Generation of tumor-targeted human T lymphocytes from induced pluripotent stem cells for cancer therapy

Current approaches to adoptive T-cell therapy require the labor-intensive generation of T-cell lines from carefully selected donors or the genetic engineering of autologous T cells from each individual patient, hindering the facile and broad use of T cells with pre-determined antigen specificity. Having rapid access to unlimited antigen-specific T lymphocytes with optimized therapeutic features would greatly advance the scope and delivery of T-cell therapies. Previous studies support the feasibility of generating T lymphocytes from human embryonic stem cells (ESCs) and iPSCs in vitro, although the yield of lymphoid cells has been low and their nature only partially defined ^3,4. More specifically, the functional characterization of T cells derived from ESCs and iPSCs is complicated by not knowing their antigen specificity and HLA restriction. For example, T cells generated in vitro from ESCs or iPSCs have an unpredictable T-cell receptor (TCR) repertoire because TCR gene rearrangements are random³ and the cells are positively selected by unclear mechanisms during their in vitro differentiation. This limitation can be circumvented by using iPSCs bearing a rearranged endogenous TCR of known antigen specificity^5–6. Unfortunately, this approach requires laborious cloning of antigen-specific T cells and is limited to antigens for which patient-specific T cells can be detected. Furthermore, as TCRs recognize antigens presented by specific HLA molecules, the clinical use of T cells that recognize antigen through an endogenous TCR is constrained by the need to match their specificity to the HLA of the recipient patient.

Genetic engineering of T lymphocytes to express CARs has recently emerged as a promising approach to rapidly generate tumor-targeted T cells endowed with enhanced anti-tumor properties⁸. For example, CARs redirect T-cell specificity in HLA-independent fashion, thereby eliminating the need to consider HLA restriction and overcoming some tumor escape mechanisms⁸. We previously demonstrated that human T cells expressing a CAR targeted to the CD19 antigen, which is expressed on the vast majority of leukemias and lymphomas, can eradicate B-cell malignancies in mice⁹. Importantly, second-generation CARs, combining both activation and co-stimulatory signaling domains, enhanced T-cell expansion and in vivo persistence ^8–10. We and others, recently demonstrated in clinical trials that second-generation CD19 CAR-modified T cells efficiently induce complete remissions in patients with acute or chronic lymphoblastic leukemias^11–14.

Here we hypothesized that genetic engineering of iPSCs with second-generation CARs⁸ would be an efficient strategy to concomitantly harness the unlimited availability of iPSCs and to generate phenotypically defined, functional and expandable T cells that are genetically targeted to a tumor antigen of interest (Fig. 1a). To this end, we generated iPSC clones (T-iPSCs) by transducing peripheral blood T lymphocytes (PBL) from a healthy volunteer with two retroviral vectors each encoding two of the reprogramming factors KLF4, SOX2, OCT-4 and C-MYC (Supplementary Fig. 1a)⁷. Multiple randomly selected T-iPSC clones were analyzed, and their pluripotency (Supplementary Fig. 1b–g) and T-cell origin (Supplementary Fig. 2a, b) were confirmed. Clone T-iPSC-1.10 was stably transduced with a bicistronic lentiviral vector encoding 19–28z (1928z-T-iPSC), a second-generation CAR specific for CD19 (ref. 14), and the fluorescent marker mCherry (Supplementary Fig. 3a–c). To direct the differentiation of 1928z-T-iPSC to the T-lymphoid lineage, we first optimized a serum-and feeder-free in vitro differentiation protocol for the generation of hematopoietic precursors through embryoid body formation (Fig. 1b). Similar to previous reports^3,4,15, we found that CD34⁺ cells from day 10 embryoid bodies expressed the highest levels of key transcription factors for lymphoid differentiation (Supplementary Fig. 4a), specifically showing increased expression of Notch 1 and CD127 (IL7Rα) in the CD34⁺CD43⁻ subset compared to CD34⁻CD43⁻ cells (Supplementary Fig. 4b). We therefore dissociated day 10 embryoid bodies and transferred the hematopoietic precursors onto Delta-like 1–expressing OP9 (OP9-DL1) feeder cells to induce T-lymphoid differentiation in an established co-culture system in the presence of the cytokines stem cell factor (SCF), Flt3L and interleukin (IL)-7 (Fig. 1b). mCherry expression was ascertained throughout the differentiation process and we did not detect substantial silencing of mCherry expression (Fig. 1b). As early as day 25 of differentiation, we could detect CD7⁺CD3⁺TCRαβ⁺ cells (Supplementary Table 1). These cells harbored the same TCR β and γ chain rearrangements as the parental T-iPSC-1.10 line (Supplementary Fig. 2c). By day 30, CD3⁺TCRαβ⁺ cells typically accounted for ~80% of the cultures (Fig. 1c and Supplementary Table 1), and all of them expressed the CD19-specific CAR on their surface; day 30 cells are referred to hereafter as 1928z-T-iPSC-T cells (Fig. 1c). A substantial fraction expressed CD8α (10.4 ± 3.5%) and CD56 (20.7 ± 9.5%), whereas very few cells expressed low amounts of CD4 and almost no cells expressed detectable CD8β (Fig. 1c and Supplementary Table 1). Further surface phenotyping showed most cells to be CD5^low and negative for CD122 and TCRγδ (Fig. 1c, Supplementary Fig. 5a, b and Supplementary Table 1).

