The Antigen ASB4 on Cancer Stem Cells Serves as a Target for CTL Immunotherapy of Colorectal Cancer

Solid tumors are heterogeneous, consisting of a variety of cell types. One of these cell types, the cancer stem cells (CSC), or cancer-initiating cells, comprise a small subset of tumor cells and are responsible for tumorigenesis (1–3). The existence of CSCs was first reported in a hematologic tumor, and thereafter in a variety of solid tumors, including colon, breast, head and neck, uterine, brain, pancreas, and prostate cancers (4–10). Intratumoral heterogeneity and the presence of CSC subsets have been established in primary solid-cancer tissues (11). CSCs are resistant to conventional chemotherapy and radiotherapy, most likely due to their quiescent status. CSCs may also have mechanisms to activate drug transporters or DNA-damage checkpoint responses (12–14). Leukemic stem cells are resistant to a targeted agent, imatinib (15). With these attributes, CSCs may contribute to relapse or metastasis in clinical settings, increasing the demand for a therapy focused on CSCs.

Our group and others have proposed that cytotoxic CD8⁺ T lymphocytes (CTL) could be used for the immunotherapeutic targeting of CSCs. Indeed, CTL responses against CSCs have been demonstrated in the context of colon, kidney, cervical, brain, head and neck, and breast cancers (16–24). Thus, host CTLs may target CSCs, thereby protecting against tumor growth in vivo. Meanwhile, it remains unclear whether CSCs are an important therapeutic target. In some studies, CTLs responding to CSC antigens also recognized or lysed non-CSC counterparts, which account for the majority of cells in solid cancers. To disambiguate effects of CTLs on CSCs from effects on non-CSCs, we searched colorectal cancer cells for CSC-specific antigens that were naturally processed and elicited CTL responses only to CSCs, leaving non-CSCs intact.

In this study, we took advantage of a pair of cell lines that arose from side-population (SP) and main-population (MP) cells of colorectal cancer, which provided us with a consistent source of CSCs (25, 26). We then mapped the HLA-A24 peptide landscapes of the CSCs and non-CSCs with HLA ligandome analysis using mass spectrometry. The comparison between those two cellular subsets revealed that many HLA-A24 peptides were expressed by both cell types. A few peptides were identified only in CSCs. We further analyzed one of these peptides: ankyrin repeat and SOCS box protein 4 (ASB4). CSCs but not non-CSCs expressed the gene and the peptide. Here, we found that the CTLs responding to the ASB4 antigen discriminate CSCs from non-CSCs of colorectal cancer and controlled colorectal cancer growth in vivo. These data may suggest that control of CSCs is required to prevent colorectal cancer growth. The CSC-specific antigen ASB4 may be useful as a therapeutic target in immunotherapy against colorectal cancer.

Cell lines were maintained in RPMI 1640 or DMEM supplemented with 10% FBS and 1% antibiotics. The human cell lines used in this study were as follows: colon carcinoma (SW480, SW620, Colo205, HCT15, Colo320, HCT116, HT29, and CRC21), lung carcinoma (A549, SBC1, SBC5, and Lc817), kidney carcinoma (Caki-1 and ACHN), liver carcinoma (HepG2), breast carcinoma (MCF7 and MDA-MB-468), ovary carcinoma (ES2), prostate carcinoma (PC3), pancreas carcinoma (panc1), melanoma 1102-MEL, cervix carcinoma (HeLa), bladder carcinoma (UM-UC-3), osteosarcoma (U2OS), erythroleukemia (K562), and TAP-deficient T2 cell line stably expressing HLA-A*24:02 (T2-A24). SW480, SW620, Colo205, HCT15, Colo320, HCT116, HT29, A549, Caki-1, ACHN, HepG2, MCF7, MDA-MB-468, ES2, PC3, panc1, HeLa, UM-UC-3, U2OS, and K562 were purchased from ATCC. SBC1, SBC5, and Lc817 were purchased from the Japanese Cancer Research Resources Bank (JCRB, Osaka, Japan). 1102-MEL was a gift from Dr. F.M. Marincola (National Cancer Institute, Bethesda, MD). T2-A24 was a gift from Dr. K. Kuzushima (Aichi Cancer Center Research Institute). CRC21 was established in our lab. The cell lines were immediately frozen in batches on arrival, and the culture period of each batch was limited to a maximum of 8 weeks. Because the HCT-15 line lacks β 2-microglobulin (β2m) expression, we used HCT-15 cells stably expressing β2m throughout the whole study.

