Single-cell CAR T atlas reveals type 2 function in 8-year leukaemia remission

Abstract

Despite a high response rate in chimeric antigen receptor (CAR) T cell therapy for acute lymphocytic leukaemia (ALL)^1-3, approximately 50% of patients relapse within the first year^4-6, representing an urgent question to address in the next stage of cellular immunotherapy. Here, to investigate the molecular determinants of ultralong CAR T cell persistence, we obtained a single-cell multi-omics atlas from 695,819 pre-infusion CAR T cells at the basal level or after CAR-specific stimulation from 82 paediatric patients with ALL enrolled in the first two CAR T ALL clinical trials and 6 healthy donors. We identified that elevated type 2 functionality in CAR T infusion products is significantly associated with patients maintaining a median B cell aplasia duration of 8.4 years. Analysis of ligand-receptor interactions revealed that type 2 cells regulate a dysfunctional subset to maintain whole-population homeostasis, and the addition of IL-4 during antigen-specific activation alleviates CAR T cell dysfunction while enhancing fitness at both transcriptomic and epigenomic levels. Serial proteomic profiling of sera after treatment revealed a higher level of circulating type 2 cytokines in 5-year or 8-year relapse-free responders. In a leukaemic mouse model, type 2^high CAR T cell products demonstrated superior expansion and antitumour activity, particularly after leukaemia rechallenge. Restoring antitumour efficacy in type 2^low CAR T cells was attainable by enhancing their type 2 functionality, either through incorporating IL-4 into the manufacturing process or by priming manufactured CAR T products with IL-4 before infusion. Our findings provide insights into the mediators of durable CAR T therapy response and suggest potential therapeutic strategies to sustain long-term remission by boosting type 2 functionality in CAR T cells.

Fig. 1. Single-cell atlas of 695,819 CAR T cells from 82 patients and 6 HDs.

a, Schematic of the experimental design. The diagram was created using BioRender. b, UMAP visualization of 695,819 high-quality single CAR T cells, filtered from 1,029,340 sequenced cells across all patients and donors. Unsupervised clustering identified 17 distinct clusters. c, The patient demographics and clinical documentation. Patients were divided into five persistence groups on the basis of their durations of BCA. The discovery cohort includes 42 patients from clinical trial NCT01626495, and the validation cohort consists of 40 patients from clinical trial NCT02906371. IPA, ingenuity pathway analysis; T_c, cytotoxic T cell; T_reg, regulatory T cell.

Fig. 2. Elevated type 2 functionality of CAR T products is associated with 8-year ultralong remission.

a–d, Unsupervised clustering analysis of CD19-3T3-stimulated CAR T cells from the discovery cohort (a) or validation cohort (c), along with the expression distribution of surface protein CD4 and CD8. The expression pattern of type 1 and type 2 score and cell proportion comparisons in specific clusters on the discovery cohort (b) or validation cohort (d) UMAP are presented. e, Frequency comparison of the type-2-cytokine⁺ population in CD4⁺CAR⁺ cells, assessed using intracellular flow cytometry. f, Comparison of type 2 cytokine secretion levels, assessed using a multiplexed secretomic assay on a cohort of 32 patients. g, Integrated ATAC–gene UMAP clustering of CD19-3T3-stimulated CAR T cells derived from six patients, with an enrichment of type 2 cells identified in cluster A3. A comparison of cell proportions in cluster A3 is presented between the patient groups. h, Pseudo-bulk chromatin accessibility tracks in the genomic region of GATA3, depicted separately for each patient. Enhancer elements predicted by ENCODE are highlighted in a light yellow. i, Differential motif activities in BCA-L versus BCA2 or 1 CAR T cells (left). Right, the expression profile of type 2 motif MA0037.3 (GATA3) and type 1 motif MA0690.1 (TBX21) across each patient. FC, fold change. j, Motif footprinting trace showing the transcription-factor-binding dynamics of GATA3 in the two patient groups, alongside its position weight matrices. k, Evaluation of tumour cell lysis efficacy and CAR T cell count in an in vitro repeat stimulation assay using CAR T cells with knockdown of STAT6 or GATA3. KD, knockdown. The diagram was created using BioRender. Data are mean ± s.e.m. from n = 48 (b), n = 46 (d), n = 42 (e), n = 32 (f) and n = 6 (g) patients or HDs, or n = 3 technical replicates for each condition (k). Significance levels were calculated using two-tailed Mann–Whitney U-tests (b, d, e, f, h and i), two-tailed unpaired Student’s t-tests (g) and one-way analysis of variance (ANOVA) with Tukey’s multiple-comparison test (k).