Taking advantage of the CD19-specific CAR, we evaluated the functional response of 1928z-T-iPSC-T cells to cell-bound CD19. 1928z-T-iPSC-T cells harvested on days 30–35 of differentiation were cultured on NIH-3T3–based artificial antigen-presenting cells (AAPCs) expressing the CD19 antigen (3T3-CD19) where indicated⁹. The 1928z-T-iPSC-T cells rapidly bound to 3T3-CD19 cells, forming clusters and eliminating the 3T3-CD19 monolayer (Fig. 1d). No such adhesion was observed when 1928z-T-iPSC-T cells were placed on CD19-negative 3T3 cells (Fig. 1d). Exposure to 3T3-CD19 cells also prompted 1928z-T-iPSC-T-cell surface expression of T-cell activation markers CD25 and CD69 (Fig. 1e) and secretion of type 1 cytokines such as IL-2, tumor-necrosis factor (TNF)-α and interferon (IFN)-γ (Fig. 1f). These results show that 1928z-T-iPSC-T cells displayed canonical features of T-lymphocyte function and specificity for the CD19 antigen.

To better elucidate the phenotype of 1928z-T-iPSC-T cells, we carried out a gene expression microarray, and we compared the mRNA expression profile of days 30–35 1928z-T-iPSC-T cells to that of naive B cells, CD4 T cells, CD8 T cells, CD3⁺CD56⁺T cells and natural killer (NK) cells isolated from peripheral blood. We also compared the profile to freshly isolated or in vitro–activated peripheral blood γδ T cells. Hierarchical clustering using the set of genes with most variable mRNA expression (s.d. > 0.75) showed that 1928z-T-iPSC-T cells were distinct from B cells and more closely related to the other T-lymphoid subsets and NK cells (Fig. 2a). Next, we directly compared 1928z-T-iPSC-T cells to particular lymphoid cell subsets by correlating mRNA expression levels of the most variable genes in our data set. This pairwise correlation analysis indicated that 1928z-T-iPSC-T cells were more similar to fresh or activated (for 7 d) γδ T cells (Supplementary Fig. 6a). This correlation was further confirmed upon examining expression of key lymphoid differentiation genes. The 1928z-T-iPSC-T cells expressed genes characteristic of the T-lymphoid lineage (e.g., GATA3, CD3δ, CD3ε, LEF1, LCK and BCL11B) at levels comparable to those of peripheral blood γδ T cells; however, the 1928z-T-iPSC-T cells did not express many genes characteristic of the NK cell lineage (e.g., CD94, CD16 and killer-cell immunoglobulin-like receptors) (Fig. 2b, Supplementary Fig. 5 and Supplementary Fig. 6b). Moreover, we detected pronounced expression of FASLG, TYROBP, CCL20, TNFSF11 (RANKL), CXCR6 and RORC, genes that are highly expressed in γδ T cells versus αβ T cells and/or NK cells¹⁶, in 1928z-T-iPSC-T cells (Fig. 2b and Supplementary Fig. 6b). The innate immune cell property of 1928z-T-iPSC-T cells is further supported by their expression of the transcription factor PLZF and the surface marker CD161 (Fig. 2c). Interestingly, 1928z-T-iPSC-T cells also