Generation of antigen-specific mature T cells from RAG1−/−RAG2−/−B2M−/− stem cells by engineering their microenvironment

Pluripotent stem cells (PSCs) are a promising source of allogeneic T cells for off-the-shelf immunotherapies. However, the process of differentiating genetically engineered PSCs to generate mature T cells requires that the same molecular elements that are crucial for the selection of these cells be removed to prevent alloreactivity. Here we show that antigen-restricted mature T cells can be generated in vitro from PSCs edited via CRISPR to lack endogenous T cell receptors (TCRs) and class I major histocompatibility complexes. Specifically, we used T cell precursors from RAG1−/−RAG2−/−B2M−/− human PSCs expressing a single TCR, and a murine stromal cell line providing the cognate human major histocompatibility complex molecule and other critical signals for T cell maturation. Possibly owing to the absence of TCR mispairing, the generated T cells showed substantially better tumour control in mice than T cells with an intact endogenous TCR. Introducing the T cell selection components into the stromal microenvironment of the PSCs overcomes inherent biological challenges associated with the development of T cell immunotherapies from allogeneic PSCs.

Pluripotent stem cells (PSCs) are a promising source of allogeneic T cells for off-the-shelf immunotherapies.However, the process of differentiating genetically engineered PSCs to generate mature T cells requires that the same molecular elements that are crucial for the selection of these cells be removed to prevent alloreactivity.Here we show that antigen-restricted mature T cells can be generated in vitro from PSCs edited via CRISPR to lack endogenous T cell receptors (TCRs) and class I major histocompatibility complexes.Specifically, we used T cell precursors from RAG1 −/− RAG2 −/− B2M −/− human PSCs expressing a single TCR, and a murine stromal cell line providing the cognate human major histocompatibility complex molecule and other critical signals for T cell maturation.Possibly owing to the absence of TCR mispairing, the generated T cells showed substantially better tumour control in mice than T cells with an intact endogenous TCR. Introducing the T cell selection components into the stromal microenvironment of the PSCs overcomes inherent biological challenges associated with the development of T cell immunotherapies from allogeneic PSCs.
Cell-based immunotherapies in which autologous T cells have been engineered to express chimaeric antigen receptors and antigen-specific T cell receptors (TCRs) have produced impressive clinical responses in patients with otherwise treatment-refractory diseases [1][2][3][4] .Despite producing curative results, the widespread application of adoptive cell therapies is limited by the complex and costly manufacturing processes needed for the delivery of patient-specific treatments 2 .Currently, autologously engineered T cell products rely on adequate patient lymphocyte counts and on a clinical status that allows the incorporation of cell therapy [5][6][7] .In addition, clinical responses are uneven, at least in part because of the variability of the T cell composition of each patient product 8 .
To overcome these constraints, development of an 'off-the-shelf' engineered T cell product has garnered increasing interest.However, the requirement for an allogeneic source with this approach presents two main barriers: donor T cell-mediated alloreactivity (causing graft-versus-host disease) 9 and host-mediated rejection of human leucocyte antigen (HLA)-mismatched allogeneic donor cells 10 .

Article
https://doi.org/10.1038/s41551-023-01146-7(Fig. 1b).Although there was an early, transient population of CD8α + cells, they did not persist in cultures and lacked surface TCR expression (CD45 + CD5 + CD7 + CD8α + TCRαβ − CD3 − ), indicating that they did not undergo positive selection.Differentiation of DKO PSCs proceeded no further than the double-positive (DP; CD45 + CD5 + CD7 + CD8α + CD4 + ) T precursor stage, indicating a failure to undergo positive selection in the absence of endogenous TCRs (Fig. 1c).Output of DKO cultures peaked at 3 weeks, with a dramatic decrease in DP T precursors over the remainder of differentiation, consistent with 'death by neglect' (Fig. 1c,d and Supplementary Fig. 2d,e).Together, the differentiation profile of DKO PSCs showed that RAG1 and RAG2 deletion halts conventional T cell development at the DP T precursor stage and prevents positive selection, similar to previous studies characterizing T cell development from RAG-deficient PSCs 17,18 .

Positive selection of RAG1 and RAG2 knockout PSCs is MHC restricted
We next investigated whether lentiviral expression of a single exogenous TCR could overcome the halt in differentiation seen in DKO PSCs and restore positive selection.The WT and DKO clones of H1, an HLA-A*02:01 positive (A*0201 pos ) PSC line, and ESI017, an HLA-A*02:01 negative (A*0201 neg ) PSC line, were transduced with a lentiviral vector encoding the α-and β-chains of the HLA-A*02:01-restricted 1G4 TCR specific for the tumour-associated NYESO 157-165 peptide (hereafter, WT + TCR and DKO + TCR PSCs, respectively) 19,20 .
Lentiviral transduction of the 1G4 TCR in both H1 and ESI017 DKO PSC lines produced high levels of expression of the transgene in all resulting T cells after differentiation in ATOs (Fig. 2a).Expression of the transgenic TCR did not impair differentiation from either of the WT + TCR PSC lines, with robust production of mature, conventional, single-positive CD8αβ + T cells (SP8), regardless of MHC expression (Fig. 2b,c and Supplementary Fig. 3).
In contrast, when 1G4 TCR was expressed in DKO lines, only the A*0201 pos line (H1) generated SP8 T cells (Fig. 2b and Supplementary Fig. 3).The A*0201 neg (ESI017) line did not proceed beyond the DP T precursor stage (Fig. 2b and Supplementary Fig. 3), and cultures were dominated by CD4 − CD8 − (double negative) cells (Fig. 2b,d).Thus, in the RAG-deficient setting, positive selection only occurred in cultures initiated from PSCs that expressed the cognate MHC to which the transgenic TCR was restricted, showing that 'self-selection' occurs during T cell differentiation.
Although positively selected SP8s generated from H1 DKO + TCR PSCs showed a conventional CD8αβ + phenotype and naive T cell markers (CD8αβ + CD45RO − CD45RA + CD62L + ; Fig. 2c and Supplementary Fig. 3), the overall SP8 percentage and yield per ATO aggregate was significantly lower compared with that of the WT line, indicating that rescue of positive selection was incomplete with the exogenous 1G4 TCR in a RAG-deficient setting (Fig. 2d,e).

