TET2 guards against unchecked BATF3-induced CAR T cell expansion

Further advances in cell engineering are needed to increase the efficacy of chimeric antigen receptor (CAR) and other T cell-based therapies1–5. As T cell differentiation and functional states are associated with distinct epigenetic profiles6,7, we hypothesized that epigenetic programming may provide a means to improve CAR T cell performance. Targeting the gene that encodes the epigenetic regulator ten–eleven translocation 2 (TET2)8 presents an interesting opportunity as its loss may enhance T cell memory9,10, albeit not cause malignancy9,11,12. Here we show that disruption of TET2 enhances T cell-mediated tumour rejection in leukaemia and prostate cancer models. However, loss of TET2 also enables antigen-independent CAR T cell clonal expansions that may eventually result in prominent systemic tissue infiltration. These clonal proliferations require biallelic TET2 disruption and sustained expression of the AP-1 factor BATF3 to drive a MYC-dependent proliferative program. This proliferative state is associated with reduced effector function that differs from both canonical T cell memory13,14 and exhaustion15,16 states, and is prone to the acquisition of secondary somatic mutations, establishing TET2 as a guardian against BATF3-induced CAR T cell proliferation and ensuing genomic instability. Our findings illustrate the potential of epigenetic programming to enhance T cell immunity but highlight the risk of unleashing unchecked proliferative responses. Disruption of TET2 increases the antitumour efficacy of CAR T cells, but establishes an epigenetic state that is prone to hyperproliferation and accumulation of secondary mutations.

Further advances in cell engineering are needed to increase the efficacy of chimeric antigen receptor (CAR) and other T cell-based therapies [1][2][3][4][5] . As T cell differentiation and functional states are associated with distinct epigenetic profiles 6,7 , we hypothesized that epigenetic programming may provide a means to improve CAR T cell performance. Targeting the gene that encodes the epigenetic regulator ten-eleven translocation 2 (TET2) 8 presents an interesting opportunity as its loss may enhance T cell memory 9,10 , albeit not cause malignancy 9,11,12 . Here we show that disruption of TET2 enhances T cellmediated tumour rejection in leukaemia and prostate cancer models. However, loss of TET2 also enables antigen-independent CAR T cell clonal expansions that may eventually result in prominent systemic tissue infiltration. These clonal proliferations require biallelic TET2 disruption and sustained expression of the AP-1 factor BATF3 to drive a MYC-dependent proliferative program. This proliferative state is associated with reduced effector function that differs from both canonical T cell memory 13,14 and exhaustion 15,16 states, and is prone to the acquisition of secondary somatic mutations, establishing TET2 as a guardian against BATF3-induced CAR T cell proliferation and ensuing genomic instability. Our findings illustrate the potential of epigenetic programming to enhance T cell immunity but highlight the risk of unleashing unchecked proliferative responses.
CARs are synthetic receptors for antigens that instruct T cell specificity and augment antitumour functions 2,4 . CAR T cell therapy for relapsed and refractory acute lymphoblastic leukaemia, non-Hodgkin lymphoma and multiple myeloma yields a high rate of complete responses, although a large fraction of patients will eventually relapse from their disease 3,17 . Novel strategies are needed to augment the overall efficacy of CAR T cells to prevent these relapses and tackle solid tumour therapy 2,3,18 . We hypothesized that epigenome programming could act in concert with CARs to promote CAR T cell activity by supporting T cell proliferation and functional persistence. TET2 is a member of the TET family of epigenetic regulators that successively oxidize 5-methyl cytosine in DNA 19 . A study of the T cell receptor in transgenic mice 9 and a case report of a patient with lymphoma with a hypomorphic TET2 allele treated with CAR T cells 10 suggest that loss of TET2 may enhance T cell responses. Mutations in TET2 are frequent in myeloid and lymphoid malignancies but are not sufficient to establish a malignant state [20][21][22] . Here we report unexpected antigen-independent clonal expansions of CAR T cells lacking TET2, which is dependent on sustained expression of BATF3.