Hybridomas for anti-HLA-A24 (C7709A2, a gift from Dr. P.G. Coulie, Ludwig Institute for Cancer Research, Brussels, Belgium), pan HLA-class I (W6/32, ATCC), and anti–HLA-DR (L243, ATCC) were cultured in Hybridoma-SFM (Gibco) supplemented with 1% penicillin/streptomycin in CELLine bioreactor flasks (Corning). Produced mAbs were condensed and collected through a semipermeable membrane during cell culture.

The original protocol has been previously described (46, 47). Briefly, cells were labeled with 5 μg/mL Hoechst 33342 dye (Lonza) for 90 minutes in the presence or absence of 50 μmol/L verapamil (Sigma-Aldrich). Verapamil was used to inhibit ABCG2 activity. Approximately 1 × 10⁶ viable cells were then analyzed and sorted using a FACSAria II (BD Biosciences). The Hoechst dye was excited with the UV laser at 355 nm and its fluorescence was measured using a 450/20 nm band-pass filter (Hoechst blue) and a 670 nm long-pass filter (Hoechst red). Dead cells were labeled with 1 μg/mL propidium iodide and excluded from the analysis.

Cells were cultured in ultra-low attachment plates (Corning Life Sciences) in serum-free Dulbecco's Modified Eagle Medium/F12 medium (Life Technologies) supplemented with 20 ng/mL human recombinant epidermal growth factor, 10 ng/mL human recombinant basic fibroblast growth factor (R&D Systems), and 1% N2 supplement (Invitrogen). On day 7, we collected spheres or counted the number of spheres of which the diameter was >100 μm.

The peptides bound to HLA-A24 of SP-H and MP-A cells were isolated and sequenced using mass spectrometry. We used two different mass spectrometers, Q-Exactive Plus (Thermo Fisher Scientific) and 4800 Plus MALDI-TOF/TOF Analyzer (AB Sciex), and the detected peptides in three technical replicates each for SP-H and MP-A were collectively analyzed. The precise workflow and procedure have been previously described (27, 48). Cell lysates were prepared from approximately 1.0 × 10⁹ SP-H or MP-A cells with lysate buffer containing 0.6% CHAPS (Dojindo Molecular Technologies) and 1 × complete protease inhibitor (Roche) in PBS. The peptide–HLA-A24 complexes included in the samples were then isolated using affinity chromatography of an HLA-A24 specific C7709A2 mAb coupled to cyanogen bromide-activated Sepharose 4B (GE Healthcare). The HLA-bound peptides were then eluted with 0.2% TFA and purified using a 10 kDa ultra-centrifugal filter (Millipore), desalted using C18 ZipTip (Millipore), and condensed by vacuum centrifugation.

Samples were then loaded into a nano-flow UHPLC (Easy-nLC 1000 system, Thermo Fisher Scientific) online-coupled to an Orbitrap mass spectrometer equipped with a nanospray ion source (Q-Exactive Plus, Thermo Fisher Scientific). We separated the samples using a 75 μm × 20 cm capillary column with a particle size of 3 μm (NTCC-360, Nikkyo Technos) by applying a linear gradient ranging from 3% to 30% buffer B (100% acetonitrile and 0.1% formic acid) at a flow rate of 300 nL/minute for 80 minutes. In mass spectrometry analysis, survey scan spectra were acquired at a resolution of 70,000 at 200 m/z with a target value of 3e6 ions, ranging from 350 to 2,000 m/z with charge states between 1+ and 4+. We applied a data-dependent top 10 method, which generates high-energy collision dissociation (HCD) fragments for the 10 most intense precursor ions per survey scan. MS/MS resolution was 17,500 at 200 m/z with a target value of 1e5 ions.

For MS/MS data analysis, we used the Sequest HT (Thermo Fisher Scientific) and Mascot ver 2.5 (Matrix Science) algorithms embedded in the Proteome Discoverer 2.1 platform (Thermo Fisher Scientific), and the peak lists were searched against the UniProt human databases. The tolerance of precursor ions and fragment ions were set to 10 ppm and 0.02 Da, respectively. More than 80% of the peptides including IV9 listed in Supplementary Table S1 were detected with a false discovery rate (FDR) of 0.01; however, we applied a less stringent FDR of 0.05 as a threshold to avoid overlooking potential CSC antigens.

