Regeneration of CD8αβ T Cells from T-cell–Derived iPSC Imparts Potent Tumor Antigen-Specific Cytotoxicity

During the last several years, cancer immunotherapy using a variety of strategies has progressed remarkably, for example, blockade of inhibitory signals in the immune system (1–3), adoptive transfer of cytotoxic T lymphocytes (CTL; refs. 4–6), transfer of T-cell receptor (TCR) genes (7–9), etc. Although these achievements may suggest that CTLs present in cancer patients have the potential to cure them, researchers and clinicians have been facing a major obstacle in getting enough antigen-specific CTLs for therapy. Some groups have reported that tumor infiltrating lymphocytes (TIL) or tumor antigen specific CTLs can be efficiently expanded (10–13), but in general it is not so easy. This problem is mainly attributable to the nature of CTLs; they easily become exhausted or die during cultivation.

To solve this problem, we came to the idea of utilizing induced pluripotent stem cell (iPSC) technology for the cloning and expansion of CTLs. When iPSCs are established from antigen-specific T cells (T-iPSC), they should inherit rearranged TCR genes, and thus all T cells regenerated from T-iPSCs should express the same TCR. Because iPSC expansion in vitro is almost unlimited, it should be possible to obtain as many fresh CTLs as needed. In keeping with this idea, we have recently succeeded in regenerating MART1-specific CTLs from iPSCs originally derived from CTLs of a melanoma patient (14). Other groups also have regenerated viral antigen-specific T cells (15, 16), T cells expressing an invariant TCRs (17–19), and T cells that were genetically engineered to express a so-called chimeric antigen receptor (CAR; ref. 20).

None of the previous studies, however, showed whether these regenerated CTLs are as good as primary CTLs in terms of antigen-specific cytotoxic activity. Indeed, in the study using CAR-T-iPSCs, the authors stated that regenerated CTLs are phenotypically similar to γδT cells (20).

Authentic CTLs are CD4⁻ CD8α⁺ CD8β⁺ (CD8αβ heterodimer⁺). CD8αα type T cells represent a minor population in most lymphoid tissues but are one of the major populations in the mucosal tissues, where they are regarded as “innate type” T cells. Of note, the CD8αα homodimer does not function well as a TCR coreceptor, because it does not efficiently bind to MHC class I molecules (21). Although regenerated CD8 T cells in the previous studies all seemed to be of the innate type, in this study we describe a novel yet simple method to generate CD8αβ T cells and show that they are active against leukemia.

This study was approved by the Institutional Review Board of the Graduate School of Medicine, Kyoto University (approval number: G761) and abided by the tenets of the Declaration of Helsinki. All specimens from healthy individuals and patients were collected after written informed consent was obtained.

Op9, OP9/DLL1, and 409B2 were purchased from RIKEN BRC. THP1 was purchased from ATCC. HL60, HEL, and KG1a were maintained in our department. OUN1 and MEG01 were gifts from Dr. Masaki Yasukawa (Ehime University, Ehime, Japan). K562 was a gift from Dr. Kiyotaka Kuzushima (Aichi Cancer Center, Aichi, Japan). C1R-A*24:02 was a gift from Dr. Masafumi Takiguchi (Kumamoto University, Kumamoto, Japan). Upon receiving the cell lines, frozen stocks were prepared within one to five passages and new stocks were thawed frequently to keep the original condition. The cell lines were passaged for less than 3 months after receipt or resuscitation. They were also authenticated by morphology, growth rate, and surface phenotype, especially expression of HLA class I.