Effect of TET2 on CAR T cell efficacy
To assess the effect of TET2 on CAR T cell efficacy, we disrupted TET2 in human T cells before retrovirally transducing them with either FDA-approved CD28 or 4-1BB-based CD19 CARs, hereafter designated Rv-1928z and Rv-19BBz, respectively, and compared their activity in the well-established B cell acute lymphoblastic leukaemia NALM6 model in NSG mice (Fig. 1a). Human peripheral blood T cells typically showed a CRISPR-Cas9-mediated TET2 editing efficiency of approximately 67% (Fig. 1b) and retroviral CAR transduction efficiency on the order of approximately 50%. No discernible phenotypic differences were observed between infused edited and control CAR T cells (Extended Data Fig. 1a,b). CAR T cells were administered at low doses to better compare their antitumour efficacy ('stress test' condition 23 ). TET2-edited Rv-19BBz CAR T cells afforded greater survival of tumour-bearing mice than their unedited counterparts ( Fig. 1d and Extended Data Fig. 1d). By contrast, no survival difference was observed between recipients of TET2-edited or unedited Rv-1928z CAR T cells ( Fig. 1c and Extended Data Fig. 1c). Flow cytometric quantification and phenotyping of CAR T cells isolated from bone marrow and the spleen 3 weeks after their infusion revealed no significant difference in quantity (Extended Data Fig. 1e,f) and differentiation state (Extended Data Fig. 1g,h) between Rv-1928z CAR T cells. However, TET2-edited Rv-19BBz CAR T cells were more abundant than their unedited counterparts (Extended Data Fig. 1e,f). TET2-edited CAR T cells showed increased expression of CCR7 in Rv-19BBz CAR T cells but not in Rv-1928z CAR T cells (Extended Data Fig. 1g,h), whereas inhibitory receptor expression (PD1, LAG3 Article and TIM3) was indistinguishable between unedited and TET2-edited groups for both CAR designs (Extended Data Fig. 1i). Intrigued by the different outcome between the two CARs, we further evaluated the 1928z CAR design in two distinct contexts that extend its persistence, by either co-expressing 4-1BBL (Rv-1928z + 4-1BBL) 23 or by transcribing the CAR from the TRAC locus (TRAC-1928z) 24 (Fig. 1a). As with Rv-1928z and Rv-19BBz, TET2 editing did not affect CAR transduction efficiency or the pre-infusion T cell phenotype of either CAR T cell populations (Extended Data Fig. 2a,b), but their efficacy was increased relative to their non-edited counterparts (Fig. 1e,f and Extended Data Fig. 2c,d). This increased efficacy was associated with increased expression of CCR7 in Rv-1928z + 4-1BBL andTRAC-1928z CAR T cells (Extended Data Fig. 2e,f). Although, it did not reach statistical significance for Rv-1928z + 4-1BBL (Extended Data Fig. 2f). Inhibitory receptor expression was similar between wild-type (WT) and TET2-edited groups for both Rv-1928z + 4-1BBL andTRAC-1928z CARs (Extended Data Fig. 2g). These findings thus established that disruption of TET2 could augment therapeutic efficacy of either CAR, albeit depending on CAR expression.

Hyperproliferative CAR T cells emerge
Continued follow-up of these mice uncovered signs of clinical distress developing after 50 days in the absence of detectable tumour in mice treated with TET2-edited T cells ( Fig. 2a and Extended Data Fig. 2c,d). Gross pathology revealed an enlarged spleen and liver, pale kidneys, and lungs with extensive T cell infiltration and absence of CD19 + leukaemia. The infiltrating T cells were CAR + and Ki67 + (Fig. 2b). This prompted us to treat additional cohorts of mice with all four CAR T cell types (Rv-19BBz, Rv-1928z, Rv-1928z + 4-1BBL and TRAC-1928z), administering 2-5 × 10 5 CAR T cells to ensure tumour elimination in most mice to allow for long-term follow-up of all four groups (Fig. 2c). All CARs maintained long-term tumour remission as assessed by bioluminescence imaging (BLI), tumour was eliminated within 2-3 weeks of CAR T cell administration. Mice treated with Rv-CARs were euthanized on day 90 and TRAC-CAR T cells recipients on day 75. Bone marrow and splenic CAR T cell numbers were considerably increased in mice treated with TET2-edited Rv-19BBz, Rv-1928z + 4-1BBL andTRAC-1928z CAR T cells, compared with their unedited counterparts (Fig. 2d). CAR T cell numbers in recipients of TET2-edited and unedited Rv-1928z in both the bone marrow and the spleen, however, did not significantly differ (Fig. 2d), except for a single mouse (1 out of 10) that showed an increase in TET2-edited CAR T cells. Flow cytometric analysis of CAR T cells isolated from the bone marrow confirmed increased expression of CCR7 in TET2-edited Rv-19BBz, Rv-1928z + 4-1BBL and TRAC-1928z CAR T cells, but not in TET2-edited Rv-1928z CAR T cells (Extended Data Fig. 3a,b). Inhibitory receptor expression was again unchanged upon TET2 editing across all four CAR designs (Extended Data Fig. 3c). This long-term follow-up thus confirmed that TET2 editing increases therapeutic efficacy and T cell accumulation, but with pathological consequences appearing weeks or months after tumour clearance.    TRAC-1928z (dose: 1 × 10 5 ; n = 15) (f) CAR T cells. Data were collated from two donors. Untreated n = 5. Log-rank Mantel-Cox test was used; P < 0.05 was considered statistically significant. P values are denoted: not significant (NS) P > 0.05, *P < 0.05, **P < 0.01 and ***P < 0.001 (c-f). The human, mouse and lipid bilayer illustrations in part a were generated using Servier Medical Art, CC BY 3.0.
To ascertain that acquisition of this hyperproliferative phenotype was not specific to a single tumour model or a particular guide RNA (gRNA), we established a human prostate cancer model in NSG mice (Extended Data Fig. 4a), targeting prostate-specific membrane antigen (PSMA) in PC3-bearing mice with PSMA-28z + 4-1BBL CAR T cells. Peripheral blood PSMA-28z + 4-1BBL CAR T cells edited with either gRNA-g1 or gRNA-g2 were tenfold more abundant than control PSMA-28z + 4-1BBL CAR T cells edited with a scrambled gRNA by day 30 (Extended Data Fig. 4b).
Splenic CAR T cell quantification revealed over 100 million CAR T cells per spleen in recipients of TET2-edited PSMA-28z + 4-1BBL by day 45 (Extended Data Fig. 4c), establishing that late acquisition of a hyperproliferative phenotype is not specific to a tumour model or gRNA.
As Cas9-mediated TET2 editing resulted in either unedited, monoallelic or biallelic disruption in individual T cells, we could test whether total loss of TET2 is required for achieving sustained proliferation. For this analysis, we focused on two Rv-1928z CAR populations yielding different rates of T cell accumulation: Rv-1928z andRv-1928z + 4-1BBL (Extended Data Figs. 5a and7a). Pre-infusion, Rv-1928z andRv-1928z + 4-1BBL CAR T cells showed similar TET2-editing efficiency (Fig. 3a,b). By day 21 post-infusion, TET2 editing was enriched for Rv-1928z + 4-1BBL but not Rv-1928z (Extended Data Fig. 5b). In subsequent follow-up, very high T cell counts were reached in 12 out of 15 mice treated with TET2-edited Rv-1928z + 4-1BBL, but only in 2 of 15 mice treated with TET2-edited Rv-1928z CAR T cells, becoming apparent by day 90 and day 200. In these latter two cases (2-2 and 2-00), we found a 19-bp deletion in both alleles in 2-2 ( Fig. 3e) and a biallelic integration of a partial retroviral vector fragment in 2-00 (Extended Data Fig. 5c). Five of the expanded TET2-edited Rv-1928z + 4-1BBL populations harvested at day 90 were randomly selected for analysis and all were found to be nearly entirely (more than 98%) biallelically TET2-edited ( Fig. 3f and Extended Data Fig. 5d-g). Western blot analysis showed an absence of TET2 protein in biallelically edited CAR T cells (Extended Data Fig. 5h). Thus, biallelic TET2 editing (TET2 bed ) is enriched over time, irrespective of CAR design, consistent with it being required for achieving a hyperproliferative T cell state.
We assessed clonal composition in the hyperproliferative CAR T cell populations by TCRvβ sequencing. All five Rv-1928z + 4-1BBL populations were multiclonal, with no clone constituting more than 50% of the total CAR product, except for sample 17-1 in which a single clone accounted for approximately 82% of the CAR T cells ( Fig. 3f and Extended Data Fig. 5d-g). By contrast, both Rv-1928z populations (2-2 and 2-00) largely consisted in a single clone (more than 95%; Fig. 3e and Extended Data Fig. 5c), consistent with the lesser probability of 1928z CAR T cells achieving clonal expansion. TCRvβ sequencing of hyperproliferative TRAC-1928z and Rv-19BBz also revealed multiclonal expansion (Extended Data Fig. 5i,j).
Lack of shared TCRs between different hyperproliferative populations, the absence of graft-versus-host disease in mice bearing hyperproliferative CAR T cell population and the emergence of clonal dominance in TRAC-1928z CAR T cell-treated mice (in which CAR T cells lack TCR expression 24 ) strongly suggested that the TCR is not required for acquisition of a hyperproliferative phenotype. To further exclude a role for TCR in sustained clonal expansion, we ablated TCR expression in conjunction with TET2 disruption before transduction of Rv-1928z + 4-1BBL and compared the frequency of emergence of the hyperproliferative phenotype in recipient mice. Long-term follow-up of TCR + TET2-edited and TCR − TET2-edited Rv-1928z + 4-1BBL CAR T cells revealed no differences in frequency of CAR T cells achieving a hyperproliferative state and their differentiation state (Extended Data Fig. 6a-c), confirming that TCR is not required for sustained proliferation.

