CRISPR-based gene disruption and integration of high-avidity, WT1-specific T cell receptors improve antitumor T cell function

T cell receptor (TCR)–based therapy has the potential to induce durable clinical responses in patients with cancer by targeting intracellular tumor antigens with high sensitivity and by promoting T cell survival. However, the need for TCRs specific for shared oncogenic antigens and the need for manufacturing protocols able to redirect T cell specificity while preserving T cell fitness remain limiting factors. By longitudinal monitoring of T cell functionality and dynamics in 15 healthy donors, we isolated 19 TCRs specific for Wilms’ tumor antigen 1 (WT1), which is overexpressed by several tumor types. TCRs recognized several peptides restricted by common human leukocyte antigen (HLA) alleles and displayed a wide range of functional avidities. We selected five high-avidity HLA-A*02:01–restricted TCRs, three that were specific to the less explored immunodominant WT137–45 and two that were specific to the noncanonical WT1−78–64 epitopes, both naturally processed by primary acute myeloid leukemia (AML) blasts. With CRISPR-Cas9 genome editing tools, we combined TCR-targeted integration into the TCR α constant (TRAC) locus with TCR β constant (TRBC) knockout, thus avoiding TCRαβ mispairing and maximizing TCR expression and function. The engineered lymphocytes were enriched in memory stem T cells. A unique WT137–45-specific TCR showed antigen-specific responses and efficiently killed AML blasts, acute lymphoblastic leukemia blasts, and glioblastoma cells in vitro and in vivo in the absence of off-tumor toxicity. T cells engineered to express this receptor are being advanced into clinical development for AML immunotherapy and represent a candidate therapy for other WT1-expressing tumors. Description T cells engineered to express a high-avidity WT1-specific TCR and no competing endogenous TCRs can eliminate cancer cells in vitro and in vivo. Alternative WT1 targets Adoptive cell therapy using T cell receptor (TCR)–modified T cells has shown clinical success. However, it remains a challenge to find TCRs that respond to shared tumor antigens presented in common human leukocyte antigen molecules. To that end, Ruggiero et al. and Lahman et al. identified and tested TCR constructs targeting alternative epitopes in the Wilms’ tumor antigen 1 (WT1) protein. Ruggiero and colleagues identified WT137–45 as a candidate peptide for TCR-based therapies and used CRISPR-Cas9 editing to target WT137–45-specific TCR sequences to the TRAC locus. Lahman and colleagues showed that, whereas a TCR-targeted WT1 epitope was lost because of immunoproteasome down-regulation, WT137–45 retains its expression and could be targeted by T cells expressing TCRs specific to that epitope, even in cancer cells that lose immunoproteasome activity. In both cases, T cells expressing WT137–45-specific TCRs controlled tumor burden in murine models. Thus, WT137–45 represents a candidate target for TCR-based immunotherapies that is resistant to immunoproteasome down-regulation.


INTRODUCTION
Adoptive cell therapy (ACT) using transfer of genetically engineered T cells is part of the frontline of cancer immunotherapy and has already produced convincing clinical results, especially against B cell tumors (1)(2)(3). In high-risk acute myeloid leukemia (AML), the beneficial impact of T cells has been documented by the graftversus-leukemia effect observed after allogeneic hematopoietic stem cell transplantation (HSCT) (4) but has been limited by the detrimental effects of alloreactivity (5). Although the infusion of engineered tumor-specific T cells may overcome this hurdle, its use is still limited. First, the expression of leukemic antigens in healthy hematopoietic cells, and the consequent risk of myelosuppression, has so far hampered the use of chimeric antigen receptor (CAR)-T cells in AML (6)(7)(8). Second, the genomic instability intrinsic to AML blasts results in an elevated risk of clonal evolution and relapse (9). Third, poor CAR-T cell persistence is linked to loss of immunosurveillance and disease recurrence (6). Thus, the generation of a T cell product should focus on antigens relevant for blast survival, which may be more likely found intracellularly. In contrast to CARs, T cell receptors (TCRs) can target virtually every protein, independent of their subcellular localization, and are sensitive to very low antigen densities (10). Moreover, TCR signaling promotes T cell survival, leading not only to the generation of antitumor effectors but also to the establishment of long-term immunological memory, necessary for counteracting tumor relapse. In this study, we focused on Wilms' tumor antigen 1 (WT1), a zinc finger transcription factor of proven immunogenicity, which has restricted expression in healthy tissues, a strong correlation with oncogenesis, and is expressed by a wide range of hematological and solid tumors (11,12). WT1 expression is associated with poor disease outcome and is rarely lost, supporting its use as a pan-leukemic marker for minimal residual disease detection (13,14). Accordingly, a low risk of tumor escape through antigen loss can be predicted if WT1 is targeted (15). Spontaneous T cell responses against WT1 have been observed in patients with cancer upon HSCT and have been found to correlate with disease regression (16)(17)(18), validating both the immunogenicity of the antigen and its clinical relevance. As a result, several clinical trials targeting WT1 have been conducted (19)(20)(21). TCR-based therapies have highlighted the safety profile of this therapeutic approach, but efficacy data are limited to date, which in part may be attributable to the characteristics of the epitope targeted in these trials. Most published studies relied on high-avidity TCRs specific for the WT1 126-134 human leukocyte antigen (HLA)-A*02:01restricted peptide, an epitope found to require the immunoproteasome for natural processing (22), which may not be operational in all tumors. Hence, the full exploitation of ACT targeting WT1 requires (i) the identification and validation of high-avidity TCRs specific for epitopes naturally processed by the standard proteasome, and thus by most AML blasts, and restricted by different HLA alleles and (ii) the engineering of more potent and durable T cells.
To this end, we used healthy donor (HD) T cells to identify 19 TCRs, including receptors with high functional avidity, recognizing multiple WT1-derived epitopes with varying HLA restrictions. Through a funnel-based approach, we selected the lead TCRs from our library to engineer TCR-transgenic (Tg) T lymphocytes using genome editing (23,24), an emerging technology in cancer immunotherapy (25)(26)(27). Through clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9-based targeted integration (TI) of the WT1-TCR genes into the TCR  constant (TRAC) locus, combined with efficient knockout (KO) of the endogenous  and  TCR chain genes, we generated homogeneous T cell products, each expressing a nonmodified, high-avidity TCR specific for the immunodominant and naturally processed HLA-A*02:01-restricted WT1 [37][38][39][40][41][42][43][44][45] epitope. The TCR with the highest avidity showed antigen-specific responses in both CD4 + and CD8 + T cells. WT1-TCR-edited lymphocytes efficiently killed leukemic blasts and glioblastoma cells in vitro and in vivo in the absence of unwanted off-tumor effects.