T-iPSCs were differentiated to CD4/8 double-positive (DP) cells using the OP9 and OP9/DLL1 stromal cell coculture systems as described (14, 22, 23), with slight modification. In brief, iPSC colonies were dissociated using trypsin (0.25%) and collagenase IV (1 mg/mL) and mechanically disrupted into small clumps by pipetting. About 600 iPSC clumps were collected and plated on gelatin precoated OP9 dishes filled with OP9 medium, that is, α-MEM (Invitrogen) with 20% FCS. On day 13, colonies were treated with collagenase Type IV (50 U/mL) and trypsin-EDTA (0.05%). Cells were plated in a OP9/DLL1 semiconfluent dish on OP9 medium containing hIL-7 (5 ng/mL), hFlt-3L (5 ng/mL), and hSCF (5 ng/mL). On day 15, semiadherent cells were collected and passage into a new dish layered with OP9/DLL1 cells. From this point, passage was done every 7 days. On day 40, floating cells were collected and CD4/8 DP cells were enriched by using CD4 microbeads (Milteny Biotec). DP cells were stimulated with CD3 Ab (15 ng/mL; OKT-3, eBioscience) in the presence of hIL-2 (100 U/mL) and hIL-7 (5 ng/mL). CD8 SP cells were stimulated one to five times by HLA-A*24:02⁺ lymphoblastoid cell line (LCL) pulsed with the peptide in the presence of hIL-7 (5 ng/mL) and hIL-21 (10 ng/mL).

Induced DP and double-negative (DN) cells (day 40) from T-iPSCs LMP2#1-1 were isolated and labeled with Cell Trace Violet and CFSE, respectively. A total of 10⁵ cells of the DP and DN cell mixture at the indicated ratios were cultured with CD3 Ab (15 ng/mL). After 6 days, cells were analyzed by flow cytometer.

The transwell system was used for the detection of soluble mediator effects. Labeled DP and DN cells (5 × 10⁴ cells each) as above were seeded in the lower and upper compartments, respectively, of a 0.4 μm pore size transwell plate (Coaster), and cultured in the presence of CD3 Ab (15 ng/mL). In control wells, DP or DN cells (5 × 10⁴ cells) or their mixture (5 × 10⁴ cells each) were cultured with CD3 Ab. After 6 days, cells were analyzed by flow cytometer.

Labeled DP and DN cells as above were cocultured at the indicated ratio with or without CD3 Ab for 5 hours. Annexin V (BioLegend) and PI double-negative cells were considered as alive.

Regenerated LMP2-specific CD8αα T cells and CD8αβ T cells were isolated by FACS Aria III (BD Bioscience). Isolated peripheral blood CD8αα T cells (TCRαβ⁺ CD4⁻ CD8α⁺ CD8β⁻), CD8αβ T cells (TCRαβ⁺ CD4⁻ CD8α⁺ CD8β⁺), and γδT cells (TCRγδ⁺ TCRαβ⁻) were stimulated with CD3/28 stimulator (Milteny Biotec) for 7 days according to the manufacturer's instructions in the presence of hIL-7 (10 ng/mL) and hIL-15 (20 ng/mL). Total RNA was isolated using an RNeasy Plus Mini Kit (QIAGEN). Libraries for RNA sequence were prepared using NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs). Sequence alignment was performed using our in-house pipelines as previously described (24). Fragments per kilobase of exon per million fragments mapped (FPKM) was calculated for Refseq genes and then log-transformed. The hierarchical clustering of gene expression data was performed using Pearson's dissimilarity as distance measure and centroid method for linkage. For this analysis, we removed genes with no or low expression (mean FPKM < 2), then identified the most variant 5% of genes (n = 527). Multi Experiment Viewer software was used to generate heat maps and hierarchical clustering. GEO accession number is GSE 81975.

All data with error bars are presented as mean ± SD. Difference was assessed using paired t-test using Prism (GraphPad software). Values of P < 0.05 were considered significant.

Latent membrane protein 2 (LMP2) is an Epstein–Barr virus (EBV)-encoded antigen that is considered as a good CTL target in EBV-related tumors (25, 26). LMP2-specific CTLs, defined by LMP2 tetramer staining, from a healthy volunteer were expanded by stimulation with antigen-presenting cells loaded with LMP2 peptide (Fig. 1A). After several weeks of expansion, these antigen-specific CTLs efficiently killed leukemic cells loaded with LMP2 peptide at 1 nmol/L concentration (Fig. 1B).