Engineering ATOs to induce positive selection from TKO PSCs
As class I MHC molecules were ablated in TKO + TCR PSCs, rescue of positive selection via the transgenic TCR could not be achieved through self-selection on MHC from PSC-derived cells (Fig. 3b).Therefore, to provide a positive selection signal for developing DP T Although the manipulation of these allogeneic mechanisms via gene editing of healthy donor peripheral blood (PB) T cells has been performed 7 , current techniques do not guarantee complete ablation of endogenous TCRs, and the process of gene editing carries the risk of genomic translocations and deleterious off-target events.Thus, there is a need to identify self-renewing sources of engineered T cells that allow for the validation and expansion of suitable clones after multiple cycles of gene editing.
Owing to their self-renewing capacity, human pluripotent stem cells (PSCs) are especially amenable to gene editing and clonal selection.A unique hurdle for the use of PSCs, however, is the complex in vitro process of haematopoietic and T cell development that must follow genetic manipulation of PSCs.Recent development of a three-dimensional culture method called the artificial thymic organoid (ATO) allows highly efficient and reproducible T cell differentiation with positive selection of mature, conventional CD8αβ + (CD3 + TCRαβ + CD8αβ + , hereafter SP8) T cells from multiple haematopoietic and PSC sources 11,12 .Induction of T cell development in the ATO system is accomplished using the murine MS5 stromal cell line engineered to express human delta-like ligand 4 (hDLL4) as a transgene.PSCs engineered using lentivirus to express a transgenic tumour-specific TCR 12 or chimaeric antigen receptor 13 can produce antigen-specific T cells after differentiation in ATOs; however, these previous studies were performed in cells with intact endogenous TCR and major histocompatibility complex (MHC) expression, thus allowing positive selection through a broad repertoire of endogenous TCRs and a diverse range of peptide MHCs (pMHCs) presented by PSC-derived cells (self-selection).
In this article, we report a method that provides all the necessary signals required for the positive selection of antigen-specific, mature, conventional CD8αβ + T cells from PSCs that lack all endogenous TCR rearrangements and MHC class I expression.To prevent the generation of any endogenous TCRs, recombination activation gene 1 (RAG1) and RAG2, which together facilitate V(D)J recombination of germline TCR loci 14 , was ablated via CRISPR-Cas9 gene editing in PSCs.To generate the class I MHC-null phenotype, β 2 -microglobulin (B2M) was targeted for knockout (KO) 15,16 .Full T cell differentiation, including positive selection, was achieved from RAG1 −/− RAG2 −/− B2M −/− PSCs by combining transgenic expression of a single TCR in PSCs, with exogenous expression of the cognate human MHC in the MS5-hDLL4 stromal line.The resulting SP8 T cells showed potent antigen-specific cytotoxicity in vitro and tumour control in vivo that was superior to PSC-derived T cells with intact endogenous TCR rearrangement.Overall, these results show that the presentation of a single human MHC complex in a murine stromal line can rescue developing human T cells that express a single, exogenous TCR.

KO of RAG1 and RAG2 prevents TCR rearrangement and T cell maturation
As RAG1 and RAG2 are located within an ~30 kb region of chromosome 11, complete deletion of their coding sequences was achieved with two single guide RNAs (sgRNAs; Fig. 1a and Supplementary Fig. 1a,b).Following identification of optimal sgRNAs (Supplementary Fig. 1c-g), single-cell RAG1 and RAG2 double KO (DKO) clones were generated from both the H1 and ESI017 parent embryonic stem cell lines, and then differentiated towards the T cell lineage using our previously developed protocol for the ATO system 12 (Fig. 1a and Supplementary Fig. 2a).Unedited (wild type, hereafter WT) PSCs proceeded through development normally, with surface expression of TCR and CD3 and spontaneous positive selection predominantly towards the CD8αβ + T cell lineage (Fig. 1b,c and Supplementary Fig. 2b,c).Differentiation of both DKO PSC lines proceeded normally into the T lineage-committed phase (CD45 + CD5 + CD7 + ; Supplementary Fig. 2d,e), but did not express surface TCR at any stage of differentiation, confirming a functional loss of endogenous TCR rearrangement      precursors, the MS5-hDLL4 line (hDLL4 hereafter) was further transduced to express HLA-A*02:01 (the cognate MHC for the 1G4 TCR), with or without human B2M (hB2M), generating the hDLL4-A*0201-hB2M (hDLL4-A02B) and hDLL4-A*0201 (hDLL4-A02) lines, respectively (Supplementary Fig. 5).A subset of the hDLL4-A02B stroma was transduced to also express cell adhesion molecule intercellular adhesion molecule 1 (ICAM1), which has been shown to interact with lymphocyte function-associated antigen 1 (LFA1) on T cells to increase the strength of signalling through the TCR and survival of thymocytes [21][22][23] .The resulting hDLL4-A*0201-hB2M-ICAM1 stroma is hereafter referred to as 'hDLL4-A02BI'.
Full differentiation of WT + TCR PSCs was induced in the ATO using either standard hDLL4 or modified stromal lines expressing the additional transgenes mentioned above (Fig. 3a,b and Supplementary Fig. 6a-c).In contrast, development of T cells from TKO + TCR PSCs was halted at the DP stage of development using either hDLL4 (Fig. 3b and Supplementary Fig. 6d-f) or hDLL4-A02 stroma (Supplementary Fig. 7a).In these conditions, DP T precursors from TKO + TCR PSCs decreased rapidly (Fig. 3b,c and Supplementary Fig. 7b), similar to previous experiments with DKO + TCR PSCs (Fig. 2d).

Transcriptomic analysis of TKO + TCR SP8 T cells
To characterize the transcriptomic identity of TKO + TCR SP8 T cells, single-cell RNA sequencing (scRNA-seq) was performed on isolated SP8 cells from TKO + TCR ATOs along with SP8s from WT PSC-derived controls (WT with hDLL4 stroma, WT + TCR with hDLL4 stroma and WT + TCR with hDLL4-A02BI stroma).In addition, scRNA-seq datasets from thymic SP8 (ref.27), whole PB 28,29 and unpublished PB SP8 datasets were incorporated for comparison.Dimensionality reduction through uniform manifold approximation and projection (UMAP) (Fig. 5a and Supplementary Fig. 10a,b) and quantification of SP8 T cell lineage genes (Fig. 5b and Supplementary Fig. 10c,d) indicated that TKO + TCR SP8 T cells closely resemble SP8s derived from unedited PSCs from thymus and PB.In addition, unsupervised hierarchical clustering and pairwise Pearson's correlation analysis of global gene expression for each sample showed that TKO + TCR SP8 T cells cluster and closely correlate with WT ATO-derived SP8 T cells, and with isolated thymic and PB SP8s (Fig. 5c).

