALDH^bright cells by the ALDH1A_88–96 peptide–specific CD8⁺ T cells used in this study is independent of ALDH1A3 expression.

Although ALDH1A1 is expressed by many cell types, it is highly unlikely that ALDH1A1-based immunotherapy would induce toxicity. Normal stem cells such as hematopoietic stem cells, which express ALDH1A1 but at a lower level than detected in tumors, have been shown not to be recognized by ALDH1A1-specific CD8⁺ T cells (14). Furthermore, although ALDH1A1 is expressed by normal hepatocytes, in agreement with the information in the literature, we have shown that these cells express little to no HLA class I Ag on their cell surface; as a result, normal hepatocytes are highly unlikely to be recognized by HLA class I–restricted, ALDH1A1-specific CD8⁺ T cells (31–33).

Presently, there is little information about the recognition of CICs by HLA class I–restricted, CD8⁺ T cell effectors. To the best of our knowledge, only 3 studies have investigated this subject: 2 involve glioblastoma multiforme (GBM) stem cells isolated under selective culture conditions and the third involves sorted colon cancer stem cells identified as a side-staining population. Using a nontumor-related cytomegalovirus (CMV) antigen as a model tumor antigen (TA), Brown and colleagues (34) showed recognition of CMV-transfected GBM stem cells by CMV pp65 peptide–specific CTLs. Recognition of the targets, however, required targets pulsed with exogenous CMV pp65 peptide. This finding suggests that GBM stem cells expressed HLA class I Ag but required the exogenous peptide to form a sufficient level of HLA class I Ag–peptide complexes for recognition by the cognate CTLs. Di Tomaso and colleagues (35) detected defects in HLA class I Ag and APM component expression in the cultured population of GBM stem cells. As a result, recognition of these target cells by autologous antitumor CTLs required pretreatment with IFNγ to upregulate HLA class I Ag expression and, presumably, HLA class I Ag/TA peptide complexes, a common situation observed in targeting tumor cells with HLA class I–restricted, TA peptide–specific T cell effectors (36, 37). In the third study, Inoda and colleagues (38) showed that colon carcinoma stem cells are sensitive in vitro and in vivo to HLA class I–restricted CTLs recognizing an epitope derived from the tumor-associated centrosomal protein, 55-kDa protein CEP55, which is expressed by the tumor-initiating cells as well as the bulk population of cells in the colon carcinoma cell lines studied. Because ALDH1A1 is expressed by CICs present in colon carcinomas and gliomas (10, 39), targeting CIC populations in these tumors with ALDH1A1-specific CD8⁺ T cells is also possible and should be more selective.

Our results strongly support further development of strategies that would incorporate ALDH1A1-based immunotherapy to target CICs. The constraints of a practical evaluation of a T-cell–based immunotherapy using human xenograft mouse models required adoptive transfer of the immune effector cells. Using recombinant DNA or optimized traditional protocols, sufficient numbers of TA-specific T cells can be generated in vitro for adoptive T-cell–based immunotherapy; this strategy has been shown in recent years to yield beneficial clinical responses in subjects with cancer (40, 41). Nonetheless, the development and implantation of ALDH1A1-based immunotherapy need not preclude a vaccine-based approach.

In accordance with the cancer stem cell theory, the elimination of CICs should be the critical criteria used to define the efficacy of a therapy rather than only reduction in tumor volume. Our research shows for the first time the potential ability of an immunotherapy to achieve the objective of targeting CICs in tumors. Furthermore, because therapeutic protocols can promote tumor escape, our findings highlight the potential benefit that combining T-cell–based immunotherapy with other independent therapeutic modalities, such as TA-specific mAb and/or inhibitors of aberrantly regulated stem cell signaling pathways (13), could have on minimizing tumor escape.

No potential conflicts of interest were disclosed.

This work was supported by National Cancer Institute at the NIH grants DE12321 and CA109688 (to C. Visus, A.B. DeLeo, and T.L. Whiteside), P50 CA097190 (to C. Visus, A.B. DeLeo, and R.L. Ferris), and CA138188 (to Y. Wang and S. Ferrone), the Hillman Foundation (to A.B. DeLeo), Hirshberg Foundation for Pancreatic Cancer Research (to C. Visus), DOD Concept Award BC085485 (to C. Visus, X. Wang, and A.B. DeLeo), the Elsa U. Pardee Foundation (to X. Wang), RO3 CA141086 (to C.R. Ferrone and X. Wang), and the Pennsylvania Department of Health (to A.B. DeLeo and S. Ferrone), which specifically disclaims responsibility for any analyses, interpretations, or conclusions detailed in this report.

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.