Total RNA was isolated from cancer cell lines and cancer tissues using an RNeasy Mini Kit (Qiagen) or an AllPrep DNA/RNA Mini Kit (Qiagen) according to the manufacturer's instructions. cDNA was synthesized from 2 μg of total RNA by reverse transcription with Superscript III and oligo(dT) primers (Life Technologies). cDNA from human fetal and adult tissues was purchased from Clontech and Bio Chain. RT-PCR mixtures were initially incubated at 94°C for 2 minutes, followed by 35 cycles of denaturation at 94°C for 15 seconds, annealing at 63°C for 30 seconds, and extension at 72°C for 30 seconds. Primer pairs were as follows: ASB4, 5′-CTGTCTTGTTTGGCCATGTG-3′ and 5′-GCGTCTCCTCATCTTGGTTG-3′ (product size 288 bp); G3PDH, 5′-ACCACAGTCCATGCCATCAC-3′ and 5′-TCCACCACCCTGTTGCTGTA-3′ (product size 452 bp).

Gene expression was also quantified using a StepOne Real-Time PCR System (Applied Biosystems) with PowerUp SYBR Green Master Mix (Thermo Fisher Scientific). An initial denaturation step of 95°C for 10 minutes was followed by 40 cycles of denaturation at 95°C for 15 seconds and annealing/extension at 60°C for 60 seconds. Primer pairs for qPCR were as follows: ASB4, 5′-CTGTCTTGTTTGGCCATGTG-3′ and 5′-GCGTCTCCTCATCTTGGTTG-3′; G3PDH, 5′-GGATTTGGTCGTATTGGG-3′ and 5′-GGAAGATGGTGATGGGATT-3′. Each sample was analyzed in triplicate and the threshold cycle values (Ct) of ASB4 were normalized according to those of G3PDH.

Synthetic peptides of the following purities were purchased from Sigma Genosis [IV9 (IYPPQFHKV), 91.7%; HIV_env584-594 (RYLRDQQLL), 81.5%; GK12 (GYISPYFINTSK), 89.7%]. Peptides in a range of the indicated concentrations were pulsed onto T2-A24 cells, incubated for 3 hours at 27°C, and then incubated for 2.5 hours at 37°C. Cells were incubated with C7709A2, followed by a secondary FITC-conjugated antibody, and then analyzed using flow cytometry. The difference in mean fluorescence intensity (ΔMFI) indicates the difference in MFI values between samples with and without the primary antibody.

We used the established method for CTL induction (49). Briefly, CD8⁺ cells were isolated from peripheral blood mononuclear cells (PBMC) of HLA-A*24:02-positive healthy donors or colorectal cancer patients using anti-CD8 coupled to magnetic microbeads (Miltenyi Biotec). Remaining CD8⁻ cells were PHA-activated and used as PHA blasts. CD8⁺ cells were cultured in AIM-V medium (Life Technologies) containing 10% human serum (kindly provided by Dr. Takamoto, Japanese Red Cross Hokkaido Block Blood Center), 20 U/mL human IL2 (kindly provided by Takeda Pharmaceutical), and 10 ng/mL IL7 (R&D Systems) for 4 weeks. The CD8⁺ cells were repeatedly stimulated with autologous PHA-blasts pulsed with 20 μmol/L of the IV9 peptide. To generate CD8⁺ T-cell clones, single cells binding both anti-CD8 (Beckman Coulter) and an IV9-HLA-A24 tetramer (MBL, Japan), as well as the single cells positive for CD8 but negative for binding to the tetramer, were isolated using FACS Aria II (BD Biosciences). They were expanded in AIM-V medium containing 100 U/mL IL2, 1 μg/mL PHA, and X-ray–irradiated PBMCs from healthy volunteers. The SM4 and control CTL clones were selected from the tetramer-positive and -negative clones, respectively.

Tetramer-positive CTLs were added to ELISPOT plates coated with antihuman IFNγ (BD Biosciences) at 5.0 × 10⁴ cells/mL per well. T2-A24 or the indicated cancer lines at 5.0 × 10⁴ was added to the corresponding wells. T2-A24 was preincubated at room temperature for 2 hours with 20 μmol/L of IV9 or irrelevant HIV peptide. After incubation in a 5% CO₂-incubator at 37°C for 24 hours, the wells were incubated with a biotinylated antihuman IFNγ antibody for 2 hours at room temperature, followed by the ELISPOT Streptavidin-HRP antibody for 1 hour. Spots were visualized using the ELISPOT AEC Substrate Set according to the manufacturer's instructions (BD Biosciences).