Fig. 3. Type 2 CAR T cells regulate dysfunctional subpopulation.

a, Identification of L–R interactions originating from type 2 enriched cluster 7 cells, primarily directed towards cluster 2. b, Identification of L–R interactions targeting cluster 2 cells. c, The expression profile of type 2 receptor genes. d, DEGs specific to cluster 2 cells, along with the expression pattern of cytotoxic score. Genes defining this module include GZMA, GZMB, GZMH, GNLY, PRF1 and NKG7. e, Signalling pathways regulated by the DEGs identified in cluster 2. f, Pseudo-bulk average expression of coinhibitory-related genes or proteins within each cluster. g, The expression pattern of proliferation score. h, Comparison of cell proportion in cluster 2. i, Experimental schematic for assessing the impact of IL-4 supplementation on the functional profile of CAR T cells derived from patients in the short-term BCA2 group. The diagram was created using BioRender. j, UMAP clustering analysis of CAR T cells with and without 10 ng ml⁻¹ IL-4 added. k, Comparison of cell proportions in specific clusters. l, Comparison of regulatory pathways between CAR T cells from patients in the BCA-L group and six patients in the BCA2 group under the original and 10 ng ml⁻¹ IL-4 conditions. For e and l, z > 0, activated/upregulated; z < 0, inhibited/downregulated; z ≥ 2 or z ≤ −2, significant. Data are mean ± s.e.m. from n = 48 (h) and n = 6 (k) patients or HDs. Significance levels were calculated using two-tailed Mann–Whitney U-tests (d and h), right-tailed Fisher’s exact tests (e) or two-tailed Wilcoxon matched-pairs signed-rank tests (k).

Fig. 4. Long-term responders exhibit higher type 2 cytokine levels in post-treatment sera.

a, Schematic of the serial proteomic profiling to measure serum proteins in 33 patients from the discovery cohort and 8 patients from the validation cohort. Timepoints are relative to the day of first infusion of CTL019 cells (day 0). The diagram was created using BioRender. b, Longitudinal levels of type 2 cytokines in patients from the discovery cohort. Four patients (with patient ID) in the BCA-L group are individually labelled with coloured lines. c, Comparison of the average type 2 cytokine levels at multiple timepoints between persistent groups in the discovery cohort. The value is the averaged expression of IL-4, IL-5 and IL-13. Data are mean ± s.e.m. from n = 45 (baseline), n = 54 (day 1–5), n = 38 (day 6–8), n = 35 (day 9–11), n = 36 (day 12–15), n = 32 (day 16–19), n = 32 (day 20–23), n = 54 (day 25–35) and n = 18 (day 36–63) measurements. Significance levels were calculated using two-tailed Mann–Whitney U-tests. d, Unsupervised clustering analysis of 345 measurements of serum samples from 33 patients in the discovery cohort, grouped by cluster ID or BCA response. Each dot represents one measurement of an individual patient at each timepoint. Cluster 2 exhibits enrichment in BCA-L patients. e, The differentially expressed proteins defining each cluster. Stars highlight type 2 cytokines with notable expression levels.

Fig. 5. Type 2^high CAR T cells demonstrate enhanced antitumour activity after leukaemia rechallenge.

a, Schematic of the in vivo leukaemia model treated with type 2^low (ND585) or type 2^high (ND463) CAR T cells. NSG mice were i.v. injected with 1 × 10⁶ Nalm6 cells. Then, 7 days later, mice were randomly assigned to three groups and were infused i.v. with 2 × 10⁶ CAR T cells or PBS (control). The surviving mice were rechallenged with 1 × 10⁶ Nalm6 cells 17 days after the CAR T infusion. The diagram was created using BioRender. b, CAR T cell expansion in the peripheral blood of Nalm6-bearing mice was measured at different timepoints after infusion. c, The tumour burden (total flux) was quantified as photons per second in mice after CAR T treatment. d, Kaplan–Meyer analysis of mouse survival. Data are mean ± s.e.m. from n = 5 mice for each group (b and c). Significance levels were calculated using two-tailed unpaired Student’s t-tests (b and c) or log-rank Mantel–Cox tests (d).

