Integrative single-cell multi-omics of CD19-CARpos and CARneg T cells suggest drivers of immunotherapy response in B cell neoplasias

Abstract

The impact of phenotypic, clonal, and functional heterogeneity of chimeric antigen receptor (CAR)-T cells on clinical outcome remains understudied. Here, we integrate clonal kinetics with transcriptomic heterogeneity resolved by single-cell omics to interrogate cellular dynamics of non-transduced (CAR^neg) and transduced (CAR^pos) T cells, in the infusion product (IP) and at the CAR-T cell expansion peak in five B cell acute lymphoblastic leukemia (B-ALL) patients treated with CD19CAR-T cells (varni-cel). We identify significant differences in cellular dynamics in response to therapy. CAR^pos T cells at IP of complete response patients exhibit a significantly higher CD4:CD8 ratio, validated in a larger cohort B-ALL patients (n = 47). Conversely, at the expansion peak, there is a clonal expansion of CD8⁺ effector memory and cytotoxic T cells. Cytotoxic CAR^pos γδ-T cells expansion correlates with treatment efficacy validated in a cohort of B-ALL (n = 18) and diffuse large B cell lymphoma (DLBCL) patients (n = 58). Our data provide insights into the complexity of T cell responses following CAR-T cell therapy and suggest drivers of immunotherapy response.

Graphical abstract

Figure 1

The landscape of T cell populations in r/r B-ALL patients treated with varni-cel (A) Schematic overview of the experimental design. CAR-T cells were manufactured from leukapheresis products. CD4/CD8 T cells were selected, activated, transduced with the CD19CAR construct, and expanded in vitro. Upon manufacturing, a sample was collected from each patient’s product prior to infusion, conforming the “Infusion Product” (IP). Subsequently, patients were monitored weekly to assess the CAR-T cell expansion, and the sample with the highest level of CAR^pos T cells over this period conformed the peak of expansion (Peak). CAR^neg and CAR^pos T cell fractions were fluorescence-activated cell sorting-purified from IP and peak samples and prepared in single-cell suspensions for scRNA-seq and scTCR-seq. (B) Flow cytometry monitoring for CAR^pos T cell expansion over the first four weeks after infusion. The percentage of CAR^pos T cells in the IP is indicated as a green dot whereas peak samples are represented as red dots. (C) Summary of the clinical evolution of the patients over a ten-month follow-up. Solid green and white bars indicate the presence or absence (no detection) of CAR^pos T cells in disease-free patients, respectively. Stripped red lines indicate patient relapse. (D–H) Uniform manifold approximation and projection (UMAP) of the 37,100 cells pooled from all the samples from the 5 patients colored as follows: (D) CAR^pos vs. CAR^neg cells and (E) IP vs. peak, (F and G) scaled from high (yellow) to low expression (magenta) of cycling score (F), and effector/cytotoxic score (G). (H) UMAP representing clonotype sizes (from gray to purple). (I) UMAP colored by the 11 different cell clusters identified by unsupervised clustering both combined (left) and separated by patient (right). (J) Dot plot of the top gene markers by cluster identified by differential expression analysis.