showed high cytotoxic potential as indicated by high expression of TNFSF10 (TRAIL), GNLY, GZMB, FASL, LTA and low expression of co-inhibitory or exhaustion markers PD1, CTLA-4 and LAG3 (Fig. 2b). The majority of the CD3⁺ cells had a CD45RA⁺CD62L⁻CCR7⁻ effector memory phenotype (T_EMRA), although a small percentage (~6%) had a more naive CD45RA⁺CD62L⁺ phenotype (Fig. 2d). We did not detect expression of CD27 or CD28 receptors on the surface of 1928z-T-iPSC-T cells (Supplementary Fig. 5c).

iPSC-derived T cells will be therapeutically relevant only if they can be expanded while retaining functional properties. We therefore set out to expand 1928z-T-iPSC-T cells using 3T3-CD19 cells and characterize the expanded T cells. Starting from 3 × 10⁶ 1928z-T-iPSC, ~1–2 × 10⁵ 1928z-T-iPSC-T cells were obtained by day 30 of differentiation. We were able to expand those 1928z-T-iPSC-T cells 10- to 50-fold (mean = 20, s.d. = 15, n = 6) after one stimulation and up to ~1,000-fold after three weekly stimulations (Fig. 2e). Therefore, although the differentiation efficiency at day 30 is around 0.05, we were able to increase it to 0.5–1.0 by 1 week and up to 50.0 after 3 weeks of expansion. The expanded cells maintained their effector memory phenotype and upregulated the expression of natural cytotoxicity receptors such as NKp44, NKp46 and NKG2D (Fig. 2e). Interestingly, the expanded cells upregulated expression of T-lymphoid lineage–specific genes (ZAP70, GATA3, CD3δ, CD3ε, TRGC2) and downregulated expression of RORC, indicative of a switch toward a type 1 (Tbet/IFN-γ expressing) phenotype (Fig. 2f, g). CD161 surface expression was also reduced after expansion (Fig. 2d, e). In aggregate these findings suggest that CAR-mediated proliferation polarized the 1928z-T-iPSC-T cells toward a type 1 response. A similar phenotype switch has been shown for RORC-expressing T17 γδ T cells¹⁷, IL-17–producing fetal innate lymphoid cells¹⁸ and murine TCR-transgenic Th17 cells¹⁹, which polarize to type 1 cells after antigen or cytokine stimulation.

We first evaluated the cytotoxic potential of expanded 1928z-T-iPSC-T cells using an in vitro ⁵¹Cr release assay with EL4 murine lymphoma cells expressing CD19 or ovalbumin (nonspecific negative control) as targets⁹. Expanded 1928z-T-iPSC-T cells displayed high antigen-specific cytotoxic activity, even at low effector-to-target (E/T) ratios (Fig. 3a). To investigate the anti-tumor activity of 1928z-T-iPSC-T cells in vivo, we established a xenogeneic tumor model. Nonobese diabetic–severe combined immunodeficient NOD-SCID IL2Rγc^null mice were inoculated with the CD19⁺ Raji human Burkitt lymphoma cell line expressing a fluorescent luciferase fusion protein. For comparison to 1928z-T-iPSC-T cells, we transduced TCR-αβ and TCR-γδ peripheral blood lymphocytes from the same donor as the T-iPSC line expressing the 1928z CD19 CAR.