Fig. 6. Revitalizing type 2^low CAR T cells through enhanced type 2 functionality boost.

a, Schematic of the two strategies used to enhance the type 2 functionality of type 2^low CAR T cells derived from the donor ND585. The diagram was created using BioRender. b, The tumour burden was measured using bioluminescence at the indicated days after CAR T cell infusion. GVHD, graft-versus-host disease. c, CAR T cell expansion in the peripheral blood of Nalm6-bearing mice measured at different timepoints after infusion. Data are mean ± s.e.m. from n = 5 mice for each group. d, Kaplan–Meyer analysis of mouse survival. Significance levels were calculated using one-way ANOVA with Tukey’s multiple-comparison test (c) or log-rank Mantel–Cox test (d).

Extended Data Fig. 1. Single-cell multi-omics profiling of pre-infusion CAR T cells.

a, Experimental pipeline of this study. b, Cell proportion in each identified cluster, and the ratio of CD4+ and CD8+ cells in each stimulation condition determined by CITE-seq surface protein data. c, UMAP distribution of all the single cells grouped by in vitro stimulation condition (upper panel) or cell cycle (lower panel). d, Expression of surface protein ADT-CD69 (activated T cell marker), ADT-CD62L (naïve T cell marker), ADT-CD4 and ADT-CD8 (subtype T cell marker) on the UMAP. e, UMAP distribution of 695,819 single CAR T cells split by sequencing batch, suggesting minimum batch effect. f–m, Expression distribution of memory states (f), Th1/Tc1 (g), Treg (h), Th2/Th9/Th17 (i), selected chemokines (j), cytotoxicity (k), cell cycle (l), and other (m) gene markers on the UMAP shown in Fig. 1b. n, CAR T cell signatures within each identified cluster based on cell states, functions, regulations, or subsets. The heatmap illustrates the averaged expression of curated T cell marker genes in each cluster. o, Heatmap showing the average expression level of ADT proteins of all the single cells in each identified cluster.

Extended Data Fig. 2. Clinical persistence correlates with characteristics in basal CAR T cells.

a, UMAP clustering of basal CAR T cells from the Discovery Cohort patients and donors. b, Expression distribution of surface protein ADT-CD4 and ADT-CD8 on the UMAP. c, Comparison of cell proportion in each identified cluster between persistence groups. d, Expression distribution of Memory Score on the UMAP, and the cell proportion comparison in Cluster 0 + 4. e, Expression distribution of Proliferation Score on the UMAP, and the cell proportion comparison in Cluster 1. f, Expression distribution of Cytotoxic Score on the UMAP, and the cell proportion comparison in Cluster 3. g, Heatmap showing the average expression level of feature genes of all the basal CAR T cells in each BCA group. h, Heatmap showing the average expression level of ADT proteins of all the basal CAR T cells in each BCA group. In d–f, genes defining each module are listed below. Scatter plot shows mean ± s.e.m. from n = 48 (d–f) patients or healthy donors. Significance levels were calculated with two-tailed Mann-Whitney test (d–f).

Extended Data Fig. 3. Transcriptomic clustering and proteomic analysis of activated CAR T cells.

a, Cell proportion comparison in each identified cluster of the Discovery Cohort UMAP (Fig. 2a) across persistence groups. b, Expression distribution of Memory Score on the Discovery Cohort UMAP in Fig. 2a, and the cell proportion comparison in Cluster 0. c, Heatmap showing the average expression levels of ADT proteins in all activated CAR T cells within each BCA group. d, Expression distribution of each gene defining the Type-1 and Type-2 Score on the Discovery Cohort UMAP in Fig. 2a. e, Cell proportion comparison in each identified cluster of the Validation Cohort UMAP (Fig. 2c) across persistence groups. f, Expression distribution of Memory Score on the Validation Cohort UMAP in Fig. 2c, and the cell proportion comparison in Cluster 0. g, Expression distribution of each gene defining the Type-1 and Type-2 Score on the Validation Cohort UMAP in Fig. 2c. In b and f, genes defining the module are listed below. Scatter plot shows mean ± s.e.m. from n = 48 (b) and n = 46 (f) patients or healthy donors. Significance levels were calculated with two-tailed Mann-Whitney test (b, f).

