Transcriptional signatures associated with persisting CD19 CAR-T cells in children with leukemia

Abstract

In the context of relapsed and refractory childhood pre-B cell acute lymphoblastic leukemia (R/R B-ALL), CD19-targeting chimeric antigen receptor (CAR)-T cells often induce durable remissions, which requires the persistence of CAR-T cells. In this study, we systematically analyzed CD19 CAR-T cells of 10 children with R/R B-ALL enrolled in the CARPALL trial via high-throughput single-cell gene expression and T cell receptor sequencing of infusion products and serial blood and bone marrow samples up to 5 years after infusion. We show that long-lived CAR-T cells developed a CD4/CD8 double-negative phenotype with an exhausted-like memory state and distinct transcriptional signature. This persistence signature was dominant among circulating CAR-T cells in all children with a long-lived treatment response for which sequencing data were sufficient (4/4, 100%). The signature was also present across T cell subsets and clonotypes, indicating that persisting CAR-T cells converge transcriptionally. This persistence signature was also detected in two adult patients with chronic lymphocytic leukemia with decade-long remissions who received a different CD19 CAR-T cell product. Examination of single T cell transcriptomes from a wide range of healthy and diseased tissues across children and adults indicated that the persistence signature may be specific to long-lived CAR-T cells. These findings raise the possibility that a universal transcriptional signature of clinically effective, persistent CD19 CAR-T cells exists.

Fig. 1. Study overview and workflow.

a, Swimmer plot illustrating the responses of individual pediatric patients with B-ALL to CAT CAR T-cell therapy and timepoints of sample collection. Attainment of complete remission was associated with attainment of B cell aplasia in all cases. Patient 5 had an isolated unilateral ocular relapse of CD19⁺ leukemia, which was treated with enucleation and remains in ongoing minimal residual disease (MRD) negative remission with no other intervention. Heat map to the right demonstrates timepoint representation per patient. Filled black boxes indicate the presence of the timepoint. Product = infusion products (M0); early = M1–M3; mid = M4–M6; late = M7–M60. b, Schematic workflow of study design. Samples were collected from infusion products (PROD), peripheral blood (PB) and bone marrow (BM) between M0 and M60. Samples were used either for flow-based immunophenotyping or for single-cell GEX and TCR sequencing on the Chromium 10x platform. RT, reverse transcription; Seq, sequencing. c, UMAP of all cells in the dataset highlighting cell types captured (left) and expression of the CAT-scFv CAR construct (right). CAT-scFv, low-affinity CAR (CAT) incorporating a CD19-specific scFv.

Fig. 2. Characterization of CARPALL CAR-T cells.

a, UMAP of CAT CAR-T cells demonstrates transcriptional clustering based on timepoint. Product = infusion products (M0); early = M1–M3; mid = M4–M6; late = M7–M60. b, UMAPs show scaled average expression of lymphocyte markers. c, Immunophenotyping CAT CAR-T cells by multi-parameter flow cytometry. Representative example of P01 showing cell identification using CD4 and CD8A at an early (M2) and a late (M72) timepoint. Cells were gated for CD3 and CAT CAR. d–f, Trajectory scatter plots quantify data from all patient samples for CD4, CD8 and double-negative CAR-T populations. g, Representative example of P01 showing immunophenotyping using CD45RA and CCR7 at an early (M2) and a late (M72) timepoint. Cells were gated for CD3 and CAT CAR. TCM, central memory; TEM, effector memory; TEMRA, terminally differentiated effector memory expressing CD45RA; TN, naive; TSCM, stem cell memory. Fluorochromes: BV, brilliant violet; Cy, cyanine dye; PE, R-phycoerythrin; PerCP, peridinin-chlorophyll protein. Source data

Fig. 3. Cell typing and defining a persisting CAR-T cell transcriptional signature.

a, UMAP of CAT CAR-T cells shows closest cell type matching using a PBMC reference and cell-marker-based annotation. Late (persisting, M7–M60) CAR-T cells are in green hues; mid (M4–6) CAR-T cells are in black/blue hues; and early (M1–3) CAR-T cells are in pink/red hues. b, Stacked bar plots show dynamic cell type proportions over time for a representative patient (P01). The top five abundant cell types in the dataset are shown. c, Contour plots quantify expression of GZMK against GZMH across patients per timepoint. d, Heat map shows differential gene expression results creating gene signatures of each timepoint. DN, double-negative; TCM, central memory; TEM, effector memory.

Fig. 4. CAR-T cell immunophenotyping.