These three T-cell populations showed some phenotypic similarities and some differences. When we expanded 1928z-T-iPSC-T cells for a week, they displayed a T_EMRA phenotype (CD45RA⁺CD27⁻CD28⁻CCR7⁻), similar to the expanded 1928z-γδ T cells. In contrast, a sizeable fraction (33%) of 1928z-αβ T cells displayed a CD45RA⁻CD27⁺CD28⁺CCR7⁺ phenotype indicative of central memory cells (Fig. 3b). CAR expression was much lower on 1928z-T-iPSC-T cells (mean fluorescence intensity (MFI) = 395) than on 1928z-γδ (MFI = 1,212) or 1928z-αβ cells (MFI = 2,010) (Fig. 3b), which may influence therapeutic activity²⁰.

As shown by bioluminescent imaging, infusion of 1928z-T-iPSC-T cells delayed tumor progression to an extent similar to that induced by peripheral blood 1928z-γδ cells (Fig. 3c and Supplementary Fig. 7), and resulted in a significant survival advantage compared to tumor-bearing mice that were not treated with T cells (log-rank P = 0.042, Cox proportional hazards regression P = 0.036; Fig. 3d). Bioluminescence imaging further revealed that 1928z-T-iPSC-T and 1928z-γδ T cells initiated tumor regression more rapidly than 1928z-αβ cells (Fig. 3c). However, although initially slower at inducing tumor regression, the 1928z-αβ T cells did eventually induce complete tumor regression (Fig. 3c). These findings demonstrate that CAR⁺ T-iPSC-T cells can lyse tumor cells in vitro, elicit strong anti-tumor responses in vivo and provide a survival benefit in tumor-bearing animals, to the same degree as their closest natural counterparts.

The iPSC and CAR technologies, combined as shown here, potentially provide an unlimited source of T lymphocytes targeted to a chosen antigen, independent of HLA restriction. Under the present conditions, starting from T-iPSCs encoding a rearranged endogenous αβ TCR, we determined that the generated T cells have the properties of γδ T cells^17,21, although they express their endogenous αβ TCR on their surface. A similar lineage diversion has been observed in mice expressing TCRα and β transgenes^22–24, wherein T cells distinct from wild-type NK, NK T cells, or CD4⁺ or CD8⁺ αβ T cells displayed γδ T-cell features, including expression of CD8α and low expression of CD5, CD122 and NK1.1 (ref.22). T cells differentiated in vitro from human CD34⁺ hematopoietic progenitors genetically engineered to express an antigen-specific TCR display an NK cell–like phenotype²⁵. Together these observations suggest a possible effect of premature expression of TCRαβ, which may skew development toward innate lymphoid-like lineages. Although T-iPSC–derived expanded T cells have been reported to have a CD3⁺CD7⁺CD5^lowTCRαβ⁺CD56⁺ phenotype associated with expression of CD8α but not CD8β, they were not identified as γδ-like T cells⁶. Interestingly, we observed expression of the pre-rearranged endogenous TCRαβ on day 15 of differentiation on OP9-DL1 cells (Supplementary Table 1), earlier than in some other reports describing T-cell differentiation from human ESC- or iPSC-derived T cells^3,4. Importantly, we observed the same kinetics of T-cell development as in T-iPSC-1.10 in two other independent T-iPSC lines bearing different TCR rearrangements (Supplementary Fig. 2b and Supplementary Fig. 8), but not in cord blood–derived iPSC or in ESCs (data not shown). Altogether, these prior observations^22–25 and ours suggest that early expression of a transgenic or endogenous TCRαβ influences the T-cell differentiation process. In addition, some subtle features of our 1928z-T-iPSC-T cells, such as their CD8α⁺CD8β⁻ phenotype, expression of CD161 and low expression of CD5, are shared between adult γδ T cells^17,21 and innate-like T cells generated in fetal development^18,26. Together with previous reports, these observations suggest that these lymphoid cells may originate from a fetal cell–like hematopoietic stem cell intermediate committed to innate-like lymphopoiesis^4,27 and that in vitro differentiation from pluripotent stem cells may be intrinsically skewed toward embryonic characteristics²⁸. Notably, the CAR, which effectively supported T-cell expansion, did not seem to influence the acquisition of the γδ phenotype, as non-CAR–transduced T-iPSC-1.10 cells also yielded TCRαβ-expressing, γδ-like T cells (Supplementary Fig. 9). A complete understanding of the maturation of T-iPSC–derived T lymphocytes will be needed to further optimize their development and differentiation, generate different T-cell lineages and shape their functional attributes.