Extended Data Fig. 4. Transcriptomic clustering analysis of CD4+ or CD8+ activated CAR T cells.

a, b, UMAP clustering of CD4+ (a) or CD8+ (b) CAR-specific stimulated CAR T cells from the Discovery Cohort patients and healthy donors, along with the expression distribution of Type-2 Score and the comparison of cell proportions in the type 2 cell enriched cluster. c, d, UMAP clustering of CD4+ (c) or CD8+ (d) CAR-specific stimulated CAR T cells from the Validation Cohort patients and healthy donors, along with the expression distribution of Type-2 Score and the comparison of cell proportions in the type 2 cell enriched cluster. e, Dot plot showing expression profile of type 2 related genes across patient groups, segregated by CD4+ and CD8+ subtypes. The size of circle represents proportion of single cells expressing the gene, and the colour shade indicates normalized expression level. f, The molecular portrait showing the activation of type 2 pathway in CAR-specific activated BCA-L CAR T cells. Molecules are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). The node colour indicates up-(orange) or down-(blue) regulation. Nodes are displayed using various shapes that represent the functional class of the gene product. Edges are displayed with various labels that describe the nature of the relationship between the nodes. In a–d, genes defining the Type-2 Score are listed below. Scatter plot shows mean ± s.e.m. from n = 48 (a, b) and n = 46 (c, d) patients or healthy donors. Significance levels were calculated with two-tailed Mann-Whitney test (a–d).

Extended Data Fig. 5. Independent evaluation of type 2 cytokines using flow cytometry and multiplexed secretomic assay.

a, Gating strategy employed for flow cytometry data analysis. Live Dead Blue (LDB) was utilized for live cell selection, followed by CD3 + CD14/CD19 − T cell gating and intact, single-cell filtering. CAR (FMC63) expression was employed for the selection of successfully transduced CAR+ cells, with subsequent analysis of CD4+ and CD8+ subpopulations conducted separately. b, Frequency comparison of the type-2-cytokine+ population in CD8 + CAR+ cells between persistence groups in the Discovery Cohort. c, Frequency comparison of the type-2-cytokine+ population in CD4+ or CD8 + CAR+ cells between persistence groups in the Validation Cohort. d, Schematic principle of multiplexed secretomic assay to measure functional cytokines. The diagram was created using BioRender. Scatter plot shows mean ± s.e.m. from n = 42 (b) and n = 40 (c) patients. Significance levels were calculated with two-tailed Mann-Whitney test (b, c).

Extended Data Fig. 6. Single-cell ATAC analysis of type 1 or type 2 markers of activated CAR T cells.

a–d, Pseudo-bulk chromatin accessibility tracks in the genomic region of type 2 marker genes (a), type 2 master regulator (b), type 1 marker gene (c), or type 1 master regulators (d), depicted separately for each patient. The enhancer elements predicted by ENCODE within the region of each gene are highlighted in a light-yellow shade. e, Motif footprinting trace showing transcription factor binding dynamics of STAT1 or TBX21 in the two patient groups, alongside their respective position weight matrices.

Extended Data Fig. 7. Ligand-receptor regulatory analysis of type 2 CAR T cells.

a, Correlation between Cytotoxic Score and Coinhibitory Score of all the Cluster 2 cells in Fig. 2a. The regression line is indicated, with the 95% confidence area shown in shaded colour. Spearman correlation coefficient and the associated p-value are shown. Genes defining each module are listed. b, Single-cell expression comparisons of Coinhibitory Score and Coinhibitory ADT Score in Cluster 2 cells between persistence groups in the Discovery Cohort. c, Identification of ligand-receptor (L-R) interactions originating from type 2 enriched Cluster 7 cells in Fig. 2c, predominantly interacting with Cluster 1 cells through type 2 L-R pairs. The thickness of edges is proportional to correlation weights, and edge colour corresponds to the Cluster ID. d, Identification of L-R interactions targeting Cluster 1 cells in Fig. 2c, revealing that cells from the majority of other clusters predominantly regulate these cells through type 2 L-R pairs. e, Differentially expressed genes (DEGs) specific to Cluster 1 in comparison to all other clusters in Fig. 2c, along with the expression distribution of the Cytotoxic Score. Genes defining this module include GZMA, GZMB, GZMH, GNLY, PRF1, and NKG7. f, Corresponding signalling pathways regulated by the DEGs identified in Cluster 1. Pathway terms are ranked by –log 10 (p-value). z score is computed and used to reflect the predicted activation level (z > 0, activated/upregulated; z < 0, inhibited/downregulated; z ≥ 2 or z ≤ −2 can be considered significant). g, Heatmap showing the average expression levels of coinhibitory-related genes or ADT proteins across all single cells within each identified cluster in Fig. 2c. The expression of TIM-3 and its encoding gene HAVCR2 is observed to be high in Cluster 1. h, Correlation between Cytotoxic Score and Coinhibitory Score of all the Cluster 1 cells in Fig. 2c. The regression line is indicated, with the 95% confidence area shown in shaded colour. Spearman correlation coefficient and the associated p-value are shown. Genes defining each module are listed in (a). i, Comparison of cell proportion in Cluster 1 between persistence groups in the Validation Cohort. j, Single-cell expression comparisons of Coinhibitory Score and Coinhibitory ADT Score in Cluster 1 cells between persistence groups in the Validation Cohort. Violin plot shows expression distribution of the top 10% single cells from each individual among n = 48 (b) and n = 46 (j) patients or healthy donors. Scatter plot shows mean ± s.e.m. from n = 46 (i) patients or healthy donors. Significance levels were calculated with two-tailed Spearman’s rank correlation test (a, h), two-tailed Mann-Whitney test (b, e, i, and j), or right-tailed Fisher’s Exact Test (f).

