CAR-T cells with the CD38^-CD73^-Tim-3^-HLA-DR⁺ phenotype predict the efficacy of tisagenlecleucel as a treatment for B cell precursor ALL

Summary

Anti-CD19 chimeric antigen receptor T cell (CAR-T) therapy is highly effective for B cell precursor acute lymphoblastic leukemia (BCP-ALL); however, approximately half of the patients relapse. Thus, there is an urgent need to identify factors that improve efficacy. This study enrolls 19 patients with BCP-ALL (16 children and 3 young adults) who receive tisagenlecleucel. Infusion products, peripheral blood, and bone marrow samples are obtained before and after CAR-T cell infusion. Single-cell analysis reveals that central memory CAR^pos T cells increase in long-term responders, whereas CXCR3⁺CD38^highPD-1^high effector CAR^pos T cells are enriched in relapsed patients, post-infusion. By contrast, CAR^pos T cells obtained from infusion products in long-term responders are enriched in the CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ phenotype, characterized by a decreased ability to produce adenosine, memory-like transcriptomic characteristics, and leveraging of mitochondrial metabolism and oxidative phosphorylation. Our study reveals that the CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ phenotype contributes to long-term remission in patients with BCP-ALL who receive tisagenlecleucel.

Graphical abstract

Highlights

• •The CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ (4MD) phenotype is important for the efficacy of tisa-cel
• •Adenosine production via the CD38−CD73 axis is reduced in the 4MD CAR-T cells
• •The 4MD CAR-T cells exhibit memory T cell features with high oxidative phosphorylation

Mikami et al. perform serial single-cell analysis of CAR-T cells in patients with BCP-ALL who receive tisagenlecleucel. The abundance of CAR-T cells with a specific phenotype correlates with long-term treatment response. These cells have metabolic characteristics indicating low adenosine production and gene expression similar to that of memory T cells.

Introduction

The survival rate of patients with acute lymphoblastic leukemia (ALL) is >90%^1,2,3; however, about 15% of patients experience relapse, the outcome of which remains unsatisfactory.^4,5,6 Anti-CD19 chimeric antigen receptor T cell (CAR-T) therapy has shown remarkable therapeutic effects, providing hope of complete recovery from relapsed/refractory B cell precursor (BCP) ALL. Clinical trials demonstrate that approximately 80% of patients with BCP-ALL achieve complete remission (CR) after anti-CD19 CAR-T therapy.^7,8,9 Despite this, some patients eventually suffer relapse; indeed, the 5-year relapse-free survival rate of patients who achieve CR is 49%.^10,11 A recent study suggested risk stratification with monitoring to predict relapse and perform consolidative hematopoietic stem cell transplantation before it occurs¹²; however, the pathophysiology responsible for failure of CAR-T therapy is not clear.

Studies have begun to decipher the efficacy-associated signatures of anti-CD19 CAR-T cells. Treatment failure caused by CAR-T cell factors is assumed to be multifactorial, e.g., product heterogeneity, lack of expansion, inadequate effector differentiation, exhaustion, and immune escape.¹³ Single-cell transcriptomics revealed that expansion of proliferative memory-like CD8 clones was a hallmark of a good response to tisagenlecleucel (tisa-cel), while an increase in CAR-T regulatory cells (CAR-Tregs) was detected among patients who did not respond to axicabtagene ciloleucel.¹⁴ Chronic interferon (IFN) signaling regulated by IRF7, and the presence of inhibitory exhaustion-associated markers such as LAG-3, Tim-3, and TIGIT, are associated with poor persistence of CAR-T cells, whereas TCF7 expression contributes to maintenance of a naive T cell state.^15,16,17 Nevertheless, it is difficult to apply them to clinical practice; therefore, identifying characteristics that improve the quality of CAR-T cells remains a pressing issue.

Here, we conducted an analysis of 19 patients with BCP-ALL receiving tisa-cel. We analyzed infusion products (IPs), peripheral blood (PB), and bone marrow (BM) samples before and after anti-CD19 CAR-T cell infusion. Post-infusion CAR^pos T cells collected from PB/BM samples exhibited a more central memory phenotype in long-term responders, in contrast to the activated effector memory phenotype (CXCR3⁺CD38^highPD-1^high) observed in relapsed patients. Notably, we found that in long-term responders, CAR^pos T cells obtained from IPs were characterized by a CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ phenotype with a reduced ability to produce adenosine, a potent immunosuppressant; this deficit in adenosine production occurred via the CD38-CD73 axis. Additionally, we found that these CAR^pos T cells in IPs were characterized by the gene expression pattern of upregulated oxidative phosphorylation (OXPHOS) and mitochondrial metabolism, similar to memory T cells. These results provide insights that may enable construction of CAR-T products with lasting efficacy.

Results

Clinical course and prognostic factors of patients with BCP-ALL who received tisa-cel therapy

To identify determinants of effective anti-CD19 CAR-T therapy, we enrolled 16 children and three young adults with relapsed/refractory BCP-ALL, all of whom received tisa-cel between 2019 and 2023 at Kyoto University Hospital. The patient characteristics are shown in Table S1. In this study, CR was defined as less than 5% blasts in BM as assessed by both microscopy and flow cytometry. In addition, in patients with specific genetic abnormalities for whom molecular monitoring was feasible, achievement of molecular remission (minimal residual disease [MRD] negativity) was also confirmed.

The leukemia blast % in the BM before CAR-T cell infusion varied from 0% to 97.6%; however, 18 out of 19 patients (94.7%) achieved CR at 1 month after CAR-T cell infusion, confirming its high efficacy as an induction therapy for remission (Figure 1A). This finding is consistent with that reported previously.¹⁸ When patients were divided into two groups according to their long-term response to CAR-T therapy, we found that nine patients (hereafter referred to as patients in remission) maintained CR for more than 1 year after CAR-T cell infusion, and 10 experienced relapse (relapsed patients) within 1 year (Figure 1B). In our cohort, no patients had low-level MRD recurrence while maintaining morphological remission. Event-free survival and overall survival of the two groups are shown in Figures S1A and S1B. Multivariate analysis revealed that ≥5% BM blasts before CAR-T therapy was a predictor of relapse, a finding mentioned as a risk in previous reports (Figures S1C, S1D, and S1E).^10,19,20,21 In terms of absolute lymphocyte counts and levels of C-reactive protein (CRP), which are indicators of an inflammatory response, both were elevated within 20 days post-infusion and were associated with tumor burden but not with prognosis (Figures 1C and 1D).

Figure 1.

Clinical information and monitoring of CAR transgene DNA from patients with BCP-ALL who received a tisa-cel infusion

(A) Blast ratio (%) before and 1 month after chimeric antigen receptor T cell (CAR-T) therapy. Blast ratios were assessed primarily from the bone marrow (BM), but in cases where BM aspiration was not performed, blast ratios were assessed in the peripheral blood (PB). CR, complete response.

(B) Clinical course of 19 patients with B cell precursor acute lymphoblastic leukemia (BCP-ALL) who received a tisagenlecleucel (tisa-cel) infusion. Remission was defined as the absence of relapse for at least 1 year.

(C) Absolute lymphocyte counts after CAR-T cell infusion in patients with BCP-ALL. In the line graph, the line for each patient is color-coded according to prognosis and BM blast ratio. The boxplot below compares the maximum lymphocyte counts within 1 month after CAR-T cell infusion for patients in remission and relapsed patients with a BM blast ratio ≥5% and <5%. ∗∗p < 0.01; ns: not significant by the Mann-Whitney test.

(D) Changes in C-reactive protein (CRP) values within 1 month from the CAR-T cell infusion. The boxplot below compares the maximum CRP values within 1 month after CAR-T cell infusion for patients in remission and relapsed patients with a BM blast ratio ≥5% and <5%. ∗∗p < 0.01; ns: not significant by the Mann-Whitney test.

(E) Schematic showing the sample collection points.

(F) Long-term monitoring of tisa-cel transgene DNA using droplet digital PCR (ddPCR). The cyclin-dependent kinase inhibitor 1A (CDKN1A) copy number was used as a reference to correct for variability in sample DNA input and quality. The lower limit of quantification is estimated as 50–100 transgene copies/μg human genomic DNA. The graph below compares the amount of tisa-cel transgene DNA between patients in remission and relapsed patients. Data are presented as median ±95% CI. A mixed-effects model with Sidak’s multiple comparisons test was used for statistical analysis.