This analysis followed the original protocol with modifications (50). Peptide extracts of 2 × 10⁸ SP-H or MP-A were prepared by acid extraction using 10% formic acid in the presence of 2 μmol/L of an irrelevant peptide followed by filtration using a <10k Da cutoff spin-column (Amicon). The samples were fractionated using RP-HPLC equipped a 2.1 × 250 mm C18 column with a particle size of 2 μm (ZORBAX 300-SB C18, Agilent) by applying a linear gradient ranging from 23% to 45% buffer B for 45 minutes. Each fraction was then dried using vacuum centrifugation overnight. Each fraction was incubated with 5 × 10⁴ SM4 and 5 × 10⁴ T2-A24. The IFNγ produced by SM4 responding to its naturally processed epitope was detected using the ELISPOT assay.

The amount of lactate dehydrogenase (LDH) released from lysed target cells was measured using an LDH cytotoxicity detection kit according to the manufacturer's instructions (Takara Bio). Target cells (1.0 × 10⁴) were co-incubated with the indicated numbers (E/T ratio) of CTLs at 37°C for 6 hours. The percentage of LDH released (cytotoxicity) was calculated as follows: % LDH release = 100 × (experimental LDH release − spontaneous LDH release)/(maximal LDH release − spontaneous LDH release). LDH values from CTL alone and target cells treated with 2% Triton X-100 (Sigma-Aldrich) were used to estimate spontaneous and maximal LDH releases, respectively. In the HLA-blocking assay, target cells were preincubated for an hour with 100 μg/mL of anti-HLA-class I (W6/32), anti-HLA-2402 (C7709A2), or anti-HLA-DR (L243).

NSG mice were purchased from The Jackson Laboratory. The mice were maintained in the animal facility of Sapporo Medical University and all procedures were performed in accordance with the institutional animal care guidelines. To evaluate the tumorigenicity of SP-H and MP-A, NSG mice were subcutaneously injected with 1.0 × 10³ and 1.0 × 10⁴ of SP-H or MP-A cells. In tumor-rejection models, 1.0 × 10³ SW480 cells were subcutaneously injected, followed by the adoptive intravenous transfer of 5.0 × 10⁵ CTLs at the indicated time points. The major (x) and minor (y) axes of the tumors were routinely measured. Tumor volume was calculated as follows: volume = xy²/2. To assess CTL infiltration into tumors, NSG mice were subcutaneously injected with 1.0 × 10⁶ SW480 cells. Subsequently, the mice were intravenously injected with 1.0 × 10⁵ and 2.0 × 10⁶ CTLs, on days 28 and 29, respectively. The tumors and spleens were removed and fixed with 10% formalin on day 31. The paraffin-embedded tissues were immunohistochemically stained with a human CD8 antibody (DAKO). The numbers of CD8⁺ cells per HPF were manually counted.

The study was performed with approval of the Institutional Review Board of Sapporo Medical University. Clinical samples of patients with colorectal cancer were included in this study, with informed consent according to the guidelines of the Declaration of Helsinki. PBMCs were isolated using Lymphoprep (Nycomed) from whole blood samples of patients and healthy donors.

PBMCs from HLA-A24-positive colorectal cancer patients were distributed over 24 wells per patient and cultured in AIM-V medium supplemented with 5% human serum and 50 U/mL IL2. PBMCs were stimulated with 20 μmol/L IV9 peptide on day 0 and day 8, and then stained with anti-CD8, an IV9-HLA-A24 tetramer, and an irrelevant HIV-HLA-A24 tetramer on day 15. The frequency of anti-IV9 CD8⁺ T-cell precursors was calculated as follows: frequency = the number of wells positive to IV9-HLA-A24/(24 × the initial number of CD8⁺ cells per well). Samples showing >0.02% positivity were counted as positive to the IV9-HLA-A24 tetramer.

Various types of malignant tumors contain a CSC subset. However, the proportion of CSCs is low and unstable through long-term in vitro culture, making their assessment challenging (28). To address this issue, we generated both CSC and non-CSC lines from a single-cell clone of human colorectal cancer SW480 cells (26). In that study, we isolated both side and main population cells using Hoechst 33342 dye staining, and demonstrated that both CSC-like (SW480-SP-A, SP-B, and SP-H) and non-CSC-like (SW480-MP-A, MP-D, and MP-K) phenotypes were sustained in culture. Although 1% to 2% of the parent SW480 cell lines were side-population cells, a side-population assay demonstrated that the representative clone SP-H was 32.6% side-population cells, and the enriched side-population cells were maintained after 4 months in vitro serum culture (Fig. 1A). Likewise, <0.1% of a representative clone of MP-A were side-population cells and this clone never generated new side-population cells (Fig. 1B). The number of spheres formed in nonserum culture was approximately 10.6 times higher in SP-H than MP-A (Fig. 1C). In addition, SP-H formed significantly larger masses in mouse xenograft models in vivo (Fig. 1D–F). Tumor masses were palpable on days 28 and 20 of 1.0 × 10³ and 1.0 × 10⁴ SP-H implantation, respectively, whereas MP-A never formed palpable tumors by day 42. These data indicate that the tumor-initiating ability of SP-H was higher than that of MP-A both in vitro and in vivo. We conclude that SP-H and MP-A clones represent CSC and non-CSC phenotypes. Both clones maintained stable phenotypes.