Repetitive stimulations of HLA-matched HDs' cells enable the generation of a library of WT1-specific TCRs
To identify WT1-specific T cells and TCRs, we designed a multistep workflow and applied it to 15 HDs (Fig. 1A). T cells from HD1 to HD12 were stimulated with a pool of 141 15-mer peptides that span the entire standard exon 5 + KTS + WT1 isoform, with the addition of an extra 126-amino acid-long fragment at the N terminus (WT1 pool) (28). T cells from HD13 to HD15 were stimulated with a restricted pool of 11 15-mers (WT1 HLA-A*02:01-restricted pool) previously reported as immunogenic and capable of binding HLA-A*02:01 (29). A peptide length of 15 amino acids was used to elicit both CD4 + and CD8 + responses (30). The activation marker CD137 was used to isolate responding T cells, and their expansion was promoted by serial stimulations with either autologous CD3 − peripheral blood mononuclear cells (PBMCs) or immortalized B lymphocytes (31,32), which provide an unlimited reservoir of autologous antigen-presenting cells (APCs) ( fig. S1). WT1-specific T cells were successfully expanded in 14 of 15 HDs (success rate: 93%) with varying kinetics and expansion rates, as shown by longitudinal monitoring of CD107a expression and interferon- (IFN-) production ( Fig. 1, B and C, and fig. S2, A and B, left). In most cases, a clear HLA class I-restricted responding population could be observed after four rounds of in vitro stimulation (11 of 14 HDs, about 70%; Fig. 1D); fewer cultures expanded HLA class II-restricted T cells (about 30%). To identify the recognized epitopes, we cocultured responding T cells with irradiated APCs pulsed with individual peptide subpools (SPs), according to a defined mapping grid (HD1 to HD12), or with individual peptides (HD13 to HD15). A skewed T cell response toward SP4, SP5, and SP16 was observed for HD1, HD3, HD6, HD7, and HD10, whereas HD13 and HD14 responded to peptide WT1 −78-64 . For the remaining donors, a T cell response directed against different portions of the WT1 protein was identified ( Fig. 1, B and C, and fig. S2, A and B, right). From two HDs (HD4 and HD7), multiple T cell clones reacting to more than one HLA class I-restricted epitope were detected and further expanded by subculturing T cells with each recognized WT1 SP ( fig. S2C). Flow cytometry analyses of the TCR repertoire (Fig. 1E) were performed on 11 responding T cell cultures and confirmed a clear oligoclonal T cell subpopulation in HD1 to HD3, HD5, HD12, and HD14. For HD4, HD6, HD10, and HD13, no dominant V (variable region of TCR chain) was detected, probably due to the limitation of the cytofluorimetric assay, which covers only about 70% of the whole V repertoire, or to the presence of multiple clones. For functional validation of individual TCRs, HLA class Irestricted candidates (13 of 19 clonotypes) were prioritized and classified in three groups: (i) TCRs recognizing WT1 37-45 , (ii) TCRs recognizing WT1 −78-64 of the WT1 HLA-A*02:01-restricted pool, and (iii) other WT1-specific HLA class I-restricted TCRs.

of 16
The TCR hunting procedure identified HLA-A*02:01-restricted TCRs specific for a noncanonical WT1 peptide HD13 and HD14 (group ii) were characterized by a preferential recognition of WT1 −78-64 (Fig. 2F). In silico prediction identified the nonamer LLAA-ILDFL as a high-affinity HLA-A*02:01 binder ( fig. S3, F and G). Although HLA restriction was confirmed by the specific recognition of WT1 −78-64pulsed HLA-A*02:01 + APCs (Fig. 2G), natural processing and presentation of the cognate peptide was initially questionable because expanded T cells barely recognized K562-Tg cells (Fig. 2G). WT1 −78-64 is located in the N-terminal noncanonical portion of WT1, in a splice variant reported in AML samples (29). We observed that this variant is not efficiently expressed by this cell line and was heterogeneously expressed by leukemic blasts (fig. S3H). We observed specific and potent killing of primary WT1 + AML blasts harvested from three HLA-A*02:01 + patients by HD14-expanded T cells, whereas HD13-expanded T cells showed a weak response (Fig. 2H). Clonal tracking of the TCR repertoire overtime revealed WT1-specific TCR genes ( Fig. 2I) even in the case of HD13, for which characterization of the TCR clonal expansion by flow cytometry proved unsuccessful. Hence, we identified two TCRs targeting a splice variant-specific naturally processed HLA-A*02:01-restricted WT1 peptide.