Bar plots show custom gene modules that functionally characterize CAR-T cells. The height of the bar refers to the fraction of cells per cell type that express the gene. Higher bars indicate that more cells of that cell type are expressing that gene. The shading refers to the average scaled expression of those genes for that cell type. Darker shading indicates that the expression of that gene is above the average expression of that gene across all cell types. All CAR-T cells (product, early, mid and late) from all patients (n = 10) are analyzed. DN, double-negative; TCM, central memory; TEM, effector memory.

Fig. 5. Population structures of CAR-T cells.

a, Representative sample (P02) illustrating the changes in clonal architecture of CAR-T cells over time. Packed circle plots show the size of each clonotype. Filled-in black circles represent clonotypes that are not unique, as they are observed across timepoints. Conversely, pink donut circles represent clonotypes that are unique to that timepoint (and not observed across time). Blue/purple/gray colored circles represent the dominant clonotype at that timepoint that corresponds with the clonal trajectories below. b, Cell type composition stacked bar plots demonstrate the shift in cell type abundances between early and late timepoints and between unique and non-unique clonotypes. TEM, effector memory.

Fig. 6. Evaluating the persisting transcriptional signature across T cells, including adult long-lived CAR-T cells.

a, UMAP plots show expression of the strongest markers of the late-persisting CAR-T signature. Red arrows highlight persisting CAR-T cells between the CARPALL (CAT CAR-T) and CTL019 studies. b, Heat map demonstrates logistic regression cell-to-cell matching. CTL019-persisting CD4 CAR-T cells match strongly to late double-negative CAT CAR-T cells from the CARPALL study, as indicated by the red arrows. c, Box plots show the per-cell distribution of the late score as defined by the late-persisting CAR-T signature. Scores are shown for CAR-T, cancer and normal datasets. Publicly available datasets analyzed in this study are described in Supplementary Table 6. The red line represents the median of the late double-negative (DN TEM) cells from the CARPALL study. Colored cell types refer to CARPALL CAR-T cells. Red arrows indicate persisting CAT CAR-T cells from CARPALL and persisting CTL019 cells. Box plots show the first quartile (the lower end of the box) and the third quartile (the upper other end of the box) as well as the median values (center line) per dataset. The ‘whiskers’ extend from the ends of the box to a maximum and minimum of 1.5 times the interquartile range beyond the box. Outliers are shown as dots. AXICEL, axicabtagene ciloleucel CAR-T cells (infusion products); BRCA, breast cancer; ChiCTR, BCMA CAR-T (ChiCTR, 1800017404); CRC, colorectal cancer; DN, double-negative; GBM, glioblastoma; HCC, hepatocellular carcinoma; JCAR014, Fred Hutchinson Cancer Center CAR-T cells; NB, neuroblastoma; NPC, nasopharyngeal carcinoma; NSCLC, non-small lung cell carcinoma; SKCM, skin cutaneous melanoma (1 = Li; 2 = Yost; 3 = Jerby-Anon); TCM, central memory; TEM, effector memory. The numbers of cells and samples used in this figure are described in Supplementary Table 6.

Extended Data Fig. 1. Sample breakdown.

Bar charts show the number of experimental samples derived from each patient sample. Each patient sample, defined as an individual, timepoint and sample source (that is P01 M01 PB) can be used for gene expression (GEX) and T-cell receptor (TCR) sequencing and/or flow-based immunophenotyping. For those samples with GEX/TCR sequencing, they are flow-sorted by CD3 and CAR prior to sequencing, with the exception of P02 M02 PB which was sorted for CAR-T cells, T cells and other leukocytes, thus they only contribute 1 experimental sample each. For two samples (P02 M06 BM and P01 M06 BM), no cells were recovered from the CAR T channel. M=month, PB = peripheral blood, BM = bone marrow.

Extended Data Fig. 2. Global cluster annotation.

Uniform Manifold Approximation Projections (UMAPs) show all cells in the dataset coloured by (a) patient of origin, (b) sample source, and (c) timepoint.

Extended Data Fig. 3. Cluster markers and cell type trajectories.

(a) Heatmap shows the average scaled expression of marker genes used for cell type annotation. (b–f) Stacked bar plots show dynamic cell type proportions over time for all patients. Top 5 abundant cell types in the dataset are shown.

Extended Data Fig. 4. General flow-based immunophenotyping.

(a) Immunophenotyping non-CAR and CAR-T cells by multi-parameter flow cytometry. All cells were initially gated for CD3 and the CAR. Red boxes and arrows highlight subsequent gating strategies for representative examples illustrating double-negative CAR-T cells are αβ T cells and CD8A/CD8B negative, in contrast to non-CAR-T cells. Fluorochromes: PerCP-Cy5.5 = Peridinin chlorophyll protein-Cyanine5.5, FITC = Fluorescein isothiocyanate, Brilliant Violet (BV) 650, BV510, BV786. (b) Trajectory scatter plots quantify flow plots signals across patient samples for CD45RA and CCR7. Cells were gated for CD3 and the CAR. TN = Naïve, TSCM = Stem Cell Memory, TCM = Central Memory, TEM = Effector Memory, TEMRA = Terminally Differentiated Effector Memory expressing CD45RA.