Extended Data Fig. 8. Clustering analysis of donor CAR T cell population response to type 2 cytokines.

a, Schematic outlining the experimental design for optimizing IL-4 concentration using donor CAR T cells. Three donors (ND317, ND365, ND502) were used for this study. The diagram was created using BioRender. b, UMAP clustering of CAR T cells from healthy donors, comparing conditions with and without varying concentrations of IL-4 added during in vitro CAR-specific activation. Characteristic clusters enriched for high proliferative (Cluster 1), type 1 (Cluster 2), type 2 (Cluster 7), and dysfunctional cytotoxic (Cluster 4) CAR T are indicated. c, Expression distribution of Proliferation Score, Type-1 Score, Type-2 Score, and Cytotoxic Score on the UMAP in (b). d, Corresponding signalling pathways regulated by the DEGs identified in Cluster 4 in (b). Functional pathways are downregulated in this cluster, whereas T cell exhaustion and death receptor signalling are activated. Pathway terms are ranked by –log 10 (p-value). z score is computed and used to reflect the predicted activation level (z > 0, activated/upregulated; z < 0, inhibited/downregulated; z ≥ 2 or z ≤ −2 can be considered significant). e, Comparison of cell proportions in Clusters 1, 2, 4, and 7 identified in (b) between conditions. “Original” denotes no IL-4 added during CAR-specific activation. f, UMAP clustering of CAR T cells from healthy donors, with and without 10 ng/mL of IL-5 added during in vitro CAR-specific activation. Characteristic clusters enriched for high proliferative (Cluster 1), type 1 (Cluster 3), type 2 (Cluster 8), and dysfunctional cytotoxic (Cluster 4) CAR T are indicated. g, Expression distribution of Proliferation Score, Type-1 Score, Type-2 Score, and Cytotoxic Score on the UMAP in (f). h, Comparison of cell proportions in Clusters 1, 3, 4, and 8 identified in (f) between conditions. “Original” denotes no IL-5 added during CAR-specific activation. i, UMAP clustering of CAR T cells from healthy donors, with and without 10 ng/mL of IL-13 added during in vitro CAR-specific activation. Characteristic clusters enriched for high proliferative (Cluster 1), type 1 (Cluster 5), type 2 (Cluster 7), and dysfunctional cytotoxic (Cluster 4) CAR T are indicated. j, Expression distribution of Proliferation Score, Type-1 Score, Type-2 Score, and Cytotoxic Score on the UMAP in (i). k, Comparison of cell proportions in Clusters 1, 4, 5, and 7 identified in (i) between conditions. “Original” denotes no IL-13 added during CAR-specific activation. In c, g, and j, Genes defining each module are listed below. Scatter plot shows mean ± s.e.m. from n = 3 (e, h, k) healthy donors. Significance levels were calculated with right-tailed Fisher’s Exact Test (d), or two-tailed Wilcoxon matched-pairs signed rank test (e, h, and k).