TCR mispairing enables MHC-independent positive selection
Despite the presence of positive selection from TKO + TCR PSCs, overall SP8 T cell output was reduced compared with WT + TCR PSCs (Fig. 4a,b).As positive selection in the WT + TCR (A*0201 neg ) ES017 line occurred in the absence of the 1G4 TCR's cognate MHC (Fig. 2b,d), we hypothesized that in the presence of competent RAG1 and RAG2, mispairing between endogenous and exogenous TCR chains would broaden the range of pMHCs that could induce positive selection and circumvent the cognate MHC restriction of the 1G4 TCR.
To investigate the mechanism by which WT T cells carrying a transgenic TCR escaped MHC restriction, SP8 T cells were isolated from ATOs generated from WT, WT + TCR and TKO + TCR PSCs, and sequenced using the 10X Genomics platform to detect exogenous and endogenous TCR V α and V β chains at single-cell resolution.
Interestingly, endogenously rearranged TCR V β chains were also detected in addition to the exogenous 1G4 TCR in 21.00% ± 9.18% of WT + TCR SP8 T cells generated with hDLL4 stroma and in 13.32% ± 3.57% of WT + TCR SP8 T cells generated with hDLL4-A02BI stroma (Fig. 6a,c).Typically, allelic exclusion of endogenous TCR V β loci is observed when an exogenous TCR is expressed before V(D)J recombination 30,31 ; however, these results indicate that allelic exclusion of the endogenous TCR V β was not complete in the context of in vitro PSC differentiation of T cells (Fig. 6a).Whereas only a minority of T cells co-expressed endogenous TCR V β in addition to the exogenous TCR (Fig. 6a), the TCR V β repertoire was still broadly distributed (Fig. 6c).
As expected, the TKO + TCR SP8 T cells expressed only the transgenic 1G4 TCR, confirming that RAG1 and RAG2 deletion prevented all endogenous TCR V α and TCR V β rearrangement (n = 3, 7,930 cells total; Fig. 6a-c).Combined with the previous observation that selection in DKO + TCR PSCs was MHC restricted (Fig. 2a,b and Supplementary

Functional characterization of TKO PSC-derived SP8 T cells
To explore the functional consequences of TCR mispairing, in vitro and in vivo assays were performed on WT + TCR and TKO + TCR SP8 T cells isolated from week 6 ATO cultures.Before the functional assays, SP8 T cells were expanded with aAPCs expressing the cognate antigen (NYESO 157-165 ).Similar to WT + TCR SP8 T cells, TKO + TCR SP8 T cells showed polyfunctional cytokine production and CD107α mobilization in response to stimulation with aAPCs expressing the cognate antigen but not an irrelevant (MART1 26-35 ) peptide (Fig. 7a).Both WT + TCR and TKO + TCR SP8 T cells upregulated surface CD25 and 4-1BB after overnight stimulation with cognate antigen (Fig. 7b) and underwent antigen-specific proliferation (Fig. 7c).Despite uniform expression of the transgenic V β 13.1 TCR chain, ~60% of WT + TCR SP8 cells did not stain with tetramer for the 1G4 TCR, consistent with mispairing of TCRs at the cell surface (Fig. 7d).In contrast, TKO + TCR SP8 T cells retained a 1:1 ratio of 1G4 TCR tetramer to transgenic V β 13.1 staining (Fig. 7d).TKO + TCR SP8 T cells showed robust loss of surface TCR in response to cognate antigen (consistent with intracellular relocation), whereas 60% of WT + TCR SP8 T cells retained surface TCR (Fig. 7e).Expanded WT + TCR and TKO + TCR T cells also showed potent antigen-specific cytotoxicity in vitro (Fig. 7f).Both WT + TCR and TKO + TCR SP8 T cells were able to undergo robust expansion over input SP8, reaching >250-fold expansion over 5 cycles of antigen-specific activation in the presence of interleukin-7 (IL-7) and interleukin-2 (IL-2) (Fig. 7g).
Expanded TKO + TCR and WT + TCR SP8 T cells were subjected to one freeze-thaw cycle and then further expanded with aAPCs expressing NYESO 157-165 peptide before testing in an in vivo tumour challenge.Immune-deficient (NOD scid gamma (NSG)) mice were intravenously (I.V.; tail vein) engrafted with luciferase-expressing NALM6 tumour cells engineered to express NYESO 157-165 peptide as a single-chain trimer

Article
https://doi.org/10.1038/s41551-023-01146-7 (SCT; Fig. 8a).Engraftment of NALM6 cells was confirmed after 5 days by imaging, before mice were stratified into groups with equivalent tumour signal (n = 9 for each group; Fig. 8b).After stratification, SP8 T cells were injected I.V. (retro-orbitally) and monitoring of tumour progression was performed by imaging (Fig. 8a-c).IL-2 (10,000 IU per mouse) was provided intraperitoneally during the course of the experiment (Fig. 8a).At the conclusion of the experiment, TKO + TCR SP8 T cells showed significantly improved tumour control compared with mice that received either no T cells or WT + TCR SP8 T cells (Fig. 8b-d), and they showed improved survival compared with mice that received no T cells (Supplementary Fig. 10e).