T cells)
% IFN--secreting cells 0 S P 1 S P 2 S P 3 S P 4 S P 5 S P 6 S P 7 S P 8 S P 9 S P 1 0 S P 1 1 S P 1 2 S P 1 3 S P 1 4 S P 1 5 S P 1 6 S P 1 7 S P 1 8 S P 1 9 S P 2 0 S P 2 1 S P 2 2 S P 2 3 S P 2 4

HD1 T cells -S8
WT1 Our TCR identification efforts were culminated in a library of 19 TCR clonotypes, specific for 14 WT1 peptides and encompassing different HLA restrictions (Fig. 2J). Out of these, seven TCRs were specific for two epitopes (five for the WT1 37-45 peptide and two for WT1 −78-64 ), naturally processed by leukemic blasts and presented on HLA-A*02:01, the most common allele in the Caucasian population (36). Overall, five (HD1, HD3, HD7, HD13, and HD14) HLA-A*02:01-restricted TCRs, for which a clear enrichment of the dominant TCR sequences could be tracked in the T cell cultures, were selected for further validation.

High-avidity WT1-TCR CRISPR-Cas9 genome-edited T cells efficiently kill primary leukemias
To redirect T cell specificity, we modified our previously reported TCR gene editing protocol (23,24), leveraging on the ability of CRISPR-Cas9 genome editing technology to precisely and efficiently disrupt multiple genes simultaneously in one manipulation step. We deleted the endogenous TCR genes, thus abrogating the risks of TCR mispairing and achieving maximal surface expression of our WT1-specific TCRs (Fig. 3A). To KO the endogenous TCR chain, we identified a highly active guide RNA (gRNA) targeting exon 1 of the TRAC locus (TRAC023; fig. S7A). In addition, to account for the evolutionary duplication of the TCR chain, gRNAs were designed, which target both TCR  constant 1 (TRBC1) and TRBC2 loci, and selected gRNA TRBC019 ( fig. S7B). Two days after activation, cells were electroporated with preassembled ribonucleoprotein (RNP) complexes. KO efficiencies at each locus, assessed on the protein abundance using CD3 surface expression, were >99% for the  chain and >90% for the  chain, with an overall disruption efficiency of 98 ± 0.6% (means ± SEM) when concomitant targeting of both loci was performed (Fig. 3B). On day 3, we redirected T cells with WT1 37-45 -specific receptors upon transduction with bidirectional lentiviral vectors (LVs) (fig. S7C) (37) encoding for HD1-, HD3-, or HD7-derived TCRs. Transduction efficiency, as evaluated by determining the rescue of CD3 surface expression, was about 50% (HD1-TCR: 41.0 ± 3%; HD3-TCR: 51.5 ± 3.5%; HD7-TCR: 58.5 ± 3.5%; means ± SEM) ( Fig. 3B) for the three WT1 37-45 -specific TCRs. Engineered cells displayed a stem memory T cell (T SCM ) phenotype ( Fig. 3C and fig. S7D) and a CD4 + -to-CD8 + T cell ratio similar to unedited lymphocytes, indicating that gene editing does not affect T cell subsets or differentiation. Homogeneous and high expression of the three different TCRs was further confirmed by high WT1 [37][38][39][40][41][42][43][44][45] dextramer binding on CD8 + cells ( fig. S7E). HD1 LV TCR-edited T cells showed a superior degranulation activity and cytokine production compared to HD3 LV and HD7 LV TCRs when challenged with T2 cells pulsed with decreasing concentrations (1 M to 1 pM) of WT1 37-45 peptide ( Fig. 3D and fig. S7F). HD7-TCR required saturating peptide concentrations (100 M) for recognition, highlighting an extremely low functional avidity ( fig. S7G). The halfmaximal effective concentration (EC 50 ) indicates the superior performance of HD1-TCR, followed by HD3-and HD7-TCR (fig .  S7H). Adoptive T cell therapy benefits from combined activity of CD4 + and CD8 + T cells (38)(39)(40); however, the activation of CD4 + cells by a HLA class I-restricted TCR requires high avidity, as described for some virus-specific lymphocytes (41). Of interest, HD1-TCR was able to activate both T cell subsets, although with an about 100-fold higher peptide threshold for CD4 + cells. This characteristic was not observed with HD3-TCR (Fig. 3E). Both HD1-and HD3-TCR-engineered T cells specifically killed blasts harvested from three HLA-A*02:01 + patients with AML, with an efficiency of up to 75% at an effector-to-target ratio of 5:1 (Fig. 3F). Overall, HD1-TCR-edited T cells exhibited superior cytotoxic ability, supporting an enhanced fitness of this cellular product.
To further highlight the relevance of the manipulation protocol implemented here, peripheral blood T cells harvested from patients with AML were cultured and engineered with the same protocol used for HDs. Robust gene editing efficiency and preferential expansion of T SCM cells were observed ( fig. S8, A to C). CD4 + -and CD8 + -edited HD1-TCR-expressing patients' T cells recognized T2 cells pulsed with the WT1 37-45 peptide ( fig. S8D) and killed HLA-matched primary leukemic blasts ( fig. S8E). These results highlight the feasibility of our protocol in generating highly functional early memory TCR-edited T cells.