Article
(pre-infusion CAR T cells) to day 21 (Extended Data Fig. 7b,c). This divergent evolution is illustrated by tracking the persistence of the 100 most frequent clones in the Rv-1928z pre-infusion cell population, all of which were also present in the Rv-1928z + 4-1BBL pre-infusion product (Extended Data Fig. 7d). By day 21, most (70 out of 100) of these clones were still detected in Rv-1928z + 4-1BBL CAR T cells, whereas only 3 out of 100 were detectable in recipients of Rv-1928z CAR T cells (Extended Data Fig. 7d). By retro-tracking clones present in hyperproliferative populations (day 90) to pre-infusion, we found few persisting clones for Rv-1928z in contrast to Rv-1928z + 4-1BBL (Extended Data Fig. 7e,f), even though both Rv-1928z and Rv-1928z + 4-1BBL had similar pre-infusion clonal diversity (Extended Data Fig. 7g).
The difference between Rv-1928z andRv-1928z + 4-1BBL in their respective clonal longevity was further evidenced by tracking the 100 most frequent shared clones from the pre-infusion Rv-1928z and Rv-1928z + 4-1BBL CAR populations up to day 90. None was detected in Rv-1928z (Extended Data Fig. 7h), whereas some of the earliest clones detected on day 0 in the Rv-1928z + 4-1BBL population remained detectable by day 90 (Extended Data Fig. 7i), although they were not dominant (Extended Data Fig. 7j). These tracking data confirmed that the probability of a given clonotype acquiring a hyperproliferative phenotype upon loss of TET2 is determined by the CAR and that the relative resistance imparted by Rv-1928z could be overcome on engaging the 4-1BB pathway by overexpressing 4-1BBL.

