B-cell-directed CAR T-cell therapy activates CD8+ cytotoxic CARneg bystander T cells in patients and nonhuman primates

Abstract

Chimeric antigen receptor (CAR) T cells hold promise as a therapy for B-cell-derived malignancies, and despite their impressive initial response rates, a significant proportion of patients ultimately experience relapse. Although recent studies have explored the mechanisms of in vivo CAR T-cell function, little is understood about the activation of surrounding CARneg bystander T cells and their potential to enhance tumor responses. We performed single-cell RNA sequencing on nonhuman primate (NHP) and patient-derived T cells to identify the phenotypic and transcriptomic hallmarks of bystander activation of CARneg T cells following B-cell-targeted CAR T-cell therapy. Using a highly translatable CD20 CAR NHP model, we observed a distinct population of activated CD8+ CARneg T cells emerging during CAR T-cell expansion. These bystander CD8+ CARneg T cells exhibited a unique transcriptional signature with upregulation of natural killer-cell markers (KIR3DL2, CD160, and KLRD1), chemokines, and chemokine receptors (CCL5, XCL1, and CCR9), and downregulation of naïve T-cell-associated genes (SELL and CD28). A transcriptionally similar population was identified in patients after a tisagenlecleucel infusion. Mechanistic studies revealed that interleukin-2 (IL-2) and IL-15 exposure induced bystander-like CD8+ T cells in a dose-dependent manner. In vitro activated and patient-derived T cells with a bystander phenotype efficiently killed leukemic cells through a T-cell receptor-independent mechanism. Collectively, to our knowledge, these data provide the first comprehensive identification and profiling of CARneg bystander CD8+ T cells following B-cell-targeting CAR T-cell therapy and suggest a novel mechanism through which CAR T-cell infusion might trigger enhanced antileukemic responses. Patient samples were obtained from the trial #NCT03369353, registered at www.ClinicalTrials.gov.

Graphical abstract

Figure 1.

NHP model of CAR T-cell therapy reveals CD8⁺ CAR^neg T cells with an activation signature. (A) Schematic of sample preparation: T cells, CAR^pos, and CAR^neg cells were flow cytometrically sorted before infusion, from the product, at day 10 (peak expansion) and day 14 (the start of CAR T-cell contraction), and prepared for scRNA-seq and scTCR-seq. (B) CAR T-cell expansion and B-cell aplasia were tracked in animal R.315. (C) Uniform manifold approximation and projection (UMAP) with Leiden clustering of the scRNA-seq data set, colored by the time point. (D) UMAP of the scRNA-seq data set colored by flow cytometrically sorted CAR^pos and CAR^neg cells. (E) UMAP of the scRNA-seq data set, colored by the normalized CAR transcript counts. (F) Unadjusted CAR transcript counts in the sorted CAR^pos and CAR^neg populations. (G) decoupleR weighted sum (WSUM) analysis of a naïve vs memory gene signature in CD4+ CAR^pos and CAR^neg T cells and a naïve vs effector gene signature in CD8⁺ CAR^pos and CAR^neg T cells, with the analysis performed with cells isolated at the time of peak expansion. (H) Violin plots of normalized expression of the CD8⁺ effector molecules granzyme B (GZMB), IFN-γ, and perforin 1 (PRF1) in CD4⁺ and CD8⁺ CAR^pos and CAR^neg T cells.

Figure 2.

Identification of CD8⁺ bystander signature in the NHP model. (A) UMAP with Leiden clustering from NHP recipient R.315. Sixteen clusters are denoted by colors and labeled with inferred cell states. (B) UMAP with normalized CD8a and CD4 expression. (C) Heat map demonstrating normalized and log-scaled expression of selected T-cell activation, effector, and known bystander genes. (D) Cluster composition based on the time point of sample collection. (E) Differential expression of the bystander cluster (cluster 14) vs all other clusters in the data set. The top 6 upregulated (green) and downregulated (red) genes by median log2 fold change are labeled. (F) The 42 genes that exhibited upregulation in bystander signature 1 were classified based on their proposed function in T cells. (G) Heat map of Morisita index for all samples, comparing cluster 14 cells (at contraction and peak expansion) with all other cells using the clonotype ID inferred by Cell Ranger to group cells into clones.

Figure 3.