Discussion
The challenges inherent in using an allogeneic source for adoptive T cell therapy take on a new level of complexity in the setting of PSC-derived T cells, where the recapitulation of T cell differentiation requires the same TCR and MHC machinery that must be removed to avoid alloreactivity.Here we report that, by introducing the components required for positive selection into separate cellular sources, it is possible to generate antigen-restricted, mature, conventional T cells from PSCs in which all endogenous TCR and class I MHC expression has been removed.Specifically, two transgenic class-I-restricted TCRs were expressed in PSCs to circumvent the loss of endogenous TCR rearrangement, and cognate MHC was provided by a supportive stromal line, also engineered to provide other critical external signals during T cell development.Whereas SP8 differentiation efficiency varied based on the TCR, the dependence of in vitro T cell differentiation on the MHC expressed by PSC-derived cells (self-selection) was removed, and the murine stromal cells were used as a surrogate for the thymic epithelial cells that normally provide these critical signals in the thymus.Production of SP8 T cells from PSCs has been achieved by other groups after TRAC or RAG disruption to prevent expression of endogenous TCRs; all such approaches used agonist or antigen-specific stimulation of cells arrested at the DP stage in monolayer or non-stromal T cell differentiation cultures [24][25][26] .Although CD8αβ T cells could be generated using such approaches with ATO-derived DPs, the phenotypes were different from CD8αβ T cells spontaneously generated in ATO cultures, both after initial DP stimulation (based on lack of CD62L expression) and further expansion (based on persistent CD45RA expression in DP stimulated and expanded cells).In contrast, the ATO-derived

Article
https://doi.org/10.1038/s41551-023-01146-7 SP8s initially appeared to be phenotypically closer to mature, naive T cells, and converted to a T effector memory phenotype with further expansion; the functional relevance of these phenotypic differences will require rigorous analysis in future studies.By deleting both RAG1 and RAG2 genes, we avoided the possibility of residual recombination activity that can occur with single gene disruption 25,26,[32][33][34] .Furthermore, the absence of observable defects in function or cytotoxicity with full RAG ablation allowed us to manipulate T cell differentiation in the complete absence of endogenous TCRα and TCRβ rearrangement.
Consistent with data from patient samples 17,18 , RAG1 −/− RAG2 −/− PSCs were able to differentiate as far as the DP stage and cell cultures failed early.Although expression of a transgenic TCR produced some mature SP8s from RAG1 −/− RAG2 −/− PSCs, cell output was low and most cells failed to advance beyond the double-negative stage, suggesting that self-selection is inefficient when it relies on a single TCR.The more robust output of mature T cells from unedited PSCs that were transduced with the same A*0201-restricted 1G4 TCR was likely due to the wider array of MHC interactions provided by mispairing between the 1G4 TCR and the broad repertoire of endogenous TCRs produced in RAG-competent cells.This conclusion is further strengthened by the efficient T cell differentiation even from A*0201 neg RAG-competent PSCs that expressed the 1G4 TCR.
It was previously shown that TCRα rearrangement occurs from induced PSCs from reprogrammed T cells in which a pre-rearranged TCR exists 25 .The TCR V β rearrangements that we noted to also be present in the RAG1-and RAG2-competent cells showed that allelic exclusion induced by the exogenous 1G4 V β chain was not absolute; however, expression of these low-frequency V β chains was barely, if at all, detectable on the cell surface by flow cytometry, and the functional relevance of potential V β mispairing is unclear 11,12 .
Although conventional, mature SP8 T cells generated from TKO + TCR PSCs were transcriptionally and functionally similar to WT + TCR PSC-derived SP8 T cells based on in vitro proliferation and killing, TKO + TCR T cells proved to be significantly superior in controlling tumour growth in vivo.In line with previous studies 25,[35][36][37][38] , our results are consistent with decreased potency of antigen-specific killing due to TCR mispairing, and we found that this effect was particularly important under in vivo conditions; however, we note the ability of such models to predict clinical responses remains to be determined.
In contrast to the poor efficiency of positive selection through self-selection of a single transgenic TCR and intact MHC in the PSCs, a significant improvement in positive selection and cell output was seen when fully human class I MHC (including hB2M) was instead provided by the stromal component of the cultures.It is intriguing to speculate how the MHC-TCR interaction may be quantitively or even qualitatively different when it involves stromal-T, rather than T-T, interaction, but it is conceivable that the mechanosensory stimulation of the TCR is different in each case 39,40 .
The murine stromal cell line was further engineered to provide an additional co-factor that could enhance positive selection.ICAM1 has been reported to potentiate TCR activation and antigen sampling through LFA1 interactions at low pMHC densities [21][22][23] , which is an inherent nature of positive selection when T cells are scanning for low-affinity, 'weak' peptides.It has also been shown that ICAM1-LFA1 signalling leads to cytoskeletal remodelling, increased TCR signal transduction, and both NOTCH-responsive and survival gene expression 41 .Although the direct mechanism of ICAM1 during T cell development was not characterized, its inclusion in the murine MS5 stromal cells significantly increased output of SP8 T cells, while having no observable effects on the ratios of developing T cells.Future studies could focus on delineating the effects of including additional signals and co-factors at the immune synapse similar to those provided by thymic epithelial cells during physiological thymopoiesis.In addition, further studies to understand how these manipulations affect TCR signalling during positive selection would be of interest, although the analyses will be challenging given the dynamic nature of complex in vitro development systems, such as the ATO, which comprise heterogeneous populations of differentiating cells.
In summary, we have shown that both stem cells and the in vitro microenvironment can be modified to deliver the necessary signals when alloreactive mechanisms are disrupted.To avoid all immune-mediated rejection of allogeneic T cells, additional manipulations of the class I MHC-null T cells to avoid natural killer (NK)-mediated rejection will be required [42][43][44] , along with the deletion of class II MHC.Notably, we have shown that the block to positive selection caused by the removal of class I MHC expression in allogeneic PSCs can be rescued by pairing an exogenous TCR in developing T cells with an engineered murine stromal microenvironment.In addition to providing a cognate MHC, the accessory stromal cells of the ATO system can be modified to deliver other extrinsic signals or co-stimulation factors that improve T cell function and specification, providing an experimental method to dissect T cell differentiation and positive selection in vitro.Owing to their inherent self-renewal capacity, gene-edited PSCs and their stromal counterparts can be readily engineered with additional strategies that will augment T cell function and immune evasion for immunotherapy.