Reduced effector function in CAR T cells
To assess the functional properties of the hyperproliferative CAR T cell population, we first evaluated the cytolytic function of hyperproliferative TET2 bed CAR T cells in vitro and in vivo. TET2 bed CAR T cells demonstrated diminished cytolytic ability and were relatively ineffective for eliminating established NALM6 in vivo ( Fig. 4a and Extended Data Fig. 8a,b), requiring a higher CAR T cell dosage to delay tumour progression (Extended Data Fig. 8c). TET2 bed CAR T cells showed a profound loss of effector cytokine secretion upon activation (Fig. 4b). For further molecular characterization, we focused on Rv-1928z + 4-1BBL

CAR T cells as the unedited Rv-1928z + 4-1BBL CAR T cells persisted
the most among the four tested CAR designs and thus could provide a matched, unedited control. Transcriptional profiling of hyperproliferative TET2 bed and WT Rv-1928z + 4-1BBL CAR T cells revealed an increased expression of cell-cycle-related factors in the former (Fig. 4c,d).  Fig. 8j,k). Collectively, these observations establish that TET2 deficiency leads to a gradual erosion of effector function but predisposes to the emergence of TET2 bed CAR T cell clones that are characterized by sustained proliferation, moderate cytolytic potential and poor cytokine responses. Consistent with this functional profile, we did not find expression of the memory-associated transcription factor TCF1 (Extended Data Fig. 8l) or an enrichment of memory gene sets in TET2 bed compared with WT Rv-1928z + 4-1BBL (Fig. 4f), despite the increased expression of some memory-associated biomarkers such as CCR7. Instead, we found enrichment in angioimmunoblastic T cell lymphoma and HTLV1-driven adult T cell leukaemia/lymphoma datasets (Fig. 4g). This led us to search for potential genetic drivers of proliferation and investigate the proliferative potential of TET2 bed CAR T cells upon secondary transplant.

BATF3 drives hyperproliferation
To assess whether TET2 bed clones had acquired mutations that could account for their clonal dominance, we performed whole-exome sequencing in three clones expressing different CARs (Extended Data Fig. 9a,c,e). Numerous non-synonymous point mutations were observed in all three dominant clones (Extended Data Fig. 9b,d,f).   Table 1). Some chromosomal amplifications and megabase-scale deletions were observed in a subset of the dominant clone population in samples 17-1 and 4-1 (Extended Data Fig. 9a,e). Given their substantially lower frequency than that of the dominant clone, these gross chromosomal defects appeared to be late occurring secondary events. For the retroviral-encoded CARs in samples 17-1 and 2-2, we identified the sites of retroviral integration. None of them disrupted or integrated next to cancer-related genes associated with angioimmunoblastic T cell lymphoma or T cell lymphoma (Supplementary Table 2). Together, we found that hyperproliferative TET2 bed T cells are prone to acquiring somatic mutations, but do not bear recurrent genetic mutations or mutations known to be associated with T cell malignancies.
Secondary transplant studies of TET2 bed CAR T cells have shown that they did not engraft on their own, but could persist with exogenous cytokine supplementation, promptly declining after cessation of cytokine administration (Extended Data Fig. 10a,b). Cell numbers remained modest and were barely detectable at day 150 when the study reached its intended end point (Extended Data Fig. 10c). These findings indicate that TET2 bed CAR T cells are unable to autonomously sustain their proliferation upon secondary transplant.
The lack of a conserved genetic driver of proliferation of TET2 bed CAR T cells prompted us to study whether their epigenetic state enables sustained proliferation. Assay for transposase-accessible chromatin using sequencing analysis revealed significant differences between accessible chromatin regions of WT and TET2 bed Rv-1928z + 4-1BBL CAR T cells ( Supplementary Fig. 1a). The AP-1 family binding motif was the most significantly enriched motif in differentially open chromatin regions of TET2 bed CAR T cells (Fig. 5a). Transcriptional analyses in these same cells revealed that, among the AP-1 factors, BATF3 was the most significantly upregulated in TET2 bed CAR T cells (Fig. 5b). BATF3 has been previously implicated as a driver of proliferation in T cell leukaemia/lymphoma 25-27 in part by inducing a MYC transcriptional program 25,26 . Distinct promoter and gene body regions of BATF3, with some encompassing consensus AP-1-binding motifs, were found to be more readily accessible in hyperproliferative TET2 bed Rv-1928z + 4-1BBL CAR T cells than WT Rv-1928z + 4-1BBL CAR T cells (Fig. 5c,d and Supplementary Fig. 1b). TET2 bed Rv-1928z + 4-1BBL CAR T cells showed a strong enrichment in hallmark MYC targets when compared with WT Rv-1928z + 4-1BBL CAR T cells (Fig. 5e). Flow cytometric analyses of unedited and hyperproliferative CAR T cells isolated at day 90 showed a higher fraction of BATF3 + MYC + in hyperproliferative CAR T cells (Fig. 5f,g). Analysis of BATF3 and MYC expression upon TET2 editing in CAR T cells ( Supplementary Fig. 1c) revealed that CAR activation induced BATF3 expression ( Supplementary Fig. 1d). The levels of BATF3 and MYC did not differ between WT and TET2-edited CAR T cells at early time points (days 1 and 8) ( Supplementary Fig. 1e), but increased after five rounds of Running enrichment score Article stimulation (day 15) in the TET2-edited group (BATF3, Supplementary  Fig. 1f; MYC, Supplementary Fig. 1g). These observations suggest that TET2 deficiency gradually establishes an epigenetic state conducive to increased BATF3 and MYC expression that may ultimately result in the sustained proliferation of TET2 bed CAR T cell clones.
To directly test the role of BATF3 in acquisition of the hyperproliferative state, we designed an in vivo study in which BATF3 and TET2 are both edited in Rv-1928z + 4-1BBL CAR T cells ( Supplementary  Fig. 2a), hypothesizing that in-frame BATF3 edits would be enriched, and out-of-frame edits would be counter-selected over time. NALM6-bearing mice were treated with a predictably curative dose of BATF3 and TET2 etd Rv-1928z + 4-1BBL CAR T cells to allow for long-term monitoring (Fig. 5h). Deep sequencing at both TET2 and BATF3 loci indeed confirmed an enrichment of out-of-frame edits at the TET2 locus and in-frame edits at the BATF3 locus when the hyperproliferative population emerged at day 50 (Fig. 5i), confirming the essential requirement for BATF3 expression to acquire the hyperproliferative phenotype.
We further corroborated this dependency on BATF3 pharmacologically. JQ1 is an inhibitor of the BET protein BRD4, which has been previously shown to inhibit BATF3 and MYC expression in adult T cell leukaemia/lymphoma cells 26 . Although JQ1 inhibited proliferation of      TET2  TET2   TET2  TET2   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET  TET2 TET2 ET   TET2  TET2   TET2  TET2   TET2  TET2 TET2 TET2 ET   TET2 TET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2  TET2   TET2  TET2   TET2 TET2 ET  TE ET2  TE ET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2 TET2 ET  TET2 TET2 ET   TET2  TET2   TET2 TET2 ET   TET2  TET2   TET2 TET2 ET  TET2 TET2 ET   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET   TET2  TET2   TET2 TET2 ET  TET2 TET2 ET   TET2  TET2   TET2 TET2 ET  TE ET2 ET2  TE ET   BATF3   MYC   TET2  TET2   TET2 TET2 ET  TET T2 T2  TET    all tested CAR populations, TET2 bed CAR T cells were more sensitive to JQ1 treatment than were pre-infusion TET2-edited CAR T cells ( Fig. 5j and Supplementary Fig. 2b,c). This heightened sensitivity to JQ1 was associated with a greater suppression of BATF3 and MYC expression in TET2 bed CAR T cells ( Fig. 5l and Supplementary Fig. 2d,e). Dexamethasone has been shown to suppress AP-1 factors 28,29 . In contrast to JQ1, dexamethasone did not limit proliferation of pre-infusion TET2-edited CAR T cells ( Fig. 5k and Supplementary Fig. 2e). However, it markedly inhibited proliferation of TET2 bed CAR T cells ( Fig. 5k and Supplementary  Fig. 2e). This increased sensitivity to dexamethasone was associated with reduction in expression of both BATF3 and MYC in TET2 bed CAR T cells (Fig. 5m and Supplementary Fig. 2f). By contrast, MYC expression remained elevated in pre-infusion TET2-edited CAR T cells despite BATF3 inhibition ( Fig. 5m and Supplementary Fig. 2f). This differential expression of MYC between pre-infusion TET2-edited CAR T cells and TET2 bed CAR T cells upon exposure to dexamethasone further supports the dependency of MYC expression on BATF3 in TET2 bed CAR T cells.