Tumor specificity is one of the essential characteristics of T lymphocytes used in adoptive T-cell therapy. Using the protocol described here, any HLA-independent antigen specificity can be imparted to any iPSC through an appropriate CAR⁸, without requiring the establishment of a patient-specific T-cell clone. To our knowledge no previously published study reported that genetic modification of human pluripotent stem cells with a receptor for antigen is an effective approach to generate T cells with therapeutic potential. 1928z-T-iPSC-T cells delayed tumor progression in vivo to a similar extent as peripheral blood–derived 1928z-γδ-T cells from the same donor. γδ T cells have some advantageous properties, such as low graft-versus-host reactivity and the ability to infiltrate solid tumors^29,30. Their anti-tumor activity has been demonstrated in several clinical settings, mainly against hematological malignancies³⁰.

CAR-modified T-iPSC–derived T cells may be especially valuable in situations where autologous or allogeneic T cells are not available. This is, for example, the case in immune-deficient patients such as HIV-infected or highly immunosuppressed patients with malignancies; in these scenarios autologous T-cell isolation and expansion is problematic or impossible. CAR-T-iPSC-T cells may also be useful in patients from whom the isolation of autologous tumor-infiltrating T lymphocytes has failed³¹, while providing the additional benefit of targeting alternative antigens recognized by CARs⁸. Other patients who could benefit from CAR-T-iPSC-T cells include those with acute leukemia who relapse after allogeneic hematopoietic cell transplantation and for whom the use of allogeneic donor lymphocyte infusions (DLI) is problematic. The efficacy of DLI in those patients is minimal, yet fraught with the risk of graft-versus-host disease³². CAR-T-iPSC-T cells could thus represent an additional option for patients who do not respond to DLI or for whom DLI use is not indicated due to high risk for graft-versus-host disease.

Several steps can be taken to avert the risks of immunological complications in the context of an off-the-shelf allogeneic CAR-T-iPSC-T therapy. The alloreactivity of T-iPSC–derived T cells, which express an endogenous TCR (Fig. 1a), can be eliminated by either disrupting the TCR, using target site–specific nucleases after T-cell differentiation³³, or by generating T-iPSCs from virus-specific T cells, which due to their recognition of a pathogen-derived antigen, are less likely to cause graft-versus-host disease³⁴. Allorejection of CAR-iPSC-T cells (which express HLA molecules) can be minimized by generating iPSCs from common HLA haplotypes (to ensure their histocompatibility with matched unrelated recipients³⁵) or by repressing HLA expression through additional genetic modification³⁶. Finally, the risk of insertional oncogenesis secondary to gene transfer can be decreased by integrating the CAR cDNA and other genes, such as suicide genes and noninvasive imaging reporters³⁷, at genomic safe harbor sites³⁸.

In summary, the combination of iPSC and CAR technologies that we describe here offers a potential new source of off-the-shelf T cells of predetermined antigen specificity. Considering the versatility of pluripotent stem cells and CAR engineering, this system may facilitate production of different T-cell subpopulations with additional genetic modifications and specificities suitable for a range of therapeutic indications.