Cell lines
The MS5-hDLL4 cell line was generated in our laboratory as previously described 12 .For clarity, we reproduce the following description from ref. 12.Briefly, MS5 cells 45 were transduced with a lentiviral vector encoding full-length hDLL4.The highest 5% of DLL4-expressing cells were isolated by fluorescence-activated cell sorting (FACS) using an anti-hDLL4 antibody and passaged in DMEM with 10% fetal calf serum (FCS).Stable expression was confirmed by flow cytometry for hDLL4 expression after several weeks of culture, along with quantitative RT-PCR and RNA sequencing.
For the cell lines including class I MHC and scaffolding proteins, the previously derived MS5-hDLL4 line was transduced with varying combinations of individual lentiviruses encoding human HLA-A*02:01, hB2M and human ICAM1 (hICAM1).The highest 5% of transduced cells were isolated by FACS using antibodies detecting HLA-A*02:01, hB2M and hICAM1, and then passaged in DMEM with 10% FCS for expansion and cryopreservation.Stable expression was also confirmed by flow cytometry for hDLL4, HLA-A*02:01, hB2M and hICAM1.aAPCs were generated in our laboratory as previously described 12 .For clarity, we reproduce the following description from ref. 12. K562 cells (catalogue number CCL-243; ATCC) were transduced with lentiviral vectors encoding full-length human CD80, CD83, CD137L and HLA-A*02:01-B2M-NYESO 157-165 SCT (gifts from David Baltimore; Caltech).Target cells for the cytotoxicity assay were created by transduction of K562 with either NYESO1 or MART1 SCTs.
The codon-optimized TCR V α and V β (including V β 13.1) chains of a TCR specific for HLA-A*02:01/NYESO 157-165 (derived from the 1G4 TCR 20 ) were previously described 19 (gift from Antoni Ribas; UCLA).TCR coding sequences and the mTagBFP2 fluorescence protein 46 , all separated by furin-SGSG-2A linkers, were subcloned into the third-generation pCCL lentiviral vector downstream of the ubiquitin C promoter with intron 1.

Human pluripotent cell lines
The human Embryonic Stem Cell Research Oversight Committee and Institutional Review Board have approved all protocols for the use of PSCs for this study.The human embryonic stem cell lines 47
For each gene target, the two gRNA candidates with the highest on-target in vitro cutting activity were chosen for off-target cleavage activity in vitro via the genome-wide, unbiased identification of double-stranded breaks enabled by sequencing (GUIDE-seq) method 56 .One gRNA with high on-and low off-target cutting activity was chosen for each target gene to proceed with editing in PSC lines.

Gene editing of human PSC lines
CRISPR-Cas9 gene editing of PSCs was performed with ribonucleoprotein (RNP) complexes 57 of purified spCas9-NLS (QB3 MacroLab, University of California, Berkeley) and custom-synthesized sgRNAs (Synthego).spCas9-NLS and gRNAs were stored at −80 °C before use for gene editing of PSCs.
Briefly, PSCs were allowed to grow in wells of a Matrigel-coated 6-well plate until reaching ~70% confluency before being dissociated with TrypLE Express (catalogue number 12604-013; Thermo Fisher Scientific) and resuspended in single-cell solution.Before nucleofection, 84 pmol of each gRNA was complexed with 60 pmol of spCas9, individually, at a ratio of 1 pmol spCas9 to 1.4 pmol gRNA for 15 min at room temperature.PSCs were resuspended at a concentration of 2 × 10 5 cells in 14 µl of P4 Primary Cell Nucleofector Solution (P4 Primary Cell 4D-Nucleofector X Kit S, catalogue number V4XP-4032; Lonza).
For sgRNA reactions, fully complexed RNPs (60 pmol spCas9:84 pmol gRNA) were added to resuspended cells and the volume was brought up to 20 µl using P4 Primary Cell Nucleofector Solution (Lonza).For dual gRNA reactions, RNPs were complexed individually and then added (2× 60 pmol spCas9 total) into the cell suspension with a custom-synthesized single-stranded oligonucleotide donor template (ssODN, 100 bp, resuspended at 100 µM; Ultramer DNA Oligo; Integrated DNA Technologies) to a final concentration of 3 µM in solution.Twenty microlitres of combined PSC, RNP and/or ssODN solutions were added into individual wells of the 16-well Nucleocuvette Strip (Lonza).Nucleofection was performed on the 4D-Nucleofector Core and X Unit (catalogue numbers AAF1003B and AAF-1003X; Lonza) using pulse and frequency code CB-150.Cells were allowed to rest in the cuvette for 10 min before transferring into 1.5 ml mTeSR Plus and ROCK inhibitor Y-27632 dihydrochloride (10 µM), and then plated in 1 well of a Matrigel-coated 12-well plate.Culture medium was changed daily, and ROCK inhibitor Y-27632 dihydrochloride (10 µM) was removed from the medium after 48 hours.Edited PSCs were allowed to reach ~70% confluency before being expanded for single-cell cloning and cryopreservation.

Single-cell cloning of edited human PSC lines
Single-cell cloning was achieved with low-density plating of expanded, edited PSCs 54 .Briefly, expanded, edited PSCs were dissociated into single-cell solution with TrypLE Express (catalogue number 12604-013; Thermo Fisher Scientific) and then plated in Matrigel-coated 10 cm dishes at a density of 0.5-1 × 10 4 cells per plate in mTeSR Plus complete culture medium with ROCK inhibitor Y-27632 dihydrochloride (10 µM).Culture medium was changed daily, and ROCK inhibitor Y-27632 dihydrochloride (10 µM) was removed from the medium after 48 hours.After colony formation, 24-48 individual colonies were scraped with a 200 µl 'P200' pipette tip under a microscope, and then they were transferred into individual wells of a Matrigel-coated 12-well plate with mTeSR Plus culture medium.Once cells reached 60-80% confluency, cells were passaged via scraping for expansion and genotyping PCRs to determine bi-allelic KO of edited genes.Clones with bi-allelic KOs were expanded, cleaned for differentiation, and then genotyped once again before cryopreservation and karyotyping.

Article
https://doi.org/10.1038/s41551-023-01146-7 Transduction of human PSC lines NYESO TCR-transduced PSC lines were generated by transduction of unedited or KO H1 or ESI017 PSCs with either lentiviral vectors encoding the 1G4 TCR (HLAA*02:01 restricted recognizing NYESO 157-165 peptide) or the F5 TCR (HLA-A*02:01 restricted recognizing MART1 26-35 peptide) and the fluorescence marker mTagBFP2.Briefly, PSCs were dissociated into single-cell suspension and plated at a density of 2 × 10 5 cells per well of a Matrigel-coated 6-well plate in mTeSR Plus culture medium with ROCK inhibitor Y-27632 dihydrochloride (10 µM).The following day, the culture medium was changed to 1 ml of mTeSR, and concentrated lentiviral supernatant was added directly into the wells.The medium was changed each day until cells reached ~70% confluency, when cells were dissociated with TypLE Express (catalogue number 12604-013; Thermo Fisher Scientific) and purified via FACS sorting using the phenotype TRA1-81 + mTagBFP2 + .Isolated cells were returned to culture on Matrigel-coated 6-well plates and mTeSR Plus culture medium for expansion and cryopreservation.
For experiments involving the use of engineered stromal lines for ATO T cell differentiation (MS5-hDLL4-A*0201, MS5-hDLL4-A*0201-hB2M and MS5-hDLL4-A*0201-hB2M-ICAM), EMOs were collected in bulk after 14 days (day 0) by adding 1 ml MACS buffer (PBS, 0.5% bovine serum albumin and 2 mM EDTA) to each filter, briefly disaggregating the ATO by scraping with a 1 ml 'P1000' pipette and then passed through a 50 µm nylon strainer.Collected cells were counted, and then 2 × 10 4 cells were combined with 2.5 × 10 5 engineered MS5-hDLL4 stromal cell lines per ATO in 1.5 ml Eppendorf tubes.Cell mixtures were centrifuged at 300g for 5 min at 4 °C in a swinging bucket centrifuge before the formation of 3 ATO aggregates (5 µl per aggregate) per 0.4 mm Millicell transwell insert (catalogue number PIMC0R5G50; EMD Millipore), similarly to the EMO aggregate formation described above.