For establishment of iPS cells from CTLs, we used SV40 Large T antigen (27, 28) in addition to Yamanaka factors to enhance the efficiency of reprogramming (Supplementary Fig. S1). For transduction of CTLs, we used Sendi virus vector, which has been shown to be safe because transferred genes as well as viral genes are not integrated into genome (29, 30). By transducing these CTLs with Yamanaka factors and SV40, two clones of iPSC lines, LMP2#1-1 and LMP2#2-13, were established. Sendai virus vectors containing Yamanaka factors and SV40 were not detected in either of these clones, and expression profiles of various stem cell genes were indistinguishable from those of control iPSCs (Supplementary Fig. S2A and S2B). Hereafter, we will focus on the clone LMP2#1-1 (Fig. 1C). Expression of stem cell genes in the LMP2#1-1 cell line was confirmed by cytochemistry, it had a normal karyotype, and was pluripotent, as judged by teratoma formation (Supplementary Fig. S2C–S2E).

We then induced T cells from LMP2-T-iPSCs using the conventional method (14). As has been reported, DP and DN cells expressed TCR on Day 40 (Fig. 1D). Upon stimulation with CD3 Ab, CD8 T cells were generated after 6 days and were exclusively of the CD8αα type and LMP2-tetramer positive (Fig. 1E). However, as much as 100 nmol/L peptide was required to induce efficient cytotoxic activity (Fig. 1F), making these cells ∼100-fold less competent than primary CTLs (Fig. 1G). However, the regenerated CTLs were able to efficiently kill K562 (Fig. 1H), indicating that they have high Natural Killer (NK)-like cytotoxicity.

We searched then for conditions in which CD8αβ T cells could be induced. We found that when purified DP cells were stimulated with CD3 Ab, CD8SP cells were generated and they were mostly of CD8αβ type (Fig. 2A). By contrast, purified DN cells gave rise to CD8αα T cells upon activation (Fig. 2A and B). The CD8αβ T cells could also be efficiently generated from purified DP cells stimulated by autologous LCLs loaded with LMP2 peptide (Fig. 2C), indicating that the absence of DN cells, rather than the stimulation method, is critical for the induction of CD8αβ T cells.

To exclude the possibility that a preexisting small number of CD8αβ cells were preferentially expanded, purified DP cells were labeled with Cell Trace Violet, and then stimulated (Fig. 2D). On day 2, DP cells started to downregulate CD4 without cell division, and then CD8αβ cells were generated on day 4 with only one or two cell divisions, indicating that CD8SP cells were induced from DP cells.

Thus, it seemed that DN cells might be exerting a suppressive effect on the generation of CD8αβ T cells. To test this possibility, isolated DP and DN cells were mixed at various ratios before stimulation, and it was seen that DN cells at a one ninth ratio to DP cells almost completely suppressed the generation of CD8αβ T cells (Fig. 3A–C).

The possibility that DN cells suppress DP-SP development by secreting soluble factors is unlikely, because virtually no suppression of CD8αβ T-cell generation was seen in a transwell experiment in which DP cells and DN cells were separated by a semipermeable membrane (Fig. 3D and E). We then observed that the DP cells died as early as 5 hours after activation in cultures with a mixture of DN and DP cells, and that the frequency of live DP cells decreased along with an increase in the DN/DP ratio (Fig. 3F–I), indicating that activated DN cells directly kill DP cells. We further found that a significant proportion of DN cells expressed perforin and granzyme A prior to activation (Fig. 3J). These results indicate that failure of DP cells to generate CD8αβ T cells after activation is attributable to the direct cytotoxicity of DN cells against DP cells. The entire culture procedure and profiles of generated cells are depicted in Supplementary Fig. S3.