Figure 2

Differences among CAR^pos and CAR^neg T cells in the IP (A) Stacked bar plot showing T cell subtype composition of CAR^pos and CAR^neg T cells in the IP. (B) Left, proportion of proliferative and non-proliferative CAR^pos and CAR^neg T cells in the IP. Right, stacked bar plots showing the T cell subtype composition within the non-proliferative (upper panels) and proliferative (lower panels) CAR^pos and CAR^neg T cells both overall and by patient in the IP. (C) Barplot (overall) and heatmap (by patient) showing the differential abundance analysis of CAR^pos and CAR^neg T cell clusters in the IP. Cluster abundance is represented as the log2 odd ratio between CAR^pos and CAR^neg subsets. Significant differences are computed with sccomp (see STAR Methods) and are indicated with ∗ (p ≤ 0.05). (D and E) Distribution of predicted CD4 and CD8 T cell classifications across all cells in the IP (D) and CAR^pos CD4 and CD8 T cell classifications, broken down by individual patient, showing how the CARpos CD4:CD8 T cell ratio varies among patients (E). (F) Boxplots depicting the differential abundance of CAR^pos CD4 and CD8 T cell populations in the IP across different patients, grouped by clinical outcome: patients 1 and 4 show early relapses and patients 2, 3, and 5 are long-term responders. Significant differences are computed with sccomp (see STAR Methods) and are indicated with ∗∗∗∗ p values ≤0.0001. (G) In a retrospective validation cohort of 47 patients with r/r B-ALL treated with Varni-cel, the CR rates were compared between two groups based on a CAR^pos CD4:CD8 ratio >3 or <3 (chi-squared test; p value = 0.03; n = 16 and n = 44). (H) Kaplan-Meier EFS curve for r/r B-ALL patients who received varni-cel products with a CAR^pos CD4:CD8 ratio >3 (gray line; n = 16) and <3 (black line; n = 44), respectively (Log rank test; p value = 0.26). Time is represented in months. (I) Violin plot showing the CAR-T cell persistence, clinically defined as loss of B cell aplasia, for varni-cel-treated r/r B-ALL patients who received products with a CAR^pos CD4:CD8 ratio >3 or <3 (n = 14 and n = 11; Mann-Whitney test; p value = 0.059). (J and K) Dot plots showing the expression of four exhaustion signatures in CAR^pos and CAR^neg T cells at the IP both overall (J) and by patient (K). The Ucell method was used to score each signature (Table S5). (L) The mean percentage of CAR-T cells expressing the indicated exhaustion markers at the IP was retrospectively compared in varni-cel-treated r/r B-ALL patients categorized based on whether their CAR-T cell persistence (left panel) or EFS (right panel) was above or below the median. Squares and circles correspond to CD4⁺ and CD8⁺ T cells, respectively.

Figure 3

Longitudinal analysis of CAR^pos T cells across time points (IP and peak) (A) Left, stacked barplots of CAR^pos T cell subtype composition at the IP and peak. Right, stacked barplots of CAR^pos T cell subtype composition analyzed within the non-proliferative T cells at the IP and peak displayed both overall and by individual patient. (B) IP-peak paired Shannon entropy measure calculated for each patient to evaluate the diversity of the TCR repertoire within the CAR^pos T cells. (C) Occupied TCR repertoire space for CAR^pos T cells by cell type/cluster and time point (IP and peak). Only small (1 < x<=5), medium (5 < x<=20), and large (20 < x<=100) clonotypes were considered. (D and E) Chord diagrams showing the interconnection of shared expanded clonotypes among the different T cell clusters at the IP (D) and peak (E). (F) Scatterplot showing the clonal expansion and the proliferation indexes of the CAR^pos T cells at IP (circles) and peak (triangles). Colors indicate the T cell cluster. Proliferation index was calculated based on scRNA-seq data, and expansion index was computed by Startrac software from scTCR-seq data (see STAR Methods). (G) Alluvial plot depicting the total number of CAR^pos T cell clonotypes per patient. Each bar is color-coded to show the proportion of the different T cell subtypes (clusters), as well as T cell expansion and sample collection time points per patient. Blue lines connect CD8⁺ cytotoxic T cell clonotypes to their corresponding clonal expansions and sample collection time points.

Figure 4

In vivo expansion of CAR^pos γδT cells correlates with treatment outcome (A) UMAP plots color-coded by the expression of the indicated cytotoxic genes. (B) Bar plots showing the absolute number of the identified CAR^pos γδT cells at the IP (light gray) and peak (gray), combining all samples (left) and separating by patient (right). (C) Similarity Jaccard index comparing CD8⁺ and γδT cell clusters. The index was computed considering the top 100 DEGs from each cluster. (D) Upset plot showing the number of shared markers between CD8⁺ and γδT cell clusters, considering the top 100 DEGs from each cluster. (E and F) Volcano plot depicting the differential expression analysis between γδT cells and CD8⁺ cytotoxic T cells at the peak of expansion (E) and between CAR^pos and CAR^neg γδT cells at the peak of expansion (F). (G) Heatmap representing the gene expression signatures related to CAR-T cell cytotoxicity in γδT cells at the peak of expansion, separated by patient. The Ucell method was used to score each signature (see STAR Methods). (H) Proportion of disease-free and relapsed varni-cel-treated B-ALL patients with persisting CAR^pos γδT cells detectable one year after treatment assessed by flow cytometry in the retrospective validation cohort (n = 18; n = 10 and n = 8). (I) Prognostic impact of the number of CAR^pos γδT cells at the IP on 5-year progression-free survival in a validation cohort of 58 r/r DLBCL patients treated with axi-cel (n = 33) and lisa-cel (n = 25) products (high γδT cells n = 28 and low γδT cells n = 30; Log rank test; p value < 0.001).