HD1-TCR-engineered T cells are superior in eliminating primary blasts compared to other WT1-specific HLA-A*02:01-restricted TCRs
In addition to the WT1 37-45 -specific TCRs, we also evaluated the two receptors specific for the noncanonical WT1  S9C). When challenged in a 24-hour coculture with primary leukemic blasts, each engineered T cell product (HD1 LV, HD13 LV, and HD14 LV TCR) was able to eliminate target cells. HD1-TCR-expressing cells were superior to the other cellular products in mediating a consistent and potent killing of leukemic blasts (Fig. 3G), thus further strengthening the selection of this receptor, specific for a canonical portion of WT1, as the lead candidate for immunotherapy of AMLs.
TI of HD1-TCR in the TRAC locus, combined with TRBC disruption, generates an efficient and specific cell product To reduce the manipulation steps, minimize insertional mutagenesis risk, and further standardize the cellular products, we explored TI of the Tg-TCR into the TRAC locus by combining CRISPR-Cas9 with an adeno-associated virus (AAV) carrying the TCR  and  genes  separated by a 2A peptide and flanked by homology arms complementary to sequences present on both sides of the nuclease cut site (Fig. 4A). We started from gRNAs selected and validated to promote specific disruption of TRAC and TRBC1/2 loci, with an efficiency greater than 90% ( fig. S7, A and B). The TRAC guides were further evaluated for insertion efficiency of the Tg-TCR, because high editing efficiency is not predictive for high insertion rates (fig. S10A). All lead guides were carefully assessed for off-target editing properties by a combination of bioinformatic (CasOFFinder) (42) and biochemical (SITE-Seq) (43) approaches ( fig. S10B). Each potential site was validated in T cells using targeted off-target sequencing ( fig. S10C). In particular, gRNA TRBC004 showed no detectable off-target activity and the gRNA TRAC002 showed a high degree of insertion and KO and also displayed no validated off-target events. To evaluate whether KO of both endogenous TCR genes is necessary, cellular products obtained upon insertion of the Tg-TCR at the TRAC locus were compared with (TI TCR) or without (TRAC-TI TCR) concomitant TRBC1/2 gene disruption. We observed a high degree of TCR insertion in the TRAC locus (60 to 70% CD3 + cells; Fig Fig. 4B, right] and nearly completely abrogated TCR mispairing (Fig. 4B, left, and fig. S11A, top versus bottom). Accordingly, TI TCR cells displayed a significantly increased functional avidity (P < 0.0001; Fig. 4C and fig. S11B) and reduced cross-reactivity in a mixed lymphocyte assay ( fig. S11C), thus corroborating the notion that TCR mispairing can lead to new, unknown reactivities (44,45).
The generation of an additional disulfide bond between TCR chains (46, 47) had no effect on reducing TCR mispairing and unexpectedly reduced the expression of our HD1-TCR ( fig. S11, A and D). TI of a CAR gene into the TRAC locus, under the control of the TCR promoter, has been reported to improve CAR-T cell function by averting tonic signaling (25). We thus explored the effects of the endogenous TCR promoter (promoterless) versus a strong exogenous promoter [full EF1 (elongation factor 1 alpha) promoter] in driving HD1-TCR expression. EF1 drove stronger, more homogeneous Tg-TCR expression (fig. S11E) to a degree comparable to natural TCRs, resulting in increased functionality ( Fig. 4D and fig. S11F). Treatment of cells with AAV in the absence of CRISPR editing did not result in any episomal WT1-TCR expression, confirming that expression is due to TI in the TRAC locus ( fig. S11E). Furthermore, we confirmed HD1 insertion at the genomic level using droplet digital polymerase chain reaction (ddPCR) with primers designed to flank the insert junction region (fig. S11G). The functional avidity of HD1-TCR proved about two logs higher than that of HD3-TCR, even when tested in TI TCR T cells, built with natural TCR constant chains driven by the EF1 promoter ( fig. S11H). HD1 TI TCR T cells efficiently killed WT1 + HLA-A*02:01 + K562 cells (Fig. 4E) and exhibited a very potent and efficient killing of primary leukemic cells at different effector-to-target ratios (Fig. 4F) to a comparable degree observed with LV-edited T lymphocytes.