We analyzed gene expression profiles of regenerated CD8αβ T cells (r-CD8αβ) in comparison with regenerated CD8αα T cells (r-CD8αα) and primary CD8αβ and CD8αα T cells (p-CD8αβ, p-CD8αα), using RNA-seq. Not only surface molecules, for example, CD3, CD8α, CD8β, and CD28 but also intracellular functional molecules, for example, GZMA, GZMB, PRF1, LCK, and BCL11B were commonly expressed in r-CD8αβ and p-CD8αβ (Fig. 4A). As to immune-checkpoint molecules, LAG3 was expressed in rCD8 cells but PDCD1 and HAVCR2 (Tim3) were very low in expression. The signature transcription factors of CD8 T cells including TBX21, RUNX3, GATA3, and IKZF1 were commonly expressed in all four groups, but ZBTB16 (PLZF), a representative transcription factor for innate type T cells (31–33), was low in both r-CD8αβ and p-CD8αβ compared with both types of CD8αα cells (Fig. 4B). Reflecting NK-like cytotoxicity of r-CD8αα, NK-related genes, for example, NCAM1, NCR1, NCR2, KLRC2, and KLRC3 were highly expressed in r-CD8αα. Although r-CD8αβ also express some of them (KLRC2 and KLRC3), but others were not expressed (Fig. 4C). However, r-CD8αα and r-CD8αβ shared some common characteristics in terms of chemokine receptor expression, such as high CXCR3 and low CCR7 expression, which are reminiscent of effector CD8 T cells (Fig. 4D and E). In global gene expression pattern, r-CD8αβ were positioned closer to p-CD8αβ compared to r-CD8αα (Fig. 4F). Taken together, these observations indicate that gene expression profiles of r-CD8αβ resembled with that of p-CD8αβ, although retaining some innate lymphocyte signature.

Regenerated CD8αβ T cells killed antigen-loaded target cells at a 1 nmol/L peptide concentration (Fig. 5A), and their antigen-specific cytotoxic activity was found to be slightly weaker than that of primary CTLs, but much stronger than that of regenerated CD8αα T cells (Fig. 5B, see Fig. 1G for comparison). Regenerated CD8αβ T cells showed comparable potential to primary CTLs in producing IFNγ, and produced IFNγ and TNFα at more than 1 nmol/L concentration of peptide (Fig. 5C and D). NK-like cytotoxicity against K562 cells was weaker than regenerated CD8αα T cells (Fig. 5E). These results collectively indicate that this novel method works well to regenerate functional CD8αβ T cells. Another LMP2-specific T-iPSC clone #2-13 also differentiated to CD8αβ T cells (Fig. 5E) and exhibited peptide-specific cytotoxicity slightly weaker than the original CTLs (Fig. 5F).

We then decided to apply this method against acute myeloid leukemia (AML) cells expressing WT1 antigen, a cancer-testis antigen broadly expressed in various types of solid tumors as well as in leukemia (34, 35). We firstly induced WT1-specific CTLs from a healthy volunteer by using WT1 peptide (Fig. 6A) and measured their cytotoxic activity (Fig. 6B). We then produced iPSCs from these CTLs, and a total of three lines were established (Supplementary Fig. S4 and Supplementary Table S1). Hereafter we will show the data using clone WT1#3-3. Regenerated cells were of the CD8αβ type and WT1-tetramer positive (Fig. 6C), and the sequence of the TCR genes in this clone was determined (Supplementary Table S1).

We then expanded CD8 T cells by stimulating with an autologous B LCL carrying WT1 peptide, or with a CD3 Ab, and a six- to eight-fold expansion was seen during 1 week in both cases (Fig. 6D). We decided to use LCL for further studies because they seemed slightly more efficient. During 6 to 8 weeks' culture, CD8 T cells were expanded by ∼10,000 fold (Fig. 6E), changing their surface phenotype from a naïve to an effector/memory profile (Fig. 6F). Regenerated CD8 T cells showed the same antigen-specific cytotoxic activity as the original CTLs (Fig. 6G and H).

Regenerated WT1-CTLs also efficiently killed THP1 and HL60 cells, HLA-matched AML cell lines expressing endogenous WT1 protein (Fig. 7A). The cytotoxic activity was almost completely eliminated by adding a blocking Ab against HLA class I, indicating that the killing was MHC restricted and dependent on TCR-MHC binding. Other AML cell lines OUN1 and MEG01 were also killed, but the cytotoxic activity was only partially suppressed by HLA blocking Ab, suggesting that they were also killed by NK-like cytotoxicity (Fig. 7B). Other AML lines, HEL and KG1a, which express WT1 but not express HLA-A*24:02, were not killed, confirming that the TCR of WT1#3-3 cells recognizes WT1 antigen presented on HLA-A*24:02. We further tested whether regenerated CTLs are able to kill primary leukemic cells. Three samples expressing the highest levels of endogenous WT1 antigen selected among several samples from HLA-A*24:02⁺ patients were killed by regenerated CTLs (Fig. 7C).