Discussion
We found that biallelic TET2 disruption can enhance the efficacy and proliferation of human CAR T cells. The increased efficacy is consistent with earlier observations in mouse TCR transgenic T cells 9 , and in a case report of a clonal expansion in a patient treated with a 4-1BB CAR 10 . The functional enhancement of CAR T cell antitumour activity, however, depends on the CAR and not merely the TET2 status. Tumour elimination by TET2-edited T cells was enhanced with Rv-19BBz, Rv-1928z + 4-1BBL and TRAC-1928z, but not Rv-1928z. Over time, TET2 bed CAR T cells repeatedly emerged in a hyperproliferative state, consistently associated with CAR expression, albeit not requiring the presence of CAR antigen, and independently of the TCR. TCRvβ sequencing identified multiple T cell clones in mice treated with TET2-edited Rv-19BBz, Rv-1928z + 4-1BBL andTRAC-1928z, whereas hyperproliferative TET2-edited Rv-1928z CAR T cells were rare and monoclonal when they occurred. These observations underscore a probabilistic fate, in which the chance of establishing a hyperproliferative state in TET2-edited T cells is low in shorter-lived Rv-1928z CAR T cells, rarely allowing rare breakthrough clones, and increased with CAR designs that autonomously promote greater persistence. Hyperproliferative TET2 bed CAR T cells bear secondary mutations, consistent with previous reports that have shown a role for TET2 in maintaining genomic integrity 30,31 . However, we did not identify a recurrent mutation among different TET2 bed CAR T cell populations or mutations known to be associated with T cell leukaemia/lymphoma 22 . Instead, we identified a strict requirement for BATF3 expression, associated with an epigenetic signature characterized by enhanced BATF3 and MYC accessibility. AP-1 factors are critically involved in distinct T cell states [32][33][34][35] . Batf overexpression in mouse CAR T cells enhances their antitumour activity 33 . BATF3 overexpression in T cells enhances their CCR7 expression and memory formation 36,37 , although high levels of BATF3 has also been associated with human CAR T cell exhaustion 32 . We found here that in the context of epigenetic changes brought on by loss of TET2, sustained BATF3 expression programs a hyperproliferative state rather than T cell memory. Furthermore, TET2 bed CAR T cells demonstrate reduced cytolytic function and poor cytokine response upon activation, despite maintaining genome accessibility in effector loci ( Supplementary Fig. 3), which suggests that effector functions are transcriptionally downregulated. These observations point to a T cell state that differs from both canonically defined T cell exhaustion 15,16 and T cell memory 13,14 . Our findings thus establish TET2 as an epigenetic regulator of BATF3 to prevent unchecked proliferation and maintain T cell genomic integrity.
Several AP-1 factors are known to potentially promote oncogenesis 38 . JUN and BATF overexpression can lead to uncontrolled proliferation 39,40 . BATF3 has been shown to drive proliferation in T cell leukaemia/ lymphoma through MYC 26 or IL-2R 27 . Sustained BATF3 expression in T cell leukaemia/lymphoma is associated with activated super-enhancers at the BATF3 locus 26,27 . The hyperproliferative CAR T cell phenotype that we report here underscores the potency of CAR T cell epigenetic programming but reveals long-term safety concerns that may arise from manipulating TET2 (ref. 41 ) and AP-1 factors 38 . Remarkably, however, TET2 bed CAR T cells remained highly sensitive to dexamethasone, which lowered the expression of both BATF3 and MYC in TET2 bed CAR T cells. This high sensitivity may explain the sudden clonal contraction upon corticosteroid administration to manage cytokine release syndrome observed in the patient bearing a TET2-deficient 19BBz CAR T cell clone 10 . The intentional disruption of TET2 for CAR T cell therapy may nonetheless be concerning, especially in elderly individuals who are more likely to have mutations in DNMT3A 42 , which can synergize with loss of TET2 to precipitate T cell oncogenesis 43 . Screening for pre-existing mutations that predispose to hyperproliferation or transformation should help to mitigate this hazard. Transient or partial suppression of TET2 during CAR T cell production 44 may eschew such a risk.
In summary, disruption of TET2 enhances CAR T cell efficacy and promotes sustained T cell accumulation but exposes to the risk of a hyperproliferative state that is prone to accumulating secondary mutations. These findings demonstrate the formidable potential of epigenetic reprogramming to alter CAR T cell fate and highlight how an AP-1 factor, such as BATF3, may direct distinct effector and proliferative states under different epigenetic contexts.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05692-z. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Data reporting
No statistical methods were used to predetermine sample size. The experiments were not randomized and, unless otherwise stated, the investigators were not blinded to allocation during experiments and outcome assessment.