To serially analyze the effects of anti-CD19 CAR-T therapy, we collected IPs and PB/BM samples before and after CAR-T cell infusion, as schematically shown in Figure 1E. Samples for mass cytometry (CyTOF) analysis were collected from pre-infusion to 1 month post-infusion, and samples for droplet digital PCR (ddPCR) were collected from 2 weeks to 1 year post-infusion.

We examined the persistence of CAR-T cells by monitoring tisa-cel transgene DNA levels using ddPCR (Figures 1F and S1F).²² As predicted, patients in remission showed continuous retention of tisa-cel transgene DNA; however, transgene DNA was also detected in all relapsed patients even at the time of CD19-positive relapse, and its levels did not differ between patients in remission and relapsed patients (Figure 1F, below). This result suggests that CAR-T cells have insufficient antitumor activity in relapsed patients.

Changes in T cell subtype and differentiation induced by anti-CD19 CAR-T therapy

A single-cell multiparameter analysis of 42 markers was performed using CyTOF; markers included FMC63 (i.e., the single-chain variable fragment of anti-CD19 CAR) (Tables S2 and S3). To better understand the overall changes in T cells before and after CAR-T cell infusion, data were obtained from the total T cell population in all 117 samples collected and visualized by Uniform Manifold Approximation and Projection (UMAP).²³ Sample type (IPs or PB/BM), major T cell subtypes, differentiation stage, and expression of CAR are shown in Figure 2A. T cell differentiation stage was defined as follows: naive = CD45RA⁺CCR7⁺; central memory (CM) = CD45RA⁻CCR7⁺; effector memory (EM) = CD45RA⁻CCR7⁻; and terminal effector (TE) = CD45RA⁺CCR7⁻.²⁴ Interestingly, T cells in IPs were different from those in PB/BM samples (Figure 2A, left), and they comprised mostly CM and EM (Figures 2A, center right and S2A), which may be due to artificial stimulation during the CAR-T manufacturing process. We then classified T cells as CAR^pos and CAR^neg according to their CAR expression status. When we compared T cell subtypes between CAR^pos and CAR^neg T cells in IPs (Figures 2A, right and S2B) and BM/PB samples (Figures 2A, right and S2C), the distribution of effector T cell subtypes (CM, EM, and TE) was not markedly different, although there were some statistically significant differences. On the other hand, in the analysis of BM/PB samples, the percentages of CD4⁺ and CD8⁺ naive subsets were smaller among CAR^pos T cells than among CAR^neg T cells, which suggests that CAR^pos T cells are more likely to become effector T cells by specifically reacting with CD19⁺ leukemia cells (Figure S2C). From a chronological point of view, T cells in patients showed a marked change from naive to EM and TE after CAR-T cell infusion in both PB and BM samples (Figures 2B, 2C, and 2D). The TE subtype increased in the CD8⁺ T cell population to a greater extent than in the CD4⁺ T cell population. There was no significant change in the percentage of Tregs (CD4⁺CD25⁺CD127⁻) before and after CAR-T cell infusion (Figures 2C and 2D).

Figure 2.

Overview of T cell subtypes and differentiation stages before and after anti-CD19 CAR-T therapy

(A) Sample types (infusion products or peripheral blood [PB]/bone marrow [BM] samples), major subtypes, differentiation stage, and expression of the chimeric antigen receptor (CAR) by T cells in all collected samples. CM, central memory; EM, effector memory; TE, terminal effector.

(B) Transition of the T cell population is illustrated by uniform manifold approximation and projection (UMAP) density plots.

(C and D) Changes in T cell differentiation stage during chimeric antigen receptor T cell (CAR-T) therapy: PB in (C); BM in (D). Adjusted p values were calculated by ANOVA with Turkey’s multiple comparison in (C) and with the Holm-Sidak method in (D). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; ns: not significant; LD, lymphodepletion.

Because the pattern of T cell marker expression in PB/BM samples was so different from that in IPs (Figure 2A), we decided to analyze T cells in PB/BM samples and IPs separately.

Post-infusion CAR^pos T cells show more of a CM phenotype in patients in remission but an activated EM phenotype in relapsed patients

We next addressed whether there was a difference in T cell subtypes in PB/BM samples from patients in remission and relapsed patients. Downstream analysis was conducted using the CATALYST R package.^25,26,27 First, clustering based on the FlowSOM algorithm using self-organizing map and minimal spanning trees identified 13 T cell subsets in PB/BM samples (Figure 3A). The optimal cluster number was determined by a delta plot showing the relative change under the cumulative distribution function (CDF) (Figure S3A). Figure 3B shows the distribution of these clusters using UMAP. As shown in Figure 2A, CAR expression in PB and BM T cells was heterogeneous and present mainly in clusters CD4⁺ EM_a and CD8⁺ EM_a (Figures 3B and 3C).

Figure 3.

Serial analysis of T cell subsets in PB/BM samples obtained before and after CAR-T cell infusion

(A) FlowSOM-based clustering identified 13 T cell subsets in peripheral blood (PB)/bone marrow (BM) samples. SCM, stem cell memory; CM, central memory; EM, effector memory; TE, terminal effector.

(B) T cell subsets in PB/BM samples were projected in uniform manifold approximation and projection (UMAP).

(C) Intensity of chimeric antigen receptor (CAR) expression by T cells in PB/BM samples.

(D) Percentage of T cell subsets in PB/BM samples after CAR-T cell infusion (CAR^pos fraction). Statistical tests were performed using the diffcyt-differential abundance-generalized linear mixed model (diffcyt-DA-GLMM). ∗FDR <0.1, ∗∗FDR <0.01; ns: not significant.

With respect to CAR^pos T cells, the most marked difference between patients in remission and relapsed patients was observed in PB samples at 1 month, in which the percentages of CD4⁺ and CD8⁺ CM were significantly higher in patients in remission and the percentages of CD4⁺ EM_a and CD8⁺ EM_a were significantly higher in relapsed patients (Figure 3D). Although not all subsets reached statistical significance, similar trends were observed in the BM (Figure 3D). Notably, T cells in CD4⁺ EM_a and CD8⁺ EM_a had a CXCR3⁺CD38^highPD-1^high phenotype (Figure 3A), suggesting overactivation.^28,29,30 These findings suggest that the balance between CM T cells and activated EM T cells (i.e., a higher proportion of CM T cells and a lower proportion of activated EM T cells) could play a role in long-term treatment responses.

By contrast, we found no clear difference in the immunophenotype of CAR^neg T cells, either before or after CAR-T cell infusion, between the remission and relapse groups (Figure S3B).

IPs from patients in remission contain more CAR^pos T cells with the CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ phenotype

The same clustering approach was used for IPs (i.e., CAR-T cells in the product before administration). FlowSOM-based clustering identified 11 T cell subsets within IPs, based on a delta plot showing the relative change under the CDF (Figure S4A). The CD4⁺ and CD8⁺ clusters were classified as either CM or EM (Figure 4A). The distribution of the T cell subsets is shown in Figure 4B. In contrast to T cells in the PB/BM samples, every T cell subset in IPs expressed CAR (Figure 4C). Overall, the percentage of CAR^pos T cells in IPs was approximately 20%, similar to that previously reported (i.e., 27.3%) by flow cytometry analysis,³¹ and there was no significant difference between patients in remission and relapsed patients (Figure 4D).

Figure 4.

T cell subsets in IPs and their relevance to the BM blast ratio before CAR-T therapy

(A) FlowSOM-based clustering identified 11 T cell subsets in infusion products (IPs). CM, central memory; EM, effector memory.

(B) T cell subsets in IPs, projected in uniform manifold approximation and projection (UMAP).

(C) Intensity of chimeric antigen receptor (CAR) expression by T cells in IPs.

(D) Percentage of CAR^pos T cells in IPs from patients in remission and relapsed patients (Mann-Whitney test). ns: not significant.

(E) Percentage of each T cell subset in IPs (CAR^pos and CAR^neg fractions). Statistical tests were performed using the diffcyt-differential abundance-generalized linear mixed model (diffcyt-DA-GLMM). ∗FDR<0.1; ns: not significant.

(F) Scatterplot of the BM blast ratio and the summed percentage of the three subsets within the CAR^pos T cell population in IPs (CD4⁺ EM_1; CD4⁺ EM_4; and CD8⁺ EM_1). A fitted curve (red) was drawn using nonlinear regression analysis (F test).

(G) The summed percentage of the three subsets was compared in terms of the BM blast ratio: ≥5% or <5% (Mann-Whitney test).

(H) Receiver operating characteristic (ROC) curves predicting disease recurrence after CAR-T therapy; based on the BM blast ratio and the sum of the three subsets. AUC, area under the ROC curve.