Finally, we examined CTL activity in vivo in a xenograft model. NOD.Cg-Prkdcscid Il2rgtm1Sug/Jic (NOG) mice inoculated with HL60 cells followed by administration of regenerated CTLs (Fig. 7D). Although leukemia cells expressing human CD33 were detected from all the control mice, inoculated regenerated CTLs were present in peripheral blood of all the treated mice (Fig. 7E). The CTL treated mice showed significantly longer survival compared with control mice (Fig. 7F).

To test the safety of transfused regenerated CTLs, we did long-term observation in vivo. NOG mice bearing no leukemic cells were inoculated with regenerated CTLs in the same manner as in xenograft model (Fig. 1D), and observed for 6 months. During observation period, no mice showed diarrhea or body weight loss. After 6 months, all mice were dissected, and mononuclear cells in bone marrow, spleen, and peripheral blood were analyzed. Small amount of regenerated CTLs were detected from spleen and bone marrow of some mice (Fig. 7G). No mice suffered from tumor derived from regenerated CTLs. We also did not see any sign of tissue damage caused by transferred CTLs.

Many groups have been struggling to produce CD8αβ type T cells from T-iPSCs, but the critical method described here turned out to be rather simple, simply include a DP cell purification step before stimulation. Our results can be summarized as three points: upon TCR stimulation, (i) DP cells give rise to CD8αβ T cells, (ii) DN cells immediately kill DP cells, (iii) DN cells give rise to CD8αα T cells. For the first point, the alternative ligand model of thymic selection may provide the explanation. According to this model, DP cells positively selected in the cortex by peptides specifically expressed there can survive in the medulla because they do not encounter the same peptide; thus, the agonist peptide can induce positive selection in the cortex (36, 37). It may be that our culture system reproduces such a cortical, positively selecting microenvironment. As to the second point, we have shown that DN cells already contain mature cells (Fig. 3J), although they do not kill other cells unless activated further via their TCR. It seems likely that DN cells kill DP cells by direct NK-like cytotoxicity; however, it remains to be studied what the target molecules are in this killing. As to the third point, in our system, the TCR is expressed at the DN stage (Supplementary Fig. S3C). When these cells are activated, they may undergo a developmental program toward γδT cells, because it is known that the DN cell fate decision is based on TCR signaling strength. DN cells take the γδT cell fate upon receiving a strong TCR signal via the γδTCR, whereas they take the αβT cell fate when there is a weak signal via the preTCR (38, 39). This model is also in line with previous reports showing that DN cells expressing a transgenic αβTCR at the DN stage come to exhibit a γδT cell-like phenotype (40).

Regenerated CD8αβ T cells showed antigen-specific cytotoxicity slightly less, but nearly comparable to primary CD8αβ T cells (Supplementary Table S2). This is a marked improvement compared to regenerated CD8αα T cells and might partly be explained because the CD8αβ heterodimer binds to MHC class I and thus functions as a co-receptor to induce lck signaling, whereas the CD8αα homodimer does not. Gene expression profiles also showed that regenerated CD8αβ T cells are closer to primary CD8αβ T cells, although still retaining some phenotypic features of innate type T cells. As activated T cells express some NK activation receptors and signal molecules like NKG2D, NKp30, DAP12, and so on (41–43), it is not surprising that these regenerated CTLs cultured in vitro also express such NK-associated molecules and possess NK-like cytotoxic potential.