HD1-TCR-edited lymphocytes display an optimal safety profile
A thorough characterization of the TCR specificity necessitates special consideration given the unexpected toxicities observed in clinical trials with affinity-enhanced TCRs (48)(49)(50). Although our TCR is natural and isolated from an HLA-matched HD and cells are engineered to minimize TCR mispairing, we sought to assess potential risks of off-target reactivity. We interrogated the residues within the WT1 37-45 epitope that are critical for HD1-TCR recognition by sequentially substituting each amino acid with an alanine (alanine scanning). All peptides but peptide VADFAPPGA (peptide 2) were predicted to bind HLA-A*02:01 with similar affinity to the wild-type peptide. No activity was detected with peptides 1 to 3 and 5, and minimal activity was detected with peptide 4. In contrast, peptides 6 and 7 exhibited activities comparable to the wild-type peptide.
These data indicate that amino acids V1, D3, and P6 are critical for TCR binding, and F4 is a strong contributor. Amino acids L2 (predicted and experimentally confirmed) and A9 (predicted) were determined critical for HLA binding, whereas P7 (peptide 6) and G8 (peptide 7) likely did not contribute to HLA or TCR binding (Fig. 4G). The alanine-substituted peptide at position 4 induced T cell reactivity only at high concentrations (500 nM); thus, the phenylalanine in position 4 was considered essential for high-avidity interactions. Positions A5 and A9 were not evaluated in this study because they are natural alanine positions. These preliminary data suggest that the minimal functional motif for the WT1 37-45 -specific HD1-TCR is VLDFAPxxA (where A is a natural alanine that was not investigated, and x is not critical for binding). In silico analysis using ScanProsite (https://prosite.expasy.org/scanprosite/) and BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to identify protein sequences that contained the minimum binding motif VLDFxPxxx with a predicted affinity for HLA-A*02:01 of at least 1000 nM revealed one other expressed protein in humans outside of WT1. Testing of cell lines expressing high abundance of this protein (PIF1) did not reveal any reactivity with the HD1-TCR T cells, suggesting high specificity of this TCR to WT1. The same analysis was performed on our second-in-rank HD13-and HD14-TCRs. Although the latter proved highly selective for the cognate epitope, HD13-TCR, despite impressive functional avidity, displayed a promiscuous recognition motif ( fig. S12), thus limiting its translational potential. Reactivity of HD1-TCR T cells to normal cells was also evaluated. WT1 is expressed at 100-to 1000-fold lower concentrations, compared to AML, in normal tissues including stromal cells within the reproductive organs, podocytes, and CD34 + cells (GTEx, Tissue Atlas) (11,51,52). Because of the relevance of potential bone marrow toxicity and the accessibility of this normal tissue, HD1-redirected T cells were tested against HLA-A*02:01 + and HLA-A*02:01 − CD34 + cells and did not induce relevant cytotoxicity (Fig. 4H). The data indicate that the WT1 37-45 peptide concentrations naturally presented by CD34 + cells are not sufficient to trigger HD1-TCR T cell-mediated recognition. Overall, results of this validation support the safety of HD1-TCR T cells and further strengthen the selection of this receptor as a candidate for clinical implementation.

EF1-driven expression of the TCR generates T cells with an optimal activation profile in vivo
We investigated the effects of the different engineering processes for their effect on T cell activity against AML in vivo. To this aim, WT1-expressing primary leukemic blasts harvested from an HLA-A*02:01 + patient (pAML1) were infused into immunodeficient NOD (non-obese diabetic) scid (severe combined immunodeficient) gamma (NSG) mice. TCR expression and phenotype of infused T cells was evaluated ( fig. S13A). Leukemia-bearing mice were treated with two doses of HD1 LV TCR, EF1 HD1 TI TCR, or (D) Promoterless (PL) and elongation factor 1 alpha (EF1) WT1 T cells were assessed for their functionality by measuring cytokine release when cocultured with WT1 37-45 -pulsed OCI-AML3 cells (n = 4). (E) Cytotoxicity of HD1-engineered T lymphocytes against hematologic tumors was shown by measuring loss of luciferase signal upon coculture with luciferase-expressing H L A -A * 0 2 : 0 1 -t r a n s d u c e d (K562-A2.1-luc), but not control (K562-luc), WT1 + chronic myelogenous leukemia (CML) (K562) cells (n = 2). Unedited T cells were used as a control. Statistics refers to the condition with unedited T cells. (F) TI of TCR-expressing cells showed a comparable killing activity to LV-engineered T lymphocytes as evaluated by measuring caspase 3 expression upon coculture with HLA-A*02:01 + WT1 + pAML blasts (total WT1 expression in parentheses) harvested from two patients at different E:T ratios (n = 2). Statistics refer to the condition with control T cells (engineered lymphocytes expressing an unrelated TCR). (G) The specificity of TCR-edited T cells was assessed by determining the critical amino acid residues involved in TCR-HLA binding. Caspase 3 and 7 induction was measured in T2 cells pulsed with the WT1 37-45 wild-type peptide or with the epitopes generated by mutating each individual non-alanine amino acid with alanine (left) upon coculture with HD1-TCR T cells (n = 2). Critical amino acid residues involved in the TCR binding are highlighted with blue circles (right). The phenylalanine in position 4 (dashed circle) was considered essential for high-avidity TCR-ligand interactions. (H) HD1-engineered T cells do not show on-target/off-tumor toxicities against CD34 + stem or progenitor cells, as evaluated by coculture of the T cell product with HLA-A*02:01 + HSCs (n = 2). Pulsing of HSC or T2 cells with WT1 37-45 peptide was used as a positive control. Functional studies were performed with T cells generated from distinct donors. Data are presented as means ± SEM [in (H), means ± SD]. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 by two-way ANOVA with Tukey's multiple comparisons test (E and F); nonlinear regression, least squares fit (C and D); or Student's t test (B).  promoterless HD1 TI TCR T cells (1 × 10 7 TCR + T cells per dose) starting 3 days after blast infusion (Fig. 5A). As control, mice were either left untreated or treated with melanoma-associated antigen recognized by T cells 1 (MART1) TI TCR T cells (recognizing an HLA-A*02:01-restricted MART1 epitope). All HD1-based cellular products significantly controlled leukemic outgrowth, compared to controls (P < 0.0001; Fig. 5B). Results highlighted a greater efficacy for HD1 TI TCR T cells than HD1 LV TCR T cells (P < 0.0001) in killing blasts in vivo, independently from the promoter used to drive TCR expression. A thorough phenotypic analysis showed that, once harvested ex vivo, LV-engineered T lymphocytes, which proved enriched in highly differentiated T EMRA (terminally differentiated effector memory T) cells at the time of infusion ( fig. S13A), were phenotypically different and displayed a reduced expression of the HLA-DR activation marker and an increased expression of TIM3 and PD-1 exhaustion markers compared to TI-based cellular products ( fig. S13, B and C). Comparative analysis of EF1-versus promoterless-based products revealed a higher MFI for surface CD3 expression (Fig. 5C), an increased proportion of activated CD4 + and CD8 + T cells (Fig. 5D), and an increased percentage of HLA-DR + TIM3 − PD-1 − T cells when the tumor-specific TCR expression was driven by the EF1 promoter ( fig. S13C). On the basis of the aggregate results, the EF1-driven T cell product was selected for further validation. The funnel diagram summarizes the filters applied to our pipeline of TCR discovery and T cell engineering to select a candidate (HD1 TI TCR) for clinical validation (fig. S13D).