Differential abundance analysis was performed for both CAR^pos T cells and CAR^neg T cells in IPs (Figure 4E). Overall, the abundance of each cluster within the CAR^pos and CAR^neg T cell populations was similar. Remarkably, there was a significant difference in the proportion of the three subsets (CD4⁺ EM_1, CD4⁺ EM_4, and CD8⁺ EM_1) within the CAR^pos T cell population. Each of the three subsets was significantly more prominent in patients in remission. When we summed the percentages of these three subsets, the total percentage was high only in patients who had low BM blast ratios before CAR-T therapy, and there was an inverse relationship between the BM blast ratio and the percentage of the three subsets (Figure 4F). Interestingly, with one exception, even in cases with very low blast ratios, those with a lower percentage of these subsets experienced relapse. If we set the cutoff for the BM blast ratio to 5% (which is suggested to predict a response to anti-CD19 CAR-T therapy in a clinical setting),¹² the total percentage was clearly different (Figure 4G). Additionally, the accuracy of the total percentage of the three subsets to predict treatment outcome was comparable with that of the BM blast ratio (Figure 4H).

Next, we defined CD4⁺ EM_1, CD4⁺ EM_4, and CD8⁺ EM_1 collectively as remission-enriched clusters and searched for common factors within these subsets. The expression histogram highlighted differences in marker expression between remission-enriched clusters and others; the expression level of HLA-DR was significantly increased, while those of CD38, CD73, and Tim-3 were significantly decreased in remission-enriched clusters (Figures 5A and 5B). We also compared markers based on patient outcomes (remission and relapse) without clustering T cells (Figure S4B). However, there were no significantly expressed markers, which indicates the importance of evaluating multiple T cell markers in combination (as T cell subsets), rather than assessing them individually. Then, we compared the UMAP distribution of remission-enriched clusters (Figure 5C) with the distribution of marker expression (Figures 5D and S4C). The expression patterns of CD38, CD73, Tim-3, and HLA-DR closely mirrored the distribution of remission-enriched clusters, supporting their utility as markers for cluster identification. Therefore, we applied a conventional gating method to the CAR^pos T cells in IPs and identified a CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ population (hereafter referred to as 4-marker-defined [4MD]) (Figure 5E). As expected, the percentage of 4MD T cells correlated strongly with that of remission-enriched clusters (Figure 5F) and was increased in patients in remission (Figure 5G). In addition, the percentage of 4MD T cells within the CAR^pos T cell populations was a good predictor of the outcome of CAR-T therapy (Figure 5H). It should be noted that the percentage of 4MD T cells in the CAR^pos T cell and CAR^neg T cell populations was similar in each IP, suggesting that transfer of the CAR gene did not impact T cell composition during the manufacturing process (Figure 5I). Furthermore, we performed multivariate analysis of identified and known risk factors that contribute to relapse (Figure S4D). The cutoff value for determining whether each IP had a 4MD high or low phenotype was calculated to be 13% based on the Youden index in the receiver operating characteristic (ROC) curve for CAR^pos T cell populations (Figure 5H). Although attention should be paid to the relatively small number of cases in the cohort, the 4MD phenotype was identified as a prognostic factor independent of the BM blast ratio. Taken together, the data suggest that the abundance of EM T cells with a 4MD phenotype in IPs is a key indicator of a durable response to tisa-cel.

Figure 5.

Detailed phenotypic and functional analysis of the T cell subsets enriched within IPs in patients in remission

(A) Histogram showing marker expression (compared with that of other clusters) by the three subsets (CD4⁺ EM_1 + CD4⁺ EM_4 + CD8⁺ EM_1) enriched in patients in remission.

(B) Volcano plot showing the significance in the difference in marker expression between remission-enriched clusters and other clusters. The threshold of the significance is set at an adjusted p value <0.05 (red dashed line; Wilcoxon matched-pairs signed rank test with the Holm-Sidak method for multiple comparisons).

(C) Remission-enriched clusters are projected in uniform manifold approximation and projection (UMAP).

(D) The expression patterns of CD38, CD73, Tim-3, and HLA-DR, projected in UMAP.

(E) An example of the gating scheme used to define CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ (four marker-defined: 4MD) T cells.

(F) Correlation between the percentage of remission-enriched clusters and that of 4MD T cells within the CAR^pos T cell population (Spearman’s rank correlation coefficient).

(G) Comparison of the percentage of 4MD T cells within the chimeric antigen receptor (CAR)^pos T cell population in patients in remission and relapsed patients (Mann-Whitney test).

(H) Receiver operating characteristic (ROC) curve showing the percentage of 4MD T cells within the CAR^pos T cell population required to predict disease recurrence. AUC, area under the ROC curve. The cutoff value was calculated based on the Youden index.

(I) Correlation between the percentage of 4MD T cells in the CAR^pos T cell and CAR^neg T cell populations (Spearman’s rank correlation coefficient).

(J) Gating scheme used to define the CD38⁻CD73⁻ population in infusion products (IPs).

(K) Percentage of CD38⁻CD73⁻ T cells within the CAR^pos T cell populations from patients in remission and relapsed patients (Mann-Whitney test).

(L) Correlation between the percentage of CD38⁻CD73⁻ T cells within the CAR^pos T cell and CAR^neg T cell populations in IPs (Spearman’s rank correlation coefficient).

(M) Schematic diagram illustrating production of adenosine (ADO) from nicotinamide adenine dinucleotide (NAD⁺). ADPR: adenosine diphosphate ribose; AMP: adenosine monophosphate.

(N) Correlation between the percentage of CD38⁻CD73⁻ T cells in IPs and ADO production (Spearman’s rank correlation coefficient).

(O) Correlation between the percentage of CD38⁺CD73⁺ T cells in IPs and ADO production (Spearman’s rank correlation coefficient).

(P) Difference in ADO production by T cells in IPs from patients in remission and relapsed patients (Student’s t test, n = 3 technical replicates from nine biological replicates per group).

(Q) Comparison of cytotoxicity against leukemia cells (SU/SR^GFP/Luc) between 4MD high and 4MD low IPs. CAR-T cells from IPs were co-cultured with SU/SR^GFP/Luc at various ratios for 24 h. To perform the experiment with three biological replicates, IPs from different patients were used (4MD high: remission d, remission f, and remission I; 4MD low: remission b, relapse d, and relapse f). The experiment was performed three times (n = 3 technical replicates). p value was calculated by a two-way repeated measures ANOVA with Sidak’s multiple comparisons test. Data are presented as mean ± SD.

Expression of CD38 and CD73 by CAR-T cells in IPs is associated with adenosine production

Among the four markers, CD38 and CD73 are involved in production of adenosine (ADO) from nicotinamide adenine dinucleotide (NAD⁺).³² Extracellular ADO facilitates immune evasion by cancer cells by suppressing immune cells such as T cells, natural killer (NK) cells, and macrophages, while increasing the activity of immunosuppressive cells such as Tregs and myeloid derived suppressor cells (MDSCs).³³ Consequently, ADO production in IPs may reduce the efficacy of CAR-T cells.³⁴ To explore this further, we conducted an additional investigation focusing on ADO production. Using conventional gating (Figure 5J), we identified the CD38⁻CD73⁻ population in IPs. The percentage of CD38⁻CD73⁻ T cells within the CAR^pos T cell population was significantly higher in patients in remission (Figure 5K). Additionally, there was a strong correlation between the percentage of CD38⁻CD73⁻ T cells within the CAR^pos and CAR^neg T cell populations within each IP (Figure 5L). To evaluate the capacity to generate ADO, we conducted an adenosine assay, in which T cells are exposed to NAD⁺ as a substrate. Sequential enzymatic reactions involving CD38 and CD73 generate ADO, which is measured in a fluorometric assay (Figure 5M). Therefore, the degree of CD38 and CD73 expression affects the amount of ADO. We found a negative correlation between the percentage of CD38⁻CD73⁻ T cells in IPs and the amount of ADO (Figure 5N), along with a strong positive correlation between the percentage of CD38⁺CD73⁺ T cells in IPs and the amount of ADO (Figure 5O). Furthermore, the amount of ADO produced by T cells in IPs from patients in remission was smaller than that by T cells from relapsed patients (Figure 5P). Donor chimerism in patients who underwent HSCT before CAR-T therapy was higher than 90% (Table S1), and donor-derived CAR-T cells in such patients might show a different response. Therefore, we compared the percentage of 4MD T cells in IPs and ADO production between patients who underwent prior HSCT and those who did not (Figures S4E and S4F). However, there was no significant difference.