CSCs are long-lived in the host environment and harbor tumor-initiating ability, hence are a target for immunotherapy (29). As antigen processing of CSCs is not fully understood, we directly surveyed the HLA-A24 peptide landscape of SP-H and MP-A (Fig. 2A). Briefly, 1 × 10⁹ cells of SP-H or MP-A were lysed, and a mixture of peptide-HLA-A24 complexes were immunoprecipitated using an HLA-A24–specific antibody. The bound peptides were then eluted and sequenced using mass spectrometry coupled with nano-flow liquid chromatography. The analysis identified 178 sequences of nonredundant HLA-A24 peptides from SP-H and MP-A (Supplementary Table S1). The average length of the detected sequences was approximately 9.1 amino acids (Fig. 2B), and both tyrosine (Y) at P2 and phenylalanine (F) or leucine (L) at PΩ were conserved across the sequences (Fig. 2C). These peptide profiles ensured that our ligandome analysis identified HLA-A24 ligands from SP-H and MP-A (30).

Although most of the peptides were MP-A–specific (86 peptides) or shared between SP-H and MP-A (57 peptides), 35 peptides were detected only in SP-H (Fig. 2D). Removal of peptides for which the source gene was expressed in normal organ tissues left a 9-mer peptide (IYPPQFHKV, “IV9”; Fig. 2E). This peptide, IV9, was repeatedly isolated from SP-H samples but never detected in MP-A samples. Although both SP and MP clones expressed HLA-A24 on the cell surface, MP clones expressed more HLA-A24, potentially explaining a difference in the number of isolated HLA-A24 peptides between SP and MP clones (Supplementary Fig. S1).

IV9 is encoded by both of two known isoforms of ASB4 (UniProt ID: Q9Y574), a potential component of the E3 ubiquitin ligase complex (Fig. 3A; ref. 31). The ASB4 gene was expressed by a variety of cancer cell lines, including lines derived from colorectal cancers (SW480, SW620, colo205, and HCT15), lung cancers (A549, SBC1, and SBC5), and a kidney cancer (Caki-1), as well as a liver cancer (HepG2; Fig. 3B). Expression was higher in clinically dissected colorectal cancer tissues, greater than 5-fold higher in 6 out of 21 cases (Fig. 3C). The ASB4 gene was little expressed by a panel of normal adult and fetal tissues as well as cultured mesenchymal stem cells derived from bone marrow (Fig. 3C and Supplementary Fig. S2). Thus, expression of the ASB4 gene was tumor specific in both cell lines and primary colorectal cancer tissues.

The synthetic IV9 peptide harboring Y at P2 and V at P9 stabilizes expression of HLA-A24 on T2-A24 cells (Fig. 4A). We stimulated PBMCs derived from a healthy donor with the IV9 peptide. From a cell population responding to an IV9-HLA-A24 tetramer, we established the CD8⁺ T-cell clone line (SM4), 98.8% of which was positive for the tetramer (Fig. 4B). The SM4 line responded to SW480 as well as to T2-A24 pulsed with 2 μmol/L IV9 peptides, producing IFNγ (Fig. 4C). SM4 produced IFNγ in response to SP-H and SW480, but not in response to MP-A cells. The SM4 line produced more IFNγ against the MP-A cells expressing ASB4, demonstrating that ASB4 was the responsible antigen encoding the CTL epitope (Fig. 4D and E).

Next, we biochemically evaluated and quantified the natural T-cell epitope produced in live SP-H and MP-A cells. We fractionated cell extracts from 2 × 10⁸ SP-H and MP-A using RP-HPLC, then incubated these extracts with SM4 in the presence of T2-A24 as antigen-presenting cells. We measured production of IFNγ using an ELISPOT assay (Fig. 4F). We found that the SP-H extract contained the natural SM4 epitope at fraction #5, the same fraction as synthetic IV9 peptide. As this assay used a consistent set of T cells and APCs across samples, only the amount of the IV9 peptide included in samples influenced the number of IFNγ spots. Therefore, we estimated the amount of IV9 in 2 × 10⁸ SP-H at approximately 12 fmol or more, because the number of positive spots at #5 was increased by 1.28-fold (96/75) compared with that of the 10 fmol synthetic peptide. MP-A extract did not contain IV9. Considering that ASB4 gene overexpression restored SM4 responses against MP-A, we assumed that the antigen-processing machinery of MP-A was capable of presenting the IV9 peptide. However, loss of the ASB4 gene expression resulted in loss of the IV9 peptide. Thus, IV9 is the natural CTL epitope and its presentation is limited to the colorectal CSC subset SP-H.