HD1-TCR-engineered T cells efficiently control human leukemia and glioblastoma outgrowth in vivo
To further evaluate the efficacy of this HD1 TI TCR T cell product in vivo, we tested engineered cells in three separate immunodeficient murine models. TCR expression and phenotype of infused T cells were evaluated ( fig. S14A). In the AML model described above (pAML1; Fig. 6A), we compared the efficacy of one versus two lymphocyte infusions. At day 20, a time point associated with a nearly complete disappearance of circulating infused T cells (Fig. 6B,  left), half of the mice belonging to the MART1-and HD1-TCR groups received an additional dose of engineered cells. After initial fluctuations in their frequencies, circulating blasts started to increase rapidly in control mice, but not in HD1-TCR-treated animals (P < 0.0001). At day 45, we observed the complete absence of circulating blasts in animals treated with two doses of HD1-TCR T cells (Fig. 6B, middle). Longitudinal profiling of circulating T cells showed a preferential accumulation of activated effector memory T cells in the HD1 TI TCR-treated animals ( fig. S14, B and C). Results were confirmed in an independent experiment with blasts (pAML5) from a different HLA-A*02:01 patient (Fig. 6B, right). At euthanasia, evaluation of the tumor burden in the bone marrow and in the spleen confirmed the efficiency of HD1-treated cells in controlling the leukemic outgrowth ( fig. S14D).
Because WT1 is overexpressed in several hematological and solid tumors (12), we reasoned to challenge HD1 TI TCR T cells in two additional models. Upon in vitro assessment of T cell-mediated cytotoxicity against an aggressive acute lymphoblastic leukemia (ALL) cell line (ALL 697) (Fig. 6C), HD1-TCR-engineered T cells were infused in tumor-bearing human interleukin-15 (hIL-15) NOG (NOD/Shi-scid/IL-2Rnull) mice. In this context, HD1 TI TCRengineered T cells proved highly effective in controlling leukemic cells compared to control groups (P < 0.0001) (Fig. 6D).
Last, as a model of solid tumor expressing WT1 (53), we selected glioblastoma, which represents a major unmet clinical need. At first, we verified the ability of HD1 TI TCR T cells to effectively kill HLA-A*02:01 + , WT1-expressing U87MG glioblastoma cells in vitro (Fig. 6E). Using an in vivo heterotopic glioblastoma NSG model, we found that HD1 TI TCR T cells infused in mice with a measurable tumor burden (≥5 × 10 9 total flux in photons per second) (fig. S14E) efficiently controlled tumor outgrowth, leading to an improved survival at day 24 (P < 0.01 versus mice infused only with the tumor and P < 0.001 versus mice treated with MART1-TCR T cells) (Fig. 6F).
Longitudinal profiling of engineered T cells circulating in mice confirmed our in vitro observations and showed a preferential enrichment of CD4 + and CD8 + activated T lymphocytes in the HD1 TI TCR group (Fig. 6, G and H, and fig. S14F) in each in vivo model tested. No substantial body weight loss ( fig. S14G), signs of graftversus-host disease (GVHD), or other treatment-related toxicities were observed in any experimental setting. Overall, these results further underline the potential of HD1-TCR-engineered T cells as a safe and potent controller of tumor outgrowth in vivo and support advancement of this product toward clinical testing.