These findings highlight that among the four T cell markers related to long-term responses to tisa-cel, CD38 and CD73 play a critical role in ADO production. The lower expression of CD38 and CD73 observed in patients in remission may contribute to reduced ADO levels in IPs, potentially promoting a sustained response to anti-CD19 CAR-T therapy.

Although the 4MD phenotype (characterized by CD38⁻CD73⁻ expression) is primarily associated with low ADO production, the other surface marker profile (Tim-3⁻HLA-DR⁺) suggests that T cells were in a non-exhausted, active state at the time of infusion. This implies that 4MD T cells more effectively contribute to early leukemia clearance. To functionally validate this, we conducted an in vitro cytotoxic T lymphocyte (CTL) assay to determine whether IPs that have a high proportion of cells with the 4MD phenotype (>13% among CAR^pos T cells) have higher tumor-killing activity. SU/SR^GFP/Luc35,36 were used as CD19⁺ BCP-ALL cells and were cultured with CAR-T cells at various ratios for 24 h. The ratio of surviving leukemia cells was calculated based on the absolute number of counting beads. As expected, 4MD high IPs showed greater cytotoxicity against CD19⁺ BCP-ALL cells than 4MD low IPs in a dose-dependent manner (Figure 5Q).

Transcriptomic characterization of 4MD T cells in IPs

Next, we focused on the characteristics of 4MD (CD38⁻CD73⁻Tim-3⁻HLA-DR⁺) T cells in IPs and sought to identify the factors that contribute to long-term CAR-T efficacy. We sorted 4MD and non-4MD T cells from IPs using a cell sorter (Table S4) and processed them for bulk RNA-sequencing (RNA-seq). Principal component analysis (PCA) showed that the 4MD and non-4MD samples had similar transcriptomic profiles, with no clear separation with respect to prognosis (i.e., remission vs. relapse) (Figure 6A). In addition, hierarchical clustering was performed, and the optimal number of clusters was calculated to be two based on the silhouette score (Figures 6B and S5). One of the two clusters consists of 4MD samples, and the other consists of non-4MD samples. This indicates that although 4MD T cells are enriched in patients in remission, their overall gene expression pattern is not significantly different between remission and relapsed cases. Next, we compared DEGs between 4MD and non-4MD T cells using the following cutoffs: adjusted p value = 0.05 and fold change = 2 (Figure 6C; Table S5). Reflecting the prior sorting scheme, we found that expression of CD38, NT5E (CD73), and HAVCR2 (Tim-3) was higher in non-4MD T cells, whereas that of HLA-DR genes (HLA-DRA, HLA-DRB1, and HLA-DRB5) was more common in 4MD T cells. Telomerase reverse transcriptase (TERT) was upregulated significantly in 4MD T cells, suggesting that they have an advantage in terms of long-term persistence.³⁷ By contrast, PIK3CG, the PI3-kinase family crucial for T cell activation,³⁸ and CBLB, the E3 ubiquitin ligase that regulates TCR signaling,³⁹ were downregulated in 4MD T cells.

Figure 6.

Transcriptomic characterization of CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ T cells in IPs

(A) Principal component analysis (PCA) plot for each patient infusion product (IP) sample. CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ (four marker-defined: 4MD) T cells and non-4MD T cells in IPs are plotted. RM, remission; RL, relapse.

(B and C) Hierarchical clustering using the Ward method (B) and a volcano plot (C) of the 4MD and non-4MD samples. The silhouette score was calculated for 2–10 clusters, and the optimal number of clusters was calculated to be two (Figure S5). Differentially expressed genes (DEGs) used for clustering met the condition of FDR-adjusted p value <0.05 (one-way ANOVA with the Benjamini-Hochberg method). VST, variance stabilizing transformation.

(D and E) Enriched terms across input gene lists for 4MD (D) and non-4MD (E) T cells were ranked according to their adjusted p values. GO, gene ontology; R-HSA, reactome-Homo sapiens; WP, WikiPathways.

(F) The results of gene set enrichment analysis (GSEA) using hallmark gene sets. NES, normalized enrichment score; FDR, false discovery rate.

(G, H, and I) GSEA (G), overrepresentative analysis (H), and gene-concept (cnet) plots (I) of reactome pathways. For GSEA, the NES and FDR q value for each gene set are as follows; mitochondrial translation: NES = 1.78 and FDR-q = 0.0041; rRNA processing: NES = 1.69 and FDR-q = 0.008; neutrophil degranulation: NES = −1.35 and FDR-q = 0.0008. Cnet plot was constructed using significantly downregulated pathways and their related genes in 4MD T cells.

(J and K) GSEA using the “adenosine metabolism and signaling” gene set and GO terms (J) and a Blue-Pink O’Gram of the leading edge subset genes related to “adenosine metabolism and signaling” (K).

Enrichment analysis using Metascape⁴⁰ revealed that 4MD T cells were enriched in “ribosome-related ontologies” and “mitochondrial” pathways, suggesting active protein synthesis and energy production, respectively, whereas non-4MD T cells were associated with “adaptive immune responses” and “inflammation” (Figures 6D and 6E). Consistent with this, hallmark gene set enrichment analysis (GSEA) showed upregulation of OXPHOS in 4MD T cells, whereas tumor necrosis factor alpha (TNF-α) signaling, interleukin (IL)-2 STAT5 signaling, and inflammatory responses were predominant in non-4MD T cells (Figure 6F). These results suggest that 4MD T cells have a more memory-T-cell-like phenotype than non-4MD T cells. Further GSEA and overrepresentative analyses of the reactome highlighted “mitochondrial translation” and “ribosomal RNA processing” in 4MD T cells, whereas “neutrophil degranulation,” “class A/1 rhodopsin like receptors,” and “inflammation-related pathways” were enriched in non-4MD T cells (Figures 6G, 6H, and 6I).

To assess whether CAR-T cells were exposed to an ADO-rich environment during IP manufacture, we compared 4MD T cells from the relapse and remission groups. Since 4MD T cells are CD38⁻CD73⁻ and do not produce much ADO by themselves, activation of ADO metabolism and associated signaling pathways would reflect exposure to exogeneous ADO.^41,42 Therefore, we created a gene set “adenosine metabolism and signaling” by combining gene ontologies related to adenosine metabolism, signaling, and immune regulation (GO:0046085, 0006382, 0032238, 0004000, 0031685, 0001609, and 0060168) and then performed GSEA using these GO terms. The analysis revealed higher enrichment of ADO metabolism and signaling in 4MD T cells from relapsed patients and identified upregulation of genes such as ADORA2A and ADA (Figures 6J and 6K). In addition, 4MD T cells in relapsed patients showed lower expression of genes related to T cell activation and proliferation than those in patients in remission (Figure 6J). These findings suggest that exposure to higher ADO levels during IP manufacture may have a negative effect on CAR-T cell efficacy in relapsed patients.

Taken together, these results demonstrate that the gene expression profile of 4MD T cells is more similar to that of memory T cells than effector T cells, suggesting their long-term efficacy for preventing BCP-ALL relapse after anti-CD19 CAR-T therapy.

Validation of the 4MD phenotype in an external cohort

The abovementioned findings suggest that a higher abundance of 4MD T cells in IPs is associated with favorable clinical outcomes. However, the limited cohort size may affect the robustness of statistical inferences (Figures S6A and S6B).

Therefore, to evaluate the generalizability of the 4MD phenotype as a predictor of clinical response, we performed validation analysis using an external cohort. Single-cell cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) data from Bai et al. 2024⁴³ were obtained from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (accession number GEO: GSE262072). Among the data, 82 samples derived from unstimulated IPs were used in subsequent analysis.

After data normalization and integration using the Seurat (v.4.4.0) pipeline,⁴⁴ clustering was performed on the RNA data using the smart local moving (SLM) algorithm. We found 21 clusters under the optimal resolution of 1.2 (Figures 7A and S7A). Expression of 4MD genes on the UMAP is presented in Figure S7B, and expression of canonical T cell markers in each cluster is shown in Figure 7B. To identify clusters with characteristics of 4MD subsets, the AddModuleScore function was used with a gene set comprising significant DEGs (adjusted p value <0.05) identified in our bulk RNA sequencing (RNA-seq) analysis (Table S5). The 4MD module score was high for some T cells in the validation IP samples (Figure 7C). Clusters with a median 4MD module score greater than 0 (higher median expression of the 4MD gene program than random genes) were annotated as 4MD clusters, which included clusters 4, 14, and 17 (Figure 7D). Cluster 4 contained a combination of CD4⁺ and CD8⁺ T cells enriched in heat shock proteins and highly expressed FABP5, whereas clusters 14 and 17 comprised CD8⁺ T cells and expressed cytotoxic genes (Figures 7B and S7C). Interestingly, clusters 4 and 17 were also enriched in gene pathways related to DNA metabolism (Figure 7E).