In accordance with the SP-H–specific detection of IV9 using mass spectrometry, ASB4 was expressed in all three SP clones, but not in any of the MP clones we had established (Fig. 5A). Although MP-A is made up mostly of main-population cells without dedifferentiation, SP-H is composed of both side- and main-population cells, suggesting that side-population cells in SP-H constantly differentiate into main-population cells (Fig. 1A). We, therefore, re-sorted SP-H cells cultured for 3 weeks into side and main populations again and determined ASB4 expression in each subset. Only the side-population descendants (SP-1, -2, and -3) but not the main-population descendants (MP-1, -2, and -3) expressed ASB4, suggesting that side-population cells of SP-H ceased ASB4 expression when differentiated into main-population cells (Fig. 5B). Moreover, the ASB4 expression was detected in 4 out of 8 colorectal cancer lines derived from 3 different colorectal cancer patients (SW480 and 620, Colo205, and HCT15; Fig. 3B). In fact, the expression levels increased to 1.9- to 3.0-fold when CSCs were enriched by sphere culture using ultra-low attachment plates without serum (Fig. 5C). Sphere culture did not influence the expression in Colo320 and HCT116, both of which were ASB4-negative. Sphere culture increased sphere formation in every colorectal cancer line as well as ASB4 expression in SW480 and HCT15 using RT-PCR (Supplementary Fig. S3A–S3C). To further investigate the enrichment of the IV9 epitope presentation by CSCs, we tested the SM4 responses. The amounts of IFNγ produced by SM4 were increased by approximately 1.6-fold against SW480 and HCT15 expressing β2m under sphere culture (Fig. 5D). These data demonstrate that ASB4 is the antigen of which the expression and following CTL responses were linked to colorectal CSC subsets.

We aimed to identify CSC antigens that are exclusively presented by CSCs and elicit CTL responses lysing only CSCs. In accordance with the ELISPOT results, SM4 lysed T2-A24 cells pulsed with the synthetic IV9 peptide, but ignored control peptides or K562 cells (Fig. 6A). SM4 successfully lysed SP clones (SP-A, SP-B, and SP-H) and, as we expected, left MP clones (MP-A, MP-D, and MP-K) intact (Fig. 6B). The killing efficacy (LDH release (%)) against MP clones was near to zero, indicating that ASB4 is a CSC-specific antigen and the CTL activity discriminates tumorigenic CSCs from non-CSCs. The result was consistent with the SP-H-specific IV9 peptide production observed in Fig. 4F. SP-H and unsorted SW480 contain approximately 30% and 1% to 2% of side-population cells, respectively (Fig. 1A and Supplementary Fig. S4). We found that SM4 lysed SW480, besides SP-H, to as low as 59.3% of SP-H at an effector/target (E/T) ratio of 9 (Fig. 6C). As SM4 activity against such a small cell population was detectable, we tested other colorectal cancer lines without CSC enrichment. We found that SM4 recognized and lysed HCT15/β2m, which expresses ASB4 and is HLA-A24-positive, but left Colo205 and Colo320 intact, both of which lack HLA-A24 (Supplementary Fig. S5). The CTL-mediated lysis of SW480 was blocked by the presence of the pan-HLA class I W6/32 mAb as well as the HLA-A24–specific C7709A2 mAb, but not by the HLA-DR–specific L243 mAb, ensuring that the response was restricted to HLA-A24 (Fig. 6D).

Finally, we investigated the effects of SM4 on colorectal cancer using in vivo models. A randomly generated CTL clone was prepared as a control, which did not respond to an IV9-HLA-A24 tetramer. This clone recognized neither SP-H nor SW480 or MP-A (Supplementary Fig. S6). Immunodeficient NSG mice were implanted with 1 × 10³ SW480 on day 0. SM4 (5 × 10⁵) was then intravenously injected before (on day −3) or after tumor implantation (on days 35 and 42; Fig. 7A). We used both models to evaluate antitumor CTL effects at early stages before tumors developed into large masses. In contrast to the control CTL clone, the adoptive transfer of SM4 significantly prevented tumor growth in both CTL injection models “before and after” tumor implantation (Fig. 7B and C, respectively). Even at the end of the time courses on day 56, both adoptive transfer models controlled tumor sizes such that implanted tumors were not palpable. Hence, SM4, which targeted only an SP subset of SW480, prevented tumor formation of SW480, which was composed of 1% to 2% of SP cells among otherwise MP cells. This result suggests that the CSC subset is a necessary and sufficient target in preventing colorectal cancer formation at the early stages.