There is growing evidence that naïve T cells, stem cell memory T cells (T_SCM), or central memory T cells (T_CM) are superior as source of adoptive T-cell transfer for tumor immunotherapy (44–47). In terms of surface markers, regenerated CD8αβ T cells expressing low CCR7 and high CXCR3 are considered as effector phenotype T cells, although we are using IL21 to inhibit terminal differentiation (5, 48). Nevertheless, our culture method has advantages that the regenerated CD8αβ T cells can be expanded more than 10,000 fold by repeated TCR stimulation while retaining comparable TCR-specific cytotoxic potential. Furthermore, it was reported to be hard to evaluate in vitro cultured CTLs just based on their surface markers with regard to the point whether they are terminally differentiated or not (49). In addition, at present, it is also difficult to appropriately address this issue in vivo in xenograft model, because microenvironments in immune system of mouse xenograft model are so far from physiological situation in human. In case of murine CTLs regenerated from T-iPSCs, it was recently shown that the regenerated CTLs exhibited memory response in vivo in mice bearing syngeneic tumor (50).

One may concern about the safety of regenerated CTLs is tumorigenecity after administration. Although we use SV40 large T antigen to establish T-iPSCs, we confirmed that SV40 was completely deleted from all the iPSCs. Also we examined the long-term safety in transfused mice, and confirmed all the mice were free of tumor derived from regenerated CTLs.

Another concern about safety is off-target specificity. In our method almost all the regenerated CD8αβ T cells after expansion by LCL pulsed with cognate peptide retain the same TCR detected as tetramer positive cells (Fig. 2C, 5F, 6C, 6F). To be exact, this is not the case at the DP stage, where RAG1 and RAG2 genes are expressed and secondly rearrangement in TCR alpha chain gene may occur. Indeed, a part of the CD8αβ T cells just regenerated by TCR stimulation have lost specificity for original epitope, resulting in the formation of negative population in tetramer staining (Supplementary Fig. S3E, top right). However, these cells can be eliminated during expansion by LCL pulsed with cognate peptide (Supplementary Fig. S3E, bottom right). We actually did not see any sign of Graft-versus-Host reaction in the mice inoculated with regenerated CTLs (Fig. 7G). In addition, it is also possible to completely ensure clonality by deleting RAG1 or RAG2 gene in T-iPSCs using genome editing technology.

In this method starting from 10⁴ T-iPSCs in one 10-cm culture dish, we can eventually obtain 10⁹ to 10¹⁰ regenerated CD8αβ T cells in one round culture, which could be sufficient for one transfusion to a patient, thus providing a strong rationale for clinical application of this strategy. We have established antigen-specific T-iPSCs from healthy volunteers, making an allogeneic approach possible. In such an allogeneic setting, it would be preferable to produce T-iPSCs from an HLA-haplotype homozygous donor, because regenerated CTLs could then be used for HLA-haplotype heterozygous patients. In such a strategy, “Off the shelf” T-cell drug would come true by generating CD8αβ T cells in advance and preserving as frozen stocks.

No potential conflicts of interest were disclosed.

Conception and design: T. Maeda, N. Kadowaki, H. Kawamoto

Development of methodology: T. Maeda, S. Nagano, H. Ichise, D. Yamada, H. Koseki, T. Kitawaki, K. Masuda

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Maeda, S. Nagano, H. Ichise, K. Kataoka, S. Ogawa, H. Koseki, K. Masuda

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Maeda, K. Kataoka, D. Yamada, K. Masuda

Writing, review, and/or revision of the manuscript: T. Maeda, A. Takaori-Kondo, K. Masuda, H. Kawamoto

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Maeda, S. Nagano, K. Kataoka, A. Takaori-Kondo, K. Masuda

Study supervision: T. Maeda, N. Kadowaki, A. Takaori-Kondo, H. Kawamoto

We thank Eri Satoh and Toshika Senba for technical support; Mahito Nakanishi and Manami Ohtaka for kindly providing Sendai virus of Yamanaka factors; Haruo Sugiyama and Fumihiro Fujiki (Osaka University), Masahiro Kawahara (Shiga University of Medical Science), and Yutaka Shimazu and Masaki Miyazaki (Kyoto University) for helpful discussion; and Peter Burrows (University of Alabama) for critical reading of the manuscript.

This work was supported by a Project for the Development of Innovative Research on Cancer Therapeutics (P-Direct) from the Ministry of Education, Culture, Sports, Science and Technology, Japan, by Astlym Co. Ltd, and by RegCell, Inc.

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.