Figure 7.

Investigation of the 4MD phenotype in IPs and the prognosis following CAR-T therapy using public CITE-seq data

(A) Twenty-one clusters of T cells were identified by the smart local moving (SLM) algorithm with a resolution parameter of 1.2. The clusters were projected on the uniform manifold approximation and projection (UMAP) image based on RNA expression.

(B) Expression of canonical T cell marker genes in each cluster.

(C) 4MD module scores were projected on the UMAP image.

(D) Violin plot showing the 4MD module score in each cluster.

(E) Top five ontology pathways significantly enriched in each cluster. The results for clusters 4, 14, and 17 are shown. For cluster 17, only one pathway had an adjusted p value <0.05. GO, gene ontology.

(F) Proportion of 4MD clusters (clusters 4, 14, and 17) in each sample. The dashed line represents the cutoff value of 0.16.

(G) Event-free survival of patients with 4MD high or low IPs. The survival curves were plotted using the Kaplan-Meier method. The p value was calculated by the Gehan-Breslow-Wilcoxon test. Patients who received CAR-T reinfusion were excluded from this analysis.

We calculated the proportion of 4MD clusters (clusters 4, 14, and 17) in each sample (Figure 7F). To adapt it to the structure of the external dataset, the cutoff value to distinguish 4MD high and 4MD low IPs was set to 16% according to the ROC curve predicting the outcome of CAR-T therapy obtained from CyTOF analysis of total T cell populations in IPs in our inhouse cohort (Figure S7D). Patients with 4MD high IPs (n = 14) had significantly better event-free survival than those with 4MD low IPs (n = 58) in the validation cohort (Figure 7G). The results obtained with the external cohort demonstrated that IPs enriched with the 4MD phenotype were associated with a better prognosis upon CAR-T therapy.

Discussion

In this study, the characteristics of post-infusion CAR-T cells in PB and BM were consistent with those reported previously, suggesting that enrichment of memory CAR-T cells, coupled with fewer exhausted effector CAR-T cells, is important for a durable treatment response.^14,15,17 Notably, we found that post-infusion CAR-T cells isolated from relapsed patients comprised mainly EM T cells with a CXCR3⁺CD38^highPD-1^high phenotype, suggesting that these T cells were strongly activated; this means that they would become exhausted over time.

CAR-T cells in IPs had different properties in patients in remission versus relapsed patients, although there were no noticeable differences in the PB and BM before CAR-T therapy (Figure S3B). This suggests that potential differences in the long-term efficacy of T cells become visible only after introduction of the CAR into T cells, followed by subsequent expansion.

CD38, a multifunctional extracellular NAD⁺ glycohydrolase, is a classical activation marker of T cells; however, it is also recognized as a marker of exhaustion that is expressed at high levels after excessive stimulation.^30,45,46 Studies show that in hemophagocytic lymphohistiocytosis, inflammation, and immune regulatory disorders, the proportion of HLA-DR⁺CD38^high T cells reflects T cell activation and correlates with sIL-2R levels.⁴⁷ Furthermore, a CD38-related pathway for producing ADO in cooperation with CD203a and CD73 has been identified³²; ADO induces T cell dysfunction by exerting an inhibitory signal within T cells following its adenosine A2 receptor (A2AR)-mediated delivery into the cytoplasm, followed by conversion to cyclic adenosine monophosphate (AMP).⁴⁸ Tim-3 is a negative immune checkpoint molecule, which together with PD-1 serves as a determinant of immune exhaustion.²⁹ Therefore, a 4MD phenotype indicates that CAR-T cells are less exhausted and maintain their immunological efficacy.

Data from in vitro tumor-killing assays and in vivo tumor-bearing mouse models show that deletion of CD38 increases the efficacy of anti-CD19 CAR-T cells by making them resistant to exhaustion.⁴⁹ In addition, inhibiting CD38 enzyme activity boosts CAR-T cell cytotoxicity by downregulating CD38-cADPR-Ca²⁺ signaling and activating the CD38-NAD⁺-SIRT1 axis to suppress glycolysis.⁵⁰ Focusing on ADO produced via the CD39–CD73 axis, Klys et al. targeted CD39 and CD73 to prevent ADO-mediated immunosuppression; however, inhibiting CD39 and CD73 did not reduce the immunosuppressive capacity, whereas they did show that inosine, a metabolite of ADO, augments CAR-T cell function and stemness through metabolic reprogramming and demonstrated that CAR-T products manufactured with inosine improved tumor killing in a mouse model.³⁴ However, no studies have shown that ADO production by CAR-T cells via the CD38-CD73 axis is associated with the efficacy of anti-CD19 CAR-T therapy.

The enriched OXPHOS and relatively low immunological activity of 4MD CAR-T cells could be conducive to formation of memory anti-CD19 CAR-T cells after infusion.⁵¹ With respect to TERT upregulation in 4MD T cells, Bai et al. found that delivery of modified TERT mRNA increases the in vivo persistence and antitumor efficacy of anti-CD19 CAR-T cells.³⁷ By contrast, CAR-T cell dysfunction is associated with an NK-like T cell transition driven by transcription factors ID3 and SOX4, which are upregulated by continuous exposure to antigens.⁵² The reason why the percentage of 4MD T cells was higher in IPs from patients in remission than in relapsed patients, despite the same manufacturing process, is still not clear. However, given that patients in remission had lower numbers of blasts in the BM before CAR-T therapy, it is likely that T cells collected from patients in remission through apheresis had fewer prior encounters with leukemia cells, and thus they were able to avoid persistent stimulation by tumor antigens, which could induce overactivation and dysfunction.³⁰ Indeed, it was proposed recently that epigenetic scarring of exhausted T cells constrains future immune responses, despite the finding that some level of phenotypic and transcriptional recovery occurs after antigen clearance.^53,54,55

In conclusion, our serial analyses of IPs in patients with BCP-ALL treated with tisa-cel revealed that enrichment of CAR-T cells with a 4MD phenotype and a low capacity to produce ADO is a characteristic of patients in remission. These findings not only enable us to predict treatment efficacy but may also improve the quality of anti-CD19 CAR-T cells used for therapy.

Limitations of the study

The number of cases in this study is not large. Because the amount of collected IPs was small, we could not demonstrate the efficacy of 4MD T cells in in vivo experiments. We did not assess whether IPs in relapsed patients contained more ADO than those in patients in remission. Since we only investigated tisa-cel, a future study needs to confirm the results with other CAR-T products.

Resource availability

Lead contact

Further information and requests for resources should be directed to the lead contact, Itaru Kato (itarkt@kuhp.kyoto-u.ac.jp).

Materials availability

This study did not generate new unique reagents and materials.

Data and code availability

• •RNA-seq data for the tisa-cel IPs are available at the Japanese Genome-phenotype Archive (JGA) managed by the DNA DataBank of Japan (DDBJ): https://www.ddbj.nig.ac.jp/index-e.html under accession number JGAS000760: https://ddbj.nig.ac.jp/search/entry/jga-study/JGAS000760, https://humandbs.dbcls.jp/en/hum0405-v3. The instruction for accessing the data is indicated in (https://humandbs.dbcls.jp/en/data-use). The CyTOF data have been deposited in Zenodo: https://zenodo.org/) under accession number 13731333 and 13731405.
• •This paper does not report original code.
• •Any additional information is available from the lead contact upon request.

Acknowledgments

This investigation was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI (no. JP20H00528, JP21K19405, JP23K18264, and JP24H00628 to J.T.; no. JP22K07211 to I.K.); the Project for Promotion of Cancer Research and Therapeutic Evolution (P-PROMOTE) from AMED (no. JP22cm0106xxxh000x, 23ama221505h0002, 24ama221531h0001, and 25ama221531h0002 to J.T.; no. 22ama221514h0001 and 23ama221514h002 to I.K.); the Practical Research for Innovative Cancer Control from AMED (no. JP21ck0106531 to I.K.); the Princess Takamatsu Cancer Research Fund to J.T.; the Takeda Hosho Grants for Research in Medicine to J.T.; the internal grant Ishizue from Kyoto University Research Administration to J.T.; the Mother and Child Health Foundation to I.K.; and the Japan Leukemia Research Fund to I.K. Graphical abstract was created in BioRender: https://BioRender.com/pcpce8e.