Extended Data Fig. 9. Clustering analysis of patient CAR T cells activated with IL-4 supplementation.

a, Expression distribution of Proliferation Score, Type-1 Score, Type-2 Score, and Cytotoxic Score on the UMAP in Fig. 3j. Genes defining each module are listed below. b, Differentially expressed genes (DEGs) specific to Cluster 4 in comparison to all other clusters in Fig. 3j. c, Corresponding signalling pathways regulated by the DEGs identified in Cluster 4 in Fig. 3j. Functional pathways are downregulated in this cluster, whereas T cell exhaustion and death receptor signalling are activated. d, Dot plot showing expression profile of type 2 receptor genes across all clusters identified in Fig. 3j, with notable high expression observed in Clusters 1 and 4. The size of circle represents proportion of single cells expressing the gene, and the colour shade indicates normalized expression level. e, Comparison of signalling pathways in BCA2 patient CAR T cells supplemented with 10 ng/mL IL-4 relative to the original condition, with or without the exclusion of dysfunctional cytotoxic Cluster 4 cells in Fig. 3j. A statistical comparison of the activation z score for functional signalling pathways (n = 21), including those regulating metabolism, immune function, and proliferation, was conducted. In c and e, pathway terms are ranked by –log 10 (p-value). z score is computed and used to reflect the predicted activation level (z > 0, activated/upregulated; z < 0, inhibited/downregulated; z ≥ 2 or z ≤ −2 can be considered significant). Significance levels were calculated with two-tailed Mann-Whitney test (b), right-tailed Fisher’s Exact Test (c), or two-tailed Wilcoxon matched-pairs signed rank test (e).

Extended Data Fig. 10. Single-cell ATAC analysis of patient CAR T cell response to IL-4.

a, Volcano plot showing differential accessible peak activities in BCA2 or 1 CAR T cells with and without the addition of 10 ng/mL IL-4. b, Pseudo-bulk chromatin accessibility tracks in the genomic region of top-ranked genes upregulated in 10 ng/mL IL-4 treated BCA2 or 1 CAR T cells compared to the original condition, including type 1 marker gene, type 2 marker gene, cytotoxicity marker genes, chemokine marker genes or activation/proliferation marker genes, depicted separately for each patient with and without the addition of IL-4. c, Pseudo-bulk chromatin accessibility tracks in the genomic region of CSF2 and Il32, depicted separately for each patient with and without the addition of IL-4. d, Expression profile of type 2 motif MA0037.3 (GATA3) and type 1 motif MA0690.1 (STAT1) across each patient with and without the addition of IL-4. In b and c, the enhancer elements predicted by ENCODE within the region of each gene are highlighted in a light-grey shade. Significance levels were calculated with two-tailed Mann-Whitney test (a).

Extended Data Fig. 11. Longitudinal evaluation of cytokine profile in post-treatment sera.

a, Longitudinal levels of type 2 cytokines in patients from the Validation Cohort. Three patients (with Patient ID) in the BCA-O group are individually labelled with colour lines. b, c, Longitudinal levels of type 1 cytokines and representative chemokines in patients from the Discovery Cohort (b) and the Validation Cohort (c). Seven patients (with Patient ID) in the BCA-L or BCA-O group are individually labelled with colour lines. d, Comparison of the average type 1 cytokine levels between persistent groups in the Discovery Cohort. The value is the averaged expression of IFN-γ, IL-2, and TNF-α. e, Comparison of the average type 2 cytokine levels at multiple time points between persistent groups in the Validation Cohort. The value represents the average of the normalized expression levels of IL-4, IL-5, and IL-13. f, Comparison of the average type 1 cytokine levels between persistent groups in the Validation Cohort. The value represents the average of the normalized expression levels of IFN-γ, IL-2, and TNF-α. Scatter plot shows mean ± s.e.m. from n = 45 (Baseline), n = 54 (Day 1–5), n =38 (Day 6–8), n = 35 (Day 9–11), n = 36 (Day 12–15), n = 32 (Day 16–19), n = 32 (Day 20–23), n = 54 (Day 25–35), n = 18 (Day 36–63) (d,) and n = 8 (e, f) measurements. Significance levels were calculated with two-tailed Mann-Whitney test (d–f).