DISCUSSION
Endowing T cells with new TCRs can effectively redirect some of the most potent players of our immune system against virtually any tumor antigen. Because of the low frequency of high-avidity cancer-specific T cells, identifying the most potent and safe tumorspecific TCRs remains a major challenge for the broad use of TCR-based therapies in clinical practice. In this study, we coupled the discovery of optimal antitumor TCRs with genome editing to generate safe and highly efficient living drugs with the potential to persist long term and patrol for tumor recurrence. We focused on WT1, a transcription factor widely overexpressed on a variety of hematological and solid tumors and collectively recognized as a high-priority antigen (11). Tumor-specific T cells in patients are rare, often exhausted (54,55), and difficult to expand. Thus, although more time consuming, we chose to stimulate T cells with autologous APCs because of reported issues of promiscuity in peptide binding observed with high-avidity TCRs generated from mismatched HLA cultures (56). Being selected from the natural immune repertoire of HDs, the newly identified high-avidity TCRs are expected to be intrinsically safer than affinity-enhanced TCRs (57,58), having gone through negative selection in the thymus.
A TCR targeting the WT1 126-134 , HLA-A*02:01-restricted epitope has been safely tested in clinical trials, but its efficacy has been limited, in part, by its requirement of processing by the immunoproteasome (22,61). In contrast, WT1 [37][38][39][40][41][42][43][44][45] and WT1 −78-64 are naturally processed and presented on the surface of primary blasts harvested from patients with leukemia, as shown by the ability of newly isolated TCRs to efficiently eliminate these targets. Despite these promising data, given the localization of WT1 −78-64 in a WT1 isoform that is not uniformly expressed in all AMLs and has limited data on its relevance in tumor biology, further investigation in different tumor types is warranted before targeting it therapeutically. In our study, although the known WT1 126-134 epitope was included in the peptide pool, T cells specific for this epitope were not retrieved, indicating that other epitopes can be more immunodominant, as also shown by vaccination studies. T cells specific for WT1 [37][38][39][40][41][42][43][44][45] could be detected in 60% of patients with cancer upon vaccination, whereas only 20% were specific for WT1 126-134 epitope (62). Overall, these data indicate the presence of an endogenous immunity toward this epitope (62,63); however, identification of high-avidity T cells has not been reported to date.
We selected the most potent TCRs recognizing the WT1 37-45 peptide (HD1-, HD3-, and HD7-TCRs) as well as the noncanonical WT1 −78-64 (HD13-and HD14-TCRs) and genetically engineered these receptors into T cells for further characterization, leveraging the CRISPR-Cas9 editing platform. Despite recognizing the same peptide, the three TCRs specific to WT1 [37][38][39][40][41][42][43][44][45] showed a wide range of functional avidities, with HD1-TCR outperforming the rest. HD1-TCR also showed a superior and more consistent killing of primary leukemic samples when compared to HD13-and HD14-TCRs. To generate a more homogeneous product and to further boost functionality and specificity, we exploited TI of the TCR genes into the TRAC locus (silencing the endogenous TCR chain) while also removing the endogenous TCR chain. TI into TRAC has already been shown beneficial for improving CAR-T cell fitness (25). In case of TCR insertion, this strategy may be particularly useful, generating a cellular product with only one TCR pair and thus avoiding mispairing issues and the need to generate modified TCR constant regions that can be immunogenic (64). This engineering approach resulted in improved TCR expression, as the transferred TCR  and  chains no longer compete with the endogenous ones for other TCR complex components. This led to increased functionality and specificity, as highlighted by the comparison of T cells engineered with and without TRBC KO. Unlike what has previously been shown for CAR-T cells, our in vivo data showed that a more efficacious cellular product, with an optimal in vivo phenotypic profile (superior engagement of activated CD4 + and CD8 + T cells, enhanced TCR surface expression, higher percentage of the HLA-DR activation marker, and reduced expression of exhaustion receptors), was obtained when TCR expression was driven by the EF1 promoter, rather than by the endogenous TCR promoter. The results reported by Eyquem et al. (25) compared random integration of a CAR using a promoter-driven LV construct to TI into the TRAC locus without an exogenous promoter and found improved in vivo activity with the latter, attributing the effect to the TCR versus EF1 promoter and the genomic context. The data reported here show a clear benefit for the TI, regardless of the promoter; however, activity was further improved with the strong exogenous promoter in our setting, which could be due to the length of the transcript, here encoding for two TCR chains instead of one CAR, or the need for two 2A self-cleaving peptides in the promoterless construct. TI of the EF1 construct generated more potent T cell activation than the construct driven by the endogenous TCR promoter while also resulting in less exhausted T cells. Recently, Müller and colleagues (65) showed that the orthotopic integration of viral specific TCR genes leads to homogeneous TCR expression and more predictable in vivo activity of engineered cells, when compared to retrovirally transduced, edited T cells. Our data suggest that optimal activation of the TCR is critical and dependent on genomic context as well as the right promoter strength. Accordingly, edited T cells, generated from HDs' peripheral blood and from circulating T cells isolated from patients with AML, displayed an early differentiated phenotype, a characteristic that has been associated with long-term T cell persistence in multiple clinical trials (66)(67)(68).
HD1-TCR-edited T cells showed high functional avidity in the low nanomolar range. They also displayed potent recognition of target cells, including primary AML blasts. HD1-TCR-edited T cells were also endowed with the unique ability to activate both CD4 + and CD8 + T cells, a characteristic not yet reported for TCRs specific for this epitope (69), which could be crucial for effective and long-lasting clinical effects.
Because a genome-edited product may contain unwanted off-target changes, and because WT1 is expressed at lower concentrations in normal tissues, we carefully evaluated its safety profile. In this regard, no off-target editing sites were observed for the gRNAs used in the genome editing process. Furthermore, the TCR recognition motif proved to be highly specific for the WT1 37-45 epitope. In addition, no killing of HSCs that express physiological concentrations of the WT1 protein was observed. Although on-target off-tumor toxicity cannot be assessed in the mouse because of the lack of HLA-A*02:01 expression, no cross-reactivity or GVHD was observed, and the infusion of HD1-TCR-edited T cells effectively controlled tumor outgrowth in different tumor settings, mirroring hematological (primary AML and ALL models) and solid (glioblastoma) tumors. T cell engraftment, activation, and persistence for about 21 days were observed in these mice, as expected in an in vivo setting without human cytokine support. No uncontrolled T cell outgrowth was detected, supporting the specificity of the TCR and the lack of chromosomal alterations due to the genome editing process, as further confirmed by detailed genomic analyses of the product being developed for clinical investigation.
Our study has some limitations. In TCR-based gene therapy studies, the TCR needs to be matched to patient HLA, thus potentially limiting its wide-range exploitation in the clinical arena. However, to overcome this hurdle, we have focused on the in-depth profiling of a TCR restricted by the HLA-A*02:01 allele, present in 40% of the Caucasian population. In addition, our pipeline has led to the generation of a library of TCRs encompassing diverse HLA restrictions, an aspect that may further enlarge the cohorts of patients who may benefit from a WT1-TCR therapy. Furthermore, our study has focused on the identification of TCRs targeting a single antigen, with the consequence that antigen loss may potentially be envisaged as an immune evasion mechanism. To limit this potential drawback, we have selected WT1, an antigen highly relevant for tumorigenesis and leukemogenesis.
As our study highlights, although a library of TCRs specific for a selected antigen can be established, the application of a systematic comparative multistep analysis of their characteristics allows for prioritization of receptors and selects for those endowed with all the relevant features necessary for its implementation in clinical trials. Hence, we here show an innovative and robust pipeline for TCR discovery able to lead to the generation of T cell products for the treatment of hematological malignancies and possibly solid tumors. These results support the development of this cellular product for clinical investigation.