Author contributions

T. Mikami, I.K., and H.H. provided leadership. J.B.W. provided key methodology. T. Mitsuyoshi, T.K., S.S., and K.U. collected samples and contributed to discussion. T. Mikami, Y.U., S.O., and M.A.L.-C. conducted transcriptomic analysis. A.T.-K., J.B.W., and J.T. interpreted the data and contributed to discussion. All authors agreed to the final version.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
BV421-anti-human CD38	Becton, Dickinson and Company (BD)	Cat#562444; RRID: AB_11151894
PE-anti-human CD73	Becton, Dickinson and Company (BD)	Cat#550257; RRID: AB_393561
APC-anti-human HLA-DR	BioLegend	Cat#307610; RRID: AB_314688
FITC-anti-human Tim-3	BioLegend	Cat#345022; RRID: AB_2563937
89Y-anti human CD45	Standard BioTools	Cat#3089003B; RRID: AB_2938863
141Pr-anti human CD196 (CCR6)	Standard BioTools	Cat#3141003A; RRID: AB_2687639
142Nd-anti-human CD19	Standard BioTools	Cat#3142001B; RRID: AB_3661857
143Nd-anti-human HLA-DR	Standard BioTools	Cat#3143013B; RRID: AB_3661844
144Nd-anti-human CD38	Standard BioTools	Cat#3144014B; RRID: AB_2687640
145Nd-anti-human CD4	Standard BioTools	Cat#3145001B; RRID: AB_3661845
146Nd-anti-human CD8a	Standard BioTools	Cat#3146001B; RRID: AB_3661846
148Nd-anti-human CD274 (PD-L1)	Standard BioTools	Cat#3148017B; RRID: AB_3677831
149Sm-anti-human CD25	Standard BioTools	Cat#3149010B; RRID: AB_2756416
150Nd-anti-human CD134 (OX40)	Standard BioTools	Cat#3150023B; RRID: AB_2938869
152Sm-anti-human TCRγδ	Standard BioTools	Cat#3152008B; RRID: AB_2687643
154Sm-anti-human CD3	Standard BioTools	Cat#3154003B; RRID: AB_2811086
155Gd-anti-human CD27	Standard BioTools	Cat#3155001B; RRID: AB_2687645
156Gd-anti-human CD183 (CXCR3)	Standard BioTools	Cat#3156004B; RRID: AB_2687646
158Gd-anti-human CD10	Standard BioTools	Cat#3158011B; RRID: AB_2921314
163Dy-anti-human CD56	Standard BioTools	Cat#3163007B; RRID: AB_3096015
164Dy-anti-human CD161	Standard BioTools	Cat#3164009B; RRID: AB_2687651
166Er-anti-human CD34	Standard BioTools	Cat#3166012B; RRID: AB_2756424
167Er-anti-human CD197 (CCR7)	Standard BioTools	Cat#3167009A; RRID: AB_2858236
170Er-anti-human CD45RA	Standard BioTools	Cat#3170010B; RRID: AB_2938862
171Yb-anti-human CD185 (CXCR5)	Standard BioTools	Cat#3171014B; RRID: AB_2858239
174Yb-anti-human CD279 (PD-1)	Standard BioTools	Cat#3174020B; RRID: AB_2868402
175Lu-anti-human CD194 (CCR4)	Standard BioTools	Cat#3175035A; RRID: AB_2921320
176Yb-anti-human CD127	Standard BioTools	Cat#3176004B; RRID: AB_3665122
209Bi-anti-human CD11b	Standard BioTools	Cat#3209003B; RRID: AB_2687654
purified-anti-human TCRαβ	BioLegend	Cat#306702; RRID: AB_314628
purified-anti-human CD33	BioLegend	Cat#303402; RRID: AB_314346
purified-anti-human CD14	BioLegend	Cat#301802; RRID: AB_314184
purified-anti-human CD147	BioLegend	Cat#306202; RRID: AB_314586
purified-anti-human β2 microgloblin	BioLegend	Cat#316302; RRID: AB_492835
Biotin CAR (FMC63)	ACROBiosystems	Cat#FM3-BY54
purified anti Biotin	BioLegend	Cat#409002; RRID: AB_10642032
purified-anti-human CD39	BioLegend	Cat#328202; RRID: AB_940438
purified-anti-human ICOS	BioLegend	Cat#313502; RRID: AB_416326
purified-anti-human CD57	BioLegend	Cat#359602; RRID: AB_2562403
purified-anti-human CD152 (CTLA-4)	BioLegend	Cat#349902; RRID: AB_10642827
purified-anti-human CD73	BioLegend	Cat#344002; RRID: AB_2154067
purified-anti-human CD22	BioLegend	Cat#302502; RRID: AB_314264
purified-anti-human Foxp3	eBioscience	Cat#14-4777-82; RRID: AB_467556
purified-anti-human Vα24	BioLegend	Cat#342902; RRID: AB_2229301
purified-anti-human CD195 (CCR5)	BioLegend	Cat#359102; RRID: AB_2562457
purified-anti-human CD16	BioLegend	Cat#302002; RRID: AB_314202
purified-anti-human Tim-3	BioLegend	Cat#345002; RRID: AB_2116574
purified-anti-human TIGIT	BioLegend	Cat#372702; RRID: AB_2632714
APC anti-human CD3	BioLegend	Cat#300412; RRID: AB_314066
Biological samples
Human bone marrow and blood samples, infusion products (tisagenlecleucel)	Kyoto University Hospital	N/A
Chemicals, peptides, and recombinant proteins
IDU	Standard BioTools	Cat#201127
Maxper MCP9 Antibody Labeling Kit, 111Cd	Standard BioTools	Cat#201111A
Maxper MCP9 Antibody Labeling Kit, 112Cd	Standard BioTools	Cat#201112A
Maxper MCP9 Antibody Labeling Kit, 114Cd	Standard BioTools	Cat#201114A
Maxper MCP9 Antibody Labeling Kit, 116Cd	Standard BioTools	Cat#201116A
Maxper X8 Antibody Labeling Kit, 147Sm	Standard BioTools	Cat#201147A
Maxper X8 Antibody Labeling Kit, 151Eu	Standard BioTools	Cat#201151A
Maxper X8 Antibody Labeling Kit, 159Tb	Standard BioTools	Cat#201159A
Maxper X8 Antibody Labeling Kit, 160Gd	Standard BioTools	Cat#201160A
Maxper X8 Antibody Labeling Kit, 161Dy	Standard BioTools	Cat#201161A
Maxper X8 Antibody Labeling Kit, 162Dy	Standard BioTools	Cat#201162A
Maxper X8 Antibody Labeling Kit, 165Ho	Standard BioTools	Cat#201165A
Maxper X8 Antibody Labeling Kit, 168Er	Standard BioTools	Cat#201168A
Maxper X8 Antibody Labeling Kit, 169Tm	Standard BioTools	Cat#201169A
Maxper X8 Antibody Labeling Kit, 172Yb	Standard BioTools	Cat#201172A
Maxper X8 Antibody Labeling Kit, 173Yb	Standard BioTools	Cat#201173A
Cell-ID Intercalator-Rh	Standard BioTools	Cat#201103A
Cell Acquisition Solution	Standard BioTools	Cat#201240
EQ Four Element Calibration Beads	Standard BioTools	Cat#201078
Lymphoprep	Alere Technologies	Cat#1858-1
CELLBANKER1	ZENOAQ RESOURCE	Cat#11910
RPMI-1640	FUJIFILM Wako	Cat#189-02025
AIM V medium	Thermo Fisher Scientific	Cat#12055083
Penicillin	Meiji Seika Pharma	JAN code:4987222637671
Streptomycin	Meiji Seika Pharma	JAN code:4987222002929
Fc receptor Binding Inhibitor Functional Grade Monoclonal Antibody	eBioscience	Cat#16-9161-73
dichloro-(ethylenediamine) palladium (II)	Sigma-Aldrich	Cat#400076-1G
FOXP3/Transcription Factor Staining Buffer Set	eBioscience	Cat#00-5523-00
ddPCR Supermix for Probes	Bio-Rad	Cat#1863026
NAD+	Selleck Biotech	Cat#S2518
RNA Screen Tape	Agilent	5067–5576
Imunace for Injection (IL-2)	KYOWA Pharmaceutical Industry	KEGG DRUG: D02749
CountBright Absolute Counting Beads	Thermo Fisher Scientific	Cat#C36950
DAPI Nucleic Acid Stain	Lonza	Cat#PA-3013
Critical commercial assays
QIAmp DNA blood mini kit	QIAGEN	Cat#51104
Qubit ds DNA Quantification Assay Kits	Thermo Fisher Scientific	Cat#Q32850
Adenosine Assay Kit	Cell Biolabs, Inc	Cat#MET-5090
MycoStrip	InvivoGen	Cat#REP-MYS-10
RNeasy UCP Micro Kit	QIAGEN	Cat#73934
NEBNext Ultra II RNA Library Kit for Illumina	NEW ENGLAND Biolabs	Cat#E7770L
MGIEasy Circularization Kit	MGI Tech	Cat#1000005259
Deposited data
Raw RNA-seq Data	This paper	DDBJ: JGAS000760
CyTOF data	This paper	Zenodo: https://doi.org/10.5281/zenodo.13731333, https://doi.org/10.5281/zenodo.13731405
Public scRNA-seq Data	Bai et al.43	NBCI Gene Expression Omnibus (accession number GSE262072)
Experimental models: Cell lines
SU/SRGFP/Luc	Hirase et al.35; Tanaka et al.36	N/A
Oligonucleotides
Primer: Tisagenlecleucel Forward: TGCCGATTTCCAGAAGAAGAAGAAG	Davis et al.22	N/A
Primer: Tisagenlecleucel Reverse: GCGCTCCTGCTGAACTTC	Davis et al.22	N/A
VIC TaqMan NFQ-MGB probe for Tisagenlecleucel: ACTCTCAGTTCACATCCTC	Davis et al.22	N/A
Primer: CDKN1A Forward: GAAAGCTGACTGCCCCTATTTG	Davis et al.22	N/A
Primer: CDKN1A Reverse: GAGAGGAAGTGCTGGGAACAAT	Davis et al.22	N/A
6FAM TaqMan NFQ-MGB probe for CDKN1A: CTCCCCAGTCTCTTT	Davis et al.22	N/A
Software and algorithms
CyTOF Software	Standard BioTools	Version 7.0.8493
Cytobank Premium	Beckman Coulter	https://premium.cytobank.org/cytobank/
CATALYST	R package	Version 1.32.0
FlowSOM	R package	Version 2.16.0
DESeq2	R package	Version 1.36.0
ggvolcanoR	R package	N/A
QuantaSoft	Bio-Rad	Version 1.7.4
Prism 9	GraphPad	Version 9.5.1
FlowJo	Becton, Dickinson and Company (BD)	Version 10.8.1
Genomon	Human Genome Center, the Institute of Medical Science, the University of Tokyo	Version 2.6.3, https://github.com/Genomon-Project/
GSEA software	Broad Institute	Version 4.3.2
Metascape software	Zhou et al.40	Version 3.5.20250101, https://metascape.org/gp/index.html#/main/step1
RNAseqChef	Etoh et al.56	Version 1.1.4
Seurat	Hao et al.57	Version 4.4.0
BioRender	BioRender	https://www.biorender.com/
BD FACSuite	Becton, Dickinson and Company (BD)	Version 1.5.0.925
Other
Helios The CyTOF system	Standard BioTools	Cat#107001
Qubit 4 Fluorometer	Thermo Fisher Scientific	Cat#Q33238
QX200 AutoDG ddPCR system	Bio-Rad	Cat#1864100JA
C1000 Touch thermal cycler	Bio-Rad	Cat#1851197JA
PX1 PCR plate sealer	Bio-Rad	Cat#1814000J1
Spark Multimode Microplate Reader	Tecan	N/A
BD FACSAria II cytometer	Becton, Dickinson and Company (BD)	N/A
BD FACSLyric	Becton, Dickinson and Company (BD)	N/A
4150 TapeStation System	Agilent	Cat#G2992AA
DNBSEQ-G400RS	MGI Tech	N/A
supercomputer SHIROKANE	Human Genome Center, the Institute of Medical Science, the University of Tokyo	N/A