We also prepared mice-bearing established masses of SW480 and counted the number of tumor-infiltrating CTLs 2 days after the adoptive transfer of SM4 (Supplementary Fig. S7A). The IHC results showed that SM4, but not control CTLs, was present inside SW480 tumors, whereas both CTLs were found inside the spleens (Fig. 7D–F). We estimate that the in vivo SW480 tumors were contained 2% or fewer SP cells, as is observed for in vitro cultures (26), because a similar analysis with SP-B tumors revealed that the percent of SP cells was stable in vivo (Supplementary Fig. 7B). The data demonstrated that SM4 was capable of homing to IV9-HLA-A24 complexes and migrating into tumors in search for a small cell population of CSCs.

Because the IV9 peptide identified in this study is displayed by HLA-A24, which is the most common HLA-A type among the East Asian population, we focused on HLA-A24–positive colorectal cancer patients and assessed the frequency of IV9-specific CD8⁺ T cells in their PBMCs. All 6 patients were histologically diagnosed with colorectal adenocarcinoma and the group consisted of 3 men and 3 women, ages between 59 and 98 (Supplementary Table S2). The primary lesion had been surgically removed in three patients, and five patients had received chemotherapy or radiotherapy. The group included five patients at stage IV and one patient at stage II. PBMCs from the patients were randomly fractionated and stimulated with 20 μg/mL of IV9 peptides for 14 days. Induced IV9-specific CD8⁺ cells were detected and counted using an IV9-HLA-A24 tetramer (Fig. 8A). We regarded the cell fraction labeled with 0.03% or more proportions as tetramer positive (32). As a result, T-cell induction was detected in 5 out of 6 patients, and the frequency of IV9-specific T-cell precursors ranged from 4.4 × 10⁻⁷ to 4.1 × 10⁻⁶ of CD8⁺ cells (Fig. 8B). The lower limit of detection in this assay was around 2.0 × 10⁻⁷ and the frequency in one patient (B05) was under the limit. Three of the patients (B01, B03, and B06) at stage IV reached long survival (>8 years); for these three patients, we detected IV9-tetramer–positive CD8⁺ T cells in multiple wells. Thus, CD8⁺ T cells of colorectal cancer patients are not immunologically tolerant to the ASB4 antigen and IV9 peptide stimulation elicited specific T-cell responses in patients with colorectal cancer.

CSC expression is prioritized in selection of antigens appropriate for cancer vaccination. Antigen expression in a CSC subset might influence antitumor CTL effects (33). In our study, not only is the IV9 peptide a natural CTL epitope directly identified using HLA-ligandome analysis, but also it is presented by an SP but not an MP subset of SW480 cells. Repeated detection by biochemical analysis also implies that the IV9 peptide is presented by HLA-A24 of CSCs, rendering it a target of CTL immune surveillance. IV9-CTLs indeed discriminated CSCs from non-CSCs, lysing only the CSC subset. The IV9-CTLs belong to a group of CTLs that recognize CSCs, but are capable of ignoring non-CSCs. The adoptive transfer of the CTLs prevented tumor formation in NSG mice implanted with 1 × 10³ unsorted SW480 cells that consisted of 1% to 2% of SP and the majority of MP cells. This result demonstrated that most SW480 cells were not the target, and the ability of IV9-CTLs to ignore non-CSCs did not harness the antitumor effect in vivo. Although non-CSCs are predominant in number in many tumors, only CSCs are able to sustain tumorigenesis (34). In fact, NSG mice implanted with 1 × 10⁴ SW480-MP cells did not develop tumors even after a month, whereas 1 × 10³ SW480-SP cells readily generated tumor masses. These data together suggest that the CSC subset is a useful target of immunotherapy despite of its rarity.