Retroviral vector constructs and retroviral production
Plasmids encoding the retroviral vector were prepared using standard molecular biology techniques 45 . LNGFR is a truncated and mutated TNFR family homologue 46 , which was used as a control molecule to ensure comparable CAR expression levels from different bicistronic vectors. Synthesis of Rv-1928z, Rv-19BBz andRv-1928z + 4-1BBL has been previously described 23,47,48 . VSV-G-pseudotyped retroviral supernatants derived from transduced gpg29 fibroblasts (H29) were used to construct stable retroviral-producing cell lines as previously described 49 .

AAV targeting for TRAC-1928z
The TRAC gRNA targets a sequence upstream of the transmembrane domain of the TCRα. This domain is required for TCRα and TCRβ assembly and trafficking to the cell surface. Both non-homologous end joining and integration of the CAR by homology-directed repair (HDR) at this locus would then efficiently disrupt the TCR complex 23 . TRAC-1928z is based on the pAAV-GFP backbone (Cell Biolabs). It contains 1.9 kb of genomic TRAC flanking the gRNA targeting sequence, a self-cleaving P2A peptide in-frame with the first exon of TRAC followed by the 1928z CAR used in clinical trials 24,50 .

Isolation and expansion of human T cells
Buffy coats from anonymous healthy donors were purchased from the New York Blood Centre (institutional review board exempted) and peripheral blood was obtained from healthy volunteers. All blood samples were handled following the required ethical and safety procedures. Peripheral blood mononuclear cells were isolated by density gradient centrifugation. T cells were then purified by using the Pan T Cell Isolation kit (Miltenyi Biotec). T cells were stimulated with CD3/CD28 T cell activator Dynabeads (Invitrogen) at 1:1 ratio and cultured in RPMI + 10% FBS, 5 ng ml −1 IL-7 and 5 ng ml −1 IL-15 (Miltenyi Biotec) for retroviral transduction (Rv-CAR) and gene targeting (TRAC-1928z) experiments. The medium was changed every 2 days, and cells were plated at 10 6 cells per millilitre.

Flow cytometry
CAR expression was measured with Alexa-Fluor-647-conjugated goat anti-mouse Fab (115-606-072, Jackson ImmunoResearch). The flow cytometry antibodies used for cell-surface phenotyping are provided in Supplementary Table 1. For intracellular staining, cells were fixed and permeabilized using Foxp3/Transcription Factor staining kit (00-5523-00, eBioscience) according to the manufacturer's protocol. The flow cytometry antibodies used for intracellular studies are provided in Supplementary Table 5. Data were analysed by FlowJo v10.1 (BD). Cell sorting was performed on a BD FACSAria cell sorter. The gating strategies for flow cytometry are provided in Supplementary Fig. 4.