Experimental model and study participant details

Description of study participants

We enrolled 16 children and three young adults with relapsed/refractory BCP-ALL, all of whom received tisa-cel between 2019 and 2023 at Kyoto University Hospital. Patients were divided into two groups (nine patients in remission, and ten relapsed patients) based on whether they maintained remission after CAR-T infusion (observation for at least one year). Due to the small number of patients, gender-specific analysis was not performed in this study. All patients or their guardians provided informed consent for sample collection. The patient characteristics are shown in Table S1. The study was approved by the Kyoto University Hospital Ethical Board (approval number: G-1030) and related institutions.

Cell line

As a CD19⁺ BCP-ALL cell line, SU/SR^GFP/Luc was used in CTL assays using IPs. SU/SR is a human Philadelphia chromosome-positive pre-B ALL cell line derived from SU-Ph2 cells by culture with imatinib,³⁵ and the SU/SR cells were provided to us by the author who created the cell line. SU/SR^GFP/Luc was established previously³⁶ in our laboratory and cryopreserved in liquid nitrogen using CELLBANKER 1 (ZENOAQ RESOURCE). SU/SR^GFP/Luc has been confirmed to be mycoplasma negative by MycoStrip (InvivoGen) before using.

Method details

Sample collection and preservation

PB and BM aspirates were processed by density gradient centrifugation using Lymphoprep (Alere Technologies). The obtained mononuclear cells were preserved in liquid nitrogen using CELLBANKER 1. Regarding T cells from IPs: after administration, product bags were washed twice with phosphate buffered saline (PBS) and leftover T cells were collected, pelleted, suspended in CELLBANKER 1, and preserved in liquid nitrogen.

Mass cytometry

Frozen samples were thawed in a 37°C water bath, resuspended in RPMI 1640 medium (FUJIFILM Wako) supplemented with 10% fetal calf serum and penicillin-streptomycin (Meiji Seika Pharma), and then washed twice (hereafter, all washing steps were performed by centrifugation at 400g, 4°C, for 5 min, in 2 mL of solvent). Cells were counted and up to 2×10⁶ cells/sample were used for subsequent sample preparation. Barcoding antibodies (anti-CD147 and anti-β2 microglobulin, in the predetermined combinations listed in Table S2) were added to cell-staining medium (CSM; PBS containing 0.1% bovine serum albumin, 2 mM ethylenediaminetetraacetic acid, and 0.01% sodium azide) to yield final reaction volumes of 100 μL, followed by incubation at room temperature (RT) for 30 min. After washing twice with CSM, barcoded samples were mixed together and pelleted. Blocking was performed for 10 min using 5 μL of Fc receptor Binding Inhibitor Functional Grade Monoclonal Antibody (eBioscience). Subsequently, anti-chemokine antibodies and IDU were added to RPMI to yield a final reaction volume of 100 μL, and the samples were incubated at 37°C for 30 min, washed twice with CSM, and stained at RT for 30 min with 100 μL CSM containing anti-cell surface marker antibodies.

Next, the cells were washed twice with PBS and stained for 10 min with PBS containing 500 nM dichloro-(ethylenediamine) palladium (II) (Sigma-Aldrich) to assess viability. Intracellular staining of Foxp3 and CTLA-4 was conducted using the FOXP3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific). Finally, 150 μL of PBS containing 2% formaldehyde and 1:500 Cell-ID Intercalator-Rh (Standard BioTools) was added to the pelleted samples, which were then stored overnight at 4°C. Before a measurement, samples were washed once with CSM and twice with Cell Acquisition Solution (Standard BioTools), filtered (35 μm) to remove aggregates, and resuspended in Cell Acquisition Solution containing 15% EQ Four Element Calibration Beads (Standard BioTools). Throughout the acquisition of data by CyTOF (Helios, Standard BioTools), cells were introduced at a constant rate of approximately 100–400 cells/s. Acquired data were normalized using CyTOF Software (v.7.0.8493) and uploaded into Cytobank Premium: https://premium.cytobank.org/cytobank/. After singlet gating, live cells were gated based on palladium staining, debarcoded, and assigned to each original sample. The CyTOF antibody panel is shown in Table S3.

T cell data were obtained by gating on the CD3⁺ fraction in each sample. CAR^pos and CAR^neg fractions were identified by expression (or not) of CAR (FMC63); T cells from the patients’ blood (obtained before CAR-T cell infusion) were used as a negative control. Downstream analysis was conducted using the R package CATALYST.^25,26,27 In the CyTOF workflow, unsupervised clustering was performed by FlowSOM,⁵⁸ and similar clusters were merged by checking their marker expression patterns. To evaluate the statistical significance of differences between patients in remission and relapsed patients, differential abundance analysis was carried out, and adjusted p-values were calculated using a generalized linear mixed model (GLMM).