Extended Data Fig. 12. Selecting type 2^low or type 2^high CAR T for in vivo leukaemia model study.

a, UMAP clustering of CAR-specific stimulated CAR T cells from six healthy donors, along with the expression distribution of surface proteins ADT-CD4 and ADT-CD8. b, Expression distribution of Type-2 Score on the UMAP in (a), with the gene module found to be enriched in Cluster 6. c, Comparison of cell proportions in each identified cluster among different donors. ND463 is selected as type 2^high CAR T, and ND585 is selected as type 2^low CAR T based on the proportion in Cluster 6. d, Single-cell expression comparison of Type-2 Score between type 2^low and type 2^high CAR T cells. e, Expression distribution of Type-1 Score on the UMAP in (a), and single-cell expression comparison between type 2^low and type 2^high CAR T cells. f, Flow cytometry analysis of CAR, CD4, and CD8 expression in type 2^low and type 2^high CAR T cells before infusion into mice. g, h, Comparison of memory marker expression (g), and coinhibitory marker expression (h) in CAR+ pre-infusion cells between type 2^low and type 2^high CAR T cells. In b and e, genes defining each module are listed below. Significance levels were calculated with two-tailed Mann-Whitney test (d, e).

Extended Data Fig. 13. In vivo leukaemia model study using type 2^low or type 2^high CAR T cells.

a, Body weight of mice since CAR T cell infusion. b, Tumour burden measured by bioluminescence at indicated days since CAR T cell infusion. c–e, Flow cytometry analysis of memory markers (c), coinhibitory markers (d), and type 1 functionality markers (e) of peripheral CAR+ cells at day 8 post-CAR T cell infusion. f, g, Flow cytometry analysis of memory markers (f) and coinhibitory markers (g) of peripheral CAR+ cells at day 12 and day 16 post-CAR T cell infusion. h, Schematic of an in vitro repeat stimulation assay using type 2^high CAR T products, with type 2 cells sorted out based on the surface markers CCR3 and CCR4. The diagram was created using BioRender. i, Evaluation of tumour cell lysis efficacy and CAR T cell count at the endpoint under different conditions. j, Flow cytometry analysis of coinhibitory, type 1 functionality, and memory markers of CAR T cells at the endpoint. Scatter plot shows mean ± s.e.m. from n = 5 mice for each group (a, c–g), or n = 3 technical replicates for each condition (i, j). Significance levels were calculated with two-tailed unpaired Student’s t-test (c–g), or one-way ANOVA with Tukey’s multiple comparisons test (i, j).

Extended Data Fig. 14. Functional evaluation of type 2 enhanced CAR T products.

a, Evaluation of the CAR T expansion, viability, and size of the newly designed “ET2-L/H CAR T” during the manufacturing process. b, Flow cytometry analysis of the CAR transduction efficiency and CD4/CD8 ratio of different conditions. c, Schematic of in vivo leukaemia model treated with type 2^low, Primed, ET2-L, or ET2-H CAR T cells. NSG mice were intravenously (i.v.) injected with 1 × 10⁶ Nalm6 cells. Seven days later, mice were randomly assigned to five groups and were infused i.v. with 2 × 10⁶ CAR T cells or PBS (control). The survival mice were rechallenged with 1 × 10⁶ Nalm6 cells 17 days post CAR T infusion, followed by another 1 × 10⁶ Nalm6 cells rechallenge 42 days post infusion. The diagram was created using BioRender. d, Tumour burden (total flux) quantified by photons/s in mice since CAR T treatment. e–g. Flow cytometry analysis of coinhibitory (e), type 1 functionality (f), and type 2 functionality markers (g) of peripheral CAR+ cells at day 8 post-CAR T cell infusion. h, Evaluation of tumour cell lysis efficacy and CAR T cell count in an in vitro repeat stimulation assay using patient-derived CAR T cells (DC80) with a BCA duration of 3 months, with or without 10 ng/mL IL-4 priming for 12 h. Nalm6 cells were introduced daily over a 9-day span at different effector-to-target (E/T) ratios as depicted in the schematic. Flow cytometry analysis was conducted on day 4 and day 9. Significance levels on specific days are denoted due to space constraints. The diagram was created using BioRender. i, j, Flow cytometry analysis of memory, type 1 functionality/cytotoxicity, type 2 functionality, proliferation, and coinhibitory markers of CAR T cells at day 4 (i) or day 9 (j) in the in vitro repeat stimulation assay. Scatter plot shows mean ± s.e.m. from n = 5 mice for each group (d–g), or n = 4 technical replicates for each condition (h–j). Significance levels were calculated with one-way ANOVA with Tukey’s multiple comparisons test (e–g), or two-tailed unpaired Student’s t-test (h–j).