The IV9 peptide is encoded by ASB4, the gene expression of which was not detected using RT-PCR in an array of normal fetal and adult tissues, including mesenchymal stem cells of the bone marrow. Expression increased in colorectal cancer tissues, and moreover, 28.6% (6 out of 21) of cases showed greater than a 5-fold increase compared with normal colon tissue without enrichment. Considering that the sphere culture of SW480, Colo205, and HCT15 enriched ASB4 gene expression, each primary colorectal cancer tissue is composed of varying proportions of the CSC subset, and in some cases, CSCs might be already enriched through their clinical courses. The exact molecular function of ASB4 in colorectal CSC function is unclear. In mice, the expression of the ASB4 gene has been reported in the developing placenta and the testis; however, adult tissues cease ASB4 gene expression, suggesting ASB4 functions early in development, perhaps at trophoblast differentiation (35–37). On the contrary, its increased expression has been reported in hepatocellular carcinomas and adrenal gland tumors, which may suggest another role of ASB4 in metastasis or tumorigenesis (38, 39). We quantitatively compared sphere formation of SP-H in the presence and absence of ASB4 gene expression; however, silencing ASB4 gene expression did not decrease the number of spheres. More persistent gene silencing other than siRNA might be necessary to observe the phenotype change, or this result could suggest that ASB4 does not serve as a driver of sphere formation. Nevertheless, the findings of the present study demonstrate that the presence of the ASB4 antigen is linked to a CSC subset of colorectal cancer. ASB4 may hold promise as a therapeutic target of CTL-based immunotherapy in colorectal cancer patients. Here, it was possible to detect the IV9-CD8⁺ T-cell precursors in 5 out of 6 colorectal cancer patients, and we knew that the remaining patient (B05) had received dexamethasone for 6 months before blood sampling, being in a clinical state of immune suppression. All three patients with long-term survival over 8 years showed higher IV9-CD8⁺ T-cell frequency in contrast to the others, suggesting a relationship between patient survival and CD8⁺ T-cell surveillance against the ASB4 antigen.

Clinical success of immune checkpoint blockades both in reducing the sizes of cancers and in prolonging cancer patient survival ensured that patients' T-cell responses are able to cope with malignancies (40). Accumulating evidence suggests that mutational load is positively correlated with clinical effects, and has sparked further research on screening neoantigens that arise from gene mutations (41–44). CTLs targeting a neoantigen discriminate it from their wild-type counterpart, and exhibit high cytotoxicity against tumors carrying the responsible gene mutation (27). However, most colorectal cancers belong to a microsatellite-stable tumor type, being an exception that is refractory to immune checkpoint blockade despite their prevalence (43). Individual variation in gene mutations also call for personalized immunotherapy, which requires detection of antigens for each patient (45). In this study, the antigen we focused on was, conversely, expressed only in a subset of colorectal cancer cells. In our SW480-derived SP- and MP-clone models, certain descendants of SP clones differentiated into MP cells during culture, keeping the proportion of SP-cells low. Therefore, we consider that eliminating such a small subset of cancer cells may not be effective in reducing the sizes of already developed bulky tumors in vivo. Instead, the CSCs responsible for the onset of tumor formation are shared among patients, generating a practical target for vaccination. The capability of IV9-CTLs to home to the target may help to identify tumor buds scattered over the body of patients at an early stage. We thus deduce that CSC-specific CTL surveillance could be useful to colorectal cancer patients with cured primary lesions in preventing relapse or metastasis.

No potential conflicts of interest were disclosed.

Conception and design: T. Kanaseki, N. Sato, T. Torigoe

Development of methodology: S. Miyamoto, V. Kochin, T. Kanaseki, Y. Hirohashi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Miyamoto, V. Kochin, A. Hongo, S. Tokita, Y. Kikuchi, A. Takaya, T. Terui, K. Ishitani, F. Hata

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Miyamoto, V. Kochin, T. Kanaseki, T. Tsukahara, I. Takemasa, A. Miyazaki, H. Hiratsuka, T. Torigoe

Writing, review, and/or revision of the manuscript: S. Miyamoto, T. Kanaseki

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Torigoe

Study supervision: T. Kanaseki, N. Sato, T. Torigoe

The authors thank Dr. Goto (Sumitomo Dainippon Pharma) for providing tetramers and Mr. Matsuo (Sapporo Clinical Laboratory) for technical support on IHC. This work was supported by Japan Society for the Promotion of Science (JSPS) to T. Kanaseki, Suhara Kinen Zaidan to T. Kanaseki, Practical Research for Innovative Cancer Control from Japan Agency for Medical Research and Development (AMED) to T. Torigoe, Grants-in-Aid of Ono Cancer Research Fund and to T. Torigoe. Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan to N. Sato, and a program for developing the supporting system for upgrading education and research from the Ministry of Education, Culture, Sports, Science and Technology of Japan to N. Sato.

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.