Isolation of ATO-derived SP8 and DP precursor cells
ATOs were collected by adding MACS buffer (PBS, 0.5% bovine serum albumin and 2 mM EDTA) to each filter, briefly disaggregating the ATO by pipetting with a 1 ml P1000 pipette, and then they were passed through a 50 µm nylon strainer; SP8 T cells were collected after 6 weeks of T cell differentiation, and DP precursors were collected after 3 weeks.For bulk-scale collection of ATOs, aggregates were collected in a similar fashion; however, up to 150 aggregates were collected in a well filled with MACS buffer and then they were transferred onto a 50 µm nylon filter, where aggregates were physically dissociated on top of the filter using the back end of a sterile 1 ml syringe.After dissociation, the filter was washed with MACS buffer and cell mixtures were centrifuged at 300g for 5 min at 4 °C in a swinging bucket centrifuge.
Activation and expansion of DP precursors was also performed using Dynabeads Human T-Activator CD3/CD28 (catalogue number 11131D; Thermo Fisher Scientific) according to the manufacturer's protocol.Briefly, beads were added in a 1:1 ratio of T cells to beads in AIM V (catalogue number 12055083; Thermo Fisher Scientific) supplemented with 5% human AB serum (catalogue number 100-512; Gemini Bio-Products), 20 ng ml −1 rhIL-2 (catalogue number 200-02; Peprotech) and 5 ng ml −1 rhIL-7 (R&D Systems).Beads were replenished every 7 days, fresh medium was added every 2-3 days and cells were replated into larger wells as necessary.

Single-cell RNA library preparation and sequencing
ATO-derived T cells were isolated as described above and then FACS sorted for ATO-derived SP8 T cells from WT (mCD29 − CD45 + CD8α + CD4 − ) and 1G4 TCR-transduced (mCD29 − mTagBFP2 + CD45 + CD8α + CD4 − ) PSC ATOs into PBS + 0.04% bovine serum albumin.Cells were counted and resuspended at a concentration of ~1,000 cells per µl and provided to the Technology Center for Genomics and Bioinformatics (TCGB) core for unique molecular identifier (UMI) tagging and generation of gene expression (GEX) and human TCR repertoire (V(D) J) libraries using the 10X Chromium Next GEM Single Cell V(D)J Reagent Kit v.1.1 (10X Genomics).Similarly, PB SP8s were FACS isolated (CD45 + CD8α + CD8β + CD4 − TCRαβ + CD3 + ) from PB mononuclear cells obtained from healthy donors, and provided to the TCGB core for UMI tagging and library generation.Fully constructed libraries for all samples were run in one S4 flow cell on the Illumina NovaSeq Platform.

GEX and TCR reference genomes
The reference genomes used for GEX and TCR alignment, both of which correspond to the GRCh38 genome, were downloaded from the 10X Genomics website.To detect the exogenous TCR, which was expressed using a codon-optimized sequence, the GRCh38 reference genome was customized using the Cell Ranger v.7.0.0 (10X Genomics) 'mkref' pipeline.

scRNA-seq data filtration and cleaning
Sequenced reads from each sample were aligned to the human reference genome GRCh38 and filtered for empty droplets using the Cell Ranger v.7.0.0 (10X Genomics) 'multi' pipeline that generated count matrices from the GEX libraries and assembled full TCR contigs from the V(D)J libraries.On average, we achieved >30,000 mean reads per cell for GEX expression libraries and >7,000 mean reads per cell for V(D)J libraries.
For samples that were acquired for this study, GEX (RNA) count matrices were loaded with Seurat v.4.3.0 (Satija Lab) 60 .As V(D)J libraries were also generated from the same complementary DNA, TCR genes aligned by whole transcriptomic sequencing were removed to prevent clustering bias from samples with intact germline TCR recombination.Count matrices were loaded separately and barcoded cells were filtered for cells with outlier UMI counts (low-quality cells and doublets), high mitochondrial gene expression (due to cellular stress or loss of cytoplasmic RNA) and low number of sequenced genes.After initial data filtration, individual datasets were bioinformatically cleaned for doublets using DoubletFinder v.2.0.3 (ref.61), and cells were cleaned for stochastic dropouts of identity genes based on their sorted SP8 phenotype (CD8α, CD8β, CD3 > 0 and CD4 < 0).Cleaned cells were then batch corrected for technical and biological variations using the reciprocal principal component analysis (RPCA) integration method in Seurat.For integration of the combined ATO-derived SP8 T cell samples, molecular count data for each sample were individually normalized and variance stabilized using SCTransform in Seurat, and then cell cycle phase scores were calculated for each individual sample based on the expression of canonical cell cycle genes within a barcoded cell.Following cell cycle scoring, raw counts were normalized and variance was stabilized again using SCTransform in Seurat with the additional steps of regressing calculated cell cycle scores and mitochondrial genes to mitigate the effects of cell cycle heterogeneity and mitochondrial gene expression.
To perform RPCA integration of ATO-derived SP8 samples, highly variable genes (nfeatures = 5,000) were identified and then used to find integration anchors between datasets.Clustering of the fully integrated object of ATO-derived SP8 samples was performed using the IKAP (Identifying K mAjor cell Population group in scRNA-seq analysis) package 62 .IKAP analysis identified the optimal principal components and clusters for the dataset.Of the three clusters identified, one cluster was removed based on high mitochondrial gene expression, indicating stressed cells or loss of cytoplasmic RNA, and the other two were identified as immature (CD45RO + ) and mature, naive (CD45RA + ) SP8 T cells based on differentially expressed genes.Mature, naive CD45RA + SP8 T cells were used for downstream analyses.
For publicly available datasets from human thymic SP8 (GSE148981; ref. 27), PB NK and CD14 + monocyte (PB monocytes) cells (10X Genomics 28,29 ), raw sequencing reads were downloaded from their respective repositories and then aligned to the human reference genome GRCh38 using the Cell Ranger v.7.0.0 (10X Genomics) 'count' pipeline, similarly to method described above, which only generated count matrices from the GEX reads.Individual samples were loaded, filtered and cleaned using Seurat v.4.3.0 (Satija Lab), similarly to that described above.Total PB samples were SCTransform normalized in Seurat and variance stabilized before integration, and principal components were identified from the integrated object.Then, dimensional reduction (UMAP) and cell clustering was performed on the integrated object using the functions included in the Seurat package (Satija Lab).PB NK and CD14 + monocytes (PB monocytes) and PB SP8 T cells were identified and subset out of the integrated object for downstream analyses.