Monitoring of tisa-cel transgene DNA using ddPCR

DNA was purified from cryopreserved mononuclear cells isolated from BM aspirates or PB using QIAmp DNA blood mini kits (QIAGEN). Isolated DNA was quantified by Qubit double-stranded DNA (dsDNA) broad-range assay kits (Thermo Fisher Scientific) and stored at −30°C.

Primers and probes targeting the CDKN1A gene and the junction region of the 4-1BB co-stimulatory domain and the CD3-zeta signaling region of the tisa-cel transgene were designed as described previously.²² The TaqMan non-fluorescent quencher-minor groove binder (NFQ-MGB) probes (fluorescence: FAM for CDKN1A, VIC for CAR) were supplied by Thermo Fisher Scientific.

ddPCR were performed using a QX200 AutoDG ddPCR system (Bio-Rad) and a C1000 Touch thermal cycler (Bio-Rad). All ddPCR reactions were prepared by mixing 12.5 μL of 2×ddPCR supermix (Bio-Rad), 1.25 μL of a 20× primer/probe mix containing both forward and reverse primers (900 nM each, final concentration) and probes (250 nM final concentration), and 10 ng of DNA in final volume of 25 μL. Droplets were created by a Droplet Generator (Bio-Rad). The PCR amplification conditions used by the C1000 Touch thermal cycler were as follows: one cycle at 95°C for 10 min, followed by 40 cycles at 94°C for 30 s and 55.5°C for 1 min, and one cycle at 98°C for 10 min. The amplified droplets were kept at 4°C until they were analyzed on a QX200 droplet reader (Bio-Rad). The type of experimental droplet reading was absolute quantification (ABS). Obtained data were analyzed using QuantaSoft (Bio-Rad, v.1.7.4) and Prism 9 (GraphPad, v.9.5.1).

Adenosine-producing assay for the IPs

Each preserved IP sample was thawed in a 37°C water bath, resuspended in 10 mL of AIM V medium (Thermo Fisher Scientific), and then washed twice with PBS. The T cells from IPs were resuspended in PBS containing 1 mM NAD⁺ (Selleck Biotech) to yield 2×10⁶ cells/ml in a final reaction volume of 330 μL, and then incubated for 2 h at 37°C/5% CO₂. The culture supernatant was then collected by centrifugation for 5 min at 700 g.

The adenosine assay was performed using an Adenosine Assay Kit (Cell Biolabs, Inc). Briefly, 50 μL of culture supernatant, serially diluted standard solution, or 1 mM NAD⁺ PBS (negative control) was mixed with the same quantity of Reaction Mix containing adenosine metabolic enzymes (adenosine deaminase, purine nucleoside phosphorylase, and xanthine oxidase) and a fluorescent probe. After a 15-min incubation at RT in the dark, the reaction solutions (in a 96-well black plate) were read by a Spark Multimode Microplate Reader (Tecan) equipped with a 535 nm excitation filter (bandwidth: 25 nm) and a 590 nm emission filter (bandwidth: 20 nm). The Net Relative Fluorescence Unit (RFU) value for each sample was converted into adenosine concentration using a standard curve. Each culture supernatant was assayed in duplicate, and the experiment was repeated three times.

In vitro CTL assays using IPs

After thawing cryopreserved CAR-T cells from IPs, CAR-T cells were cultured for 24 h in RPMI 1640 medium supplemented with 10% fetal calf serum, penicillin-streptomycin, and 50 IU/mL IL-2 (Imunace 35, KYOWA Pharmaceutical Industry). Then, SU/SR^GFP/Luc cells (3×10³) were cultured with CAR-T cells at various ratios (1:0, 1:1, and 1:2 [SU/SR^GFP/Luc:CAR-T cells]) in RPMI 1640 medium supplemented with 10% fetal calf serum and penicillin-streptomycin (final volume, 100 μL in a 96-well round bottom plate) for 24 h. Thereafter, 1.98×10⁴ CountBright Absolute Counting Beads (Thermo Fisher Scientific) were added to each well, and cells were stained at 4°C for 30 min with an anti-human CD3-APC antibody (BioLegend, 1:100) and DAPI (Lonza, 1:1000). Before flow cytometric analysis, samples were resuspended in 300 μL of PBS and measured using a FACSLyric cytometer. Acquired data were analyzed using FlowJo software (v10.8.1) (BD Biosciences), and the absolute number of SU/SR^GFP/Luc cells was calculated from the number of CountBright Absolute Counting Beads. Statistical analysis was performed using GraphPad Prism 9.5.1 (GraphPad Software).

RNA sequencing

Prior to RNA isolation, T cells in the IPs were separated into a CD38⁻CD73⁻Tim-3⁻HLA-DR⁺ fraction (4MD) and others (non-4MD) using a BD FACSAria Ⅱ cytometer. The antibodies used for separation are summarized in Table S4. T cells were stained with fluorophore-conjugated antibodies in a final reaction volume of 100 μL in PBS at RT for 30 min. After staining, the samples were washed twice in PBS and suspended in RPMI 1640 medium for cell sorting.

RNA isolation was carried out using an RNeasy UCP Micro Kit (QIAGEN). Agilent 4150 TapeStation and RNA Screen Tapes (Agilent Technologies) were used to measure RNA integrity, and libraries were prepared using the NEBNext Ultra Ⅱ RNA Library Kit for Illumina (New England Biolabs) and the MGIEasy Circularization Kit (MGI Tech). Constructed single-stranded circular DNA libraries were run on DNBSEQ-G400RS (MGI Tech) in 150-basepair (bp) paired-end mode. Sequence alignment to the GRCh37 human genome assembly was performed using STAR,⁵⁹ and read counting was performed using the in-house pipeline, Genomon v.2.6.3: https://github.com/Genomon-Project/. The expression level of each RefSeq gene was calculated from the mapped read counts, and normalized using the R package DESeq2 version 1.36.0.⁶⁰ The abovementioned alignment and data processing were performed by the supercomputer SHIROKANE at Human Genome Center, Institute of Medical Science, The University of Tokyo.

Gene set enrichment was analyzed by GSEA software (v 4.3.2) from the Broad Institute,^61,62 and differentially regulated MSigDB gene sets (h.all.v2024.1.Hs.symbols and c5.go.v2024.1.Hs.symbols) were evaluated. DEGs were extracted using the R package DESeq2, and volcano plots generated by R package ggvolcanoR.⁶³ Pathway analysis was conducted using Metascape software: https://metascape.org/gp/index.html#/main/step1.⁴⁰ Other transcriptomic outputs, including an MDS plot, heatmap, over-representative analysis of gene ontology (GO) biological process and reactome, and gene-concept (cnet) plots were produced by RNAseqChef: https://imeg-ku.shinyapps.io/RNAseqChef/ (method used for DEG detection = DESeq2; methods used for FDR control = Bonferroni/Holm; cut-off conditions = -fold change = 2, FDR = 0.05, basemean = 0).⁵⁶

Processing of public scRNA-seq data

Single-cell CITE-seq data from Bai et al.⁴³ were obtained from the NCBI Gene Expression Omnibus (accession number GSE262072) and analyzed using the Seurat (v4.4.0) pipeline. A Seurat object was reconstructed from the RNA and cell surface protein (CSP) count matrices. Only samples derived from unstimulated CAR-T products were retained for further analysis.

RNA data were log-normalized, and CSP data were normalized using the centered log-ratio method. The standard Seurat workflow for multimodal data integration was then applied.⁵⁷ Clustering was performed on the RNA data using the SLM algorithm with a resolution parameter of 1.2.

To assess expression of the 4MD gene program, the AddModuleScore function was used with a gene set comprising significantly upregulated genes identified in our bulk RNA-seq analysis (Table S5). Clusters with a median 4MD module score greater than 0 were annotated as 4MD-positive. The percentage of 4MD cells in each sample was subsequently calculated.

Quantification and statistical analysis

Since the number of pediatric and young-adult patients with BCP-ALL who were treated with tisa-cel was not large, all applicable patients were enrolled in the study. The experiments were not randomized, and investigators were not blinded to the results. The statistical methods used are two-sided and indicated in the corresponding parts of the text, figures, legends, and methods. Statistical values for GSEA, Metascape, and RNAseqChef were calculated as reported previously.^40,56,62 DEGs were compared based on the DESeq2-derived -log₁₀ adjusted p-values. The adenosine assay was repeated three times, with two biological replicates, and statistical analyses were performed using GraphPad Prism 9.5.1 (GraphPad Software). Error bars and significance values are presented in the figure legends; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. No data were excluded from the analysis.

Published: January 23, 2026