Study design
The objective of this study was to identify tumor-specific TCRs to be used in TCR gene editing with the final aim of generating cellular products for the treatment of hematological and solid malignancies. For expansion of WT1-specific T cells from HDs, cell cultures were functionally tested at different time points and analyzed by TCR sequencing. Newly identified TCRs were further tested and validated in TCR-engineered T cells. For the in vitro evaluation of TCRengineered T cells, we used as effector cells lymphocytes harvested from different healthy individuals or AML patients and engineered to redirect their TCR specificity. The detailed number of biological replicates is reported in each figure legend. As target cells, we used either leukemic blasts harvested from different AML patients or tumor cell lines. Experiments were performed at increasing effectorto-target ratios, and several parameters were evaluated (killing ability, CD107a expression, and cytokine secretion). Peripheral blood was obtained from HDs at San Raffaele Hospital (Ospedale San Raffaele Scientific Institute) upon informed consent, in agreement with the Declaration of Helsinki. PBMCs were isolated using Ficoll-Hypaque (Fresenius) density gradient centrifugation.
For in vivo studies, the number of animals was selected on the basis of variability observed in pilot experiments and on availability of primary blasts. Growth kinetics of the infused leukemias and of U87MG cells was assessed before treatment. Animals were randomized to groups (6 to 10 mice per group), and no mice were excluded from the experiment. Mice were treated by an operator who was blinded to treatment groups. All in vitro and in vivo experiments were replicated by different investigators, and all replicates were successful. Analysis of in vitro and in vivo data was based on objectively measurable data (cell counts and MFI).

Editing of T lymphocytes
PBMCs from HDs or from AML patients were activated and sorted using magnetic beads conjugated to antibodies to CD3 and CD28 (ClinExVivo CD3/CD28, Invitrogen) and seeded at a concentration of 10 6 cells/ml in X-VIVO 15 supplemented with 1% penicillinstreptomycin, 2 mM glutamine, 5% fetal bovine serum (FBS) (Lonza/Euroclone), IL-7 (5 ng/ml) (PeproTech), and IL-15 (5 ng/ml) (PeproTech) (70). Two days after stimulation, T cells were electroporated with RNP complexes [consisting of purified Spy Cas9 nuclease (Intellia Therapeutics) duplexed with synthetic gRNA] targeting the TRAC and TRBC1/2 loci simultaneously using the Lonza Nucleofector 4D Electroporation System. Upon electroporation, T cells were seeded in X-VIVO 15 supplemented with 5% FBS in the presence of IL-7 (5 ng/ml) and IL-15 (5 ng/ml). The day after, T cells were transduced with a TCR-encoding LV. Beads were detached at day 6 after stimulation. TCR KO efficiency was evaluated by measuring the loss of CD3 expression from the T cell surface. Similarly, efficiency of LV transduction was evaluated as CD3 and specific TCR V (or dextramer) expression on the cell surface. At day 15 after stimulation, T cell phenotype was evaluated by cytofluorimetric analysis. For TI of the WT1-TCR in the TRAC locus, the same experimental setting was used with the following differences: (i) A different TRAC single guide RNA (sgRNA) was used and (ii) the redirection of T cell specificity was achieved by adding AAV particles to seeded T cells directly after the electroporation step.

Statistical analysis
Raw, individual-level data are presented in data file S1. Statistical analyses were performed with Prism software (GraphPad, version 8). Student's t test, one-way analysis of variance (ANOVA), and two-way ANOVA with Tukey's or Sidak's multiple comparisons tests were performed for the analysis of the set of data throughout the study as indicated. Log-rank (Mantel-Cox) test was performed for the survival analysis. EC 50 values were calculated using a nonlinear regression model (least squares fit) by using the dose-response equation of the GraphPad Prism software. A P value of <0.05 was set as a threshold for significance.