Mouse leukaemia and prostate tumour models
We used 6-12-week-old NOD/SCID/IL-2Rγ null mice (The Jackson Laboratory), under a protocol approved by the Memorial Sloan Kettering Cancer Centre (MSKCC) Institutional Animal Care and Use Committee. All relevant animal use guidelines and ethical regulations were followed. NALM6-expressing firefly luciferase (FFLuc)-GFP cells have been previously described 48 . For the leukaemia model, either male or female mice were inoculated with 5 × 10 5 FFLuc-GFP NALM6 cells by tail vein injection; CAR T cells were then injected 4 days later at varying doses. For the prostate cancer model, male mice were inoculated with 2 × 10 6 PC3-PSMA FFLuc-GFP 51 cells by tail vein injection; CAR T cells were then injected 4 weeks later. Both NALM6 and PC3 cells produced similar tumour burdens, and no mice were excluded before treatment. For the TRAC-1928z stress test, long-term CAR T cell assessment (Fig. 2) and prostate cancer model (Extended Data Fig. 4), a scrambled gRNA GCACUACCAGAGCUAA CUCA was used as a control. No randomization or blinding methods were used. BLI was performed using the IVIS Imaging System (PerkinElmer) with the Living Image V4.4 software (PerkinElmer) for the acquisition of imaging datasets.

Secondary transplant of TET2 bed CAR T cells
A day before the transplant, NSG mice were irradiated with a cumulative dose of 200 cGy. TET2 bed CAR T cells (2 × 10 6 ) were then injected through the tail vein. For the IL-2 treatment group, mice received 1,000 U of IL-2 twice a week (intraperitoneal). For the IL-7 + IL-15 treatment group, IL-7 was subcutaneously injected at 0.5 μg per mouse per week. IL-15 and IL-15RA were pre-incubated at a 1:6 weight ratio at 37 °C for 30 min before injection (intraperitoneal) in mice at a dose of 2.5 μg (IL-15) + 15 μg (IL-15RA) per week 52 . Mice received exogenous cytokines for 60 days.

Cytotoxicity assays
The cytotoxicity of T cells transduced with a CAR was determined by a luciferase-based assay. NALM6-expressing FFLuc-GFP cells served as target cells. The effector and tumour cells were co-cultured at indicated effector to target (E/T) ratio in the black-walled 96-well plates in triplicate manner with 1 × 10 5 target cells in a total volume of 100 μl per well. Target cells alone were planted at the same cell density to determine the maximal luciferase expression (relative light units (RLU max )). Eighteen hours later, 100 μl luciferase substrate (Bright-Glo, Promega) was directly added to each well. Emitted light was measured by a luminescence plate reader or the Xenogen IVIS Imaging System (Xenogen) with Living Image V4.4 software (Xenogen) for acquisition of imaging datasets. Lysis was determined as (1 − (RLU sample )/ (RLU max )) × 100.

DNA-RNA simultaneous extraction
Cell pellets were resuspended in RLT buffer and nucleic acids were extracted using the AllPrep DNA/RNA Mini Kit (80204, Qiagen) according to the manufacturer's instructions. RNA was eluted in nuclease-free water and DNA in 0.5X buffer EB. Phase separation in cells lysed in TRIzol reagent (15596018, Thermo Fisher) was induced with chloroform. RNA was precipitated with isopropanol and linear acrylamide and washed with 75% ethanol. The samples were resuspended in RNase-free water.

Transcriptome sequencing
After RiboGreen quantification and quality control by Agilent BioAnalyser, 2 ng total RNA with RNA integrity numbers ranging from 7.3 to 9.7 underwent amplification using the SMART-seq v4 Ultra Low Input RNA Kit (63488, Clonetech), with 12 cycles of amplification. Subsequently, 10 ng of amplified cDNA was used to prepare libraries with the KAPA Hyper Prep Kit (KK8504, Kapa Biosystems) using 12 cycles of PCR. Samples were barcoded and run on a HiSeq 4000 in a PE50 run, using the HiSeq 3000/4000 SBS Kit (Illumina). An average of 40 million paired reads were generated per sample and the percent of mRNA bases per sample ranged from 31% to 69%. DESeq2 was used for normalization and differential analysis of transcriptional data.

TCR sequencing
After PicoGreen quantification and quality control by Agilent BioAnalyser, 188-200 ng of genomic DNA was split equally into six reactions and prepared using the immunoSEQ human TCRB Kit (Adaptive Biotechnologies) according to the manufacturer's instructions. In brief, multiplex PCR was used to amplify the CDR3 region for 31 cycles. After clean-up, 2 μl of PCR product was used as input into library preparation with eight cycles of PCR. Barcoded samples were pooled by volume and Article sequenced using custom primers on a NextSeq 500 in a SR155 run, using the NextSeq 500/550 Mid Output Kit v2.5 (150 cycles; Illumina). The loading concentration was 1 pM, and 20% spike-in of PhiX was added to the run to increase diversity and for quality control purposes. Raw BCL files were transferred to the immunoSEQ Analyser for processing and analysis.

Exome capture and sequencing
After PicoGreen quantification and quality control by Agilent BioAnalyser, 250 ng of DNA were used to prepare libraries using the KAPA Hyper Prep Kit with eight cycles of PCR. After sample barcoding, 100 ng of library were captured by hybridization using the xGen Exome Research Panel v1.0 (IDT) according to the manufacturer's protocol. PCR amplification of the post-capture libraries was carried out for eight cycles. Samples were run on a HiSeq 4000 in a PE100 run, using the HiSeq 3000/4000 SBS Kit (Illumina). Samples were covered to an average of 111×.