Extended Data Fig. 5. Gamma-Delta and NK Immunophenotyping CAR-T cells by multi-parameter flow cytometry.

All cells were initially gated for CD3, CAR, CD4 and CD8A. (a) Flow plots for representative samples (P06 and P09) with high populations of double-negative TCRγδ cells. (b) Trajectory scatter plots quantify flow plots signals across patient samples for TCRγδ. Cells were previously gated for CD3, CAR, CD4 and CD8. (c) Flow plots show gating for CD16 and CD56. Prior gating includes CD3, CAR, CD4, CD8. Fluorochromes: APC = Allophycocyanin, Cy = Cyanin.

Extended Data Fig. 6. Expression of key genes in the persisting CAR T-cell signature.

Boxplots show the percent of T cell expression for (a) GZMK, (b) TIGIT, and (c) GPR183 for flow-based immunophenotyping (left) and single-cell RNA sequencing (scRNA-seq) data. CAR-T cell scRNA-seq, data is derived from 60188 cells from 43 independent samples. For non-CAR T-cell scRNAseq, data is derived from 98003 cells from 43 independent samples. Flow data is derived from 15 independent samples. Box plots show the first quartile (the lower end of the box) and the third quartile (the upper other end of the box), as well as the median values (centre line) per dataset. The ‘whiskers’ extend from the ends of the box to a maximum and minimum of 1.5 times the interquartile range beyond the box. Outliers are shown as dots.

Extended Data Fig. 7. Exhaustion panel expression and flow-based immunophenotyping.

(a) Trajectory scatter plots show the percent of T cells expression for LAG3, PDCD1, and TIM3 across time for flow-based immunophenotyping (left) and single-cell RNA sequencing (scRNA-seq) data. Ribbons refer to the 95% confidence level interval for predictions using a ‘loess’ model to fit the line. (b) Trajectory scatter plots show the percent of T cells expression for the combination of LAG3, PDCD1, and TIM3 across time for flow-based immunophenotyping. Ribbons refer to the 95% confidence level interval for predictions using a ‘loess’ model to fit the line.

Extended Data Fig. 8. Clonal structure of CAR-T cells.

(a) Clonal trajectory illustrates the changes in clonal architecture of CAR-T cells over time across a second representative patients with more than 3 timepoints sequenced. Packed circle plots show the size of each clonotype. Filled-in black circles represent clonotypes that are not unique, as they are observed across timepoints. Conversely, pink donut circles represent clonotypes that are unique to that timepoint (and not observed across time). Blue/purple/grey coloured circles represent the dominant clonotype at that timepoint that correspond with the clonal trajectories below. (b) Cell type composition stacked bar plots demonstrate the shift in cell type abundances between early timepoints and late timepoints between unique and non-unique clonotypes (right). (c) Trajectory scatter plots quantify data from all patient samples for CAR expression across time.

Extended Data Fig. 9. Evaluation of the persisting CAR-T cell signature across T-cells: expanded view.

Boxplots show the per cell distribution of the late score as defined by the late, persisting CAR-T signature. Scores are shown for representative datasets encompassing (a) the CARPALL CAR-T cells, (b) CARPALL Non-CAR-T cells, CTL019 CAR-T cells, (c) Tumour-infiltrating T cells, and (d) T cells from normal datasets. The red dashed line represents the median of the late double-negative (DN TEM) cells from the CARPALL study. Across datasets, the x-axis ticks refer to Seurat-based clusters per dataset, with the exception of the CARPALL CAR-T cell dataset where the x-axis refers to annotated cell types. Box plots show the first quartile (the lower end of the box) and the third quartile (the upper other end of the box), as well as the median values (centre line) per dataset. The ‘whiskers’ extend from the ends of the box to a maximum and minimum of 1.5 times the interquartile range beyond the box. Outliers are shown as dots. BRCA = Breast Cancer, CRC=Colorectal Cancer, PBMC = Peripheral Blood Mononuclear Cells. Publicly available datasets analysed in this study and the number of cells and samples used in this figure are described in Supplementary Table 6.

Extended Data Fig. 10. Flow cytometry gating strategy for CD3⁺ CAR⁺ T cells sorting.

SSC = side scatter, FSC = forward scatter. Fluorochromes: FITC = Fluorescein isothiocyanate, PE = R-phycoerythrin, APC = Allophycocyanin.