Analysis and visualization of scRNA-seq data
Cleaned samples were merged into a single Seurat (Satija Lab) object, preserving only the raw RNA count matrices for each object.Each sample was then normalized and variance stabilized using SCTransform in Seurat before regressing cell cycle phase scores and mitochondrial genes, as described above, before remerging the list back into a single object.Given the individual SCTransform models calculated from each sample, the minimum median UMI was used to recorrect the counts and data slot for downstream analysis using PrepSCTFindMarkers in Seurat (Satija Lab).Scaled expression values of highly variable genes, which were identified as the union of the top 5,000 genes from each individual sample during SCTransform in Seurat, were used to calculate principal component analysis for dimensionality reduction using the UMAP function in Seurat (Satija Lab).Individual cell populations (SP8, NK and monocytes) clustered together, and minor outliers were manually cleaned for gene expression visualization using the DotPlot function in Seurat (Satija Lab).
To perform global gene expression analysis between individual samples and their respective sources, a pseudobulk approach was taken by extracting the raw counts and metadata from the object, described above, and creating an object using the SingleCellExperiment package 63 .Individual samples were assigned identities based on cell type identity and source to create a DEseq2 (ref.64) object in R, and pseudobulked counts were normalized using the regularized-logarithm transformation (rlog) function.Pearson's correlations of global gene expression for all pairwise combinations between each sample were calculated from the rlog-normalized data using the cor() function in R v. 4.2.1 (ref.65).Dendrogram and hierarchical clustering analysis were visualized using the pheatmap package 66 .
For the purposes of analysing SP8 T cells without the addition of PB monocytes and PB NK cells, SP8 T cells were subset out of the object and each individual sample was normalized using SCTransform in Seurat and integrated using the RPCA integration method in Seurat with nfeatures = 5,000 highly variable genes.Clustering of the fully integrated object of ATO-derived SP8 samples was performed with the IKAP package 62 , which identified the optimal principal components and clusters for UMAP dimensionality reduction of the integrated object.Dendrogram, hierarchical clustering and global gene expression analyses were performed as described above.

TCR repertoire analysis by scRNA-seq
Full, endogenous TCR V α and V β contigs were aligned and assembled using the Cell Ranger v.7.0.0 (10X Genomics) multi pipeline, as described above.As the exogenous TCR was expressed using a codon-optimized sequence, reads from the 5′ GEX libraries were aligned using a custom human genome reference (GRCh38), described above, which included codon-optimized sequence for the 1G4 TCR.After full reconstruction of TCRs and gene alignment, Cell Ranger output matrices were filtered using Seurat v.4.3.0 (Satija Lab), as described above.For samples including the exogenous 1G4 TCR, barcoded cells were also extracted for cells that expressed the exogenous transgene.Finally, the remaining barcoded cells were exported as a list and used to filter V(D)J sequencing outputs in R for TCR diversity calculations, using the number of barcoded cells remaining after quality control as a denominator, and calculations of percentages of cells expressing endogenous and/or exogenous TCR.

In vitro cytotoxicity assays
Expanded SP8 T cells from ATOs were rested as described above.For cytotoxicity assays, 2-fold serial dilutions of rested SP8 T cells were performed in 96-well U-bottom plates starting at 1 × 10 5 cells in 200 µl AIM V (Thermo Fisher Scientific) with 5% human AB serum (Gemini Bio-products).K562 aAPCs expressing cognate (NYESO) or irrelevant (MART1) SCTs were plated at 1 × 10 5 target cells per well.Apoptotic cell death of target cells was quantified by Apotracker Green (BioLegend) and DAPI staining at 6 hours.Target cell death was calculated by subtracting the percentage of Apotracker Green-positive target cells in wells receiving no T cells from wells that received T cells.

Flow cytometry and antibodies
Staining for flow cytometry was performed in PBS with 0.5% bovine serum albumin and 2 mM EDTA for 15 min at 4 °C in the dark.TruStain FcX

Fig. 1 |
Fig. 1 | Inhibition of T cell development at the DP stage of T cell differentiation by DKO of RAG1 and RAG2.a, Schematic showing the generation of clonal RAG1 and RAG2 DKO PSCs derived from the H1 or ESI017 parent embryonic stem cell lines and their differentiation within the ATO system.b, Representative flow cytometry analysis of ATOs generated from unedited (WT) and DKO PSCs from the HLA-A*02:01 positive (A*0201 pos ) H1 and HLA-A*02:01 negative (A*0201 neg ) ESI017 parent lines.Surface expression of TCRαβ and CD3 from T lineage-committed cells is shown at the indicated time points.c, Temporal dynamics of CD8α and CD4 expression in T lineage-committed cells from WT and DKO PSCs differentiated in the ATO system are shown at the indicated time points.In b and c, the gating strategy is shown above the plots, and the numbers in the plots indicate the percentage of cells within each gate.Gates for individual populations are drawn on flow cytometry plots.Fluorophores are as follows: DAPI (4′,6-diamidino-2-phenylindole); PE (phycoerythrin); APC (allophycocyanin); BV (Brilliant Violet).d, The percentage of T cell populations from the DAPI − mCD29 − CD56 − CD45 + CD5 + CD7 + gate in c at the indicated time points (data shown as mean ± s.e.m.; H1, n = 3 independent experiments; ESI017, n = 4 independent experiments).Populations are defined as DP (CD8α + CD4 + ) and mature SP8 (CD8α + CD4 − TCRαβ + CD3 + ).Chr 11, chromosome 11.