Bystander CAR T cells can be detected in a validation cohort of 4 additional NHP. (A) UMAP with shared nearest-neighbor clustering across 4 additional animals, colored to identify 23 transcriptional clusters. (B) Bystander score determined by applying bystander signature 1 to each cluster using decoupleR. (C) Heat map of row-scaled normalized expression averaged across each cluster for the top 10 upregulated and downregulated genes identified in bystander signature 1. (D) Composition of each cluster by animal, with an arrow highlighting cluster 22, the cluster with the highest signature score when applying bystander signature 1. (E) Composition of clusters based on sample collection time point, with an arrow highlighting cluster 22, the cluster with the highest signature score when applying bystander signature 1.

Figure 4.

Identification of CD8⁺ CAR^neg bystander T cells in patients after tisagenlecleucel infusion. (A) UMAP of scRNA-seq data from 6 pediatric patients after tisagenlecleucel, colored by patient ID. (B) UMAP with shared nearest-neighbor clustering across all patients, colored by 15 transcriptional clusters. (C) UMAP colored by cell source, as well as normalized expression of genes from bystander signature 4 (CD8, CD160, CCL5, KLRK1 [NKG2D], and KLRD1 [CD94]). The product was not cytometrically sorted into CAR^pos and CAR^neg T cells and therefore contained a mixed population of CAR^pos and CAR^neg T cells. (D) Composition of clusters by cell source, with an arrow highlighting clusters 0 and 12, which were enriched for bystander signature 1. (E) Bystander score determined by applying bystander signature 1 to each CD8⁺ CAR^neg cluster using decoupleR. (F) Composition of clusters by patient ID, with arrows highlighting clusters 0 and 12, which were enriched for bystander signature 1. (G) Shannon diversity of T-cell clones in bystander clusters 12 and 0 vs all CD8⁺ CAR T cells in each patient.

Figure 5.

Stimulation of primary human T cells with gamma cytokines IL-15 or IL-2 generates cells with a bystander phenotype. (A) Representative flow plots demonstrating the gating strategy used to identify T cells expressing the bystander markers CD8, CD160, NKG2D, and CCL5 in unstimulated cells and cells stimulated with IL-2. Cells were gated on CD3⁺/CD8⁺ double-positive T cells and subsequently gated on bystander markers CCL5, NKG2D, and CD160. CD8/CCL5 double-positive cells were assessed for the expression of the bystander marker of NKG2D and CD160. (B) The percentage of total bystander-phenotype-positive CD8⁺ T cells after stimulation with cytokine release syndrome (CRS)-associated cytokines and a paired t test was used for statistical analysis. (C) Percentage of total bystander-phenotype-positive CD8⁺ T cells with the bystander phenotype after titration of IL-2 and (D) IL-15. Stim, stimulated; unstim, unstimulated.

Figure 6.

Cytokine-stimulated and patient–derived primary human bystander-phenotype-positive CD8⁺ T cells are capable of killing leukemia and lymphoma cell lines. (A) Schematic of coculture cytotoxicity assay. Cells were sorted for bystander-phenotype-positive and -negative CD8⁺ T cells and cocultured with Nalm6, β2M knockout Nalm6, or Daudi cells at effector-to-target ratio of 1:1 for 16 hours. (B) Percentage killing of Nalm6 (n = 7; paired t test) and (C) β2M knockout Nalm6 cells by unstimulated (n = 3; unpaired t test), IL-15 (n = 5; paired t test), and IL-2 (n = 3; paired t test)–stimulated bystander-phenotype-positive and -negative CD8⁺ T cells. (D) Percentage killing of Daudi cells after coculture with IL15-stimulated T cells sorted for bystander-phenotype-positive and -negative CD8⁺ T cells (n = 3; paired t test). Experiments were performed in duplicates or triplicates for each donor. Individual donors are represented by the unique symbols in the figure. (E) Frequency of CAR⁺ T cells and CD8⁺, CD94⁺, CD160⁺, and NKG2D⁺ CAR-negative T cells (red cluster) were determined by flow cytometric analysis in PBMCs isolated from a healthy donor, as well as from Pt006 and Pt007, with blood drawn 6 to 9 days after CAR T-cell infusion. (F) Percentage killing of β2M knockout Nalm6 cells 16 hours after coculture with bystander-phenotype-positive and -negative CD8⁺ T cells. This experiment was done with technical triplicates, the mean of all experiments is shown as ± standard error of mean. Stim, stimulated; unstim, unstimulated; WT, wild-type.