ATAC-seq
Profiling of chromatin was performed by assay for transposaseaccessible chromatin using sequencing (ATAC-seq) as previously described 53

S-EPTS/LM-PCR integration site analysis
Shearing-extension primer tag selection ligation-mediated PCR (S-EPTS/LM-PCR) is a shearing DNA-based integration site analysis method in orientation to the original EPTS/LM-PCR 54 . S-EPTS/LM-PCR starts with shearing of genomic DNA to an intended length of 500 bp using the Covaris M220 instrument. Sheared DNA is split into three equal replicates (500 ng each) and purified, followed by primer extension using two vector, long-terminal-repeat-specific biotinylated primers. The extension product is purified, and biotinylated DNA captured by paramagnetic beads. The captured DNA is ligated to linker cassettes including a molecular barcode, and the ligation product is amplified in an exponential PCR using biotinylated vector-specific and linker-cassette-specific primers. Biotinylated PCR products are magnetically captured, washed and used as template for amplification in a second exponential PCR with barcoded primers, allowing sequencing by MiSeq technology (Illumina). Final preparation for sequencing was done as previously described 55,56 . Applied DNA double barcoding allowed for parallel sequencing of multiple samples in a single sequencing run while minimizing sample cross-contamination. Amplicons were then sequenced on the MiSeq instrument using the V2 Reagent Kit (Illumina).

Integration site computational analysis
Raw sequence data were trimmed according to sequence quality (Phred) and only sequences showing complete identity in both molecular barcodes (linker cassette barcode and sequencing barcodes) were further analysed. An in-house semi-automated bioinformatical data mining pipeline was used to analyse the data 57 . In brief, quality-filtered sequences were trimmed (vector-specific and linker-cassette-specific parts removed) and only sequences that showed at least 18 nucleotides of expected, vector-specific sequence were analysed further to ensure the analysis of true vector-genome junctions. Such trimmed sequences were further filtered in a way that only sequences equal to or larger than 25 bp were aligned to the human genome (UCSC assembly release number hg38, version 3) by Burrows-Wheeler Aligner MEM algorithm (version 0.7.17) for the initial alignment 58 . It was subsequently followed by mapping of potential integration site sequences with BLAST, in which the minimum alignment identity percentage of 95% is used, whereas nearby genes and other integrating features were annotated as previously described according to RefSeq database 59 . The relative sequence count of each detected integration site was calculated in relation to all sequences attributed to the corresponding sample.

Statistical analysis
All statistical analyses were performed using the Prism 9 (GraphPad) software. No statistical methods were used to predetermine sample size. Statistical tests are provided in the figure legends. The Kolmogorov-Smirnov test was used to determine P values in gene set enrichment analysis: *P < 0.05, **P < 0.01, ***P < 0.0001 and ****P < 0.00001.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Data generated from RNA-seq and ATAC-seq have been deposited in the Gene Expression Omnibus with the accession number GSE220259. The publicly available datasets used in this study are GSE23321 for central memory and effector memory phenotype comparison, AKL_HTLV1_UP (M7705), AKL_HTLV1_DN (M9815), the angioimmunoblastic T cell lymphoma dataset (GSE6338) and HALLMARK_MYC_V1 (M5926). Source data are provided with this paper. Data is represented as mean±SD (n = 5 for no supplement, and IL2. n = 4 for IL7/15). p values were determined by two-sided Mann-Whitney test (b). p < 0.05 was considered statistically significant. p values are denoted: p > 0.05, not significant, NS; *, p < 0.05; **, p < 0.01. (b). Exact p values are available in Supplementary Table 4. The mouse illustration in part a was generated using Servier Medical Art, CC BY 3.0.

Corresponding author(s): Michel Sadelain
Last updated by author(s): 12/8/2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above.

Software and code
Policy information about availability of computer code Data collection Xenogen IVIS Imaging System were used to image mice. BD Fortessa or Cytek Aurora cytometers were used to collect flow cytometry data.
Tecan Spark microplate reader was used to collect CTL data.

Data analysis
FlowJo ( For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy Data generated from RNAseq and ATACseq experiments has been deposited in GEO (Accession number:GSE220259). The publicly available datasets used in this study are GSE23321 for central memory and effector memory phenotype comparison, AKL_HTLV1_UP (M7705), AKL_HTLV1_DN (M9815), AITL dataset (GSE6338), HALLMARK_MYC_V1 (M5926).

Recruitment
Buffy coats from anonymous healthy donors were purchased from the New York Blood Centre. New York Blood Centre recruits healthy donors under a broad consent covering in vitro laboratory research. Peripheral blood was obtained from healthy volunteers regardless of gender and age.

Ethics oversight
All human blood samples were approved by MSKCC IRB and handled following the required safety procedures. Human buffy coats obatined from New York Blood Centre were IRB-exempt.
Note that full information on the approval of the study protocol must also be provided in the manuscript.

Flow Cytometry
Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC).
The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers).
All plots are contour plots with outliers or pseudocolor plots.
A numerical value for number of cells or percentage (with statistics) is provided.

Methodology
Sample preparation