Fig. 2 |
Fig. 2 | Class I MHC-restricted rescue of T cell development by 1G4 TCR in RAG1 and RAG2 DKO PSCs.a,b, Representative flow cytometry analysis of WT and RAG1 and RAG2 DKO PSCs from both parent lines, H1 (A*0201 pos ) and ESI017 (A*0201 neg ), stably transduced to express the HLA-A*02:01-restricted 1G4 TCR (WT + TCR and DKO + TCR).a, Surface expression of the transgenic V β 13.1 and CD3 (gated as shown) during ATO differentiation of PSC-derived cells at the indicated time points.b, Differentiation kinetics of V β 13.1 + CD3 + T cells based on CD8α and CD4 expression at the indicated time points of ATO differentiation.The gating strategy is indicated above the plots, and the numbers in the plots indicate the percentage of cells within each gate.c, At 7 weeks of T cell differentiation, ATO-derived H1 WT + TCR and DKO + TCR SP8 T cells (gated as shown) were analysed for maturation markers of conventional T cells as shown; Articlehttps://doi.org/10.1038/s41551-023-01146-7

Fig. 4 |
Fig. 4 | Output and phenotypic characterization of SP8 T cells from RAG1, RAG2 and B2M TKO PSCs.a, Output of SP8 T cells per WT + TCR or TKO + TCR PSC ATO aggregated with hDLL4 or hDLL4-A02BI stroma over the 7 week course of T cell differentiation (mean ± s.e.m.; P values are shown compared with TKO + TCR with hDLL4-A02BI stroma; *P < 0.05, **P < 0.01, ***P < 0.001, two-tailed unpaired t-test; test is not significant if no P value is indicated; WT + TCR with hDLL4 stroma, n = 4 independent experiments; WT + TCR with hDLL4-A02BI stroma, n = 3 independent experiments; TKO + TCR with hDLL4 stroma, n = 4 independent experiments; TKO + TCR with hDLL4-A02BI stroma, n = 5 independent experiments).b, SP8 max T cell output per ATO reached over the 7 week course of T cell differentiation.The mean ± s.e.m. (*P < 0.05, two-tailed Mann-Whitney U-test) is shown for each group (WT + TCR with hDLL4 stroma, n = 4 independent experiments; WT + TCR with hDLL4-A02BI stroma, n = 3 independent experiments; TKO + TCR with hDLL4 stroma, n = 4 independent experiments; TKO + TCR with hDLL4-A02BI stroma, n = 5 independent experiments).c, At 7 weeks of T cell differentiation, TKO + TCR PSC-derived SP8 T cells were analysed for maturation markers of conventional T cells.The gating strategy is indicated above the plots, and the numbers within the plots indicate the percentage of cells within each gate.

Fig. 7 |
Fig. 7 | Functional characterization of PSC-derived, antigen-specific T cells in vitro.a-f, WT + TCR and TKO + TCR (1G4 TCR) SP8 T cells were isolated from week 6 ATOs, both generated with hDLL4-A02BI stroma and expanded with K562 aAPCs expressing the cognate antigen (NYESO), IL-2 and IL-7 before in vitro functional assays.The numbers indicate the percentage of cells within each gate.a, Expanded T cells were rested and then stimulated for 6 hours with K562 aAPCs presenting non-specific (MART1) or cognate (NYESO) antigen as an SCT to assay cytokine production and CD107α upregulation (gated on Zombie NIR − mStrawberry − TCRαβ + CD3 + CD8α + cells).Data are representative of three independent experiments.b, Upregulation of activation markers CD25 and 4-1BB on WT + TCR and TKO + TCR SP8 T cells in response to stimulation with MART1 or NYESO aAPCs for 24 hours (gated on DAPI − V β 13.1 + CD3 + CD8α + cells).c, Proliferation of SP8 T cells, measured by carboxyfluorescein succinimidyl

1 PBSFig. 8 |
Fig. 8 | In vivo function of 1G4-expressing, class I MHC-null, RAG1-and RAG2null SP8 T cells.a, Experimental design for in vivo tumour challenge in NSG mice I.V. engrafted with NALM6 tumour cells (5 × 10 5 per mouse) expressing the cognate NYESO SCT and firefly luciferase.Five days after tumour engraftment, mice were injected I.V. with PBS, WT + TCR SP8 T cells (1 × 10 7 per mouse) or TKO + TCR SP8 T cells (1 × 10 7 per mouse).Tumour bioluminescence was measured every 3-4 days; IL-2 (10,000 IU per dose) was administered on the indicated days.b, Stratification of mice with equivalent tumour bioluminescence signal at day 5 into treatment groups.The mean ± s.e.m. for each group is shown; two-tailed Mann-Whitney U-test for each group showed NS difference (n = 9 individual mice, all groups).c, Representative imaging of mice from each treatment group at the indicated time points.d, Summary of bioluminescence tumour signal for all animals in each treatment group at the indicated time points.The mean ± s.e.m. (*P < 0.05, **P < 0.01, ***P < 0.001, two-tailed Mann-Whitney U-test) for each group is shown (n = 9 individual mice, all groups). Articlehttps://doi.org/10.1038/s41551-023-01146-7 (catalogue number WA01; WiCell) and ESI017 (ref.48) (catalogue number ES-700; ESI BIO) were maintained and expanded on Matrigel-coated 6-well plates (Growth Factor Reduced Matrigel Matrix; catalogue number 356231; BD Biosciences) in mTeSR Plus complete medium (mTeSR Plus Basal Medium and mTeSR 5X Supplement; catalogue number 100-0276; Stem Cell Technologies).Culture medium was changed daily.After reaching ~70% confluency, PSC cultures were dissociated with TrypLE Express (catalogue number 12604-013; Thermo Fisher Scientific) and seeded in single-cell suspension at a density of 2 × 10 5 cells per well of a Matrigel-coated 6-well plate in mTeSR Plus complete medium and ROCK inhibitor Y-27632 dihydrochloride (10 µM; catalogue number 1254; Tocris Bioscience), which was removed from culture medium after 1 day.